runtime: add network related helpers and maps (#297)

* add bpf_get_smp_processor_id

* add xdp related functions

* add helpers and map in maps

Author: yunwei37

runtime/extension/userspace_xdp.h

@@ -0,0 +1,19 @@
+#ifndef BPFTIME_EXTENSION_USERSPACE_XDP_HPP
+#define BPFTIME_EXTENSION_USERSPACE_XDP_HPP
+
+#include <cstdint>
+
+struct xdp_md_userspace
+{
+	uint64_t data;
+	uint64_t data_end;
+	uint32_t data_meta;
+	uint32_t ingress_ifindex;
+	uint32_t rx_queue_index;
+	uint32_t egress_ifindex;
+	// additional fields
+	uint64_t buffer_start; // record the start of the available buffer
+	uint64_t buffer_end; // record the end of the available buffer
+};
+
+#endif // BPFTIME_EXTENSION_USERSPACE_XDP_HPP

runtime/src/bpf_helper.cpp

@@ -23,6 +23,7 @@
 #include "bpftime.hpp"
 #include "bpftime_shm.hpp"
 #include "bpftime_internal.h"
+#include "extension/userspace_xdp.h"
 #include <spdlog/spdlog.h>
 #include <vector>
 #include <bpftime_shm_internal.hpp>
@@ -104,17 +105,17 @@ uint64_t bpf_get_current_uid_gid(uint64_t, uint64_t, uint64_t, uint64_t,
 
 uint64_t bpftime_ktime_get_ns(uint64_t, uint64_t, uint64_t, uint64_t, uint64_t)
 {
-    auto now = std::chrono::steady_clock::now();
-    auto ns = std::chrono::time_point_cast<std::chrono::nanoseconds>(now);
-    return ns.time_since_epoch().count();
+	auto now = std::chrono::steady_clock::now();
+	auto ns = std::chrono::time_point_cast<std::chrono::nanoseconds>(now);
+	return ns.time_since_epoch().count();
 }
 
 uint64_t bpftime_get_current_comm(uint64_t buf, uint64_t size, uint64_t,
 				  uint64_t, uint64_t)
 {
 	static std::string filename_buf;
 
-	if (filename_buf.empty()) {
+	if (unlikely(filename_buf.empty())) {
 		char strbuf[PATH_MAX];
 
 		auto len = readlink("/proc/self/exe", strbuf,
@@ -219,6 +220,7 @@ uint64_t bpf_ringbuf_submit(uint64_t data, uint64_t flags, uint64_t, uint64_t,
 	bpftime_ringbuf_submit(fd, (void *)(uintptr_t)data, false);
 	return 0;
 }
+
 uint64_t bpf_ringbuf_discard(uint64_t data, uint64_t flags, uint64_t, uint64_t,
 			     uint64_t)
 {
@@ -328,6 +330,7 @@ uint64_t bpftime_tail_call(uint64_t ctx, uint64_t prog_array, uint64_t index)
 	close(to_call_fd);
 	return run_opts.retval;
 }
+
 uint64_t bpftime_get_attach_cookie(uint64_t ctx, uint64_t, uint64_t, uint64_t,
 				   uint64_t)
 {
@@ -340,6 +343,95 @@ uint64_t bpftime_get_attach_cookie(uint64_t ctx, uint64_t, uint64_t, uint64_t,
 		return 0;
 	}
 }
+
+uint64_t bpftime_get_smp_processor_id()
+{
+	int cpu = sched_getcpu();
+	if (cpu == -1) {
+		SPDLOG_ERROR("sched_getcpu error");
+		return 0; // unlikely
+	}
+	return (uint64_t)cpu;
+}
+
+// From https://github.com/microsoft/ebpf-for-windows
+int64_t bpftime_csum_diff(const void *from, int from_size, const void *to,
+			  int to_size, int seed)
+{
+	int csum_diff = -EINVAL;
+
+	if ((from_size % 4 != 0) || (to_size % 4 != 0)) {
+		// size of buffers should be a multiple of 4.
+		goto Exit;
+	}
+
+	csum_diff = seed;
+	if (to != NULL) {
+		for (int i = 0; i < to_size / 2; i++) {
+			csum_diff += (uint16_t)(*((uint16_t *)to + i));
+		}
+	}
+	if (from != NULL) {
+		for (int i = 0; i < from_size / 2; i++) {
+			csum_diff += (uint16_t)(~*((uint16_t *)from + i));
+		}
+	}
+
+	// Adding 16-bit unsigned integers or their one's complement will
+	// produce a positive 32-bit integer, unless the length of the buffers
+	// is so long, that the signed 32 bit output overflows and produces a
+	// negative result.
+	if (csum_diff < 0) {
+		csum_diff = -EINVAL;
+	}
+Exit:
+	return csum_diff;
+}
+
+#define ETH_HLEN 14 /* Total octets in header.	 */
+
+long bpftime_xdp_adjust_head(struct xdp_md_userspace *xdp, int offset)
+{
+	// We don't use xdp meta data
+	uint64_t data = xdp->data + offset;
+	if (unlikely(data > xdp->data_end - ETH_HLEN) || data > xdp->buffer_end)
+		return -EINVAL;
+	if (data < xdp->buffer_start) {
+		// move the data so the buffer can place the new header
+		memmove(reinterpret_cast<void *>(xdp->buffer_start +
+						 (xdp->buffer_start - data)),
+			reinterpret_cast<void *>(xdp->data),
+			xdp->data_end - xdp->data);
+		data = xdp->buffer_start;
+	}
+	xdp->data = data;
+	return 0;
+}
+
+long bpftime_xdp_adjust_tail(struct xdp_md_userspace *xdp_md, int delta)
+{
+	// We don't use xdp meta data
+	uint64_t data = xdp_md->data_end + delta;
+	if (data < xdp_md->data || data < xdp_md->buffer_start ||
+	    data > xdp_md->buffer_end) {
+		return -EINVAL;
+	}
+	xdp_md->data_end = data;
+	return 0;
+}
+
+long bpftime_xdp_load_bytes(struct xdp_md_userspace *xdp_md, __u32 offset,
+			    void *buf, __u32 len)
+{
+	// We don't support fragmented packets
+	uint64_t data = xdp_md->data + offset;
+	if (data + len > xdp_md->data_end) {
+		return -EINVAL;
+	}
+	memcpy(buf, reinterpret_cast<void *>(data), len);
+	return 0;
+}
+
 } // extern "C"
 
 namespace bpftime
@@ -643,6 +735,30 @@ const bpftime_helper_group kernel_helper_group = {
 		    .name = "bpf_probe_read",
 		    .fn = (void *)bpftime_probe_read,
 	    } },
+	  { BPF_FUNC_get_smp_processor_id,
+	    bpftime_helper_info{
+		    .index = BPF_FUNC_get_smp_processor_id,
+		    .name = "bpf_get_smp_processor_id",
+		    .fn = (void *)bpftime_get_smp_processor_id,
+	    } },
+	  { BPF_FUNC_csum_diff,
+	    bpftime_helper_info{
+		    .index = BPF_FUNC_csum_diff,
+		    .name = "bpf_csum_diff",
+		    .fn = (void *)bpftime_csum_diff,
+	    } },
+	  { BPF_FUNC_xdp_adjust_head,
+	    bpftime_helper_info{
+		    .index = BPF_FUNC_xdp_adjust_head,
+		    .name = "bpf_xdp_adjust_head",
+		    .fn = (void *)bpftime_xdp_adjust_head,
+	    } },
+	  { BPF_FUNC_xdp_adjust_tail,
+	    bpftime_helper_info{
+		    .index = BPF_FUNC_xdp_adjust_tail,
+		    .name = "bpf_xdp_adjust_tail",
+		    .fn = (void *)bpftime_xdp_adjust_tail,
+	    } },
 	  { BPF_FUNC_probe_read_kernel,
 	    bpftime_helper_info{
 		    .index = BPF_FUNC_probe_read_kernel,

runtime/src/bpf_map/userspace/map_in_maps.hpp

@@ -0,0 +1,30 @@
+/* SPDX-License-Identifier: MIT
+ *
+ * Copyright (c) 2022, eunomia-bpf org
+ * All rights reserved.
+ */
+#ifndef _MAP_IN_MAP_HPP
+#define _MAP_IN_MAP_HPP
+
+#include <bpf_map/map_common_def.hpp>
+#include "array_map.hpp"
+
+namespace bpftime
+{
+
+// implementation of array map
+class array_map_of_maps_impl : public array_map_impl {
+    public:
+	array_map_of_maps_impl(boost::interprocess::managed_shared_memory &memory, uint32_t max_entries) : array_map_impl(memory, sizeof(int), max_entries){
+			   }
+	// TODO: add verify the correctness of the key
+	void *elem_lookup(const void *key) {
+		auto key_val = array_map_impl::elem_lookup(key);
+		int map_id = *(int *)key_val;
+		return (void*)((u_int64_t)map_id << 32);
+	}
+};
+
+} // namespace bpftime
+
+#endif // _MAP_IN_MAP_HPP

runtime/src/handler/map_handler.cpp

@@ -16,6 +16,7 @@
 #include <bpf_map/shared/percpu_array_map_kernel_user.hpp>
 #include <bpf_map/shared/perf_event_array_kernel_user.hpp>
 #include <bpf_map/userspace/prog_array.hpp>
+#include <bpf_map/userspace/map_in_maps.hpp>
 #include <unistd.h>
 
 using boost::interprocess::interprocess_sharable_mutex;
@@ -134,6 +135,11 @@ const void *bpf_map_handler::map_lookup_elem(const void *key,
 			static_cast<prog_array_map_impl *>(map_impl_ptr.get());
 		return do_lookup(impl);
 	}
+	case bpf_map_type::BPF_MAP_TYPE_ARRAY_OF_MAPS: {
+		auto impl = static_cast<array_map_of_maps_impl *>(
+			map_impl_ptr.get());
+		return do_lookup(impl);
+	}
 	default:
 		auto func_ptr = global_map_ops_table[(int)type].elem_lookup;
 		if (func_ptr) {
@@ -221,6 +227,15 @@ long bpf_map_handler::map_update_elem(const void *key, const void *value,
 			static_cast<prog_array_map_impl *>(map_impl_ptr.get());
 		return do_update(impl);
 	}
+	case bpf_map_type::BPF_MAP_TYPE_ARRAY_OF_MAPS: {
+		if (!from_syscall) {
+			// Map in maps only support update from syscall
+			return -EINVAL;
+		}
+		auto impl = static_cast<array_map_of_maps_impl *>(
+			map_impl_ptr.get());
+		return do_update(impl);
+	}
 	default:
 		auto func_ptr = global_map_ops_table[(int)type].elem_update;
 		if (func_ptr) {
@@ -297,6 +312,11 @@ int bpf_map_handler::bpf_map_get_next_key(const void *key, void *next_key,
 			static_cast<prog_array_map_impl *>(map_impl_ptr.get());
 		return do_get_next_key(impl);
 	}
+	case bpf_map_type::BPF_MAP_TYPE_ARRAY_OF_MAPS: {
+		auto impl = static_cast<array_map_of_maps_impl *>(
+			map_impl_ptr.get());
+		return do_get_next_key(impl);
+	}
 	default:
 		auto func_ptr =
 			global_map_ops_table[(int)type].map_get_next_key;
@@ -384,6 +404,15 @@ long bpf_map_handler::map_delete_elem(const void *key, bool from_syscall) const
 			static_cast<prog_array_map_impl *>(map_impl_ptr.get());
 		return do_delete(impl);
 	}
+	case bpf_map_type::BPF_MAP_TYPE_ARRAY_OF_MAPS: {
+		if (!from_syscall) {
+			// Map in maps only support update from syscall
+			return -EINVAL;
+		}
+		auto impl = static_cast<array_map_of_maps_impl *>(
+			map_impl_ptr.get());
+		return do_delete(impl);
+	}
 	default:
 		auto func_ptr = global_map_ops_table[(int)type].elem_delete;
 		if (func_ptr) {
@@ -476,7 +505,11 @@ int bpf_map_handler::map_init(managed_shared_memory &memory)
 						max_entries);
 		return 0;
 	}
-
+	case bpf_map_type::BPF_MAP_TYPE_ARRAY_OF_MAPS: {
+		map_impl_ptr = memory.construct<array_map_of_maps_impl>(
+			container_name.c_str())(memory, max_entries);
+		return 0;
+	}
 	default:
 		if (bpftime_get_agent_config().allow_non_buildin_map_types) {
 			SPDLOG_INFO("non-builtin map type: {}", (int)type);