[PATCH v2] slab: Fix nodeid bounds check for non-contiguous node IDs

Subsystems: memory management, the rest

STALE4266d

7 messages, 5 authors, 2014-12-01 · open the first message on its own page

[PATCH v2] slab: Fix nodeid bounds check for non-contiguous node IDs

From: Paul Mackerras <hidden>
Date: 2014-12-01 04:29:09 

The bounds check for nodeid in ____cache_alloc_node gives false
positives on machines where the node IDs are not contiguous, leading
to a panic at boot time.  For example, on a POWER8 machine the node
IDs are typically 0, 1, 16 and 17.  This means that num_online_nodes()
returns 4, so when ____cache_alloc_node is called with nodeid = 16 the
VM_BUG_ON triggers, like this:

kernel BUG at /home/paulus/kernel/kvm/mm/slab.c:3079!
Oops: Exception in kernel mode, sig: 5 [#1]
SMP NR_CPUS=1024 NUMA PowerNV
Modules linked in:
CPU: 0 PID: 0 Comm: swapper Not tainted 3.18.0-rc5-kvm+ #17
task: c0000000013ba230 ti: c000000001494000 task.ti: c000000001494000
NIP: c000000000264f6c LR: c000000000264f5c CTR: 0000000000000000
REGS: c0000000014979a0 TRAP: 0700   Not tainted  (3.18.0-rc5-kvm+)
MSR: 9000000002021032 <SF,HV,VEC,ME,IR,DR,RI>  CR: 28000448  XER: 20000000
CFAR: c00000000047e978 SOFTE: 0
GPR00: c000000000264f5c c000000001497c20 c000000001499d48 0000000000000004
GPR04: 0000000000000100 0000000000000010 0000000000000068 ffffffffffffffff
GPR08: 0000000000000000 0000000000000001 00000000082d0000 c000000000cca5a8
GPR12: 0000000048000448 c00000000fda0000 000001003bd44ff0 0000000010020578
GPR16: 000001003bd44ff8 000001003bd45000 0000000000000001 0000000000000000
GPR20: 0000000000000000 0000000000000000 0000000000000000 0000000000000010
GPR24: c000000ffe000080 c000000000c824ec 0000000000000068 c000000ffe000080
GPR28: 0000000000000010 c000000ffe000080 0000000000000010 0000000000000000
NIP [c000000000264f6c] .____cache_alloc_node+0x6c/0x270
LR [c000000000264f5c] .____cache_alloc_node+0x5c/0x270
Call Trace:
[c000000001497c20] [c000000000264f5c] .____cache_alloc_node+0x5c/0x270 (unreliable)
[c000000001497cf0] [c00000000026552c] .kmem_cache_alloc_node_trace+0xdc/0x360
[c000000001497dc0] [c000000000c824ec] .init_list+0x3c/0x128
[c000000001497e50] [c000000000c827b4] .kmem_cache_init+0x1dc/0x258
[c000000001497ef0] [c000000000c54090] .start_kernel+0x2a0/0x568
[c000000001497f90] [c000000000008c6c] start_here_common+0x20/0xa8
Instruction dump:
7c7d1b78 7c962378 4bda4e91 60000000 3c620004 38800100 386370d8 48219959
60000000 7f83e000 7d301026 5529effe <0b090000> 393c0010 79291f24 7d3d4a14

To fix this, we instead compare the nodeid with MAX_NUMNODES, and
additionally make sure it isn't negative (since nodeid is an int).
The check is there mainly to protect the array dereference in the
get_node() call in the next line, and the array being dereferenced is
of size MAX_NUMNODES.  If the nodeid is in range but invalid (for
example if the node is off-line), the BUG_ON in the next line will
catch that.

Signed-off-by: Paul Mackerras <redacted>
---
v2: include the oops message in the patch description

 mm/slab.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/mm/slab.c b/mm/slab.c
index eb2b2ea..f34e053 100644
--- a/mm/slab.c
+++ b/mm/slab.c
@@ -3076,7 +3076,7 @@ static void *____cache_alloc_node(struct kmem_cache *cachep, gfp_t flags,
 	void *obj;
 	int x;
 
-	VM_BUG_ON(nodeid > num_online_nodes());
+	VM_BUG_ON(nodeid < 0 || nodeid >= MAX_NUMNODES);
 	n = get_node(cachep, nodeid);
 	BUG_ON(!n);
 
-- 
2.1.1

Re: [PATCH v2] slab: Fix nodeid bounds check for non-contiguous node IDs

From: Yasuaki Ishimatsu <hidden>
Date: 2014-12-01 05:00:11 

(2014/12/01 13:28), Paul Mackerras wrote:
The bounds check for nodeid in ____cache_alloc_node gives false
positives on machines where the node IDs are not contiguous, leading
to a panic at boot time.  For example, on a POWER8 machine the node
IDs are typically 0, 1, 16 and 17.  This means that num_online_nodes()
returns 4, so when ____cache_alloc_node is called with nodeid = 16 the
VM_BUG_ON triggers, like this:

kernel BUG at /home/paulus/kernel/kvm/mm/slab.c:3079!
Oops: Exception in kernel mode, sig: 5 [#1]
SMP NR_CPUS=1024 NUMA PowerNV
Modules linked in:
CPU: 0 PID: 0 Comm: swapper Not tainted 3.18.0-rc5-kvm+ #17
task: c0000000013ba230 ti: c000000001494000 task.ti: c000000001494000
NIP: c000000000264f6c LR: c000000000264f5c CTR: 0000000000000000
REGS: c0000000014979a0 TRAP: 0700   Not tainted  (3.18.0-rc5-kvm+)
MSR: 9000000002021032 <SF,HV,VEC,ME,IR,DR,RI>  CR: 28000448  XER: 20000000
CFAR: c00000000047e978 SOFTE: 0
GPR00: c000000000264f5c c000000001497c20 c000000001499d48 0000000000000004
GPR04: 0000000000000100 0000000000000010 0000000000000068 ffffffffffffffff
GPR08: 0000000000000000 0000000000000001 00000000082d0000 c000000000cca5a8
GPR12: 0000000048000448 c00000000fda0000 000001003bd44ff0 0000000010020578
GPR16: 000001003bd44ff8 000001003bd45000 0000000000000001 0000000000000000
GPR20: 0000000000000000 0000000000000000 0000000000000000 0000000000000010
GPR24: c000000ffe000080 c000000000c824ec 0000000000000068 c000000ffe000080
GPR28: 0000000000000010 c000000ffe000080 0000000000000010 0000000000000000
NIP [c000000000264f6c] .____cache_alloc_node+0x6c/0x270
LR [c000000000264f5c] .____cache_alloc_node+0x5c/0x270
Call Trace:
[c000000001497c20] [c000000000264f5c] .____cache_alloc_node+0x5c/0x270 (unreliable)
[c000000001497cf0] [c00000000026552c] .kmem_cache_alloc_node_trace+0xdc/0x360
[c000000001497dc0] [c000000000c824ec] .init_list+0x3c/0x128
[c000000001497e50] [c000000000c827b4] .kmem_cache_init+0x1dc/0x258
[c000000001497ef0] [c000000000c54090] .start_kernel+0x2a0/0x568
[c000000001497f90] [c000000000008c6c] start_here_common+0x20/0xa8
Instruction dump:
7c7d1b78 7c962378 4bda4e91 60000000 3c620004 38800100 386370d8 48219959
60000000 7f83e000 7d301026 5529effe <0b090000> 393c0010 79291f24 7d3d4a14

To fix this, we instead compare the nodeid with MAX_NUMNODES, and
additionally make sure it isn't negative (since nodeid is an int).
The check is there mainly to protect the array dereference in the
get_node() call in the next line, and the array being dereferenced is
of size MAX_NUMNODES.  If the nodeid is in range but invalid (for
example if the node is off-line), the BUG_ON in the next line will
catch that.

Signed-off-by: Paul Mackerras <redacted>
---
Looks good to me.

Reviewed-by: Yasuaki Ishimatsu <redacted>

If you need to backport it into -stable kernel, please read
Documentation/stable_kernel_rules.txt.

Thanks,
Yasuaki Ishimatsu
quoted hunk
v2: include the oops message in the patch description

  mm/slab.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/mm/slab.c b/mm/slab.c
index eb2b2ea..f34e053 100644
--- a/mm/slab.c
+++ b/mm/slab.c
@@ -3076,7 +3076,7 @@ static void *____cache_alloc_node(struct kmem_cache *cachep, gfp_t flags,
  	void *obj;
  	int x;

-	VM_BUG_ON(nodeid > num_online_nodes());
+	VM_BUG_ON(nodeid < 0 || nodeid >= MAX_NUMNODES);
  	n = get_node(cachep, nodeid);
  	BUG_ON(!n);

Re: [PATCH v2] slab: Fix nodeid bounds check for non-contiguous node IDs

From: Michael Ellerman <mpe@ellerman.id.au>
Date: 2014-12-01 05:02:19 

On Mon, 2014-12-01 at 15:28 +1100, Paul Mackerras wrote:
The bounds check for nodeid in ____cache_alloc_node gives false
positives on machines where the node IDs are not contiguous, leading
to a panic at boot time.  For example, on a POWER8 machine the node
IDs are typically 0, 1, 16 and 17.  This means that num_online_nodes()
returns 4, so when ____cache_alloc_node is called with nodeid = 16 the
VM_BUG_ON triggers, like this:
...
To fix this, we instead compare the nodeid with MAX_NUMNODES, and
additionally make sure it isn't negative (since nodeid is an int).
The check is there mainly to protect the array dereference in the
get_node() call in the next line, and the array being dereferenced is
of size MAX_NUMNODES.  If the nodeid is in range but invalid (for
example if the node is off-line), the BUG_ON in the next line will
catch that.
When did this break? How come we only just noticed?

Also needs:

Cc: stable@vger.kernel.org

cheers

Re: [PATCH v2] slab: Fix nodeid bounds check for non-contiguous node IDs

From: Paul Mackerras <hidden>
Date: 2014-12-01 05:25:11 

On Mon, Dec 01, 2014 at 04:02:14PM +1100, Michael Ellerman wrote:
On Mon, 2014-12-01 at 15:28 +1100, Paul Mackerras wrote:
quoted
The bounds check for nodeid in ____cache_alloc_node gives false
positives on machines where the node IDs are not contiguous, leading
to a panic at boot time.  For example, on a POWER8 machine the node
IDs are typically 0, 1, 16 and 17.  This means that num_online_nodes()
returns 4, so when ____cache_alloc_node is called with nodeid = 16 the
VM_BUG_ON triggers, like this:
...
quoted
To fix this, we instead compare the nodeid with MAX_NUMNODES, and
additionally make sure it isn't negative (since nodeid is an int).
The check is there mainly to protect the array dereference in the
get_node() call in the next line, and the array being dereferenced is
of size MAX_NUMNODES.  If the nodeid is in range but invalid (for
example if the node is off-line), the BUG_ON in the next line will
catch that.
When did this break? How come we only just noticed?
Commit 14e50c6a9bc2, which went into 3.10-rc1.

You'll only notice if you have CONFIG_SLAB=y and CONFIG_DEBUG_VM=y
and you're running on a machine with discontiguous node IDs.

Also needs:

Cc: stable@vger.kernel.org
It does.  I remembered that a minute after I sent the patch.

Paul.

Re: [PATCH v2] slab: Fix nodeid bounds check for non-contiguous node IDs

From: Michael Ellerman <mpe@ellerman.id.au>
Date: 2014-12-01 08:52:25 

On Mon, 2014-12-01 at 16:24 +1100, Paul Mackerras wrote:
On Mon, Dec 01, 2014 at 04:02:14PM +1100, Michael Ellerman wrote:
quoted
On Mon, 2014-12-01 at 15:28 +1100, Paul Mackerras wrote:
quoted
The bounds check for nodeid in ____cache_alloc_node gives false
positives on machines where the node IDs are not contiguous, leading
to a panic at boot time.  For example, on a POWER8 machine the node
IDs are typically 0, 1, 16 and 17.  This means that num_online_nodes()
returns 4, so when ____cache_alloc_node is called with nodeid = 16 the
VM_BUG_ON triggers, like this:
...
quoted
To fix this, we instead compare the nodeid with MAX_NUMNODES, and
additionally make sure it isn't negative (since nodeid is an int).
The check is there mainly to protect the array dereference in the
get_node() call in the next line, and the array being dereferenced is
of size MAX_NUMNODES.  If the nodeid is in range but invalid (for
example if the node is off-line), the BUG_ON in the next line will
catch that.
When did this break? How come we only just noticed?
Commit 14e50c6a9bc2, which went into 3.10-rc1.
OK. So a Fixes tag is nice:

Fixes: 14e50c6a9bc2 ("mm: slab: Verify the nodeid passed to ____cache_alloc_node")

You'll only notice if you have CONFIG_SLAB=y and CONFIG_DEBUG_VM=y
and you're running on a machine with discontiguous node IDs.
Right. And we have SLUB=y for all the defconfigs that are likely to hit that.
quoted
Also needs:

Cc: stable@vger.kernel.org
It does.  I remembered that a minute after I sent the patch.
OK. Hopefully one of the slab maintainers will be happy to add it for us when
they merge this?

cheers

Re: [PATCH v2] slab: Fix nodeid bounds check for non-contiguous node IDs

From: Pekka Enberg <hidden>
Date: 2014-12-01 09:22:32 

On 12/1/14 6:28 AM, Paul Mackerras wrote:
quoted hunk
---
v2: include the oops message in the patch description

  mm/slab.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/mm/slab.c b/mm/slab.c
index eb2b2ea..f34e053 100644
--- a/mm/slab.c
+++ b/mm/slab.c
@@ -3076,7 +3076,7 @@ static void *____cache_alloc_node(struct kmem_cache *cachep, gfp_t flags,
  	void *obj;
  	int x;
  
-	VM_BUG_ON(nodeid > num_online_nodes());
+	VM_BUG_ON(nodeid < 0 || nodeid >= MAX_NUMNODES);
  	n = get_node(cachep, nodeid);
  	BUG_ON(!n);
Reviewed-by: Pekka Enberg <penberg@kernel.org>

Re: [PATCH v2] slab: Fix nodeid bounds check for non-contiguous node IDs

From: David Rientjes <rientjes@google.com>
Date: 2014-12-01 21:06:13 

On Mon, 1 Dec 2014, Paul Mackerras wrote:

The bounds check for nodeid in ____cache_alloc_node gives false
positives on machines where the node IDs are not contiguous, leading
to a panic at boot time.  For example, on a POWER8 machine the node
IDs are typically 0, 1, 16 and 17.  This means that num_online_nodes()
returns 4, so when ____cache_alloc_node is called with nodeid = 16 the
VM_BUG_ON triggers, like this:

kernel BUG at /home/paulus/kernel/kvm/mm/slab.c:3079!
Oops: Exception in kernel mode, sig: 5 [#1]
SMP NR_CPUS=1024 NUMA PowerNV
Modules linked in:
CPU: 0 PID: 0 Comm: swapper Not tainted 3.18.0-rc5-kvm+ #17
task: c0000000013ba230 ti: c000000001494000 task.ti: c000000001494000
NIP: c000000000264f6c LR: c000000000264f5c CTR: 0000000000000000
REGS: c0000000014979a0 TRAP: 0700   Not tainted  (3.18.0-rc5-kvm+)
MSR: 9000000002021032 <SF,HV,VEC,ME,IR,DR,RI>  CR: 28000448  XER: 20000000
CFAR: c00000000047e978 SOFTE: 0
GPR00: c000000000264f5c c000000001497c20 c000000001499d48 0000000000000004
GPR04: 0000000000000100 0000000000000010 0000000000000068 ffffffffffffffff
GPR08: 0000000000000000 0000000000000001 00000000082d0000 c000000000cca5a8
GPR12: 0000000048000448 c00000000fda0000 000001003bd44ff0 0000000010020578
GPR16: 000001003bd44ff8 000001003bd45000 0000000000000001 0000000000000000
GPR20: 0000000000000000 0000000000000000 0000000000000000 0000000000000010
GPR24: c000000ffe000080 c000000000c824ec 0000000000000068 c000000ffe000080
GPR28: 0000000000000010 c000000ffe000080 0000000000000010 0000000000000000
NIP [c000000000264f6c] .____cache_alloc_node+0x6c/0x270
LR [c000000000264f5c] .____cache_alloc_node+0x5c/0x270
Call Trace:
[c000000001497c20] [c000000000264f5c] .____cache_alloc_node+0x5c/0x270 (unreliable)
[c000000001497cf0] [c00000000026552c] .kmem_cache_alloc_node_trace+0xdc/0x360
[c000000001497dc0] [c000000000c824ec] .init_list+0x3c/0x128
[c000000001497e50] [c000000000c827b4] .kmem_cache_init+0x1dc/0x258
[c000000001497ef0] [c000000000c54090] .start_kernel+0x2a0/0x568
[c000000001497f90] [c000000000008c6c] start_here_common+0x20/0xa8
Instruction dump:
7c7d1b78 7c962378 4bda4e91 60000000 3c620004 38800100 386370d8 48219959
60000000 7f83e000 7d301026 5529effe <0b090000> 393c0010 79291f24 7d3d4a14

To fix this, we instead compare the nodeid with MAX_NUMNODES, and
additionally make sure it isn't negative (since nodeid is an int).
The check is there mainly to protect the array dereference in the
get_node() call in the next line, and the array being dereferenced is
of size MAX_NUMNODES.  If the nodeid is in range but invalid (for
example if the node is off-line), the BUG_ON in the next line will
catch that.

Signed-off-by: Paul Mackerras <redacted>
Acked-by: David Rientjes <rientjes@google.com>
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help