brk vs sbrk vs mmap on FreeBSD

Index > Unix > brk vs sbrk vs mmap on FreeBSD

Author

Thread

Chewy509

Joined: 19 Jun 2003
Posts: 297
Location: Bris-vegas, Australia

Chewy509 16 Nov 2006, 00:56

Hi Guys,

I'm in the process of porting some apps to FreeBSD, and have found that my malloc() code wasn't working.

After some testing, came across that sbrk (brk is aliased to sbrk on fbsd) is no longer supported for 64bit apps on AMD64. (Returns error 45 - EOPNOTSUPP // Operation Not Supported). However couldn't find any documentation anywhere on the internet to support this? The kernel source seems to run around in circles... (if someone could tell the exact source code file that does the actual sbrk implementation, it would be great).

So did some research and found that FreeBSD coders recommend mmap() over sbrk() anyway... However when using mmap(), I'm getting an invalid file handle message? Code is below:

Code:

memInit:
  mov r4, 0 ; don't care where the memory is allocated
  mov r5, 1048576 ; alloc 1MB
  mov r3, 3 ; RW access to memory
  mov r2, 4096 ; MAP_ANON - not a file
  mov r8, -1 ; -1 for file handle if using MAP_ANON
  mov r9, 0 ; ignored for MAP_ANON
  mov r0, 197 ; mmap();
  syscall
  mov qword [_mmap], r0 ; save address so we can release it on exit;
  ret

Note: r0 = rax, r1 = rbx, r2 = rcx, r3 = rdx, r4 = rdi, r5 = rsi, r6 = rbp, r7 = rsp.

Carry flag is set, and r0 = EBADF (9) ; Bad File Descriptor?

So does anyone have an example of a working call to mmap() or tell me what's wrong with the above code, and possibly confirm the implementation status of sbrk()?

16 Nov 2006, 00:56

jb

Joined: 26 Mar 2007
Posts: 5
Location: Holland

jb 01 Apr 2007, 00:12

According to the NetBSD 3.1 man page for mmap, MAP_SHARED or MAP_PRIVATE should be present.

Another thing is, you leave out argument #7. Your r9 is not the `offset' but the `pad' parameter. You should push a zero offset for amd64 I think.

Anyhow, in NetBSD 3.1/i386 this test program seems to work:

$ cat test.asm
format elf executable
entry start
start:
push 0 0 ;64 bit offset
push 0xdeadbeef ;pad
push -1 ;fd
push 0x1002 ;MAP_ANON|MAP_PRIVATE
push 7 ;rwx
push 2*4096 ;give me two pages
push 0 ;I don't care where
push eax
mov eax,197
int 0x80
jc error
xor eax,eax ;exit 0
error:
push eax
push eax
xor eax,eax
inc eax
int 0x80
include 'note.inc'

Just after the system call, running under a debugger, the memory map looks like this:
08048000 4K read/write/exec [ heap ]
BDBFD000 8K read/write/exec [ anon ]
BDBFF000 4K read/exec [ uvm_aobj ]
BDC00000 30720K [ stack ]
BFA00000 1984K read/write [ stack ]
BFBF0000 64K read/write [ stack ]
total 2064K

The second line shows the 8K allocated.

01 Apr 2007, 00:12

Chewy509

Joined: 19 Jun 2003
Posts: 297
Location: Bris-vegas, Australia

Chewy509 04 Apr 2007, 01:39

Thanks for the late reply, but did eventually sort this issue out...

On FreeBSD 6.1, the direct interface to mmap() syscall will always return invalid file descriptor, but if you use the generic syscall interface, it works fine. (The syscall that lets you call other syscalls through it, I'm at work and don't have the syscall number handy).

Reading through the FreeBSD libc, and it's mmap() call, it eventually maps through it the kernel mmap() syscall, with an 8th parameter as well (the mmap() man page only has 7 parameters, which is correct for the libc mmap() call, but not for the kernel mmap() call). Sad

But in the end I did get it working (somewhat).

04 Apr 2007, 01:39

jb

Joined: 26 Mar 2007
Posts: 5
Location: Holland

jb 04 Apr 2007, 18:57

Running ktruss (or strace or whatever it is called on FreeBSD) is often a convenient alternative to reading politically correct man pages.

04 Apr 2007, 18:57

< Last Thread | Next Thread >

Forum Rules:

You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum
You cannot attach files in this forum
You can download files in this forum