DSA: promisc on master, generic flow dissector code
Subsystems (9)
Vladimir Oltean
maintainercover DSA: promisc on master, generic flow dissector code
From: Vladimir Oltean <[email protected]>
The initial purpose of this series was to implement the .flow_dissect method for sja1105 and for ocelot/felix. But on Felix this posed a problem, as the DSA headers were of different lengths on RX and on TX. A better solution than to just increase the smaller one was to also try to shrink the larger one, but in turn that required the DSA master to be put in promiscuous mode (which sja1105 also needed, for other reasons).
Finally, we can add the missing .flow_dissect methods to ocelot and sja1105 (as well as generalize the formula to other taggers as well).
Vladimir Oltean (4):
net: dsa: allow drivers to request promiscuous mode on master
net: dsa: sja1105: request promiscuous mode for master
net: dsa: tag_ocelot: use a short prefix on both ingress and egress
net: dsa: implement and use a generic procedure for the flow dissector
drivers/net/dsa/ocelot/felix.c | 5 ++-- drivers/net/dsa/sja1105/sja1105_main.c | 3 ++ include/net/dsa.h | 13 +++++++++ net/dsa/dsa_priv.h | 22 +++++++++++++++ net/dsa/master.c | 39 +++++++++++++++++++++++++- net/dsa/tag_ar9331.c | 9 ++++++ net/dsa/tag_brcm.c | 5 ++-- net/dsa/tag_dsa.c | 4 +-- net/dsa/tag_edsa.c | 4 +-- net/dsa/tag_lan9303.c | 9 ++++++ net/dsa/tag_mtk.c | 3 +- net/dsa/tag_ocelot.c | 29 +++++++++++++++---- net/dsa/tag_qca.c | 3 +- net/dsa/tag_sja1105.c | 13 +++++++++ 14 files changed, 143 insertions(+), 18 deletions(-)
Signature · 1 line
2.17.1
Vladimir Oltean
maintainerPATCH 1/4 net: dsa: allow drivers to request promiscuous mode on master
From: Vladimir Oltean <[email protected]>
Currently DSA assumes that taggers don't mess with the destination MAC address of the frames on RX. That is not always the case. Some DSA headers are placed before the Ethernet header (ocelot), and others simply mangle random bytes from the destination MAC address (sja1105 with its incl_srcpt option).
Currently the DSA master goes to promiscuous mode automatically when the slave devices go too (such as when enslaved to a bridge), but in standalone mode this is a problem that needs to be dealt with.
So give drivers the possibility to signal that their tagging protocol will get randomly dropped otherwise, and let DSA deal with fixing that.
--- include/net/dsa.h | 13 +++++++++++++ net/dsa/master.c | 39 ++++++++++++++++++++++++++++++++++++++- 2 files changed, 51 insertions(+), 1 deletion(-)
include/net/dsa.h +13 -0
@@ -217,6 +217,12 @@ struct dsa_port {*/ const struct net_device_ops *orig_ndo_ops; +/* +* Original master netdev flags in case we need to put it in +* promiscuous mode +*/ +unsigned int orig_master_flags; +bool setup; };
@@ -298,6 +304,13 @@ struct dsa_switch {*/ bool mtu_enforcement_ingress; +/* Some tagging protocols either mangle or shift the destination MAC +* address, in which case the DSA master would drop packets on ingress +* if what it understands out of the destination MAC address is not in +* its RX filter. +*/ +bool promisc_on_master; +size_t num_ports; };
net/dsa/master.c +38 -1
@@ -294,6 +294,37 @@ static void dsa_master_ndo_teardown(struct net_device *dev) cpu_dp->orig_ndo_ops =NULL ; } +static void dsa_master_set_promisc (struct net_device *dev) +{ +struct dsa_port *cpu_dp = dev->dsa_ptr; +struct dsa_switch *ds = cpu_dp->ds; +unsigned int flags; + +if (!ds->promisc_on_master) +return ; + + flags =dev_get_flags (dev); + + cpu_dp->orig_master_flags = flags; + +rtnl_lock (); +dev_change_flags (dev, flags |IFF_PROMISC ,NULL ); +rtnl_unlock (); +} + +static void dsa_master_reset_promisc (struct net_device *dev) +{ +struct dsa_port *cpu_dp = dev->dsa_ptr; +struct dsa_switch *ds = cpu_dp->ds; + +if (!ds->promisc_on_master) +return ; + +rtnl_lock (); +dev_change_flags (dev, cpu_dp->orig_master_flags,NULL ); +rtnl_unlock (); +} +static ssize_t tagging_show (struct device *d,struct device_attribute *attr,char *buf) {
@@ -345,9 +376,12 @@ int dsa_master_setup(struct net_device *dev, struct dsa_port *cpu_dp)wmb (); dev->dsa_ptr = cpu_dp; + +dsa_master_set_promisc (dev); + ret =dsa_master_ethtool_setup (dev);if (ret) -return ret; + goto out_err_reset_promisc; ret =dsa_master_ndo_setup (dev);if (ret)
@@ -363,6 +397,8 @@ int dsa_master_setup(struct net_device *dev, struct dsa_port *cpu_dp)dsa_master_ndo_teardown (dev); out_err_ethtool_teardown:dsa_master_ethtool_teardown (dev); +out_err_reset_promisc: +dsa_master_reset_promisc (dev);return ret; }
@@ -372,6 +408,7 @@ void dsa_master_teardown(struct net_device *dev)dsa_master_ndo_teardown (dev);dsa_master_ethtool_teardown (dev);dsa_master_reset_mtu (dev); +dsa_master_reset_promisc (dev); dev->dsa_ptr =NULL ;
Signature · 1 line
2.17.1
Re: [PATCH net-next 1/4] net: dsa: allow drivers to request promiscuous mode on master
Quoted 8 lines · Vladimir Oltean
From: Vladimir Oltean <[email protected]>
Currently DSA assumes that taggers don't mess with the destination MAC address of the frames on RX. That is not always the case. Some DSA headers are placed before the Ethernet header (ocelot), and others simply mangle random bytes from the destination MAC address (sja1105 with its
incl_srcptoption).
Quoted 8 lines
Currently the DSA master goes to promiscuous mode automatically when the slave devices go too (such as when enslaved to a bridge), but in standalone mode this is a problem that needs to be dealt with.
So give drivers the possibility to signal that their tagging protocol will get randomly dropped otherwise, and let DSA deal with fixing that.
Signed-off-by Vladimir Oltean <[email protected]>
Signature · 1 line
Florian
Vladimir Oltean
maintainerPATCH 2/4 net: dsa: sja1105: request promiscuous mode for master
In reply to Vladimir OlteanFrom: Vladimir Oltean <[email protected]>
Currently PTP is broken when ports are in standalone mode (the tagger keeps printing this message):
sja1105 spi0.1: Expected meta frame, is 180c200000e in the DSA master multicast filter?
Sure, one might say "simply add 01-80-c2-00-00-0e to the master's RX filter" but things become more complicated because:
- Actually all frames in the 01-80-c2-xx-xx-xx and 01-1b-19-xx-xx-xx range are trapped to the CPU automatically
- The switch mangles bytes 3 and 4 of the MAC address via the incl_srcpt ("include source port [in the DMAC]") option, so an address installed to the RX filter would, at the end of the day, not correspond to the final address seen by the DSA master.
Assume RX filters on DSA masters are typically too small to include all necessary addresses for PTP to work properly on sja1105, and just request promiscuous mode unconditionally.
--- drivers/net/dsa/sja1105/sja1105_main.c | 3 +++ 1 file changed, 3 insertions(+)
drivers/net/dsa/sja1105/sja1105_main.c +3 -0
@@ -2080,6 +2080,9 @@ static int sja1105_setup(struct dsa_switch *ds)dev_err (ds->dev,"Failed to configure MII clocking: %d\n" , rc);return rc; } + + ds->promisc_on_master = true; +/* On SJA1105, VLAN filtering per se is always enabled in hardware. * The only thing we can do to disable it is lie about what the 802.1Q * EtherType is.
Signature · 1 line
2.17.1
Re: [PATCH net-next 2/4] net: dsa: sja1105: request promiscuous mode for master
Quoted 17 lines · Vladimir Oltean
From: Vladimir Oltean <[email protected]>
Currently PTP is broken when ports are in standalone mode (the tagger keeps printing this message):
sja1105 spi0.1: Expected meta frame, is
180c200000ein the DSA master multicast filter?Sure, one might say "simply add 01-80-c2-00-00-0e to the master's RX filter" but things become more complicated because:
- Actually all frames in the 01-80-c2-xx-xx-xx and 01-1b-19-xx-xx-xx range are trapped to the CPU automatically
- The switch mangles bytes 3 and 4 of the MAC address via theincl_srcpt("include source port [in the DMAC]") option, so an address installed to the RX filter would, at the end of the day, not correspond to the final address seen by the DSA master.
Quoted 5 lines
Assume RX filters on DSA masters are typically too small to include all necessary addresses for PTP to work properly on sja1105, and just request promiscuous mode unconditionally.
Signed-off-by Vladimir Oltean <[email protected]>
Signature · 1 line
Florian
Vladimir Oltean
maintainerPATCH 3/4 net: dsa: tag_ocelot: use a short prefix on both ingress and egress
In reply to Vladimir OlteanFrom: Vladimir Oltean <[email protected]>
There are 2 goals that we follow:
- Reduce the header size
- Make the header size equal between RX and TX
The issue that required long prefix on RX was the fact that the ocelot DSA tag, being put before Ethernet as it is, would overlap with the area that a DSA master uses for RX filtering (destination MAC address mainly).
Now that we can ask DSA to put the master in promiscuous mode, in theory we could remove the prefix altogether and call it a day, but it looks like we can't. Using no prefix on ingress, some packets (such as ICMP) would be received, while others (such as PTP) would not be received. This is because the DSA master we use (enetc) triggers parse errors ("MAC rx frame errors") presumably because it sees Ethernet frames with a bad length. And indeed, when using no prefix, the EtherType (bytes 12-13 of the frame, bits 96-111) falls over the REW_VAL field from the extraction header, aka the PTP timestamp.
When turning the short (32-bit) prefix on, the EtherType overlaps with bits 64-79 of the extraction header, which are a reserved area transmitted as zero by the switch. The packets are not dropped by the DSA master with a short prefix. Actually, the frames look like this in tcpdump (below is a PTP frame, with an extra dsa_8021q tag - dadb 0482 - added by a downstream sja1105).
89:0c:a9:f2:01:00 > 88:80:00:0a:00:1d, 802.3, length 0: LLC, \ dsap Unknown (0x10) Individual, ssap ProWay NM (0x0e) Response, \ ctrl 0x0004: Information, send seq 2, rcv seq 0, \ Flags [Response], length 78
0x0000: 8880 000a 001d 890c a9f2 0100 0000 100f ................
0x0010: 0400 0000 0180 c200 000e 001f 7b63 0248 ............{c.H
0x0020: dadb 0482 88f7 1202 0036 0000 0000 0000 .........6......
0x0030: 0000 0000 0000 0000 0000 001f 7bff fe63 ............{..c
0x0040: 0248 0001 1f81 0500 0000 0000 0000 0000 .H..............
0x0050: 0000 0000 0000 0000 0000 0000 ............So the short prefix is our new default: we've shortened our RX frames by 12 octets, increased TX by 4, and headers are now equal between RX and TX. Note that we still need promiscuous mode for the DSA master to not drop it.
--- drivers/net/dsa/ocelot/felix.c | 5 +++-- net/dsa/tag_ocelot.c | 20 +++++++++++++++----- 2 files changed, 18 insertions(+), 7 deletions(-)
drivers/net/dsa/ocelot/felix.c +3 -2
@@ -545,8 +545,8 @@ static int felix_setup(struct dsa_switch *ds)/* Bring up the CPU port module and configure the NPI port */ if (dsa_is_cpu_port (ds, port))ocelot_configure_cpu (ocelot, port, -OCELOT_TAG_PREFIX_NONE , -OCELOT_TAG_PREFIX_LONG ); +OCELOT_TAG_PREFIX_SHORT , +OCELOT_TAG_PREFIX_SHORT ); }/* Include the CPU port module in the forwarding mask for unknown
@@ -564,6 +564,7 @@ static int felix_setup(struct dsa_switch *ds)ANA_FLOODING_FLD_UNICAST (PGID_UC ), ANA_FLOODING, tc); + ds->promisc_on_master = true; ds->mtu_enforcement_ingress = true;/* It looks like the MAC/PCS interrupt register - PM0_IEVENT (0x8040) * isn't instantiated for the Felix PF.
net/dsa/tag_ocelot.c +15 -5
@@ -5,6 +5,9 @@#include <linux/packing.h> #include "dsa_priv.h" +#define OCELOT_PREFIX_VAL 0x8880000a +#define OCELOT_TOTAL_TAG_LEN (OCELOT_SHORT_PREFIX_LEN + OCELOT_TAG_LEN) +/* The CPU injection header and the CPU extraction header can have 3 types of * prefixes: long, short and no prefix. The format of the header itself is the * same in all 3 cases.
@@ -143,8 +146,11 @@ static struct sk_buff *ocelot_xmit(struct sk_buff *skb,struct ocelot *ocelot = ds->priv;struct ocelot_port *ocelot_port = ocelot->ports[port];u8 *injection; +u32 *prefix; +int rc; -if (unlikely (skb_cow_head (skb,OCELOT_TAG_LEN ) <0 )) { + rc =skb_cow_head (skb,OCELOT_TOTAL_TAG_LEN ); +if (unlikely (rc <0 )) {netdev_err (netdev,"Cannot make room for tag.\n" );return NULL ; }
@@ -174,6 +180,10 @@ static struct sk_buff *ocelot_xmit(struct sk_buff *skb,packing (injection, &rew_op,125 ,117 ,OCELOT_TAG_LEN ,PACK ,0 ); } + prefix =skb_push (skb,OCELOT_SHORT_PREFIX_LEN ); + + *prefix =cpu_to_be32 (OCELOT_PREFIX_VAL ); +return skb; }
@@ -189,11 +199,11 @@ static struct sk_buff *ocelot_rcv(struct sk_buff *skb,* so it points to the beginning of the frame. */ skb_push (skb,ETH_HLEN ); -/* We don't care about the long prefix, it is just for easy entrance +/* We don't care about the short prefix, it is just for easy entrance * into the DSA master's RX filter. Discard it now by moving it into * the headroom. */ -skb_pull (skb,OCELOT_LONG_PREFIX_LEN ); +skb_pull (skb,OCELOT_SHORT_PREFIX_LEN );/* And skb->data now points to the extraction frame header. * Keep a pointer to it. */
@@ -207,7 +217,7 @@ static struct sk_buff *ocelot_rcv(struct sk_buff *skb,skb_pull (skb,ETH_HLEN );/* Remove from inet csum the extraction header */ -skb_postpull_rcsum (skb, start,OCELOT_LONG_PREFIX_LEN +OCELOT_TAG_LEN ); +skb_postpull_rcsum (skb, start,OCELOT_TOTAL_TAG_LEN );packing (extraction, &src_port,46 ,43 ,OCELOT_TAG_LEN ,UNPACK ,0 );packing (extraction, &qos_class,19 ,17 ,OCELOT_TAG_LEN ,UNPACK ,0 );
@@ -233,7 +243,7 @@ static struct dsa_device_ops ocelot_netdev_ops = { .proto = DSA_TAG_PROTO_OCELOT, .xmit = ocelot_xmit, .rcv = ocelot_rcv, - .overhead = OCELOT_TAG_LEN + OCELOT_LONG_PREFIX_LEN, + .overhead = OCELOT_TOTAL_TAG_LEN, };MODULE_LICENSE ("GPL v2" );
Signature · 1 line
2.17.1
Jakub Kicinski
maintainerRe: [PATCH net-next 3/4] net: dsa: tag_ocelot: use a short prefix on both ingress and egress
Quoted 6 lines · On Mon, 11 May 2020 23:20:45 +0300 Vladimir Oltean wrote:
From: Vladimir Oltean <[email protected]>
There are 2 goals that we follow:
- Reduce the header size
- Make the header size equal between RX and TX
Getting this from sparse:
../net/dsa/tag_ocelot.c:185:17: warning: incorrect type in assignment (different base types)../net/dsa/tag_ocelot.c:185:17: expected unsigned int [usertype]../net/dsa/tag_ocelot.c:185:17: got restricted __be32 [usertype]
Vladimir Oltean
maintainerRe: [PATCH net-next 3/4] net: dsa: tag_ocelot: use a short prefix on both ingress and egress
Quoted 6 lines · Jakub Kicinski
Quoted 4 lines · On Mon, 11 May 2020 23:20:45 +0300 Vladimir Oltean wrote:
From: Vladimir Oltean <[email protected]>
There are 2 goals that we follow:
Quoted 8 lines
Quoted 2 lines
- Reduce the header size
- Make the header size equal between RX and TXGetting this from sparse:
../net/dsa/tag_ocelot.c:185:17: warning: incorrect type in assignment (different base types)../net/dsa/tag_ocelot.c:185:17: expected unsigned int [usertype]../net/dsa/tag_ocelot.c:185:17: got restricted__be32[usertype]
I hate this warning :(
David Miller
maintainerRe: [PATCH net-next 3/4] net: dsa: tag_ocelot: use a short prefix on both ingress and egress
From: Vladimir Oltean <[email protected]>
Date: Tue, 12 May 2020 01:44:53 +0300
Quoted 10 lines · Jakub Kicinski
Quoted 9 lines · On Mon, 11 May 2020 23:20:45 +0300 Vladimir Oltean wrote:
Quoted 6 lines
From: Vladimir Oltean <[email protected]>
There are 2 goals that we follow:
- Reduce the header size
- Make the header size equal between RX and TX
Quoted 7 lines
Quoted 5 lines
Getting this from sparse:
../net/dsa/tag_ocelot.c:185:17: warning: incorrect type in assignment (different base types)../net/dsa/tag_ocelot.c:185:17: expected unsigned int [usertype]../net/dsa/tag_ocelot.c:185:17: got restricted__be32[usertype]I hate this warning :(
You hate that endianness bugs are caught automatically? :-)
Vladimir Oltean
maintainerRe: [PATCH net-next 3/4] net: dsa: tag_ocelot: use a short prefix on both ingress and egress
Quoted 16 lines · David Miller
From: Vladimir Oltean <[email protected]>
Date: Tue, 12 May 2020 01:44:53 +0300Quoted 12 lines · Jakub Kicinski
Quoted 11 lines · On Mon, 11 May 2020 23:20:45 +0300 Vladimir Oltean wrote:
Quoted 6 lines
From: Vladimir Oltean <[email protected]>
There are 2 goals that we follow:
- Reduce the header size
- Make the header size equal between RX and TXGetting this from sparse:
Quoted 8 lines
Quoted 5 lines
Quoted 3 lines
../net/dsa/tag_ocelot.c:185:17: warning: incorrect type in assignment (different base types)../net/dsa/tag_ocelot.c:185:17: expected unsigned int [usertype]../net/dsa/tag_ocelot.c:185:17: got restricted__be32[usertype]I hate this warning :(
You hate that endianness bugs are caught automatically? :-)
Well, there's no bug here, really, what's annoying is that it's warning me when a convention is not being followed. It would have been smarter if it just limited itself to the situations when not following that convention actually causes a problem.
Re: [PATCH net-next 3/4] net: dsa: tag_ocelot: use a short prefix on both ingress and egress
In reply to Vladimir OlteanQuoted 41 lines · Vladimir Oltean
From: Vladimir Oltean <[email protected]>
There are 2 goals that we follow:
- Reduce the header size
- Make the header size equal between RX and TXThe issue that required long prefix on RX was the fact that the ocelot DSA tag, being put before Ethernet as it is, would overlap with the area that a DSA master uses for RX filtering (destination MAC address mainly).
Now that we can ask DSA to put the master in promiscuous mode, in theory we could remove the prefix altogether and call it a day, but it looks like we can't. Using no prefix on ingress, some packets (such as ICMP) would be received, while others (such as PTP) would not be received. This is because the DSA master we use (enetc) triggers parse errors ("MAC rx frame errors") presumably because it sees Ethernet frames with a bad length. And indeed, when using no prefix, the EtherType (bytes 12-13 of the frame, bits 96-111) falls over the
REW_VALfield from the extraction header, aka the PTP timestamp.When turning the short (32-bit) prefix on, the EtherType overlaps with bits 64-79 of the extraction header, which are a reserved area transmitted as zero by the switch. The packets are not dropped by the DSA master with a short prefix. Actually, the frames look like this in tcpdump (below is a PTP frame, with an extra
dsa_8021qtag - dadb 0482 - added by a downstream sja1105).89:0c:a9:f2:01:00 > 88:80:00:0a:00:1d, 802.3, length 0: LLC, \ dsap Unknown (0x10) Individual, ssap ProWay NM (0x0e) Response, \ ctrl 0x0004: Information, send seq 2, rcv seq 0, \ Flags [Response], length 780x0000: 8880 000a 001d 890c a9f2 0100 0000 100f ................ 0x0010: 0400 0000 0180 c200 000e 001f 7b63 0248 ............{c.H 0x0020: dadb 0482 88f7 1202 0036 0000 0000 0000 .........6...... 0x0030: 0000 0000 0000 0000 0000 001f 7bff fe63 ............{..c 0x0040: 0248 0001 1f81 0500 0000 0000 0000 0000 .H.............. 0x0050: 0000 0000 0000 0000 0000 0000 ............
Quoted 6 lines
So the short prefix is our new default: we've shortened our RX frames by 12 octets, increased TX by 4, and headers are now equal between RX and TX. Note that we still need promiscuous mode for the DSA master to not drop it.
Signed-off-by Vladimir Oltean <[email protected]>
Signature · 1 line
Florian
Vladimir Oltean
maintainerPATCH 4/4 net: dsa: implement and use a generic procedure for the flow dissector
In reply to Vladimir OlteanFrom: Vladimir Oltean <[email protected]>
For all DSA formats that don't use tail tags, it looks like behind the obscure number crunching they're all doing the same thing: locating the real EtherType behind the DSA tag. Nonetheless, this is not immediately obvious, so create a generic helper for those DSA taggers that put the header before the EtherType.
Another assumption for the generic function is that the DSA tags are of equal length on RX and on TX. Prior to the previous patch, this was not true for ocelot and for gswip. The problem was resolved for ocelot, but for gswip it still remains, so that hasn't been converted.
--- net/dsa/dsa_priv.h | 22 ++++++++++++++++++++++ net/dsa/tag_ar9331.c | 9 +++++++++ net/dsa/tag_brcm.c | 5 +++-- net/dsa/tag_dsa.c | 4 ++-- net/dsa/tag_edsa.c | 4 ++-- net/dsa/tag_lan9303.c | 9 +++++++++ net/dsa/tag_mtk.c | 3 +-- net/dsa/tag_ocelot.c | 9 +++++++++ net/dsa/tag_qca.c | 3 +-- net/dsa/tag_sja1105.c | 13 +++++++++++++ 10 files changed, 71 insertions(+), 10 deletions(-)
net/dsa/dsa_priv.h +22 -0
@@ -195,6 +195,28 @@ dsa_slave_to_master(const struct net_device *dev)return dp->cpu_dp->master; } +/* All DSA tags that push the EtherType to the right (basically all except tail +* tags, which don't break dissection) can be treated the same from the +* perspective of the flow dissector. +* +* We need to return: +* - offset: the (B - A) difference between: +* A. the position of the real EtherType and +* B. the current skb->data (aka ETH_HLEN bytes into the frame, aka 2 bytes +* after the normal EtherType was supposed to be) +* The offset in bytes is exactly equal to the tagger overhead (and half of +* that, in __be16 shorts). +* +* - proto: the value of the real EtherType. +*/ +static inline void dsa_tag_generic_flow_dissect (const struct sk_buff *skb, +__be16 *proto,int *offset, +int tag_len) +{ + *offset = tag_len; + *proto = ((__be16 *)skb->data)[(tag_len /2 ) -1 ]; +} +/* switch.c */ int dsa_switch_register_notifier (struct dsa_switch *ds);void dsa_switch_unregister_notifier (struct dsa_switch *ds);
net/dsa/tag_ar9331.c +9 -0
@@ -83,12 +83,21 @@ static struct sk_buff *ar9331_tag_rcv(struct sk_buff *skb,return skb; } +static int ar9331_flow_dissect (const struct sk_buff *skb,__be16 *proto, +int *offset) +{ +dsa_tag_generic_flow_dissect (skb, proto, offset,AR9331_HDR_LEN ); + +return 0 ; +} +static const struct dsa_device_ops ar9331_netdev_ops = { .name ="ar9331" , .proto =DSA_TAG_PROTO_AR9331 , .xmit = ar9331_tag_xmit, .rcv = ar9331_tag_rcv, .overhead =AR9331_HDR_LEN , + .flow_dissect = ar9331_flow_dissect, };MODULE_LICENSE ("GPL v2" );
net/dsa/tag_brcm.c +3 -2
@@ -160,9 +160,10 @@ static int brcm_tag_flow_dissect(const struct sk_buff *skb, __be16 *proto,* skb->data points 2 bytes before the actual Ethernet type field and * we have an offset of 4bytes between where skb->data and where the * payload starts. +* So we can use the generic helper in both cases. */ - *offset =BRCM_TAG_LEN ; - *proto = ((__be16 *)skb->data)[1 ]; +dsa_tag_generic_flow_dissect (skb, proto, offset,BRCM_TAG_LEN ); + return 0; }#endif
net/dsa/tag_dsa.c +2 -2
@@ -145,8 +145,8 @@ static struct sk_buff *dsa_rcv(struct sk_buff *skb, struct net_device *dev,static int dsa_tag_flow_dissect (const struct sk_buff *skb,__be16 *proto,int *offset) { - *offset =4 ; - *proto = ((__be16 *)skb->data)[1 ]; +dsa_tag_generic_flow_dissect (skb, proto, offset,DSA_HLEN ); +return 0 ; }
net/dsa/tag_edsa.c +2 -2
@@ -164,8 +164,8 @@ static struct sk_buff *edsa_rcv(struct sk_buff *skb, struct net_device *dev,static int edsa_tag_flow_dissect (const struct sk_buff *skb,__be16 *proto,int *offset) { - *offset =8 ; - *proto = ((__be16 *)skb->data)[3 ]; +dsa_tag_generic_flow_dissect (skb, proto, offset,EDSA_HLEN ); +return 0 ; }
net/dsa/tag_lan9303.c +9 -0
@@ -128,12 +128,21 @@ static struct sk_buff *lan9303_rcv(struct sk_buff *skb, struct net_device *dev,return skb; } +static int lan9303_flow_dissect (const struct sk_buff *skb,__be16 *proto, +int *offset) +{ +dsa_tag_generic_flow_dissect (skb, proto, offset,LAN9303_TAG_LEN ); + +return 0 ; +} +static const struct dsa_device_ops lan9303_netdev_ops = { .name ="lan9303" , .proto =DSA_TAG_PROTO_LAN9303 , .xmit = lan9303_xmit, .rcv = lan9303_rcv, .overhead =LAN9303_TAG_LEN , + .flow_dissect = lan9303_flow_dissect, };MODULE_LICENSE ("GPL" );
net/dsa/tag_mtk.c +1 -2
@@ -92,8 +92,7 @@ static struct sk_buff *mtk_tag_rcv(struct sk_buff *skb, struct net_device *dev,static int mtk_tag_flow_dissect (const struct sk_buff *skb,__be16 *proto,int *offset) { - *offset =4 ; - *proto = ((__be16 *)skb->data)[1 ]; +dsa_tag_generic_flow_dissect (skb, proto, offset,MTK_HDR_LEN );return 0 ; }
net/dsa/tag_ocelot.c +9 -0
@@ -238,12 +238,21 @@ static struct sk_buff *ocelot_rcv(struct sk_buff *skb,return skb; } +static int ocelot_flow_dissect (const struct sk_buff *skb,__be16 *proto, +int *offset) +{ +dsa_tag_generic_flow_dissect (skb, proto, offset,OCELOT_TOTAL_TAG_LEN ); + +return 0 ; +} +static struct dsa_device_ops ocelot_netdev_ops = { .name ="ocelot" , .proto =DSA_TAG_PROTO_OCELOT , .xmit = ocelot_xmit, .rcv = ocelot_rcv, .overhead =OCELOT_TOTAL_TAG_LEN , + .flow_dissect = ocelot_flow_dissect, };MODULE_LICENSE ("GPL v2" );
net/dsa/tag_qca.c +1 -2
@@ -90,8 +90,7 @@ static struct sk_buff *qca_tag_rcv(struct sk_buff *skb, struct net_device *dev,static int qca_tag_flow_dissect (const struct sk_buff *skb,__be16 *proto,int *offset) { - *offset =QCA_HDR_LEN ; - *proto = ((__be16 *)skb->data)[0 ]; +dsa_tag_generic_flow_dissect (skb, proto, offset,QCA_HDR_LEN );return 0 ; }
net/dsa/tag_sja1105.c +13 -0
@@ -304,6 +304,18 @@ static struct sk_buff *sja1105_rcv(struct sk_buff *skb, is_meta); } +static int sja1105_flow_dissect (const struct sk_buff *skb,__be16 *proto, +int *offset) +{ +/* No tag added for management frames, all ok */ +if (unlikely (sja1105_is_link_local (skb))) +return 0 ; + +dsa_tag_generic_flow_dissect (skb, proto, offset,VLAN_HLEN ); + +return 0 ; +} +static struct dsa_device_ops sja1105_netdev_ops = { .name ="sja1105" , .proto =DSA_TAG_PROTO_SJA1105 ,
@@ -311,6 +323,7 @@ static struct dsa_device_ops sja1105_netdev_ops = { .rcv = sja1105_rcv, .filter = sja1105_filter, .overhead = VLAN_HLEN, + .flow_dissect = sja1105_flow_dissect, };MODULE_LICENSE ("GPL v2" );
Signature · 1 line
2.17.1
Re: [PATCH net-next 4/4] net: dsa: implement and use a generic procedure for the flow dissector
Quoted 8 lines · Vladimir Oltean
From: Vladimir Oltean <[email protected]>
For all DSA formats that don't use tail tags, it looks like behind the obscure number crunching they're all doing the same thing: locating the real EtherType behind the DSA tag. Nonetheless, this is not immediately obvious, so create a generic helper for those DSA taggers that put the header before the EtherType.
Quoted 6 lines
Another assumption for the generic function is that the DSA tags are of equal length on RX and on TX. Prior to the previous patch, this was not true for ocelot and for gswip. The problem was resolved for ocelot, but for gswip it still remains, so that hasn't been converted.
Signed-off-by Vladimir Oltean <[email protected]>
Given that __skb_flow_dissect() already de-references dsa_device_ops from skb->dev->dsa_ptr, maybe we can go one step further and just have dsa_tag_generic_flow_dissect obtain the overhead from the SKB directly since this will already have touched the cache lines involved. This then makes it unnecessary for the various taggers to specify a custom function and instead, dsa_tag_generic_flow_dissect() can be assigned where the dsa_device_ops are declared for the various tags. Did I miss something?
It also looks like tag_ocelot.c and tag_sja1105.c should have their dsa_device_ops structures const, as a separate patch certainly.
Signature · 1 line
Florian
Vladimir Oltean
maintainerRe: [PATCH net-next 4/4] net: dsa: implement and use a generic procedure for the flow dissector
Quoted 28 lines · Florian Fainelli
Quoted 8 lines · Vladimir Oltean
From: Vladimir Oltean <[email protected]>
For all DSA formats that don't use tail tags, it looks like behind the obscure number crunching they're all doing the same thing: locating the real EtherType behind the DSA tag. Nonetheless, this is not immediately obvious, so create a generic helper for those DSA taggers that put the header before the EtherType.
Quoted 6 lines
Another assumption for the generic function is that the DSA tags are of equal length on RX and on TX. Prior to the previous patch, this was not true for ocelot and for gswip. The problem was resolved for ocelot, but for gswip it still remains, so that hasn't been converted.
Signed-off-by Vladimir Oltean <[email protected]>Given that
__skb_flow_dissect()already de-referencesdsa_device_opsfromskb->dev->dsa_ptr, maybe we can go one step further and just havedsa_tag_generic_flow_dissectobtain the overhead from the SKB directly since this will already have touched the cache lines involved. This then makes it unnecessary for the various taggers to specify a custom function and instead,dsa_tag_generic_flow_dissect()can be assigned where thedsa_device_opsare declared for the various tags. Did I miss something?
Quoted 4 lines
It also looks like
tag_ocelot.candtag_sja1105.cshould have theirdsa_device_opsstructures const, as a separate patch certainly. --
Florian
Actually I wrote this patch in such a way that the assembly code generated would remain the same, since dsa_tag_generic_flow_dissect is inline, the arithmetic would be done at compile time, which it wouldn't be if this function were to look at the tagger .overhead. I don't know, I can do either way.
Quoted 11 lines · Vladimir Oltean
From: Vladimir Oltean <[email protected]>
The initial purpose of this series was to implement the .flow_dissect method for sja1105 and for ocelot/felix. But on Felix this posed a problem, as the DSA headers were of different lengths on RX and on TX. A better solution than to just increase the smaller one was to also try to shrink the larger one, but in turn that required the DSA master to be put in promiscuous mode (which sja1105 also needed, for other reasons).
Finally, we can add the missing .flow_dissect methods to ocelot and sja1105 (as well as generalize the formula to other taggers as well).
On a separate note, do you have any systems for which it would be desirable that the DSA standalone port implemented receive filtering? On BCM7278 devices, the Ethernet MAC connected to the switch is always in promiscuous mode, and so we rely on the switch not to flood the CPU port unnecessarily with MC traffic (if nothing else), this is currently implemented in our downstream kernel, but has not made it upstream yet, previous attempt was here:
https://www.spinics.net/lists/netdev/msg544361.html
I would like to revisit that at some point.
Signature · 1 line
Florian
Vladimir Oltean
maintainerQuoted 24 lines · Florian Fainelli
Quoted 11 lines · Vladimir Oltean
From: Vladimir Oltean <[email protected]>
The initial purpose of this series was to implement the .flow_dissect method for sja1105 and for ocelot/felix. But on Felix this posed a problem, as the DSA headers were of different lengths on RX and on TX. A better solution than to just increase the smaller one was to also try to shrink the larger one, but in turn that required the DSA master to be put in promiscuous mode (which sja1105 also needed, for other reasons).
Finally, we can add the missing .flow_dissect methods to ocelot and sja1105 (as well as generalize the formula to other taggers as well).
On a separate note, do you have any systems for which it would be desirable that the DSA standalone port implemented receive filtering? On BCM7278 devices, the Ethernet MAC connected to the switch is always in promiscuous mode, and so we rely on the switch not to flood the CPU port unnecessarily with MC traffic (if nothing else), this is currently implemented in our downstream kernel, but has not made it upstream yet, previous attempt was here:
Quoted 5 lines
https://www.spinics.net/lists/netdev/msg544361.html
I would like to revisit that at some point.
--
Florian
Yes, CPU filtering of traffic (not just multicast) is one of the problems we're facing. I'll take a look at your patches and maybe I'll pick them up.
Thanks,
-Vladimir
Quoted 27 lines · Vladimir Oltean
Quoted 26 lines · Florian Fainelli
Quoted 11 lines · Vladimir Oltean
From: Vladimir Oltean <[email protected]>
The initial purpose of this series was to implement the .flow_dissect method for sja1105 and for ocelot/felix. But on Felix this posed a problem, as the DSA headers were of different lengths on RX and on TX. A better solution than to just increase the smaller one was to also try to shrink the larger one, but in turn that required the DSA master to be put in promiscuous mode (which sja1105 also needed, for other reasons).
Finally, we can add the missing .flow_dissect methods to ocelot and sja1105 (as well as generalize the formula to other taggers as well).
On a separate note, do you have any systems for which it would be desirable that the DSA standalone port implemented receive filtering? On BCM7278 devices, the Ethernet MAC connected to the switch is always in promiscuous mode, and so we rely on the switch not to flood the CPU port unnecessarily with MC traffic (if nothing else), this is currently implemented in our downstream kernel, but has not made it upstream yet, previous attempt was here:
Quoted 7 lines
Quoted 3 lines
I would like to revisit that at some point.
--
FlorianYes, CPU filtering of traffic (not just multicast) is one of the problems we're facing. I'll take a look at your patches and maybe I'll pick them up.
The part that modifies DSA to program the known MC addresses should still be largely applicable, there were essentially two problems that I was facing, which could be tackled separately.
1) flooding of unknown MC traffic on DSA standalone ports is not very intuitive if you come from NICs that did filtering before. We should leverage a DSA switch driver's ability to support port_egress_floods and support port_mdb_add and combine them to avoid flooding the CPU port.
2) Programming of known multicast addresses for VLAN devices on top of DSA standalone ports while the switch implements global VLAN filtering. In that case when we get to the DSA slave device's ndo_set_rx_mode() we have lost all information about which VID the MAC address is coming from so we cannot insert the MAC address with the correct VID to support proper filtering. TI's cpsw driver implements a super complicated scheme to solve that problem and this was worked on by Ivan in a more generic and usable form: https://lwn.net/Articles/780783/
Signature · 1 line
Florian
Vladimir Oltean
maintainerHi Florian,
Quoted 39 lines · Florian Fainelli
Quoted 27 lines · Vladimir Oltean
Quoted 26 lines · Florian Fainelli
Quoted 11 lines · Vladimir Oltean
From: Vladimir Oltean <[email protected]>
The initial purpose of this series was to implement the .flow_dissect method for sja1105 and for ocelot/felix. But on Felix this posed a problem, as the DSA headers were of different lengths on RX and on TX. A better solution than to just increase the smaller one was to also try to shrink the larger one, but in turn that required the DSA master to be put in promiscuous mode (which sja1105 also needed, for other reasons).
Finally, we can add the missing .flow_dissect methods to ocelot and sja1105 (as well as generalize the formula to other taggers as well).
On a separate note, do you have any systems for which it would be desirable that the DSA standalone port implemented receive filtering? On BCM7278 devices, the Ethernet MAC connected to the switch is always in promiscuous mode, and so we rely on the switch not to flood the CPU port unnecessarily with MC traffic (if nothing else), this is currently implemented in our downstream kernel, but has not made it upstream yet, previous attempt was here:
Quoted 7 lines
Quoted 3 lines
I would like to revisit that at some point.
--
FlorianYes, CPU filtering of traffic (not just multicast) is one of the problems we're facing. I'll take a look at your patches and maybe I'll pick them up.
Quoted 9 lines
The part that modifies DSA to program the known MC addresses should still be largely applicable, there were essentially two problems that I was facing, which could be tackled separately.
1) flooding of unknown MC traffic on DSA standalone ports is not very intuitive if you come from NICs that did filtering before. We should leverage a DSA switch driver's ability to support
port_egress_floodsand supportport_mdb_addand combine them to avoid flooding the CPU port.
Could you clarify one thing for me:
- SWITCHDEV_ATTR_ID_BRIDGE_MC_DISABLED is supposed to sync/unsync all known multicast {mac, vid} addresses to the hardware filtering table
- SWITCHDEV_ATTR_ID_BRIDGE_MROUTER is supposed to toggle flooding of unknown multicast addresses to the CPU
- What is BR_MCAST_FLOOD from SWITCHDEV_ATTR_ID_PORT_BRIDGE_FLAGS supposed to do? The same thing?
Quoted 8 lines
2) Programming of known multicast addresses for VLAN devices on top of DSA standalone ports while the switch implements global VLAN filtering. In that case when we get to the DSA slave device's
ndo_set_rx_mode()we have lost all information about which VID the MAC address is coming from so we cannot insert the MAC address with the correct VID to support proper filtering. TI's cpsw driver implements a super complicated scheme to solve that problem and this was worked on by Ivan in a more generic and usable form: https://lwn.net/Articles/780783/
So you're thinking of pulling his vlan_dev_get_addr_vid API and syncing {DMAC, VID} addresses to the DSA slave ports by installing mdb rules (or fdb in case of unicast filtering) on the CPU port?
But I think the only thing that would make global VLAN filtering more complicated than the general case is that we would have to:
- deny switch ports from being enslaved to a VLAN-unaware bridge if there is at least one other DSA slave in this switch that has addresses with a non-pvid VLAN in its RX filter
- deny changes to the VLAN filtering state of any bridge that spans a switch which has at least a port with a non-pvid VLAN in its RX filter Am I missing something?
Also, for unicast filtering, I believe you would agree to leverage port_egress_floods(unicast=false), and have the DSA core install one fdb entry on the CPU port per each slave netdevice MAC address. Sadly, for that to work, we'd have to keep our own parallel reference counting in dsa_slave_sync_unsync_fdb_addr (because the refcount in __hw_addr_sync_dev is per slave device and not per cpu port).
Last but not least, how do you see the CPU membership of VLANs problem being approached? Using Ivan's vlan_dev_get_addr_vid API, only let the CPU see traffic from a particular VLAN if there is any upper 8021q net device installed on any DSA slave? What should happen when we bridge a DSA slave port? Let the flood gates open? If the DSA slave is bridged only with other DSA slaves from the same switch, hopefully the answer is no. Hopefully we can open the flood gates only when bridging with a foreign interface.
Quoted 2 lines
--
Florian
Thanks,
-Vladimir