Message-Id: <20250603-fork-tearing-v1-1-a7f64b7cfc96@google.com>
Date: Tue, 03 Jun 2025 20:21:02 +0200
From: Jann Horn <jannh@...gle.com>
To: Andrew Morton <akpm@...ux-foundation.org>, 
 David Hildenbrand <david@...hat.com>, 
 Lorenzo Stoakes <lorenzo.stoakes@...cle.com>, 
 "Liam R. Howlett" <Liam.Howlett@...cle.com>, 
 Vlastimil Babka <vbabka@...e.cz>, Mike Rapoport <rppt@...nel.org>, 
 Suren Baghdasaryan <surenb@...gle.com>, Michal Hocko <mhocko@...e.com>, 
 linux-mm@...ck.org
Cc: Peter Xu <peterx@...hat.com>, linux-kernel@...r.kernel.org, 
 Jann Horn <jannh@...gle.com>, stable@...r.kernel.org
Subject: [PATCH 1/2] mm/memory: ensure fork child sees coherent memory
 snapshot

When fork() encounters possibly-pinned pages, those pages are immediately
copied instead of just marking PTEs to make CoW happen later. If the parent
is multithreaded, this can cause the child to see memory contents that are
inconsistent in multiple ways:

1. We are copying the contents of a page with a memcpy() while userspace
   may be writing to it. This can cause the resulting data in the child to
   be inconsistent.
2. After we've copied this page, future writes to other pages may
   continue to be visible to the child while future writes to this page are
   no longer visible to the child.

This means the child could theoretically see incoherent states where
allocator freelists point to objects that are actually in use or stuff like
that. A mitigating factor is that, unless userspace already has a deadlock
bug, userspace can pretty much only observe such issues when fancy lockless
data structures are used (because if another thread was in the middle of
mutating data during fork() and the post-fork child tried to take the mutex
protecting that data, it might wait forever).

On top of that, this issue is only observable when pages are either
DMA-pinned or appear false-positive-DMA-pinned due to a page having >=1024
references and the parent process having used DMA-pinning at least once
before.

Fixes: 70e806e4e645 ("mm: Do early cow for pinned pages during fork() for ptes")
Cc: stable@...r.kernel.org
Signed-off-by: Jann Horn <jannh@...gle.com>
---
 mm/memory.c | 18 ++++++++++++++++++
 1 file changed, 18 insertions(+)

diff --git a/mm/memory.c b/mm/memory.c
index 49199410805c..b406dfda976b 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -917,7 +917,25 @@ copy_present_page(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma
 	/*
 	 * We have a prealloc page, all good!  Take it
 	 * over and copy the page & arm it.
+	 *
+	 * One nasty aspect is that we could be in a multithreaded process or
+	 * such, where another thread is in the middle of writing to memory
+	 * while this thread is forking. As long as we're just marking PTEs as
+	 * read-only to make copy-on-write happen *later*, that's easy; we just
+	 * need to do a single TLB flush before dropping the mmap/VMA locks, and
+	 * that's enough to guarantee that the child gets a coherent snapshot of
+	 * memory.
+	 * But here, where we're doing an immediate copy, we must ensure that
+	 * threads in the parent process can no longer write into the page being
+	 * copied until we're done forking.
+	 * This means that we still need to mark the source PTE as read-only,
+	 * with an immediate TLB flush.
+	 * (To make the source PTE writable again after fork() is done, we can
+	 * rely on the page fault handler to do that lazily, thanks to
+	 * PageAnonExclusive().)
 	 */
+	ptep_set_wrprotect(src_vma->vm_mm, addr, src_pte);
+	flush_tlb_page(src_vma, addr);
 
 	if (copy_mc_user_highpage(&new_folio->page, page, addr, src_vma))
 		return -EHWPOISON;

-- 
2.49.0.1204.g71687c7c1d-goog

Powered by blists - more mailing lists