Use bpftool instead of iproute2 to load eBPF and XDP maps.

Signed-off-by: fruffy <[email protected]>

backends/ebpf/CMakeLists.txt

@@ -21,17 +21,17 @@ if(NOT APPLE)
   set(FETCHCONTENT_QUIET OFF)
   fetchcontent_declare(
     bpfrepo
-    URL https://github.com/libbpf/libbpf/archive/refs/tags/v1.4.1.tar.gz
-    URL_HASH SHA256=cc01a3a05d25e5978c20be7656f14eb8b6fcb120bb1c7e8041e497814fc273cb
-    SOURCE_DIR ${CMAKE_CURRENT_SOURCE_DIR}/runtime/contrib/libbpf
+    URL https://github.com/libbpf/bpftool/releases/download/v7.5.0/bpftool-libbpf-v7.5.0-sources.tar.gz
+    # URL_HASH SHA256=cc01a3a05d25e5978c20be7656f14eb8b6fcb120bb1c7e8041e497814fc273cb
+    SOURCE_DIR ${CMAKE_CURRENT_SOURCE_DIR}/runtime/contrib/bpftool
     USES_TERMINAL_DOWNLOAD TRUE
     GIT_PROGRESS TRUE
     DOWNLOAD_EXTRACT_TIMESTAMP TRUE
   )
   fetchcontent_makeavailable(bpfrepo)
   set(FETCHCONTENT_QUIET ${FETCHCONTENT_QUIET_PREV})
   # Check if we have already built the libbpf library.
-  find_library(LIBBPF NAMES bpf HINTS "${CMAKE_CURRENT_SOURCE_DIR}/runtime/usr/lib64/")
+  find_library(LIBBPF NAMES bpf HINTS "${CMAKE_CURRENT_SOURCE_DIR}/runtime/install/libbpf/")
   if (NOT LIBBPF)
     message("Building libbpf...")
     execute_process(

backends/ebpf/build_libbpf

@@ -10,7 +10,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-""" This programs builds a libbpf static library and places it in the runtime
+""" This programs builds a bpftool binary and the libbpf static library and places it in the runtime
     folder. The library and its headers are required by the kernel target.
 """
 
@@ -31,21 +31,24 @@ def main() -> int:
         level=logging.WARN,
         filemode="w",
     )
-    libbpf_dir = FILE_DIR.joinpath("runtime/contrib/libbpf")
-    libbpf_src_dir = libbpf_dir.joinpath("src")
-    libbpf_build_dir = libbpf_src_dir.joinpath("build")
-    libbpf_target_dir = FILE_DIR.joinpath("runtime")
-    # Create the libbpf build directory
-    testutils.check_and_create_dir(libbpf_build_dir)
-    # Build libbpf
-    mk_cmd = f"make -C {libbpf_src_dir} install "
+    bpftool_dir = FILE_DIR.joinpath("runtime/contrib/bpftool")
+    bpftool_src_dir = bpftool_dir.joinpath("src")
+    bpftool_build_dir = bpftool_src_dir.joinpath("build")
+    bpftool_target_dir = FILE_DIR.joinpath("runtime/install/")
+    bpftool_target_dir.mkdir(parents=True, exist_ok=True)
+    # Create the bpftool build directory
+    testutils.check_and_create_dir(bpftool_build_dir)
+    # Build bpftool
+    mk_cmd = f"make -C {bpftool_src_dir} install "
+    mk_cmd += f"OUTPUT={bpftool_target_dir}/ "
     mk_cmd += "BUILD_STATIC_ONLY=y "
-    mk_cmd += f"OBJDIR={libbpf_build_dir} "
-    mk_cmd += f"DESTDIR={libbpf_target_dir} "
+    mk_cmd += f"OBJDIR={bpftool_build_dir} "
+    mk_cmd += f"DESTDIR={bpftool_target_dir} "
     mk_cmd += "EXTRA_CFLAGS=-fPIE"
+    print(mk_cmd)
     result = testutils.exec_process(args=mk_cmd)
     if result.returncode != testutils.SUCCESS:
-        testutils.log.error("Could not build libbpf")
+        testutils.log.error("Could not build bpftool")
     return result.returncode
 
 

backends/ebpf/run-ebpf-test.py

@@ -123,9 +123,9 @@ def __init__(self):
         self.replace = False  # Replace previous outputs.
         self.target = "test"  # The name of the target compiler.
         # Actual location of the test framework.
-        self.testdir = str(FILE_DIR)
+        self.testdir = FILE_DIR
         # The location of the eBPF runtime, some targets may overwrite this.
-        self.runtimedir = str(FILE_DIR.joinpath("runtime"))
+        self.runtimedir = FILE_DIR.joinpath("runtime")
         self.extern = ""  # Path to C file with extern definition.
 
 

backends/ebpf/runtime/.gitignore

@@ -0,0 +1 @@
+install

backends/ebpf/runtime/contrib/.gitignore

@@ -1 +1,2 @@
 libbpf
+bpftool

backends/ebpf/runtime/ebpf_kernel.h

@@ -14,16 +14,14 @@ See the License for the specific language governing permissions and
 limitations under the License.
 */
 
-
 /// This file contains all functions and definitions necessary for the kernel target C
 /// code to compile. It must be included with any file generated by the p4c-ebpf kernel
 /// compiler.
 #ifndef BACKENDS_EBPF_RUNTIME_EBPF_KERNEL_H_
 #define BACKENDS_EBPF_RUNTIME_EBPF_KERNEL_H_
 
 #include "ebpf_common.h"
-
-#include <bpf/bpf_endian.h> // definitions for bpf_ntohs etc...
+#include "install/libbpf/include/bpf/bpf_endian.h"  // definitions for bpf_ntohs etc...
 
 #undef htonl
 #undef htons
@@ -35,158 +33,82 @@ limitations under the License.
 #define bpf_htonll(x) htonll(x)
 #endif
 
-#define load_byte(data, b) (*(((u8*)(data)) + (b)))
-#define load_half(data, b) bpf_ntohs(*(u16 *)((u8*)(data) + (b)))
-#define load_word(data, b) bpf_ntohl(*(u32 *)((u8*)(data) + (b)))
-#define load_dword(data, b) bpf_be64_to_cpu(*(u64 *)((u8*)(data) + (b)))
+#define load_byte(data, b) (*(((u8 *)(data)) + (b)))
+#define load_half(data, b) bpf_ntohs(*(u16 *)((u8 *)(data) + (b)))
+#define load_word(data, b) bpf_ntohl(*(u32 *)((u8 *)(data) + (b)))
+#define load_dword(data, b) bpf_be64_to_cpu(*(u64 *)((u8 *)(data) + (b)))
 
 /// If we operate in user space we only need to include bpf.h and
 /// define the userspace API macros.
 /// For kernel programs we need to specify a list of kernel helpers. These are
-/// taken from here: https://github.com/torvalds/linux/blob/master/tools/lib/bpf/bpf_helpers.h
+/// taken from here: https://github.com/libbpf/libbpf/blob/master/src/bpf_helpers.h
 
-#ifdef CONTROL_PLANE // BEGIN EBPF USER SPACE DEFINITIONS
+#ifdef CONTROL_PLANE  // BEGIN EBPF USER SPACE DEFINITIONS
 
-#include <bpf/bpf.h> // bpf_obj_get/pin, bpf_map_update_elem
+#include "install/libbpf/include/bpf/bpf.h"  // bpf_obj_get/pin, bpf_map_update_elem
 
-#define BPF_USER_MAP_UPDATE_ELEM(index, key, value, flags)\
+#define BPF_USER_MAP_UPDATE_ELEM(index, key, value, flags) \
     bpf_map_update_elem(index, key, value, flags)
 #define BPF_OBJ_PIN(table, name) bpf_obj_pin(table, name)
 #define BPF_OBJ_GET(name) bpf_obj_get(name)
 
-#else // BEGIN EBPF KERNEL DEFINITIONS
+#else  // BEGIN EBPF KERNEL DEFINITIONS
 
-#include <linux/pkt_cls.h>  // TC_ACT_OK, TC_ACT_SHOT
-#include "linux/bpf.h"  // types, and general bpf definitions
+#include "contrib/bpftool/include/uapi/linux/bpf.h"  // BPF_ANY,
+#include "contrib/bpftool/include/uapi/linux/pkt_cls.h"  // TC_ACT_OK, TC_ACT_SHOT
 // This file contains the definitions of all the kernel bpf essentials
-#include <bpf/bpf_helpers.h>
-
-/// A helper structure used by an eBPF C program
-/// to describe map attributes for the elf_bpf loader
-/// FIXME: We only need this because we are loading with iproute2
-struct bpf_elf_map {
-    __u32 type;
-    __u32 size_key;
-    __u32 size_value;
-    __u32 max_elem;
-    __u32 flags;
-    __u32 id;
-    __u32 pinning;
-    __u32 inner_id;
-    __u32 inner_idx;
-};
+#include "install/libbpf/include/bpf/bpf_helpers.h"
 
 /// Simple descriptor which replaces the kernel sk_buff structure.
 #define SK_BUFF struct __sk_buff
 
-/// From iproute2, annotate table with BTF which allows to read types at runtime.
-#define BPF_ANNOTATE_KV_PAIR(name, type_key, type_val)  \
-    struct ____btf_map_##name {                         \
-        type_key key;                                   \
-        type_val value;                                 \
-    };                                                  \
-    struct ____btf_map_##name                           \
-        __attribute__ ((section(".maps." #name), used)) \
-        ____btf_map_##name = {};
-
 #define REGISTER_START()
-#ifndef BTF
-/// Note: pinning exports the table name globally, do not remove.
-#define REGISTER_TABLE(NAME, TYPE, KEY_TYPE, VALUE_TYPE, MAX_ENTRIES) \
-struct bpf_elf_map SEC("maps") NAME = {          \
-    .type        = TYPE,               \
-    .size_key    = sizeof(KEY_TYPE),   \
-    .size_value  = sizeof(VALUE_TYPE), \
-    .max_elem    = MAX_ENTRIES,        \
-    .pinning     = 2,                  \
-    .flags       = 0,                  \
-};
-#define REGISTER_TABLE_INNER(NAME, TYPE, KEY_TYPE, VALUE_TYPE, MAX_ENTRIES, ID, INNER_IDX) \
-struct bpf_elf_map SEC("maps") NAME = {          \
-    .type        = TYPE,               \
-    .size_key    = sizeof(KEY_TYPE),   \
-    .size_value  = sizeof(VALUE_TYPE), \
-    .max_elem    = MAX_ENTRIES,        \
-    .pinning     = 2,                  \
-    .flags       = 0,                  \
-    .id          = ID,                 \
-    .inner_idx   = INNER_IDX,          \
-};
-#define REGISTER_TABLE_OUTER(NAME, TYPE, KEY_TYPE, VALUE_TYPE, MAX_ENTRIES, INNER_ID, INNER_NAME) \
-struct bpf_elf_map SEC("maps") NAME = {          \
-    .type        = TYPE,               \
-    .size_key    = sizeof(KEY_TYPE),   \
-    .size_value  = sizeof(VALUE_TYPE), \
-    .max_elem    = MAX_ENTRIES,        \
-    .pinning     = 2,                  \
-    .flags       = 0,                  \
-    .inner_id    = INNER_ID,           \
-};
-#define REGISTER_TABLE_FLAGS(NAME, TYPE, KEY_TYPE, VALUE_TYPE, MAX_ENTRIES, FLAGS) \
-struct bpf_elf_map SEC("maps") NAME = {          \
-    .type        = TYPE,               \
-    .size_key    = sizeof(KEY_TYPE),   \
-    .size_value  = sizeof(VALUE_TYPE), \
-    .max_elem    = MAX_ENTRIES,        \
-    .pinning     = 2,                  \
-    .flags       = FLAGS,              \
-};
-#else
 #define REGISTER_TABLE(NAME, TYPE, KEY_TYPE, VALUE_TYPE, MAX_ENTRIES) \
-struct {                                 \
-    __uint(type, TYPE);                  \
-    KEY_TYPE *key;                       \
-    VALUE_TYPE *value;                   \
-    __uint(max_entries, MAX_ENTRIES);    \
-    __uint(pinning, LIBBPF_PIN_BY_NAME); \
-} NAME SEC(".maps");
+    struct {                                                          \
+        __uint(type, TYPE);                                           \
+        KEY_TYPE *key;                                                \
+        VALUE_TYPE *value;                                            \
+        __uint(max_entries, MAX_ENTRIES);                             \
+    } NAME SEC(".maps");
 #define REGISTER_TABLE_FLAGS(NAME, TYPE, KEY_TYPE, VALUE_TYPE, MAX_ENTRIES, FLAGS) \
-struct {                                 \
-    __uint(type, TYPE);                  \
-    KEY_TYPE *key;                       \
-    VALUE_TYPE *value;                   \
-    __uint(max_entries, MAX_ENTRIES);    \
-    __uint(pinning, LIBBPF_PIN_BY_NAME); \
-    __uint(map_flags, FLAGS);            \
-} NAME SEC(".maps");
+    struct {                                                                       \
+        __uint(type, TYPE);                                                        \
+        KEY_TYPE *key;                                                             \
+        VALUE_TYPE *value;                                                         \
+        __uint(max_entries, MAX_ENTRIES);                                          \
+        __uint(map_flags, FLAGS);                                                  \
+    } NAME SEC(".maps");
 #define REGISTER_TABLE_INNER(NAME, TYPE, KEY_TYPE, VALUE_TYPE, MAX_ENTRIES, ID, INNER_IDX) \
-struct NAME {                            \
-    __uint(type, TYPE);                  \
-    KEY_TYPE *key;                       \
-    VALUE_TYPE *value;                   \
-    __uint(max_entries, MAX_ENTRIES);    \
-} NAME SEC(".maps");
+    struct NAME {                                                                          \
+        __uint(type, TYPE);                                                                \
+        KEY_TYPE *key;                                                                     \
+        VALUE_TYPE *value;                                                                 \
+        __uint(max_entries, MAX_ENTRIES);                                                  \
+    } NAME SEC(".maps");
 #define REGISTER_TABLE_OUTER(NAME, TYPE, KEY_TYPE, VALUE_TYPE, MAX_ENTRIES, INNER_ID, INNER_NAME) \
-struct {                                 \
-    __uint(type, TYPE);                  \
-    KEY_TYPE *key;                       \
-    VALUE_TYPE *value;                   \
-    __uint(max_entries, MAX_ENTRIES);    \
-    __uint(pinning, LIBBPF_PIN_BY_NAME); \
-    __array(values, struct INNER_NAME);  \
-} NAME SEC(".maps");
+    struct {                                                                                      \
+        __uint(type, TYPE);                                                                       \
+        KEY_TYPE *key;                                                                            \
+        VALUE_TYPE *value;                                                                        \
+        __uint(max_entries, MAX_ENTRIES);                                                         \
+        __array(values, struct INNER_NAME);                                                       \
+    } NAME SEC(".maps") = {.values = {&INNER_NAME}};
 #define REGISTER_TABLE_NO_KEY_TYPE(NAME, TYPE, KEY_SIZE, VALUE_TYPE, MAX_ENTRIES) \
-struct {                                 \
-    __uint(type, TYPE);                  \
-    __uint(key_size, KEY_SIZE);          \
-    VALUE_TYPE *value;                   \
-    __uint(max_entries, MAX_ENTRIES);    \
-    __uint(pinning, LIBBPF_PIN_BY_NAME); \
-} NAME SEC(".maps");
-#endif
+    struct {                                                                      \
+        __uint(type, TYPE);                                                       \
+        __uint(key_size, KEY_SIZE);                                               \
+        VALUE_TYPE *value;                                                        \
+        __uint(max_entries, MAX_ENTRIES);                                         \
+    } NAME SEC(".maps");
 #define REGISTER_END()
 
-#define BPF_MAP_LOOKUP_ELEM(table, key) \
-    bpf_map_lookup_elem(&table, key)
-#define BPF_MAP_UPDATE_ELEM(table, key, value, flags) \
-    bpf_map_update_elem(&table, key, value, flags)
-#define BPF_MAP_DELETE_ELEM(table, key) \
-    bpf_map_delete_elem(&table, key)
-#define BPF_USER_MAP_UPDATE_ELEM(index, key, value, flags)\
-    bpf_update_elem(index, key, value, flags)
+#define BPF_MAP_LOOKUP_ELEM(table, key) bpf_map_lookup_elem(&table, key)
+#define BPF_MAP_UPDATE_ELEM(table, key, value, flags) bpf_map_update_elem(&table, key, value, flags)
+#define BPF_MAP_DELETE_ELEM(table, key) bpf_map_delete_elem(&table, key)
+#define BPF_USER_MAP_UPDATE_ELEM(index, key, value, flags) bpf_update_elem(index, key, value, flags)
 #define BPF_OBJ_PIN(table, name) bpf_obj_pin(table, name)
 #define BPF_OBJ_GET(name) bpf_obj_get(name)
 
-#endif // END EBPF KERNEL DEFINITIONS
+#endif  // END EBPF KERNEL DEFINITIONS
 
 #endif  // BACKENDS_EBPF_RUNTIME_EBPF_KERNEL_H_

backends/ebpf/runtime/kernel.mk

@@ -3,7 +3,7 @@ ROOT_DIR := $(dir $(abspath $(lastword $(MAKEFILE_LIST))))
 # Argument for the CLANG compiler
 LLC ?= llc
 CLANG ?= clang
-override INCLUDES+= -I$(ROOT_DIR) -I$(ROOT_DIR)usr/include/ -I$(ROOT_DIR)contrib/libbpf/include/uapi/
+override INCLUDES+= -I$(ROOT_DIR)
 override LIBS+=
 # Optimization flags to save space
 override CFLAGS+= -O2 -g -c -D__KERNEL__ -D__ASM_SYSREG_H \

backends/ebpf/target.cpp

@@ -98,7 +98,6 @@ void KernelSamplesTarget::emitTableDecl(Util::SourceCodeBuilder *builder, cstrin
                               flags);
     }
     builder->newline();
-    annotateTableWithBTF(builder, tblName, keyType, valueType);
 }
 
 void KernelSamplesTarget::emitTableDeclSpinlock(Util::SourceCodeBuilder *builder, cstring tblName,
@@ -131,14 +130,12 @@ void KernelSamplesTarget::emitMapInMapDecl(Util::SourceCodeBuilder *builder, cst
     builder->appendFormat(registerInnerTable, innerName, kind, innerKeyType, innerValueType,
                           innerSize, innerMapIndex, innerMapIndex);
     builder->newline();
-    annotateTableWithBTF(builder, innerName, innerKeyType, innerValueType);
 
     kind = getBPFMapType(outerTableKind);
     cstring keyType = outerTableKind == TableArray ? "__u32"_cs : outerKeyType;
     builder->appendFormat(registerOuterTable, outerName, kind, keyType, "__u32", outerSize,
                           innerMapIndex, innerName);
     builder->newline();
-    annotateTableWithBTF(builder, outerName, keyType, "__u32"_cs);
 }
 
 void KernelSamplesTarget::emitLicense(Util::SourceCodeBuilder *builder, cstring license) const {
@@ -202,12 +199,6 @@ void KernelSamplesTarget::emitTraceMessage(Util::SourceCodeBuilder *builder, con
     builder->newline();
 }
 
-void KernelSamplesTarget::annotateTableWithBTF(Util::SourceCodeBuilder *builder, cstring name,
-                                               cstring keyType, cstring valueType) const {
-    builder->appendFormat("BPF_ANNOTATE_KV_PAIR(%v, %v, %v)", name, keyType, valueType);
-    builder->newline();
-}
-
 //////////////////////////////////////////////////////////////
 void XdpTarget::emitResizeBuffer(Util::SourceCodeBuilder *builder, cstring buffer,
                                  cstring offsetVar) const {

backends/ebpf/target.h

@@ -195,9 +195,6 @@ class KernelSamplesTarget : public Target {
     cstring sysMapPath() const override { return "/sys/fs/bpf/tc/globals"_cs; }
 
     cstring packetDescriptorType() const override { return "struct __sk_buff"_cs; }
-
-    void annotateTableWithBTF(Util::SourceCodeBuilder *builder, cstring name, cstring keyType,
-                              cstring valueType) const;
 };
 
 class P4TCTarget : public KernelSamplesTarget {

backends/ebpf/targets/ebpfenv.py

@@ -96,7 +96,7 @@ def ns_proc_append(self, proc: subprocess.Popen, cmd: str) -> int:
     def ns_proc_close(self, proc: subprocess.Popen, **extra_args: Any) -> int:
         """Close and actually run the process in the namespace. Returns the
         exit code."""
-        testutils.log.info("Executing command: %s", proc)
+        testutils.log.info("Executing command: %s", proc.args)
         result = testutils.run_process(proc, timeout=testutils.TIMEOUT, **extra_args)
         if result.returncode != testutils.SUCCESS:
             testutils.log.error(
@@ -109,7 +109,7 @@ def _configure_bridge(self, br_name: str) -> int:
         avoid ICMPv6 spam."""
         # We do not care about failures here
         self.ns_exec(f"ip link set dev {br_name} up")
-        self.ns_exec(f"ip link set dev {br_name} mtu 9000")
+        self.ns_exec(f"ip link set dev {br_name} mtu 1500")
         # Prevent the broadcasting of ipv6 link discovery messages
         self.ns_exec("sysctl -w net.ipv6.conf.all.disable_ipv6=1")
         self.ns_exec("sysctl -w net.ipv6.conf.default.disable_ipv6=1")
@@ -130,7 +130,7 @@ def _configure_bridge_port(self, port_name: str) -> int:
         result = self.ns_exec(cmd)
         if result != testutils.SUCCESS:
             return result
-        cmd = f"ip link set dev {port_name} mtu 9000"
+        cmd = f"ip link set dev {port_name} mtu 1500"
         return self.ns_exec(cmd)
 
     def attach_interfaces(self, num_ifaces: int) -> int: