Openwall GNU/*/Linux - a small security-enhanced Linux distro for servers
[<prev] [next>] [day] [month] [year] [list]
Date: Thu, 4 Jun 2020 19:01:12 +0800
From: Fan Yang <fan_yang@...U.EDU.CN>
To: oss-security@...ts.openwall.com
Subject: CVE-2020-10757 Linux kernel: mremap hugepage mmaped DAX nvdimm may
 cause corrupted page table

Hi all,

NOTE: this bug have been assigned CVE id CVE-2020-10757.  Its impact dates
back to the commit 5c7fb56e5e3f ("mm, dax: dax-pmd vs thp-pmd vs
hugetlbfs-pmd”), at kernel version v4.5.  And it can be fixed by the patch
here (https://lkml.org/lkml/2020/6/4/314).

Description
===========

I observed this bug when mremap a mmaped DAX nvdimm to a mmaped anonymous
memory region.  The mremap system call returns successfully but when access
the region afterwards, the program get killed due to corrupted page table:

try_mremap: Corrupted page table at address 7facc4fd1000
PGD 800000015beee067 P4D 800000015beee067 PUD 40695a067 PMD 1614ec067 PTE 6969696969696969
Bad pagetable: 000f [#3] SMP PTI
CPU: 6 PID: 11264 Comm: try_mremap Tainted: G    B D W         5.6.6-300.fc32.x86_64 #1
Hardware name: System manufacturer System Product Name/PRIME Z270M-PLUS, BIOS 0601 01/13/2017
RIP: 0033:0x4012fc
Code: 00 00 e8 87 fd ff ff 48 89 45 e8 48 8b 45 e8 48 3b 45 f8 74 11 bf 3a 20 40 00 e8 8f fd ff ff b8 ff ff ff ff eb 0c 48 8b 45 f8 <c6> 00 aa b8 00 00 00 00 c9 c3 66 2e 0f 1f 84 00 00 00 00 00 f3 0f
RSP: 002b:00007ffde40d8d00 EFLAGS: 00010246
RAX: 00007facc4fd1000 RBX: 0000000000000000 RCX: 00007facc4f0255e
RDX: 0000000000001000 RSI: 0000000000001000 RDI: 00007facc4a00000
RBP: 00007ffde40d8d30 R08: 00007facc4fd1000 R09: 0000000000000000
R10: 0000000000000003 R11: 0000000000000202 R12: 00000000004010a0
R13: 0000000000000000 R14: 0000000000000000 R15: 0000000000000000
...
BUG: Bad page map in process try_mremap  pte:6969696969696969 pmd:1614ec067
addr:00007facc4fd1000 vm_flags:288000fb anon_vma:0000000000000000 mapping:ffff9badd87a28a0 index:0
file:try_mremap fault:ext4_dax_fault mmap:ext4_file_mmap readpage:0x0
CPU: 6 PID: 11264 Comm: try_mremap Tainted: G    B D W         5.6.6-300.fc32.x86_64 #1
Hardware name: System manufacturer System Product Name/PRIME Z270M-PLUS, BIOS 0601 01/13/2017
Call Trace:
dump_stack+0x64/0x88
print_bad_pte.cold+0x95/0xbf
vm_normal_page+0xbe/0xd0
unmap_page_range+0x68b/0xeb0
unmap_vmas+0x6a/0xd0
exit_mmap+0x97/0x170
mmput+0x61/0x140
do_exit+0x2f3/0xae0
rewind_stack_do_exit+0x17/0x20

Note that the weird 0x69 in the pte value is the ascii of "i" which I had
written to the memory.  This bug is due to in move_page_tables:mm/mremap.c,
the condition to handle a huge pmd is as follows:

		if (is_swap_pmd(*old_pmd) || pmd_trans_huge(*old_pmd)) {

However, the DAX file is mapped as huge page but it is not transparent huge
page.  So the huge pmd is not split, the physical page the pmd points to is
treated as a page table (but actually it is a 2M data page).  Then
move_ptes uses the value of the "pte" to update the pte where the page
remap to, and the mremap system call returns successfully.  Afterwards, the
access to the new address incurs corrupted page table.

Re-produce Instructions
=======================

1. one need to have a machine with Intel Optane DC Persistent Memory
(https://www.intel.com/content/www/us/en/architecture-and-technology/optane-dc-persistent-memory.html),
or run a VM with a virtualized NVDIMM
(https://software.intel.com/content/www/us/en/develop/articles/how-to-emulate-persistent-memory-on-an-intel-architecture-server.html).

2. mount a DAX file system (e.g., I use ext4).

3. Write a userspace program to mremap a DAX mmaped file to a mmaped
  anonymous memory region.

Here is the code I use:

#define _GNU_SOURCE
#include <sys/mman.h>
#include <stdio.h>
#include <unistd.h>
#include <string.h>
#include <fcntl.h>
#include <stdlib.h>
#include <errno.h>

#define PROT			PROT_READ|PROT_WRITE

#define REGION_PM_TMP_PATH	"/mnt/pmem0/try_mremap"

#define REGION_MEM_SIZE 4096*4
#define REGION_PM_SIZE	4096*512
#define REMAP_MEM_OFF   0
#define REMAP_PM_OFF    0
#define REMAP_SIZE      4096

char * map_tmp_pm_region(void)
{
	int fd;

	fd = open(REGION_PM_TMP_PATH, O_RDWR|O_CREAT, 0644);
	if (fd < 0) {
		perror(REGION_PM_TMP_PATH);
		exit(-1);
	}

	if (ftruncate(fd, REGION_PM_SIZE)) {
		perror("ftruncate");
		exit(-1);
	}

	return mmap(NULL, REGION_PM_SIZE, PROT, MAP_SHARED_VALIDATE|MAP_SYNC,
		    fd, 0);
}

int main(int argc, char **argv)
{
	char *regm, *regp, *remap;
	int ret;

	regm = mmap(NULL, REGION_MEM_SIZE, PROT, MAP_PRIVATE|MAP_ANONYMOUS,
		    -1, 0);
	if (regm == MAP_FAILED) {
		perror("regm");
		return -1;
	}

	regp = map_tmp_pm_region();
	if (regp == MAP_FAILED) {
		perror("regp");
		return -1;
	}

	memset(regm, 'a', REGION_MEM_SIZE);
	memset(regp, 'i', REGION_PM_SIZE);

	remap = mremap(regp + REMAP_PM_OFF, REMAP_SIZE, REMAP_SIZE,
		       MREMAP_MAYMOVE|MREMAP_FIXED, regm + REMAP_MEM_OFF);
	if (remap != regm + REMAP_MEM_OFF) {
		perror("mremap");
		return -1;
	}

	*regm = 0xAA;		/* write anything to the address */
	return 0;
}

The Patch
=========

arch/x86/include/asm/pgtable.h | 1 +
mm/mremap.c                    | 2 +-
2 files changed, 2 insertions(+), 1 deletion(-)

diff --git a/arch/x86/include/asm/pgtable.h b/arch/x86/include/asm/pgtable.h
index 4d02e64af1b3..19cdeebfbde6 100644
--- a/arch/x86/include/asm/pgtable.h
+++ b/arch/x86/include/asm/pgtable.h
@@ -257,6 +257,7 @@ static inline int pmd_large(pmd_t pte)
}

#ifdef CONFIG_TRANSPARENT_HUGEPAGE
+/* NOTE: when predicate huge page, consider also pmd_devmap, or use pmd_large */
static inline int pmd_trans_huge(pmd_t pmd)
{
	return (pmd_val(pmd) & (_PAGE_PSE|_PAGE_DEVMAP)) == _PAGE_PSE;
diff --git a/mm/mremap.c b/mm/mremap.c
index 6aa6ea605068..57b1f999f789 100644
--- a/mm/mremap.c
+++ b/mm/mremap.c
@@ -266,7 +266,7 @@ unsigned long move_page_tables(struct vm_area_struct *vma,
		new_pmd = alloc_new_pmd(vma->vm_mm, vma, new_addr);
		if (!new_pmd)
			break;
-		if (is_swap_pmd(*old_pmd) || pmd_trans_huge(*old_pmd)) {
+		if (is_swap_pmd(*old_pmd) || pmd_trans_huge(*old_pmd) || pmd_devmap(*old_pmd)) {
			if (extent == HPAGE_PMD_SIZE) {
				bool moved;
				/* See comment in move_ptes() */
-- 
2.25.4


Powered by blists - more mailing lists

Your e-mail address:

Please check out the Open Source Software Security Wiki, which is counterpart to this mailing list.

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.