selftests/bpf: Check returned keys in map batch validators
Subsystems (3)
PATCH 1/1 selftests/bpf: Check returned keys in map batch validators
The hash and array batch validators mark visited entries by output position, making the subsequent completeness checks always succeed. Duplicate keys with matching values can therefore hide missing entries.
Check that returned keys fall within the populated range and index the visited array by key. This detects missing entries while preserving unordered results and existing per-CPU value validation.
Controlled inputs confirm that the old validators accept duplicates and out-of-range keys with matching values, while the updated checks reject them. Ordered, reversed and shuffled valid keys still pass, and scalar and per-CPU value corruption is still rejected. The full test_maps suite passes with no skips on the matching kernel.
--- tools/testing/selftests/bpf/map_tests/array_map_batch_ops.c | 5 ++++- tools/testing/selftests/bpf/map_tests/htab_map_batch_ops.c | 4 +++- 2 files changed, 7 insertions(+), 2 deletions(-)
tools/testing/selftests/bpf/map_tests/array_map_batch_ops.c +4 -1
@@ -45,6 +45,9 @@ static void map_batch_verify(int *visited, __u32 max_entries, int *keys,memset (visited,0 , max_entries *sizeof (*visited));for (i =0 ; i < max_entries; i++) { +CHECK (keys[i] <0 || keys[i] >= max_entries,"key checking" , +"error: i %d key %d out of range\n" , i, keys[i]); +if (is_pcpu) { cpu_offset = i * nr_cpus;for (j =0 ; j < nr_cpus; j++) {
@@ -59,7 +62,7 @@ static void map_batch_verify(int *visited, __u32 max_entries, int *keys,"error: i %d key %d value %lld\n" , i, keys[i], values[i]); } - visited[i] =1 ; + visited[keys[i]] =1 ; }for (i =0 ; i < max_entries; i++) { CHECK(visited[i] != 1, "visited checking",
tools/testing/selftests/bpf/map_tests/htab_map_batch_ops.c +3 -1
@@ -50,6 +50,8 @@ static void map_batch_verify(int *visited, __u32 max_entries,memset (visited,0 , max_entries *sizeof (*visited));for (i =0 ; i < max_entries; i++) { +CHECK (keys[i] <1 || keys[i] > max_entries,"key checking" , +"error: i %d key %d out of range\n" , i, keys[i]);if (is_pcpu) {for (j =0 ; j <bpf_num_possible_cpus (); j++) {
@@ -65,7 +67,7 @@ static void map_batch_verify(int *visited, __u32 max_entries, ((int *)values)[i]); } - visited[i] =1 ; + visited[keys[i] -1 ] =1 ; }for (i =0 ; i < max_entries; i++) {
Signature · 1 line
2.55.0
[email protected]
AI reviewThank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [Low] The patch introduces new usages of the deprecated CHECK() macro instead of the modern ASSERT_*()…
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Low] The patch introduces new usages of the deprecated CHECK() macro instead of the modern ASSERT_*() macros, violating BPF subsystem selftest guidelines. --
commit a52968798d334a652d5c678ee6009b5d4bc96738
Author: Tianyi Chen <[email protected]>
selftests/bpf: Check returned keys in map batch validators
This commit updates map batch validators for hash and array maps by indexing the visited array by key and verifying that returned keys fall within the valid range. This prevents duplicate keys with matching values from hiding missing entries.
tools/testing/selftests/bpf/map_tests/array_map_batch_ops.c static void map_batch_verify(int *visited, __u32 max_entries, int *keys, · line 50
Quoted 11 lines
tools/testing/selftests/bpf/map_tests/array_map_batch_ops.c +3
-0@@ -45,6 +45,9 @@ static void map_batch_verify(int *visited, __u32 max_entries, int *keys,memset (visited,0 , max_entries *sizeof (*visited));for (i =0 ; i < max_entries; i++) { +CHECK (keys[i] <0 || keys[i] >= max_entries,"key checking" , +"error: i %d key %d out of range\n" , i, keys[i]); +
[Severity: Low]
Could this use the modern ASSERT_*() macros instead of the deprecated CHECK() macro?
[ ... ]
tools/testing/selftests/bpf/map_tests/htab_map_batch_ops.c static void map_batch_verify(int *visited, __u32 max_entries, · line 54
Quoted 10 lines
tools/testing/selftests/bpf/map_tests/htab_map_batch_ops.c +2
-0@@ -50,6 +50,8 @@ static void map_batch_verify(int *visited, __u32 max_entries,memset (visited,0 , max_entries *sizeof (*visited));for (i =0 ; i < max_entries; i++) { +CHECK (keys[i] <1 || keys[i] > max_entries,"key checking" , +"error: i %d key %d out of range\n" , i, keys[i]);
[Severity: Low]
Could this also use the modern ASSERT_*() macros rather than CHECK()?
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1
Thanks for the review.
These files are built into the legacy test_maps runner and include test_maps.h. Its CHECK() macro reports the failure and exits immediately; that header does not provide ASSERT_*() helpers.
The ASSERT_*() helpers in test_progs.h belong to a different test runner and depend on its failure-reporting state. Replacing these checks would require a separate framework conversion. The range checks must also stop execution before the following visited-array access.
I have therefore retained CHECK() in v2, consistent with the existing map_tests code. The revision also covers the LPM trie validator:
https://lore.kernel.org/bpf/[email protected]/
Thanks,
Tianyi
[email protected]
AI reviewThis isn't a bug, but map_tests/ has three near-identical batch validators, and lpm_trie_map_batch_ops.c still has the same position-indexed visited[i] = 1 marking that this patch replaces in the…
Quoted 37 lines
tools/testing/selftests/bpf/map_tests/array_map_batch_ops.c +4
-1@@ -45,6 +45,9 @@ static void map_batch_verify(int *visited, __u32 max_entries, int *keys,memset (visited,0 , max_entries *sizeof (*visited));for (i =0 ; i < max_entries; i++) { +CHECK (keys[i] <0 || keys[i] >= max_entries,"key checking" , +"error: i %d key %d out of range\n" , i, keys[i]); +if (is_pcpu) { cpu_offset = i * nr_cpus;for (j =0 ; j < nr_cpus; j++) {@@ -59,7 +62,7 @@ static void map_batch_verify(int *visited, __u32 max_entries, int *keys,"error: i %d key %d value %lld\n" , i, keys[i], values[i]); } - visited[i] =1 ; + visited[keys[i]] =1 ; }for (i =0 ; i < max_entries; i++) { CHECK(visited[i] != 1, "visited checking",tools/testing/selftests/bpf/map_tests/htab_map_batch_ops.c +2
-0@@ -50,6 +50,8 @@ static void map_batch_verify(int *visited, __u32 max_entries,memset (visited,0 , max_entries *sizeof (*visited));for (i =0 ; i < max_entries; i++) { +CHECK (keys[i] <1 || keys[i] > max_entries,"key checking" , +"error: i %d key %d out of range\n" , i, keys[i]);if (is_pcpu) {for (j =0 ; j <bpf_num_possible_cpus (); j++) {
tools/testing/selftests/bpf/map_tests/htab_map_batch_ops.c static void map_batch_verify(int *visited, __u32 max_entries, · line 73
Quoted 8 lines
tools/testing/selftests/bpf/map_tests/htab_map_batch_ops.c +1
-1@@ -65,7 +67,7 @@ static void map_batch_verify(int *visited, __u32 max_entries, ((int *)values)[i]); } - visited[i] =1 ; + visited[keys[i] -1 ] =1 ; }for (i =0 ; i < max_entries; i++) {
This isn't a bug, but map_tests/ has three near-identical batch validators, and lpm_trie_map_batch_ops.c still has the same position-indexed visited[i] = 1 marking that this patch replaces in the array and htab copies. The changelog scopes itself to 'the hash and array batch validators', and the LPM key is a struct so it cannot index visited directly, but would it be worth covering the third validator in the same series (perhaps keyed off the parsed last octet), or is the struct key a reason to leave it alone?
The first half of the changelog states the problem and the fix clearly.
Quoted 8 lines
The hash and array batch validators mark visited entries by output position, making the subsequent completeness checks always succeed. Duplicate keys with matching values can therefore hide missing entries.
Check that returned keys fall within the populated range and index the visited array by key. This detects missing entries while preserving unordered results and existing per-CPU value validation.
Quoted 5 lines
Controlled inputs confirm that the old validators accept duplicates and out-of-range keys with matching values, while the updated checks reject them. Ordered, reversed and shuffled valid keys still pass, and scalar and per-CPU value corruption is still rejected. The full
test_mapssuite passes with no skips on the matching kernel.
This isn't a bug, but the last five lines describe an ad hoc validation harness that is not part of the patch and cannot be reproduced from it. Would this validation narrative read better below the --- line, leaving the changelog to the problem and the fix?
---
AI reviewed your patch. Please fix the bug or email reply why it's not a bug.
See: https://github.com/kernel-patches/vmtest/blob/master/ci/claude/README.md
CI run summary: https://github.com/kernel-patches/bpf/actions/runs/34040182269
Thanks for pointing out the third validator.
I have included the LPM trie case in v2. The struct key is not a reason to leave it out: the updated check validates the /32 prefix and complete 192.168.1.x address, then checks the host octet against the inserted key range before using it to index visited.
I also moved the validation notes below the --- separator, leaving the commit message focused on the problem and the fix.
The revised full test_maps suite passes with no skips on the matching kernel, including array, per-CPU array, hash, per-CPU hash and LPM batch tests. The revised patch is here:
https://lore.kernel.org/bpf/[email protected]/
Thanks,
Tianyi