Merge branch 'bpf_enable_stats'

Song Liu says:

====================
run_time_ns is a useful stats for BPF programs. However, it is gated by
sysctl kernel.bpf_stats_enabled. When multiple user space tools are
toggling kernl.bpf_stats_enabled at the same time, they may confuse each
other.

Solve this problem with a new BPF command BPF_ENABLE_STATS.

Changes v8 => v9:
  1. Clean up in selftest (Andrii).
  2. Not using static variable in test program (Andrii).

Changes v7 => v8:
  1. Change name BPF_STATS_RUNTIME_CNT => BPF_STATS_RUN_TIME (Alexei).
  2. Add CHECK_ATTR to bpf_enable_stats() (Alexei).
  3. Rebase (Andrii).
  4. Simplfy the selftest (Alexei).

Changes v6 => v7:
  1. Add test to verify run_cnt matches count measured by the program.

Changes v5 => v6:
  1. Simplify test program (Yonghong).
  2. Rebase (with some conflicts).

Changes v4 => v5:
  1. Use memset to zero bpf_attr in bpf_enable_stats() (Andrii).

Changes v3 => v4:
  1. Add libbpf support and selftest;
  2. Avoid cleaning trailing space.

Changes v2 => v3:
  1. Rename the command to BPF_ENABLE_STATS, and make it extendible.
  2. fix commit log;
  3. remove unnecessary headers.
====================

Signed-off-by: Alexei Starovoitov <ast@kernel.org>

1 files changed, 57 insertions, 0 deletions

diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c
index c75b2dd2459c..4f34eecec9ce 100644
--- a/kernel/bpf/syscall.c
+++ b/kernel/bpf/syscall.c
@@ -3872,6 +3872,60 @@ 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;
+}
+
 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
 {
 	union bpf_attr attr;
@@ -3996,6 +4050,9 @@ 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;
 	default:
 		err = -EINVAL;
 		break;