Userfaultfd did not create private memory when UFFDIO_COPY was invoked
on a MAP_PRIVATE shmem mapping. Instead it wrote to the shmem file,
even when that had not been opened for writing. Though, fortunately,
that could only happen where there was a hole in the file.
Fix the shmem-backed implementation of UFFDIO_COPY to create private
memory for MAP_PRIVATE mappings. The hugetlbfs-backed implementation
was already correct.
This change is visible to userland, if userfaultfd has been used in
unintended ways: so it introduces a small risk of incompatibility, but
is necessary in order to respect file permissions.
An app that uses UFFDIO_COPY for anything like postcopy live migration
won't notice the difference, and in fact it'll run faster because
there will be no copy-on-write and memory waste in the tmpfs pagecache
anymore.
Userfaults on MAP_PRIVATE shmem keep triggering only on file holes
like before.
The real zeropage can also be built on a MAP_PRIVATE shmem mapping
through UFFDIO_ZEROPAGE and that's safe because the zeropage pte is
never dirty, in turn even an mprotect upgrading the vma permission
from PROT_READ to PROT_READ|PROT_WRITE won't make the zeropage pte
writable.
Reported-by: Mike Rapoport <r...@linux.ibm.com>
Reviewed-by: Hugh Dickins <hu...@google.com>
Cc: sta...@vger.kernel.org
Fixes: 4c27fe4c4c84 ("userfaultfd: shmem: add shmem_mcopy_atomic_pte for
userfaultfd support")
Signed-off-by: Andrea Arcangeli <aarca...@redhat.com>
---
mm/userfaultfd.c | 15 +++++++++++++--
1 file changed, 13 insertions(+), 2 deletions(-)
diff --git a/mm/userfaultfd.c b/mm/userfaultfd.c
index 46c8949e5f8f..471b6457f95f 100644
--- a/mm/userfaultfd.c
+++ b/mm/userfaultfd.c
@@ -380,7 +380,17 @@ static __always_inline ssize_t mfill_atomic_pte(struct
mm_struct *dst_mm,
{
ssize_t err;
- if (vma_is_anonymous(dst_vma)) {
+ /*
+ * The normal page fault path for a shmem will invoke the
+ * fault, fill the hole in the file and COW it right away. The
+ * result generates plain anonymous memory. So when we are
+ * asked to fill an hole in a MAP_PRIVATE shmem mapping, we'll
+ * generate anonymous memory directly without actually filling
+ * the hole. For the MAP_PRIVATE case the robustness check
+ * only happens in the pagetable (to verify it's still none)
+ * and not in the radix tree.
+ */
+ if (!(dst_vma->vm_flags & VM_SHARED)) {
if (!zeropage)
err = mcopy_atomic_pte(dst_mm, dst_pmd, dst_vma,
dst_addr, src_addr, page);
@@ -489,7 +499,8 @@ static __always_inline ssize_t __mcopy_atomic(struct
mm_struct *dst_mm,
* dst_vma.
*/
err = -ENOMEM;
- if (vma_is_anonymous(dst_vma) && unlikely(anon_vma_prepare(dst_vma)))
+ if (!(dst_vma->vm_flags & VM_SHARED) &&
+ unlikely(anon_vma_prepare(dst_vma)))
goto out_unlock;
while (src_addr < src_start + len) {
