Skip to contentSkip to conversation

net: sched: fix some issues

PATCH Superseded Yang Yingliang 2 participants 5 messages Nov 18, 2013 – Nov 19, 2013

Subsystems (3)

v2 of 7 Superseded Latest is v7 →

lore
Share & tools
lore
#0 lore raw

v2 cover net: sched: fix some issues

fix a regression introduced by commit b757c9336d63f94c6b57532 (tbf: improved accuracy at high rates).
fix quantum calculation introduced by 64bit rates.

v2
patch 1/2: redescribe the regression
patch 2/2: add Eric's ack

Yang Yingliang (2):
net: sched: tbf: fix calculation of max_size
net: sched: htb: fix calculation of quantum

 include/net/sch_generic.h | 46 ++++++++++++++++++++++++++++++++
 net/sched/sch_htb.c       | 18 +++++++------
 net/sched/sch_tbf.c       | 67 ++++++++++++++++++++++++++---------------------
 3 files changed, 93 insertions(+), 38 deletions(-)
Signature · 1 line
1.8.0
#1 lore raw

PATCH v2 1/2 net: sched: tbf: fix calculation of max_size

commit b757c9336d63f94c6b57532(tbf: improved accuracy at high rates) introduce a regression.

With the follow command:
tc qdisc add dev eth1 root handle 1: tbf latency 50ms burst 10KB rate 30gbit mtu 64k

Without this patch, the max_size value is 10751(bytes).
But, in fact, the real max_size value should be smaller than 7440(bytes). Or a packet whose length is bigger than 7440 will cause network congestion. Because the packet is so big that can't get enough tokens. Even all the tokens in the buffer is given to the packet.

With this patch, the max_size value is 7440(bytes).
The packets whose length is bigger than 7440(bytes) will be dropped or reshape in tbf_enqueue().

---
 include/net/sch_generic.h | 46 ++++++++++++++++++++++++++++++++
 net/sched/sch_tbf.c       | 67 ++++++++++++++++++++++++++---------------------
 2 files changed, 83 insertions(+), 30 deletions(-)
include/net/sch_generic.h +46 -0
index d0a6321..8da64f3 100644
@@ -701,6 +701,52 @@ static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
 	return ((u64)len * r->mult) >> r->shift;
 }
 
+/* Time to Length, convert time in ns to length in bytes
+ * to determinate how many bytes can be sent in given time.
+ */
+static inline u64 psched_ns_t2l(const struct psched_ratecfg *r,
+				u64 time_in_ns)
+{
+	u64 len = time_in_ns;
+	u8 shift = r->shift;
+	bool is_div = false;
+
+	/* The formula is :
+	 * len = (time_in_ns << shift) / mult
+	 * when time_in_ns does shift, it would overflow.
+	 * If overflow happens first time, do division.
+	 * Then do shift. If it happens again,
+	 * set lenth to ~0ULL.
+	 */
+	while (shift) {
+		if (len & (1ULL << 63)) {
+			if (!is_div) {
+				len = div64_u64(len, r->mult);
+				is_div = true;
+			} else {
+				/* overflow happens */
+				len = ~0ULL;
+				is_div = true;
+				break;
+			}
+		}
+		len <<= 1;
+		shift--;
+	}
+	if (!is_div)
+		len = div64_u64(len, r->mult);
+
+	if (unlikely(r->linklayer == TC_LINKLAYER_ATM))
+		len = (len / 53) * 48;
+
+	if (len > r->overhead)
+		len -= r->overhead;
+	else
+		len = 0;
+
+	return len;
+}
+
 void psched_ratecfg_precompute(struct psched_ratecfg *r,
 			       const struct tc_ratespec *conf,
 			       u64 rate64);
net/sched/sch_tbf.c +37 -30

102 lines not on this page · raw message

Signature · 1 line
1.8.0
#2 lore raw

Re: [PATCH net v2 1/2] net: sched: tbf: fix calculation of max_size

Quoted 12 lines · On Mon, 18 Nov 2013 16:39:23 +0800 Yang Yingliang <[email protected]>…

commit b757c9336d63f94c6b57532(tbf: improved accuracy at high rates) introduce a regression.

With the follow command:
tc qdisc add dev eth1 root handle 1: tbf latency 50ms burst 10KB rate 30gbit mtu 64k

Without this patch, the max_size value is 10751(bytes).
But, in fact, the real max_size value should be smaller than 7440(bytes). Or a packet whose length is bigger than 7440 will cause network congestion. Because the packet is so big that can't get enough tokens. Even all the tokens in the buffer is given to the packet.

Quoted 3 lines

With this patch, the max_size value is 7440(bytes).
The packets whose length is bigger than 7440(bytes) will be dropped or reshape in tbf_enqueue().

I acknowledge that TBF seems to have some dependencies to the userspace constructed rate table (which we do NOT use anymore in the kernel). And that these should be fixed.

But I'm not sure that your patch is the best solution... and the patch also contains some issues, see inlined comments.

The main annoying problem is *again* how the rate table system got removed, in the kernel, but nobody fixed userspace.

So, the main problem is that qopt->buffer (send from userspace/tc) is in a "time-format" (user input "burst" in bytes). Which used-to, make sense because the rate table used the same "time-format".

Now you are reversing this calculation of "q->buffer" (token burst) back into bytes, so we can choose "max_size" (to avoid a problem in tbf_dequeue()).

I don't like this converting back-and-forth, I'm worried about rounding errors.

The easiest "hack" solution would be:

for (n = 0; n < 65535; n++)

     if (psched_l2t_ns(&qopt->rate, n) > q->buffer)
         break;

max_size = n;

Unfortunately we have to keep backward compat with iproute2/tc, but IMHO it would be a lot easier, if we could fix userspace, and remove all the length-to-time calculations, as they should now be the responsibility of the kernel. Well, wishful thinking...

Quoted 50 lines
---
 include/net/sch_generic.h | 46 ++++++++++++++++++++++++++++++++
 net/sched/sch_tbf.c       | 67 ++++++++++++++++++++++++++---------------------
 2 files changed, 83 insertions(+), 30 deletions(-)
include/net/sch_generic.h +36 -0
index d0a6321..8da64f3 100644
@@ -701,6 +701,52 @@ static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
 	return ((u64)len * r->mult) >> r->shift;
 }
 
+/* Time to Length, convert time in ns to length in bytes
+ * to determinate how many bytes can be sent in given time.
+ */
+static inline u64 psched_ns_t2l(const struct psched_ratecfg *r,
+				u64 time_in_ns)
+{
+	u64 len = time_in_ns;
+	u8 shift = r->shift;
+	bool is_div = false;
+
+	/* The formula is :
+	 * len = (time_in_ns << shift) / mult
+	 * when time_in_ns does shift, it would overflow.
+	 * If overflow happens first time, do division.
+	 * Then do shift. If it happens again,
+	 * set lenth to ~0ULL.
+	 */
+	while (shift) {
+		if (len & (1ULL << 63)) {
+			if (!is_div) {
+				len = div64_u64(len, r->mult);
+				is_div = true;
+			} else {
+				/* overflow happens */
+				len = ~0ULL;
+				is_div = true;
+				break;
+			}
+		}
+		len <<= 1;
+		shift--;
+	}
+	if (!is_div)
+		len = div64_u64(len, r->mult);
+
+	if (unlikely(r->linklayer == TC_LINKLAYER_ATM))

include/net/sch_generic.h static inline u64 psched_l2t_ns(const struct psched_ratecfg *r, · line 749

Quoted 10 lines
include/net/sch_generic.h +10 -0
+		len = (len / 53) * 48;
+
+	if (len > r->overhead)
+		len -= r->overhead;
+	else
+		len = 0;
+
+	return len;
+}
+

Are we 100% sure, that the conversion is accurate between psched_l2t_ns() and psched_ns_t2l for all possible rates.

E.g. why is it that r->shift have to be recalculate (orig created in psched_ratecfg_precompute()).

Quoted 25 lines
include/net/sch_generic.h +0 -0
 void psched_ratecfg_precompute(struct psched_ratecfg *r,
 			       const struct tc_ratespec *conf,
 			       u64 rate64);
net/sched/sch_tbf.c +1 -6
index 68f9859..eb9ce7b 100644
@@ -279,7 +279,7 @@ static int tbf_change(struct Qdisc *sch, struct nlattr *opt)
 	struct qdisc_rate_table *rtab = NULL;
 	struct qdisc_rate_table *ptab = NULL;
 	struct Qdisc *child = NULL;
-	int max_size, n;
+	u32 max_size = 0;
 	u64 rate64 = 0, prate64 = 0;
 
 	err = nla_parse_nested(tb, TCA_TBF_MAX, opt, tbf_policy);
@@ -291,33 +291,20 @@ static int tbf_change(struct Qdisc *sch, struct nlattr *opt)
 		goto done;
 
 	qopt = nla_data(tb[TCA_TBF_PARMS]);
-	rtab = qdisc_get_rtab(&qopt->rate, tb[TCA_TBF_RTAB]);
-	if (rtab == NULL)
-		goto done;
-
-	if (qopt->peakrate.rate) {

net/sched/sch_tbf.c static int tbf_change(struct Qdisc *sch, struct nlattr *opt) · line 299

Quoted 10 lines
net/sched/sch_tbf.c +6 -4
-		if (qopt->peakrate.rate > qopt->rate.rate)
-			ptab = qdisc_get_rtab(&qopt->peakrate, tb[TCA_TBF_PTAB]);
-		if (ptab == NULL)
-			goto done;
+	if (qopt->rate.linklayer == TC_LINKLAYER_UNAWARE) {
+		rtab = qdisc_get_rtab(&qopt->rate, tb[TCA_TBF_RTAB]);
+		if (rtab) {
+			qdisc_put_rtab(rtab);
+			rtab = NULL;
+		}

This is correct code construct, for backward compatible reading of the rate table.

But, notice how your free this at once, which means you also should cleanup the exit/done: section.

net/sched/sch_tbf.c static int tbf_change(struct Qdisc *sch, struct nlattr *opt) · line 308

Quoted 6 lines
net/sched/sch_tbf.c +0 -5
 	}
-
-	for (n = 0; n < 256; n++)
-		if (rtab->data[n] > qopt->buffer)
-			break;
-	max_size = (n << qopt->rate.cell_log) - 1;

This is here we could do the quick-and-dirty solution:

for (n = 0; n < 65535; n++)

     if (psched_l2t_ns(&qopt->rate, n) > q->buffer)
         break;

max_size = n;

Quoted 29 lines
net/sched/sch_tbf.c +6 -12
-	if (ptab) {
-		int size;
-
-		for (n = 0; n < 256; n++)
-			if (ptab->data[n] > qopt->mtu)
-				break;
-		size = (n << qopt->peakrate.cell_log) - 1;
-		if (size < max_size)
-			max_size = size;
+	if (qopt->peakrate.linklayer == TC_LINKLAYER_UNAWARE) {
+		ptab = qdisc_get_rtab(&qopt->peakrate, tb[TCA_TBF_PTAB]);
+		if (ptab) {
+			qdisc_put_rtab(ptab);
+			ptab = NULL;
+		}
 	}
-	if (max_size < 0)
-		goto done;
 
 	if (q->qdisc != &noop_qdisc) {
 		err = fifo_set_limit(q->qdisc, qopt->limit);
@@ -339,25 +326,45 @@ static int tbf_change(struct Qdisc *sch, struct nlattr *opt)
 	}
 	q->limit = qopt->limit;
 	q->mtu = PSCHED_TICKS2NS(qopt->mtu);
-	q->max_size = max_size;
 	q->buffer = PSCHED_TICKS2NS(qopt->buffer);
 	q->tokens = q->buffer;
 	q->ptokens = q->mtu;

net/sched/sch_tbf.c static int tbf_change(struct Qdisc *sch, struct nlattr *opt) · line 339

Quoted 10 lines
net/sched/sch_tbf.c +5 -2
 
 	if (tb[TCA_TBF_RATE64])
 		rate64 = nla_get_u64(tb[TCA_TBF_RATE64]);
-	psched_ratecfg_precompute(&q->rate, &rtab->rate, rate64);
-	if (ptab) {
+	psched_ratecfg_precompute(&q->rate, &qopt->rate, rate64);
+	if (!q->rate.rate_bytes_ps)
+		goto unlock_done;
+	max_size = min_t(u64, psched_ns_t2l(&q->rate, q->buffer), ~0);
+	max_size = min_t(u32, max_size, (256 << qopt->rate.cell_log) - 1);

The rate system and the rate.cell_log is not really used anymore, so its a bit strange to use it. Perhaps it's even a bug to base a calculation on this.

Quoted 15 lines
net/sched/sch_tbf.c +9 -1
+
+	if (qopt->peakrate.rate) {
+		u64 size = 0;
 		if (tb[TCA_TBF_PRATE64])
 			prate64 = nla_get_u64(tb[TCA_TBF_PRATE64]);
-		psched_ratecfg_precompute(&q->peak, &ptab->rate, prate64);
+		psched_ratecfg_precompute(&q->peak, &qopt->peakrate, prate64);
+		size = psched_ns_t2l(&q->peak, q->mtu);
+		max_size = min_t(u64, max_size, size);
+		max_size = min_t(u32,
+				 max_size,
+				 (256 << qopt->peakrate.cell_log) - 1);
 		q->peak_present = true;
 	} else {
 		q->peak_present = false;

net/sched/sch_tbf.c static int tbf_change(struct Qdisc *sch, struct nlattr *opt) · line 363

Quoted 10 lines
net/sched/sch_tbf.c +6 -0
 	}
 
+	if (!max_size)
+		goto unlock_done;
+	q->max_size = max_size;
+
 	sch_tree_unlock(sch);
 	err = 0;
+
+	if (0) {

I really dislike this construct. I'm afraid what a dumb compile would optimized this to.

net/sched/sch_tbf.c static int tbf_change(struct Qdisc *sch, struct nlattr *opt) · line 370

Quoted 7 lines
net/sched/sch_tbf.c +4 -0
+unlock_done:
+		sch_tree_unlock(sch);
+		err = -EINVAL;
+	}
 done:
 	if (rtab)
 		qdisc_put_rtab(rtab);

This could be cleaned up, as we already have released the rtab's.

Signature · 5 lines
Best regards,
  Jesper Dangaard Brouer
  MSc.CS, Sr. Network Kernel Developer at Red Hat
  Author of http://www.iptv-analyzer.org
  LinkedIn: http://www.linkedin.com/in/brouer
#3 lore raw

Re: [PATCH net v2 1/2] net: sched: tbf: fix calculation of max_size

Quoted 120 lines · Jesper Dangaard Brouer
Quoted 12 lines · On Mon, 18 Nov 2013 16:39:23 +0800 Yang Yingliang <[email protected]>…

commit b757c9336d63f94c6b57532(tbf: improved accuracy at high rates) introduce a regression.

With the follow command:
tc qdisc add dev eth1 root handle 1: tbf latency 50ms burst 10KB rate 30gbit mtu 64k

Without this patch, the max_size value is 10751(bytes).
But, in fact, the real max_size value should be smaller than 7440(bytes). Or a packet whose length is bigger than 7440 will cause network congestion. Because the packet is so big that can't get enough tokens. Even all the tokens in the buffer is given to the packet.

Quoted 3 lines

With this patch, the max_size value is 7440(bytes).
The packets whose length is bigger than 7440(bytes) will be dropped or reshape in tbf_enqueue().

I acknowledge that TBF seems to have some dependencies to the userspace constructed rate table (which we do NOT use anymore in the kernel). And that these should be fixed.

But I'm not sure that your patch is the best solution... and the patch also contains some issues, see inlined comments.

The main annoying problem is *again* how the rate table system got removed, in the kernel, but nobody fixed userspace.

So, the main problem is that qopt->buffer (send from userspace/tc) is in a "time-format" (user input "burst" in bytes). Which used-to, make sense because the rate table used the same "time-format".

Now you are reversing this calculation of "q->buffer" (token burst) back into bytes, so we can choose "max_size" (to avoid a problem in tbf_dequeue()).

I don't like this converting back-and-forth, I'm worried about rounding errors.

The easiest "hack" solution would be:

for (n = 0; n < 65535; n++)

     if (psched_l2t_ns(&qopt->rate, n) > q->buffer)
         break;

max_size = n;

Unfortunately we have to keep backward compat with iproute2/tc, but IMHO it would be a lot easier, if we could fix userspace, and remove all the length-to-time calculations, as they should now be the responsibility of the kernel. Well, wishful thinking...

Quoted 50 lines
---
 include/net/sch_generic.h | 46 ++++++++++++++++++++++++++++++++
 net/sched/sch_tbf.c       | 67 ++++++++++++++++++++++++++---------------------
 2 files changed, 83 insertions(+), 30 deletions(-)
include/net/sch_generic.h +36 -0
index d0a6321..8da64f3 100644
@@ -701,6 +701,52 @@ static inline u64 psched_l2t_ns(const struct psched_ratecfg *r,
 	return ((u64)len * r->mult) >> r->shift;
 }
 
+/* Time to Length, convert time in ns to length in bytes
+ * to determinate how many bytes can be sent in given time.
+ */
+static inline u64 psched_ns_t2l(const struct psched_ratecfg *r,
+				u64 time_in_ns)
+{
+	u64 len = time_in_ns;
+	u8 shift = r->shift;
+	bool is_div = false;
+
+	/* The formula is :
+	 * len = (time_in_ns << shift) / mult
+	 * when time_in_ns does shift, it would overflow.
+	 * If overflow happens first time, do division.
+	 * Then do shift. If it happens again,
+	 * set lenth to ~0ULL.
+	 */
+	while (shift) {
+		if (len & (1ULL << 63)) {
+			if (!is_div) {
+				len = div64_u64(len, r->mult);
+				is_div = true;
+			} else {
+				/* overflow happens */
+				len = ~0ULL;
+				is_div = true;
+				break;
+			}
+		}
+		len <<= 1;
+		shift--;
+	}
+	if (!is_div)
+		len = div64_u64(len, r->mult);
+
+	if (unlikely(r->linklayer == TC_LINKLAYER_ATM))

include/net/sch_generic.h static inline u64 psched_l2t_ns(const struct psched_ratecfg *r, · line 749

Quoted 10 lines
include/net/sch_generic.h +10 -0
+		len = (len / 53) * 48;
+
+	if (len > r->overhead)
+		len -= r->overhead;
+	else
+		len = 0;
+
+	return len;
+}
+

Are we 100% sure, that the conversion is accurate between psched_l2t_ns() and psched_ns_t2l for all possible rates.

Quoted 2 lines

E.g. why is it that r->shift have to be recalculate (orig created in psched_ratecfg_precompute()).

It don't recalculate r->shift, it just do shift bit by bit in case that len overflows. If len overflows, set it to ~0ULL.
I would like to use the way you suggest to calculate the max_size. So forget about this function.:)

Quoted 41 lines
Quoted 27 lines
include/net/sch_generic.h +0 -0
 void psched_ratecfg_precompute(struct psched_ratecfg *r,
 			       const struct tc_ratespec *conf,
 			       u64 rate64);
net/sched/sch_tbf.c +1 -6
index 68f9859..eb9ce7b 100644
@@ -279,7 +279,7 @@ static int tbf_change(struct Qdisc *sch, struct nlattr *opt)
 	struct qdisc_rate_table *rtab = NULL;
 	struct qdisc_rate_table *ptab = NULL;
 	struct Qdisc *child = NULL;
-	int max_size, n;
+	u32 max_size = 0;
 	u64 rate64 = 0, prate64 = 0;
 
 	err = nla_parse_nested(tb, TCA_TBF_MAX, opt, tbf_policy);
@@ -291,33 +291,20 @@ static int tbf_change(struct Qdisc *sch, struct nlattr *opt)
 		goto done;
 
 	qopt = nla_data(tb[TCA_TBF_PARMS]);
-	rtab = qdisc_get_rtab(&qopt->rate, tb[TCA_TBF_RTAB]);
-	if (rtab == NULL)
-		goto done;
-
-	if (qopt->peakrate.rate) {

net/sched/sch_tbf.c static int tbf_change(struct Qdisc *sch, struct nlattr *opt) · line 299

Quoted 10 lines
net/sched/sch_tbf.c +6 -4
-		if (qopt->peakrate.rate > qopt->rate.rate)
-			ptab = qdisc_get_rtab(&qopt->peakrate, tb[TCA_TBF_PTAB]);
-		if (ptab == NULL)
-			goto done;
+	if (qopt->rate.linklayer == TC_LINKLAYER_UNAWARE) {
+		rtab = qdisc_get_rtab(&qopt->rate, tb[TCA_TBF_RTAB]);
+		if (rtab) {
+			qdisc_put_rtab(rtab);
+			rtab = NULL;
+		}

This is correct code construct, for backward compatible reading of the rate table.

Quoted 2 lines

But, notice how your free this at once, which means you also should cleanup the exit/done: section.

Agreed, I will change it in v3.
Thanks!

Quoted 49 lines

net/sched/sch_tbf.c static int tbf_change(struct Qdisc *sch, struct nlattr *opt) · line 308

Quoted 9 lines
net/sched/sch_tbf.c +0 -5
 	}
-
-	for (n = 0; n < 256; n++)
-		if (rtab->data[n] > qopt->buffer)
-			break;
-	max_size = (n << qopt->rate.cell_log) - 1;

This is here we could do the quick-and-dirty solution:

for (n = 0; n < 65535; n++)

     if (psched_l2t_ns(&qopt->rate, n) > q->buffer)
         break;

max_size = n;

Quoted 30 lines
net/sched/sch_tbf.c +6 -12
-	if (ptab) {
-		int size;
-
-		for (n = 0; n < 256; n++)
-			if (ptab->data[n] > qopt->mtu)
-				break;
-		size = (n << qopt->peakrate.cell_log) - 1;
-		if (size < max_size)
-			max_size = size;
+	if (qopt->peakrate.linklayer == TC_LINKLAYER_UNAWARE) {
+		ptab = qdisc_get_rtab(&qopt->peakrate, tb[TCA_TBF_PTAB]);
+		if (ptab) {
+			qdisc_put_rtab(ptab);
+			ptab = NULL;
+		}
 	}
-	if (max_size < 0)
-		goto done;
 
 	if (q->qdisc != &noop_qdisc) {
 		err = fifo_set_limit(q->qdisc, qopt->limit);
@@ -339,25 +326,45 @@ static int tbf_change(struct Qdisc *sch, struct nlattr *opt)
 	}
 	q->limit = qopt->limit;
 	q->mtu = PSCHED_TICKS2NS(qopt->mtu);
-	q->max_size = max_size;
 	q->buffer = PSCHED_TICKS2NS(qopt->buffer);
 	q->tokens = q->buffer;
 	q->ptokens = q->mtu;
 
Quoted 13 lines

net/sched/sch_tbf.c static int tbf_change(struct Qdisc *sch, struct nlattr *opt) · line 339

Quoted 9 lines
net/sched/sch_tbf.c +5 -2
 	if (tb[TCA_TBF_RATE64])
 		rate64 = nla_get_u64(tb[TCA_TBF_RATE64]);
-	psched_ratecfg_precompute(&q->rate, &rtab->rate, rate64);
-	if (ptab) {
+	psched_ratecfg_precompute(&q->rate, &qopt->rate, rate64);
+	if (!q->rate.rate_bytes_ps)
+		goto unlock_done;
+	max_size = min_t(u64, psched_ns_t2l(&q->rate, q->buffer), ~0);
+	max_size = min_t(u32, max_size, (256 << qopt->rate.cell_log) - 1);

The rate system and the rate.cell_log is not really used anymore, so its a bit strange to use it. Perhaps it's even a bug to base a calculation on this.

cell_log is calculated from mtu(user input in bytes) in userspace. In the old calculation, max_size should be smaller than (256 << cell_log) which means it's smaller than mtu.

Quoted 18 lines
Quoted 18 lines
net/sched/sch_tbf.c +9 -1
+
+	if (qopt->peakrate.rate) {
+		u64 size = 0;
 		if (tb[TCA_TBF_PRATE64])
 			prate64 = nla_get_u64(tb[TCA_TBF_PRATE64]);
-		psched_ratecfg_precompute(&q->peak, &ptab->rate, prate64);
+		psched_ratecfg_precompute(&q->peak, &qopt->peakrate, prate64);
+		size = psched_ns_t2l(&q->peak, q->mtu);
+		max_size = min_t(u64, max_size, size);
+		max_size = min_t(u32,
+				 max_size,
+				 (256 << qopt->peakrate.cell_log) - 1);
 		q->peak_present = true;
 	} else {
 		q->peak_present = false;
 	}
 
Quoted 11 lines

net/sched/sch_tbf.c static int tbf_change(struct Qdisc *sch, struct nlattr *opt) · line 363

Quoted 8 lines
net/sched/sch_tbf.c +6 -0
+	if (!max_size)
+		goto unlock_done;
+	q->max_size = max_size;
+
 	sch_tree_unlock(sch);
 	err = 0;
+
+	if (0) {

I really dislike this construct. I'm afraid what a dumb compile would optimized this to.

OK, change it in v3.
Thanks!

Quoted 12 lines

net/sched/sch_tbf.c static int tbf_change(struct Qdisc *sch, struct nlattr *opt) · line 370

Quoted 8 lines
net/sched/sch_tbf.c +4 -0
+unlock_done:
+		sch_tree_unlock(sch);
+		err = -EINVAL;
+	}
 done:
 	if (rtab)
 		qdisc_put_rtab(rtab);

This could be cleaned up, as we already have released the rtab's.

#4 lore raw

PATCH v2 2/2 net: sched: htb: fix calculation of quantum

In reply to Yang Yingliang

Now, 32bit rates may be not the true rate.
So use rate_bytes_ps which is from
max(rate32, rate64) to calcualte quantum.

---
 net/sched/sch_htb.c | 18 ++++++++++--------
 1 file changed, 10 insertions(+), 8 deletions(-)
net/sched/sch_htb.c +10 -8
index 0e1e38b..57c6678 100644
@@ -1477,11 +1477,20 @@ static int htb_change_class(struct Qdisc *sch, u32 classid,
 		sch_tree_lock(sch);
 	}
 
+	rate64 = tb[TCA_HTB_RATE64] ? nla_get_u64(tb[TCA_HTB_RATE64]) : 0;
+
+	ceil64 = tb[TCA_HTB_CEIL64] ? nla_get_u64(tb[TCA_HTB_CEIL64]) : 0;
+
+	psched_ratecfg_precompute(&cl->rate, &hopt->rate, rate64);
+	psched_ratecfg_precompute(&cl->ceil, &hopt->ceil, ceil64);
+
 	/* it used to be a nasty bug here, we have to check that node
 	 * is really leaf before changing cl->un.leaf !
 	 */
 	if (!cl->level) {
-		cl->quantum = hopt->rate.rate / q->rate2quantum;
+		u64 quantum = div64_u64(cl->rate.rate_bytes_ps,
+					q->rate2quantum);
+		cl->quantum = min_t(u64, quantum, INT_MAX);
 		if (!hopt->quantum && cl->quantum < 1000) {
 			pr_warning(
 			       "HTB: quantum of class %X is small. Consider r2q change.\n",
@@ -1500,13 +1509,6 @@ static int htb_change_class(struct Qdisc *sch, u32 classid,
 			cl->prio = TC_HTB_NUMPRIO - 1;
 	}
 
-	rate64 = tb[TCA_HTB_RATE64] ? nla_get_u64(tb[TCA_HTB_RATE64]) : 0;
-
-	ceil64 = tb[TCA_HTB_CEIL64] ? nla_get_u64(tb[TCA_HTB_CEIL64]) : 0;
-
-	psched_ratecfg_precompute(&cl->rate, &hopt->rate, rate64);
-	psched_ratecfg_precompute(&cl->ceil, &hopt->ceil, ceil64);
-
 	cl->buffer = PSCHED_TICKS2NS(hopt->buffer);
 	cl->cbuffer = PSCHED_TICKS2NS(hopt->cbuffer);
 
Signature · 1 line
1.8.0