powerpc: Avoid code patching freed init sections
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
@@ -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
@@ -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
@@ -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
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()andkvm_guest_init()is an init function. This whole chain gets inlined, so when we free the init section (hencekvm_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.cc [email protected] # 4.13+Signed-off-by Michael Neuling <[email protected]>---
For stable I've marked this as v4.13+ since that's when we refactoredcode-patching.cbut 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 -> %pxv3:
Addinit_mem_freeflag to avoid potential race.
Feedback from Christophe Leroy:
- useinit_section_contains()
- change order of init test for performance
- usepr_debug()
- remove blank linev2:
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@@ -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
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,
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()andkvm_guest_init()is an init function. This whole chain gets inlined, so when we free the init section (hencekvm_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.cc [email protected] # 4.13+Signed-off-by Michael Neuling <[email protected]>---
For stable I've marked this as v4.13+ since that's when we refactoredcode-patching.cbut 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 -> %pxv3:
Addinit_mem_freeflag to avoid potential race.
Feedback from Christophe Leroy:
- useinit_section_contains()
- change order of init test for performance
- usepr_debug()
- remove blank linev2:
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@@ -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
-0arch/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
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:
=20Quoted 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()andkvm_guest_init()is an init function. This whole chain gets inlined, so when we free the init section (hencekvm_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.cc [email protected] # 4.13+Signed-off-by Michael Neuling <[email protected]>---
For stable I've marked this as v4.13+ since that's when we refactoredcode-patching.cbut 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 -> %pxv3:
Addinit_mem_freeflag to avoid potential race.
Feedback from Christophe Leroy:
- useinit_section_contains()
- change order of init test for performance
- usepr_debug()
- remove blank linev2:
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.hindex1a951b0046..1fffbba8d6100644 ---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, unsignedshort hex);=20
extern unsigned intrtas_data;
extern unsigned long longmemory_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.cindex850f3b8f4d..6ae2777c22100644 ---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 usingkernel_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
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()andkvm_guest_init()is an init function. This whole chain gets inlined, so when we free the init section (hencekvm_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.cc [email protected] # 4.13+Signed-off-by Michael Neuling <[email protected]>---
For stable I've marked this as v4.13+ since that's when we refactoredcode-patching.cbut 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 -> %pxv3:
Addinit_mem_freeflag to avoid potential race.
Feedback from Christophe Leroy:
- useinit_section_contains()
- change order of init test for performance
- usepr_debug()
- remove blank linev2:
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@@ -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@@ -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@@ -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 ); }
Michael Ellerman
AppliedQuoted 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()andkvm_guest_init()is an init function. This whole chain gets inlined, so when we free the init section (hencekvm_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
cc [email protected] # 4.13+Signed-off-by Michael Neuling <[email protected]>Reviewed-by Nicholas Piggin <[email protected]>Reviewed-by Christophe Leroy <[email protected]>
Applied to powerpc fixes, thanks.
https://git.kernel.org/powerpc/c/51c3c62b58b357e8d35e4cc32f7b4e
cheers
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()andkvm_guest_init()is an init function. This whole chain gets inlined, so when we free the init section (hencekvm_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
cc [email protected] # 4.13+Signed-off-by Michael Neuling <[email protected]>Reviewed-by Nicholas Piggin <[email protected]>Reviewed-by Christophe Leroy <[email protected]>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
@@ -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
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()andkvm_guest_init()is an init function. This whole chain gets inlined, so when we free the init section (hencekvm_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
cc [email protected] # 4.13+Signed-off-by Michael Neuling <[email protected]>Reviewed-by Nicholas Piggin <[email protected]>Reviewed-by Christophe Leroy <[email protected]>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@@ -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