Merge branch 'bpf: Add missed stats for kprobes'

Jiri Olsa says:

====================
hi,
at the moment we can't retrieve the number of missed kprobe
executions and subsequent execution of BPF programs.

This patchset adds:
  - counting of missed execution on attach layer for:
    . kprobes attached through perf link (kprobe/ftrace)
    . kprobes attached through kprobe.multi link (fprobe)
  - counting of recursion_misses for BPF kprobe programs

It's still technically possible to create kprobe without perf link (using
SET_BPF perf ioctl) in which case we don't have a way to retrieve the kprobe's
'missed' count. However both libbpf and cilium/ebpf libraries use perf link
if it's available, and for old kernels without perf link support we can use
BPF program to retrieve the kprobe missed count.

v3 changes:
  - added acks [Song]
  - make test_missed not serial [Andrii]

Also available at:
  https://git.kernel.org/pub/scm/linux/kernel/git/jolsa/perf.git
  bpf/missed_stats

thanks,
jirka
====================

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>

3 files changed, 27 insertions, 11 deletions

diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c
index 85c1d908f70f..6b5280f14a53 100644
--- a/kernel/bpf/syscall.c
+++ b/kernel/bpf/syscall.c
@@ -3374,7 +3374,7 @@ static void bpf_perf_link_dealloc(struct bpf_link *link)
 static int bpf_perf_link_fill_common(const struct perf_event *event,
 				     char __user *uname, u32 ulen,
 				     u64 *probe_offset, u64 *probe_addr,
-				     u32 *fd_type)
+				     u32 *fd_type, unsigned long *missed)
 {
 	const char *buf;
 	u32 prog_id;
@@ -3385,7 +3385,7 @@ static int bpf_perf_link_fill_common(const struct perf_event *event,
 		return -EINVAL;
 
 	err = bpf_get_perf_event_info(event, &prog_id, fd_type, &buf,
-				      probe_offset, probe_addr);
+				      probe_offset, probe_addr, missed);
 	if (err)
 		return err;
 	if (!uname)
@@ -3408,6 +3408,7 @@ static int bpf_perf_link_fill_common(const struct perf_event *event,
 static int bpf_perf_link_fill_kprobe(const struct perf_event *event,
 				     struct bpf_link_info *info)
 {
+	unsigned long missed;
 	char __user *uname;
 	u64 addr, offset;
 	u32 ulen, type;
@@ -3416,7 +3417,7 @@ static int bpf_perf_link_fill_kprobe(const struct perf_event *event,
 	uname = u64_to_user_ptr(info->perf_event.kprobe.func_name);
 	ulen = info->perf_event.kprobe.name_len;
 	err = bpf_perf_link_fill_common(event, uname, ulen, &offset, &addr,
-					&type);
+					&type, &missed);
 	if (err)
 		return err;
 	if (type == BPF_FD_TYPE_KRETPROBE)
@@ -3425,6 +3426,7 @@ static int bpf_perf_link_fill_kprobe(const struct perf_event *event,
 		info->perf_event.type = BPF_PERF_EVENT_KPROBE;
 
 	info->perf_event.kprobe.offset = offset;
+	info->perf_event.kprobe.missed = missed;
 	if (!kallsyms_show_value(current_cred()))
 		addr = 0;
 	info->perf_event.kprobe.addr = addr;
@@ -3444,7 +3446,7 @@ static int bpf_perf_link_fill_uprobe(const struct perf_event *event,
 	uname = u64_to_user_ptr(info->perf_event.uprobe.file_name);
 	ulen = info->perf_event.uprobe.name_len;
 	err = bpf_perf_link_fill_common(event, uname, ulen, &offset, &addr,
-					&type);
+					&type, NULL);
 	if (err)
 		return err;
 
@@ -3480,7 +3482,7 @@ static int bpf_perf_link_fill_tracepoint(const struct perf_event *event,
 	uname = u64_to_user_ptr(info->perf_event.tracepoint.tp_name);
 	ulen = info->perf_event.tracepoint.name_len;
 	info->perf_event.type = BPF_PERF_EVENT_TRACEPOINT;
-	return bpf_perf_link_fill_common(event, uname, ulen, NULL, NULL, NULL);
+	return bpf_perf_link_fill_common(event, uname, ulen, NULL, NULL, NULL, NULL);
 }
 
 static int bpf_perf_link_fill_perf_event(const struct perf_event *event,
@@ -4813,7 +4815,7 @@ static int bpf_task_fd_query(const union bpf_attr *attr,
 
 		err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
 					      &buf, &probe_offset,
-					      &probe_addr);
+					      &probe_addr, NULL);
 		if (!err)
 			err = bpf_task_fd_query_copy(attr, uattr, prog_id,
 						     fd_type, buf,
diff --git a/kernel/trace/bpf_trace.c b/kernel/trace/bpf_trace.c
index 868008f56fec..df697c74d519 100644
--- a/kernel/trace/bpf_trace.c
+++ b/kernel/trace/bpf_trace.c
@@ -117,6 +117,9 @@ unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
 		 * and don't send kprobe event into ring-buffer,
 		 * so return zero here
 		 */
+		rcu_read_lock();
+		bpf_prog_inc_misses_counters(rcu_dereference(call->prog_array));
+		rcu_read_unlock();
 		ret = 0;
 		goto out;
 	}
@@ -2384,7 +2387,8 @@ int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
 
 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
 			    u32 *fd_type, const char **buf,
-			    u64 *probe_offset, u64 *probe_addr)
+			    u64 *probe_offset, u64 *probe_addr,
+			    unsigned long *missed)
 {
 	bool is_tracepoint, is_syscall_tp;
 	struct bpf_prog *prog;
@@ -2419,7 +2423,7 @@ int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
 #ifdef CONFIG_KPROBE_EVENTS
 		if (flags & TRACE_EVENT_FL_KPROBE)
 			err = bpf_get_kprobe_info(event, fd_type, buf,
-						  probe_offset, probe_addr,
+						  probe_offset, probe_addr, missed,
 						  event->attr.type == PERF_TYPE_TRACEPOINT);
 #endif
 #ifdef CONFIG_UPROBE_EVENTS
@@ -2614,6 +2618,7 @@ static int bpf_kprobe_multi_link_fill_link_info(const struct bpf_link *link,
 	kmulti_link = container_of(link, struct bpf_kprobe_multi_link, link);
 	info->kprobe_multi.count = kmulti_link->cnt;
 	info->kprobe_multi.flags = kmulti_link->flags;
+	info->kprobe_multi.missed = kmulti_link->fp.nmissed;
 
 	if (!uaddrs)
 		return 0;
@@ -2710,6 +2715,7 @@ kprobe_multi_link_prog_run(struct bpf_kprobe_multi_link *link,
 	int err;
 
 	if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
+		bpf_prog_inc_misses_counter(link->link.prog);
 		err = 0;
 		goto out;
 	}
diff --git a/kernel/trace/trace_kprobe.c b/kernel/trace/trace_kprobe.c
index 3d7a180a8427..961a78ffd6d2 100644
--- a/kernel/trace/trace_kprobe.c
+++ b/kernel/trace/trace_kprobe.c
@@ -1189,6 +1189,12 @@ static const struct file_operations kprobe_events_ops = {
 	.write		= probes_write,
 };
 
+static unsigned long trace_kprobe_missed(struct trace_kprobe *tk)
+{
+	return trace_kprobe_is_return(tk) ?
+		tk->rp.kp.nmissed + tk->rp.nmissed : tk->rp.kp.nmissed;
+}
+
 /* Probes profiling interfaces */
 static int probes_profile_seq_show(struct seq_file *m, void *v)
 {
@@ -1200,8 +1206,7 @@ static int probes_profile_seq_show(struct seq_file *m, void *v)
 		return 0;
 
 	tk = to_trace_kprobe(ev);
-	nmissed = trace_kprobe_is_return(tk) ?
-		tk->rp.kp.nmissed + tk->rp.nmissed : tk->rp.kp.nmissed;
+	nmissed = trace_kprobe_missed(tk);
 	seq_printf(m, "  %-44s %15lu %15lu\n",
 		   trace_probe_name(&tk->tp),
 		   trace_kprobe_nhit(tk),
@@ -1547,7 +1552,8 @@ NOKPROBE_SYMBOL(kretprobe_perf_func);
 
 int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
 			const char **symbol, u64 *probe_offset,
-			u64 *probe_addr, bool perf_type_tracepoint)
+			u64 *probe_addr, unsigned long *missed,
+			bool perf_type_tracepoint)
 {
 	const char *pevent = trace_event_name(event->tp_event);
 	const char *group = event->tp_event->class->system;
@@ -1566,6 +1572,8 @@ int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
 	*probe_addr = kallsyms_show_value(current_cred()) ?
 		      (unsigned long)tk->rp.kp.addr : 0;
 	*symbol = tk->symbol;
+	if (missed)
+		*missed = trace_kprobe_missed(tk);
 	return 0;
 }
 #endif	/* CONFIG_PERF_EVENTS */