mm: numa: recheck for transhuge pages under lock during protection changes

Sasha reported the following bug using trinity

[  886.745765] kernel BUG at mm/mprotect.c:149!
[  886.746831] invalid opcode: 0000 [#1] PREEMPT SMP DEBUG_PAGEALLOC
[  886.748511] Dumping ftrace buffer:
[  886.749998]    (ftrace buffer empty)
[  886.750110] Modules linked in:
[  886.751396] CPU: 20 PID: 26219 Comm: trinity-c216 Tainted: G        W    3.14.0-rc5-next-20140305-sasha-00011-ge06f5f3-dirty torvalds#105
[  886.751396] task: ffff8800b6c80000 ti: ffff880228436000 task.ti: ffff880228436000
[  886.751396] RIP: 0010:[<ffffffff812aab33>]  [<ffffffff812aab33>] change_protection_range+0x3b3/0x500
[  886.751396] RSP: 0000:ffff880228437da8  EFLAGS: 00010282
[  886.751396] RAX: 8000000527c008e5 RBX: 00007f647916e000 RCX: 0000000000000000
[  886.751396] RDX: ffff8802ef488e40 RSI: 00007f6479000000 RDI: 8000000527c008e5
[  886.751396] RBP: ffff880228437e78 R08: 0000000000000000 R09: 0000000000000000
[  886.751396] R10: 0000000000000001 R11: 0000000000000000 R12: ffff8802ef488e40
[  886.751396] R13: 00007f6479000000 R14: 00007f647916e000 R15: 00007f646e34e000
[  886.751396] FS:  00007f64b28d4700(0000) GS:ffff88052ba00000(0000) knlGS:0000000000000000
[  886.751396] CS:  0010 DS: 0000 ES: 0000 CR0: 000000008005003b
[  886.751396] CR2: 00007f64aed83af8 CR3: 0000000206e52000 CR4: 00000000000006a0
[  886.751396] Stack:
[  886.751396]  ffff880200000001 ffffffff8447f152 ffff880228437dd8 ffff880228af3000
[  886.769142]  00007f646e34e000 00007f647916dfff 0000000000000000 00007f647916e000
[  886.769142]  ffff880206e527f0 00007f647916dfff 0000000000000000 0000000000000000
[  886.769142] Call Trace:
[  886.769142]  [<ffffffff8447f152>] ? preempt_count_sub+0xe2/0x120
[  886.769142]  [<ffffffff812aaca5>] change_protection+0x25/0x30
[  886.769142]  [<ffffffff812c3eeb>] change_prot_numa+0x1b/0x30
[  886.769142]  [<ffffffff8118df49>] task_numa_work+0x279/0x360
[  886.769142]  [<ffffffff8116c57e>] task_work_run+0xae/0xf0
[  886.769142]  [<ffffffff8106ffbe>] do_notify_resume+0x8e/0xe0
[  886.769142]  [<ffffffff8447a93b>] retint_signal+0x4d/0x92
[  886.769142] Code: 49 8b 3c 24 48 83 3d fc 2e ba 04 00 75 12 0f 0b 0f 1f 84 00 00 00 00 00 eb fe 66 0f 1f 44 00 00 48 89 f8 66 66 66 90 84 c0 79 0d <0f> 0b 0f 1f 00 eb fe 66 0f 1f 44 00 00 8b 4d 9c 44 8b 4d 98
+89
[  886.769142] RIP  [<ffffffff812aab33>] change_protection_range+0x3b3/0x500
[  886.769142]  RSP <ffff880228437da8>

The VM_BUG_ON was added by the patch "mm,numa: reorganize
change_pmd_range". The race existed without the patch but was just harder
to hit.

The problem is that a transhuge check is made without holding the PTL.
It's possible at the time of the check that a parallel fault clears the
pmd and inserts a new one which then triggers the VM_BUG_ON check.  This
patch removes the VM_BUG_ON but fixes the race by rechecking transhuge
under the PTL when marking page tables for NUMA hinting and bailing if a
race occurred.  It is not a problem for calls to mprotect() as they hold
mmap_sem for write.

Signed-off-by: Mel Gorman <mgorman@suse.de>
Reported-by: Sasha Levin <sasha.levin@oracle.com>
Reviewed-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

Author: Mel Gorman

mm/mprotect.c

@@ -46,6 +46,17 @@ static unsigned long change_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
 	unsigned long pages = 0;
 
 	pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
+
+	/*
+	 * For a prot_numa update we only hold mmap_sem for read so there is a
+	 * potential race with faulting where a pmd was temporarily none so
+	 * recheck it under the lock and bail if we race
+	 */
+	if (prot_numa && unlikely(pmd_trans_huge(*pmd))) {
+		pte_unmap_unlock(pte, ptl);
+		return 0;
+	}
+
 	arch_enter_lazy_mmu_mode();
 	do {
 		oldpte = *pte;
@@ -132,12 +143,13 @@ static inline unsigned long change_pmd_range(struct vm_area_struct *vma,
 						pages += HPAGE_PMD_NR;
 						nr_huge_updates++;
 					}
+
+					/* huge pmd was handled */
 					continue;
 				}
 			}
 			/* fall through, the trans huge pmd just split */
 		}
-		VM_BUG_ON(pmd_trans_huge(*pmd));
 		this_pages = change_pte_range(vma, pmd, addr, next, newprot,
 				 dirty_accountable, prot_numa);
 		pages += this_pages;