[lkml]   [2019]   [Jun]   [24]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectCVE-2019-12817: Linux kernel: powerpc: Unrelated processes may be able to read/write to each other's virtual memory
The Linux kernel for powerpc since 4.17 has a bug where unrelated processes may
be able to read/write to each other's virtual memory under certain conditions.

This bug only affects machines using 64-bit CPUs with the hash page table MMU,
see below for more detail on affected CPUs.

To trigger the bug a process must allocate memory above 512TB. That only happens
if userspace explicitly requests it with mmap(). That process must then fork(),
at this point the child incorrectly inherits the "context id" of the parent
associated with the mapping above 512TB. It may then be possible for the
parent/child to write to each other's mappings above 512TB, which should not be
possible, and constitutes memory corruption.

If instead the child process exits, all its context ids are freed, including the
context id that is still in use by the parent for the mapping above 512TB. That
id can then be reallocated to a third process, that process can then read/write
to the parent's mapping above 512TB. Additionally if the freed id is used for
the third process's primary context id, then the parent is able to read/write to
the third process's mappings *below* 512TB.

If the parent and child both exit before another process is allocated the freed
context id, the kernel will notice the double free of the id and print a warning
such as:

ida_free called for id=103 which is not allocated.
WARNING: CPU: 8 PID: 7293 at lib/idr.c:520 ida_free_rc+0x1b4/0x1d0

The bug was introduced in commit:
f384796c40dc ("powerpc/mm: Add support for handling > 512TB address in SLB miss")

Which was originally merged in v4.17.

Only machines using the hash page table (HPT) MMU are affected, eg. PowerPC 970
(G5), PA6T, Power5/6/7/8/9. By default Power9 bare metal machines (powernv) use
the Radix MMU and are not affected, unless the machine has been explicitly
booted in HPT mode (using disable_radix on the kernel command line). KVM guests
on Power9 may be affected if the host or guest is configured to use the HPT MMU.
LPARs under PowerVM on Power9 are affected as they always use the HPT MMU.
Kernels built with PAGE_SIZE=4K are not affected.

The upstream fix is here:

There's also a kernel selftest to verify the fix:

Or a similar standalone version is included below.


cat > test.c <<EOF
#undef NDEBUG

#include <assert.h>
#include <stdio.h>
#include <stdlib.h>
#include <sys/mman.h>
#include <sys/types.h>
#include <sys/wait.h>
#include <unistd.h>

#define MAP_FIXED_NOREPLACE MAP_FIXED // "Should be safe" above 512TB

int main(void)
int p2c[2], c2p[2], rc, status, c, *p;
unsigned long page_size;
pid_t pid;

page_size = sysconf(_SC_PAGESIZE);
if (page_size != 65536) {
printf("Unsupported page size - not affected\n");
return 1;

// Create a mapping at 512TB to allocate an extended_id
p = mmap((void *)(512ul << 40), page_size, PROT_READ | PROT_WRITE,
if (p == MAP_FAILED) {
printf("Error: couldn't mmap(), confirm kernel has 4TB support\n");
return 1;

printf("parent writing %p = 1\n", p);
*p = 1;

assert(pipe(p2c) != -1 && pipe(c2p) != -1);

pid = fork();
if (pid == 0) {
assert(read(p2c[0], &c, 1) == 1);

pid = getpid();
printf("child writing %p = %d\n", p, pid);
*p = pid;

assert(write(c2p[1], &c, 1) == 1);
assert(read(p2c[0], &c, 1) == 1);

c = 0;
assert(write(p2c[1], &c, 1) == 1);
assert(read(c2p[0], &c, 1) == 1);

// Prevent compiler optimisation
asm volatile("" : : : "memory");

rc = 0;
printf("parent reading %p = %d\n", p, *p);
if (*p != 1) {
printf("Error: BUG! parent saw child's write! *p = %d\n", *p);
rc = 1;

assert(write(p2c[1], &c, 1) == 1);
assert(waitpid(pid, &status, 0) != -1);
assert(WIFEXITED(status) && WEXITSTATUS(status) == 0);

if (rc == 0)
printf("success: test completed OK\n");

return rc;
[unhandled content-type:application/pgp-signature]
 \ /
  Last update: 2019-06-24 16:45    [W:0.037 / U:2.680 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site