Skip to contentSkip to conversation

hugetlb: clean up potential spectre issue warnings

PATCH Applied Mike Kravetz 5 participants 12 messages Feb 18, 2022 – Feb 24, 2022

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 hugetlb: clean up potential spectre issue warnings

Recently introduced code allows numa nodes to be specified on the kernel command line for hugetlb allocations or CMA reservations. The node values are user specified and used as indicies into arrays. This generated the following smatch warnings:

mm/hugetlb.c:4170 hugepages_setup() warn: potential spectre issue 'default_hugepages_in_node' [w] mm/hugetlb.c:4172 hugepages_setup() warn: potential spectre issue 'parsed_hstate->max_huge_pages_node' [w] mm/hugetlb.c:6898 cmdline_parse_hugetlb_cma() warn: potential spectre issue 'hugetlb_cma_size_in_node' [w] (local cap)

Clean up by using array_index_nospec to sanitize array indicies.

The routine cmdline_parse_hugetlb_cma has the same overflow/truncation issue addressed in [1]. That is also fixed with this change.

[1] https://lore.kernel.org/linux-mm/[email protected]/

---
 mm/hugetlb.c | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)
mm/hugetlb.c +4 -3
index 1f0cca036f7f..55abf4e31603 100644
@@ -31,6 +31,7 @@
 #include <linux/llist.h>
 #include <linux/cma.h>
 #include <linux/migrate.h>
+#include <linux/nospec.h>
 
 #include <asm/page.h>
 #include <asm/pgalloc.h>
@@ -4161,7 +4162,7 @@ static int __init hugepages_setup(char *s)
 			}
 			if (tmp >= nr_online_nodes)
 				goto invalid;
-			node = tmp;
+			node = array_index_nospec(tmp, nr_online_nodes);
 			p += count + 1;
 			/* Parse hugepages */
 			if (sscanf(p, "%lu%n", &tmp, &count) != 1)
@@ -6889,9 +6890,9 @@ static int __init cmdline_parse_hugetlb_cma(char *p)
 			break;
 
 		if (s[count] == ':') {
-			nid = tmp;
-			if (nid < 0 || nid >= MAX_NUMNODES)
+			if (tmp >= MAX_NUMNODES)
 				break;
+			nid = array_index_nospec(tmp, MAX_NUMNODES);
 
 			s += count + 1;
 			tmp = memparse(s, &s);
Signature · 1 line
2.34.1
#1 lore raw

On Fri 18-02-22 13:29:46, Mike Kravetz wrote:
[...]

Quoted 11 lines
mm/hugetlb.c +1 -1
@@ -4161,7 +4162,7 @@ static int __init hugepages_setup(char *s)
 			}
 			if (tmp >= nr_online_nodes)
 				goto invalid;
-			node = tmp;
+			node = array_index_nospec(tmp, nr_online_nodes);
 			p += count + 1;
 			/* Parse hugepages */
 			if (sscanf(p, "%lu%n", &tmp, &count) != 1)
@@ -6889,9 +6890,9 @@ static int __init cmdline_parse_hugetlb_cma(char *p)
 			break;

mm/hugetlb.c static int __init cmdline_parse_hugetlb_cma(char *p) · line 6898

Quoted 10 lines
mm/hugetlb.c +2 -2
 
 		if (s[count] == ':') {
-			nid = tmp;
-			if (nid < 0 || nid >= MAX_NUMNODES)
+			if (tmp >= MAX_NUMNODES)
 				break;
+			nid = array_index_nospec(tmp, MAX_NUMNODES);
 
 			s += count + 1;
 			tmp = memparse(s, &s);

This is an early boot code, how is this supposed to be used as a side channel?

Signature · 2 lines
Michal Hocko
SUSE Labs
#2 lore raw
Quoted 21 lines · Michal Hocko

On Fri 18-02-22 13:29:46, Mike Kravetz wrote:
[...]

Quoted 19 lines
mm/hugetlb.c +3 -3
@@ -4161,7 +4162,7 @@ static int __init hugepages_setup(char *s)
 			}
 			if (tmp >= nr_online_nodes)
 				goto invalid;
-			node = tmp;
+			node = array_index_nospec(tmp, nr_online_nodes);
 			p += count + 1;
 			/* Parse hugepages */
 			if (sscanf(p, "%lu%n", &tmp, &count) != 1)
@@ -6889,9 +6890,9 @@ static int __init cmdline_parse_hugetlb_cma(char *p)
 			break;
 
 		if (s[count] == ':') {
-			nid = tmp;
-			if (nid < 0 || nid >= MAX_NUMNODES)
+			if (tmp >= MAX_NUMNODES)
 				break;
+			nid = array_index_nospec(tmp, MAX_NUMNODES);
 
Quoted 5 lines

mm/hugetlb.c static int __init cmdline_parse_hugetlb_cma(char *p) · line 6898

Quoted 2 lines
mm/hugetlb.c +0 -0
 			s += count + 1;
 			tmp = memparse(s, &s);

This is an early boot code, how is this supposed to be used as a side channel?

I do not have an evil hacker mind, but I can not think of a way this one time use of a user specified index could be an issue. It does add noise to the BUILD REGRESSION emails sent to Andrew.

Signature · 1 line
Mike Kravetz
#3 lore raw
Quoted 28 lines · Mike Kravetz
Quoted 21 lines · Michal Hocko

On Fri 18-02-22 13:29:46, Mike Kravetz wrote:
[...]

Quoted 19 lines
mm/hugetlb.c +3 -3
@@ -4161,7 +4162,7 @@ static int __init hugepages_setup(char *s)
 			}
 			if (tmp >= nr_online_nodes)
 				goto invalid;
-			node = tmp;
+			node = array_index_nospec(tmp, nr_online_nodes);
 			p += count + 1;
 			/* Parse hugepages */
 			if (sscanf(p, "%lu%n", &tmp, &count) != 1)
@@ -6889,9 +6890,9 @@ static int __init cmdline_parse_hugetlb_cma(char *p)
 			break;
 
 		if (s[count] == ':') {
-			nid = tmp;
-			if (nid < 0 || nid >= MAX_NUMNODES)
+			if (tmp >= MAX_NUMNODES)
 				break;
+			nid = array_index_nospec(tmp, MAX_NUMNODES);
 
Quoted 5 lines

mm/hugetlb.c static int __init cmdline_parse_hugetlb_cma(char *p) · line 6898

Quoted 2 lines
mm/hugetlb.c +0 -0
 			s += count + 1;
 			tmp = memparse(s, &s);

This is an early boot code, how is this supposed to be used as a side channel?

Quoted 3 lines

I do not have an evil hacker mind, but I can not think of a way this one time use of a user specified index could be an issue. It does add noise to the BUILD REGRESSION emails sent to Andrew.

Maybe Smack can be taught to ignore __init and other early boot functions.

I do not have any strong objections to using array_index_nospec because it won't do any harm. Except that it makes a security measure a normal comodity so any future changes to array_index_nospec and its users will have to consult additional callers. Whether that is something we should deeply care about, I don't know.

At minimum make sure to be explicit that this can hardly be a Spectre gadget as it is a _one_ time early boot call. If there is a scenario where this could be really abused then it should be mentioned explicitly.

Signature · 2 lines
Michal Hocko
SUSE Labs
#4 lore raw
Quoted 41 lines

-----Original Message-----
From: Michal Hocko <[email protected]>
Sent: Monday, February 21, 2022 11:48 PM
To: Mike Kravetz <[email protected]>
Cc: [email protected]; [email protected]; Baolin Wang <[email protected]>; Zhenguo Yao <[email protected]>; Liu Yuntao <[email protected]>; Dan Carpenter <[email protected]>; Andrew Morton <akpm@linux- foundation.org>
Subject: Re: [PATCH v2] hugetlb: clean up potential spectre issue warnings

Quoted 13 lines · Mike Kravetz
Quoted 12 lines · Michal Hocko

On Fri 18-02-22 13:29:46, Mike Kravetz wrote:
[...]

mm/hugetlb.c static int __init

Quoted 10 lines
mm/hugetlb.c +1 -1
@@ -4161,7 +4162,7 @@ static int __init hugepages_setup(char *s)
 			}
 			if (tmp >= nr_online_nodes)
 				goto invalid;
-			node = tmp;
+			node = array_index_nospec(tmp, nr_online_nodes);
 			p += count + 1;
 			/* Parse hugepages */
 			if (sscanf(p, "%lu%n", &tmp, &count) != 1)
@@ -6889,9 +6890,9 @@ static int __init

cmdline_parse_hugetlb_cma(char *p)

Quoted 15 lines
Quoted 14 lines

mm/hugetlb.c static int __init cmdline_parse_hugetlb_cma(char *p) · line 6898

Quoted 11 lines
mm/hugetlb.c +2 -2
 			break;
 
 		if (s[count] == ':') {
-			nid = tmp;
-			if (nid < 0 || nid >= MAX_NUMNODES)
+			if (tmp >= MAX_NUMNODES)
 				break;
+			nid = array_index_nospec(tmp, MAX_NUMNODES);
 
 			s += count + 1;
 			tmp = memparse(s, &s);

This is an early boot code, how is this supposed to be used as a side channel?

Quoted 6 lines
Quoted 3 lines

I do not have an evil hacker mind, but I can not think of a way this one time use of a user specified index could be an issue. It does add noise to the BUILD REGRESSION emails sent to Andrew.

Maybe Smack can be taught to ignore __init and other early boot functions.

Why is Smack getting called out? The relationship is not obvious.

Quoted 14 lines

I do not have any strong objections to using array_index_nospec because it won't do any harm. Except that it makes a security measure a normal comodity so any future changes to array_index_nospec and its users will have to consult additional callers. Whether that is something we should deeply care about, I don't know.

At minimum make sure to be explicit that this can hardly be a Spectre gadget as it is a _one_ time early boot call. If there is a scenario where this could be really abused then it should be mentioned explicitly.
--
Michal Hocko
SUSE Labs

#5 lore raw
Quoted 41 lines · Casey
Quoted 41 lines

-----Original Message-----
From: Michal Hocko <[email protected]>
Sent: Monday, February 21, 2022 11:48 PM
To: Mike Kravetz <[email protected]>
Cc: [email protected]; [email protected]; Baolin Wang <[email protected]>; Zhenguo Yao <[email protected]>; Liu Yuntao <[email protected]>; Dan Carpenter <[email protected]>; Andrew Morton <akpm@linux- foundation.org>
Subject: Re: [PATCH v2] hugetlb: clean up potential spectre issue warnings

Quoted 13 lines · Mike Kravetz
Quoted 12 lines · Michal Hocko

On Fri 18-02-22 13:29:46, Mike Kravetz wrote:
[...]

mm/hugetlb.c static int __init

Quoted 10 lines
mm/hugetlb.c +1 -1
@@ -4161,7 +4162,7 @@ static int __init hugepages_setup(char *s)
 			}
 			if (tmp >= nr_online_nodes)
 				goto invalid;
-			node = tmp;
+			node = array_index_nospec(tmp, nr_online_nodes);
 			p += count + 1;
 			/* Parse hugepages */
 			if (sscanf(p, "%lu%n", &tmp, &count) != 1)
@@ -6889,9 +6890,9 @@ static int __init

cmdline_parse_hugetlb_cma(char *p)

Quoted 15 lines
Quoted 14 lines

mm/hugetlb.c static int __init cmdline_parse_hugetlb_cma(char *p) · line 6898

Quoted 11 lines
mm/hugetlb.c +2 -2
 			break;
 
 		if (s[count] == ':') {
-			nid = tmp;
-			if (nid < 0 || nid >= MAX_NUMNODES)
+			if (tmp >= MAX_NUMNODES)
 				break;
+			nid = array_index_nospec(tmp, MAX_NUMNODES);
 
 			s += count + 1;
 			tmp = memparse(s, &s);

This is an early boot code, how is this supposed to be used as a side channel?

Quoted 9 lines
Quoted 6 lines
Quoted 3 lines

I do not have an evil hacker mind, but I can not think of a way this one time use of a user specified index could be an issue. It does add noise to the BUILD REGRESSION emails sent to Andrew.

Maybe Smack can be taught to ignore __init and other early boot functions.

Why is Smack getting called out? The relationship is not obvious.

He meant Smatch. It's a really common mistake that I did not anticipate in 2002.

I can probably silence the spectre warnings for __init functions. TBH, I don't really understand spectre at all so I mostly ignore those warnings. :/

regards,
dan carpenter

#6 lore raw
Quoted 46 lines · Dan Carpenter
Quoted 45 lines · Casey
Quoted 45 lines

-----Original Message-----
From: Michal Hocko <[email protected]>
Sent: Monday, February 21, 2022 11:48 PM
To: Mike Kravetz <[email protected]>
Cc: [email protected]; [email protected]; Baolin Wang <[email protected]>; Zhenguo Yao <[email protected]>; Liu Yuntao <[email protected]>; Dan Carpenter <[email protected]>; Andrew Morton <akpm@linux- foundation.org>
Subject: Re: [PATCH v2] hugetlb: clean up potential spectre issue warnings

Quoted 13 lines · Mike Kravetz
Quoted 12 lines · Michal Hocko

On Fri 18-02-22 13:29:46, Mike Kravetz wrote:
[...]

mm/hugetlb.c static int __init

Quoted 10 lines
mm/hugetlb.c +1 -1
@@ -4161,7 +4162,7 @@ static int __init hugepages_setup(char *s)
 			}
 			if (tmp >= nr_online_nodes)
 				goto invalid;
-			node = tmp;
+			node = array_index_nospec(tmp, nr_online_nodes);
 			p += count + 1;
 			/* Parse hugepages */
 			if (sscanf(p, "%lu%n", &tmp, &count) != 1)
@@ -6889,9 +6890,9 @@ static int __init

cmdline_parse_hugetlb_cma(char *p)

Quoted 18 lines
Quoted 9 lines
Quoted 9 lines
mm/hugetlb.c +2 -2
 			break;
 
 		if (s[count] == ':') {
-			nid = tmp;
-			if (nid < 0 || nid >= MAX_NUMNODES)
+			if (tmp >= MAX_NUMNODES)
 				break;
+			nid = array_index_nospec(tmp, MAX_NUMNODES);
 
Quoted 5 lines

mm/hugetlb.c static int __init cmdline_parse_hugetlb_cma(char *p) · line 6898

Quoted 2 lines
mm/hugetlb.c +0 -0
 			s += count + 1;
 			tmp = memparse(s, &s);

This is an early boot code, how is this supposed to be used as a side channel?

I do not have an evil hacker mind, but I can not think of a way this one time use of a user specified index could be an issue. It does add noise to the BUILD REGRESSION emails sent to Andrew.

Quoted 8 lines
Quoted 5 lines
Quoted 2 lines

Maybe Smack can be taught to ignore __init and other early boot functions.

Why is Smack getting called out? The relationship is not obvious.

He meant Smatch. It's a really common mistake that I did not anticipate in 2002.

Right. My bad.

Quoted 3 lines

I can probably silence the spectre warnings for __init functions. TBH, I don't really understand spectre at all so I mostly ignore those warnings. :/

AFAIU the spectre gadget would need to be called repeatedly to be usable for a side channel attack. I might be really missing some scenario but to me it seems that __init functions cannot really be used for that.

Signature · 2 lines
Michal Hocko
SUSE Labs
#7 lore raw
In reply to Michal Hocko
Quoted 42 lines · Michal Hocko
Quoted 28 lines · Mike Kravetz
Quoted 21 lines · Michal Hocko

On Fri 18-02-22 13:29:46, Mike Kravetz wrote:
[...]

Quoted 19 lines
mm/hugetlb.c +3 -3
@@ -4161,7 +4162,7 @@ static int __init hugepages_setup(char *s)
 			}
 			if (tmp >= nr_online_nodes)
 				goto invalid;
-			node = tmp;
+			node = array_index_nospec(tmp, nr_online_nodes);
 			p += count + 1;
 			/* Parse hugepages */
 			if (sscanf(p, "%lu%n", &tmp, &count) != 1)
@@ -6889,9 +6890,9 @@ static int __init cmdline_parse_hugetlb_cma(char *p)
 			break;
 
 		if (s[count] == ':') {
-			nid = tmp;
-			if (nid < 0 || nid >= MAX_NUMNODES)
+			if (tmp >= MAX_NUMNODES)
 				break;
+			nid = array_index_nospec(tmp, MAX_NUMNODES);
 
Quoted 5 lines

mm/hugetlb.c static int __init cmdline_parse_hugetlb_cma(char *p) · line 6898

Quoted 2 lines
mm/hugetlb.c +0 -0
 			s += count + 1;
 			tmp = memparse(s, &s);

This is an early boot code, how is this supposed to be used as a side channel?

Quoted 3 lines

I do not have an evil hacker mind, but I can not think of a way this one time use of a user specified index could be an issue. It does add noise to the BUILD REGRESSION emails sent to Andrew.

Maybe Smack can be taught to ignore __init and other early boot functions.

I do not have any strong objections to using array_index_nospec because it won't do any harm. Except that it makes a security measure a normal comodity so any future changes to array_index_nospec and its users will have to consult additional callers. Whether that is something we should deeply care about, I don't know.

Quoted 4 lines

At minimum make sure to be explicit that this can hardly be a Spectre gadget as it is a _one_ time early boot call. If there is a scenario where this could be really abused then it should be mentioned explicitly.

How about adding this note to the commit message?

Note: these routines take a user specified value used as an index ONCE during the boot process. As a result, they can not be used as a general method of exploitation. Code changes are being made to eliminate warnings.

Andrew, would you like me to send a v3?

Signature · 1 line
Mike Kravetz
#8 lore raw
Quoted 48 lines · Mike Kravetz
Quoted 42 lines · Michal Hocko
Quoted 28 lines · Mike Kravetz
Quoted 21 lines · Michal Hocko

On Fri 18-02-22 13:29:46, Mike Kravetz wrote:
[...]

Quoted 19 lines
mm/hugetlb.c +3 -3
@@ -4161,7 +4162,7 @@ static int __init hugepages_setup(char *s)
 			}
 			if (tmp >= nr_online_nodes)
 				goto invalid;
-			node = tmp;
+			node = array_index_nospec(tmp, nr_online_nodes);
 			p += count + 1;
 			/* Parse hugepages */
 			if (sscanf(p, "%lu%n", &tmp, &count) != 1)
@@ -6889,9 +6890,9 @@ static int __init cmdline_parse_hugetlb_cma(char *p)
 			break;
 
 		if (s[count] == ':') {
-			nid = tmp;
-			if (nid < 0 || nid >= MAX_NUMNODES)
+			if (tmp >= MAX_NUMNODES)
 				break;
+			nid = array_index_nospec(tmp, MAX_NUMNODES);
 
Quoted 5 lines

mm/hugetlb.c static int __init cmdline_parse_hugetlb_cma(char *p) · line 6898

Quoted 2 lines
mm/hugetlb.c +0 -0
 			s += count + 1;
 			tmp = memparse(s, &s);

This is an early boot code, how is this supposed to be used as a side channel?

Quoted 3 lines

I do not have an evil hacker mind, but I can not think of a way this one time use of a user specified index could be an issue. It does add noise to the BUILD REGRESSION emails sent to Andrew.

Maybe Smack can be taught to ignore __init and other early boot functions.

I do not have any strong objections to using array_index_nospec because it won't do any harm. Except that it makes a security measure a normal comodity so any future changes to array_index_nospec and its users will have to consult additional callers. Whether that is something we should deeply care about, I don't know.

Quoted 4 lines

At minimum make sure to be explicit that this can hardly be a Spectre gadget as it is a _one_ time early boot call. If there is a scenario where this could be really abused then it should be mentioned explicitly.

Quoted 5 lines

How about adding this note to the commit message?

Note: these routines take a user specified value used as an index ONCE during the boot process. As a result, they can not be used as a general method of exploitation. Code changes are being made to eliminate warnings.

This would help but the question whether the change is worth remains. Does this change have any other advantage than silencing the warning?

Signature · 2 lines
Michal Hocko
SUSE Labs
#9 lore raw
Quoted 11 lines · Michal Hocko
Quoted 6 lines · Mike Kravetz

On 2/21/22 23:47, Michal Hocko wrote:
How about adding this note to the commit message?

Note: these routines take a user specified value used as an index ONCE during the boot process. As a result, they can not be used as a general method of exploitation. Code changes are being made to eliminate warnings.

This would help but the question whether the change is worth remains. Does this change have any other advantage than silencing the warning?

Silencing the warnings was the primary motivation for the change. If Dan has a plan to change smatch so that they are silenced for __init functions, then it would be better to not make the changes to use array_index_nospec.

While making the changes, I shuffled the code a little and did not immediately notice that it also 'fixes' an overflow/truncation issue when assigning an unsigned long to int as addressed in [1]. We should probably make this change whether or not we use array_index_nospec to silence warnings.

[1] https://lore.kernel.org/linux-mm/[email protected]/

Thanks for your comments and suggestions!

Signature · 1 line
Mike Kravetz
#10 lore raw
Quoted 17 lines · Mike Kravetz
Quoted 11 lines · Michal Hocko
Quoted 6 lines · Mike Kravetz

On 2/21/22 23:47, Michal Hocko wrote:
How about adding this note to the commit message?

Note: these routines take a user specified value used as an index ONCE during the boot process. As a result, they can not be used as a general method of exploitation. Code changes are being made to eliminate warnings.

This would help but the question whether the change is worth remains. Does this change have any other advantage than silencing the warning?

Silencing the warnings was the primary motivation for the change. If Dan has a plan to change smatch so that they are silenced for __init functions, then it would be better to not make the changes to use array_index_nospec.

Quoted 6 lines

While making the changes, I shuffled the code a little and did not immediately notice that it also 'fixes' an overflow/truncation issue when assigning an unsigned long to int as addressed in [1]. We should probably make this change whether or not we use array_index_nospec to silence warnings.

[1] https://lore.kernel.org/linux-mm/[email protected]/

Yeah, this makes sense to me.

Signature · 2 lines
Michal Hocko
SUSE Labs
maintainer
#11 lore raw
In reply to Mike Kravetz
Quoted 5 lines · On Tue, 22 Feb 2022 13:53:56 -0800 Mike Kravetz <[email protected]> wrote:
Quoted 3 lines

gadget as it is a _one_ time early boot call. If there is a scenario where this could be really abused then it should be mentioned explicitly.

How about adding this note to the commit message?

I have already done that. I added

"As Michal pointed out, this is unlikely to be exploitable because it is __init code. But the patch suppresses the warnings."