On Sat, 2010-05-22 at 11:30 +0700, Mulyadi Santosa wrote: > On Fri, May 21, 2010 at 19:33, Shyam Burkule <shyam.burkule@xxxxxxxxx> wrote: > > Hi All, > > I am using Linux kernel 2.6.27(SLES11). I am getting one strange problem. > > I am using task_pid function followed by get_pid function to get pid > > structure. > > Here is code snippet, > > struct *pid = get_pid(task_pid(current)); > > > > I am saving this pid structure in a variable which is used later to kill the > > same process using kill_pid providing the same "pid *" which we saved > > earlier. Since kill_pid() requires "pid *", we have save this somewhere > > during process creation. Whenever we use get(task_pid(current)) in the code, > > I observed that the process is not visible in "ps -ef" as well as "/proc" > > filesystem. > > The codes say that get_pid increments the "count" field of the pid > struct. I failed to find the related code in fs/proc that says > anything about "don't show a task with count higher than 0". > > Thus, I could offer two hypotheses: > a. The code that "hides" the pid in /proc is something else, probably > additional code not included in mainline (since you're saying it's > SLES kernel, right?) > > b. it's just coincidence, that when you do something on this pid, you > accidentally remove the task structure, or at least the pid struct. Shyam, did you mean that (a) as soon as you call get_pid(task_pid(current)) the pid disappears from ps and proc, or (b) that you expect the pid to remain listed in ps and proc after the task exits, because you earlier called get_pid()? If (a) then I'm puzzled. Like Mulyadi, I also can't see anything that might explain why ps would be affected by get_pid. The only effect of get_pid is to increment the "count" member; I changed the name of this field and recompiled the kernel and this verified that the only places using count are: * kernel/pid.c * include/linux/pid.h * include/linux/init_task.h pid.c only uses it in get_pid and put_pid to reference-count the pid struct and free it when count=0. So there is clearly nothing in the proc filesystem or elsewhere that is looking at the count. And get_pid isn't moving structures from one list to another or anything like that. If (b), then that's not what get_pid is for. It ensures that the struct pid continues to exist, but the pid# itself becomes "unused", and is free for the OS to reuse. NB: I'm looking at 2.6.34... Cheers, Simon -- To unsubscribe from this list: send an email with "unsubscribe kernelnewbies" to ecartis@xxxxxxxxxxxx Please read the FAQ at http://kernelnewbies.org/FAQ
