Skip to contentSkip to conversation

bpftool: Fix undefined behavior caused by shifting into the sign bit

PATCH Applied Kuan-Wei Chiu 3 participants 5 messages Sep 8, 2024 – Sep 10, 2024

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

PATCH 1/1 bpftool: Fix undefined behavior caused by shifting into the sign bit

Replace shifts of '1' with '1U' in bitwise operations within __show_dev_tc_bpf() to prevent undefined behavior caused by shifting into the sign bit of a signed integer. By using '1U', the operations are explicitly performed on unsigned integers, avoiding potential integer overflow or sign-related issues.

---
 tools/bpf/bpftool/net.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)
tools/bpf/bpftool/net.c +4 -4
index 968714b4c3d4..ad2ea6cf2db1 100644
@@ -482,9 +482,9 @@ static void __show_dev_tc_bpf(const struct ip_devname_ifindex *dev,
 		if (prog_flags[i] || json_output) {
 			NET_START_ARRAY("prog_flags", "%s ");
 			for (j = 0; prog_flags[i] && j < 32; j++) {
-				if (!(prog_flags[i] & (1 << j)))
+				if (!(prog_flags[i] & (1U << j)))
 					continue;
-				NET_DUMP_UINT_ONLY(1 << j);
+				NET_DUMP_UINT_ONLY(1U << j);
 			}
 			NET_END_ARRAY("");
 		}
@@ -493,9 +493,9 @@ static void __show_dev_tc_bpf(const struct ip_devname_ifindex *dev,
 			if (link_flags[i] || json_output) {
 				NET_START_ARRAY("link_flags", "%s ");
 				for (j = 0; link_flags[i] && j < 32; j++) {
-					if (!(link_flags[i] & (1 << j)))
+					if (!(link_flags[i] & (1U << j)))
 						continue;
-					NET_DUMP_UINT_ONLY(1 << j);
+					NET_DUMP_UINT_ONLY(1U << j);
 				}
 				NET_END_ARRAY("");
 			}
Signature · 1 line
2.34.1
maintainer Acked-by on v1 1/1
#1 lore raw
Quoted 7 lines · Kuan-Wei Chiu

Replace shifts of '1' with '1U' in bitwise operations within __show_dev_tc_bpf() to prevent undefined behavior caused by shifting into the sign bit of a signed integer. By using '1U', the operations are explicitly performed on unsigned integers, avoiding potential integer overflow or sign-related issues.

Looks good, thank you.

How did you find these?

#2 lore raw
Quoted 7 lines · Quentin Monnet
Quoted 6 lines · Kuan-Wei Chiu

Replace shifts of '1' with '1U' in bitwise operations within __show_dev_tc_bpf() to prevent undefined behavior caused by shifting into the sign bit of a signed integer. By using '1U', the operations are explicitly performed on unsigned integers, avoiding potential integer overflow or sign-related issues.

Quoted 8 lines
Quoted 1 line

Looks good, thank you.

How did you find these?

TL;DR: I discovered this issue through code review.

I am a student developer trying to contribute to the Linux kernel. I was attempting to compile bpftool with ubsan enabled, and while running ./bpftool net list, I encountered the following error message:

net.c:827:2: runtime error: null pointer passed as argument 1, which is declared to never be null

This prompted me to review the code in net.c, and during that process, I unexpectedly came across the bug that this patch addresses.

As for the ubsan complaint mentioned above, it was triggered because qsort is being called as qsort(NULL, 0, ...) when netfilter has no entries to display. In glibc, qsort is marked with __nonnull ((1, 4)). However, I found conflicting information on cppreference.com [1], which states that when count is zero, both ptr and comp can be NULL. This confused me, so I will need to check the C standard to clarify this. If it turns out that qsort(NULL, 0, ...) is invalid, I will submit a separate patch to fix it.

BTW, should this patch include a Fixes tag and a Cc @stable?

[1]: https://en.cppreference.com/w/c/algorithm/qsort

Regards,
Kuan-Wei

maintainer
#3 lore raw

2024-09-09 04:52 UTC+0800 ~ Kuan-Wei Chiu <[email protected]>

Quoted 19 lines · Quentin Monnet
Quoted 7 lines · Kuan-Wei Chiu
Quoted 6 lines

Replace shifts of '1' with '1U' in bitwise operations within __show_dev_tc_bpf() to prevent undefined behavior caused by shifting into the sign bit of a signed integer. By using '1U', the operations are explicitly performed on unsigned integers, avoiding potential integer overflow or sign-related issues.

Quoted 8 lines
Quoted 1 line

Looks good, thank you.

How did you find these?

TL;DR: I discovered this issue through code review.

Quoted 8 lines

I am a student developer trying to contribute to the Linux kernel. I was attempting to compile bpftool with ubsan enabled, and while running ./bpftool net list, I encountered the following error message:

net.c:827:2: runtime error: null pointer passed as argument 1, which is declared to never be null

This prompted me to review the code in net.c, and during that process, I unexpectedly came across the bug that this patch addresses.

Nice

Quoted 9 lines

As for the ubsan complaint mentioned above, it was triggered because qsort is being called as qsort(NULL, 0, ...) when netfilter has no entries to display. In glibc, qsort is marked with __nonnull ((1, 4)). However, I found conflicting information on cppreference.com [1], which states that when count is zero, both ptr and comp can be NULL. This confused me, so I will need to check the C standard to clarify this. If it turns out that qsort(NULL, 0, ...) is invalid, I will submit a separate patch to fix it.

OK, thanks for looking into it!

Quoted 3 lines

BTW, should this patch include a Fixes tag and a Cc @stable?

We could maybe have used a Fixes:, but the patch is merged already so never mind. As for stable, I don't think this is necessary. I don't believe we can hit the undefined behaviour today; and we encourage people to package bpftool from the GitHub mirror anyway, where your patch will land soon.

Thanks,
Quentin

#4 lore raw
In reply to Kuan-Wei Chiu
This patch was applied to bpf/bpf-next.git (master) by Andrii Nakryiko <[email protected]>:

Hello:

This patch was applied to bpf/bpf-next.git (master) by Andrii Nakryiko <[email protected]>:

Quoted 9 lines · On Sun, 8 Sep 2024 22:00:09 +0800 you wrote:

Replace shifts of '1' with '1U' in bitwise operations within __show_dev_tc_bpf() to prevent undefined behavior caused by shifting into the sign bit of a signed integer. By using '1U', the operations are explicitly performed on unsigned integers, avoiding potential integer overflow or sign-related issues.

[...]

Here is the summary with links:
- bpftool: Fix undefined behavior caused by shifting into the sign bit

    https://git.kernel.org/bpf/bpf-next/c/4cdc0e4ce5e8

You are awesome, thank you!

Signature · 2 lines
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html