Merge branch 'bpf-tracing-multiprog-tp-query'

Yonghong Song says:

====================
Commit e87c6bc ("bpf: permit multiple bpf attachments
for a single perf event") added support to attach multiple
bpf programs to a single perf event. Given a perf event
(kprobe, uprobe, or kernel tracepoint), the perf ioctl interface
is used to query bpf programs attached to the same trace event.

There already exists a BPF_PROG_QUERY command for introspection
currently used by cgroup+bpf. We did have an implementation for
querying tracepoint+bpf through the same interface. However, it
looks cleaner to use ioctl() style of api here, since attaching
bpf prog to tracepoint/kuprobe is also done via ioctl.

Patch #1 had the core implementation and patch #2 added
a test case in tools bpf selftests suite.

Changelogs:
v3 -> v4:
  - Fix a compilation error with newer gcc like 6.3.1 while
    old gcc 4.8.5 is okay. I was using &uquery->ids to represent
    the address to the ids array to make it explicit that the
    address is passed, and this syntax is rightly rejected
    by gcc 6.3.1.
v2 -> v3:
  - Change uapi structure perf_event_query_bpf to be more
    clearer based on Peter's suggestion, and adjust
    other codes accordingly.
v1 -> v2:
  - Rebase on top of net-next.
  - Use existing bpf_prog_array_length function instead of
    implementing the same functionality in function
    bpf_prog_array_copy_info.
====================

Signed-off-by: Alexei Starovoitov <[email protected]>

include/linux/bpf.h

@@ -254,6 +254,7 @@ typedef unsigned long (*bpf_ctx_copy_t)(void *dst, const void *src,
 
 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
 		     void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy);
+int bpf_event_query_prog_array(struct perf_event *event, void __user *info);
 
 int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr,
 			  union bpf_attr __user *uattr);
@@ -285,6 +286,9 @@ int bpf_prog_array_copy_to_user(struct bpf_prog_array __rcu *progs,
 
 void bpf_prog_array_delete_safe(struct bpf_prog_array __rcu *progs,
 				struct bpf_prog *old_prog);
+int bpf_prog_array_copy_info(struct bpf_prog_array __rcu *array,
+			     __u32 __user *prog_ids, u32 request_cnt,
+			     __u32 __user *prog_cnt);
 int bpf_prog_array_copy(struct bpf_prog_array __rcu *old_array,
 			struct bpf_prog *exclude_prog,
 			struct bpf_prog *include_prog,

include/uapi/linux/perf_event.h

@@ -418,6 +418,27 @@ struct perf_event_attr {
 	__u16	__reserved_2;	/* align to __u64 */
 };
 
+/*
+ * Structure used by below PERF_EVENT_IOC_QUERY_BPF command
+ * to query bpf programs attached to the same perf tracepoint
+ * as the given perf event.
+ */
+struct perf_event_query_bpf {
+	/*
+	 * The below ids array length
+	 */
+	__u32	ids_len;
+	/*
+	 * Set by the kernel to indicate the number of
+	 * available programs
+	 */
+	__u32	prog_cnt;
+	/*
+	 * User provided buffer to store program ids
+	 */
+	__u32	ids[0];
+};
+
 #define perf_flags(attr)	(*(&(attr)->read_format + 1))
 
 /*
@@ -433,6 +454,7 @@ struct perf_event_attr {
 #define PERF_EVENT_IOC_ID		_IOR('$', 7, __u64 *)
 #define PERF_EVENT_IOC_SET_BPF		_IOW('$', 8, __u32)
 #define PERF_EVENT_IOC_PAUSE_OUTPUT	_IOW('$', 9, __u32)
+#define PERF_EVENT_IOC_QUERY_BPF	_IOWR('$', 10, struct perf_event_query_bpf *)
 
 enum perf_event_ioc_flags {
 	PERF_IOC_FLAG_GROUP		= 1U << 0,

kernel/bpf/core.c

@@ -1462,6 +1462,8 @@ int bpf_prog_array_copy_to_user(struct bpf_prog_array __rcu *progs,
 	rcu_read_lock();
 	prog = rcu_dereference(progs)->progs;
 	for (; *prog; prog++) {
+		if (*prog == &dummy_bpf_prog.prog)
+			continue;
 		id = (*prog)->aux->id;
 		if (copy_to_user(prog_ids + i, &id, sizeof(id))) {
 			rcu_read_unlock();
@@ -1545,6 +1547,25 @@ int bpf_prog_array_copy(struct bpf_prog_array __rcu *old_array,
 	return 0;
 }
 
+int bpf_prog_array_copy_info(struct bpf_prog_array __rcu *array,
+			     __u32 __user *prog_ids, u32 request_cnt,
+			     __u32 __user *prog_cnt)
+{
+	u32 cnt = 0;
+
+	if (array)
+		cnt = bpf_prog_array_length(array);
+
+	if (copy_to_user(prog_cnt, &cnt, sizeof(cnt)))
+		return -EFAULT;
+
+	/* return early if user requested only program count or nothing to copy */
+	if (!request_cnt || !cnt)
+		return 0;
+
+	return bpf_prog_array_copy_to_user(array, prog_ids, request_cnt);
+}
+
 static void bpf_prog_free_deferred(struct work_struct *work)
 {
 	struct bpf_prog_aux *aux;

kernel/events/core.c

@@ -4723,6 +4723,9 @@ static long _perf_ioctl(struct perf_event *event, unsigned int cmd, unsigned lon
 		rcu_read_unlock();
 		return 0;
 	}
+
+	case PERF_EVENT_IOC_QUERY_BPF:
+		return bpf_event_query_prog_array(event, (void __user *)arg);
 	default:
 		return -ENOTTY;
 	}

kernel/trace/bpf_trace.c

@@ -820,3 +820,26 @@ void perf_event_detach_bpf_prog(struct perf_event *event)
 unlock:
 	mutex_unlock(&bpf_event_mutex);
 }
+
+int bpf_event_query_prog_array(struct perf_event *event, void __user *info)
+{
+	struct perf_event_query_bpf __user *uquery = info;
+	struct perf_event_query_bpf query = {};
+	int ret;
+
+	if (!capable(CAP_SYS_ADMIN))
+		return -EPERM;
+	if (event->attr.type != PERF_TYPE_TRACEPOINT)
+		return -EINVAL;
+	if (copy_from_user(&query, uquery, sizeof(query)))
+		return -EFAULT;
+
+	mutex_lock(&bpf_event_mutex);
+	ret = bpf_prog_array_copy_info(event->tp_event->prog_array,
+				       uquery->ids,
+				       query.ids_len,
+				       &uquery->prog_cnt);
+	mutex_unlock(&bpf_event_mutex);
+
+	return ret;
+}

tools/include/uapi/linux/perf_event.h

@@ -418,6 +418,27 @@ struct perf_event_attr {
 	__u16	__reserved_2;	/* align to __u64 */
 };
 
+/*
+ * Structure used by below PERF_EVENT_IOC_QUERY_BPF command
+ * to query bpf programs attached to the same perf tracepoint
+ * as the given perf event.
+ */
+struct perf_event_query_bpf {
+	/*
+	 * The below ids array length
+	 */
+	__u32	ids_len;
+	/*
+	 * Set by the kernel to indicate the number of
+	 * available programs
+	 */
+	__u32	prog_cnt;
+	/*
+	 * User provided buffer to store program ids
+	 */
+	__u32	ids[0];
+};
+
 #define perf_flags(attr)	(*(&(attr)->read_format + 1))
 
 /*
@@ -433,6 +454,7 @@ struct perf_event_attr {
 #define PERF_EVENT_IOC_ID		_IOR('$', 7, __u64 *)
 #define PERF_EVENT_IOC_SET_BPF		_IOW('$', 8, __u32)
 #define PERF_EVENT_IOC_PAUSE_OUTPUT	_IOW('$', 9, __u32)
+#define PERF_EVENT_IOC_QUERY_BPF	_IOWR('$', 10, struct perf_event_query_bpf *)
 
 enum perf_event_ioc_flags {
 	PERF_IOC_FLAG_GROUP		= 1U << 0,

tools/testing/selftests/bpf/Makefile

@@ -17,7 +17,7 @@ TEST_GEN_PROGS = test_verifier test_tag test_maps test_lru_map test_lpm_map test
 
 TEST_GEN_FILES = test_pkt_access.o test_xdp.o test_l4lb.o test_tcp_estats.o test_obj_id.o \
 	test_pkt_md_access.o test_xdp_redirect.o test_xdp_meta.o sockmap_parse_prog.o     \
-	sockmap_verdict_prog.o dev_cgroup.o sample_ret0.o
+	sockmap_verdict_prog.o dev_cgroup.o sample_ret0.o test_tracepoint.o
 
 TEST_PROGS := test_kmod.sh test_xdp_redirect.sh test_xdp_meta.sh \
 	test_offload.py

tools/testing/selftests/bpf/test_progs.c

@@ -21,8 +21,10 @@ typedef __u16 __sum16;
 #include <linux/ipv6.h>
 #include <linux/tcp.h>
 #include <linux/filter.h>
+#include <linux/perf_event.h>
 #include <linux/unistd.h>
 
+#include <sys/ioctl.h>
 #include <sys/wait.h>
 #include <sys/resource.h>
 #include <sys/types.h>
@@ -617,6 +619,136 @@ static void test_obj_name(void)
 	}
 }
 
+static void test_tp_attach_query(void)
+{
+	const int num_progs = 3;
+	int i, j, bytes, efd, err, prog_fd[num_progs], pmu_fd[num_progs];
+	__u32 duration = 0, info_len, saved_prog_ids[num_progs];
+	const char *file = "./test_tracepoint.o";
+	struct perf_event_query_bpf *query;
+	struct perf_event_attr attr = {};
+	struct bpf_object *obj[num_progs];
+	struct bpf_prog_info prog_info;
+	char buf[256];
+
+	snprintf(buf, sizeof(buf),
+		 "/sys/kernel/debug/tracing/events/sched/sched_switch/id");
+	efd = open(buf, O_RDONLY, 0);
+	if (CHECK(efd < 0, "open", "err %d errno %d\n", efd, errno))
+		return;
+	bytes = read(efd, buf, sizeof(buf));
+	close(efd);
+	if (CHECK(bytes <= 0 || bytes >= sizeof(buf),
+		  "read", "bytes %d errno %d\n", bytes, errno))
+		return;
+
+	attr.config = strtol(buf, NULL, 0);
+	attr.type = PERF_TYPE_TRACEPOINT;
+	attr.sample_type = PERF_SAMPLE_RAW | PERF_SAMPLE_CALLCHAIN;
+	attr.sample_period = 1;
+	attr.wakeup_events = 1;
+
+	query = malloc(sizeof(*query) + sizeof(__u32) * num_progs);
+	for (i = 0; i < num_progs; i++) {
+		err = bpf_prog_load(file, BPF_PROG_TYPE_TRACEPOINT, &obj[i],
+				    &prog_fd[i]);
+		if (CHECK(err, "prog_load", "err %d errno %d\n", err, errno))
+			goto cleanup1;
+
+		bzero(&prog_info, sizeof(prog_info));
+		prog_info.jited_prog_len = 0;
+		prog_info.xlated_prog_len = 0;
+		prog_info.nr_map_ids = 0;
+		info_len = sizeof(prog_info);
+		err = bpf_obj_get_info_by_fd(prog_fd[i], &prog_info, &info_len);
+		if (CHECK(err, "bpf_obj_get_info_by_fd", "err %d errno %d\n",
+			  err, errno))
+			goto cleanup1;
+		saved_prog_ids[i] = prog_info.id;
+
+		pmu_fd[i] = syscall(__NR_perf_event_open, &attr, -1 /* pid */,
+				    0 /* cpu 0 */, -1 /* group id */,
+				    0 /* flags */);
+		if (CHECK(pmu_fd[i] < 0, "perf_event_open", "err %d errno %d\n",
+			  pmu_fd[i], errno))
+			goto cleanup2;
+		err = ioctl(pmu_fd[i], PERF_EVENT_IOC_ENABLE, 0);
+		if (CHECK(err, "perf_event_ioc_enable", "err %d errno %d\n",
+			  err, errno))
+			goto cleanup3;
+
+		if (i == 0) {
+			/* check NULL prog array query */
+			query->ids_len = num_progs;
+			err = ioctl(pmu_fd[i], PERF_EVENT_IOC_QUERY_BPF, query);
+			if (CHECK(err || query->prog_cnt != 0,
+				  "perf_event_ioc_query_bpf",
+				  "err %d errno %d query->prog_cnt %u\n",
+				  err, errno, query->prog_cnt))
+				goto cleanup3;
+		}
+
+		err = ioctl(pmu_fd[i], PERF_EVENT_IOC_SET_BPF, prog_fd[i]);
+		if (CHECK(err, "perf_event_ioc_set_bpf", "err %d errno %d\n",
+			  err, errno))
+			goto cleanup3;
+
+		if (i == 1) {
+			/* try to get # of programs only */
+			query->ids_len = 0;
+			err = ioctl(pmu_fd[i], PERF_EVENT_IOC_QUERY_BPF, query);
+			if (CHECK(err || query->prog_cnt != 2,
+				  "perf_event_ioc_query_bpf",
+				  "err %d errno %d query->prog_cnt %u\n",
+				  err, errno, query->prog_cnt))
+				goto cleanup3;
+
+			/* try a few negative tests */
+			/* invalid query pointer */
+			err = ioctl(pmu_fd[i], PERF_EVENT_IOC_QUERY_BPF,
+				    (struct perf_event_query_bpf *)0x1);
+			if (CHECK(!err || errno != EFAULT,
+				  "perf_event_ioc_query_bpf",
+				  "err %d errno %d\n", err, errno))
+				goto cleanup3;
+
+			/* no enough space */
+			query->ids_len = 1;
+			err = ioctl(pmu_fd[i], PERF_EVENT_IOC_QUERY_BPF, query);
+			if (CHECK(!err || errno != ENOSPC || query->prog_cnt != 2,
+				  "perf_event_ioc_query_bpf",
+				  "err %d errno %d query->prog_cnt %u\n",
+				  err, errno, query->prog_cnt))
+				goto cleanup3;
+		}
+
+		query->ids_len = num_progs;
+		err = ioctl(pmu_fd[i], PERF_EVENT_IOC_QUERY_BPF, query);
+		if (CHECK(err || query->prog_cnt != (i + 1),
+			  "perf_event_ioc_query_bpf",
+			  "err %d errno %d query->prog_cnt %u\n",
+			  err, errno, query->prog_cnt))
+			goto cleanup3;
+		for (j = 0; j < i + 1; j++)
+			if (CHECK(saved_prog_ids[j] != query->ids[j],
+				  "perf_event_ioc_query_bpf",
+				  "#%d saved_prog_id %x query prog_id %x\n",
+				  j, saved_prog_ids[j], query->ids[j]))
+				goto cleanup3;
+	}
+
+	i = num_progs - 1;
+	for (; i >= 0; i--) {
+ cleanup3:
+		ioctl(pmu_fd[i], PERF_EVENT_IOC_DISABLE);
+ cleanup2:
+		close(pmu_fd[i]);
+ cleanup1:
+		bpf_object__close(obj[i]);
+	}
+	free(query);
+}
+
 int main(void)
 {
 	struct rlimit rinf = { RLIM_INFINITY, RLIM_INFINITY };
@@ -630,6 +762,7 @@ int main(void)
 	test_bpf_obj_id();
 	test_pkt_md_access();
 	test_obj_name();
+	test_tp_attach_query();
 
 	printf("Summary: %d PASSED, %d FAILED\n", pass_cnt, error_cnt);
 	return error_cnt ? EXIT_FAILURE : EXIT_SUCCESS;

tools/testing/selftests/bpf/test_tracepoint.c

@@ -0,0 +1,26 @@
+// SPDX-License-Identifier: GPL-2.0
+// Copyright (c) 2017 Facebook
+
+#include <linux/bpf.h>
+#include "bpf_helpers.h"
+
+/* taken from /sys/kernel/debug/tracing/events/sched/sched_switch/format */
+struct sched_switch_args {
+	unsigned long long pad;
+	char prev_comm[16];
+	int prev_pid;
+	int prev_prio;
+	long long prev_state;
+	char next_comm[16];
+	int next_pid;
+	int next_prio;
+};
+
+SEC("tracepoint/sched/sched_switch")
+int oncpu(struct sched_switch_args *ctx)
+{
+	return 0;
+}
+
+char _license[] SEC("license") = "GPL";
+__u32 _version SEC("version") = 1; /* ignored by tracepoints, required by libbpf.a */