diff options
| -rw-r--r-- | Documentation/networking/ip-sysctl.rst | 75 | ||||
| -rw-r--r-- | include/net/netns/ipv4.h | 5 | ||||
| -rw-r--r-- | net/ipv4/sysctl_net_ipv4.c | 43 | ||||
| -rw-r--r-- | net/ipv4/tcp_ipv4.c | 8 |
4 files changed, 131 insertions, 0 deletions
diff --git a/Documentation/networking/ip-sysctl.rst b/Documentation/networking/ip-sysctl.rst index e7b3fa7bb3f7..815efc89ad73 100644 --- a/Documentation/networking/ip-sysctl.rst +++ b/Documentation/networking/ip-sysctl.rst @@ -1069,6 +1069,81 @@ tcp_child_ehash_entries - INTEGER Default: 0 +tcp_plb_enabled - BOOLEAN + If set and the underlying congestion control (e.g. DCTCP) supports + and enables PLB feature, TCP PLB (Protective Load Balancing) is + enabled. PLB is described in the following paper: + https://doi.org/10.1145/3544216.3544226. Based on PLB parameters, + upon sensing sustained congestion, TCP triggers a change in + flow label field for outgoing IPv6 packets. A change in flow label + field potentially changes the path of outgoing packets for switches + that use ECMP/WCMP for routing. + + PLB changes socket txhash which results in a change in IPv6 Flow Label + field, and currently no-op for IPv4 headers. It is possible + to apply PLB for IPv4 with other network header fields (e.g. TCP + or IPv4 options) or using encapsulation where outer header is used + by switches to determine next hop. In either case, further host + and switch side changes will be needed. + + When set, PLB assumes that congestion signal (e.g. ECN) is made + available and used by congestion control module to estimate a + congestion measure (e.g. ce_ratio). PLB needs a congestion measure to + make repathing decisions. + + Default: FALSE + +tcp_plb_idle_rehash_rounds - INTEGER + Number of consecutive congested rounds (RTT) seen after which + a rehash can be performed, given there are no packets in flight. + This is referred to as M in PLB paper: + https://doi.org/10.1145/3544216.3544226. + + Possible Values: 0 - 31 + + Default: 3 + +tcp_plb_rehash_rounds - INTEGER + Number of consecutive congested rounds (RTT) seen after which + a forced rehash can be performed. Be careful when setting this + parameter, as a small value increases the risk of retransmissions. + This is referred to as N in PLB paper: + https://doi.org/10.1145/3544216.3544226. + + Possible Values: 0 - 31 + + Default: 12 + +tcp_plb_suspend_rto_sec - INTEGER + Time, in seconds, to suspend PLB in event of an RTO. In order to avoid + having PLB repath onto a connectivity "black hole", after an RTO a TCP + connection suspends PLB repathing for a random duration between 1x and + 2x of this parameter. Randomness is added to avoid concurrent rehashing + of multiple TCP connections. This should be set corresponding to the + amount of time it takes to repair a failed link. + + Possible Values: 0 - 255 + + Default: 60 + +tcp_plb_cong_thresh - INTEGER + Fraction of packets marked with congestion over a round (RTT) to + tag that round as congested. This is referred to as K in the PLB paper: + https://doi.org/10.1145/3544216.3544226. + + The 0-1 fraction range is mapped to 0-256 range to avoid floating + point operations. For example, 128 means that if at least 50% of + the packets in a round were marked as congested then the round + will be tagged as congested. + + Setting threshold to 0 means that PLB repaths every RTT regardless + of congestion. This is not intended behavior for PLB and should be + used only for experimentation purpose. + + Possible Values: 0 - 256 + + Default: 128 + UDP variables ============= diff --git a/include/net/netns/ipv4.h b/include/net/netns/ipv4.h index 1b8004679445..25f90bba4889 100644 --- a/include/net/netns/ipv4.h +++ b/include/net/netns/ipv4.h @@ -183,6 +183,11 @@ struct netns_ipv4 { unsigned long tfo_active_disable_stamp; u32 tcp_challenge_timestamp; u32 tcp_challenge_count; + u8 sysctl_tcp_plb_enabled; + u8 sysctl_tcp_plb_idle_rehash_rounds; + u8 sysctl_tcp_plb_rehash_rounds; + u8 sysctl_tcp_plb_suspend_rto_sec; + int sysctl_tcp_plb_cong_thresh; int sysctl_udp_wmem_min; int sysctl_udp_rmem_min; diff --git a/net/ipv4/sysctl_net_ipv4.c b/net/ipv4/sysctl_net_ipv4.c index 9b8a6db7a66b..0af28cedd071 100644 --- a/net/ipv4/sysctl_net_ipv4.c +++ b/net/ipv4/sysctl_net_ipv4.c @@ -40,6 +40,8 @@ static int one_day_secs = 24 * 3600; static u32 fib_multipath_hash_fields_all_mask __maybe_unused = FIB_MULTIPATH_HASH_FIELD_ALL_MASK; static unsigned int tcp_child_ehash_entries_max = 16 * 1024 * 1024; +static int tcp_plb_max_rounds = 31; +static int tcp_plb_max_cong_thresh = 256; /* obsolete */ static int sysctl_tcp_low_latency __read_mostly; @@ -1384,6 +1386,47 @@ static struct ctl_table ipv4_net_table[] = { .extra1 = SYSCTL_ZERO, .extra2 = SYSCTL_TWO, }, + { + .procname = "tcp_plb_enabled", + .data = &init_net.ipv4.sysctl_tcp_plb_enabled, + .maxlen = sizeof(u8), + .mode = 0644, + .proc_handler = proc_dou8vec_minmax, + .extra1 = SYSCTL_ZERO, + .extra2 = SYSCTL_ONE, + }, + { + .procname = "tcp_plb_idle_rehash_rounds", + .data = &init_net.ipv4.sysctl_tcp_plb_idle_rehash_rounds, + .maxlen = sizeof(u8), + .mode = 0644, + .proc_handler = proc_dou8vec_minmax, + .extra2 = &tcp_plb_max_rounds, + }, + { + .procname = "tcp_plb_rehash_rounds", + .data = &init_net.ipv4.sysctl_tcp_plb_rehash_rounds, + .maxlen = sizeof(u8), + .mode = 0644, + .proc_handler = proc_dou8vec_minmax, + .extra2 = &tcp_plb_max_rounds, + }, + { + .procname = "tcp_plb_suspend_rto_sec", + .data = &init_net.ipv4.sysctl_tcp_plb_suspend_rto_sec, + .maxlen = sizeof(u8), + .mode = 0644, + .proc_handler = proc_dou8vec_minmax, + }, + { + .procname = "tcp_plb_cong_thresh", + .data = &init_net.ipv4.sysctl_tcp_plb_cong_thresh, + .maxlen = sizeof(int), + .mode = 0644, + .proc_handler = proc_dointvec_minmax, + .extra1 = SYSCTL_ZERO, + .extra2 = &tcp_plb_max_cong_thresh, + }, { } }; diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c index 87d440f47a70..58b838b56c7f 100644 --- a/net/ipv4/tcp_ipv4.c +++ b/net/ipv4/tcp_ipv4.c @@ -3218,6 +3218,14 @@ static int __net_init tcp_sk_init(struct net *net) net->ipv4.sysctl_tcp_fastopen_blackhole_timeout = 0; atomic_set(&net->ipv4.tfo_active_disable_times, 0); + /* Set default values for PLB */ + net->ipv4.sysctl_tcp_plb_enabled = 0; /* Disabled by default */ + net->ipv4.sysctl_tcp_plb_idle_rehash_rounds = 3; + net->ipv4.sysctl_tcp_plb_rehash_rounds = 12; + net->ipv4.sysctl_tcp_plb_suspend_rto_sec = 60; + /* Default congestion threshold for PLB to mark a round is 50% */ + net->ipv4.sysctl_tcp_plb_cong_thresh = 128; + /* Reno is always built in */ if (!net_eq(net, &init_net) && bpf_try_module_get(init_net.ipv4.tcp_congestion_control, |