Skip to contentSkip to conversation

possible kernel bug in signal transit.

DISCUSSION Alex Lyashkov 2 participants 10 messages Mar 13, 2004 – Mar 14, 2004

lore
Share & tools
lore
#0 lore raw

Hello All

I analyze kernel vanila 2.6.4 and found one possible bug in __kill_pg_info function.

        for_each_task_pid(pgrp, PIDTYPE_PGID, p, l, pid) {
                err = group_send_sig_info(sig, info, p);
                if (retval)
                        retval = err;
        }
but I think if (retval) is incorrect check. possible this cycle must be
        for_each_task_pid(pgrp, PIDTYPE_PGID, p, l, pid) {
                err = group_send_sig_info(sig, info, p);
                if (ret) {
                        retval = err;
			break;
		}
        }

because in original variant me assign to retval only first value from ret and other be ignored if this value be 0.

Signature · 2 lines
Alex Lyashkov <[email protected]>
PSoft
#1 lore raw
Quoted 19 lines · Alex Lyashkov

Hello All

I analyze kernel vanila 2.6.4 and found one possible bug in __kill_pg_info function.

        for_each_task_pid(pgrp, PIDTYPE_PGID, p, l, pid) {
                err = group_send_sig_info(sig, info, p);
                if (retval)
                        retval = err;
        }
but I think if (retval) is incorrect check. possible this cycle must be
        for_each_task_pid(pgrp, PIDTYPE_PGID, p, l, pid) {
                err = group_send_sig_info(sig, info, p);
                if (ret) {
                        retval = err;
			break;
		}
        }
Quoted 3 lines

because in original variant me assign to retval only first value from ret and other be ignored if this value be 0.

No, the code's OK, albeit undesirably obscure. It will return -ESRCH if none of the tasks had a matching pgrp and will return the result of the final non-zero-returning group_send_sig_info() if one or more of the group_send_sig_info() calls failed, and will return zero if all of the group_send_sig_info() calls returned zero.

Thanks for checking though..

#2 lore raw

В Вск, 14.03.2004, в 03:18, Andrew Morton пишет:

Quoted 24 lines
Quoted 19 lines · Alex Lyashkov

Hello All

I analyze kernel vanila 2.6.4 and found one possible bug in __kill_pg_info function.

        for_each_task_pid(pgrp, PIDTYPE_PGID, p, l, pid) {
                err = group_send_sig_info(sig, info, p);
                if (retval)
                        retval = err;
        }
but I think if (retval) is incorrect check. possible this cycle must be
        for_each_task_pid(pgrp, PIDTYPE_PGID, p, l, pid) {
                err = group_send_sig_info(sig, info, p);
                if (ret) {
                        retval = err;
			break;
		}
        }
Quoted 3 lines

because in original variant me assign to retval only first value from ret and other be ignored if this value be 0.

Quoted 7 lines

No, the code's OK, albeit undesirably obscure. It will return -ESRCH if none of the tasks had a matching pgrp and will return the result of the final non-zero-returning group_send_sig_info() if one or more of the group_send_sig_info() calls failed, and will return zero if all of the group_send_sig_info() calls returned zero.

Thanks for checking though..

No. it can`t return final non-zero-returning group_send_sig_info() if first call group_send_sig_info return 0.
Example me have 3 groups it will return
1 - zero
2 - nonzero (example -1)
3 - nonzero (example -2)
original code will return zero.
but you say it will return last non zero - "-2" in example. (do only first assignment after it retval is zero and no assignments does.)
For her logic me can write.
=====

	int retval = 0;
	int err = -1;
        for_each_task_pid(pgrp, PIDTYPE_PGID, p, l, pid) {
                 err = group_send_sig_info(sig, info, p);
		if( err )
			retval = err;

         }
	return err==-1 ? -ESRCH : retval;

===
If me not enter to this cycle - retval = -ESRCH, and last non zero err if cycle is worked or zero if all group_send_sig_info() return zero.

Signature · 2 lines
Alex Lyashkov <[email protected]>
PSoft
#3 lore raw
Quoted 4 lines · Alex Lyashkov
Quoted 2 lines

Thanks for checking though..

No. it can`t return final non-zero-returning group_send_sig_info() if first call group_send_sig_info return 0.

you're right. How about the nice and simple version?

int __kill_pg_info(int sig, struct siginfo *info, pid_t pgrp)

{
	struct task_struct *p;
	struct list_head *l;
	struct pid *pid;
	int retval;
	int found;

	if (pgrp <= 0)
		return -EINVAL;

	found = 0;
	retval = 0;
	for_each_task_pid(pgrp, PIDTYPE_PGID, p, l, pid) {
		int err;

		found = 1;
		err = group_send_sig_info(sig, info, p);
		if (!retval)
			retval = err;
	}
	return found ? retval : -ESRCH;
}
#4 lore raw

В Вск, 14.03.2004, в 07:00, Andrew Morton пишет:

Quoted 24 lines
Quoted 4 lines · Alex Lyashkov
Quoted 2 lines

Thanks for checking though..

No. it can`t return final non-zero-returning group_send_sig_info() if first call group_send_sig_info return 0.

you're right. How about the nice and simple version?

int __kill_pg_info(int sig, struct siginfo *info, pid_t pgrp)

{
	struct task_struct *p;
	struct list_head *l;
	struct pid *pid;
	int retval;
	int found;

	if (pgrp <= 0)
		return -EINVAL;

	found = 0;
	retval = 0;
	for_each_task_pid(pgrp, PIDTYPE_PGID, p, l, pid) {
		int err;
Quoted 7 lines
		found = 1;
		err = group_send_sig_info(sig, info, p);
		if (!retval)
			retval = err;
	}
	return found ? retval : -ESRCH;
}

not. it error. At this code you save first non zero value err but other been ignored.

Signature · 2 lines
Alex Lyashkov <[email protected]>
PSoft
#5 lore raw
Quoted 24 lines · Alex Lyashkov
Quoted 17 lines

int __kill_pg_info(int sig, struct siginfo *info, pid_t pgrp)

{
	struct task_struct *p;
	struct list_head *l;
	struct pid *pid;
	int retval;
	int found;

	if (pgrp <= 0)
		return -EINVAL;

	found = 0;
	retval = 0;
	for_each_task_pid(pgrp, PIDTYPE_PGID, p, l, pid) {
		int err;
Quoted 7 lines
		found = 1;
		err = group_send_sig_info(sig, info, p);
		if (!retval)
			retval = err;
	}
	return found ? retval : -ESRCH;
}
Quoted 2 lines

not. it error. At this code you save first non zero value err but other been ignored.

Well we can only return one error code. Or are you suggesting that we should terminate the loop early on error? If so, why?

#6 lore raw

В Вск, 14.03.2004, в 07:47, Andrew Morton пишет:

Quoted 28 lines
Quoted 24 lines · Alex Lyashkov
Quoted 17 lines

int __kill_pg_info(int sig, struct siginfo *info, pid_t pgrp)

{
	struct task_struct *p;
	struct list_head *l;
	struct pid *pid;
	int retval;
	int found;

	if (pgrp <= 0)
		return -EINVAL;

	found = 0;
	retval = 0;
	for_each_task_pid(pgrp, PIDTYPE_PGID, p, l, pid) {
		int err;
Quoted 7 lines
		found = 1;
		err = group_send_sig_info(sig, info, p);
		if (!retval)
			retval = err;
	}
	return found ? retval : -ESRCH;
}
Quoted 2 lines

not. it error. At this code you save first non zero value err but other been ignored.

Quoted 2 lines

Well we can only return one error code. Or are you suggesting that we should terminate the loop early on error? If so, why?

You say me can return _last_ error core. but this function return _first_.

I write second variant where not terminate loop and save _last_ error code (i was sending in previous mail). but if you have i write full function:
====
int __kill_pg_info(int sig, struct siginfo *info, pid_t pgrp)

{
	struct task_struct *p;
	struct list_head *l;
	struct pid *pid;

int retval = 0;
int err = -1;

if (pgrp <= 0)
return -EINVAL;

        for_each_task_pid(pgrp, PIDTYPE_PGID, p, l, pid) {
                 err = group_send_sig_info(sig, info, p);
                if( err )
                        retval = err;

         }
        return err==-1 ? -ESRCH : retval;
}

===
what you think about its code ?

Signature · 2 lines
Alex Lyashkov <[email protected]>
PSoft
#7 lore raw
Quoted 5 lines · Alex Lyashkov
Quoted 3 lines

Well we can only return one error code. Or are you suggesting that we should terminate the loop early on error? If so, why?

You say me can return _last_ error core. but this function return _first_.

It doesn't matter, really. Other parts of the kernel will generally return the first-encountered error because at times it _does_ matter. But here it doesn't.

#8 lore raw

В Вск, 14.03.2004, в 08:09, Andrew Morton пишет:

Quoted 10 lines
Quoted 5 lines · Alex Lyashkov
Quoted 3 lines

Well we can only return one error code. Or are you suggesting that we should terminate the loop early on error? If so, why?

You say me can return _last_ error core. but this function return _first_.

It doesn't matter, really. Other parts of the kernel will generally return the first-encountered error because at times it _does_ matter. But here it doesn't.

okey. second question.
a really need extra variable instead change conditions in return ?

Signature · 2 lines
Alex Lyashkov <[email protected]>
PSoft
#9 lore raw
Quoted 9 lines · Alex Lyashkov

__ ______, 14.03.2004, __ 08:09, Andrew Morton __________:

Quoted 7 lines
Quoted 5 lines · Alex Lyashkov
Quoted 3 lines

Well we can only return one error code. Or are you suggesting that we should terminate the loop early on error? If so, why?

You say me can return _last_ error core. but this function return _first_.

Quoted 5 lines
Quoted 3 lines

It doesn't matter, really. Other parts of the kernel will generally return the first-encountered error because at times it _does_ matter. But here it doesn't.

okey. second question.
a really need extra variable instead change conditions in return ?

I think it's clearer that way, don't you?