htb_dequeue_tree assertion (kernel 2.4.21-ac4)
Hello,
I think the BUG_TRAP() in the htb_dequeue_tree() is wrong. First it checks if the class pointer "cl" is NULL, which is obviously right. But I do not understand why we also check whenever the queue length of the leaf queue is zero "cl->un.leaf.q->q.qlen". I would have put that in the expression of the "if" statements that comes afterwards. A queue length of 0 is not an error condition that should be reported (please, correct me if I misunderstood the code).
I can pretty much reliably trigger the assertion with a well utilized gigabit ethernet link when I flush and reactivate the TC configuration every 3 seconds. It looks like the error occurs only when confiuration changes are made.
I will some some more tests on monday when I am back at the office to verify that the queue length is (not) the problem.
bye,
wilfried
static struct sk_buff *htb_dequeue_tree(struct htb_sched *q,int prio,int level)
{
struct sk_buff *skb = NULL;
//struct htb_sched *q = (struct htb_sched *)sch->data;
struct htb_class *cl,*start;
/* look initial class up in the row */
start = cl = htb_lookup_leaf (q->row[level]+prio,prio,q->ptr[level]+prio);
do {
BUG_TRAP(cl && cl->un.leaf.q->q.qlen); if (!cl) return NULL;
HTB_DBG(4,1,"htb_deq_tr prio=%d lev=%d cl=%X defic=%d\n",
prio,level,cl->classid,cl->un.leaf.deficit[level]);
if (likely((skb = cl->un.leaf.q->dequeue(cl->un.leaf.q)) != NULL))
break;
if (!cl->warned) {_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
If you read comment above htb_dequeue_tree, it should be called only when it is sure that there are packets inside of the level/prio. It is known by other HTB mechanism (per-level activity lists).
Thus the bugtrap is to catch case where class was inserted into activity list because it had packets in its sub-qdisc but when we actually decide to dequeue - it has no packet. It is weird - can qdisc lose packets even when dequeue was not called ??
-------------------------------
Martin Devera aka devik
Linux kernel QoS/HTB maintainer
http://luxik.cdi.cz/~devik/
Quoted 37 lines · Wilfried Weissmann
Hello,
I think the
BUG_TRAP()in thehtb_dequeue_tree()is wrong. First it checks if the class pointer "cl" is NULL, which is obviously right. But I do not understand why we also check whenever the queue length of the leaf queue is zero "cl->un.leaf.q->q.qlen". I would have put that in the expression of the "if" statements that comes afterwards. A queue length of 0 is not an error condition that should be reported (please, correct me if I misunderstood the code).
I can pretty much reliably trigger the assertion with a well utilized gigabit ethernet link when I flush and reactivate the TC configuration every 3 seconds. It looks like the error occurs only when confiuration changes are made.
I will some some more tests on monday when I am back at the office to verify that the queue length is (not) the problem.bye,
wilfriedstatic struct
sk_buff*htb_dequeue_tree(struct htb_sched *q,int prio,int level){ struct sk_buff *skb = NULL; //struct htb_sched *q = (struct htb_sched *)sch->data; struct htb_class *cl,*start; /* look initial class up in the row */ start = cl = htb_lookup_leaf (q->row[level]+prio,prio,q->ptr[level]+prio); do { BUG_TRAP(cl && cl->un.leaf.q->q.qlen); if (!cl) return NULL; HTB_DBG(4,1,"htb_deq_tr prio=%d lev=%d cl=%X defic=%d\n", prio,level,cl->classid,cl->un.leaf.deficit[level]); if (likely((skb = cl->un.leaf.q->dequeue(cl->un.leaf.q)) != NULL)) break; if (!cl->warned) {
Quoted 5 lines
_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
Quoted 9 lines · devik
If you read comment above
htb_dequeue_tree, it should be called only when it is sure that there are packets inside of the level/prio. It is known by other HTB mechanism (per-level activity lists).Thus the bugtrap is to catch case where class was inserted into activity list because it had packets in its sub-qdisc but when we actually decide to dequeue - it has no packet. It is weird - can qdisc lose packets even when dequeue was not called ??
If you change the depth of the leave queue then it is possible to drop packets or if you completely exchange the queue. Which would also explain why the assertion only occurs when the configuration is altered.
Greetings,
Wilfried
Quoted 45 lines
-------------------------------
Martin Devera aka devik
Linux kernel QoS/HTB maintainer
http://luxik.cdi.cz/~devik/Quoted 36 lines · Wilfried Weissmann
Hello,
I think the
BUG_TRAP()in thehtb_dequeue_tree()is wrong. First it checks if the class pointer "cl" is NULL, which is obviously right. But I do not understand why we also check whenever the queue length of the leaf queue is zero "cl->un.leaf.q->q.qlen". I would have put that in the expression of the "if" statements that comes afterwards. A queue length of 0 is not an error condition that should be reported (please, correct me if I misunderstood the code).
I can pretty much reliably trigger the assertion with a well utilized gigabit ethernet link when I flush and reactivate the TC configuration every 3 seconds. It looks like the error occurs only when confiuration changes are made.
I will some some more tests on monday when I am back at the office to verify that the queue length is (not) the problem.bye,
wilfriedstatic struct
sk_buff*htb_dequeue_tree(struct htb_sched *q,int prio,int level){ struct sk_buff *skb = NULL; //struct htb_sched *q = (struct htb_sched *)sch->data; struct htb_class *cl,*start; /* look initial class up in the row */ start = cl = htb_lookup_leaf (q->row[level]+prio,prio,q->ptr[level]+prio); do { BUG_TRAP(cl && cl->un.leaf.q->q.qlen); if (!cl) return NULL; HTB_DBG(4,1,"htb_deq_tr prio=%d lev=%d cl=%X defic=%d\n", prio,level,cl->classid,cl->un.leaf.deficit[level]); if (likely((skb = cl->un.leaf.q->dequeue(cl->un.leaf.q)) != NULL)) break; if (!cl->warned) {
_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
Quoted 10 lines
Quoted 9 lines
If you read comment above
htb_dequeue_tree, it should be called only when it is sure that there are packets inside of the level/prio. It is known by other HTB mechanism (per-level activity lists).Thus the bugtrap is to catch case where class was inserted into activity list because it had packets in its sub-qdisc but when we actually decide to dequeue - it has no packet. It is weird - can qdisc lose packets even when dequeue was not called ??
Quoted 3 lines
If you change the depth of the leave queue then it is possible to drop packets or if you completely exchange the queue. Which would also explain why the assertion only occurs when the configuration is altered.
Well, I agree that there is something wrong. Now it is neccessary to find scenario where it does happen so that it is fixed in right way. I have not much time these days to test these cases but your informations would lead to following hypothesis:
Classe's child qdisc is replaced while old one still has nonzero queue. New empty qdisc is grafted under class instead. What about attached patch (it is against my latest version so you can see offset warnings) ?
devik
sch_htb.c +4 -0
@@ -1286,6 +1286,10 @@ static int htb_graft(struct Qdisc *sch,return -ENOBUFS ;sch_tree_lock (sch);if ((*old =xchg (&cl->un.leaf.q, new)) !=NULL ) { +/* TODO: test it */ +if (cl->prio_activity) +htb_deactivate ((struct htb_sched *)sch->data,cl); +/* TODO: is it correct ? Why CBQ doesn't do it ? */ sch->q.qlen -= (*old)->q.qlen;qdisc_reset (*old);
Quoted 15 lines · devik
Quoted 10 lines
Quoted 9 lines
If you read comment above
htb_dequeue_tree, it should be called only when it is sure that there are packets inside of the level/prio. It is known by other HTB mechanism (per-level activity lists).Thus the bugtrap is to catch case where class was inserted into activity list because it had packets in its sub-qdisc but when we actually decide to dequeue - it has no packet. It is weird - can qdisc lose packets even when dequeue was not called ??
Quoted 3 lines
If you change the depth of the leave queue then it is possible to drop packets or if you completely exchange the queue. Which would also explain why the assertion only occurs when the configuration is altered.
Quoted 8 lines
Well, I agree that there is something wrong. Now it is neccessary to find scenario where it does happen so that it is fixed in right way. I have not much time these days to test these cases but your informations would lead to following hypothesis:
Classe's child qdisc is replaced while old one still has nonzero queue. New empty qdisc is grafted under class instead. What about attached patch (it is against my latest version so you can see offset warnings) ?
This would not work if there are several intermediates HTB queues from the device to the leave queue. In this case every queue from the queue that was changed to the root has to be notified about the change. (The setup we want to use involves such a configuration.) Maybe it is better to just deactivate a class when a dequeue from its leave failes due to a zero queue length. If you are concerned about performance then an audit process could be implemented. For example to check one leave queue every 64 packets +/- initial random offset to create some entropy similar to the maximum mount count in the ext2 filesystem. Maybe there are better ways to do this. I am not so familiar with the code.
I will make some tests with the patch tomorrow. If my theory is true then it should still help a lot.
bye,
wilfried
Quoted 9 lines
devik
------------------------------------------------------------------------
sch_htb.c static int htb_graft(struct Qdisc *sch, · line 1295
Quoted 10 lines
sch_htb.c +4
-0@@ -1286,6 +1286,10 @@ static int htb_graft(struct Qdisc *sch,return -ENOBUFS ;sch_tree_lock (sch);if ((*old =xchg (&cl->un.leaf.q, new)) !=NULL ) { +/* TODO: test it */ +if (cl->prio_activity) +htb_deactivate ((struct htb_sched *)sch->data,cl); +/* TODO: is it correct ? Why CBQ doesn't do it ? */ sch->q.qlen -= (*old)->q.qlen;qdisc_reset (*old);
_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
Quoted 11 lines
Quoted 10 lines · devik
Quoted 10 lines
Quoted 9 lines
If you read comment above
htb_dequeue_tree, it should be called only when it is sure that there are packets inside of the level/prio. It is known by other HTB mechanism (per-level activity lists).Thus the bugtrap is to catch case where class was inserted into activity list because it had packets in its sub-qdisc but when we actually decide to dequeue - it has no packet. It is weird - can qdisc lose packets even when dequeue was not called ??
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
If you change the depth of the leave queue then it is possible to drop packets or if you completely exchange the queue. Which would also explain why the assertion only occurs when the configuration is altered.
Now I verified that the problem is indeed the 0 queue length and not a NULL class pointer.
Quoted 23 lines
Quoted 5 lines
Well, I agree that there is something wrong. Now it is neccessary to find scenario where it does happen so that it is fixed in right way. I have not much time these days to test these cases but your
informations
Quoted 5 lines
would lead to following hypothesis:
Classe's child qdisc is replaced while old one still has nonzero queue. New empty qdisc is grafted under class instead. What about attached patch (it is against my latest version so you can see offset warnings) ?
This would not work if there are several intermediates HTB queues from the device to the leave queue. In this case every queue from the queue that was changed to the root has to be notified about the change. (The setup we want to use involves such a configuration.) Maybe it is better to just deactivate a class when a dequeue from its leave failes due to a zero queue length. If you are concerned about performance then an audit process could be implemented. For example to check one leave queue every 64 packets +/- initial random offset to create some entropy similar to the maximum mount count in the ext2 filesystem. Maybe there are better ways to do this. I am not so familiar with the code.
Quoted 2 lines
I will make some tests with the patch tomorrow. If my theory is true then it should still help a lot.
With the patch applied it is much harder to find the right ceil settings to trigger the assertion, however it does not fix the problem. I also got the following log entry:
HTB: dequeue bug (8,270045,270045), report it please !
Maybe this massages is just a side effect of the bug.
Greetings,
Wilfried
Quoted 23 lines
bye,
wilfriedQuoted 19 lines
devik
------------------------------------------------------------------------
sch_htb.c +4
-0@@ -1286,6 +1286,10 @@ static int htb_graft(struct Qdisc *sch,return -ENOBUFS ;sch_tree_lock (sch);if ((*old =xchg (&cl->un.leaf.q, new)) !=NULL ) { +/* TODO: test it */ +if (cl->prio_activity) +htb_deactivate ((struct htb_sched *)sch->data,cl); +/* TODO: is it correct ? Why CBQ doesn't do it ? */ sch->q.qlen -= (*old)->q.qlen;qdisc_reset (*old);
Signature · 7 lines
+++ GMX - Mail, Messaging & more http://www.gmx.net +++ Jetzt ein- oder umsteigen und USB-Speicheruhr als Prämie sichern! _______________________________________________ LARTC mailing list / [email protected] http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
Yes I agree with you regarding zero queue size. I plan to make patch similar to your proposal. I hope it will be today.
-------------------------------
Martin Devera aka devik
Linux kernel QoS/HTB maintainer
http://luxik.cdi.cz/~devik/
Quoted 82 lines · On Mon, 21 Jul 2003 [email protected] wrote:
Quoted 11 lines
Quoted 10 lines · devik
Quoted 10 lines
Quoted 9 lines
If you read comment above
htb_dequeue_tree, it should be called only when it is sure that there are packets inside of the level/prio. It is known by other HTB mechanism (per-level activity lists).Thus the bugtrap is to catch case where class was inserted into activity list because it had packets in its sub-qdisc but when we actually decide to dequeue - it has no packet. It is weird - can qdisc lose packets even when dequeue was not called ??
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
If you change the depth of the leave queue then it is possible to drop packets or if you completely exchange the queue. Which would also explain why the assertion only occurs when the configuration is altered.
Now I verified that the problem is indeed the 0 queue length and not a NULL class pointer.
Quoted 23 lines
Quoted 5 lines
Well, I agree that there is something wrong. Now it is neccessary to find scenario where it does happen so that it is fixed in right way. I have not much time these days to test these cases but your
informations
Quoted 5 lines
would lead to following hypothesis:
Classe's child qdisc is replaced while old one still has nonzero queue. New empty qdisc is grafted under class instead. What about attached patch (it is against my latest version so you can see offset warnings) ?
This would not work if there are several intermediates HTB queues from the device to the leave queue. In this case every queue from the queue that was changed to the root has to be notified about the change. (The setup we want to use involves such a configuration.) Maybe it is better to just deactivate a class when a dequeue from its leave failes due to a zero queue length. If you are concerned about performance then an audit process could be implemented. For example to check one leave queue every 64 packets +/- initial random offset to create some entropy similar to the maximum mount count in the ext2 filesystem. Maybe there are better ways to do this. I am not so familiar with the code.
Quoted 2 lines
I will make some tests with the patch tomorrow. If my theory is true then it should still help a lot.
With the patch applied it is much harder to find the right ceil settings to trigger the assertion, however it does not fix the problem. I also got the following log entry:
HTB: dequeue bug (8,270045,270045), report it please !
Maybe this massages is just a side effect of the bug.
Greetings,
WilfriedQuoted 10 lines
bye,
wilfriedQuoted 6 lines
devik
------------------------------------------------------------------------
Quoted 13 lines
sch_htb.c static int htb_graft(struct Qdisc *sch, · line 1295
Quoted 13 lines
sch_htb.c +4
-0@@ -1286,6 +1286,10 @@ static int htb_graft(struct Qdisc *sch,return -ENOBUFS ;sch_tree_lock (sch);if ((*old =xchg (&cl->un.leaf.q, new)) !=NULL ) { +/* TODO: test it */ +if (cl->prio_activity) +htb_deactivate ((struct htb_sched *)sch->data,cl); +/* TODO: is it correct ? Why CBQ doesn't do it ? */ sch->q.qlen -= (*old)->q.qlen;qdisc_reset (*old);--
+++ GMX - Mail, Messaging & more http://www.gmx.net +++
Quoted 4 lines
Jetzt ein- oder umsteigen und USB-Speicheruhr als Prämie sichern!
_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
Hi,
try attached fix please (it duplicates last one too so that you might get a reject).
-------------------------------
Martin Devera aka devik
Linux kernel QoS/HTB maintainer
http://luxik.cdi.cz/~devik/
Quoted 48 lines · Wilfried Weissmann
Quoted 15 lines · devik
Quoted 10 lines
Quoted 9 lines
If you read comment above
htb_dequeue_tree, it should be called only when it is sure that there are packets inside of the level/prio. It is known by other HTB mechanism (per-level activity lists).Thus the bugtrap is to catch case where class was inserted into activity list because it had packets in its sub-qdisc but when we actually decide to dequeue - it has no packet. It is weird - can qdisc lose packets even when dequeue was not called ??
Quoted 3 lines
If you change the depth of the leave queue then it is possible to drop packets or if you completely exchange the queue. Which would also explain why the assertion only occurs when the configuration is altered.
Quoted 8 lines
Well, I agree that there is something wrong. Now it is neccessary to find scenario where it does happen so that it is fixed in right way. I have not much time these days to test these cases but your informations would lead to following hypothesis:
Classe's child qdisc is replaced while old one still has nonzero queue. New empty qdisc is grafted under class instead. What about attached patch (it is against my latest version so you can see offset warnings) ?
This would not work if there are several intermediates HTB queues from the device to the leave queue. In this case every queue from the queue that was changed to the root has to be notified about the change. (The setup we want to use involves such a configuration.) Maybe it is better to just deactivate a class when a dequeue from its leave failes due to a zero queue length. If you are concerned about performance then an audit process could be implemented. For example to check one leave queue every 64 packets +/- initial random offset to create some entropy similar to the maximum mount count in the ext2 filesystem. Maybe there are better ways to do this. I am not so familiar with the code.
I will make some tests with the patch tomorrow. If my theory is true then it should still help a lot.
bye,
wilfriedQuoted 6 lines
devik
------------------------------------------------------------------------
Quoted 19 lines
sch_htb.c static int htb_graft(struct Qdisc *sch, · line 1295
Quoted 13 lines
sch_htb.c +4
-0@@ -1286,6 +1286,10 @@ static int htb_graft(struct Qdisc *sch,return -ENOBUFS ;sch_tree_lock (sch);if ((*old =xchg (&cl->un.leaf.q, new)) !=NULL ) { +/* TODO: test it */ +if (cl->prio_activity) +htb_deactivate ((struct htb_sched *)sch->data,cl); +/* TODO: is it correct ? Why CBQ doesn't do it ? */ sch->q.qlen -= (*old)->q.qlen;qdisc_reset (*old);
sch_htb.c +16 -2
@@ -947,15 +947,24 @@ static struct sk_buff * htb_dequeue_tree(struct htb_sched *q,int prio,int level) { struct sk_buff *skb = NULL; -//struct htb_sched *q = (struct htb_sched *)sch->data; struct htb_class *cl,*start;/* look initial class up in the row */ start = cl =htb_lookup_leaf (q->row[level]+prio,prio,q->ptr[level]+prio); do { - BUG_TRAP(cl && cl->un.leaf.q->q.qlen); if (!cl) return NULL; + BUG_TRAP(cl); + if (!cl) return NULL; HTB_DBG(4,1,"htb_deq_tr prio=%d lev=%d cl=%X defic=%d\n", prio,level,cl->classid,cl->un.leaf.deficit[level]); + + /* class can be empty - it is unlikely but can be true if leaf + qdisc drops packets in enqueue routine or if someone used + graft operation on the leaf since last dequeue; + simply deactivate and skip such class */ + if (unlikely(cl->un.leaf.q->q.qlen == 0)) { + htb_deactivate(q,cl); + goto new_lookup; + } if (likely((skb = cl->un.leaf.q->dequeue(cl->un.leaf.q)) != NULL)) break;
@@ -965,6 +974,7 @@ htb_dequeue_tree(struct htb_sched *q,int } q->nwc_hit++;htb_next_rb_node ((level?cl->parent->un.inner.ptr:q->ptr[0 ])+prio); +new_lookup: cl =htb_lookup_leaf (q->row[level]+prio,prio,q->ptr[level]+prio); }while (cl != start);
@@ -1286,6 +1296,10 @@ static int htb_graft(struct Qdisc *sch,return -ENOBUFS ;sch_tree_lock (sch);if ((*old =xchg (&cl->un.leaf.q, new)) !=NULL ) { +/* TODO: test it */ +if (cl->prio_activity) +htb_deactivate ((struct htb_sched *)sch->data,cl); +/* TODO: is it correct ? Why CBQ doesn't do it ? */ sch->q.qlen -= (*old)->q.qlen;qdisc_reset (*old);
Quoted 4 lines · devik
Hi,
try attached fix please (it duplicates last one too so that you might get a reject).
Thanks, but now the rb_tree may become empty and this causes an oops in htb_lookup_leaf() (tree-rb_node = NULL). I think the kernel crashes in "while ((*sp->pptr)->rb_left)". Catching that case is easy. But we must not forget to leave the do{}while() loop in htb_dequeue_tree() when an empty tree is detected.
I cannot provide you any patches right now. I will send them tomorrow if everything works.
Greetings,
Wilfried
Quoted 117 lines
-------------------------------
Martin Devera aka devik
Linux kernel QoS/HTB maintainer
http://luxik.cdi.cz/~devik/Quoted 108 lines · Wilfried Weissmann
Quoted 15 lines · devik
Quoted 10 lines
Quoted 9 lines
If you read comment above
htb_dequeue_tree, it should be called only when it is sure that there are packets inside of the level/prio. It is known by other HTB mechanism (per-level activity lists).Thus the bugtrap is to catch case where class was inserted into activity list because it had packets in its sub-qdisc but when we actually decide to dequeue - it has no packet. It is weird - can qdisc lose packets even when dequeue was not called ??
Quoted 3 lines
If you change the depth of the leave queue then it is possible to drop packets or if you completely exchange the queue. Which would also explain why the assertion only occurs when the configuration is altered.
Quoted 8 lines
Well, I agree that there is something wrong. Now it is neccessary to find scenario where it does happen so that it is fixed in right way. I have not much time these days to test these cases but your informations would lead to following hypothesis:
Classe's child qdisc is replaced while old one still has nonzero queue. New empty qdisc is grafted under class instead. What about attached patch (it is against my latest version so you can see offset warnings) ?
This would not work if there are several intermediates HTB queues from the device to the leave queue. In this case every queue from the queue that was changed to the root has to be notified about the change. (The setup we want to use involves such a configuration.) Maybe it is better to just deactivate a class when a dequeue from its leave failes due to a zero queue length. If you are concerned about performance then an audit process could be implemented. For example to check one leave queue every 64 packets +/- initial random offset to create some entropy similar to the maximum mount count in the ext2 filesystem. Maybe there are better ways to do this. I am not so familiar with the code.
I will make some tests with the patch tomorrow. If my theory is true then it should still help a lot.
bye,
wilfriedQuoted 8 lines
devik
------------------------------------------------------------------------
sch_htb.c static int htb_graft(struct Qdisc *sch, · line 1295
Quoted 10 lines
sch_htb.c +4
-0@@ -1286,6 +1286,10 @@ static int htb_graft(struct Qdisc *sch,return -ENOBUFS ;sch_tree_lock (sch);if ((*old =xchg (&cl->un.leaf.q, new)) !=NULL ) { +/* TODO: test it */ +if (cl->prio_activity) +htb_deactivate ((struct htb_sched *)sch->data,cl); +/* TODO: is it correct ? Why CBQ doesn't do it ? */ sch->q.qlen -= (*old)->q.qlen;qdisc_reset (*old);------------------------------------------------------------------------
sch_htb.c +12
-2@@ -947,15 +947,24 @@ static struct sk_buff * htb_dequeue_tree(struct htb_sched *q,int prio,int level) { struct sk_buff *skb = NULL; -//struct htb_sched *q = (struct htb_sched *)sch->data; struct htb_class *cl,*start;/* look initial class up in the row */ start = cl =htb_lookup_leaf (q->row[level]+prio,prio,q->ptr[level]+prio); do { - BUG_TRAP(cl && cl->un.leaf.q->q.qlen); if (!cl) return NULL; + BUG_TRAP(cl); + if (!cl) return NULL; HTB_DBG(4,1,"htb_deq_tr prio=%d lev=%d cl=%X defic=%d\n", prio,level,cl->classid,cl->un.leaf.deficit[level]); + + /* class can be empty - it is unlikely but can be true if leaf + qdisc drops packets in enqueue routine or if someone used + graft operation on the leaf since last dequeue; + simply deactivate and skip such class */ + if (unlikely(cl->un.leaf.q->q.qlen = 0)) { + htb_deactivate(q,cl); + goto new_lookup; + } if (likely((skb = cl->un.leaf.q->dequeue(cl->un.leaf.q)) != NULL)) break;@@ -965,6 +974,7 @@ htb_dequeue_tree(struct htb_sched *q,int } q->nwc_hit++;htb_next_rb_node ((level?cl->parent->un.inner.ptr:q->ptr[0 ])+prio); +new_lookup: cl =htb_lookup_leaf (q->row[level]+prio,prio,q->ptr[level]+prio); }while (cl != start);
Quoted 12 lines
sch_htb.c static int htb_graft(struct Qdisc *sch, · line 1305
Quoted 11 lines
sch_htb.c +4
-0@@ -1286,6 +1296,10 @@ static int htb_graft(struct Qdisc *sch,return -ENOBUFS ;sch_tree_lock (sch);if ((*old =xchg (&cl->un.leaf.q, new)) !=NULL ) { +/* TODO: test it */ +if (cl->prio_activity) +htb_deactivate ((struct htb_sched *)sch->data,cl); +/* TODO: is it correct ? Why CBQ doesn't do it ? */ sch->q.qlen -= (*old)->q.qlen;qdisc_reset (*old);
_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/