tracing/probes: Support BTF argument on module functions
authorMasami Hiramatsu (Google) <mhiramat@kernel.org>
Tue, 22 Aug 2023 16:25:42 +0000 (01:25 +0900)
committerMasami Hiramatsu (Google) <mhiramat@kernel.org>
Wed, 23 Aug 2023 00:39:15 +0000 (09:39 +0900)
commitb1d1e90490b671444ebf66292201572c1059d323
tree592078e0509a6d9ffa75586e1a78c1a3d8a8feff
parentf8bbf8b9901a09915d7b466ceb2bccc1fb5abea8
tracing/probes: Support BTF argument on module functions

Since the btf returned from bpf_get_btf_vmlinux() only covers functions in
the vmlinux, BTF argument is not available on the functions in the modules.
Use bpf_find_btf_id() instead of bpf_get_btf_vmlinux()+btf_find_name_kind()
so that BTF argument can find the correct struct btf and btf_type in it.
With this fix, fprobe events can use `$arg*` on module functions as below

 # grep nf_log_ip_packet /proc/kallsyms
ffffffffa0005c00 t nf_log_ip_packet [nf_log_syslog]
ffffffffa0005bf0 t __pfx_nf_log_ip_packet [nf_log_syslog]
 # echo 'f nf_log_ip_packet $arg*' > dynamic_events
 # cat dynamic_events
f:fprobes/nf_log_ip_packet__entry nf_log_ip_packet net=net pf=pf hooknum=hooknum skb=skb in=in out=out loginfo=loginfo prefix=prefix

To support the module's btf which is removable, the struct btf needs to be
ref-counted. So this also records the btf in the traceprobe_parse_context
and returns the refcount when the parse has done.

Link: https://lore.kernel.org/all/169272154223.160970.3507930084247934031.stgit@devnote2/
Suggested-by: Alexei Starovoitov <alexei.starovoitov@gmail.com>
Signed-off-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Acked-by: Steven Rostedt (Google) <rostedt@goodmis.org>
include/linux/btf.h
kernel/bpf/btf.c
kernel/trace/trace_eprobe.c
kernel/trace/trace_fprobe.c
kernel/trace/trace_kprobe.c
kernel/trace/trace_probe.c
kernel/trace/trace_probe.h
kernel/trace/trace_uprobe.c