Skip to contentSkip to conversation

: xarray: Fix potential out of bounds access

PATCH In review Cyrill Gorcunov 2 participants 3 messages Jan 14, 2019

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 : xarray: Fix potential out of bounds access

Since the mark is used as an array index we should use preincrement to not access the XA_MARK_MAX index.

---
Matthew, take a look please, I suspect we may access the mark index out of allocated one. Compile tested only. It comes from 58d6ea3085f2e53714810a513c61629f6d2be0a6

 lib/xarray.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
lib/xarray.c +1 -1
Index: linux-next.git/lib/xarray.c
===================================================================
@@ -129,7 +129,7 @@ static void xas_squash_marks(const struc
 			continue;
 		__set_bit(xas->xa_offset, marks);
 		bitmap_clear(marks, xas->xa_offset + 1, xas->xa_sibs);
-	} while (mark++ != (__force unsigned)XA_MARK_MAX);
+	} while (++mark != (__force unsigned)XA_MARK_MAX);
 }
 
 /* extracts the offset within this node from the index */
maintainer
#1 lore raw
Quoted 2 lines · Cyrill Gorcunov

Since the mark is used as an array index we should use preincrement to not access the XA_MARK_MAX index.

But XA_MARK_MAX is inclusive:

include/linux/xarray.h:#define XA_MARK_MAX XA_MARK_2

so we actually want to access XA_MARK_MAX. Now, we don't have a test in the test-suite that fails as a result of your patch, so that needs to get fixed. How about this:

From: Matthew Wilcox <[email protected]>
Date: Mon, 14 Jan 2019 13:57:31 -0500
Subject: [PATCH] XArray tests: Check mark 2 gets squashed

We do not currently check that the loop in xas_squash_marks() doesn't have an off-by-one error in it. It didn't, but a patch which introduced an off-by-one error wasn't caught by any existing test. Switch the roles of XA_MARK_1 and XA_MARK_2 to catch that bug.

---
 lib/test_xarray.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
lib/test_xarray.c +3 -3
index 3cf17338b0a4..c596a957f764 100644
@@ -199,7 +199,7 @@ static noinline void check_xa_mark_1(struct xarray *xa, unsigned long index)
 		XA_BUG_ON(xa, xa_store_index(xa, index + 1, GFP_KERNEL));
 		xa_set_mark(xa, index + 1, XA_MARK_0);
 		XA_BUG_ON(xa, xa_store_index(xa, index + 2, GFP_KERNEL));
-		xa_set_mark(xa, index + 2, XA_MARK_1);
+		xa_set_mark(xa, index + 2, XA_MARK_2);
 		XA_BUG_ON(xa, xa_store_index(xa, next, GFP_KERNEL));
 		xa_store_order(xa, index, order, xa_mk_index(index),
 				GFP_KERNEL);
@@ -209,8 +209,8 @@ static noinline void check_xa_mark_1(struct xarray *xa, unsigned long index)
 			void *entry;
 
 			XA_BUG_ON(xa, !xa_get_mark(xa, i, XA_MARK_0));
-			XA_BUG_ON(xa, !xa_get_mark(xa, i, XA_MARK_1));
-			XA_BUG_ON(xa, xa_get_mark(xa, i, XA_MARK_2));
+			XA_BUG_ON(xa, xa_get_mark(xa, i, XA_MARK_1));
+			XA_BUG_ON(xa, !xa_get_mark(xa, i, XA_MARK_2));
 
 			/* We should see two elements in the array */
 			rcu_read_lock();
Signature · 1 line
2.20.1
#2 lore raw
Quoted 7 lines · Matthew Wilcox
Quoted 2 lines · Cyrill Gorcunov

Since the mark is used as an array index we should use preincrement to not access the XA_MARK_MAX index.

But XA_MARK_MAX is inclusive:

include/linux/xarray.h:#define XA_MARK_MAX XA_MARK_2

Indeed, I misread the variable name.

Quoted 3 lines

so we actually want to access XA_MARK_MAX. Now, we don't have a test in the test-suite that fails as a result of your patch, so that needs to get fixed. How about this:

Looks great. Thank you!