bpf: Sharing bpf runtime stats with BPF_ENABLE_STATS

[upstream commit d46edd671a147032e22cfeb271a5734703093649]

Currently, sysctl kernel.bpf_stats_enabled controls BPF runtime stats.
Typical userspace tools use kernel.bpf_stats_enabled as follows:

  1. Enable kernel.bpf_stats_enabled;
  2. Check program run_time_ns;
  3. Sleep for the monitoring period;
  4. Check program run_time_ns again, calculate the difference;
  5. Disable kernel.bpf_stats_enabled.

The problem with this approach is that only one userspace tool can toggle
this sysctl. If multiple tools toggle the sysctl at the same time, the
measurement may be inaccurate.

To fix this problem while keep backward compatibility, introduce a new
bpf command BPF_ENABLE_STATS. On success, this command enables stats and
returns a valid fd. BPF_ENABLE_STATS takes argument "type". Currently,
only one type, BPF_STATS_RUN_TIME, is supported. We can extend the
command to support other types of stats in the future.

With BPF_ENABLE_STATS, user space tool would have the following flow:

  1. Get a fd with BPF_ENABLE_STATS, and make sure it is valid;
  2. Check program run_time_ns;
  3. Sleep for the monitoring period;
  4. Check program run_time_ns again, calculate the difference;
  5. Close the fd.

Signed-off-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200430071506.1408910-2-songliubraving@fb.com

include/linux/bpf.h

@@ -956,6 +956,7 @@ _out:							\
 
 #ifdef CONFIG_BPF_SYSCALL
 DECLARE_PER_CPU(int, bpf_prog_active);
+extern struct mutex bpf_stats_enabled_mutex;
 
 extern const struct file_operations bpf_map_fops;
 extern const struct file_operations bpf_prog_fops;

include/uapi/linux/bpf.h

@@ -406,6 +406,12 @@ enum {
 /* flags for BPF_PROG_QUERY */
 #define BPF_F_QUERY_EFFECTIVE	(1U << 0)
 
+/* type for BPF_ENABLE_STATS */
+enum bpf_stats_type {
+	/* enabled run_time_ns and run_cnt */
+	BPF_STATS_RUN_TIME = 0,
+};
+
 enum bpf_stack_build_id_status {
 	/* user space need an empty entry to identify end of a trace */
 	BPF_STACK_BUILD_ID_EMPTY = 0,
@@ -600,6 +606,10 @@ union bpf_attr {
 		__u32		old_prog_fd;
 	} link_update;
 
+	struct { /* struct used by BPF_ENABLE_STATS command */
+		__u32		type;
+	} enable_stats;
+
 	struct { /* struct used by BPF_ITER_CREATE command */
 		__u32		link_fd;
 		__u32		flags;

kernel/bpf/syscall.c

@@ -3617,6 +3617,60 @@ static int bpf_link_get_fd_by_id(const union bpf_attr *attr)
 	return fd;
 }
 
+DEFINE_MUTEX(bpf_stats_enabled_mutex);
+
+static int bpf_stats_release(struct inode *inode, struct file *file)
+{
+	mutex_lock(&bpf_stats_enabled_mutex);
+	static_key_slow_dec(&bpf_stats_enabled_key.key);
+	mutex_unlock(&bpf_stats_enabled_mutex);
+	return 0;
+}
+
+static const struct file_operations bpf_stats_fops = {
+	.release = bpf_stats_release,
+};
+
+static int bpf_enable_runtime_stats(void)
+{
+	int fd;
+
+	mutex_lock(&bpf_stats_enabled_mutex);
+
+	/* Set a very high limit to avoid overflow */
+	if (static_key_count(&bpf_stats_enabled_key.key) > INT_MAX / 2) {
+		mutex_unlock(&bpf_stats_enabled_mutex);
+		return -EBUSY;
+	}
+
+	fd = anon_inode_getfd("bpf-stats", &bpf_stats_fops, NULL, O_CLOEXEC);
+	if (fd >= 0)
+		static_key_slow_inc(&bpf_stats_enabled_key.key);
+
+	mutex_unlock(&bpf_stats_enabled_mutex);
+	return fd;
+}
+
+#define BPF_ENABLE_STATS_LAST_FIELD enable_stats.type
+
+static int bpf_enable_stats(union bpf_attr *attr)
+{
+
+	if (CHECK_ATTR(BPF_ENABLE_STATS))
+		return -EINVAL;
+
+	if (!capable(CAP_SYS_ADMIN))
+		return -EPERM;
+
+	switch (attr->enable_stats.type) {
+	case BPF_STATS_RUN_TIME:
+		return bpf_enable_runtime_stats();
+	default:
+		break;
+	}
+	return -EINVAL;
+}
+
 #define BPF_ITER_CREATE_LAST_FIELD iter_create.flags
 
 static int bpf_iter_create(union bpf_attr *attr)
@@ -3751,6 +3805,9 @@ SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, siz
 		err = bpf_obj_get_next_id(&attr, uattr,
 					  &link_idr, &link_idr_lock);
 		break;
+	case BPF_ENABLE_STATS:
+		err = bpf_enable_stats(&attr);
+		break;
 	case BPF_ITER_CREATE:
 		err = bpf_iter_create(&attr);
 		break;

kernel/sysctl.c

@@ -368,6 +368,41 @@ static int min_extfrag_threshold;
 static int max_extfrag_threshold = 1000;
 #endif
 
+
+#ifdef CONFIG_BPF_SYSCALL
+static int bpf_stats_handler(struct ctl_table *table, int write,
+			     void __user *buffer, size_t *lenp,
+			     loff_t *ppos)
+{
+	struct static_key *key = (struct static_key *)table->data;
+	static int saved_val;
+	int val, ret;
+	struct ctl_table tmp = {
+		.data   = &val,
+		.maxlen = sizeof(val),
+		.mode   = table->mode,
+		.extra1 = SYSCTL_ZERO,
+		.extra2 = SYSCTL_ONE,
+	};
+
+	if (write && !capable(CAP_SYS_ADMIN))
+		return -EPERM;
+
+	mutex_lock(&bpf_stats_enabled_mutex);
+	val = saved_val;
+	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
+	if (write && !ret && val != saved_val) {
+		if (val)
+			static_key_slow_inc(key);
+		else
+			static_key_slow_dec(key);
+		saved_val = val;
+	}
+	mutex_unlock(&bpf_stats_enabled_mutex);
+	return ret;
+}
+#endif
+
 static struct ctl_table kern_table[] = {
 	{
 		.procname       = "container_cpuquota_aware",
@@ -1442,7 +1477,7 @@ static struct ctl_table kern_table[] = {
 		.data		= &bpf_stats_enabled_key.key,
 		.maxlen		= sizeof(bpf_stats_enabled_key),
 		.mode		= 0644,
-		.proc_handler	= proc_do_static_key,
+		.proc_handler	= bpf_stats_handler,
 	},
 #endif
 #if defined(CONFIG_TREE_RCU) || defined(CONFIG_PREEMPT_RCU)