From 29ba4fffd396f4beefe34d24d7bcd86cb5c3e492 Mon Sep 17 00:00:00 2001 From: Daniel Borkmann Date: Mon, 5 Jan 2015 23:57:45 +0100 Subject: net: tcp: refactor reinitialization of congestion control We can just move this to an extra function and make the code a bit more readable, no functional change. Joint work with Florian Westphal. Signed-off-by: Florian Westphal Signed-off-by: Daniel Borkmann Signed-off-by: David S. Miller --- net/ipv4/tcp_cong.c | 24 ++++++++++++++---------- 1 file changed, 14 insertions(+), 10 deletions(-) (limited to 'net/ipv4/tcp_cong.c') diff --git a/net/ipv4/tcp_cong.c b/net/ipv4/tcp_cong.c index 27ead0dd16bc..38f2f8aa4ceb 100644 --- a/net/ipv4/tcp_cong.c +++ b/net/ipv4/tcp_cong.c @@ -107,6 +107,18 @@ void tcp_init_congestion_control(struct sock *sk) icsk->icsk_ca_ops->init(sk); } +static void tcp_reinit_congestion_control(struct sock *sk, + const struct tcp_congestion_ops *ca) +{ + struct inet_connection_sock *icsk = inet_csk(sk); + + tcp_cleanup_congestion_control(sk); + icsk->icsk_ca_ops = ca; + + if (sk->sk_state != TCP_CLOSE && icsk->icsk_ca_ops->init) + icsk->icsk_ca_ops->init(sk); +} + /* Manage refcounts on socket close. */ void tcp_cleanup_congestion_control(struct sock *sk) { @@ -262,21 +274,13 @@ int tcp_set_congestion_control(struct sock *sk, const char *name) #endif if (!ca) err = -ENOENT; - else if (!((ca->flags & TCP_CONG_NON_RESTRICTED) || ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))) err = -EPERM; - else if (!try_module_get(ca->owner)) err = -EBUSY; - - else { - tcp_cleanup_congestion_control(sk); - icsk->icsk_ca_ops = ca; - - if (sk->sk_state != TCP_CLOSE && icsk->icsk_ca_ops->init) - icsk->icsk_ca_ops->init(sk); - } + else + tcp_reinit_congestion_control(sk, ca); out: rcu_read_unlock(); return err; -- cgit v1.2.3 From c5c6a8ab45ec0f18733afb4aaade0d4a139d80b3 Mon Sep 17 00:00:00 2001 From: Daniel Borkmann Date: Mon, 5 Jan 2015 23:57:46 +0100 Subject: net: tcp: add key management to congestion control This patch adds necessary infrastructure to the congestion control framework for later per route congestion control support. For a per route congestion control possibility, our aim is to store a unique u32 key identifier into dst metrics, which can then be mapped into a tcp_congestion_ops struct. We argue that having a RTAX key entry is the most simple, generic and easy way to manage, and also keeps the memory footprint of dst entries lower on 64 bit than with storing a pointer directly, for example. Having a unique key id also allows for decoupling actual TCP congestion control module management from the FIB layer, i.e. we don't have to care about expensive module refcounting inside the FIB at this point. We first thought of using an IDR store for the realization, which takes over dynamic assignment of unused key space and also performs the key to pointer mapping in RCU. While doing so, we stumbled upon the issue that due to the nature of dynamic key distribution, it just so happens, arguably in very rare occasions, that excessive module loads and unloads can lead to a possible reuse of previously used key space. Thus, previously stale keys in the dst metric are now being reassigned to a different congestion control algorithm, which might lead to unexpected behaviour. One way to resolve this would have been to walk FIBs on the actually rare occasion of a module unload and reset the metric keys for each FIB in each netns, but that's just very costly. Therefore, we argue a better solution is to reuse the unique congestion control algorithm name member and map that into u32 key space through jhash. For that, we split the flags attribute (as it currently uses 2 bits only anyway) into two u32 attributes, flags and key, so that we can keep the cacheline boundary of 2 cachelines on x86_64 and cache the precalculated key at registration time for the fast path. On average we might expect 2 - 4 modules being loaded worst case perhaps 15, so a key collision possibility is extremely low, and guaranteed collision-free on LE/BE for all in-tree modules. Overall this results in much simpler code, and all without the overhead of an IDR. Due to the deterministic nature, modules can now be unloaded, the congestion control algorithm for a specific but unloaded key will fall back to the default one, and on module reload time it will switch back to the expected algorithm transparently. Joint work with Florian Westphal. Signed-off-by: Florian Westphal Signed-off-by: Daniel Borkmann Signed-off-by: David S. Miller --- include/net/inet_connection_sock.h | 3 +- include/net/tcp.h | 9 +++- net/ipv4/tcp_cong.c | 97 +++++++++++++++++++++++++++++++------- 3 files changed, 91 insertions(+), 18 deletions(-) (limited to 'net/ipv4/tcp_cong.c') diff --git a/include/net/inet_connection_sock.h b/include/net/inet_connection_sock.h index 848e85cb5c61..5976bdecf58b 100644 --- a/include/net/inet_connection_sock.h +++ b/include/net/inet_connection_sock.h @@ -98,7 +98,8 @@ struct inet_connection_sock { const struct tcp_congestion_ops *icsk_ca_ops; const struct inet_connection_sock_af_ops *icsk_af_ops; unsigned int (*icsk_sync_mss)(struct sock *sk, u32 pmtu); - __u8 icsk_ca_state; + __u8 icsk_ca_state:7, + icsk_ca_dst_locked:1; __u8 icsk_retransmits; __u8 icsk_pending; __u8 icsk_backoff; diff --git a/include/net/tcp.h b/include/net/tcp.h index f50f29faf76f..135b70c9a734 100644 --- a/include/net/tcp.h +++ b/include/net/tcp.h @@ -787,6 +787,8 @@ enum tcp_ca_ack_event_flags { #define TCP_CA_MAX 128 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) +#define TCP_CA_UNSPEC 0 + /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ #define TCP_CONG_NON_RESTRICTED 0x1 /* Requires ECN/ECT set on all packets */ @@ -794,7 +796,8 @@ enum tcp_ca_ack_event_flags { struct tcp_congestion_ops { struct list_head list; - unsigned long flags; + u32 key; + u32 flags; /* initialize private data (optional) */ void (*init)(struct sock *sk); @@ -841,6 +844,10 @@ u32 tcp_reno_ssthresh(struct sock *sk); void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); extern struct tcp_congestion_ops tcp_reno; +struct tcp_congestion_ops *tcp_ca_find_key(u32 key); +u32 tcp_ca_get_key_by_name(const char *name); +char *tcp_ca_get_name_by_key(u32 key, char *buffer); + static inline bool tcp_ca_needs_ecn(const struct sock *sk) { const struct inet_connection_sock *icsk = inet_csk(sk); diff --git a/net/ipv4/tcp_cong.c b/net/ipv4/tcp_cong.c index 38f2f8aa4ceb..63c29dba68a8 100644 --- a/net/ipv4/tcp_cong.c +++ b/net/ipv4/tcp_cong.c @@ -13,6 +13,7 @@ #include #include #include +#include #include static DEFINE_SPINLOCK(tcp_cong_list_lock); @@ -31,6 +32,34 @@ static struct tcp_congestion_ops *tcp_ca_find(const char *name) return NULL; } +/* Must be called with rcu lock held */ +static const struct tcp_congestion_ops *__tcp_ca_find_autoload(const char *name) +{ + const struct tcp_congestion_ops *ca = tcp_ca_find(name); +#ifdef CONFIG_MODULES + if (!ca && capable(CAP_NET_ADMIN)) { + rcu_read_unlock(); + request_module("tcp_%s", name); + rcu_read_lock(); + ca = tcp_ca_find(name); + } +#endif + return ca; +} + +/* Simple linear search, not much in here. */ +struct tcp_congestion_ops *tcp_ca_find_key(u32 key) +{ + struct tcp_congestion_ops *e; + + list_for_each_entry_rcu(e, &tcp_cong_list, list) { + if (e->key == key) + return e; + } + + return NULL; +} + /* * Attach new congestion control algorithm to the list * of available options. @@ -45,9 +74,12 @@ int tcp_register_congestion_control(struct tcp_congestion_ops *ca) return -EINVAL; } + ca->key = jhash(ca->name, sizeof(ca->name), strlen(ca->name)); + spin_lock(&tcp_cong_list_lock); - if (tcp_ca_find(ca->name)) { - pr_notice("%s already registered\n", ca->name); + if (ca->key == TCP_CA_UNSPEC || tcp_ca_find_key(ca->key)) { + pr_notice("%s already registered or non-unique key\n", + ca->name); ret = -EEXIST; } else { list_add_tail_rcu(&ca->list, &tcp_cong_list); @@ -70,9 +102,50 @@ void tcp_unregister_congestion_control(struct tcp_congestion_ops *ca) spin_lock(&tcp_cong_list_lock); list_del_rcu(&ca->list); spin_unlock(&tcp_cong_list_lock); + + /* Wait for outstanding readers to complete before the + * module gets removed entirely. + * + * A try_module_get() should fail by now as our module is + * in "going" state since no refs are held anymore and + * module_exit() handler being called. + */ + synchronize_rcu(); } EXPORT_SYMBOL_GPL(tcp_unregister_congestion_control); +u32 tcp_ca_get_key_by_name(const char *name) +{ + const struct tcp_congestion_ops *ca; + u32 key; + + might_sleep(); + + rcu_read_lock(); + ca = __tcp_ca_find_autoload(name); + key = ca ? ca->key : TCP_CA_UNSPEC; + rcu_read_unlock(); + + return key; +} +EXPORT_SYMBOL_GPL(tcp_ca_get_key_by_name); + +char *tcp_ca_get_name_by_key(u32 key, char *buffer) +{ + const struct tcp_congestion_ops *ca; + char *ret = NULL; + + rcu_read_lock(); + ca = tcp_ca_find_key(key); + if (ca) + ret = strncpy(buffer, ca->name, + TCP_CA_NAME_MAX); + rcu_read_unlock(); + + return ret; +} +EXPORT_SYMBOL_GPL(tcp_ca_get_name_by_key); + /* Assign choice of congestion control. */ void tcp_assign_congestion_control(struct sock *sk) { @@ -253,25 +326,17 @@ out: int tcp_set_congestion_control(struct sock *sk, const char *name) { struct inet_connection_sock *icsk = inet_csk(sk); - struct tcp_congestion_ops *ca; + const struct tcp_congestion_ops *ca; int err = 0; - rcu_read_lock(); - ca = tcp_ca_find(name); + if (icsk->icsk_ca_dst_locked) + return -EPERM; - /* no change asking for existing value */ + rcu_read_lock(); + ca = __tcp_ca_find_autoload(name); + /* No change asking for existing value */ if (ca == icsk->icsk_ca_ops) goto out; - -#ifdef CONFIG_MODULES - /* not found attempt to autoload module */ - if (!ca && capable(CAP_NET_ADMIN)) { - rcu_read_unlock(); - request_module("tcp_%s", name); - rcu_read_lock(); - ca = tcp_ca_find(name); - } -#endif if (!ca) err = -ENOENT; else if (!((ca->flags & TCP_CONG_NON_RESTRICTED) || -- cgit v1.2.3 From e73ebb0881ea5534ce606c1d71b4ac44db5c6930 Mon Sep 17 00:00:00 2001 From: Neal Cardwell Date: Wed, 28 Jan 2015 20:01:35 -0500 Subject: tcp: stretch ACK fixes prep LRO, GRO, delayed ACKs, and middleboxes can cause "stretch ACKs" that cover more than the RFC-specified maximum of 2 packets. These stretch ACKs can cause serious performance shortfalls in common congestion control algorithms that were designed and tuned years ago with receiver hosts that were not using LRO or GRO, and were instead politely ACKing every other packet. This patch series fixes Reno and CUBIC to handle stretch ACKs. This patch prepares for the upcoming stretch ACK bug fix patches. It adds an "acked" parameter to tcp_cong_avoid_ai() to allow for future fixes to tcp_cong_avoid_ai() to correctly handle stretch ACKs, and changes all congestion control algorithms to pass in 1 for the ACKed count. It also changes tcp_slow_start() to return the number of packet ACK "credits" that were not processed in slow start mode, and can be processed by the congestion control module in additive increase mode. In future patches we will fix tcp_cong_avoid_ai() to handle stretch ACKs, and fix Reno and CUBIC handling of stretch ACKs in slow start and additive increase mode. Reported-by: Eyal Perry Signed-off-by: Neal Cardwell Signed-off-by: Yuchung Cheng Signed-off-by: Eric Dumazet Signed-off-by: David S. Miller --- include/net/tcp.h | 4 ++-- net/ipv4/tcp_bic.c | 2 +- net/ipv4/tcp_cong.c | 11 +++++++---- net/ipv4/tcp_cubic.c | 2 +- net/ipv4/tcp_scalable.c | 3 ++- net/ipv4/tcp_veno.c | 2 +- net/ipv4/tcp_yeah.c | 2 +- 7 files changed, 15 insertions(+), 11 deletions(-) (limited to 'net/ipv4/tcp_cong.c') diff --git a/include/net/tcp.h b/include/net/tcp.h index f50f29faf76f..9d9111ef43ae 100644 --- a/include/net/tcp.h +++ b/include/net/tcp.h @@ -834,8 +834,8 @@ void tcp_get_available_congestion_control(char *buf, size_t len); void tcp_get_allowed_congestion_control(char *buf, size_t len); int tcp_set_allowed_congestion_control(char *allowed); int tcp_set_congestion_control(struct sock *sk, const char *name); -void tcp_slow_start(struct tcp_sock *tp, u32 acked); -void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w); +u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); +void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); u32 tcp_reno_ssthresh(struct sock *sk); void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); diff --git a/net/ipv4/tcp_bic.c b/net/ipv4/tcp_bic.c index bb395d46a389..c037644eafb7 100644 --- a/net/ipv4/tcp_bic.c +++ b/net/ipv4/tcp_bic.c @@ -150,7 +150,7 @@ static void bictcp_cong_avoid(struct sock *sk, u32 ack, u32 acked) tcp_slow_start(tp, acked); else { bictcp_update(ca, tp->snd_cwnd); - tcp_cong_avoid_ai(tp, ca->cnt); + tcp_cong_avoid_ai(tp, ca->cnt, 1); } } diff --git a/net/ipv4/tcp_cong.c b/net/ipv4/tcp_cong.c index 27ead0dd16bc..6826017c12d1 100644 --- a/net/ipv4/tcp_cong.c +++ b/net/ipv4/tcp_cong.c @@ -291,25 +291,28 @@ int tcp_set_congestion_control(struct sock *sk, const char *name) * ABC caps N to 2. Slow start exits when cwnd grows over ssthresh and * returns the leftover acks to adjust cwnd in congestion avoidance mode. */ -void tcp_slow_start(struct tcp_sock *tp, u32 acked) +u32 tcp_slow_start(struct tcp_sock *tp, u32 acked) { u32 cwnd = tp->snd_cwnd + acked; if (cwnd > tp->snd_ssthresh) cwnd = tp->snd_ssthresh + 1; + acked -= cwnd - tp->snd_cwnd; tp->snd_cwnd = min(cwnd, tp->snd_cwnd_clamp); + + return acked; } EXPORT_SYMBOL_GPL(tcp_slow_start); /* In theory this is tp->snd_cwnd += 1 / tp->snd_cwnd (or alternative w) */ -void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w) +void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked) { if (tp->snd_cwnd_cnt >= w) { if (tp->snd_cwnd < tp->snd_cwnd_clamp) tp->snd_cwnd++; tp->snd_cwnd_cnt = 0; } else { - tp->snd_cwnd_cnt++; + tp->snd_cwnd_cnt += acked; } } EXPORT_SYMBOL_GPL(tcp_cong_avoid_ai); @@ -333,7 +336,7 @@ void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked) tcp_slow_start(tp, acked); /* In dangerous area, increase slowly. */ else - tcp_cong_avoid_ai(tp, tp->snd_cwnd); + tcp_cong_avoid_ai(tp, tp->snd_cwnd, 1); } EXPORT_SYMBOL_GPL(tcp_reno_cong_avoid); diff --git a/net/ipv4/tcp_cubic.c b/net/ipv4/tcp_cubic.c index 6b6002416a73..df4bc4d87e58 100644 --- a/net/ipv4/tcp_cubic.c +++ b/net/ipv4/tcp_cubic.c @@ -320,7 +320,7 @@ static void bictcp_cong_avoid(struct sock *sk, u32 ack, u32 acked) tcp_slow_start(tp, acked); } else { bictcp_update(ca, tp->snd_cwnd); - tcp_cong_avoid_ai(tp, ca->cnt); + tcp_cong_avoid_ai(tp, ca->cnt, 1); } } diff --git a/net/ipv4/tcp_scalable.c b/net/ipv4/tcp_scalable.c index 6824afb65d93..333bcb2415ff 100644 --- a/net/ipv4/tcp_scalable.c +++ b/net/ipv4/tcp_scalable.c @@ -25,7 +25,8 @@ static void tcp_scalable_cong_avoid(struct sock *sk, u32 ack, u32 acked) if (tp->snd_cwnd <= tp->snd_ssthresh) tcp_slow_start(tp, acked); else - tcp_cong_avoid_ai(tp, min(tp->snd_cwnd, TCP_SCALABLE_AI_CNT)); + tcp_cong_avoid_ai(tp, min(tp->snd_cwnd, TCP_SCALABLE_AI_CNT), + 1); } static u32 tcp_scalable_ssthresh(struct sock *sk) diff --git a/net/ipv4/tcp_veno.c b/net/ipv4/tcp_veno.c index a4d2d2d88dca..112151eeee45 100644 --- a/net/ipv4/tcp_veno.c +++ b/net/ipv4/tcp_veno.c @@ -159,7 +159,7 @@ static void tcp_veno_cong_avoid(struct sock *sk, u32 ack, u32 acked) /* In the "non-congestive state", increase cwnd * every rtt. */ - tcp_cong_avoid_ai(tp, tp->snd_cwnd); + tcp_cong_avoid_ai(tp, tp->snd_cwnd, 1); } else { /* In the "congestive state", increase cwnd * every other rtt. diff --git a/net/ipv4/tcp_yeah.c b/net/ipv4/tcp_yeah.c index cd7273218598..17d35662930d 100644 --- a/net/ipv4/tcp_yeah.c +++ b/net/ipv4/tcp_yeah.c @@ -92,7 +92,7 @@ static void tcp_yeah_cong_avoid(struct sock *sk, u32 ack, u32 acked) } else { /* Reno */ - tcp_cong_avoid_ai(tp, tp->snd_cwnd); + tcp_cong_avoid_ai(tp, tp->snd_cwnd, 1); } /* The key players are v_vegas.beg_snd_una and v_beg_snd_nxt. -- cgit v1.2.3 From 814d488c61260521b1b3cc97063700a5a6667c8f Mon Sep 17 00:00:00 2001 From: Neal Cardwell Date: Wed, 28 Jan 2015 20:01:36 -0500 Subject: tcp: fix the timid additive increase on stretch ACKs tcp_cong_avoid_ai() was too timid (snd_cwnd increased too slowly) on "stretch ACKs" -- cases where the receiver ACKed more than 1 packet in a single ACK. For example, suppose w is 10 and we get a stretch ACK for 20 packets, so acked is 20. We ought to increase snd_cwnd by 2 (since acked/w = 20/10 = 2), but instead we were only increasing cwnd by 1. This patch fixes that behavior. Reported-by: Eyal Perry Signed-off-by: Neal Cardwell Signed-off-by: Yuchung Cheng Signed-off-by: Eric Dumazet Signed-off-by: David S. Miller --- net/ipv4/tcp_cong.c | 15 +++++++++------ 1 file changed, 9 insertions(+), 6 deletions(-) (limited to 'net/ipv4/tcp_cong.c') diff --git a/net/ipv4/tcp_cong.c b/net/ipv4/tcp_cong.c index 6826017c12d1..faaee5338bea 100644 --- a/net/ipv4/tcp_cong.c +++ b/net/ipv4/tcp_cong.c @@ -304,16 +304,19 @@ u32 tcp_slow_start(struct tcp_sock *tp, u32 acked) } EXPORT_SYMBOL_GPL(tcp_slow_start); -/* In theory this is tp->snd_cwnd += 1 / tp->snd_cwnd (or alternative w) */ +/* In theory this is tp->snd_cwnd += 1 / tp->snd_cwnd (or alternative w), + * for every packet that was ACKed. + */ void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked) { + tp->snd_cwnd_cnt += acked; if (tp->snd_cwnd_cnt >= w) { - if (tp->snd_cwnd < tp->snd_cwnd_clamp) - tp->snd_cwnd++; - tp->snd_cwnd_cnt = 0; - } else { - tp->snd_cwnd_cnt += acked; + u32 delta = tp->snd_cwnd_cnt / w; + + tp->snd_cwnd_cnt -= delta * w; + tp->snd_cwnd += delta; } + tp->snd_cwnd = min(tp->snd_cwnd, tp->snd_cwnd_clamp); } EXPORT_SYMBOL_GPL(tcp_cong_avoid_ai); -- cgit v1.2.3 From c22bdca94782f05b9337d8548bde51b2f38ef17f Mon Sep 17 00:00:00 2001 From: Neal Cardwell Date: Wed, 28 Jan 2015 20:01:37 -0500 Subject: tcp: fix stretch ACK bugs in Reno Change Reno to properly handle stretch ACKs in additive increase mode by passing in the count of ACKed packets to tcp_cong_avoid_ai(). In addition, if snd_cwnd crosses snd_ssthresh during slow start processing, and we then exit slow start mode, we need to carry over any remaining "credit" for packets ACKed and apply that to additive increase by passing this remaining "acked" count to tcp_cong_avoid_ai(). Reported-by: Eyal Perry Signed-off-by: Neal Cardwell Signed-off-by: Yuchung Cheng Signed-off-by: Eric Dumazet Signed-off-by: David S. Miller --- net/ipv4/tcp_cong.c | 10 ++++++---- 1 file changed, 6 insertions(+), 4 deletions(-) (limited to 'net/ipv4/tcp_cong.c') diff --git a/net/ipv4/tcp_cong.c b/net/ipv4/tcp_cong.c index faaee5338bea..8670e68e2ce6 100644 --- a/net/ipv4/tcp_cong.c +++ b/net/ipv4/tcp_cong.c @@ -335,11 +335,13 @@ void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked) return; /* In "safe" area, increase. */ - if (tp->snd_cwnd <= tp->snd_ssthresh) - tcp_slow_start(tp, acked); + if (tp->snd_cwnd <= tp->snd_ssthresh) { + acked = tcp_slow_start(tp, acked); + if (!acked) + return; + } /* In dangerous area, increase slowly. */ - else - tcp_cong_avoid_ai(tp, tp->snd_cwnd, 1); + tcp_cong_avoid_ai(tp, tp->snd_cwnd, acked); } EXPORT_SYMBOL_GPL(tcp_reno_cong_avoid); -- cgit v1.2.3