Skip to contentSkip to conversation

bpf, cgroup: Don't lose errors in effective BPF_LSM_CGROUP queries

PATCH Applied Shakeel Butt 2 participants 2 messages Nov 12 – Nov 14 BPF

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 bpf, cgroup: Don't lose errors in effective BPF_LSM_CGROUP queries

An effective BPF_LSM_CGROUP query walks one attach type per LSM hook. __cgroup_bpf_query() sets ret to -ENOSPC before the loop when prog_cnt is too small, but the effective branch then overwrites ret with each type's copy result:

- If the buffer runs out exactly at the end of a type, the last copy returns 0 and the query returns 0 instead of -ENOSPC.
- If one type's copy faults and a later one succeeds, the query returns 0 instead of -EFAULT.

Return -EFAULT right away when a copy faults, as the non-effective branch does, and leave ret alone otherwise. The copy helper then no longer needs to return -ENOSPC. That also removes a false -ENOSPC it gave when a dummy entry followed the last copied program.

Reproduced in a VM with two BPF_LSM_CGROUP programs on different hooks. An effective query with prog_cnt=1 returned 0, and so did one whose first id slot was not writable. With this patch they fail with -ENOSPC and -EFAULT. Queries of other attach types already did.

Fixes b79c9fc9551b ("bpf: implement BPF_PROG_QUERY for BPF_LSM_CGROUP")

---
This is independent of the "bpf, cgroup: Don't wait for cgroup_mutex to read or free cgroup bpf state" series [1], which changes nearby lines. The two apply cleanly in either order.

[1] https://lore.kernel.org/bpf/[email protected]/

 kernel/bpf/cgroup.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
kernel/bpf/cgroup.c +3 -2
index 57a785241d42..2bffa9f659cd 100644
@@ -424,7 +424,7 @@ static int bpf_cgroup_array_copy_to_user(struct bpf_prog_array *array,
 			return -EFAULT;
 		i++;
 	}
-	return item->prog ? -ENOSPC : 0;
+	return 0;
 }
 
 static int bpf_cgroup_array_delete_safe_at(struct bpf_prog_array *array,
@@ -1505,7 +1505,8 @@ static int __cgroup_bpf_query(struct cgroup *cgrp, const union bpf_attr *attr,
 			effective = rcu_dereference_protected(cgrp->bpf.effective[atype],
 							      lockdep_is_held(&cgroup_mutex));
 			cnt = min_t(int, bpf_cgroup_array_length(effective, atype), total_cnt);
-			ret = bpf_cgroup_array_copy_to_user(effective, prog_ids, cnt, atype);
+			if (bpf_cgroup_array_copy_to_user(effective, prog_ids, cnt, atype))
+				return -EFAULT;
 		} else {
 			struct hlist_head *progs;
 			struct bpf_prog_list *pl;

base-commit: 2e14a351662dc5a3b7421d210049c59edd0425d2

Signature · 1 line
2.53.0-Meta
#1 lore raw
This patch was applied to bpf/bpf.git (master) by Alexei Starovoitov <[email protected]>:

Hello:

This patch was applied to bpf/bpf.git (master)
by Alexei Starovoitov <[email protected]>:

Quoted 11 lines · On Thu, 12 Nov 2026 10:38:03 -0800 you wrote:

An effective BPF_LSM_CGROUP query walks one attach type per LSM hook. __cgroup_bpf_query() sets ret to -ENOSPC before the loop when prog_cnt is too small, but the effective branch then overwrites ret with each type's copy result:

- If the buffer runs out exactly at the end of a type, the last copy returns 0 and the query returns 0 instead of -ENOSPC.
- If one type's copy faults and a later one succeeds, the query returns 0 instead of -EFAULT.

[...]

Here is the summary with links:
- [bpf-next] bpf, cgroup: Don't lose errors in effective BPF_LSM_CGROUP queries

    https://git.kernel.org/bpf/bpf/c/e6cd32b8dbea

You are awesome, thank you!

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