Angelo P. Castellani | 5ef8147 | 2007-02-22 00:23:05 -0800 | [diff] [blame] | 1 | #include <linux/mm.h> |
| 2 | #include <linux/module.h> |
| 3 | #include <linux/skbuff.h> |
| 4 | #include <linux/inet_diag.h> |
Stephen Hemminger | 3927f2e | 2007-03-25 19:54:23 -0700 | [diff] [blame] | 5 | #include <asm/div64.h> |
Angelo P. Castellani | 5ef8147 | 2007-02-22 00:23:05 -0800 | [diff] [blame] | 6 | |
| 7 | #include <net/tcp.h> |
| 8 | |
| 9 | /* Vegas variables */ |
| 10 | struct vegas { |
| 11 | u32 beg_snd_nxt; /* right edge during last RTT */ |
| 12 | u32 beg_snd_una; /* left edge during last RTT */ |
| 13 | u32 beg_snd_cwnd; /* saves the size of the cwnd */ |
| 14 | u8 doing_vegas_now;/* if true, do vegas for this RTT */ |
| 15 | u16 cntRTT; /* # of RTTs measured within last RTT */ |
| 16 | u32 minRTT; /* min of RTTs measured within last RTT (in usec) */ |
| 17 | u32 baseRTT; /* the min of all Vegas RTT measurements seen (in usec) */ |
| 18 | }; |
| 19 | |
| 20 | /* There are several situations when we must "re-start" Vegas: |
| 21 | * |
| 22 | * o when a connection is established |
| 23 | * o after an RTO |
| 24 | * o after fast recovery |
| 25 | * o when we send a packet and there is no outstanding |
| 26 | * unacknowledged data (restarting an idle connection) |
| 27 | * |
| 28 | * In these circumstances we cannot do a Vegas calculation at the |
| 29 | * end of the first RTT, because any calculation we do is using |
| 30 | * stale info -- both the saved cwnd and congestion feedback are |
| 31 | * stale. |
| 32 | * |
| 33 | * Instead we must wait until the completion of an RTT during |
| 34 | * which we actually receive ACKs. |
| 35 | */ |
| 36 | static inline void vegas_enable(struct sock *sk) |
| 37 | { |
| 38 | const struct tcp_sock *tp = tcp_sk(sk); |
| 39 | struct vegas *vegas = inet_csk_ca(sk); |
| 40 | |
| 41 | /* Begin taking Vegas samples next time we send something. */ |
| 42 | vegas->doing_vegas_now = 1; |
| 43 | |
| 44 | /* Set the beginning of the next send window. */ |
| 45 | vegas->beg_snd_nxt = tp->snd_nxt; |
| 46 | |
| 47 | vegas->cntRTT = 0; |
| 48 | vegas->minRTT = 0x7fffffff; |
| 49 | } |
| 50 | |
| 51 | /* Stop taking Vegas samples for now. */ |
| 52 | static inline void vegas_disable(struct sock *sk) |
| 53 | { |
| 54 | struct vegas *vegas = inet_csk_ca(sk); |
| 55 | |
| 56 | vegas->doing_vegas_now = 0; |
| 57 | } |
| 58 | |
| 59 | static void tcp_vegas_init(struct sock *sk) |
| 60 | { |
| 61 | struct vegas *vegas = inet_csk_ca(sk); |
| 62 | |
| 63 | vegas->baseRTT = 0x7fffffff; |
| 64 | vegas_enable(sk); |
| 65 | } |
| 66 | |
| 67 | static void tcp_vegas_state(struct sock *sk, u8 ca_state) |
| 68 | { |
| 69 | |
| 70 | if (ca_state == TCP_CA_Open) |
| 71 | vegas_enable(sk); |
| 72 | else |
| 73 | vegas_disable(sk); |
| 74 | } |
| 75 | |
| 76 | /* Do RTT sampling needed for Vegas. |
| 77 | * Basically we: |
| 78 | * o min-filter RTT samples from within an RTT to get the current |
| 79 | * propagation delay + queuing delay (we are min-filtering to try to |
| 80 | * avoid the effects of delayed ACKs) |
| 81 | * o min-filter RTT samples from a much longer window (forever for now) |
| 82 | * to find the propagation delay (baseRTT) |
| 83 | */ |
| 84 | static void tcp_vegas_rtt_calc(struct sock *sk, u32 usrtt) |
| 85 | { |
| 86 | struct vegas *vegas = inet_csk_ca(sk); |
| 87 | u32 vrtt = usrtt + 1; /* Never allow zero rtt or baseRTT */ |
| 88 | |
| 89 | /* Filter to find propagation delay: */ |
| 90 | if (vrtt < vegas->baseRTT) |
| 91 | vegas->baseRTT = vrtt; |
| 92 | |
| 93 | /* Find the min RTT during the last RTT to find |
| 94 | * the current prop. delay + queuing delay: |
| 95 | */ |
| 96 | vegas->minRTT = min(vegas->minRTT, vrtt); |
| 97 | vegas->cntRTT++; |
| 98 | } |
| 99 | |
| 100 | /* |
| 101 | * If the connection is idle and we are restarting, |
| 102 | * then we don't want to do any Vegas calculations |
| 103 | * until we get fresh RTT samples. So when we |
| 104 | * restart, we reset our Vegas state to a clean |
| 105 | * slate. After we get acks for this flight of |
| 106 | * packets, _then_ we can make Vegas calculations |
| 107 | * again. |
| 108 | */ |
| 109 | static void tcp_vegas_cwnd_event(struct sock *sk, enum tcp_ca_event event) |
| 110 | { |
| 111 | if (event == CA_EVENT_CWND_RESTART || |
| 112 | event == CA_EVENT_TX_START) |
| 113 | tcp_vegas_init(sk); |
| 114 | } |
| 115 | |
| 116 | /* Extract info for Tcp socket info provided via netlink. */ |
| 117 | static void tcp_vegas_get_info(struct sock *sk, u32 ext, |
| 118 | struct sk_buff *skb) |
| 119 | { |
| 120 | const struct vegas *ca = inet_csk_ca(sk); |
| 121 | if (ext & (1 << (INET_DIAG_VEGASINFO - 1))) { |
| 122 | struct tcpvegas_info *info; |
| 123 | |
| 124 | info = RTA_DATA(__RTA_PUT(skb, INET_DIAG_VEGASINFO, |
| 125 | sizeof(*info))); |
| 126 | |
| 127 | info->tcpv_enabled = ca->doing_vegas_now; |
| 128 | info->tcpv_rttcnt = ca->cntRTT; |
| 129 | info->tcpv_rtt = ca->baseRTT; |
| 130 | info->tcpv_minrtt = ca->minRTT; |
| 131 | rtattr_failure: ; |
| 132 | } |
| 133 | } |
| 134 | |
| 135 | |