x86/mm: fix use-after-free of vma during userfaultfd fault

Syzkaller with KASAN has reported a use-after-free of vma->vm_flags in
__do_page_fault() with the following reproducer:

  mmap(&(0x7f0000000000/0xfff000)=nil, 0xfff000, 0x3, 0x32, 0xffffffffffffffff, 0x0)
  mmap(&(0x7f0000011000/0x3000)=nil, 0x3000, 0x1, 0x32, 0xffffffffffffffff, 0x0)
  r0 = userfaultfd(0x0)
  ioctl$UFFDIO_API(r0, 0xc018aa3f, &(0x7f0000002000-0x18)={0xaa, 0x0, 0x0})
  ioctl$UFFDIO_REGISTER(r0, 0xc020aa00, &(0x7f0000019000)={{&(0x7f0000012000/0x2000)=nil, 0x2000}, 0x1, 0x0})
  r1 = gettid()
  syz_open_dev$evdev(&(0x7f0000013000-0x12)="2f6465762f696e7075742f6576656e742300", 0x0, 0x0)
  tkill(r1, 0x7)

The vma should be pinned by mmap_sem, but handle_userfault() might (in a
return to userspace scenario) release it and then acquire again, so when
we return to __do_page_fault() (with other result than VM_FAULT_RETRY),
the vma might be gone.

Specifically, per Andrea the scenario is
 "A return to userland to repeat the page fault later with a
  VM_FAULT_NOPAGE retval (potentially after handling any pending signal
  during the return to userland). The return to userland is identified
  whenever FAULT_FLAG_USER|FAULT_FLAG_KILLABLE are both set in
  vmf->flags"

However, since commit a3c4fb7c9c2e ("x86/mm: Fix fault error path using
unsafe vma pointer") there is a vma_pkey() read of vma->vm_flags after
that point, which can thus become use-after-free.  Fix this by moving
the read before calling handle_mm_fault().

Reported-by: syzbot <bot+6a5269ce759a7bb12754ed9622076dc93f65a1f6@syzkaller.appspotmail.com>
Reported-by: Dmitry Vyukov <dvyukov@google.com>
Suggested-by: Kirill A. Shutemov <kirill@shutemov.name>
Fixes: 3c4fb7c9c2e ("x86/mm: Fix fault error path using unsafe vma pointer")
Reviewed-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

diff --git a/arch/x86/mm/fault.c b/arch/x86/mm/fault.c
index e2baeaa053a5b9feb76a1587fb4756786ffd76db..7101c281c7ce244b78d78326d9628734573e3e09 100644 (file)
--- a/arch/x86/mm/fault.c
+++ b/arch/x86/mm/fault.c
@@ -1440,7 +1440,17 @@ good_area:
         * make sure we exit gracefully rather than endlessly redo
         * the fault.  Since we never set FAULT_FLAG_RETRY_NOWAIT, if
         * we get VM_FAULT_RETRY back, the mmap_sem has been unlocked.
         * make sure we exit gracefully rather than endlessly redo
         * the fault.  Since we never set FAULT_FLAG_RETRY_NOWAIT, if
         * we get VM_FAULT_RETRY back, the mmap_sem has been unlocked.

+        *
+        * Note that handle_userfault() may also release and reacquire mmap_sem
+        * (and not return with VM_FAULT_RETRY), when returning to userland to
+        * repeat the page fault later with a VM_FAULT_NOPAGE retval
+        * (potentially after handling any pending signal during the return to
+        * userland). The return to userland is identified whenever
+        * FAULT_FLAG_USER|FAULT_FLAG_KILLABLE are both set in flags.
+        * Thus we have to be careful about not touching vma after handling the
+        * fault, so we read the pkey beforehand.

         */
         */

+       pkey = vma_pkey(vma);

        fault = handle_mm_fault(vma, address, flags);
        major |= fault & VM_FAULT_MAJOR;
 
        fault = handle_mm_fault(vma, address, flags);
        major |= fault & VM_FAULT_MAJOR;
 


@@ -1467,7 +1477,6 @@ good_area:
                return;
        }
 
                return;
        }
 

-       pkey = vma_pkey(vma);

        up_read(&mm->mmap_sem);
        if (unlikely(fault & VM_FAULT_ERROR)) {
                mm_fault_error(regs, error_code, address, &pkey, fault);
        up_read(&mm->mmap_sem);
        if (unlikely(fault & VM_FAULT_ERROR)) {
                mm_fault_error(regs, error_code, address, &pkey, fault);