Extracting only my function names from ELF binary

Question

Im writing a script for extracting all the functions (written by user) in a binary.

The following shell script extracts my function names as well as some library functions which start with __.

readelf -s ./a.out | gawk '
{ 
  if ($4 == "FUNC" && $3 != "0" && $7 == "13" && $8 != "main") { 
    print "b " $NF; //***Updated
  } 
}' &> function_names; 

Output of function_names file:

b __libc_csu_fini
b PrintDivider    
b PrintFooter    
b __libc_csu_init    
b PrintHeader

I would like to extract only my functions. So how to check whether function name starts with __ or else any other alternatives also highly appreciated.

Update::

@djf solution works fine. What if .c files which are compiled also may contain a function which starts with __? In that case, how to differentiate?

Recto

Answer 1

What about using readelf on your object file(s) instead of the linked executable? Then there's no spam from the library functions. Use the -c flag to compile to an object file and not link immediately.

PS: The proper tool to extract names from an executable or object file is nm, not readelf. Using nm -P file has everything you want.

$ nm -P tst.o | awk '$2 == "T" {print "b " $1}'
b foo
b main

EDIT: To ignore main and symbols starting with an underscore, use

$ nm -P a.out | awk '$2 == "T" && $1 !~ /^_/ && $1 != "main" {print "b " $1}'

Vampirism

nm also gives _init , _start, _fini – Recto

@Recto No it doesn't if you use it on the object files instead of the executable. But see my edit. – Vampirism

nm -P tst.o | awk '$2 == "T" && $1 != "main" {print "b " $1}' is exactly wanted.. :) – Recto

Answer 2

You could add a regex check to make sure that the function name starts with a letter.

I presume that $8 contains the function name:

readelf -s ./a.out | gawk '
{
  if($4 == "FUNC" && $3 != "0" && $7 == "13" && $8 != "main" && $8~/^[[:alpha:]]/) {
    print $NF;
  }
}'

Freighter

If user defined function also starts with __, then how to differntiate ??? any idea?? pls share. – Recto

@Recto it's really bad practice to name your functions with leading underscores since these names are reserved for the compiler and core libraries... avoid doing that if you can - in C++ its forbidden by the language standard. I'm not sure about C, though. I can't tell you how to differentiate because you haven't told us what your function names look like... C-libraries usually add a prefix to all function names to avoid name clashes with other libraries. Maybe you could filter on such a prefix? – Freighter

Answer 3

Pipe it through grep ^[^_]. [30 char]

Collard

Thanks for your response. Can u please explain bit more ?? – Recto

@Recto Explain what? You pipe the output through this command and BOOM you get what you want. – Collard

@Recto Now has your question changed? Or what? I don't understand. For me, it's still the same: grep for the names which don't begin by an underscore. – Collard

b is breakpoint command for gdb. so b <space> <function_name>. I want to use this file for gdb as input --command option. – Recto

@Recto Non sequitur. What is the problem now? – Collard

Any particular reason for the .*$ part in the regex? – Vampirism

@Vampirism Yes, unless I include that, the answer is 28 chars long and SO doesn't let me submit it. – Collard

Ah, then I would have single quoted the regex to avoid the shell from globbing ^[^_] in case there's a file named ^x around... :-) – Vampirism