2018-03-23 18:12:32

by Dave Hansen

[permalink] [raw]
Subject: [PATCH 0/9] x86, pkeys: two protection keys bug fixes

This fixes two bugs, and adds selftests to make sure they stay fixed:

1. pkey 0 was not usable via mprotect_pkey() because it had never
been explicitly allocated.
2. mprotect(PROT_EXEC) memory could sometimes be left with the
implicit exec-only protection key assigned.

I already posted #1 previously. I'm including them both here because
I don't think it's been picked up in case folks want to pull these
all in a single bundle.

Cc: Ram Pai <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Dave Hansen <[email protected]>
Cc: Michael Ellermen <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Andrew Morton <[email protected]>p
Cc: Shuah Khan <[email protected]>
Cc: Shakeel Butt <[email protected]>


2018-03-23 18:12:39

by Dave Hansen

[permalink] [raw]
Subject: [PATCH 8/9] x86, pkeys, selftests: add allow faults on unknown keys


From: Dave Hansen <[email protected]>

The exec-only pkey is allocated inside the kernel and userspace
is not told what it is. So, allow PK faults to occur that have
an unknown key.

Signed-off-by: Dave Hansen <[email protected]>
Cc: Ram Pai <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Dave Hansen <[email protected]>
Cc: Michael Ellermen <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Andrew Morton <[email protected]>
Cc: Shuah Khan <[email protected]>
---

b/tools/testing/selftests/x86/protection_keys.c | 10 +++++++++-
1 file changed, 9 insertions(+), 1 deletion(-)

diff -puN tools/testing/selftests/x86/protection_keys.c~pkeys-selftests-unknown-exec-only-key tools/testing/selftests/x86/protection_keys.c
--- a/tools/testing/selftests/x86/protection_keys.c~pkeys-selftests-unknown-exec-only-key 2018-03-21 15:47:51.985198917 -0700
+++ b/tools/testing/selftests/x86/protection_keys.c 2018-03-21 15:47:51.988198917 -0700
@@ -922,13 +922,21 @@ void *malloc_pkey(long size, int prot, u
}

int last_pkru_faults;
+#define UNKNOWN_PKEY -2
void expected_pk_fault(int pkey)
{
dprintf2("%s(): last_pkru_faults: %d pkru_faults: %d\n",
__func__, last_pkru_faults, pkru_faults);
dprintf2("%s(%d): last_si_pkey: %d\n", __func__, pkey, last_si_pkey);
pkey_assert(last_pkru_faults + 1 == pkru_faults);
- pkey_assert(last_si_pkey == pkey);
+
+ /*
+ * For exec-only memory, we do not know the pkey in
+ * advance, so skip this check.
+ */
+ if (pkey != UNKNOWN_PKEY)
+ pkey_assert(last_si_pkey == pkey);
+
/*
* The signal handler shold have cleared out PKRU to let the
* test program continue. We now have to restore it.
_

2018-03-23 18:12:45

by Dave Hansen

[permalink] [raw]
Subject: [PATCH 9/9] x86, pkeys, selftests: add PROT_EXEC test


From: Dave Hansen <[email protected]>

Under the covers, implement executable-only memory with
protection keys when userspace calls mprotect(PROT_EXEC).

But, we did not have a selftest for that. Now we do.

Signed-off-by: Dave Hansen <[email protected]>
Cc: Ram Pai <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Dave Hansen <[email protected]>
Cc: Michael Ellermen <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Andrew Morton <[email protected]>
Cc: Shuah Khan <[email protected]>
---

b/tools/testing/selftests/x86/protection_keys.c | 51 ++++++++++++++++++++++--
1 file changed, 47 insertions(+), 4 deletions(-)

diff -puN tools/testing/selftests/x86/protection_keys.c~pkeys-selftests-prot_exec tools/testing/selftests/x86/protection_keys.c
--- a/tools/testing/selftests/x86/protection_keys.c~pkeys-selftests-prot_exec 2018-03-23 10:46:03.976813119 -0700
+++ b/tools/testing/selftests/x86/protection_keys.c 2018-03-23 10:46:03.980813119 -0700
@@ -930,10 +930,10 @@ void expected_pk_fault(int pkey)
dprintf2("%s(%d): last_si_pkey: %d\n", __func__, pkey, last_si_pkey);
pkey_assert(last_pkru_faults + 1 == pkru_faults);

- /*
- * For exec-only memory, we do not know the pkey in
- * advance, so skip this check.
- */
+ /*
+ * For exec-only memory, we do not know the pkey in
+ * advance, so skip this check.
+ */
if (pkey != UNKNOWN_PKEY)
pkey_assert(last_si_pkey == pkey);

@@ -1335,6 +1335,49 @@ void test_executing_on_unreadable_memory
expected_pk_fault(pkey);
}

+void test_implicit_mprotect_exec_only_memory(int *ptr, u16 pkey)
+{
+ void *p1;
+ int scratch;
+ int ptr_contents;
+ int ret;
+
+ dprintf1("%s() start\n", __func__);
+
+ p1 = get_pointer_to_instructions();
+ lots_o_noops_around_write(&scratch);
+ ptr_contents = read_ptr(p1);
+ dprintf2("ptr (%p) contents@%d: %x\n", p1, __LINE__, ptr_contents);
+
+ /* Use a *normal* mprotect(), not mprotect_pkey(): */
+ ret = mprotect(p1, PAGE_SIZE, PROT_EXEC);
+ pkey_assert(!ret);
+
+ dprintf2("pkru: %x\n", rdpkru());
+
+ /* Make sure this is an *instruction* fault */
+ madvise(p1, PAGE_SIZE, MADV_DONTNEED);
+ lots_o_noops_around_write(&scratch);
+ do_not_expect_pk_fault();
+ ptr_contents = read_ptr(p1);
+ dprintf2("ptr (%p) contents@%d: %x\n", p1, __LINE__, ptr_contents);
+ expected_pk_fault(UNKNOWN_PKEY);
+
+ /*
+ * Put the memory back to non-PROT_EXEC. Should clear the
+ * exec-only pkey off the VMA and allow it to be readable
+ * again. Go to PROT_NONE first to check for a kernel bug
+ * that did not clear the pkey when doing PROT_NONE.
+ */
+ ret = mprotect(p1, PAGE_SIZE, PROT_NONE);
+ pkey_assert(!ret);
+
+ ret = mprotect(p1, PAGE_SIZE, PROT_READ|PROT_EXEC);
+ pkey_assert(!ret);
+ ptr_contents = read_ptr(p1);
+ do_not_expect_pk_fault();
+}
+
void test_mprotect_pkey_on_unsupported_cpu(int *ptr, u16 pkey)
{
int size = PAGE_SIZE;
_

2018-03-23 18:13:11

by Dave Hansen

[permalink] [raw]
Subject: [PATCH 6/9] x86, pkeys, selftests: fix pkey exhaustion test off-by-one


From: Dave Hansen <[email protected]>

In our "exhaust all pkeys" test, we make sure that there
is the expected number available. Turns out that the
test did not cover the execute-only key, but discussed
it anyway. It did *not* discuss the test-allocated
key.

Now that we have a test for the mprotect(PROT_EXEC) case,
this off-by-one issue showed itself. Correct the off-by-
one and add the explanation for the case we missed.

Signed-off-by: Dave Hansen <[email protected]>
Cc: Ram Pai <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Dave Hansen <[email protected]>
Cc: Michael Ellermen <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Andrew Morton <[email protected]>
Cc: Shuah Khan <[email protected]>
---

b/tools/testing/selftests/x86/protection_keys.c | 13 ++++++++-----
1 file changed, 8 insertions(+), 5 deletions(-)

diff -puN tools/testing/selftests/x86/protection_keys.c~pkeys-selftests-exhaust-off-by-one tools/testing/selftests/x86/protection_keys.c
--- a/tools/testing/selftests/x86/protection_keys.c~pkeys-selftests-exhaust-off-by-one 2018-03-21 15:47:50.910198920 -0700
+++ b/tools/testing/selftests/x86/protection_keys.c 2018-03-21 15:47:50.914198920 -0700
@@ -1155,12 +1155,15 @@ void test_pkey_alloc_exhaust(int *ptr, u
pkey_assert(i < NR_PKEYS*2);

/*
- * There are 16 pkeys supported in hardware. One is taken
- * up for the default (0) and another can be taken up by
- * an execute-only mapping. Ensure that we can allocate
- * at least 14 (16-2).
+ * There are 16 pkeys supported in hardware. Three are
+ * allocated by the time we get here:
+ * 1. The default key (0)
+ * 2. One possibly consumed by an execute-only mapping.
+ * 3. One allocated by the test code and passed in via
+ * 'pkey' to this function.
+ * Ensure that we can allocate at least another 13 (16-3).
*/
- pkey_assert(i >= NR_PKEYS-2);
+ pkey_assert(i >= NR_PKEYS-3);

for (i = 0; i < nr_allocated_pkeys; i++) {
err = sys_pkey_free(allocated_pkeys[i]);
_

2018-03-23 18:13:31

by Dave Hansen

[permalink] [raw]
Subject: [PATCH 7/9] x86, pkeys, selftests: factor out "instruction page"


From: Dave Hansen <[email protected]>

We currently have an execute-only test, but it is for
the explicit mprotect_pkey() interface. We will soon
add a test for the implicit mprotect(PROT_EXEC)
enterface. We need this code in both tests.

Signed-off-by: Dave Hansen <[email protected]>
Cc: Ram Pai <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Dave Hansen <[email protected]>
Cc: Michael Ellermen <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Andrew Morton <[email protected]>
Cc: Shuah Khan <[email protected]>
---

b/tools/testing/selftests/x86/protection_keys.c | 21 +++++++++++++++++----
1 file changed, 17 insertions(+), 4 deletions(-)

diff -puN tools/testing/selftests/x86/protection_keys.c~pkeys-selftests-get_pointer_to_instructions tools/testing/selftests/x86/protection_keys.c
--- a/tools/testing/selftests/x86/protection_keys.c~pkeys-selftests-get_pointer_to_instructions 2018-03-21 15:47:51.447198918 -0700
+++ b/tools/testing/selftests/x86/protection_keys.c 2018-03-21 15:47:51.450198918 -0700
@@ -1277,12 +1277,9 @@ void test_ptrace_of_child(int *ptr, u16
free(plain_ptr_unaligned);
}

-void test_executing_on_unreadable_memory(int *ptr, u16 pkey)
+void *get_pointer_to_instructions(void)
{
void *p1;
- int scratch;
- int ptr_contents;
- int ret;

p1 = ALIGN_PTR_UP(&lots_o_noops_around_write, PAGE_SIZE);
dprintf3("&lots_o_noops: %p\n", &lots_o_noops_around_write);
@@ -1292,7 +1289,23 @@ void test_executing_on_unreadable_memory
/* Point 'p1' at the *second* page of the function: */
p1 += PAGE_SIZE;

+ /*
+ * Try to ensure we fault this in on next touch to ensure
+ * we get an instruction fault as opposed to a data one
+ */
madvise(p1, PAGE_SIZE, MADV_DONTNEED);
+
+ return p1;
+}
+
+void test_executing_on_unreadable_memory(int *ptr, u16 pkey)
+{
+ void *p1;
+ int scratch;
+ int ptr_contents;
+ int ret;
+
+ p1 = get_pointer_to_instructions();
lots_o_noops_around_write(&scratch);
ptr_contents = read_ptr(p1);
dprintf2("ptr (%p) contents@%d: %x\n", p1, __LINE__, ptr_contents);
_

2018-03-23 18:13:59

by Dave Hansen

[permalink] [raw]
Subject: [PATCH 5/9] x86, pkeys, selftests: fix pointer math


From: Dave Hansen <[email protected]>

We dump out the entire area of the siginfo where the
si_pkey_ptr is supposed to be. But, we do some math
on the poitner, which is a u32. We intended to do
byte math, not u32 math on the pointer.

Cast it over to a u8* so it works.

Also, move this block of code to below th si_code
check. It doesn't hurt anything, but the si_pkey
field is gibberish for other signal types.

Signed-off-by: Dave Hansen <[email protected]>
Cc: Ram Pai <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Dave Hansen <[email protected]>
Cc: Michael Ellermen <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Andrew Morton <[email protected]>
Cc: Shuah Khan <[email protected]>
---

b/tools/testing/selftests/x86/protection_keys.c | 14 +++++++-------
1 file changed, 7 insertions(+), 7 deletions(-)

diff -puN tools/testing/selftests/x86/protection_keys.c~pkeys-selftests-fix-pointer-math tools/testing/selftests/x86/protection_keys.c
--- a/tools/testing/selftests/x86/protection_keys.c~pkeys-selftests-fix-pointer-math 2018-03-21 15:47:50.374198921 -0700
+++ b/tools/testing/selftests/x86/protection_keys.c 2018-03-21 15:47:50.377198921 -0700
@@ -289,13 +289,6 @@ void signal_handler(int signum, siginfo_
dump_mem(pkru_ptr - 128, 256);
pkey_assert(*pkru_ptr);

- si_pkey_ptr = (u32 *)(((u8 *)si) + si_pkey_offset);
- dprintf1("si_pkey_ptr: %p\n", si_pkey_ptr);
- dump_mem(si_pkey_ptr - 8, 24);
- siginfo_pkey = *si_pkey_ptr;
- pkey_assert(siginfo_pkey < NR_PKEYS);
- last_si_pkey = siginfo_pkey;
-
if ((si->si_code == SEGV_MAPERR) ||
(si->si_code == SEGV_ACCERR) ||
(si->si_code == SEGV_BNDERR)) {
@@ -303,6 +296,13 @@ void signal_handler(int signum, siginfo_
exit(4);
}

+ si_pkey_ptr = (u32 *)(((u8 *)si) + si_pkey_offset);
+ dprintf1("si_pkey_ptr: %p\n", si_pkey_ptr);
+ dump_mem((u8 *)si_pkey_ptr - 8, 24);
+ siginfo_pkey = *si_pkey_ptr;
+ pkey_assert(siginfo_pkey < NR_PKEYS);
+ last_si_pkey = siginfo_pkey;
+
dprintf1("signal pkru from xsave: %08x\n", *pkru_ptr);
/* need __rdpkru() version so we do not do shadow_pkru checking */
dprintf1("signal pkru from pkru: %08x\n", __rdpkru());
_

2018-03-23 18:14:16

by Dave Hansen

[permalink] [raw]
Subject: [PATCH 3/9] x86, pkeys, selftests: add a test for pkey 0


From: Dave Hansen <[email protected]>

Protection key 0 is the default key for all memory and will
not normally come back from pkey_alloc(). But, you might
still want pass it to mprotect_pkey().

This check ensures that you can use pkey 0.

Signed-off-by: Dave Hansen <[email protected]>
Cc: Ram Pai <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Dave Hansen <[email protected]>
Cc: Michael Ellermen <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Andrew Morton <[email protected]>
Cc: Shuah Khan <[email protected]>
---

b/tools/testing/selftests/x86/protection_keys.c | 30 ++++++++++++++++++++++++
1 file changed, 30 insertions(+)

diff -puN tools/testing/selftests/x86/protection_keys.c~pkeys-update-selftests-with-pkey-0-test tools/testing/selftests/x86/protection_keys.c
--- a/tools/testing/selftests/x86/protection_keys.c~pkeys-update-selftests-with-pkey-0-test 2018-03-21 15:47:49.273198924 -0700
+++ b/tools/testing/selftests/x86/protection_keys.c 2018-03-21 15:47:49.277198924 -0700
@@ -1169,6 +1169,35 @@ void test_pkey_alloc_exhaust(int *ptr, u
}
}

+/*
+ * pkey 0 is special. It is allocated by default, so you do not
+ * have to call pkey_alloc() to use it first. Make sure that it
+ * is usable.
+ */
+void test_mprotect_with_pkey_0(int *ptr, u16 pkey)
+{
+ long size;
+ int prot;
+
+ assert(pkey_last_malloc_record);
+ size = pkey_last_malloc_record->size;
+ /*
+ * This is a bit of a hack. But mprotect() requires
+ * huge-page-aligned sizes when operating on hugetlbfs.
+ * So, make sure that we use something that's a multiple
+ * of a huge page when we can.
+ */
+ if (size >= HPAGE_SIZE)
+ size = HPAGE_SIZE;
+ prot = pkey_last_malloc_record->prot;
+
+ /* Use pkey 0 */
+ mprotect_pkey(ptr, size, prot, 0);
+
+ /* Make sure that we can set it back to the original pkey. */
+ mprotect_pkey(ptr, size, prot, pkey);
+}
+
void test_ptrace_of_child(int *ptr, u16 pkey)
{
__attribute__((__unused__)) int peek_result;
@@ -1306,6 +1335,7 @@ void (*pkey_tests[])(int *ptr, u16 pkey)
test_kernel_gup_of_access_disabled_region,
test_kernel_gup_write_to_write_disabled_region,
test_executing_on_unreadable_memory,
+ test_mprotect_with_pkey_0,
test_ptrace_of_child,
test_pkey_syscalls_on_non_allocated_pkey,
test_pkey_syscalls_bad_args,
_

2018-03-23 18:14:18

by Dave Hansen

[permalink] [raw]
Subject: [PATCH 1/9] x86, pkeys: do not special case protection key 0


From: Dave Hansen <[email protected]>

mm_pkey_is_allocated() treats pkey 0 as unallocated. That is
inconsistent with the manpages, and also inconsistent with
mm->context.pkey_allocation_map. Stop special casing it and only
disallow values that are actually bad (< 0).

The end-user visible effect of this is that you can now use
mprotect_pkey() to set pkey=0.

This is a bit nicer than what Ram proposed because it is simpler
and removes special-casing for pkey 0. On the other hand, it does
allow applciations to pkey_free() pkey-0, but that's just a silly
thing to do, so we are not going to protect against it.

Signed-off-by: Dave Hansen <[email protected]>
Cc: Ram Pai <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Dave Hansen <[email protected]>
Cc: Michael Ellermen <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Andrew Morton <[email protected]>p
Cc: Shuah Khan <[email protected]>
---

b/arch/x86/include/asm/mmu_context.h | 2 +-
b/arch/x86/include/asm/pkeys.h | 6 +++---
2 files changed, 4 insertions(+), 4 deletions(-)

diff -puN arch/x86/include/asm/mmu_context.h~x86-pkey-0-default-allocated arch/x86/include/asm/mmu_context.h
--- a/arch/x86/include/asm/mmu_context.h~x86-pkey-0-default-allocated 2018-03-21 15:47:48.182198927 -0700
+++ b/arch/x86/include/asm/mmu_context.h 2018-03-21 15:47:48.187198927 -0700
@@ -192,7 +192,7 @@ static inline int init_new_context(struc

#ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
if (cpu_feature_enabled(X86_FEATURE_OSPKE)) {
- /* pkey 0 is the default and always allocated */
+ /* pkey 0 is the default and allocated implicitly */
mm->context.pkey_allocation_map = 0x1;
/* -1 means unallocated or invalid */
mm->context.execute_only_pkey = -1;
diff -puN arch/x86/include/asm/pkeys.h~x86-pkey-0-default-allocated arch/x86/include/asm/pkeys.h
--- a/arch/x86/include/asm/pkeys.h~x86-pkey-0-default-allocated 2018-03-21 15:47:48.184198927 -0700
+++ b/arch/x86/include/asm/pkeys.h 2018-03-21 15:47:48.188198927 -0700
@@ -49,10 +49,10 @@ bool mm_pkey_is_allocated(struct mm_stru
{
/*
* "Allocated" pkeys are those that have been returned
- * from pkey_alloc(). pkey 0 is special, and never
- * returned from pkey_alloc().
+ * from pkey_alloc() or pkey 0 which is allocated
+ * implicitly when the mm is created.
*/
- if (pkey <= 0)
+ if (pkey < 0)
return false;
if (pkey >= arch_max_pkey())
return false;
_

2018-03-23 18:14:23

by Dave Hansen

[permalink] [raw]
Subject: [PATCH 4/9] x86, pkeys: override pkey when moving away from PROT_EXEC


From: Dave Hansen <[email protected]>

I got a bug report that the following code (roughly) was
causing a SIGSEGV:

mprotect(ptr, size, PROT_EXEC);
mprotect(ptr, size, PROT_NONE);
mprotect(ptr, size, PROT_READ);
*ptr = 100;

The problem is hit when the mprotect(PROT_EXEC)
is implicitly assigned a protection key to the VMA, and made
that key ACCESS_DENY|WRITE_DENY. The PROT_NONE mprotect()
failed to remove the protection key, and the PROT_NONE->
PROT_READ left the PTE usable, but the pkey still in place
and left the memory inaccessible.

To fix this, we ensure that we always "override" the pkee
at mprotect() if the VMA does not have execute-only
permissions, but the VMA has the execute-only pkey.

We had a check for PROT_READ/WRITE, but it did not work
for PROT_NONE. This entirely removes the PROT_* checks,
which ensures that PROT_NONE now works.

Reported-by: Shakeel Butt <[email protected]>

Signed-off-by: Dave Hansen <[email protected]>
Cc: Ram Pai <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Dave Hansen <[email protected]>
Cc: Michael Ellermen <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Andrew Morton <[email protected]>
Cc: Shuah Khan <[email protected]>
---

b/arch/x86/include/asm/pkeys.h | 12 +++++++++++-
b/arch/x86/mm/pkeys.c | 19 ++++++++++---------
2 files changed, 21 insertions(+), 10 deletions(-)

diff -puN arch/x86/include/asm/pkeys.h~pkeys-abandon-exec-only-pkey-more-aggressively arch/x86/include/asm/pkeys.h
--- a/arch/x86/include/asm/pkeys.h~pkeys-abandon-exec-only-pkey-more-aggressively 2018-03-21 15:47:49.810198922 -0700
+++ b/arch/x86/include/asm/pkeys.h 2018-03-21 15:47:49.816198922 -0700
@@ -2,6 +2,8 @@
#ifndef _ASM_X86_PKEYS_H
#define _ASM_X86_PKEYS_H

+#define ARCH_DEFAULT_PKEY 0
+
#define arch_max_pkey() (boot_cpu_has(X86_FEATURE_OSPKE) ? 16 : 1)

extern int arch_set_user_pkey_access(struct task_struct *tsk, int pkey,
@@ -15,7 +17,7 @@ extern int __execute_only_pkey(struct mm
static inline int execute_only_pkey(struct mm_struct *mm)
{
if (!boot_cpu_has(X86_FEATURE_OSPKE))
- return 0;
+ return ARCH_DEFAULT_PKEY;

return __execute_only_pkey(mm);
}
@@ -56,6 +58,14 @@ bool mm_pkey_is_allocated(struct mm_stru
return false;
if (pkey >= arch_max_pkey())
return false;
+ /*
+ * The exec-only pkey is set in the allocation map, but
+ * is not available to any of the user interfaces like
+ * mprotect_pkey().
+ */
+ if (pkey == mm->context.execute_only_pkey)
+ return false;
+
return mm_pkey_allocation_map(mm) & (1U << pkey);
}

diff -puN arch/x86/mm/pkeys.c~pkeys-abandon-exec-only-pkey-more-aggressively arch/x86/mm/pkeys.c
--- a/arch/x86/mm/pkeys.c~pkeys-abandon-exec-only-pkey-more-aggressively 2018-03-21 15:47:49.812198922 -0700
+++ b/arch/x86/mm/pkeys.c 2018-03-21 15:47:49.816198922 -0700
@@ -94,15 +94,7 @@ int __arch_override_mprotect_pkey(struct
*/
if (pkey != -1)
return pkey;
- /*
- * Look for a protection-key-drive execute-only mapping
- * which is now being given permissions that are not
- * execute-only. Move it back to the default pkey.
- */
- if (vma_is_pkey_exec_only(vma) &&
- (prot & (PROT_READ|PROT_WRITE))) {
- return 0;
- }
+
/*
* The mapping is execute-only. Go try to get the
* execute-only protection key. If we fail to do that,
@@ -113,7 +105,16 @@ int __arch_override_mprotect_pkey(struct
pkey = execute_only_pkey(vma->vm_mm);
if (pkey > 0)
return pkey;
+ } else if (vma_is_pkey_exec_only(vma)) {
+ /*
+ * Protections are *not* PROT_EXEC, but the mapping
+ * is using the exec-only pkey. This mapping was
+ * PROT_EXEC and will no longer be. Move back to
+ * the default pkey.
+ */
+ return ARCH_DEFAULT_PKEY;
}
+
/*
* This is a vanilla, non-pkey mprotect (or we failed to
* setup execute-only), inherit the pkey from the VMA we
_

2018-03-23 18:15:07

by Dave Hansen

[permalink] [raw]
Subject: [PATCH 2/9] x86, pkeys, selftests: save off 'prot' for allocations


From: Dave Hansen <[email protected]>

This makes it possible to to tell what 'prot' a given allocation
is supposed to have. That way, if we want to change just the
pkey, we know what 'prot' to pass to mprotect_pkey().

Also, keep a record of the most recent allocation so the tests
can easily find it.

Signed-off-by: Dave Hansen <[email protected]>
Cc: Ram Pai <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Dave Hansen <[email protected]>
Cc: Michael Ellermen <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Andrew Morton <[email protected]>
Cc: Shuah Khan <[email protected]>
---

b/tools/testing/selftests/x86/protection_keys.c | 14 +++++++++-----
1 file changed, 9 insertions(+), 5 deletions(-)

diff -puN tools/testing/selftests/x86/protection_keys.c~pkeys-update-selftests-store-malloc-record tools/testing/selftests/x86/protection_keys.c
--- a/tools/testing/selftests/x86/protection_keys.c~pkeys-update-selftests-store-malloc-record 2018-03-21 15:47:48.740198925 -0700
+++ b/tools/testing/selftests/x86/protection_keys.c 2018-03-21 15:47:48.743198925 -0700
@@ -674,10 +674,12 @@ int mprotect_pkey(void *ptr, size_t size
struct pkey_malloc_record {
void *ptr;
long size;
+ int prot;
};
struct pkey_malloc_record *pkey_malloc_records;
+struct pkey_malloc_record *pkey_last_malloc_record;
long nr_pkey_malloc_records;
-void record_pkey_malloc(void *ptr, long size)
+void record_pkey_malloc(void *ptr, long size, int prot)
{
long i;
struct pkey_malloc_record *rec = NULL;
@@ -709,6 +711,8 @@ void record_pkey_malloc(void *ptr, long
(int)(rec - pkey_malloc_records), rec, ptr, size);
rec->ptr = ptr;
rec->size = size;
+ rec->prot = prot;
+ pkey_last_malloc_record = rec;
nr_pkey_malloc_records++;
}

@@ -753,7 +757,7 @@ void *malloc_pkey_with_mprotect(long siz
pkey_assert(ptr != (void *)-1);
ret = mprotect_pkey((void *)ptr, PAGE_SIZE, prot, pkey);
pkey_assert(!ret);
- record_pkey_malloc(ptr, size);
+ record_pkey_malloc(ptr, size, prot);
rdpkru();

dprintf1("%s() for pkey %d @ %p\n", __func__, pkey, ptr);
@@ -774,7 +778,7 @@ void *malloc_pkey_anon_huge(long size, i
size = ALIGN_UP(size, HPAGE_SIZE * 2);
ptr = mmap(NULL, size, PROT_NONE, MAP_ANONYMOUS|MAP_PRIVATE, -1, 0);
pkey_assert(ptr != (void *)-1);
- record_pkey_malloc(ptr, size);
+ record_pkey_malloc(ptr, size, prot);
mprotect_pkey(ptr, size, prot, pkey);

dprintf1("unaligned ptr: %p\n", ptr);
@@ -847,7 +851,7 @@ void *malloc_pkey_hugetlb(long size, int
pkey_assert(ptr != (void *)-1);
mprotect_pkey(ptr, size, prot, pkey);

- record_pkey_malloc(ptr, size);
+ record_pkey_malloc(ptr, size, prot);

dprintf1("mmap()'d hugetlbfs for pkey %d @ %p\n", pkey, ptr);
return ptr;
@@ -869,7 +873,7 @@ void *malloc_pkey_mmap_dax(long size, in

mprotect_pkey(ptr, size, prot, pkey);

- record_pkey_malloc(ptr, size);
+ record_pkey_malloc(ptr, size, prot);

dprintf1("mmap()'d for pkey %d @ %p\n", pkey, ptr);
close(fd);
_

2018-03-23 19:16:16

by Shakeel Butt

[permalink] [raw]
Subject: Re: [PATCH 4/9] x86, pkeys: override pkey when moving away from PROT_EXEC

On Fri, Mar 23, 2018 at 11:09 AM, Dave Hansen
<[email protected]> wrote:
>
> From: Dave Hansen <[email protected]>
>
> I got a bug report that the following code (roughly) was
> causing a SIGSEGV:
>
> mprotect(ptr, size, PROT_EXEC);
> mprotect(ptr, size, PROT_NONE);
> mprotect(ptr, size, PROT_READ);
> *ptr = 100;
>
> The problem is hit when the mprotect(PROT_EXEC)
> is implicitly assigned a protection key to the VMA, and made
> that key ACCESS_DENY|WRITE_DENY. The PROT_NONE mprotect()
> failed to remove the protection key, and the PROT_NONE->
> PROT_READ left the PTE usable, but the pkey still in place
> and left the memory inaccessible.
>
> To fix this, we ensure that we always "override" the pkee
> at mprotect() if the VMA does not have execute-only
> permissions, but the VMA has the execute-only pkey.
>
> We had a check for PROT_READ/WRITE, but it did not work
> for PROT_NONE. This entirely removes the PROT_* checks,
> which ensures that PROT_NONE now works.
>
> Reported-by: Shakeel Butt <[email protected]>
>
> Signed-off-by: Dave Hansen <[email protected]>

Should there be a 'Fixes' tag? Also should this patch go to stable?

> Cc: Ram Pai <[email protected]>
> Cc: Thomas Gleixner <[email protected]>
> Cc: Dave Hansen <[email protected]>
> Cc: Michael Ellermen <[email protected]>
> Cc: Ingo Molnar <[email protected]>
> Cc: Andrew Morton <[email protected]>
> Cc: Shuah Khan <[email protected]>
> ---
>
> b/arch/x86/include/asm/pkeys.h | 12 +++++++++++-
> b/arch/x86/mm/pkeys.c | 19 ++++++++++---------
> 2 files changed, 21 insertions(+), 10 deletions(-)
>
> diff -puN arch/x86/include/asm/pkeys.h~pkeys-abandon-exec-only-pkey-more-aggressively arch/x86/include/asm/pkeys.h
> --- a/arch/x86/include/asm/pkeys.h~pkeys-abandon-exec-only-pkey-more-aggressively 2018-03-21 15:47:49.810198922 -0700
> +++ b/arch/x86/include/asm/pkeys.h 2018-03-21 15:47:49.816198922 -0700
> @@ -2,6 +2,8 @@
> #ifndef _ASM_X86_PKEYS_H
> #define _ASM_X86_PKEYS_H
>
> +#define ARCH_DEFAULT_PKEY 0
> +
> #define arch_max_pkey() (boot_cpu_has(X86_FEATURE_OSPKE) ? 16 : 1)
>
> extern int arch_set_user_pkey_access(struct task_struct *tsk, int pkey,
> @@ -15,7 +17,7 @@ extern int __execute_only_pkey(struct mm
> static inline int execute_only_pkey(struct mm_struct *mm)
> {
> if (!boot_cpu_has(X86_FEATURE_OSPKE))
> - return 0;
> + return ARCH_DEFAULT_PKEY;
>
> return __execute_only_pkey(mm);
> }
> @@ -56,6 +58,14 @@ bool mm_pkey_is_allocated(struct mm_stru
> return false;
> if (pkey >= arch_max_pkey())
> return false;
> + /*
> + * The exec-only pkey is set in the allocation map, but
> + * is not available to any of the user interfaces like
> + * mprotect_pkey().
> + */
> + if (pkey == mm->context.execute_only_pkey)
> + return false;
> +
> return mm_pkey_allocation_map(mm) & (1U << pkey);
> }
>
> diff -puN arch/x86/mm/pkeys.c~pkeys-abandon-exec-only-pkey-more-aggressively arch/x86/mm/pkeys.c
> --- a/arch/x86/mm/pkeys.c~pkeys-abandon-exec-only-pkey-more-aggressively 2018-03-21 15:47:49.812198922 -0700
> +++ b/arch/x86/mm/pkeys.c 2018-03-21 15:47:49.816198922 -0700
> @@ -94,15 +94,7 @@ int __arch_override_mprotect_pkey(struct
> */
> if (pkey != -1)
> return pkey;
> - /*
> - * Look for a protection-key-drive execute-only mapping
> - * which is now being given permissions that are not
> - * execute-only. Move it back to the default pkey.
> - */
> - if (vma_is_pkey_exec_only(vma) &&
> - (prot & (PROT_READ|PROT_WRITE))) {
> - return 0;
> - }
> +
> /*
> * The mapping is execute-only. Go try to get the
> * execute-only protection key. If we fail to do that,
> @@ -113,7 +105,16 @@ int __arch_override_mprotect_pkey(struct
> pkey = execute_only_pkey(vma->vm_mm);
> if (pkey > 0)
> return pkey;
> + } else if (vma_is_pkey_exec_only(vma)) {
> + /*
> + * Protections are *not* PROT_EXEC, but the mapping
> + * is using the exec-only pkey. This mapping was
> + * PROT_EXEC and will no longer be. Move back to
> + * the default pkey.
> + */
> + return ARCH_DEFAULT_PKEY;
> }
> +
> /*
> * This is a vanilla, non-pkey mprotect (or we failed to
> * setup execute-only), inherit the pkey from the VMA we
> _

2018-03-23 19:24:17

by Dave Hansen

[permalink] [raw]
Subject: Re: [PATCH 4/9] x86, pkeys: override pkey when moving away from PROT_EXEC

On 03/23/2018 12:15 PM, Shakeel Butt wrote:
>> We had a check for PROT_READ/WRITE, but it did not work
>> for PROT_NONE. This entirely removes the PROT_* checks,
>> which ensures that PROT_NONE now works.
>>
>> Reported-by: Shakeel Butt <[email protected]>
>> Signed-off-by: Dave Hansen <[email protected]>
> Should there be a 'Fixes' tag? Also should this patch go to stable?

There could be, but I'm to lazy to dig up the original commit. Does it
matter?

And, yes, I think it probably makes sense for -stable. I'll add that if
I resend this series.


2018-03-23 19:28:37

by Shakeel Butt

[permalink] [raw]
Subject: Re: [PATCH 4/9] x86, pkeys: override pkey when moving away from PROT_EXEC

On Fri, Mar 23, 2018 at 12:23 PM, Dave Hansen <[email protected]> wrote:
> On 03/23/2018 12:15 PM, Shakeel Butt wrote:
>>> We had a check for PROT_READ/WRITE, but it did not work
>>> for PROT_NONE. This entirely removes the PROT_* checks,
>>> which ensures that PROT_NONE now works.
>>>
>>> Reported-by: Shakeel Butt <[email protected]>
>>> Signed-off-by: Dave Hansen <[email protected]>
>> Should there be a 'Fixes' tag? Also should this patch go to stable?
>
> There could be, but I'm to lazy to dig up the original commit. Does it
> matter?
>

I think for stable 'Fixes' is usually preferable.

2018-03-23 19:31:24

by Dave Hansen

[permalink] [raw]
Subject: Re: [PATCH 4/9] x86, pkeys: override pkey when moving away from PROT_EXEC

On 03/23/2018 12:27 PM, Shakeel Butt wrote:
> On Fri, Mar 23, 2018 at 12:23 PM, Dave Hansen <[email protected]> wrote:
>> On 03/23/2018 12:15 PM, Shakeel Butt wrote:
>>>> We had a check for PROT_READ/WRITE, but it did not work
>>>> for PROT_NONE. This entirely removes the PROT_* checks,
>>>> which ensures that PROT_NONE now works.
>>>>
>>>> Reported-by: Shakeel Butt <[email protected]>
>>>> Signed-off-by: Dave Hansen <[email protected]>
>>> Should there be a 'Fixes' tag? Also should this patch go to stable?
>> There could be, but I'm to lazy to dig up the original commit. Does it
>> matter?
>>
> I think for stable 'Fixes' is usually preferable.

This one is a no-brainer. If pkeys.c is there, it's necesary.

2018-03-23 19:39:52

by Thomas Gleixner

[permalink] [raw]
Subject: Re: [PATCH 4/9] x86, pkeys: override pkey when moving away from PROT_EXEC

On Fri, 23 Mar 2018, Dave Hansen wrote:

> On 03/23/2018 12:15 PM, Shakeel Butt wrote:
> >> We had a check for PROT_READ/WRITE, but it did not work
> >> for PROT_NONE. This entirely removes the PROT_* checks,
> >> which ensures that PROT_NONE now works.
> >>
> >> Reported-by: Shakeel Butt <[email protected]>
> >> Signed-off-by: Dave Hansen <[email protected]>
> > Should there be a 'Fixes' tag? Also should this patch go to stable?
>
> There could be, but I'm to lazy to dig up the original commit. Does it
> matter?
>
> And, yes, I think it probably makes sense for -stable. I'll add that if
> I resend this series.

The fixes tag makes sense in general even if the patch is not tagged for
stable. It gives you immediate context and I use it a lot to look why this
went unnoticed or what the context of that change was.

Thanks,

tglx

2018-03-23 19:47:11

by Thomas Gleixner

[permalink] [raw]
Subject: Re: [PATCH 4/9] x86, pkeys: override pkey when moving away from PROT_EXEC

On Fri, 23 Mar 2018, Thomas Gleixner wrote:

> On Fri, 23 Mar 2018, Dave Hansen wrote:
>
> > On 03/23/2018 12:15 PM, Shakeel Butt wrote:
> > >> We had a check for PROT_READ/WRITE, but it did not work
> > >> for PROT_NONE. This entirely removes the PROT_* checks,
> > >> which ensures that PROT_NONE now works.
> > >>
> > >> Reported-by: Shakeel Butt <[email protected]>
> > >> Signed-off-by: Dave Hansen <[email protected]>
> > > Should there be a 'Fixes' tag? Also should this patch go to stable?
> >
> > There could be, but I'm to lazy to dig up the original commit. Does it
> > matter?
> >
> > And, yes, I think it probably makes sense for -stable. I'll add that if
> > I resend this series.
>
> The fixes tag makes sense in general even if the patch is not tagged for
> stable. It gives you immediate context and I use it a lot to look why this
> went unnoticed or what the context of that change was.

That said, I'm even lazier than you and prefer you to dig up the original
commit :)

Thanks,

tglx

2018-03-23 19:50:24

by Dave Hansen

[permalink] [raw]
Subject: Re: [PATCH 4/9] x86, pkeys: override pkey when moving away from PROT_EXEC

On 03/23/2018 12:45 PM, Thomas Gleixner wrote:
>> The fixes tag makes sense in general even if the patch is not tagged for
>> stable. It gives you immediate context and I use it a lot to look why this
>> went unnoticed or what the context of that change was.
> That said, I'm even lazier than you and prefer you to dig up the original
> commit :)

I'll have these tags in the next repost.

2018-03-26 17:36:56

by Ram Pai

[permalink] [raw]
Subject: Re: [PATCH 1/9] x86, pkeys: do not special case protection key 0

On Fri, Mar 23, 2018 at 11:09:05AM -0700, Dave Hansen wrote:
>
> From: Dave Hansen <[email protected]>
>
> mm_pkey_is_allocated() treats pkey 0 as unallocated. That is
> inconsistent with the manpages, and also inconsistent with
> mm->context.pkey_allocation_map. Stop special casing it and only
> disallow values that are actually bad (< 0).
>
> The end-user visible effect of this is that you can now use
> mprotect_pkey() to set pkey=0.
>
> This is a bit nicer than what Ram proposed because it is simpler
> and removes special-casing for pkey 0. On the other hand, it does
> allow applciations to pkey_free() pkey-0, but that's just a silly
> thing to do, so we are not going to protect against it.
>
> Signed-off-by: Dave Hansen <[email protected]>
> Cc: Ram Pai <[email protected]>
> Cc: Thomas Gleixner <[email protected]>
> Cc: Dave Hansen <[email protected]>
> Cc: Michael Ellermen <[email protected]>
> Cc: Ingo Molnar <[email protected]>
> Cc: Andrew Morton <[email protected]>p
> Cc: Shuah Khan <[email protected]>
> ---
>
> b/arch/x86/include/asm/mmu_context.h | 2 +-
> b/arch/x86/include/asm/pkeys.h | 6 +++---
> 2 files changed, 4 insertions(+), 4 deletions(-)
>
> diff -puN arch/x86/include/asm/mmu_context.h~x86-pkey-0-default-allocated arch/x86/include/asm/mmu_context.h
> --- a/arch/x86/include/asm/mmu_context.h~x86-pkey-0-default-allocated 2018-03-21 15:47:48.182198927 -0700
> +++ b/arch/x86/include/asm/mmu_context.h 2018-03-21 15:47:48.187198927 -0700
> @@ -192,7 +192,7 @@ static inline int init_new_context(struc
>
> #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
> if (cpu_feature_enabled(X86_FEATURE_OSPKE)) {
> - /* pkey 0 is the default and always allocated */
> + /* pkey 0 is the default and allocated implicitly */
> mm->context.pkey_allocation_map = 0x1;

In the second patch, you introduce DEFAULT_KEY. Maybe you
should introduce here and express the above code as

mm->context.pkey_allocation_map = (0x1 << DEFAULT_KEY);

Incase your default key changes to something else, you are still good.

> /* -1 means unallocated or invalid */
> mm->context.execute_only_pkey = -1;
> diff -puN arch/x86/include/asm/pkeys.h~x86-pkey-0-default-allocated arch/x86/include/asm/pkeys.h
> --- a/arch/x86/include/asm/pkeys.h~x86-pkey-0-default-allocated 2018-03-21 15:47:48.184198927 -0700
> +++ b/arch/x86/include/asm/pkeys.h 2018-03-21 15:47:48.188198927 -0700
> @@ -49,10 +49,10 @@ bool mm_pkey_is_allocated(struct mm_stru
> {
> /*
> * "Allocated" pkeys are those that have been returned
> - * from pkey_alloc(). pkey 0 is special, and never
> - * returned from pkey_alloc().
> + * from pkey_alloc() or pkey 0 which is allocated
> + * implicitly when the mm is created.
> */
> - if (pkey <= 0)
> + if (pkey < 0)
> return false;
> if (pkey >= arch_max_pkey())
> return false;
> _

--
Ram Pai


2018-03-26 17:40:36

by Dave Hansen

[permalink] [raw]
Subject: Re: [PATCH 1/9] x86, pkeys: do not special case protection key 0

On 03/26/2018 10:35 AM, Ram Pai wrote:
>> #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
>> if (cpu_feature_enabled(X86_FEATURE_OSPKE)) {
>> - /* pkey 0 is the default and always allocated */
>> + /* pkey 0 is the default and allocated implicitly */
>> mm->context.pkey_allocation_map = 0x1;
> In the second patch, you introduce DEFAULT_KEY. Maybe you
> should introduce here and express the above code as
>
> mm->context.pkey_allocation_map = (0x1 << DEFAULT_KEY);
>
> Incase your default key changes to something else, you are still good.

That's a good cleanup, but I'd rather limit _this_ set to bug fixes.


2018-03-27 02:28:59

by Ram Pai

[permalink] [raw]
Subject: Re: [PATCH 1/9] x86, pkeys: do not special case protection key 0

On Fri, Mar 23, 2018 at 11:09:05AM -0700, Dave Hansen wrote:
>
> From: Dave Hansen <[email protected]>
>
> mm_pkey_is_allocated() treats pkey 0 as unallocated. That is
> inconsistent with the manpages, and also inconsistent with
> mm->context.pkey_allocation_map. Stop special casing it and only
> disallow values that are actually bad (< 0).
>
> The end-user visible effect of this is that you can now use
> mprotect_pkey() to set pkey=0.
>
> This is a bit nicer than what Ram proposed because it is simpler
> and removes special-casing for pkey 0. On the other hand, it does
> allow applciations to pkey_free() pkey-0, but that's just a silly
> thing to do, so we are not going to protect against it.

The more I think about this, the more I feel we are opening up a can
of worms. I am ok with a bad application, shooting itself in its feet.
But I am worried about all the bug reports and support requests we
will encounter when applications inadvertently shoot themselves
and blame it on the kernel.

a warning in dmesg logs indicating a free-of-pkey-0 can help deflect
the blame from the kernel.

RP


2018-03-27 04:12:56

by Dave Hansen

[permalink] [raw]
Subject: Re: [PATCH 1/9] x86, pkeys: do not special case protection key 0

On 03/26/2018 07:27 PM, Ram Pai wrote:
>> This is a bit nicer than what Ram proposed because it is simpler
>> and removes special-casing for pkey 0. On the other hand, it does
>> allow applciations to pkey_free() pkey-0, but that's just a silly
>> thing to do, so we are not going to protect against it.
> The more I think about this, the more I feel we are opening up a can
> of worms. I am ok with a bad application, shooting itself in its feet.
> But I am worried about all the bug reports and support requests we
> will encounter when applications inadvertently shoot themselves
> and blame it on the kernel.
>
> a warning in dmesg logs indicating a free-of-pkey-0 can help deflect
> the blame from the kernel.

I think it's OK to leave it. A legit, very careful app could decide not
to use pkey 0. It might even be fun to write that in the selftests for
sheer entertainment value.

Although, it _could_ be a bit more debuggable than it is now. A
tracepoint that dumps out the pkey that got faulted on along with the
PKRU value at fault time might be nice to have. That's mildly difficult
to do from outside the app.