bpf: show real jited prog address in /proc/kallsyms

Currently, /proc/kallsyms shows page address of jited bpf program. The
main reason here is to not expose randomized start address. However,
This is not ideal for detailed profiling (find hot instructions from
stack traces). This patch replaces the page address with real prog start
address.

This change is OK because these addresses are still protected by sysctl
kptr_restrict (see kallsyms_show_value()), and only programs loaded by
root are added to kallsyms (see bpf_prog_kallsyms_add()).

Signed-off-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>

diff --git a/kernel/bpf/core.c b/kernel/bpf/core.c
index 6377225b208204c1c2d8829a778f50ebaa7d816d..1a796e0799ec4a524aee5c325734f0ab22a410ac 100644 (file)
--- a/kernel/bpf/core.c
+++ b/kernel/bpf/core.c
@@ -553,7 +553,6 @@ bool is_bpf_text_address(unsigned long addr)
 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
                    char *sym)
 {
-       unsigned long symbol_start, symbol_end;
        struct bpf_prog_aux *aux;
        unsigned int it = 0;
        int ret = -ERANGE;
@@ -566,10 +565,9 @@ int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
                if (it++ != symnum)
                        continue;
 
-               bpf_get_prog_addr_region(aux->prog, &symbol_start, &symbol_end);
                bpf_get_prog_name(aux->prog, sym);
 
-               *value = symbol_start;
+               *value = (unsigned long)aux->prog->bpf_func;
                *type  = BPF_SYM_ELF_TYPE;
 
                ret = 0;