Linux Kernel Internals

altosys · 06-08-2004, 06:19 AM

Can any1 provide me with details of :
strace ./a.out

where ./a.out is the output of a simple program printf-ing Welcome ONLY.

I would like a detailed explaination of all the memory mapping, et al. of the strace command.

infamous41md · 06-08-2004, 12:26 PM

the mapping is for libC and ld.

Mara · 06-08-2004, 02:52 PM

Strace is not about memory mappings. It just shows system function calls. Memory mappings of a running program can be found in /proc/PID/maps.

aluser · 06-08-2004, 05:16 PM

strace shows every *syscall* that your program makes, along with the arguments to it and the return value. You can look up any syscall in the man pages like "man 2 open" "man 2 mmap" etc. I'm not sure why I get "old_mmap" in my output; there is only a page for mmap. Other than that you should be able to look up all the calls your program is making.

infamous41md · 06-08-2004, 07:40 PM

u guys seemed to have missed his question. he is asking why mmap() is called, even tho your program doesn't call it. and the answer is as i said above, every program that is linked with libC maps the libc text segment into its address space, and also the libc data and bss sections; in addition the dynamic linker, ld has it's text and data areas mapped as well.

Code:

[n00b@localho.outernet] cat > hw.c
int main() { printf("foo\n"); return 0;}
[n00b@localho.outernet] gcc hw.c 
[n00b@localho.outernet] strace -v ./a.out 
execve("./a.out", ["./a.out"], [/* 31 vars */]) = 0
brk(0)                                  = 0x80494a8
open("/etc/ld.so.preload", O_RDONLY)    = -1 ENOENT (No such file or directory)
#open up ld and get its size for mapping
open("/etc/ld.so.cache", O_RDONLY)      = 3
fstat64(3, {st_dev=makedev(3, 67), st_ino=586277, st_mode=S_IFREG|0644, st_nlink=1, st_uid=0, st_gid=0, st_blksize=4096, st_blocks=72, st_size=33730, st_atime=2004/06/08-20:34:14, st_mtime=2004/06/08-14:13:40, st_ctime=2004/06/08-14:13:40}) = 0
old_mmap(NULL, 33730, PROT_READ, MAP_PRIVATE, 3, 0) = 0x40015000
close(3)                                = 0 
#open up libc and get its size
open("/lib/libc.so.6", O_RDONLY)        = 3
read(3, "\177ELF\1\1\1\0\0\0\0\0\0\0\0\0\3\0\3\0\1\0\0\0p\\\1\000"..., 1024) = 1024
fstat64(3, {st_dev=makedev(3, 67), st_ino=649660, st_mode=S_IFREG|0755, st_nlink=1, st_uid=0, st_gid=0, st_blksize=4096, st_blocks=2816, st_size=1435624, st_atime=2004/06/08-20:34:14, st_mtime=2003/03/05-00:57:19, st_ctime=2004/01/01-13:42:43}) = 0
old_mmap(NULL, 1256740, PROT_READ|PROT_EXEC, MAP_PRIVATE, 3, 0) = 0x4001e000
mprotect(0x40148000, 36132, PROT_NONE)  = 0
#now map its data section
old_mmap(0x40148000, 20480, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_FIXED, 3, 0x12a000) = 0x40148000
#and .bss
old_mmap(0x4014d000, 15652, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0) = 0x4014d000
close(3)                                = 0
#and ld's data section 
old_mmap(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x40151000
munmap(0x40015000, 33730)               = 0
fstat64(1, {st_dev=makedev(0, 6), st_ino=2, st_mode=S_IFCHR|0620, st_nlink=1, st_uid=1000, st_gid=5, st_blksize=1024, st_blocks=0, st_rdev=makedev(136, 0), st_atime=2004/06/08-20:34:14, st_mtime=2004/06/08-20:34:14, st_ctime=2004/06/08-15:20:16}) = 0
old_mmap(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x40015000 
write(1, "foo\n", 4foo
)                    = 4
munmap(0x40015000, 4096)                = 0
semget(IPC_PRIVATE, 4096, IPC_CREAT|0x40149140|0400) = -1 ENOSYS (Function not implemented)
_exit(0)                                = ?

aluser · 06-08-2004, 08:57 PM

mmap() is being called to load shared library code, but I read the "et al" part as meaning he's only using mmap as one example of strace output.

altosys · 07-23-2004, 10:00 AM

Thanks ppl,.......

I googled around a lot on the basis of all you rreplies..........
I guess....all of you helped be in building the answer i wanted.

Anyways, Im back with a nw query :-)

Inside the Kernel sourse there are a lot of macros which resemble the following :
"__KERNEL__" (double under scores) and even "_something_"
Do these have any special meanings

Also when a source says :
#ifndef _LINUX_FS_H
#define _LINUX_FS_H

what exactly is it trying to say or do?

btmiller · 07-23-2004, 11:08 AM

The

#ifndef SOME_MACRO
#define SOME_MACRO

/* stuff */

#endif

is a header guard. It's so the file can be included multiple times without causing multiple definition errors.

aluser · 07-23-2004, 12:36 PM

To elaborate on what btmiller said,
when you write foo.h, you should always wrap it in

Code:

#ifndef _FOO_H
# define _FOO_H

....

#endif /* _FOO_H */

This causes the entire foo.h to only be read once. e.g. you could have bar.c #include foo.h and baz.h, and baz.h #includes foo.h, Then foo.h would be #included twice and you'd get compiler warnings about previous declarations of everything in foo.h.

The whole problem exists because the #include directive is very literal. The compiler sees it as if everything in the #included file is cut and pasted directly in place of the #include.

It's also common to have

Code:

#ifndef SOMETHING
# define SOMETHING blah blah
#endif

which simply defines SOMETHING if it isn't already defined; this has nothing to do with protecting the entire file from being included multiple times.

altosys · 07-24-2004, 01:51 AM

wait wait.......

I'm still a bit confused !

So I write a file foo.h and I include
#ifndef _FOO_H
#define _FOO_H
#endif

Now, I have another file bar.c which:
#include foo.h
#define baz.h

baz.h , in turn, might :
#include foo.h

So the above #ifndef and #define pair will prevent the warnings of "previous declarations of foo.h"

Is that correct? If so, how is that actually taking place? and does the macro _FOO_H expand on its own to foo.h ?

fr0zen · 07-24-2004, 03:13 PM

#defines are available to all subsequent included files.

Since baz.h is included after foo.h, the _FOO_H #define is within its scope.