Deadlock occurs rarely when traversing task's open-fds

repdetect · 09-05-2014, 11:17 AM

Hello,

I am a newbie about the Linux Kernel Module and I need some help.
Please let me know the problem with the below code.

Code:

struct fdtable *fdt = NULL;
struct file *file = NULL;

rcu_read_lock();
for_each_process(ptask) {
   if (!ptask->files) continue;
   
   spin_lock_irqsave(&f->file_lock, iflags);
   fdt = files_fdtable(f);
   if (fdt) {
      for (i = 0; i < fdt->max_fds; i++) {
         file = fcheck_files(f, i);
         if (!file) continue;
         if (file == some filepointer)  break;
      }
   }
   spin_unlock_irqrestore(&f->file_lock, iflags);
}
rcu_read_unlock();

I just want to traverse opened files from tasks but my system (CentOS 6.5, Kernel Version is 2.6.32-431.el6.i686) rarely hangs.
By the way I am working on netfilter hook module.
And I guess the deadlock a bit more happens when I try to open, close and switch between websites very quickly in Firefox.

I have already spent the number of days, please help me.
Thank you for any help.

xode · 09-08-2014, 12:10 AM

This is a wild guess on my part coming from basic multitasking principles. The rcu_read_lock() function might not be checking the state of the lock before turning the lock on. If that is the case, you need to add an additional code line to check the lock state and, if the lock is on, make the process wait until the lock is off.