The question is very simple:

How does a C program start in linux? Is main() the first function called in the a c application by kernel. I understand it is the first called function written by the application programmer, but the question is to understand the what all kernel does and what all functions it calls before calling main()

I think you're asking about the executable format and how executables are loaded:

http://en.wikipedia.org/wiki/Executa...inkable_Format
http://www.linuxjournal.com/article/1059
http://wiki.osdev.org/ELF#Loading_ELF_Binaries

There is something called the C startup code. It is executed before main(), and may perform any number of things, and is compiler and architecture dependent. It may do such things as create argc & argv & initialize global variables.
--- rod.

Thank you for the links.
After allocating space for these sections inside memory, how the OS launches main()? can you explain this part?

---------- Post added 03-01-11 at 11:17 PM ----------

I really would like to more know about those things... Any good read you can suggest on this?

The source code for your compiler. Often, the startup code is customizable, and is distributed in source form (often in assembler language), especially for cases like embedded systems, where there may be hardware-specific requirements, such as initializing hardware.

The OS does not call main(), the startup code does. It isn't necessarily the case that there even is an OS. Compilers exist for 'bare metal' architectures.

--- rod.

It puts the address of the first instruction in the program counter.

The OS doesn't launch main(). It is two steps removed from that.

I have an example handy from a recent thread on a related subject.
http://www.linuxquestions.org/questi...8/#post4272950

(For the binary in that thread) the beginning of the output from the readelf -l command is:
From the kernel's point of view, what matters is the line that says: Requesting program interpreter: /lib/ld-linux.so.2
IIUC, that means the kernel actually loads and gives control to that .so file, rather than really starting your executable. That .so file finishes the job of loading your program (finding and loading all other .so files that must be present before your code starts).

Next look at the lines
Entry point 0x80482f0
...
LOAD 0x000000 0x08048000 0x08048000 0x00588 0x00588 R E 0x1000
IIUC, once ld-linux.so has finished loading your program it transfers control to that entry point, which is in that one of the sections loaded from the main executable file.

But that still isn't main()

You also should see in the link time details of the post I indicated above, where it says:
/usr/lib/gcc/x86_64-redhat-linux/4.1.2/32/crtbegin.o
There is a C startup module included by gcc when it gives the command to the linker. I believe that is the module named crtbegin.o in that example. So the linker makes the entry point of that module be the entry point of the entire binary.

The code in crtbegin.o does various initialization required by the C runtime library. Then it calls main()

So if I'm getting the details right, the OS gives control to ld-linux, which gives control to crtbegin, which calls main().

1 members found this post helpful.

Thank you for the detailed answer.
But one more question, from where the new process starts. Is it when ctbegin calls main or from inside ld-linux library ? Who is forking and exec ing?

I'm not sure there is an actual technical question there vs. a point of view on the meaning of "process starts".

There might be some explicit definition of that inside the Linux kernel. I don't know.

Any reasonable definition I can think of for the point at which a process starts would be somewhere before the kernel gives control to ld-linux inside that process.

I not sure I understand that question either:
The fork occurs in a different process (that is the parent of this process) and the exec occurs in this process while this process is still executing in the parent's image. That is all well before the kernel gives control to ld-linux in this process.

1 members found this post helpful.

Sorry for confusing you.
Even though you did not understand the question, I think I got my answer right here