[PATCH -next] mm/page_poison: fix a false memory corruption

[Qian Cai]

The linux-next commit "mm: security: introduce init_on_alloc=1 and
init_on_free=1 boot options" [1] introduced a false positive when
init_on_free=1 and page_poison=on, due to the page_poison expects the
pattern 0xaa when allocating pages which were overwritten by
init_on_free=1 with 0.

It is not possible to switch the order between kernel_init_free_pages()
and kernel_poison_pages() in free_pages_prepare(), because at least on
powerpc the formal will call clear_page() and the subsequence access by
kernel_poison_pages() will trigger the kernel access of bad area errors.

Fix it by treating init_on_free=1 the same as
CONFIG_PAGE_POISONING_ZERO=y.

[1] https://patchwork.kernel.org/patch/10999465/

Signed-off-by: Qian Cai <cai@xxxxxx>
---
 mm/page_poison.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)

diff --git a/mm/page_poison.c b/mm/page_poison.c
index 21d4f97cb49b..272403b992d3 100644
--- a/mm/page_poison.c
+++ b/mm/page_poison.c
@@ -68,22 +68,26 @@ static void check_poison_mem(unsigned char *mem, size_t bytes)
 	static DEFINE_RATELIMIT_STATE(ratelimit, 5 * HZ, 10);
 	unsigned char *start;
 	unsigned char *end;
+	int pattern = PAGE_POISON;
 
 	if (IS_ENABLED(CONFIG_PAGE_POISONING_NO_SANITY))
 		return;
 
-	start = memchr_inv(mem, PAGE_POISON, bytes);
+	if (static_branch_unlikely(&init_on_free))
+		pattern = 0;
+
+	start = memchr_inv(mem, pattern, bytes);
 	if (!start)
 		return;
 
 	for (end = mem + bytes - 1; end > start; end--) {
-		if (*end != PAGE_POISON)
+		if (*end != pattern)
 			break;
 	}
 
 	if (!__ratelimit(&ratelimit))
 		return;
-	else if (start == end && single_bit_flip(*start, PAGE_POISON))
+	else if (start == end && single_bit_flip(*start, pattern))
 		pr_err("pagealloc: single bit error\n");
 	else
 		pr_err("pagealloc: memory corruption\n");
-- 
1.8.3.1