Re: [PATCH bpf-next v3 2/7] bpf: introduce bpf subcommand BPF_TASK_FD_QUERY

On 5/23/18 2:04 PM, Alexei Starovoitov wrote:

On Wed, May 23, 2018 at 10:13:22AM -0700, Martin KaFai Lau wrote:

quoted

quoted

+		__u32		prog_id;	/* output: prod_id */
+		__u32		attach_info;	/* output: BPF_ATTACH_* */
+		__u64		probe_offset;	/* output: probe_offset */
+		__u64		probe_addr;	/* output: probe_addr */
+	} task_fd_query;
  } __attribute__((aligned(8)));
  
  /* The description below is an attempt at providing documentation to eBPF

@@ -2458,4 +2475,14 @@ struct bpf_fib_lookup {
  	__u8	dmac[6];     /* ETH_ALEN */
  };
  
+/* used by <task, fd> based query */
+enum {

Nit. Instead of a comment, is it better to give this
enum a descriptive name?

quoted

+	BPF_ATTACH_RAW_TRACEPOINT,	/* tp name */
+	BPF_ATTACH_TRACEPOINT,		/* tp name */
+	BPF_ATTACH_KPROBE,		/* (symbol + offset) or addr */
+	BPF_ATTACH_KRETPROBE,		/* (symbol + offset) or addr */
+	BPF_ATTACH_UPROBE,		/* filename + offset */
+	BPF_ATTACH_URETPROBE,		/* filename + offset */
+};

One more nit here.
Can we come up with better names for the above?
'attach' is a verb. I cannot help but read above as it's an action
for the kernel to attach to something and not the type of event
where the program was attached to.
Since we pass task+fd into that BPF_TASK_FD_QUERY command how
about returning BPF_FD_TYPE_KPROBE, BPF_FD_TYPE_TRACEPOINT, ... ?

Okay will use BPF_FD_TYPE_*... which is indeed better than
BPF_ATTACH_*.