Open vSwitch upcall optimiziations
Subsystems (4)
PATCH 2/8 netlink: Avoid netlink mmap alloc if msg size exceeds frame size
In reply to a message not in the archiveAn insufficent ring frame size configuration can lead to an unnecessary skb allocation for every Netlink message. Check frame size before taking the queue lock and allocating the skb and re-check with lock to be safe.
--- net/netlink/af_netlink.c | 4 ++++ 1 file changed, 4 insertions(+)
net/netlink/af_netlink.c +4 -0
@@ -1769,6 +1769,9 @@ struct sk_buff *netlink_alloc_skb(struct sock *ssk, unsigned int size,if (ring->pg_vec ==NULL ) goto out_put; +if (ring->frame_size -NL_MMAP_HDRLEN < size) + goto out_put; + skb =alloc_skb_head (gfp_mask);if (skb ==NULL ) goto err1;
@@ -1778,6 +1781,7 @@ struct sk_buff *netlink_alloc_skb(struct sock *ssk, unsigned int size,if (ring->pg_vec ==NULL ) goto out_free; +/* check again under lock */ maxlen = ring->frame_size -NL_MMAP_HDRLEN ;if (maxlen < size) goto out_free;
Signature · 1 line
1.8.3.1
v7 cover Open vSwitch upcall optimiziations
Reposting this before the merge window as it will go through Jesse's tree.
Given jumbo frames, the capacity of the slow path is improved by a factor of 2.5x.
V7: - removed unintential kernel-doc comment
- WARN_ONCE() -> WARN(), message on single line, added \n
V6: - Added memory mapped netlink i/o support
- Drop user_features if old user space not aware of user features reuses an existing datapath
V5: - Removed padding requirement in user space
- Added OVS_DP_F_UNALIGNED flag let user space signal ability to receive unaligned Netlink messages, fall back to linear copy otherwise.
V4: - Daniel Borkmann pointed out that the style in skbuff.h has changed
in net-next, adapted to no longer using extern in headers.
V3: - Removed unneeded alignment of nlmsg_len after padding
V2: - Added skb_zerocopy_headlen() to calculate headroom of destination
buffer. This also takes care of the from->head_frag issue.
- Attribute alignment for frag_list case
- API documentation
- performance data for CHECKSUM_PARTIAL tx case
Thomas Graf (8):
genl: Add genlmsg_new_unicast() for unicast message allocation
netlink: Avoid netlink mmap alloc if msg size exceeds frame size
openvswitch: Enable memory mapped Netlink i/o
net: Export skb_zerocopy() to zerocopy from one skb to another
openvswitch: Allow user space to announce ability to accept unaligned
Netlink messages
openvswitch: Allow update of dp with OVS_DP_CMD_NEW if NLM_F_REPLACE
is set
openvswitch: Drop user features if old user space attempted to create
datapath
openvswitch: Use skb_zerocopy() for upcall
include/linux/skbuff.h | 3 + include/net/genetlink.h | 4 + include/uapi/linux/openvswitch.h | 14 ++- net/core/skbuff.c | 85 ++++++++++++++++ net/netfilter/nfnetlink_queue_core.c | 59 +---------- net/netlink/af_netlink.c | 4 + net/netlink/genetlink.c | 21 ++++ net/openvswitch/datapath.c | 192 ++++++++++++++++++++++------------- net/openvswitch/datapath.h | 2 + 9 files changed, 255 insertions(+), 129 deletions(-)
Signature · 1 line
1.8.3.1
PATCH v7 1/8 genl: Add genlmsg_new_unicast() for unicast message allocation
Allocates a new sk_buff large enough to cover the specified payload plus required Netlink headers. Will check receiving socket for memory mapped i/o capability and use it if enabled. Will fall back to non-mapped skb if message size exceeds the frame size of the ring.
--- include/net/genetlink.h | 4 ++++ net/netlink/genetlink.c | 21 +++++++++++++++++++++ 2 files changed, 25 insertions(+)
include/net/genetlink.h +4 -0
@@ -73,6 +73,7 @@ struct genl_family {* @attrs: netlink attributes * @_net: network namespace * @user_ptr: user pointers +* @dst_sk: destination socket */ struct genl_info {u32 snd_seq;
@@ -85,6 +86,7 @@ struct genl_info {struct net * _net;#endif void * user_ptr[2 ]; +struct sock * dst_sk; };static inline struct net *genl_info_net (struct genl_info *info)
@@ -177,6 +179,8 @@ void genl_notify(struct genl_family *family, struct sk_buff *skb, struct net *net, u32 portid, u32 group, struct nlmsghdr *nlh, gfp_t flags); +struct sk_buff *genlmsg_new_unicast (size_t payload,struct genl_info *info, +gfp_t flags);void *genlmsg_put (struct sk_buff *skb,u32 portid,u32 seq,struct genl_family *family,int flags,u8 cmd);
net/netlink/genetlink.c +21 -0
@@ -454,6 +454,26 @@ int genl_unregister_family(struct genl_family *family)EXPORT_SYMBOL (genl_unregister_family);/** +* genlmsg_new_unicast - Allocate generic netlink message for unicast +* @payload: size of the message payload +* @info: information on destination +* @flags: the type of memory to allocate +* +* Allocates a new sk_buff large enough to cover the specified payload +* plus required Netlink headers. Will check receiving socket for +* memory mapped i/o capability and use it if enabled. Will fall back +* to non-mapped skb if message size exceeds the frame size of the ring. +*/ +struct sk_buff *genlmsg_new_unicast (size_t payload,struct genl_info *info, +gfp_t flags) +{ +size_t len =nlmsg_total_size (genlmsg_total_size (payload)); + +return netlink_alloc_skb (info->dst_sk, len, info->snd_portid, flags); +} +EXPORT_SYMBOL_GPL (genlmsg_new_unicast); + +/** * genlmsg_put - Add generic netlink header to netlink message * @skb: socket buffer holding the message * @portid: netlink portid the message is addressed to
@@ -593,6 +613,7 @@ static int genl_family_rcv_msg(struct genl_family *family, info.genlhdr =nlmsg_data (nlh); info.userhdr =nlmsg_data (nlh) +GENL_HDRLEN ; info.attrs = attrbuf; + info.dst_sk = skb->sk;genl_info_net_set (&info, net);memset (&info.user_ptr,0 ,sizeof (info.user_ptr));
Signature · 1 line
1.8.3.1
PATCH v7 3/8 openvswitch: Enable memory mapped Netlink i/o
In reply to Thomas GrafUse memory mapped Netlink i/o for all unicast openvswitch communication if a ring has been set up.
Benchmark
* pktgen -> ovs internal port
* 5M pkts, 5M flows
* 4 threads, 8 cores
Before:
Result: OK: 67418743(c67108212+d310530) usec, 5000000 (9000byte,0frags)
74163pps 5339Mb/sec (5339736000bps) errors: 0
+ 2.98% ovs-vswitchd [k] copy_user_generic_string + 2.49% ovs-vswitchd [k] memcpy + 1.84% kpktgend_2 [k] memcpy + 1.81% kpktgend_1 [k] memcpy + 1.81% kpktgend_3 [k] memcpy + 1.78% kpktgend_0 [k] memcpy
After:
Result: OK: 24229690(c24127165+d102524) usec, 5000000 (9000byte,0frags)
206358pps 14857Mb/sec (14857776000bps) errors: 0
+ 2.80% ovs-vswitchd [k] memcpy + 1.31% kpktgend_2 [k] memcpy + 1.23% kpktgend_0 [k] memcpy + 1.09% kpktgend_1 [k] memcpy + 1.04% kpktgend_3 [k] memcpy + 0.96% ovs-vswitchd [k] copy_user_generic_string
--- net/openvswitch/datapath.c | 56 ++++++++++++++++++++++++---------------------- 1 file changed, 29 insertions(+), 27 deletions(-)
net/openvswitch/datapath.c +29 -27
@@ -402,6 +402,11 @@ static int queue_userspace_packet(struct net *net, int dp_ifindex,struct sk_buff *nskb =NULL ;struct sk_buff *user_skb;/* to be queued to userspace */ struct nlattr *nla; +struct genl_info info = { + .dst_sk = net->genl_sock, + .snd_portid = upcall_info->portid, + }; +size_t len;int err;if (vlan_tx_tag_present (skb)) {
@@ -422,7 +427,8 @@ static int queue_userspace_packet(struct net *net, int dp_ifindex, goto out; } - user_skb =genlmsg_new (upcall_msg_size (skb, upcall_info->userdata),GFP_ATOMIC ); + len =upcall_msg_size (skb, upcall_info->userdata); + user_skb =genlmsg_new_unicast (len, &info,GFP_ATOMIC );if (!user_skb) { err = -ENOMEM ; goto out;
@@ -726,27 +732,30 @@ error:return err; } -static struct sk_buff *ovs_flow_cmd_alloc_info (struct sw_flow *flow) +static struct sk_buff *ovs_flow_cmd_alloc_info (struct sw_flow *flow, +struct genl_info *info) { -const struct sw_flow_actions *sf_acts; +size_t len; - sf_acts =ovsl_dereference (flow->sf_acts); + len =ovs_flow_cmd_msg_size (ovsl_dereference (flow->sf_acts)); -return genlmsg_new (ovs_flow_cmd_msg_size (sf_acts),GFP_KERNEL ); +return genlmsg_new_unicast (len, info,GFP_KERNEL ); }static struct sk_buff *ovs_flow_cmd_build_info (struct sw_flow *flow,struct datapath *dp, -u32 portid,u32 seq,u8 cmd) +struct genl_info *info, +u8 cmd) {struct sk_buff *skb;int retval; - skb =ovs_flow_cmd_alloc_info (flow); + skb =ovs_flow_cmd_alloc_info (flow, info);if (!skb)return ERR_PTR (-ENOMEM ); - retval =ovs_flow_cmd_fill_info (flow, dp, skb, portid, seq,0 , cmd); + retval =ovs_flow_cmd_fill_info (flow, dp, skb, info->snd_portid, + info->snd_seq,0 , cmd);BUG_ON (retval <0 );return skb; }
@@ -835,8 +844,7 @@ static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info) goto err_flow_free; } - reply =ovs_flow_cmd_build_info (flow, dp, info->snd_portid, - info->snd_seq,OVS_FLOW_CMD_NEW ); + reply =ovs_flow_cmd_build_info (flow, dp, info,OVS_FLOW_CMD_NEW ); }else {/* We found a matching flow. */ struct sw_flow_actions *old_acts;
@@ -864,8 +872,7 @@ static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)rcu_assign_pointer (flow->sf_acts, acts);ovs_nla_free_flow_actions (old_acts); - reply =ovs_flow_cmd_build_info (flow, dp, info->snd_portid, - info->snd_seq,OVS_FLOW_CMD_NEW ); + reply =ovs_flow_cmd_build_info (flow, dp, info,OVS_FLOW_CMD_NEW );/* Clear stats. */ if (a[OVS_FLOW_ATTR_CLEAR ]) {
@@ -927,8 +934,7 @@ static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info) goto unlock; } - reply =ovs_flow_cmd_build_info (flow, dp, info->snd_portid, - info->snd_seq,OVS_FLOW_CMD_NEW ); + reply =ovs_flow_cmd_build_info (flow, dp, info,OVS_FLOW_CMD_NEW );if (IS_ERR (reply)) { err =PTR_ERR (reply); goto unlock;
@@ -975,7 +981,7 @@ static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info) goto unlock; } - reply =ovs_flow_cmd_alloc_info (flow); + reply =ovs_flow_cmd_alloc_info (flow, info);if (!reply) { err = -ENOMEM ; goto unlock;
@@ -1127,17 +1133,17 @@ error:return -EMSGSIZE ; } -static struct sk_buff *ovs_dp_cmd_build_info (struct datapath *dp,u32 portid, -u32 seq,u8 cmd) +static struct sk_buff *ovs_dp_cmd_build_info (struct datapath *dp, +struct genl_info *info,u8 cmd) {struct sk_buff *skb;int retval; - skb =genlmsg_new (ovs_dp_cmd_msg_size (),GFP_KERNEL ); + skb =genlmsg_new_unicast (ovs_dp_cmd_msg_size (), info,GFP_KERNEL );if (!skb)return ERR_PTR (-ENOMEM ); - retval =ovs_dp_cmd_fill_info (dp, skb, portid, seq,0 , cmd); + retval =ovs_dp_cmd_fill_info (dp, skb, info->snd_portid, info->snd_seq,0 , cmd);if (retval <0 ) {kfree_skb (skb);return ERR_PTR (retval);
@@ -1232,8 +1238,7 @@ static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info) goto err_destroy_ports_array; } - reply =ovs_dp_cmd_build_info (dp, info->snd_portid, - info->snd_seq,OVS_DP_CMD_NEW ); + reply =ovs_dp_cmd_build_info (dp, info,OVS_DP_CMD_NEW ); err =PTR_ERR (reply);if (IS_ERR (reply)) goto err_destroy_local_port;
@@ -1299,8 +1304,7 @@ static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)if (IS_ERR (dp)) goto unlock; - reply =ovs_dp_cmd_build_info (dp, info->snd_portid, - info->snd_seq,OVS_DP_CMD_DEL ); + reply =ovs_dp_cmd_build_info (dp, info,OVS_DP_CMD_DEL ); err =PTR_ERR (reply);if (IS_ERR (reply)) goto unlock;
@@ -1328,8 +1332,7 @@ static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)if (IS_ERR (dp)) goto unlock; - reply =ovs_dp_cmd_build_info (dp, info->snd_portid, - info->snd_seq,OVS_DP_CMD_NEW ); + reply =ovs_dp_cmd_build_info (dp, info,OVS_DP_CMD_NEW );if (IS_ERR (reply)) { err =PTR_ERR (reply); genl_set_err(&dp_datapath_genl_family, sock_net(skb->sk), 0,
@@ -1360,8 +1363,7 @@ static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info) goto unlock; } - reply =ovs_dp_cmd_build_info (dp, info->snd_portid, - info->snd_seq,OVS_DP_CMD_NEW ); + reply =ovs_dp_cmd_build_info (dp, info,OVS_DP_CMD_NEW );if (IS_ERR (reply)) { err =PTR_ERR (reply); goto unlock;
Signature · 1 line
1.8.3.1
PATCH v7 4/8 net: Export skb_zerocopy() to zerocopy from one skb to another
In reply to Thomas GrafMake the skb zerocopy logic written for nfnetlink queue available for use by other modules.
--- include/linux/skbuff.h | 3 ++ net/core/skbuff.c | 85 ++++++++++++++++++++++++++++++++++++ net/netfilter/nfnetlink_queue_core.c | 59 ++----------------------- 3 files changed, 92 insertions(+), 55 deletions(-)
include/linux/skbuff.h +3 -0
@@ -2345,6 +2345,9 @@ int skb_splice_bits(struct sk_buff *skb, unsigned int offset, struct pipe_inode_info *pipe, unsigned int len, unsigned int flags);void skb_copy_and_csum_dev (const struct sk_buff *skb,u8 *to); +unsigned int skb_zerocopy_headlen (const struct sk_buff *from); +void skb_zerocopy (struct sk_buff *to,const struct sk_buff *from, +int len,int hlen);void skb_split (struct sk_buff *skb,struct sk_buff *skb1,const u32 len);int skb_shift (struct sk_buff *tgt,struct sk_buff *skb,int shiftlen);void skb_scrub_packet (struct sk_buff *skb,bool xnet);
net/core/skbuff.c +85 -0
91 lines not on this page · raw message
net/netfilter/nfnetlink_queue_core.c +4 -55
83 lines not on this page · raw message
Signature · 1 line
1.8.3.1
PATCH v7 5/8 openvswitch: Allow user space to announce ability to accept unaligned Netlink messages
In reply to Thomas Graf--- include/uapi/linux/openvswitch.h | 4 ++++ net/openvswitch/datapath.c | 14 ++++++++++++++ net/openvswitch/datapath.h | 2 ++ 3 files changed, 20 insertions(+)
include/uapi/linux/openvswitch.h +4 -0
@@ -75,6 +75,7 @@ enum ovs_datapath_attr {OVS_DP_ATTR_UPCALL_PID ,/* Netlink PID to receive upcalls */ OVS_DP_ATTR_STATS ,/* struct ovs_dp_stats */ OVS_DP_ATTR_MEGAFLOW_STATS ,/* struct ovs_dp_megaflow_stats */ +OVS_DP_ATTR_USER_FEATURES ,/* OVS_DP_F_* */ __OVS_DP_ATTR_MAX };
@@ -106,6 +107,9 @@ struct ovs_vport_stats {__u64 tx_dropped;/* no space available in linux */ }; +/* Allow last Netlink attribute to be unaligned */ +#define OVS_DP_F_UNALIGNED (1 << 0) +/* Fixed logical ports. */ #define OVSP_LOCAL ((__u32)0)
net/openvswitch/datapath.c +14 -0
@@ -1067,6 +1067,7 @@ static const struct genl_ops dp_flow_genl_ops[] = {static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX +1 ] = { [OVS_DP_ATTR_NAME ] = { .type =NLA_NUL_STRING , .len =IFNAMSIZ -1 }, [OVS_DP_ATTR_UPCALL_PID ] = { .type =NLA_U32 }, + [OVS_DP_ATTR_USER_FEATURES ] = { .type =NLA_U32 }, };static struct genl_family dp_datapath_genl_family = {
@@ -1125,6 +1126,9 @@ static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb, &dp_megaflow_stats)) goto nla_put_failure; +if (nla_put_u32 (skb,OVS_DP_ATTR_USER_FEATURES , dp->user_features)) + goto nla_put_failure; +return genlmsg_end (skb, ovs_header); nla_put_failure:
@@ -1171,6 +1175,12 @@ static struct datapath *lookup_datapath(struct net *net,return dp ? dp :ERR_PTR (-ENODEV ); } +static void ovs_dp_change (struct datapath *dp,struct nlattr **a) +{ +if (a[OVS_DP_ATTR_USER_FEATURES ]) + dp->user_features =nla_get_u32 (a[OVS_DP_ATTR_USER_FEATURES ]); +} +static int ovs_dp_cmd_new (struct sk_buff *skb,struct genl_info *info) {struct nlattr **a = info->attrs;
@@ -1229,6 +1239,8 @@ static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info) parms.port_no =OVSP_LOCAL ; parms.upcall_portid =nla_get_u32 (a[OVS_DP_ATTR_UPCALL_PID ]); +ovs_dp_change (dp, a); + vport =new_vport (&parms);if (IS_ERR (vport)) { err =PTR_ERR (vport);
@@ -1332,6 +1344,8 @@ static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)if (IS_ERR (dp)) goto unlock; +ovs_dp_change (dp, info->attrs); + reply =ovs_dp_cmd_build_info (dp, info,OVS_DP_CMD_NEW );if (IS_ERR (reply)) { err =PTR_ERR (reply);
net/openvswitch/datapath.h +2 -0
@@ -88,6 +88,8 @@ struct datapath {/* Network namespace ref. */ struct net *net;#endif + +u32 user_features; };/**
Signature · 1 line
1.8.3.1
PATCH v7 6/8 openvswitch: Allow update of dp with OVS_DP_CMD_NEW if NLM_F_REPLACE is set
In reply to Thomas GrafConsolidates ovs_dp_cmd_new() and ovs_dp_cmd_set() to simplify handling and avoid code duplication.
Allows user space to specify NLM_F_REPLACE with OVS_DP_CMD_NEW and overwrite the settings such as the user features of an existing datapath.
--- net/openvswitch/datapath.c | 83 +++++++++++++++++++++++----------------------- 1 file changed, 42 insertions(+), 41 deletions(-)
net/openvswitch/datapath.c +42 -41
@@ -1175,13 +1175,8 @@ static struct datapath *lookup_datapath(struct net *net,return dp ? dp :ERR_PTR (-ENODEV ); } -static void ovs_dp_change (struct datapath *dp,struct nlattr **a) -{ -if (a[OVS_DP_ATTR_USER_FEATURES ]) - dp->user_features =nla_get_u32 (a[OVS_DP_ATTR_USER_FEATURES ]); -} - -static int ovs_dp_cmd_new (struct sk_buff *skb,struct genl_info *info) +static int ovs_dp_create_or_update (struct sk_buff *skb,struct genl_info *info, +bool create) {struct nlattr **a = info->attrs;struct vport_parms parms;
@@ -1190,6 +1185,7 @@ static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)struct vport *vport;struct ovs_net *ovs_net;int err, i; +bool allocated = false; err = -EINVAL ;if (!a[OVS_DP_ATTR_NAME ] || !a[OVS_DP_ATTR_UPCALL_PID ])
@@ -1197,11 +1193,26 @@ static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)ovs_lock (); + dp =lookup_datapath (sock_net (skb->sk), info->userhdr, info->attrs); +if (!IS_ERR (dp)) { +if (info->nlhdr->nlmsg_flags &NLM_F_REPLACE ) + goto update; + + err = -EEXIST ; + goto err_unlock_ovs; + } + +if (!create) { + err =PTR_ERR (dp); + goto err_unlock_ovs; + } + err = -ENOMEM ; dp =kzalloc (sizeof (*dp),GFP_KERNEL );if (dp ==NULL ) goto err_unlock_ovs; + allocated = true;ovs_dp_set_net (dp,hold_net (sock_net (skb->sk)));/* Allocate table. */
@@ -1239,8 +1250,6 @@ static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info) parms.port_no =OVSP_LOCAL ; parms.upcall_portid =nla_get_u32 (a[OVS_DP_ATTR_UPCALL_PID ]); -ovs_dp_change (dp, a); - vport =new_vport (&parms);if (IS_ERR (vport)) { err =PTR_ERR (vport);
@@ -1250,13 +1259,27 @@ static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info) goto err_destroy_ports_array; } +update: +if (a[OVS_DP_ATTR_USER_FEATURES ]) + dp->user_features =nla_get_u32 (a[OVS_DP_ATTR_USER_FEATURES ]); + reply =ovs_dp_cmd_build_info (dp, info,OVS_DP_CMD_NEW ); err =PTR_ERR (reply); -if (IS_ERR (reply)) - goto err_destroy_local_port; +if (IS_ERR (reply)) { +if (allocated) + goto err_destroy_local_port; +else { + err =0 ; +genl_set_err (&dp_datapath_genl_family,sock_net (skb->sk),0 , +0 , err); + goto err_unlock_ovs; + } + } - ovs_net =net_generic (ovs_dp_get_net (dp), ovs_net_id); -list_add_tail_rcu (&dp->list_node, &ovs_net->dps); +if (allocated) { + ovs_net =net_generic (ovs_dp_get_net (dp), ovs_net_id); +list_add_tail_rcu (&dp->list_node, &ovs_net->dps); + }ovs_unlock ();
@@ -1280,6 +1303,11 @@ err:return err; } +static int ovs_dp_cmd_new (struct sk_buff *skb,struct genl_info *info) +{ +return ovs_dp_create_or_update (skb, info, true); +} +/* Called with ovs_mutex. */ static void __dp_destroy (struct datapath *dp) {
@@ -1334,34 +1362,7 @@ unlock:static int ovs_dp_cmd_set (struct sk_buff *skb,struct genl_info *info) { -struct sk_buff *reply; -struct datapath *dp; -int err; - -ovs_lock (); - dp =lookup_datapath (sock_net (skb->sk), info->userhdr, info->attrs); - err =PTR_ERR (dp); -if (IS_ERR (dp)) - goto unlock; - -ovs_dp_change (dp, info->attrs); - - reply =ovs_dp_cmd_build_info (dp, info,OVS_DP_CMD_NEW ); -if (IS_ERR (reply)) { - err =PTR_ERR (reply); -genl_set_err (&dp_datapath_genl_family,sock_net (skb->sk),0 , -0 , err); - err =0 ; - goto unlock; - } - -ovs_unlock (); -ovs_notify (&dp_datapath_genl_family, reply, info); - -return 0 ; -unlock: -ovs_unlock (); -return err; +return ovs_dp_create_or_update (skb, info, false); }static int ovs_dp_cmd_get (struct sk_buff *skb,struct genl_info *info)
Signature · 1 line
1.8.3.1
PATCH v7 7/8 openvswitch: Drop user features if old user space attempted to create datapath
In reply to Thomas GrafDrop user features if an outdated user space instance that does not understand the concept of user_features attempted to create a new datapath.
--- include/uapi/linux/openvswitch.h | 10 +++++++++- net/openvswitch/datapath.c | 9 +++++++++ 2 files changed, 18 insertions(+), 1 deletion(-)
include/uapi/linux/openvswitch.h +9 -1
@@ -40,7 +40,15 @@ struct ovs_header {#define OVS_DATAPATH_FAMILY "ovs_datapath"#define OVS_DATAPATH_MCGROUP "ovs_datapath" -#define OVS_DATAPATH_VERSION 0x1 + +/* V2: +* - API users are expected to provide OVS_DP_ATTR_USER_FEATURES +* with NLM_F_REPLACE when creating the datapath. +*/ +#define OVS_DATAPATH_VERSION 2 + +/* First OVS datapath version to support features */ +#define OVS_DP_VER_FEATURES 2enum ovs_datapath_cmd {OVS_DP_CMD_UNSPEC ,
net/openvswitch/datapath.c +9 -0
@@ -1198,6 +1198,15 @@ static int ovs_dp_create_or_update(struct sk_buff *skb, struct genl_info *info,if (info->nlhdr->nlmsg_flags &NLM_F_REPLACE ) goto update; +/* An outdated user space instance that does not understand +* the concept of user_features has attempted to create a new +* datapath. Drop all user features. +*/ +if (info->genlhdr->version <OVS_DP_VER_FEATURES ) { +WARN (dp->user_features,"Dropping previously announced user features\n" ); + dp->user_features =0 ; + } + err = -EEXIST ; goto err_unlock_ovs; }
Signature · 1 line
1.8.3.1
PATCH v7 8/8 openvswitch: Use skb_zerocopy() for upcall
In reply to Thomas GrafUse of skb_zerocopy() can avoids the expensive call to memcpy() when copying the packet data into the Netlink skb. Completes checksum through skb_checksum_help() if needed.
Zerocopy is only performed if user space supported unaligned Netlink messages. memory mapped netlink i/o is preferred over zerocopy if it is set up.
Cost of upcall is significantly reduced from:
+ 7.48% vhost-8471 [k] memcpy + 5.57% ovs-vswitchd [k] memcpy + 2.81% vhost-8471 [k] csum_partial_copy_generic
to:
+ 5.72% ovs-vswitchd [k] memcpy + 3.32% vhost-5153 [k] memcpy + 0.68% vhost-5153 [k] skb_zerocopy
(megaflows disabled)
--- net/openvswitch/datapath.c | 54 +++++++++++++++++++++++++++++++--------------- 1 file changed, 37 insertions(+), 17 deletions(-)
net/openvswitch/datapath.c +37 -17
@@ -108,10 +108,10 @@ int lockdep_ovsl_is_held(void)#endif static struct vport *new_vport (const struct vport_parms *); -static int queue_gso_packets (struct net *,int dp_ifindex,struct sk_buff *, -const struct dp_upcall_info *); -static int queue_userspace_packet (struct net *,int dp_ifindex, -struct sk_buff *, +static int queue_gso_packets (struct datapath *,struct net *,int dp_ifindex, +struct sk_buff *,const struct dp_upcall_info *); +static int queue_userspace_packet (struct datapath *,struct net *, +int dp_ifindex,struct sk_buff *,const struct dp_upcall_info *);/* Must be called with rcu_read_lock or ovs_mutex. */
@@ -292,9 +292,9 @@ int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb, }if (!skb_is_gso (skb)) - err =queue_userspace_packet (ovs_dp_get_net (dp), dp_ifindex, skb, upcall_info); + err =queue_userspace_packet (dp,ovs_dp_get_net (dp), dp_ifindex, skb, upcall_info);else - err =queue_gso_packets (ovs_dp_get_net (dp), dp_ifindex, skb, upcall_info); + err =queue_gso_packets (dp,ovs_dp_get_net (dp), dp_ifindex, skb, upcall_info);if (err) goto err;
@@ -310,7 +310,7 @@ err:return err; } -static int queue_gso_packets (struct net *net,int dp_ifindex, +static int queue_gso_packets (struct datapath *dp,struct net *net,int dp_ifindex,struct sk_buff *skb,const struct dp_upcall_info *upcall_info) {
@@ -327,7 +327,7 @@ static int queue_gso_packets(struct net *net, int dp_ifindex,/* Queue all of the segments. */ skb = segs; do { - err = queue_userspace_packet(net, dp_ifindex, skb, upcall_info); + err = queue_userspace_packet(dp, net, dp_ifindex, skb, upcall_info); if (err) break;
@@ -381,10 +381,11 @@ static size_t key_attr_size(void) }static size_t upcall_msg_size (const struct sk_buff *skb, -const struct nlattr *userdata) +const struct nlattr *userdata, +unsigned int hdrlen) {size_t size =NLMSG_ALIGN (sizeof (struct ovs_header )) - +nla_total_size (skb->len)/* OVS_PACKET_ATTR_PACKET */ + +nla_total_size (hdrlen)/* OVS_PACKET_ATTR_PACKET */ +nla_total_size (key_attr_size ());/* OVS_PACKET_ATTR_KEY */ /* OVS_PACKET_ATTR_USERDATA */
@@ -394,8 +395,8 @@ static size_t upcall_msg_size(const struct sk_buff *skb,return size; } -static int queue_userspace_packet (struct net *net,int dp_ifindex, -struct sk_buff *skb, +static int queue_userspace_packet (struct datapath *dp,struct net *net, +int dp_ifindex,struct sk_buff *skb,const struct dp_upcall_info *upcall_info) {struct ovs_header *upcall;
@@ -407,6 +408,7 @@ static int queue_userspace_packet(struct net *net, int dp_ifindex, .snd_portid = upcall_info->portid, };size_t len; +unsigned int hlen;int err;if (vlan_tx_tag_present (skb)) {
@@ -427,7 +429,21 @@ static int queue_userspace_packet(struct net *net, int dp_ifindex, goto out; } - len =upcall_msg_size (skb, upcall_info->userdata); +/* Complete checksum if needed */ +if (skb->ip_summed ==CHECKSUM_PARTIAL && + (err =skb_checksum_help (skb))) + goto out; + +/* Older versions of OVS user space enforce alignment of the last +* Netlink attribute to NLA_ALIGNTO which would require extensive +* padding logic. Only perform zerocopy if padding is not required. +*/ +if (dp->user_features &OVS_DP_F_UNALIGNED ) + hlen =skb_zerocopy_headlen (skb); +else + hlen = skb->len; + + len =upcall_msg_size (skb, upcall_info->userdata, hlen); user_skb =genlmsg_new_unicast (len, &info,GFP_ATOMIC );if (!user_skb) { err = -ENOMEM ;
@@ -447,13 +463,17 @@ static int queue_userspace_packet(struct net *net, int dp_ifindex,nla_len (upcall_info->userdata), nla_data(upcall_info->userdata)); - nla =__nla_reserve (user_skb,OVS_PACKET_ATTR_PACKET , skb->len); +/* Only reserve room for attribute header, packet data is added +* in skb_zerocopy() */ +if (!(nla =nla_reserve (user_skb,OVS_PACKET_ATTR_PACKET ,0 ))) + goto out; + nla->nla_len =nla_attr_size (skb->len); + +skb_zerocopy (user_skb, skb, skb->len, hlen); -skb_copy_and_csum_dev (skb,nla_data (nla)); + ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len; -genlmsg_end (user_skb, upcall); err =genlmsg_unicast (net, user_skb, upcall_info->portid); - out:kfree_skb (nskb);return err;
Signature · 1 line
1.8.3.1
Re: [PATCH net-next 6/8] openvswitch: Allow update of dp with OVS_DP_CMD_NEW if NLM_F_REPLACE is set
In reply to a message not in the archivenet/openvswitch/datapath.c
Quoted 4 lines · Thomas Graf
net/openvswitch/datapath.c +0
-0
I'm a little worried that this introduces some quirks to the interface. Such as:
net/openvswitch/datapath.c static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info) · line 1191
Quoted 8 lines
net/openvswitch/datapath.c +1
-0@@ -1190,6 +1185,7 @@ static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)struct vport *vport;struct ovs_net *ovs_net;int err, i; +bool allocated = false; err = -EINVAL ;if (!a[OVS_DP_ATTR_NAME ] || !a[OVS_DP_ATTR_UPCALL_PID ])
This requires that DP_CMD_SET supply an OVS_DP_ATTR_UPCALL_PID although I don't think that it's really necessary. In fact, we used to completely ignore that field on SET since it's really only useful on datapath creation and can otherwise be done more naturally through the vport interface.
net/openvswitch/datapath.c static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info) · line 1199
Quoted 8 lines
net/openvswitch/datapath.c +4
-0@@ -1197,11 +1193,26 @@ static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)ovs_lock (); + dp =lookup_datapath (sock_net (skb->sk), info->userhdr, info->attrs); +if (!IS_ERR (dp)) { +if (info->nlhdr->nlmsg_flags &NLM_F_REPLACE ) + goto update;
Conversely, this won't respect the UPCALL_PID field, which I would expect it to since I think NLM_F_REPLACE should be roughly equivalent to a delete and create. I believe that's the only field that is missing although it seems easy to have the same problem as others are added in the future.
Re: [PATCH net-next 7/8] openvswitch: Drop user features if old user space attempted to create datapath
In reply to Thomas GrafQuoted 5 lines · Thomas Graf
include/uapi/linux/openvswitch.h +0
-0
include/uapi/linux/openvswitch.h struct ovs_header { · line 48
Quoted 10 lines
include/uapi/linux/openvswitch.h +6
-1@@ -40,7 +40,15 @@ struct ovs_header {#define OVS_DATAPATH_FAMILY "ovs_datapath"#define OVS_DATAPATH_MCGROUP "ovs_datapath" -#define OVS_DATAPATH_VERSION 0x1 + +/* V2: +* - API users are expected to provide OVS_DP_ATTR_USER_FEATURES +* with NLM_F_REPLACE when creating the datapath. +*/ +#define OVS_DATAPATH_VERSION 2
I'm not sure that I understand this comment - I guess it means that there's no legacy path when using NLM_F_REPLACE?
Re: [PATCH net-next 6/8] openvswitch: Allow update of dp with OVS_DP_CMD_NEW if NLM_F_REPLACE is set
In reply to Jesse GrossQuoted 14 lines · Jesse Gross
net/openvswitch/datapath.c
Quoted 4 lines · Thomas Graf
net/openvswitch/datapath.c +0
-0I'm a little worried that this introduces some quirks to the interface. Such as:
Quoted 6 lines
net/openvswitch/datapath.c +1
-0@@ -1190,6 +1185,7 @@ static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)struct vport *vport;struct ovs_net *ovs_net;int err, i; +bool allocated = false;
Quoted 8 lines
net/openvswitch/datapath.c static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info) · line 1191
Quoted 2 lines
net/openvswitch/datapath.c +0
-0err = -EINVAL ;if (!a[OVS_DP_ATTR_NAME ] || !a[OVS_DP_ATTR_UPCALL_PID ])This requires that
DP_CMD_SETsupply anOVS_DP_ATTR_UPCALL_PIDalthough I don't think that it's really necessary. In fact, we used to completely ignore that field on SET since it's really only useful on datapath creation and can otherwise be done more naturally through the vport interface.
It's a theoretical exercise since nobody is using OVS_DP_CMD_SET but I agree, it should be optional in the update path. I'll update the patch.
Quoted 4 lines
Quoted 4 lines
net/openvswitch/datapath.c +0
-0@@ -1197,11 +1193,26 @@ static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)ovs_lock ();
Quoted 10 lines
net/openvswitch/datapath.c static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info) · line 1199
Quoted 4 lines
net/openvswitch/datapath.c +4
-0+ dp =lookup_datapath (sock_net (skb->sk), info->userhdr, info->attrs); +if (!IS_ERR (dp)) { +if (info->nlhdr->nlmsg_flags &NLM_F_REPLACE ) + goto update;Conversely, this won't respect the
UPCALL_PIDfield, which I would expect it to since I thinkNLM_F_REPLACEshould be roughly equivalent to a delete and create. I believe that's the only field that is missing although it seems easy to have the same problem as others are added in the future.
It's up to us to define that.
What the patch proposes is what OVS has been doing already: Try to find a datapath with matching name and reuse it. Otherwise create a new data path. Disregard UPCALL_PID.
As you state above, the upcall pid can be modified through the vport interface which in my opinion is the correct way to modify it if needed.
I have no problem with adding upcall pid overwrite logic to the NLM_F_REPLACE path but it _changes_ the existing semantics in terms of "opening" a datapath.
Re: [PATCH net-next 7/8] openvswitch: Drop user features if old user space attempted to create datapath
In reply to Jesse GrossQuoted 7 lines · Jesse Gross
Quoted 6 lines · Thomas Graf
include/uapi/linux/openvswitch.h +0
-0@@ -40,7 +40,15 @@ struct ovs_header {
Quoted 12 lines
include/uapi/linux/openvswitch.h struct ovs_header { · line 48
Quoted 9 lines
include/uapi/linux/openvswitch.h +6
-1#define OVS_DATAPATH_FAMILY "ovs_datapath"#define OVS_DATAPATH_MCGROUP "ovs_datapath" -#define OVS_DATAPATH_VERSION 0x1 + +/* V2: +* - API users are expected to provide OVS_DP_ATTR_USER_FEATURES +* with NLM_F_REPLACE when creating the datapath. +*/ +#define OVS_DATAPATH_VERSION 2I'm not sure that I understand this comment - I guess it means that there's no legacy path when using
NLM_F_REPLACE?
It means that if the V2 caller does not provide user features the existing user features are preserved upon NLM_F_REPLACE. V1 users can't be aware and the user features are always reset.