inserting a library(*.so) into a runing processe's adress space

[Andrew Haley]

ranjith kumar wrote:

> I have c file (1.c) and a sharedlibrary file (libmy.so).
> I know that using dlopen() in 1.c we can insert libmy.so into address
> space of 1.c and call any function defined in that library.
> 
> 1)My question is how to insert a library into a running process when
> it doen not contain dlopen() inside its source code???
>  I have little bit information about this. By adjusting heap area of
> the process we can insert a call to dlopen(). I want to know
>   how to access heap area of  a running process and insert a call to
> dlopen(). I know about ptrace() can be used to stop a process and look
> inside its code.
>  Any links or material please.
> Is there any other method to insert a library into a process???

You could certainly do it with ptrace by calling dlopen() in the
context of the process.  To figure out how, read the ptrace docs and
inspect gdb source.

> 2) If a  library is inserted in a process dynamically  how and who
> will replace a function call ie)  jump address(in the context  of
> binary code) with correct address?

ptrace can do this too.

> 3) I know how to produce a lib******.so files.
>     example: gcc -shared -o libmy.so my.o
> But what is the significanc of 'lib' prefix in libmy.so
> I saw some files(.so) without 'lib' prefix(for example  my.so).
>  How to produce such files and what is the difference between libmy.so
> and my.so??

It's just a naming convention that the tools use.

> 4) I produced an executable by
>      gcc -o explample1 1.o lib.o
>    and another by
>  gccc -o exmple2 1.o libmy.so (libmy.so is made from lib.o)
> Which will run faster example1 or example2? why?

It's generally faster not to use shared libraries.

Your concerns cross a number of groups.  binutils can help you for
some of this.

Andrew.