]> git.baikalelectronics.ru Git - kernel.git/commitdiff
mm: do not rely on mm == current->mm in __get_user_pages_locked
authorJason A. Donenfeld <Jason@zx2c4.com>
Mon, 28 Sep 2020 10:35:07 +0000 (12:35 +0200)
committerLinus Torvalds <torvalds@linux-foundation.org>
Mon, 28 Sep 2020 16:21:50 +0000 (09:21 -0700)
It seems likely this block was pasted from internal_get_user_pages_fast,
which is not passed an mm struct and therefore uses current's.  But
__get_user_pages_locked is passed an explicit mm, and current->mm is not
always valid. This was hit when being called from i915, which uses:

  pin_user_pages_remote->
    __get_user_pages_remote->
      __gup_longterm_locked->
        __get_user_pages_locked

Before, this would lead to an OOPS:

  BUG: kernel NULL pointer dereference, address: 0000000000000064
  #PF: supervisor write access in kernel mode
  #PF: error_code(0x0002) - not-present page
  CPU: 10 PID: 1431 Comm: kworker/u33:1 Tainted: P S   U     O      5.9.0-rc7+ #140
  Hardware name: LENOVO 20QTCTO1WW/20QTCTO1WW, BIOS N2OET47W (1.34 ) 08/06/2020
  Workqueue: i915-userptr-acquire __i915_gem_userptr_get_pages_worker [i915]
  RIP: 0010:__get_user_pages_remote+0xd7/0x310
  Call Trace:
   __i915_gem_userptr_get_pages_worker+0xc8/0x260 [i915]
   process_one_work+0x1ca/0x390
   worker_thread+0x48/0x3c0
   kthread+0x114/0x130
   ret_from_fork+0x1f/0x30
  CR2: 0000000000000064

This commit fixes the problem by using the mm pointer passed to the
function rather than the bogus one in current.

Fixes: 5d134b292004 ("mm: Introduce mm_struct.has_pinned")
Tested-by: Chris Wilson <chris@chris-wilson.co.uk>
Reported-by: Harald Arnesen <harald@skogtun.org>
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/gup.c

index dfe781d2ad4c16ddaa09aea3c476e6d1cb8fe953..e869c634cc9a683fa739cf2fa4aa99be204df210 100644 (file)
--- a/mm/gup.c
+++ b/mm/gup.c
@@ -1256,7 +1256,7 @@ static __always_inline long __get_user_pages_locked(struct mm_struct *mm,
        }
 
        if (flags & FOLL_PIN)
-               atomic_set(&current->mm->has_pinned, 1);
+               atomic_set(&mm->has_pinned, 1);
 
        /*
         * FOLL_PIN and FOLL_GET are mutually exclusive. Traditional behavior