Skip to contentSkip to conversation

powerpc: Avoid code patching freed init sections

PATCH Applied Michael Neuling 5 participants 8 messages Sep 14, 2018 – Oct 1, 2018 PowerPC

lore
Share & tools
lore
#0 lore raw

PATCH v4 1/1 powerpc: Avoid code patching freed init sections

This stops us from doing code patching in init sections after they've been freed.

In this chain:
kvm_guest_init() ->

    kvm_use_magic_page() ->
      fault_in_pages_readable() ->
	 __get_user() ->
	   __get_user_nocheck() ->
	     barrier_nospec();

We have a code patching location at barrier_nospec() and kvm_guest_init() is an init function. This whole chain gets inlined, so when we free the init section (hence kvm_guest_init()), this code goes away and hence should no longer be patched.

We seen this as userspace memory corruption when using a memory checker while doing partition migration testing on powervm (this starts the code patching post migration via /sys/kernel/mobility/migration). In theory, it could also happen when using /sys/kernel/debug/powerpc/barrier_nospec.

---
For stable I've marked this as v4.13+ since that's when we refactored code-patching.c but it could go back even further than that. In reality though, I think we can only hit this since the first spectre/meltdown changes.

v4:
Feedback from Christophe Leroy:
- init_mem_free -> init_mem_is_free
- prlog %lx -> %px

v3:
Add init_mem_free flag to avoid potential race.
Feedback from Christophe Leroy:
- use init_section_contains()
- change order of init test for performance
- use pr_debug()
- remove blank line

v2:
Print when we skip an address

---
 arch/powerpc/include/asm/setup.h | 1 +
 arch/powerpc/lib/code-patching.c | 6 ++++++
 arch/powerpc/mm/mem.c            | 2 ++
 3 files changed, 9 insertions(+)
arch/powerpc/include/asm/setup.h +1 -0
index 1a951b0046..1fffbba8d6 100644
@@ -9,6 +9,7 @@ extern void ppc_printk_progress(char *s, unsigned short hex);
 
 extern unsigned int rtas_data;
 extern unsigned long long memory_limit;
+extern bool init_mem_is_free;
 extern unsigned long klimit;
 extern void *zalloc_maybe_bootmem(size_t size, gfp_t mask);
 
arch/powerpc/lib/code-patching.c +6 -0
index 850f3b8f4d..6ae2777c22 100644
@@ -28,6 +28,12 @@ static int __patch_instruction(unsigned int *exec_addr, unsigned int instr,
 {
 	int err;
 
+	/* Make sure we aren't patching a freed init section */
+	if (init_mem_is_free && init_section_contains(exec_addr, 4)) {
+		pr_debug("Skipping init section patching addr: 0x%px\n", exec_addr);
+		return 0;
+	}
+
 	__put_user_size(instr, patch_addr, 4, err);
 	if (err)
 		return err;
arch/powerpc/mm/mem.c +2 -0
index 5c8530d0c6..04ccb274a6 100644
@@ -63,6 +63,7 @@
 #endif
 
 unsigned long long memory_limit;
+bool init_mem_is_free;
 
 #ifdef CONFIG_HIGHMEM
 pte_t *kmap_pte;
@@ -396,6 +397,7 @@ void free_initmem(void)
 {
 	ppc_md.progress = ppc_printk_progress;
 	mark_initmem_nx();
+	init_mem_is_free = true;
 	free_initmem_default(POISON_FREE_INITMEM);
 }
 
Signature · 1 line
2.17.1
reviewer Reviewed-by on v4 1/1
#1 lore raw
Quoted 69 lines · On Fri, 14 Sep 2018 11:14:11 +1000 Michael Neuling <[email protected]> wrote:

This stops us from doing code patching in init sections after they've been freed.

In this chain:
kvm_guest_init() ->

    kvm_use_magic_page() ->
      fault_in_pages_readable() ->
	 __get_user() ->
	   __get_user_nocheck() ->
	     barrier_nospec();

We have a code patching location at barrier_nospec() and kvm_guest_init() is an init function. This whole chain gets inlined, so when we free the init section (hence kvm_guest_init()), this code goes away and hence should no longer be patched.

We seen this as userspace memory corruption when using a memory checker while doing partition migration testing on powervm (this starts the code patching post migration via /sys/kernel/mobility/migration). In theory, it could also happen when using /sys/kernel/debug/powerpc/barrier_nospec.

---
For stable I've marked this as v4.13+ since that's when we refactored code-patching.c but it could go back even further than that. In reality though, I think we can only hit this since the first spectre/meltdown changes.

v4:
Feedback from Christophe Leroy:
- init_mem_free -> init_mem_is_free
- prlog %lx -> %px

v3:
Add init_mem_free flag to avoid potential race.
Feedback from Christophe Leroy:
- use init_section_contains()
- change order of init test for performance
- use pr_debug()
- remove blank line

v2:
Print when we skip an address

---
 arch/powerpc/include/asm/setup.h | 1 +
 arch/powerpc/lib/code-patching.c | 6 ++++++
 arch/powerpc/mm/mem.c            | 2 ++
 3 files changed, 9 insertions(+)
arch/powerpc/include/asm/setup.h +1 -0
index 1a951b0046..1fffbba8d6 100644
@@ -9,6 +9,7 @@ extern void ppc_printk_progress(char *s, unsigned short hex);
 
 extern unsigned int rtas_data;
 extern unsigned long long memory_limit;
+extern bool init_mem_is_free;
 extern unsigned long klimit;
 extern void *zalloc_maybe_bootmem(size_t size, gfp_t mask);
 
arch/powerpc/lib/code-patching.c +0 -0
index 850f3b8f4d..6ae2777c22 100644

arch/powerpc/lib/code-patching.c static int __patch_instruction(unsigned int *exec_addr, unsigned int instr, · line 35

Quoted 8 lines
arch/powerpc/lib/code-patching.c +5 -0
@@ -28,6 +28,12 @@ static int __patch_instruction(unsigned int *exec_addr, unsigned int instr,
 {
 	int err;
 
+	/* Make sure we aren't patching a freed init section */
+	if (init_mem_is_free && init_section_contains(exec_addr, 4)) {
+		pr_debug("Skipping init section patching addr: 0x%px\n", exec_addr);
+		return 0;
+	}

What we should do is a whitelist, make sure it's only patching the sections we want it to.

That's a bigger job when you consider modules and things too though, so this looks good for now. Thanks,

#2 lore raw

Le 14/09/2018 à 06:22, Nicholas Piggin a écrit :

Quoted 84 lines
Quoted 69 lines · On Fri, 14 Sep 2018 11:14:11 +1000 Michael Neuling <[email protected]> wrote:

This stops us from doing code patching in init sections after they've been freed.

In this chain:

   kvm_guest_init() ->
     kvm_use_magic_page() ->
       fault_in_pages_readable() ->
	 __get_user() ->
	   __get_user_nocheck() ->
	     barrier_nospec();

We have a code patching location at barrier_nospec() and kvm_guest_init() is an init function. This whole chain gets inlined, so when we free the init section (hence kvm_guest_init()), this code goes away and hence should no longer be patched.

We seen this as userspace memory corruption when using a memory checker while doing partition migration testing on powervm (this starts the code patching post migration via /sys/kernel/mobility/migration). In theory, it could also happen when using /sys/kernel/debug/powerpc/barrier_nospec.

---
For stable I've marked this as v4.13+ since that's when we refactored code-patching.c but it could go back even further than that. In reality though, I think we can only hit this since the first spectre/meltdown changes.

v4:
Feedback from Christophe Leroy:
- init_mem_free -> init_mem_is_free
- prlog %lx -> %px

v3:
Add init_mem_free flag to avoid potential race.
Feedback from Christophe Leroy:
- use init_section_contains()
- change order of init test for performance
- use pr_debug()
- remove blank line

v2:
Print when we skip an address

---
  arch/powerpc/include/asm/setup.h | 1 +
  arch/powerpc/lib/code-patching.c | 6 ++++++
  arch/powerpc/mm/mem.c            | 2 ++
  3 files changed, 9 insertions(+)
arch/powerpc/include/asm/setup.h +1 -0
index 1a951b0046..1fffbba8d6 100644
@@ -9,6 +9,7 @@ extern void ppc_printk_progress(char *s, unsigned short hex);
 
  extern unsigned int rtas_data;
  extern unsigned long long memory_limit;
+extern bool init_mem_is_free;
  extern unsigned long klimit;
  extern void *zalloc_maybe_bootmem(size_t size, gfp_t mask);
 
arch/powerpc/lib/code-patching.c +0 -0
index 850f3b8f4d..6ae2777c22 100644

arch/powerpc/lib/code-patching.c static int __patch_instruction(unsigned int *exec_addr, unsigned int instr, · line 35

Quoted 8 lines
arch/powerpc/lib/code-patching.c +5 -0
@@ -28,6 +28,12 @@ static int __patch_instruction(unsigned int *exec_addr, unsigned int instr,
  {
  	int err;
 
+	/* Make sure we aren't patching a freed init section */
+	if (init_mem_is_free && init_section_contains(exec_addr, 4)) {
+		pr_debug("Skipping init section patching addr: 0x%px\n", exec_addr);
+		return 0;
+	}

What we should do is a whitelist, make sure it's only patching the sections we want it to.

Quoted 2 lines

That's a bigger job when you consider modules and things too though, so this looks good for now. Thanks,

What about using kernel_text_address() for it then ? It also handles modules, is it more complicated than that ?

Christophe

#3 lore raw
Quoted 92 lines · On Tue, 18 Sep 2018 10:52:09 +0200 Christophe LEROY <[email protected]>…

=20
=20
Le 14/09/2018 =C3=A0 06:22, Nicholas Piggin a =C3=A9crit=C2=A0:

Quoted 81 lines

On Fri, 14 Sep 2018 11:14:11 +1000
Michael Neuling <[email protected]> wrote:
=20

Quoted 78 lines

This stops us from doing code patching in init sections after they've been freed.

In this chain:

   kvm_guest_init() ->
     kvm_use_magic_page() ->
       fault_in_pages_readable() ->
	 __get_user() ->
	   __get_user_nocheck() ->
	     barrier_nospec();

We have a code patching location at barrier_nospec() and kvm_guest_init() is an init function. This whole chain gets inlined, so when we free the init section (hence kvm_guest_init()), this code goes away and hence should no longer be patched.

We seen this as userspace memory corruption when using a memory checker while doing partition migration testing on powervm (this starts the code patching post migration via /sys/kernel/mobility/migration). In theory, it could also happen when using /sys/kernel/debug/powerpc/barrier_nospec.

---
For stable I've marked this as v4.13+ since that's when we refactored code-patching.c but it could go back even further than that. In reality though, I think we can only hit this since the first spectre/meltdown changes.

v4:
Feedback from Christophe Leroy:
- init_mem_free -> init_mem_is_free
- prlog %lx -> %px

v3:
Add init_mem_free flag to avoid potential race.
Feedback from Christophe Leroy:
- use init_section_contains()
- change order of init test for performance
- use pr_debug()
- remove blank line

v2:
Print when we skip an address

---
  arch/powerpc/include/asm/setup.h | 1 +
  arch/powerpc/lib/code-patching.c | 6 ++++++
  arch/powerpc/mm/mem.c            | 2 ++
  3 files changed, 9 insertions(+)
(quoted diff) +0 -0

b/arch/powerpc/include/asm/setup.h index 1a951b0046..1fffbba8d6 100644 --- a/arch/powerpc/include/asm/setup.h
+++ b/arch/powerpc/include/asm/setup.h

(quoted diff) +0 -0
@@ -9,6 +9,7 @@ extern void ppc_printk_progress(char *s, unsigned

short hex);=20
extern unsigned int rtas_data;
extern unsigned long long memory_limit;

arch/powerpc/include/asm/setup.h +1 -0
+extern bool init_mem_is_free;
  extern unsigned long klimit;
  extern void *zalloc_maybe_bootmem(size_t size, gfp_t mask);
 =20
(quoted diff) +0 -0

b/arch/powerpc/lib/code-patching.c index 850f3b8f4d..6ae2777c22 100644 --- a/arch/powerpc/lib/code-patching.c
+++ b/arch/powerpc/lib/code-patching.c

(quoted diff) +3 -0
@@ -28,6 +28,12 @@ static int __patch_instruction(unsigned int
 *exec_addr, unsigned int instr, {
  	int err;
 =20
+	/* Make sure we aren't patching a freed init section */
+	if (init_mem_is_free && init_section_contains(exec_addr,
 4)) {
+		pr_debug("Skipping init section patching addr:
 0x%px\n", exec_addr);
Quoted 8 lines

arch/powerpc/lib/code-patching.c static int __patch_instruction(unsigned int *exec_addr, unsigned int instr, · line 35

Quoted 2 lines
arch/powerpc/lib/code-patching.c +2 -0
+		return 0;
+	}

=20
What we should do is a whitelist, make sure it's only patching the sections we want it to.
=20
That's a bigger job when you consider modules and things too though, so this looks good for now. Thanks,

Quoted 3 lines

=20
What about using kernel_text_address() for it then ? It also handles=20 modules, is it more complicated than that ?

Modules are patched separately so should not need to be excluded here.

There is a different problem with modules: when the mitigation type changes the modules are not re-patched with the new settings.

Thanks

Michal

Reviewed-by on v4 1/1
#4 lore raw
In reply to Michael Neuling

Le 14/09/2018 à 03:14, Michael Neuling a écrit :

Quoted 95 lines

This stops us from doing code patching in init sections after they've been freed.

In this chain:

   kvm_guest_init() ->
     kvm_use_magic_page() ->
       fault_in_pages_readable() ->
	 __get_user() ->
	   __get_user_nocheck() ->
	     barrier_nospec();

We have a code patching location at barrier_nospec() and kvm_guest_init() is an init function. This whole chain gets inlined, so when we free the init section (hence kvm_guest_init()), this code goes away and hence should no longer be patched.

We seen this as userspace memory corruption when using a memory checker while doing partition migration testing on powervm (this starts the code patching post migration via /sys/kernel/mobility/migration). In theory, it could also happen when using /sys/kernel/debug/powerpc/barrier_nospec.

---
For stable I've marked this as v4.13+ since that's when we refactored code-patching.c but it could go back even further than that. In reality though, I think we can only hit this since the first spectre/meltdown changes.

v4:
Feedback from Christophe Leroy:
- init_mem_free -> init_mem_is_free
- prlog %lx -> %px

v3:
Add init_mem_free flag to avoid potential race.
Feedback from Christophe Leroy:
- use init_section_contains()
- change order of init test for performance
- use pr_debug()
- remove blank line

v2:
Print when we skip an address

---
  arch/powerpc/include/asm/setup.h | 1 +
  arch/powerpc/lib/code-patching.c | 6 ++++++
  arch/powerpc/mm/mem.c            | 2 ++
  3 files changed, 9 insertions(+)
arch/powerpc/include/asm/setup.h +1 -0
index 1a951b0046..1fffbba8d6 100644
@@ -9,6 +9,7 @@ extern void ppc_printk_progress(char *s, unsigned short hex);
 
  extern unsigned int rtas_data;
  extern unsigned long long memory_limit;
+extern bool init_mem_is_free;
  extern unsigned long klimit;
  extern void *zalloc_maybe_bootmem(size_t size, gfp_t mask);
 
arch/powerpc/lib/code-patching.c +6 -0
index 850f3b8f4d..6ae2777c22 100644
@@ -28,6 +28,12 @@ static int __patch_instruction(unsigned int *exec_addr, unsigned int instr,
  {
  	int err;
 
+	/* Make sure we aren't patching a freed init section */
+	if (init_mem_is_free && init_section_contains(exec_addr, 4)) {
+		pr_debug("Skipping init section patching addr: 0x%px\n", exec_addr);
+		return 0;
+	}
+
  	__put_user_size(instr, patch_addr, 4, err);
  	if (err)
  		return err;
arch/powerpc/mm/mem.c +1 -0
index 5c8530d0c6..04ccb274a6 100644
@@ -63,6 +63,7 @@
  #endif
 
  unsigned long long memory_limit;
+bool init_mem_is_free;
 
  #ifdef CONFIG_HIGHMEM
  pte_t *kmap_pte;

arch/powerpc/mm/mem.c void free_initmem(void) · line 403

Quoted 7 lines
arch/powerpc/mm/mem.c +1 -0
@@ -396,6 +397,7 @@ void free_initmem(void)
  {
  	ppc_md.progress = ppc_printk_progress;
  	mark_initmem_nx();
+	init_mem_is_free = true;
  	free_initmem_default(POISON_FREE_INITMEM);
  }
 
#5 lore raw
In reply to Michael Neuling
Quoted 22 lines · Michael Neuling

This stops us from doing code patching in init sections after they've been freed.

In this chain:
kvm_guest_init() ->

    kvm_use_magic_page() ->
      fault_in_pages_readable() ->
	 __get_user() ->
	   __get_user_nocheck() ->
	     barrier_nospec();

We have a code patching location at barrier_nospec() and kvm_guest_init() is an init function. This whole chain gets inlined, so when we free the init section (hence kvm_guest_init()), this code goes away and hence should no longer be patched.

We seen this as userspace memory corruption when using a memory checker while doing partition migration testing on powervm (this starts the code patching post migration via /sys/kernel/mobility/migration). In theory, it could also happen when using /sys/kernel/debug/powerpc/barrier_nospec.

Quoted 4 lines

Applied to powerpc fixes, thanks.

https://git.kernel.org/powerpc/c/51c3c62b58b357e8d35e4cc32f7b4e

cheers

#6 lore raw

Le 21/09/2018 à 13:59, Michael Ellerman a écrit :

Quoted 23 lines · Michael Neuling
Quoted 22 lines

This stops us from doing code patching in init sections after they've been freed.

In this chain:

   kvm_guest_init() ->
     kvm_use_magic_page() ->
       fault_in_pages_readable() ->
	 __get_user() ->
	   __get_user_nocheck() ->
	     barrier_nospec();

We have a code patching location at barrier_nospec() and kvm_guest_init() is an init function. This whole chain gets inlined, so when we free the init section (hence kvm_guest_init()), this code goes away and hence should no longer be patched.

We seen this as userspace memory corruption when using a memory checker while doing partition migration testing on powervm (this starts the code patching post migration via /sys/kernel/mobility/migration). In theory, it could also happen when using /sys/kernel/debug/powerpc/barrier_nospec.

Quoted 9 lines
Quoted 4 lines

Applied to powerpc fixes, thanks.

https://git.kernel.org/powerpc/c/51c3c62b58b357e8d35e4cc32f7b4e

This patch breaks booting on my MPC83xx board (book3s/32) very early (before console is active), provoking restart. u-boot reports a checkstop reset at restart.

Reverting this commit fixes the issue.

The following patch fixes the issue as well, but I think it is not the best solution. I still think the test should be in patch_instruction() instead of being in __patch_instruction(), see my comment on v2

Christophe

arch/powerpc/lib/code-patching.c +1 -1
index 6ae2777..6192fda 100644
@@ -29,7 +29,7 @@ static int __patch_instruction(unsigned int *exec_addr, unsigned int instr,
        int err;
 
        /* Make sure we aren't patching a freed init section */
-       if (init_mem_is_free && init_section_contains(exec_addr, 4)) {
+       if (*PTRRELOC(&init_mem_is_free) && init_section_contains(exec_addr, 4)) {
                pr_debug("Skipping init section patching addr: 
0x%px\n", exec_addr);
                return 0;
        }

Christophe

#7 lore raw
Quoted 39 lines · Christophe LEROY
Quoted 23 lines · Michael Ellerman
Quoted 22 lines · Michael Neuling

This stops us from doing code patching in init sections after they've been freed.

In this chain:

   kvm_guest_init() ->
     kvm_use_magic_page() ->
       fault_in_pages_readable() ->
	 __get_user() ->
	   __get_user_nocheck() ->
	     barrier_nospec();

We have a code patching location at barrier_nospec() and kvm_guest_init() is an init function. This whole chain gets inlined, so when we free the init section (hence kvm_guest_init()), this code goes away and hence should no longer be patched.

We seen this as userspace memory corruption when using a memory checker while doing partition migration testing on powervm (this starts the code patching post migration via /sys/kernel/mobility/migration). In theory, it could also happen when using /sys/kernel/debug/powerpc/barrier_nospec.

Quoted 9 lines
Quoted 4 lines

Applied to powerpc fixes, thanks.

https://git.kernel.org/powerpc/c/51c3c62b58b357e8d35e4cc32f7b4e

This patch breaks booting on my MPC83xx board (book3s/32) very early (before console is active), provoking restart.
u-boot reports a checkstop reset at restart.

Quoted 5 lines

Reverting this commit fixes the issue.

The following patch fixes the issue as well, but I think it is not the best solution. I still think the test should be in patch_instruction() instead of being in __patch_instruction(), see my comment on v2

Arrh, sorry.

Can you write this up as a real patch with a signed off by so mpe can take it?

Mikey

Quoted 24 lines

Christophe

arch/powerpc/lib/code-patching.c +1 -1
index 6ae2777..6192fda 100644
@@ -29,7 +29,7 @@ static int __patch_instruction(unsigned int 
 *exec_addr, unsigned int instr,
         int err;
 
         /* Make sure we aren't patching a freed init section */
-       if (init_mem_is_free && init_section_contains(exec_addr, 4)) {
+       if (*PTRRELOC(&init_mem_is_free) && 
 init_section_contains(exec_addr, 4)) {
                 pr_debug("Skipping init section patching addr: 
0x%px\n", exec_addr);
                 return 0;
         }

Christophe