introduce BPF_F_SHARE_PE

.travis.yml

@@ -0,0 +1,33 @@
+sudo: required
+language: bash
+dist: bionic
+services:
+    - docker
+
+env:
+    global:
+        - PROJECT_NAME='libbpf'
+        - AUTHOR_EMAIL="$(git log -1 --pretty=\"%aE\")"
+        - REPO_ROOT="$TRAVIS_BUILD_DIR"
+        - CI_ROOT="$REPO_ROOT/travis-ci"
+        - VMTEST_ROOT="$CI_ROOT/vmtest"
+
+addons:
+    apt:
+        packages:
+            - qemu-kvm
+            - zstd
+            - binutils-dev
+            - elfutils
+            - libcap-dev
+            - libelf-dev
+            - libdw-dev
+            - python3-docutils
+
+jobs:
+    include:
+        - stage: Builds & Tests
+          name: Kernel LATEST + selftests
+          language: bash
+          env: KERNEL=LATEST
+          script:  $CI_ROOT/vmtest/run_vmtest.sh || travis_terminate 1

include/uapi/linux/bpf.h

@@ -414,6 +414,9 @@ enum {
 
 /* Enable memory-mapping BPF map */
 	BPF_F_MMAPABLE		= (1U << 10),
+
+/* Share perf_event among processes */
+	BPF_F_SHARE_PE		= (1U << 11),
 };
 
 /* Flags for BPF_PROG_QUERY. */

kernel/bpf/arraymap.c

@@ -15,7 +15,7 @@
 #include "map_in_map.h"
 
 #define ARRAY_CREATE_FLAG_MASK \
-	(BPF_F_NUMA_NODE | BPF_F_MMAPABLE | BPF_F_ACCESS_MASK)
+	(BPF_F_NUMA_NODE | BPF_F_MMAPABLE | BPF_F_ACCESS_MASK | BPF_F_SHARE_PE)
 
 static void bpf_array_free_percpu(struct bpf_array *array)
 {
@@ -64,6 +64,10 @@ int array_map_alloc_check(union bpf_attr *attr)
 	    attr->map_flags & BPF_F_MMAPABLE)
 		return -EINVAL;
 
+	if (attr->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY &&
+	    attr->map_flags & BPF_F_SHARE_PE)
+		return -EINVAL;
+
 	if (attr->value_size > KMALLOC_MAX_SIZE)
 		/* if value_size is bigger, the user space won't be able to
 		 * access the elements.
@@ -778,6 +782,26 @@ static int fd_array_map_delete_elem(struct bpf_map *map, void *key)
 	}
 }
 
+static void perf_event_fd_array_map_free(struct bpf_map *map)
+{
+	struct bpf_event_entry *ee;
+	struct bpf_array *array;
+	int i;
+
+	if ((map->map_flags & BPF_F_SHARE_PE) == 0) {
+		fd_array_map_free(map);
+		return;
+	}
+
+	array = container_of(map, struct bpf_array, map);
+	for (i = 0; i < array->map.max_entries; i++) {
+		ee = READ_ONCE(array->ptrs[i]);
+		if (ee)
+			fd_array_map_delete_elem(map, &i);
+	}
+	bpf_map_area_free(array);
+}
+
 static void *prog_fd_array_get_ptr(struct bpf_map *map,
 				   struct file *map_file, int fd)
 {
@@ -1134,6 +1158,9 @@ static void perf_event_fd_array_release(struct bpf_map *map,
 	struct bpf_event_entry *ee;
 	int i;
 
+	if (map->map_flags & BPF_F_SHARE_PE)
+		return;
+
 	rcu_read_lock();
 	for (i = 0; i < array->map.max_entries; i++) {
 		ee = READ_ONCE(array->ptrs[i]);
@@ -1148,7 +1175,7 @@ const struct bpf_map_ops perf_event_array_map_ops = {
 	.map_meta_equal = bpf_map_meta_equal,
 	.map_alloc_check = fd_array_map_alloc_check,
 	.map_alloc = array_map_alloc,
-	.map_free = fd_array_map_free,
+	.map_free = perf_event_fd_array_map_free,
 	.map_get_next_key = array_map_get_next_key,
 	.map_lookup_elem = fd_array_map_lookup_elem,
 	.map_delete_elem = fd_array_map_delete_elem,

tools/include/uapi/linux/bpf.h

@@ -414,6 +414,9 @@ enum {
 
 /* Enable memory-mapping BPF map */
 	BPF_F_MMAPABLE		= (1U << 10),
+
+/* Share perf_event among processes */
+	BPF_F_SHARE_PE		= (1U << 11),
 };
 
 /* Flags for BPF_PROG_QUERY. */

tools/testing/selftests/bpf/prog_tests/perf_event_share.c

@@ -0,0 +1,68 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/* Copyright (c) 2019 Facebook */
+#include <test_progs.h>
+#include <linux/bpf.h>
+#include "test_perf_event_share.skel.h"
+
+static int duration;
+
+static void test_one_map(struct bpf_map *map, struct bpf_program *prog,
+			 bool has_share_pe)
+{
+	int err, key = 0, pfd = -1, mfd = bpf_map__fd(map);
+	DECLARE_LIBBPF_OPTS(bpf_test_run_opts, opts);
+	struct perf_event_attr attr = {
+		.size = sizeof(struct perf_event_attr),
+		.type = PERF_TYPE_SOFTWARE,
+		.config = PERF_COUNT_SW_CPU_CLOCK,
+	};
+
+	pfd = syscall(__NR_perf_event_open, &attr, 0 /* pid */,
+		      -1 /* cpu 0 */, -1 /* group id */, 0 /* flags */);
+	if (CHECK(pfd < 0, "perf_event_open", "failed\n"))
+		return;
+
+	err = bpf_map_update_elem(mfd, &key, &pfd, BPF_ANY);
+	if (CHECK(err < 0, "bpf_map_update_elem", "failed\n"))
+		goto cleanup;
+
+	err = bpf_prog_test_run_opts(bpf_program__fd(prog), &opts);
+	if (CHECK(err < 0, "bpf_prog_test_run_opts", "failed\n"))
+		goto cleanup;
+	if (CHECK(opts.retval != 0, "bpf_perf_event_read_value",
+		  "failed with %d\n", opts.retval))
+		goto cleanup;
+
+	/* closing mfd, prog still holds a reference on map */
+	close(mfd);
+
+	err = bpf_prog_test_run_opts(bpf_program__fd(prog), &opts);
+	if (CHECK(err < 0, "bpf_prog_test_run_opts", "failed\n"))
+		goto cleanup;
+
+	if (has_share_pe) {
+		CHECK(opts.retval != 0, "bpf_perf_event_read_value",
+		      "failed with %d\n", opts.retval);
+	} else {
+		CHECK(opts.retval != -ENOENT, "bpf_perf_event_read_value",
+		      "should have failed with %d, but got %d\n", -ENOENT,
+		      opts.retval);
+	}
+
+cleanup:
+	close(pfd);
+}
+
+void test_perf_event_share(void)
+{
+	struct test_perf_event_share *skel;
+
+	skel = test_perf_event_share__open_and_load();
+	if (CHECK(!skel, "skel_open", "failed to open skeleton\n"))
+		return;
+
+	test_one_map(skel->maps.array_1, skel->progs.read_array_1, false);
+	test_one_map(skel->maps.array_2, skel->progs.read_array_2, true);
+
+	test_perf_event_share__destroy(skel);
+}

tools/testing/selftests/bpf/progs/test_perf_event_share.c

@@ -0,0 +1,44 @@
+// SPDX-License-Identifier: GPL-2.0
+// Copyright (c) 2020 Facebook
+#include "vmlinux.h"
+#include <bpf/bpf_helpers.h>
+#include <bpf/bpf_tracing.h>
+
+struct {
+	__uint(type, BPF_MAP_TYPE_PERF_EVENT_ARRAY);
+	__uint(max_entries, 1);
+	__uint(key_size, sizeof(int));
+	__uint(value_size, sizeof(int));
+} array_1 SEC(".maps");
+
+struct {
+	__uint(type, BPF_MAP_TYPE_PERF_EVENT_ARRAY);
+	__uint(max_entries, 1);
+	__uint(key_size, sizeof(int));
+	__uint(value_size, sizeof(int));
+	__uint(map_flags, BPF_F_SHARE_PE);  /* array_2 has BPF_F_SHARE_PE */
+} array_2 SEC(".maps");
+
+SEC("raw_tp/sched_switch")
+int BPF_PROG(read_array_1)
+{
+	struct bpf_perf_event_value val;
+	long ret;
+
+	ret = bpf_perf_event_read_value(&array_1, 0, &val, sizeof(val));
+	bpf_printk("read_array_1 returns %ld", ret);
+	return ret;
+}
+
+SEC("raw_tp/task_rename")
+int BPF_PROG(read_array_2)
+{
+	struct bpf_perf_event_value val;
+	long ret;
+
+	ret = bpf_perf_event_read_value(&array_2, 0, &val, sizeof(val));
+	bpf_printk("read_array_2 returns %ld", ret);
+	return ret;
+}
+
+char LICENSE[] SEC("license") = "GPL";

travis-ci/managers/debian.sh

@@ -0,0 +1,84 @@
+#!/bin/bash
+
+PHASES=(${@:-SETUP RUN RUN_ASAN CLEANUP})
+DEBIAN_RELEASE="${DEBIAN_RELEASE:-testing}"
+CONT_NAME="${CONT_NAME:-libbpf-debian-$DEBIAN_RELEASE}"
+ENV_VARS="${ENV_VARS:-}"
+DOCKER_RUN="${DOCKER_RUN:-docker run}"
+REPO_ROOT="${REPO_ROOT:-$PWD}"
+ADDITIONAL_DEPS=(clang pkg-config gcc-8)
+CFLAGS="-g -O2 -Werror -Wall"
+
+function info() {
+    echo -e "\033[33;1m$1\033[0m"
+}
+
+function error() {
+    echo -e "\033[31;1m$1\033[0m"
+}
+
+function docker_exec() {
+    docker exec $ENV_VARS -it $CONT_NAME "$@"
+}
+
+set -e
+
+source "$(dirname $0)/travis_wait.bash"
+
+for phase in "${PHASES[@]}"; do
+    case $phase in
+        SETUP)
+            info "Setup phase"
+            info "Using Debian $DEBIAN_RELEASE"
+
+            sudo apt-get -y -o Dpkg::Options::="--force-confnew" install docker-ce
+            docker --version
+
+            docker pull debian:$DEBIAN_RELEASE
+            info "Starting container $CONT_NAME"
+            $DOCKER_RUN -v $REPO_ROOT:/build:rw \
+                        -w /build --privileged=true --name $CONT_NAME \
+                        -dit --net=host debian:$DEBIAN_RELEASE /bin/bash
+            docker_exec bash -c "echo deb-src http://deb.debian.org/debian $DEBIAN_RELEASE main >>/etc/apt/sources.list"
+            docker_exec apt-get -y update
+            docker_exec apt-get -y build-dep libelf-dev
+            docker_exec apt-get -y install libelf-dev
+            docker_exec apt-get -y install "${ADDITIONAL_DEPS[@]}"
+            ;;
+        RUN|RUN_CLANG|RUN_GCC8|RUN_ASAN|RUN_CLANG_ASAN|RUN_GCC8_ASAN)
+            if [[ "$phase" = *"CLANG"* ]]; then
+                ENV_VARS="-e CC=clang -e CXX=clang++"
+                CC="clang"
+            elif [[ "$phase" = *"GCC8"* ]]; then
+                ENV_VARS="-e CC=gcc-8 -e CXX=g++-8"
+                CC="gcc-8"
+            else
+                CFLAGS="${CFLAGS} -Wno-stringop-truncation"
+            fi
+            if [[ "$phase" = *"ASAN"* ]]; then
+                CFLAGS="${CFLAGS} -fsanitize=address,undefined"
+            fi
+            docker_exec mkdir build install
+            docker_exec ${CC:-cc} --version
+            info "build"
+	    docker_exec make -j$((4*$(nproc))) CFLAGS="${CFLAGS}" -C ./src -B OBJDIR=../build
+            info "ldd build/libbpf.so:"
+            docker_exec ldd build/libbpf.so
+            if ! docker_exec ldd build/libbpf.so | grep -q libelf; then
+                error "No reference to libelf.so in libbpf.so!"
+                exit 1
+            fi
+            info "install"
+            docker_exec make -j$((4*$(nproc))) -C src OBJDIR=../build DESTDIR=../install install
+            docker_exec rm -rf build install
+            ;;
+        CLEANUP)
+            info "Cleanup phase"
+            docker stop $CONT_NAME
+            docker rm -f $CONT_NAME
+            ;;
+        *)
+            echo >&2 "Unknown phase '$phase'"
+            exit 1
+    esac
+done

travis-ci/managers/travis_wait.bash

@@ -0,0 +1,61 @@
+# This was borrowed from https://github.com/travis-ci/travis-build/tree/master/lib/travis/build/bash
+# to get around https://github.com/travis-ci/travis-ci/issues/9979. It should probably be removed
+# as soon as Travis CI has started to provide an easy way to export the functions to bash scripts.
+
+travis_jigger() {
+  local cmd_pid="${1}"
+  shift
+  local timeout="${1}"
+  shift
+  local count=0
+
+  echo -e "\\n"
+
+  while [[ "${count}" -lt "${timeout}" ]]; do
+    count="$((count + 1))"
+    echo -ne "Still running (${count} of ${timeout}): ${*}\\r"
+    sleep 60
+  done
+
+  echo -e "\\n${ANSI_RED}Timeout (${timeout} minutes) reached. Terminating \"${*}\"${ANSI_RESET}\\n"
+  kill -9 "${cmd_pid}"
+}
+
+travis_wait() {
+  local timeout="${1}"
+
+  if [[ "${timeout}" =~ ^[0-9]+$ ]]; then
+    shift
+  else
+    timeout=20
+  fi
+
+  local cmd=("${@}")
+  local log_file="travis_wait_${$}.log"
+
+  "${cmd[@]}" &>"${log_file}" &
+  local cmd_pid="${!}"
+
+  travis_jigger "${!}" "${timeout}" "${cmd[@]}" &
+  local jigger_pid="${!}"
+  local result
+
+  {
+    set +e
+    wait "${cmd_pid}" 2>/dev/null
+    result="${?}"
+    ps -p"${jigger_pid}" &>/dev/null && kill "${jigger_pid}"
+    set -e
+  }
+
+  if [[ "${result}" -eq 0 ]]; then
+    echo -e "\\n${ANSI_GREEN}The command ${cmd[*]} exited with ${result}.${ANSI_RESET}"
+  else
+    echo -e "\\n${ANSI_RED}The command ${cmd[*]} exited with ${result}.${ANSI_RESET}"
+  fi
+
+  echo -e "\\n${ANSI_GREEN}Log:${ANSI_RESET}\\n"
+  cat "${log_file}"
+
+  return "${result}"
+}