Why the funny kernel always crashes ??

[gazolinia]

Hi,

Since 2.6.18 and above kernel not exporting symbol table and tasklist_lock , I have patched the vanilla kernel to export them manually. Now I am rewriting the fork method , My code compiles fine and works properly for a program which has single fork, when program calls repeatedly called fork without latency , my system crashes , I am sure that the code breaks within the list loop .My pseudo code is as follows

asmlinkage int my_fork(struct pt_regs regs)
{
int ret;

ret = call_original_fork(fork_saved, regs);
if (ret > 0) {
struct list_head *this,*next;
struct task_struct *child_task= NULL;
lock_kernel();
read_lock(&tasklist_lock);

list_for_each_safe(this, next, &current->children) {
child_task = list_entry(this, struct task_struct, sibling);
if (child_task->pid == ret) {
/* This guy is still running, even though it got waited? */
printkDEBUG("Yes we found a match\n");
break;
}
}
unlock_kernel();
read_unlock(&tasklist_lock);
return ret;
}

Please let us know if you need any more info.

Thanks in Advance,
Jai

[gazolinia]

Hi geeks,

Today I found out the cause is because of write lock . Let myself be more clear .

1) I have kernel 2.6.18 with SMP enabled (CONFIG_SMP =Y}
2) I am writing a kernel module to manipulate the fork system call (Just to keep track of all forks happening in my system) .

This is my code for fork call, I am exporting the symbol table and everytime I call fork in my user program this method gets called without any issue .

asmlinkage int my_fork(struct pt_regs regs)
{
int ret;
rwlock_t chroots_lock = RW_LOCK_UNLOCKED;
unsigned long flags;
ret = call_original_fork(fork_saved, regs);
if (ret > 0) {

struct task_struct *child_task= NULL;

//READ LOCKING

lock_kernel();
read_lock_irqsave(&chroots_lock,flags);
child_task = find_task_by_pid (ret);
unlock_kernel();
read_lock_irqsave(&chroots_lock,flags);

//WRITE LCOKING cool the problem is here

lock_kernel();
write_lock_irqsave(&chroots_lock,flags);
//Add the child_task to main list (I am 100 % sure this method is working fine )
write_lock_irqrestore(&chroots_lock,flags);
unlock_kernel();
return ret;

}

}

This module and everything worked fine , until 2.4 kernel now I am migrating to 2.6 kernel where after making few changes to code it get compiled and loads fine. I didn't touch the locking part of the code after this migration.

After loading the module , even now it works fine for example 1( delays between the fork calls) , but the system hangs for the example 2 when there is no delay between the fork calls.

Example 1: prog1.pl

fork();
sleep(5);
fork();
sleep(5);
fork();

Example 2:

fork();
fork();
fork();


Does anything I need to change in my locking part of code to make it work as expected in 2.6 kernel . Any little help will be great for me , Please let me know if you need any more info.

Thanks,
Jai

[gazolinia]

Hi guys,
My module was intended to run on Uniprocessor machine, but I compiled the kernel with SMP mode . This all caused the issue, I just recompiled with SMP disabled. Now it looks little ok but not 100% .

~Jai