Skip to contentSkip to conversation

bonding: modify the current and add new hash functions

PATCH Superseded Nikolay Aleksandrov 3 participants 7 messages Sep 26, 2013 – Oct 1, 2013

Subsystems (4)

v4 of 5 Superseded Latest is v5 →

lore
Share & tools
lore
#0 lore raw

v4 cover bonding: modify the current and add new hash functions

Hi all,
This is a complete remake of my old patch that modified the bonding hash functions to use skb_flow_dissect which was suggested by Eric Dumazet. This time around I've left the old modes although using a new hash function again suggested by Eric, which is the same for all modes. The only difference is the way the headers are obtained. The old modes obtain them as before in order to address concerns about speed, but the 2 new ones use skb_flow_dissect. The unification of the hash function allows to remove a pointer from struct bonding and also a few extra functions that dealt with it. Two new functions are added which take care of the hashing based on bond->params.xmit_policy only:
bond_xmit_hash() - global function, used by XOR and 3ad modes bond_flow_dissect() - used by bond_xmit_hash() to obtain the necessary headers and combine them according to bond->params.xmit_policy. Also factor out the ports extraction from skb_flow_dissect and add a new function - skb_flow_get_ports() which can be re-used.

v2: add the flow_dissector patch and use skb_flow_get_ports in patch 02
v3: fix a bug in the flow_dissector patch that caused a different thoff
by modifying the thoff argument in skb_flow_get_ports directly, most of the users already do it anyway.
Also add the necessary export symbol for skb_flow_get_ports.
v4: integrate the thoff bug fix in patch 01

Best regards,
Nikolay Aleksandrov

Nikolay Aleksandrov (3):
flow_dissector: factor out the ports extraction in skb_flow_get_ports
bonding: modify the old and add new xmit hash policies
bonding: document the new xmit policy modes and update the changed
ones

 Documentation/networking/bonding.txt |  66 ++++++------
 drivers/net/bonding/bond_3ad.c       |   2 +-
 drivers/net/bonding/bond_main.c      | 197 ++++++++++++-----------------------
 drivers/net/bonding/bond_sysfs.c     |   2 -
 drivers/net/bonding/bonding.h        |   3 +-
 include/net/flow_keys.h              |   1 +
 include/uapi/linux/if_bonding.h      |   2 +
 net/core/flow_dissector.c            |  39 +++++--
 8 files changed, 137 insertions(+), 175 deletions(-)
Signature · 1 line
1.8.1.4
#1 lore raw

PATCH v4 1/3 flow_dissector: factor out the ports extraction in skb_flow_get_ports

Factor out the code that extracts the ports from skb_flow_dissect and add a new function skb_flow_get_ports which can be re-used. Also fix a bug that caused thoff to be wrong in the AH case, spotted by Eric Dumazet.

---
v2: new patch
v3: fix a bug in skb_flow_dissect where thoff didn't have poff added by
modifying thoff directly in skb_flow_get_ports as it's done anyway. Also add the necessary export symbol for skb_flow_get_ports.
v4: integrate the thoff fix in skb_flow_get_ports
This seems like a good idea because there're other users that can re-use it later as well.

 include/net/flow_keys.h   |  1 +
 net/core/flow_dissector.c | 39 ++++++++++++++++++++++++++++-----------
 2 files changed, 29 insertions(+), 11 deletions(-)
include/net/flow_keys.h +1 -0
index ac2439d..7e64bd8 100644
@@ -14,4 +14,5 @@ struct flow_keys {
 };
 
 bool skb_flow_dissect(const struct sk_buff *skb, struct flow_keys *flow);
+__be32 skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto);
 #endif
net/core/flow_dissector.c +28 -11
index 1929af8..ec9919b 100644
@@ -25,9 +25,35 @@ static void iph_to_flow_copy_addrs(struct flow_keys *flow, const struct iphdr *i
 	memcpy(&flow->src, &iph->saddr, sizeof(flow->src) + sizeof(flow->dst));
 }
 
+/**
+ * skb_flow_get_ports - extract the upper layer ports and return them
+ * @skb: buffer to extract the ports from
+ * @thoff: transport header offset
+ * @ip_proto: protocol for which to get port offset
+ *
+ * The function will try to retrieve the ports at offset thoff + poff where poff
+ * is the protocol port offset returned from proto_ports_offset
+ */
+__be32 skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto)
+{
+	int poff = proto_ports_offset(ip_proto);
+
+	if (poff >= 0) {
+		__be32 *ports, _ports;
+
+		ports = skb_header_pointer(skb, thoff + poff, sizeof(_ports),
+					   &_ports);
+		if (ports)
+			return *ports;
+	}
+
+	return 0;
+}
+EXPORT_SYMBOL(skb_flow_get_ports);
+
 bool skb_flow_dissect(const struct sk_buff *skb, struct flow_keys *flow)
 {
-	int poff, nhoff = skb_network_offset(skb);
+	int nhoff = skb_network_offset(skb);
 	u8 ip_proto;
 	__be16 proto = skb->protocol;
 
@@ -150,16 +176,7 @@ ipv6:
 	}
 
 	flow->ip_proto = ip_proto;
-	poff = proto_ports_offset(ip_proto);
-	if (poff >= 0) {
-		__be32 *ports, _ports;
-
-		nhoff += poff;
-		ports = skb_header_pointer(skb, nhoff, sizeof(_ports), &_ports);
-		if (ports)
-			flow->ports = *ports;
-	}
-
+	flow->ports = skb_flow_get_ports(skb, nhoff, ip_proto);
 	flow->thoff = (u16) nhoff;
 
 	return true;
Signature · 1 line
1.8.1.4
#2 lore raw

Re: [PATCH net-next v4 1/3] flow_dissector: factor out the ports extraction in skb_flow_get_ports

Quoted 5 lines · Nikolay Aleksandrov
net/core/flow_dissector.c +0 -4
 	flow->ip_proto = ip_proto;
-	poff = proto_ports_offset(ip_proto);
-	if (poff >= 0) {
-		__be32 *ports, _ports;
-

net/core/flow_dissector.c ipv6: · line 182

Quoted 10 lines
net/core/flow_dissector.c +1 -6
-		nhoff += poff;
-		ports = skb_header_pointer(skb, nhoff, sizeof(_ports), &_ports);
-		if (ports)
-			flow->ports = *ports;
-	}
-
+	flow->ports = skb_flow_get_ports(skb, nhoff, ip_proto);
 	flow->thoff = (u16) nhoff;
 
 	return true;

The best way is to ask David to apply my fix in net tree, then your patch will be on top of it.

Mixing a fix and a factoring seems not right.

#3 lore raw

Re: [PATCH net-next v4 1/3] flow_dissector: factor out the ports extraction in skb_flow_get_ports

Quoted 16 lines · Eric Dumazet
Quoted 14 lines · Nikolay Aleksandrov
net/core/flow_dissector.c +1 -10
 	flow->ip_proto = ip_proto;
-	poff = proto_ports_offset(ip_proto);
-	if (poff >= 0) {
-		__be32 *ports, _ports;
-
-		nhoff += poff;
-		ports = skb_header_pointer(skb, nhoff, sizeof(_ports), &_ports);
-		if (ports)
-			flow->ports = *ports;
-	}
-
+	flow->ports = skb_flow_get_ports(skb, nhoff, ip_proto);
 	flow->thoff = (u16) nhoff;
Quoted 10 lines
Quoted 1 line

return true;

The best way is to ask David to apply my fix in net tree, then your patch will be on top of it.

Mixing a fix and a factoring seems not right.

Ah, okay, sorry about that, I'm hurrying again :-) Dave would you apply Eric's fix in the net-next tree ? I'll then re-base on it.

Thanks,
Nik

maintainer
#4 lore raw

Re: [PATCH net-next v4 1/3] flow_dissector: factor out the ports extraction in skb_flow_get_ports

From: Nikolay Aleksandrov <[email protected]>
Date: Thu, 26 Sep 2013 18:35:33 +0200

Quoted 2 lines

Dave would you apply Eric's fix in the net-next tree ? I'll then re-base on it.

I just merged net into net-next, please rebase.

Thanks.

#5 lore raw

PATCH v4 2/3 bonding: modify the old and add new xmit hash policies

In reply to Nikolay Aleksandrov

This patch adds two new hash policy modes which use skb_flow_dissect:
3 - Encapsulated layer 2+3
4 - Encapsulated layer 3+4
There should be a good improvement for tunnel users in those modes. It also changes the old hash functions to:
hash ^= (__force u32)flow.dst ^ (__force u32)flow.src; hash ^= (hash >> 16);
hash ^= (hash >> 8);

Where hash will be initialized either to L2 hash, that is SRCMAC[5] XOR DSTMAC[5], or to flow->ports which should be extracted from the upper layer. Flow's dst and src are also extracted based on the xmit policy either directly from the buffer or by using skb_flow_dissect, but in both cases if the protocol is IPv6 then dst and src are obtained by ipv6_addr_hash() on the real addresses. In case of a non-dissectable packet, the algorithms fall back to L2 hashing.
The bond_set_mode_ops() function is now obsolete and thus deleted because it was used only to set the proper hash policy. Also we trim a pointer from struct bonding because we no longer need to keep the hash function, now there's only a single hash function - bond_xmit_hash that works based on bond->params.xmit_policy.

The hash function and skb_flow_dissect were suggested by Eric Dumazet. The layer names were suggested by Andy Gospodarek, because I suck at semantics.

---
v2: fix a bug in bond_flow_dissect which might've caused the use of
uninitalized flow_keys and make use of skb_flow_get_ports
v3, v4: no change
One line is intentionally left at 82 chars since it's the whole function and IMO looks better that way.

 drivers/net/bonding/bond_3ad.c   |   2 +-
 drivers/net/bonding/bond_main.c  | 197 ++++++++++++++-------------------------
 drivers/net/bonding/bond_sysfs.c |   2 -
 drivers/net/bonding/bonding.h    |   3 +-
 include/uapi/linux/if_bonding.h  |   2 +
 5 files changed, 72 insertions(+), 134 deletions(-)
drivers/net/bonding/bond_3ad.c +1 -1
index 0d8f427..b3ab703 100644
@@ -2442,7 +2442,7 @@ int bond_3ad_xmit_xor(struct sk_buff *skb, struct net_device *dev)
 		goto out;
 	}
 
-	slave_agg_no = bond->xmit_hash_policy(skb, slaves_in_agg);
+	slave_agg_no = bond_xmit_hash(bond, skb, slaves_in_agg);
 
 	bond_for_each_slave(bond, slave) {
 		struct aggregator *agg = SLAVE_AD_INFO(slave).port.aggregator;
drivers/net/bonding/bond_main.c +68 -129

269 lines not on this page · raw message

drivers/net/bonding/bond_sysfs.c +0 -2
index c29b836..dba3b9b 100644
@@ -352,7 +352,6 @@ static ssize_t bonding_store_mode(struct device *d,
 	/* don't cache arp_validate between modes */
 	bond->params.arp_validate = BOND_ARP_VALIDATE_NONE;
 	bond->params.mode = new_value;
-	bond_set_mode_ops(bond, bond->params.mode);
 	pr_info("%s: setting mode to %s (%d).\n",
 		bond->dev->name, bond_mode_tbl[new_value].modename,
 		new_value);
@@ -392,7 +391,6 @@ static ssize_t bonding_store_xmit_hash(struct device *d,
 		ret = -EINVAL;
 	} else {
 		bond->params.xmit_policy = new_value;
-		bond_set_mode_ops(bond, bond->params.mode);
 		pr_info("%s: setting xmit hash policy to %s (%d).\n",
 			bond->dev->name,
 			xmit_hashtype_tbl[new_value].modename, new_value);
drivers/net/bonding/bonding.h +1 -2
index 03cf3fd..4db9ec4 100644
@@ -245,7 +245,6 @@ struct bonding {
 	char     proc_file_name[IFNAMSIZ];
 #endif /* CONFIG_PROC_FS */
 	struct   list_head bond_list;
-	int      (*xmit_hash_policy)(struct sk_buff *, int);
 	u16      rr_tx_counter;
 	struct   ad_bond_info ad_info;
 	struct   alb_bond_info alb_info;
@@ -446,7 +445,7 @@ int bond_release(struct net_device *bond_dev, struct net_device *slave_dev);
 void bond_mii_monitor(struct work_struct *);
 void bond_loadbalance_arp_mon(struct work_struct *);
 void bond_activebackup_arp_mon(struct work_struct *);
-void bond_set_mode_ops(struct bonding *bond, int mode);
+int bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, int count);
 int bond_parse_parm(const char *mode_arg, const struct bond_parm_tbl *tbl);
 void bond_select_active_slave(struct bonding *bond);
 void bond_change_active_slave(struct bonding *bond, struct slave *new_active);
include/uapi/linux/if_bonding.h +2 -0
index a17edda..9635a62 100644
@@ -91,6 +91,8 @@
 #define BOND_XMIT_POLICY_LAYER2		0 /* layer 2 (MAC only), default */
 #define BOND_XMIT_POLICY_LAYER34	1 /* layer 3+4 (IP ^ (TCP || UDP)) */
 #define BOND_XMIT_POLICY_LAYER23	2 /* layer 2+3 (IP ^ MAC) */
+#define BOND_XMIT_POLICY_ENCAP23	3 /* encapsulated layer 2+3 */
+#define BOND_XMIT_POLICY_ENCAP34	4 /* encapsulated layer 3+4 */
 
 typedef struct ifbond {
 	__s32 bond_mode;
Signature · 1 line
1.8.1.4
#6 lore raw

PATCH v4 3/3 bonding: document the new xmit policy modes and update the changed ones

In reply to Nikolay Aleksandrov

Add new documentation for encap2+3 and encap3+4, also update the formula for the old modes due to the changes.

---
v2, v3, v4: no change

 Documentation/networking/bonding.txt | 66 ++++++++++++++++++++----------------
 1 file changed, 36 insertions(+), 30 deletions(-)
Documentation/networking/bonding.txt +36 -30
index 9b28e71..3856ed2 100644
@@ -743,21 +743,16 @@ xmit_hash_policy
 		protocol information to generate the hash.
 
 		Uses XOR of hardware MAC addresses and IP addresses to
-		generate the hash.  The IPv4 formula is
+		generate the hash.  The formula is
 
-		(((source IP XOR dest IP) AND 0xffff) XOR
-			( source MAC XOR destination MAC ))
-				modulo slave count
+		hash = source MAC XOR destination MAC
+		hash = hash XOR source IP XOR destination IP
+		hash = hash XOR (hash RSHIFT 16)
+		hash = hash XOR (hash RSHIFT 8)
+		And then hash is reduced modulo slave count.
 
-		The IPv6 formula is
-
-		hash = (source ip quad 2 XOR dest IP quad 2) XOR
-		       (source ip quad 3 XOR dest IP quad 3) XOR
-		       (source ip quad 4 XOR dest IP quad 4)
-
-		(((hash >> 24) XOR (hash >> 16) XOR (hash >> 8) XOR hash)
-			XOR (source MAC XOR destination MAC))
-				modulo slave count
+		If the protocol is IPv6 then the source and destination
+		addresses are first hashed using ipv6_addr_hash.
 
 		This algorithm will place all traffic to a particular
 		network peer on the same slave.  For non-IP traffic,
@@ -779,21 +774,16 @@ xmit_hash_policy
 		slaves, although a single connection will not span
 		multiple slaves.
 
-		The formula for unfragmented IPv4 TCP and UDP packets is
-
-		((source port XOR dest port) XOR
-			 ((source IP XOR dest IP) AND 0xffff)
-				modulo slave count
+		The formula for unfragmented TCP and UDP packets is
 
-		The formula for unfragmented IPv6 TCP and UDP packets is
+		hash = source port, destination port (as in the header)
+		hash = hash XOR source IP XOR destination IP
+		hash = hash XOR (hash RSHIFT 16)
+		hash = hash XOR (hash RSHIFT 8)
+		And then hash is reduced modulo slave count.
 
-		hash = (source port XOR dest port) XOR
-		       ((source ip quad 2 XOR dest IP quad 2) XOR
-			(source ip quad 3 XOR dest IP quad 3) XOR
-			(source ip quad 4 XOR dest IP quad 4))
-
-		((hash >> 24) XOR (hash >> 16) XOR (hash >> 8) XOR hash)
-			modulo slave count
+		If the protocol is IPv6 then the source and destination
+		addresses are first hashed using ipv6_addr_hash.
 
 		For fragmented TCP or UDP packets and all other IPv4 and
 		IPv6 protocol traffic, the source and destination port
@@ -801,10 +791,6 @@ xmit_hash_policy
 		formula is the same as for the layer2 transmit hash
 		policy.
 
-		The IPv4 policy is intended to mimic the behavior of
-		certain switches, notably Cisco switches with PFC2 as
-		well as some Foundry and IBM products.
-
 		This algorithm is not fully 802.3ad compliant.  A
 		single TCP or UDP conversation containing both
 		fragmented and unfragmented packets will see packets
@@ -815,6 +801,26 @@ xmit_hash_policy
 		conversations.  Other implementations of 802.3ad may
 		or may not tolerate this noncompliance.
 
+	encap2+3
+
+		This policy uses the same formula as layer2+3 but it
+		relies on skb_flow_dissect to obtain the header fields
+		which might result in the use of inner headers if an
+		encapsulation protocol is used. For example this will
+		improve the performance for tunnel users because the
+		packets will be distributed according to the encapsulated
+		flows.
+
+	encap3+4
+
+		This policy uses the same formula as layer3+4 but it
+		relies on skb_flow_dissect to obtain the header fields
+		which might result in the use of inner headers if an
+		encapsulation protocol is used. For example this will
+		improve the performance for tunnel users because the
+		packets will be distributed according to the encapsulated
+		flows.
+
 	The default value is layer2.  This option was added in bonding
 	version 2.6.3.  In earlier versions of bonding, this parameter
 	does not exist, and the layer2 policy is the only policy.  The
Signature · 1 line
1.8.1.4