net: tcp: make txhash use consistent for IPv4
Subsystems (5)
cover net: tcp: make txhash use consistent for IPv4
Hello,
Series is divided in two parts. First two commits make the txhash (used for the skb hash in TCP) to be consistent for all IPv4/TCP packets (IPv6 doesn't have the same issue). Last two commits improve doc/comment hash-related parts.
One example is when using OvS with dp_hash, which uses skb->hash, to select a path. We'd like packets from the same flow to be consistent, as well as the hash being stable over time when using net.core.txrehash=0. Same applies for kernel ECMP which also can use skb->hash.
IMHO the series makes sense in net-next, but we could argue (some) commits be seen as fixes and I can resend if necessary.
Thanks!
Antoine
Antoine Tenart (4):
net: tcp: make the txhash available in TIME_WAIT sockets for IPv4 too
net: ipv4: use consistent txhash in TIME_WAIT and SYN_RECV
Documentation: net: net.core.txrehash is not specific to listening
sockets
net: skbuff: fix l4_hash comment
Documentation/admin-guide/sysctl/net.rst | 4 ++-- include/linux/skbuff.h | 4 ++-- include/net/ip.h | 2 +- net/ipv4/ip_output.c | 4 +++- net/ipv4/tcp_ipv4.c | 14 +++++++++----- net/ipv4/tcp_minisocks.c | 2 +- 6 files changed, 18 insertions(+), 12 deletions(-)
Signature · 1 line
2.40.0
PATCH 1/4 net: tcp: make the txhash available in TIME_WAIT sockets for IPv4 too
Commit c67b85558ff2 ("ipv6: tcp: send consistent autoflowlabel in TIME_WAIT state") made the socket txhash also available in TIME_WAIT sockets but for IPv6 only. Make it available for IPv4 too as we'll use it in later commits.
--- net/ipv4/tcp_minisocks.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
net/ipv4/tcp_minisocks.c +1 -1
@@ -303,6 +303,7 @@ void tcp_time_wait(struct sock *sk, int state, int timeo) tcptw->tw_ts_offset = tp->tsoffset; tcptw->tw_last_oow_ack_time =0 ; tcptw->tw_tx_delay = tp->tcp_tx_delay; + tw->tw_txhash = sk->sk_txhash;#if IS_ENABLED (CONFIG_IPV6 )if (tw->tw_family ==PF_INET6 ) {struct ipv6_pinfo *np =inet6_sk (sk);
@@ -311,7 +312,6 @@ void tcp_time_wait(struct sock *sk, int state, int timeo) tw->tw_v6_rcv_saddr = sk->sk_v6_rcv_saddr; tw->tw_tclass = np->tclass; tw->tw_flowlabel =be32_to_cpu (np->flow_label &IPV6_FLOWLABEL_MASK ); - tw->tw_txhash = sk->sk_txhash; tw->tw_ipv6only = sk->sk_ipv6only; }#endif
Signature · 1 line
2.40.0
PATCH 2/4 net: ipv4: use consistent txhash in TIME_WAIT and SYN_RECV
In reply to Antoine TenartWhen using IPv4/TCP, skb->hash comes from sk->sk_txhash except in TIME_WAIT and SYN_RECV where it's not set in the reply skb from ip_send_unicast_reply. Those packets will have a mismatched hash with others from the same flow as their hashes will be 0. IPv6 does not have the same issue as the hash is set from the socket txhash in those cases.
This commits sets the hash in the reply skb from ip_send_unicast_reply, which makes the IPv4 code behaving like IPv6.
--- include/net/ip.h | 2 +- net/ipv4/ip_output.c | 4 +++- net/ipv4/tcp_ipv4.c | 14 +++++++++----- 3 files changed, 13 insertions(+), 7 deletions(-)
include/net/ip.h +1 -1
@@ -280,7 +280,7 @@ void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb,const struct ip_options *sopt, __be32 daddr, __be32 saddr, const struct ip_reply_arg *arg, - unsigned int len, u64 transmit_time); + unsigned int len, u64 transmit_time, u32 txhash);#define IP_INC_STATS (net, field) SNMP_INC_STATS64((net)->mib.ip_statistics, field)#define __IP_INC_STATS (net, field) __SNMP_INC_STATS64((net)->mib.ip_statistics, field)
net/ipv4/ip_output.c +3 -1
@@ -1682,7 +1682,7 @@ void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb,const struct ip_options *sopt, __be32 daddr, __be32 saddr, const struct ip_reply_arg *arg, - unsigned int len, u64 transmit_time) + unsigned int len, u64 transmit_time, u32 txhash) { struct ip_options_data replyopts;struct ipcm_cookie ipc;
@@ -1745,6 +1745,8 @@ void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb, arg->csum)); nskb->ip_summed =CHECKSUM_NONE ; nskb->mono_delivery_time = !!transmit_time; +if (txhash) +skb_set_hash (nskb, txhash,PKT_HASH_TYPE_L4 );ip_push_pending_frames (sk, &fl4); } out:
net/ipv4/tcp_ipv4.c +9 -5
@@ -692,6 +692,7 @@ static void tcp_v4_send_reset(const struct sock *sk, struct sk_buff *skb)u64 transmit_time =0 ;struct sock *ctl_sk;struct net *net; +u32 txhash =0 ;/* Never send a reset in response to a reset. */ if (th->rst)
@@ -829,12 +830,14 @@ static void tcp_v4_send_reset(const struct sock *sk, struct sk_buff *skb) inet_twsk(sk)->tw_priority : sk->sk_priority; transmit_time =tcp_transmit_time (sk);xfrm_sk_clone_policy (ctl_sk, sk); + txhash = (sk->sk_state ==TCP_TIME_WAIT ) ? +inet_twsk (sk)->tw_txhash : sk->sk_txhash; }ip_send_unicast_reply (ctl_sk, skb, &TCP_SKB_CB (skb)->header.h4.opt,ip_hdr (skb)->saddr,ip_hdr (skb)->daddr, &arg, arg.iov[0 ].iov_len, - transmit_time); + transmit_time, txhash); ctl_sk->sk_mark =0 ;xfrm_sk_free_policy (ctl_sk);
@@ -857,7 +860,7 @@ static void tcp_v4_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq, u32 ack, u32 win, u32 tsval, u32 tsecr, int oif, struct tcp_md5sig_key *key, - int reply_flags, u8 tos) + int reply_flags, u8 tos, u32 txhash) { const struct tcphdr *th = tcp_hdr(skb); struct {
@@ -933,7 +936,7 @@ static void tcp_v4_send_ack(const struct sock *sk, skb, &TCP_SKB_CB (skb)->header.h4.opt,ip_hdr (skb)->saddr,ip_hdr (skb)->daddr, &arg, arg.iov[0 ].iov_len, - transmit_time); + transmit_time, txhash); ctl_sk->sk_mark =0 ;sock_net_set (ctl_sk, &init_net);
@@ -954,7 +957,8 @@ static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb) tw->tw_bound_dev_if,tcp_twsk_md5_key (tcptw), tw->tw_transparent ?IP_REPLY_ARG_NOSRCCHECK :0 , - tw->tw_tos + tw->tw_tos, + tw->tw_txhash );inet_twsk_put (tw);
@@ -987,7 +991,7 @@ static void tcp_v4_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,0 ,tcp_md5_do_lookup (sk, l3index, addr,AF_INET ),inet_rsk (req)->no_srccheck ?IP_REPLY_ARG_NOSRCCHECK :0 , - ip_hdr(skb)->tos); + ip_hdr(skb)->tos, tcp_rsk(req)->txhash); }/*
Signature · 1 line
2.40.0
PATCH 3/4 Documentation: net: net.core.txrehash is not specific to listening sockets
In reply to Antoine TenartThe net.core.txrehash documentation mentions this knob is for listening sockets only, while sk_rethink_txhash can be called on SYN and RTO retransmits on all TCP sockets.
Remove the listening socket part.
--- Documentation/admin-guide/sysctl/net.rst | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
Documentation/admin-guide/sysctl/net.rst +2 -2
@@ -386,8 +386,8 @@ Default : 0 (for compatibility reasons) txrehash -------- -Controls default hash rethink behaviour on listening socket when SO_TXREHASH -option is set to SOCK_TXREHASH_DEFAULT (i. e. not overridden by setsockopt). +Controls default hash rethink behaviour on socket when SO_TXREHASH option is set +to SOCK_TXREHASH_DEFAULT (i. e. not overridden by setsockopt). If set to 1 (default), hash rethink is performed on listening socket. If set to 0, hash rethink is not performed.
Signature · 1 line
2.40.0
PATCH 4/4 net: skbuff: fix l4_hash comment
In reply to Antoine TenartSince commit 877d1f6291f8 ("net: Set sk_txhash from a random number") sk->sk_txhash is not a canonical 4-tuple hash. sk->sk_txhash is used in the TCP Tx path to populate skb->hash, with skb->l4_hash=1. With this, skb->l4_hash does not always indicate the hash is a "canonical 4-tuple hash over transport ports" but rather a hash from L4 layer to provide a uniform distribution over flows. Reword the comment accordingly, to avoid misunderstandings.
--- include/linux/skbuff.h | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
include/linux/skbuff.h +2 -2
@@ -791,8 +791,8 @@ typedef unsigned char *sk_buff_data_t;* @active_extensions: active extensions (skb_ext_id types) * @ndisc_nodetype: router type (from link layer) * @ooo_okay: allow the mapping of a socket to a queue to be changed -* @l4_hash: indicate hash is a canonical 4-tuple hash over transport -* ports. +* @l4_hash: indicate hash is from layer 4 and provides a uniform +* distribution over flows. * @sw_hash: indicates hash was computed in software stack * @wifi_acked_valid: wifi_acked was set * @wifi_acked: whether frame was acked on wifi or not
Signature · 1 line
2.40.0
Quoted 8 lines · Antoine Tenart
Hello,
Series is divided in two parts. First two commits make the txhash (used for the skb hash in TCP) to be consistent for all IPv4/TCP packets (IPv6 doesn't have the same issue). Last two commits improve doc/comment hash-related parts.
Quoted 4 lines
One example is when using OvS with
dp_hash, which usesskb->hash, to select a path. We'd like packets from the same flow to be consistent, as well as the hash being stable over time when using net.core.txrehash=0. Same applies for kernel ECMP which also can useskb->hash.
How do you plan to test these patches ?
Quoted 3 lines
IMHO the series makes sense in net-next, but we could argue (some) commits be seen as fixes and I can resend if necessary.
net-next is closed...
Quoting Eric Dumazet (2023-04-27 17:15:27)
Quoted 7 lines · Antoine Tenart
Quoted 6 lines
Series is divided in two parts. First two commits make the txhash (used for the skb hash in TCP) to be consistent for all IPv4/TCP packets (IPv6 doesn't have the same issue). Last two commits improve doc/comment hash-related parts.
Quoted 6 lines
Quoted 4 lines
One example is when using OvS with
dp_hash, which usesskb->hash, to select a path. We'd like packets from the same flow to be consistent, as well as the hash being stable over time when using net.core.txrehash=0. Same applies for kernel ECMP which also can useskb->hash.How do you plan to test these patches ?
I did perform manual checks (with net.core.txrehash=0 to make sure the hash was consistent) using a setup with OvS and dp_hash (~ECMP like) and looking which path packets took. Not sure if there is a simpler test that could be automated, we can't use autoflowlabel to make simple scripts like for IPv6. Anything you'd like to see specifically?
Quoted 4 lines
Quoted 2 lines
IMHO the series makes sense in net-next, but we could argue (some) commits be seen as fixes and I can resend if necessary.
net-next is closed...
I was convinced to have checked, but well, completely missed it. Sorry about that... Will resend when appropriate.
Thanks,
Antoine