build: add USE_HASHSET option to switch to hashset implementation

When cmake -DUSE_HASHSET=1 is passed, set.h redirects all bf_set_*
symbols to their bf_hashset_* counterparts via preprocessor macros.
The CI workflow is extended to test both paths. USE_HASHSET defaults
to off for now.

Author: pzmarzly

.github/workflows/ci.yaml

@@ -94,24 +94,25 @@ jobs:
         host:
           - { name: 8-core-ubuntu, arch: x64 }
           - { name: 4-core-ubuntu-arm, arch: arm64 }
+        use_hashset: [ 0, 1 ]
     runs-on: [ "${{ matrix.host.name }}" ]
     container:
       image: ghcr.io/facebook/bpfilter:fedora-43-${{ matrix.host.arch }}
       options: --privileged
-    name: "Test: ${{ matrix.host.arch }}"
+    name: "Test: ${{ matrix.host.arch }}${{ matrix.use_hashset == 1 && ' (hashset)' || '' }}"
     steps:
       - name: Checkout bpfilter
         uses: actions/checkout@v2
       - name: Restore the cached test results
         uses: actions/cache@v4
-        if: matrix.host.arch == 'x64'
+        if: matrix.host.arch == 'x64' && matrix.use_hashset == 0
         with:
           path: build/coverage
           key: tests-results-${{ github.run_id }}
       - name: Mount bpffs
         run: mount bpffs /sys/fs/bpf -t bpf
       - name: Configure the build
-        run: cmake -S $GITHUB_WORKSPACE -B $GITHUB_WORKSPACE/build -DWITH_COVERAGE=1
+        run: cmake -S $GITHUB_WORKSPACE -B $GITHUB_WORKSPACE/build -DWITH_COVERAGE=1 -DUSE_HASHSET=${{ matrix.use_hashset }}
       - name: Build tests
         run: make -C $GITHUB_WORKSPACE/build -j `nproc` test_bin
 
@@ -125,7 +126,7 @@ jobs:
         run: ctest --test-dir $GITHUB_WORKSPACE/build -L fuzzing --verbose
       - name: Upload fuzzer findings
         uses: actions/upload-artifact@v4
-        if: always()
+        if: always() && matrix.use_hashset == 0
         with:
           name: fuzzer-findings-${{ matrix.host.arch }}
           path: ${{ github.workspace }}/build/findings
@@ -136,6 +137,7 @@ jobs:
       - name: Run checks
         run: ctest --test-dir $GITHUB_WORKSPACE/build -L check --verbose
       - name: Generate the coverage report
+        if: matrix.use_hashset == 0
         run: make -C $GITHUB_WORKSPACE/build coverage
 
   benchmark:

CMakeLists.txt

@@ -34,6 +34,7 @@ option(NO_TESTS "Disable unit, end-to-end, and integration tests" 0)
 option(NO_CHECKS "Disable the check target (clang-tidy and clang-format" 0)
 option(NO_BENCHMARKS "Disable the benchmark" 0)
 option(WITH_COVERAGE "Build with code coverage support. Disabled by default" 0)
+option(USE_HASHSET "Use bf_hashset instead of bf_set for set implementation. Disabled by default" 0)
 
 set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
 

src/libbpfilter/CMakeLists.txt

@@ -55,7 +55,6 @@ set(libbpfilter_srcs
     ${CMAKE_CURRENT_SOURCE_DIR}/request.c
     ${CMAKE_CURRENT_SOURCE_DIR}/response.c
     ${CMAKE_CURRENT_SOURCE_DIR}/rule.c
-    ${CMAKE_CURRENT_SOURCE_DIR}/set.c
     ${CMAKE_CURRENT_SOURCE_DIR}/vector.c
     ${CMAKE_CURRENT_SOURCE_DIR}/verdict.c
     ${CMAKE_CURRENT_SOURCE_DIR}/version.c
@@ -65,6 +64,15 @@ set(libbpfilter_srcs
     ${CMAKE_SOURCE_DIR}/src/external/mpack.c
 )
 
+if (NOT USE_HASHSET)
+    list(APPEND libbpfilter_srcs ${CMAKE_CURRENT_SOURCE_DIR}/set.c)
+endif ()
+
+set(BF_PC_EXTRA_CFLAGS "")
+if (USE_HASHSET)
+    string(APPEND BF_PC_EXTRA_CFLAGS " -DBF_USE_HASHSET")
+endif ()
+
 configure_file(
     ${CMAKE_CURRENT_SOURCE_DIR}/bpfilter.pc.in
     ${CMAKE_BINARY_DIR}/output/lib/pkgconfig/bpfilter.pc
@@ -86,6 +94,8 @@ target_compile_definitions(libbpfilter
     PRIVATE
         # MPack should use the C standard library API
         MPACK_STDLIB
+    PUBLIC
+        $<$<BOOL:${USE_HASHSET}>:BF_USE_HASHSET>
 )
 
 target_include_directories(libbpfilter

src/libbpfilter/bpfilter.pc.in

@@ -6,5 +6,5 @@ Name: bpfilter
 Description: BPF-based packet filtering framework
 URL: https://github.com/facebook/bpfilter
 Version: @PROJECT_VERSION@@PROJECT_VERSION_SUFFIX@
-Cflags: -I${includedir}
+Cflags: -I${includedir}@BF_PC_EXTRA_CFLAGS@
 Libs: -L${libdir} -lbpfilter

src/libbpfilter/include/bpfilter/bpfilter.h

@@ -9,10 +9,10 @@
 #include <stddef.h>
 
 #include <bpfilter/list.h>
+#include <bpfilter/set.h>
 
 struct bf_response;
 struct bf_chain;
-struct bf_set;
 struct bf_hookopts;
 
 /**

src/libbpfilter/include/bpfilter/chain.h

@@ -9,12 +9,12 @@
 #include <bpfilter/hook.h>
 #include <bpfilter/list.h>
 #include <bpfilter/pack.h>
+#include <bpfilter/set.h>
 #include <bpfilter/verdict.h>
 
 struct bf_hookopts;
 struct bf_matcher;
 struct bf_rule;
-struct bf_set;
 
 #define _free_bf_chain_ __attribute__((cleanup(bf_chain_free)))
 

src/libbpfilter/include/bpfilter/set.h

@@ -5,6 +5,30 @@
 
 #pragma once
 
+#ifdef BF_USE_HASHSET
+
+#include <bpfilter/hashset.h>
+
+#define bf_set bf_hashset
+#define _free_bf_set_ _free_bf_hashset_
+#define BF_SET_MAX_N_COMPS BF_HASHSET_MAX_N_COMPS
+#define bf_set_new bf_hashset_new
+#define bf_set_new_from_raw bf_hashset_new_from_raw
+#define bf_set_new_from_pack bf_hashset_new_from_pack
+#define bf_set_free bf_hashset_free
+#define bf_set_pack bf_hashset_pack
+#define bf_set_dump bf_hashset_dump
+#define bf_set_is_empty bf_hashset_is_empty
+#define bf_set_add_elem bf_hashset_add_elem
+#define bf_set_add_elem_raw bf_hashset_add_elem_raw
+#define bf_set_add_many bf_hashset_add_many
+#define bf_set_remove_many bf_hashset_remove_many
+#define bf_set_foreach bf_hashset_foreach
+#define bf_set_size bf_hashset_size
+#define bf_set_contains bf_hashset_contains
+
+#else
+
 #include <stdbool.h>
 #include <stddef.h>
 
@@ -193,3 +217,14 @@ int bf_set_add_many(struct bf_set *dest, struct bf_set **to_add);
  * - `-EINVAL`: set key format doesn't match between dest and to_remove.
  */
 int bf_set_remove_many(struct bf_set *dest, struct bf_set **to_remove);
+
+/**
+ * @brief Check whether an element exists in a set.
+ *
+ * @param set Initialised set. Can't be NULL.
+ * @param elem Element to look up. Can't be NULL.
+ * @return True if @p elem is present.
+ */
+bool bf_set_contains(const struct bf_set *set, const void *elem);
+
+#endif

src/libbpfilter/set.c

@@ -523,3 +523,16 @@ int bf_set_remove_many(struct bf_set *dest, struct bf_set **to_remove)
 
     return 0;
 }
+
+bool bf_set_contains(const struct bf_set *set, const void *elem)
+{
+    assert(set);
+    assert(elem);
+
+    bf_list_foreach (&set->elems, node) {
+        if (memcmp(bf_list_node_get_data(node), elem, set->elem_size) == 0)
+            return true;
+    }
+
+    return false;
+}

tests/harness/fake.c

@@ -221,6 +221,7 @@ struct bf_matcher *bft_matcher_dummy(const void *data, size_t data_len)
     return TAKE_PTR(matcher);
 }
 
+#ifndef BF_USE_HASHSET
 struct bf_set *bft_set_dummy(size_t n_elems)
 {
     _free_bf_set_ struct bf_set *set = NULL;
@@ -245,6 +246,7 @@ struct bf_set *bft_set_dummy(size_t n_elems)
 
     return TAKE_PTR(set);
 }
+#endif
 
 struct bf_hashset *bft_hashset_dummy(size_t n_elems)
 {

tests/harness/fake.h

@@ -53,5 +53,9 @@ const void *bft_get_randomly_filled_buffer(size_t len);
 struct bf_chain *bft_chain_dummy(bool with_rules);
 struct bf_rule *bft_rule_dummy(size_t n_matchers);
 struct bf_matcher *bft_matcher_dummy(const void *data, size_t data_len);
+#ifdef BF_USE_HASHSET
+#define bft_set_dummy bft_hashset_dummy
+#else
 struct bf_set *bft_set_dummy(size_t n_elems);
+#endif
 struct bf_hashset *bft_hashset_dummy(size_t n_elems);