Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * INET An implementation of the TCP/IP protocol suite for the LINUX |
| 3 | * operating system. INET is implemented using the BSD Socket |
| 4 | * interface as the means of communication with the user level. |
| 5 | * |
| 6 | * Implementation of the Transmission Control Protocol(TCP). |
| 7 | * |
Jesper Juhl | 02c30a8 | 2005-05-05 16:16:16 -0700 | [diff] [blame] | 8 | * Authors: Ross Biro |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 9 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
| 10 | * Mark Evans, <evansmp@uhura.aston.ac.uk> |
| 11 | * Corey Minyard <wf-rch!minyard@relay.EU.net> |
| 12 | * Florian La Roche, <flla@stud.uni-sb.de> |
| 13 | * Charles Hedrick, <hedrick@klinzhai.rutgers.edu> |
| 14 | * Linus Torvalds, <torvalds@cs.helsinki.fi> |
| 15 | * Alan Cox, <gw4pts@gw4pts.ampr.org> |
| 16 | * Matthew Dillon, <dillon@apollo.west.oic.com> |
| 17 | * Arnt Gulbrandsen, <agulbra@nvg.unit.no> |
| 18 | * Jorge Cwik, <jorge@laser.satlink.net> |
| 19 | */ |
| 20 | |
| 21 | /* |
| 22 | * Changes: |
| 23 | * Pedro Roque : Fast Retransmit/Recovery. |
| 24 | * Two receive queues. |
| 25 | * Retransmit queue handled by TCP. |
| 26 | * Better retransmit timer handling. |
| 27 | * New congestion avoidance. |
| 28 | * Header prediction. |
| 29 | * Variable renaming. |
| 30 | * |
| 31 | * Eric : Fast Retransmit. |
| 32 | * Randy Scott : MSS option defines. |
| 33 | * Eric Schenk : Fixes to slow start algorithm. |
| 34 | * Eric Schenk : Yet another double ACK bug. |
| 35 | * Eric Schenk : Delayed ACK bug fixes. |
| 36 | * Eric Schenk : Floyd style fast retrans war avoidance. |
| 37 | * David S. Miller : Don't allow zero congestion window. |
| 38 | * Eric Schenk : Fix retransmitter so that it sends |
| 39 | * next packet on ack of previous packet. |
| 40 | * Andi Kleen : Moved open_request checking here |
| 41 | * and process RSTs for open_requests. |
| 42 | * Andi Kleen : Better prune_queue, and other fixes. |
Stephen Hemminger | caa20d9a | 2005-11-10 17:13:47 -0800 | [diff] [blame] | 43 | * Andrey Savochkin: Fix RTT measurements in the presence of |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 44 | * timestamps. |
| 45 | * Andrey Savochkin: Check sequence numbers correctly when |
| 46 | * removing SACKs due to in sequence incoming |
| 47 | * data segments. |
| 48 | * Andi Kleen: Make sure we never ack data there is not |
| 49 | * enough room for. Also make this condition |
| 50 | * a fatal error if it might still happen. |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 51 | * Andi Kleen: Add tcp_measure_rcv_mss to make |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 52 | * connections with MSS<min(MTU,ann. MSS) |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 53 | * work without delayed acks. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 54 | * Andi Kleen: Process packets with PSH set in the |
| 55 | * fast path. |
| 56 | * J Hadi Salim: ECN support |
| 57 | * Andrei Gurtov, |
| 58 | * Pasi Sarolahti, |
| 59 | * Panu Kuhlberg: Experimental audit of TCP (re)transmission |
| 60 | * engine. Lots of bugs are found. |
| 61 | * Pasi Sarolahti: F-RTO for dealing with spurious RTOs |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 62 | */ |
| 63 | |
Joe Perches | afd46503 | 2012-03-12 07:03:32 +0000 | [diff] [blame] | 64 | #define pr_fmt(fmt) "TCP: " fmt |
| 65 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 66 | #include <linux/mm.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 67 | #include <linux/slab.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 68 | #include <linux/module.h> |
| 69 | #include <linux/sysctl.h> |
Ilpo Järvinen | a0bffff | 2009-03-21 13:36:17 -0700 | [diff] [blame] | 70 | #include <linux/kernel.h> |
Satoru SATOH | 5ffc02a | 2008-05-04 22:14:42 -0700 | [diff] [blame] | 71 | #include <net/dst.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 72 | #include <net/tcp.h> |
| 73 | #include <net/inet_common.h> |
| 74 | #include <linux/ipsec.h> |
| 75 | #include <asm/unaligned.h> |
Chris Leech | 1a2449a | 2006-05-23 18:05:53 -0700 | [diff] [blame] | 76 | #include <net/netdma.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 77 | |
Brian Haley | ab32ea5 | 2006-09-22 14:15:41 -0700 | [diff] [blame] | 78 | int sysctl_tcp_timestamps __read_mostly = 1; |
| 79 | int sysctl_tcp_window_scaling __read_mostly = 1; |
| 80 | int sysctl_tcp_sack __read_mostly = 1; |
| 81 | int sysctl_tcp_fack __read_mostly = 1; |
| 82 | int sysctl_tcp_reordering __read_mostly = TCP_FASTRETRANS_THRESH; |
Eric Dumazet | 4bc2f18 | 2010-07-09 21:22:10 +0000 | [diff] [blame] | 83 | EXPORT_SYMBOL(sysctl_tcp_reordering); |
Brian Haley | ab32ea5 | 2006-09-22 14:15:41 -0700 | [diff] [blame] | 84 | int sysctl_tcp_dsack __read_mostly = 1; |
| 85 | int sysctl_tcp_app_win __read_mostly = 31; |
Eric Dumazet | b49960a | 2012-05-02 02:28:41 +0000 | [diff] [blame] | 86 | int sysctl_tcp_adv_win_scale __read_mostly = 1; |
Eric Dumazet | 4bc2f18 | 2010-07-09 21:22:10 +0000 | [diff] [blame] | 87 | EXPORT_SYMBOL(sysctl_tcp_adv_win_scale); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 88 | |
Eric Dumazet | 282f23c | 2012-07-17 10:13:05 +0200 | [diff] [blame] | 89 | /* rfc5961 challenge ack rate limiting */ |
| 90 | int sysctl_tcp_challenge_ack_limit = 100; |
| 91 | |
Brian Haley | ab32ea5 | 2006-09-22 14:15:41 -0700 | [diff] [blame] | 92 | int sysctl_tcp_stdurg __read_mostly; |
| 93 | int sysctl_tcp_rfc1337 __read_mostly; |
| 94 | int sysctl_tcp_max_orphans __read_mostly = NR_FILE; |
Ilpo Järvinen | c96fd3d | 2007-09-20 11:36:37 -0700 | [diff] [blame] | 95 | int sysctl_tcp_frto __read_mostly = 2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 96 | |
Andreas Petlund | 7e38017 | 2010-02-18 04:48:19 +0000 | [diff] [blame] | 97 | int sysctl_tcp_thin_dupack __read_mostly; |
| 98 | |
Brian Haley | ab32ea5 | 2006-09-22 14:15:41 -0700 | [diff] [blame] | 99 | int sysctl_tcp_moderate_rcvbuf __read_mostly = 1; |
Nandita Dukkipati | 6ba8a3b | 2013-03-11 10:00:43 +0000 | [diff] [blame] | 100 | int sysctl_tcp_early_retrans __read_mostly = 3; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 101 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 102 | #define FLAG_DATA 0x01 /* Incoming frame contained data. */ |
| 103 | #define FLAG_WIN_UPDATE 0x02 /* Incoming ACK was a window update. */ |
| 104 | #define FLAG_DATA_ACKED 0x04 /* This ACK acknowledged new data. */ |
| 105 | #define FLAG_RETRANS_DATA_ACKED 0x08 /* "" "" some of which was retransmitted. */ |
| 106 | #define FLAG_SYN_ACKED 0x10 /* This ACK acknowledged SYN. */ |
| 107 | #define FLAG_DATA_SACKED 0x20 /* New SACK. */ |
| 108 | #define FLAG_ECE 0x40 /* ECE in this ACK */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 109 | #define FLAG_SLOWPATH 0x100 /* Do not skip RFC checks for window update.*/ |
Yuchung Cheng | e33099f | 2013-03-20 13:33:00 +0000 | [diff] [blame] | 110 | #define FLAG_ORIG_SACK_ACKED 0x200 /* Never retransmitted data are (s)acked */ |
Ilpo Järvinen | 2e60529 | 2007-08-02 19:46:58 -0700 | [diff] [blame] | 111 | #define FLAG_SND_UNA_ADVANCED 0x400 /* Snd_una was changed (!= FLAG_DATA_ACKED) */ |
Ryousei Takano | 564262c1 | 2007-10-25 23:03:52 -0700 | [diff] [blame] | 112 | #define FLAG_DSACKING_ACK 0x800 /* SACK blocks contained D-SACK info */ |
Ilpo Järvinen | cadbd03 | 2007-12-31 04:49:21 -0800 | [diff] [blame] | 113 | #define FLAG_SACK_RENEGING 0x2000 /* snd_una advanced to a sacked seq */ |
Eric Dumazet | 12fb3dd | 2013-04-19 07:19:48 +0000 | [diff] [blame] | 114 | #define FLAG_UPDATE_TS_RECENT 0x4000 /* tcp_replace_ts_recent() */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 115 | |
| 116 | #define FLAG_ACKED (FLAG_DATA_ACKED|FLAG_SYN_ACKED) |
| 117 | #define FLAG_NOT_DUP (FLAG_DATA|FLAG_WIN_UPDATE|FLAG_ACKED) |
| 118 | #define FLAG_CA_ALERT (FLAG_DATA_SACKED|FLAG_ECE) |
| 119 | #define FLAG_FORWARD_PROGRESS (FLAG_ACKED|FLAG_DATA_SACKED) |
| 120 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 121 | #define TCP_REMNANT (TCP_FLAG_FIN|TCP_FLAG_URG|TCP_FLAG_SYN|TCP_FLAG_PSH) |
Ilpo Järvinen | bdf1ee5 | 2007-05-27 02:04:16 -0700 | [diff] [blame] | 122 | #define TCP_HP_BITS (~(TCP_RESERVED_BITS|TCP_FLAG_PSH)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 123 | |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 124 | /* Adapt the MSS value used to make delayed ack decision to the |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 125 | * real world. |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 126 | */ |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 127 | static void tcp_measure_rcv_mss(struct sock *sk, const struct sk_buff *skb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 128 | { |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 129 | struct inet_connection_sock *icsk = inet_csk(sk); |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 130 | const unsigned int lss = icsk->icsk_ack.last_seg_size; |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 131 | unsigned int len; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 132 | |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 133 | icsk->icsk_ack.last_seg_size = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 134 | |
| 135 | /* skb->len may jitter because of SACKs, even if peer |
| 136 | * sends good full-sized frames. |
| 137 | */ |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 138 | len = skb_shinfo(skb)->gso_size ? : skb->len; |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 139 | if (len >= icsk->icsk_ack.rcv_mss) { |
| 140 | icsk->icsk_ack.rcv_mss = len; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 141 | } else { |
| 142 | /* Otherwise, we make more careful check taking into account, |
| 143 | * that SACKs block is variable. |
| 144 | * |
| 145 | * "len" is invariant segment length, including TCP header. |
| 146 | */ |
Arnaldo Carvalho de Melo | 9c70220 | 2007-04-25 18:04:18 -0700 | [diff] [blame] | 147 | len += skb->data - skb_transport_header(skb); |
William Allen Simpson | bee7ca9 | 2009-11-10 09:51:18 +0000 | [diff] [blame] | 148 | if (len >= TCP_MSS_DEFAULT + sizeof(struct tcphdr) || |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 149 | /* If PSH is not set, packet should be |
| 150 | * full sized, provided peer TCP is not badly broken. |
| 151 | * This observation (if it is correct 8)) allows |
| 152 | * to handle super-low mtu links fairly. |
| 153 | */ |
| 154 | (len >= TCP_MIN_MSS + sizeof(struct tcphdr) && |
Arnaldo Carvalho de Melo | aa8223c | 2007-04-10 21:04:22 -0700 | [diff] [blame] | 155 | !(tcp_flag_word(tcp_hdr(skb)) & TCP_REMNANT))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 156 | /* Subtract also invariant (if peer is RFC compliant), |
| 157 | * tcp header plus fixed timestamp option length. |
| 158 | * Resulting "len" is MSS free of SACK jitter. |
| 159 | */ |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 160 | len -= tcp_sk(sk)->tcp_header_len; |
| 161 | icsk->icsk_ack.last_seg_size = len; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 162 | if (len == lss) { |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 163 | icsk->icsk_ack.rcv_mss = len; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 164 | return; |
| 165 | } |
| 166 | } |
Alexey Kuznetsov | 1ef9696 | 2006-09-19 12:52:50 -0700 | [diff] [blame] | 167 | if (icsk->icsk_ack.pending & ICSK_ACK_PUSHED) |
| 168 | icsk->icsk_ack.pending |= ICSK_ACK_PUSHED2; |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 169 | icsk->icsk_ack.pending |= ICSK_ACK_PUSHED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 170 | } |
| 171 | } |
| 172 | |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 173 | static void tcp_incr_quickack(struct sock *sk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 174 | { |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 175 | struct inet_connection_sock *icsk = inet_csk(sk); |
Eric Dumazet | 95c9617 | 2012-04-15 05:58:06 +0000 | [diff] [blame] | 176 | unsigned int quickacks = tcp_sk(sk)->rcv_wnd / (2 * icsk->icsk_ack.rcv_mss); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 177 | |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 178 | if (quickacks == 0) |
| 179 | quickacks = 2; |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 180 | if (quickacks > icsk->icsk_ack.quick) |
| 181 | icsk->icsk_ack.quick = min(quickacks, TCP_MAX_QUICKACKS); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 182 | } |
| 183 | |
stephen hemminger | 1b9f409 | 2010-09-28 19:30:14 +0000 | [diff] [blame] | 184 | static void tcp_enter_quickack_mode(struct sock *sk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 185 | { |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 186 | struct inet_connection_sock *icsk = inet_csk(sk); |
| 187 | tcp_incr_quickack(sk); |
| 188 | icsk->icsk_ack.pingpong = 0; |
| 189 | icsk->icsk_ack.ato = TCP_ATO_MIN; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 190 | } |
| 191 | |
| 192 | /* Send ACKs quickly, if "quick" count is not exhausted |
| 193 | * and the session is not interactive. |
| 194 | */ |
| 195 | |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 196 | static inline bool tcp_in_quickack_mode(const struct sock *sk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 197 | { |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 198 | const struct inet_connection_sock *icsk = inet_csk(sk); |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 199 | |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 200 | return icsk->icsk_ack.quick && !icsk->icsk_ack.pingpong; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 201 | } |
| 202 | |
Ilpo Järvinen | bdf1ee5 | 2007-05-27 02:04:16 -0700 | [diff] [blame] | 203 | static inline void TCP_ECN_queue_cwr(struct tcp_sock *tp) |
| 204 | { |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 205 | if (tp->ecn_flags & TCP_ECN_OK) |
Ilpo Järvinen | bdf1ee5 | 2007-05-27 02:04:16 -0700 | [diff] [blame] | 206 | tp->ecn_flags |= TCP_ECN_QUEUE_CWR; |
| 207 | } |
| 208 | |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 209 | static inline void TCP_ECN_accept_cwr(struct tcp_sock *tp, const struct sk_buff *skb) |
Ilpo Järvinen | bdf1ee5 | 2007-05-27 02:04:16 -0700 | [diff] [blame] | 210 | { |
| 211 | if (tcp_hdr(skb)->cwr) |
| 212 | tp->ecn_flags &= ~TCP_ECN_DEMAND_CWR; |
| 213 | } |
| 214 | |
| 215 | static inline void TCP_ECN_withdraw_cwr(struct tcp_sock *tp) |
| 216 | { |
| 217 | tp->ecn_flags &= ~TCP_ECN_DEMAND_CWR; |
| 218 | } |
| 219 | |
Eric Dumazet | 7a269ff | 2011-09-22 20:02:19 +0000 | [diff] [blame] | 220 | static inline void TCP_ECN_check_ce(struct tcp_sock *tp, const struct sk_buff *skb) |
Ilpo Järvinen | bdf1ee5 | 2007-05-27 02:04:16 -0700 | [diff] [blame] | 221 | { |
Eric Dumazet | 7a269ff | 2011-09-22 20:02:19 +0000 | [diff] [blame] | 222 | if (!(tp->ecn_flags & TCP_ECN_OK)) |
| 223 | return; |
| 224 | |
Eric Dumazet | b82d1bb | 2011-09-27 02:20:08 -0400 | [diff] [blame] | 225 | switch (TCP_SKB_CB(skb)->ip_dsfield & INET_ECN_MASK) { |
Eric Dumazet | 7a269ff | 2011-09-22 20:02:19 +0000 | [diff] [blame] | 226 | case INET_ECN_NOT_ECT: |
Ilpo Järvinen | bdf1ee5 | 2007-05-27 02:04:16 -0700 | [diff] [blame] | 227 | /* Funny extension: if ECT is not set on a segment, |
Eric Dumazet | 7a269ff | 2011-09-22 20:02:19 +0000 | [diff] [blame] | 228 | * and we already seen ECT on a previous segment, |
| 229 | * it is probably a retransmit. |
| 230 | */ |
| 231 | if (tp->ecn_flags & TCP_ECN_SEEN) |
Ilpo Järvinen | bdf1ee5 | 2007-05-27 02:04:16 -0700 | [diff] [blame] | 232 | tcp_enter_quickack_mode((struct sock *)tp); |
Eric Dumazet | 7a269ff | 2011-09-22 20:02:19 +0000 | [diff] [blame] | 233 | break; |
| 234 | case INET_ECN_CE: |
Eric Dumazet | aae06bf | 2012-08-06 11:04:43 +0000 | [diff] [blame] | 235 | if (!(tp->ecn_flags & TCP_ECN_DEMAND_CWR)) { |
| 236 | /* Better not delay acks, sender can have a very low cwnd */ |
| 237 | tcp_enter_quickack_mode((struct sock *)tp); |
| 238 | tp->ecn_flags |= TCP_ECN_DEMAND_CWR; |
| 239 | } |
Eric Dumazet | 7a269ff | 2011-09-22 20:02:19 +0000 | [diff] [blame] | 240 | /* fallinto */ |
| 241 | default: |
| 242 | tp->ecn_flags |= TCP_ECN_SEEN; |
Ilpo Järvinen | bdf1ee5 | 2007-05-27 02:04:16 -0700 | [diff] [blame] | 243 | } |
| 244 | } |
| 245 | |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 246 | static inline void TCP_ECN_rcv_synack(struct tcp_sock *tp, const struct tcphdr *th) |
Ilpo Järvinen | bdf1ee5 | 2007-05-27 02:04:16 -0700 | [diff] [blame] | 247 | { |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 248 | if ((tp->ecn_flags & TCP_ECN_OK) && (!th->ece || th->cwr)) |
Ilpo Järvinen | bdf1ee5 | 2007-05-27 02:04:16 -0700 | [diff] [blame] | 249 | tp->ecn_flags &= ~TCP_ECN_OK; |
| 250 | } |
| 251 | |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 252 | static inline void TCP_ECN_rcv_syn(struct tcp_sock *tp, const struct tcphdr *th) |
Ilpo Järvinen | bdf1ee5 | 2007-05-27 02:04:16 -0700 | [diff] [blame] | 253 | { |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 254 | if ((tp->ecn_flags & TCP_ECN_OK) && (!th->ece || !th->cwr)) |
Ilpo Järvinen | bdf1ee5 | 2007-05-27 02:04:16 -0700 | [diff] [blame] | 255 | tp->ecn_flags &= ~TCP_ECN_OK; |
| 256 | } |
| 257 | |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 258 | static bool TCP_ECN_rcv_ecn_echo(const struct tcp_sock *tp, const struct tcphdr *th) |
Ilpo Järvinen | bdf1ee5 | 2007-05-27 02:04:16 -0700 | [diff] [blame] | 259 | { |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 260 | if (th->ece && !th->syn && (tp->ecn_flags & TCP_ECN_OK)) |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 261 | return true; |
| 262 | return false; |
Ilpo Järvinen | bdf1ee5 | 2007-05-27 02:04:16 -0700 | [diff] [blame] | 263 | } |
| 264 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 265 | /* Buffer size and advertised window tuning. |
| 266 | * |
| 267 | * 1. Tuning sk->sk_sndbuf, when connection enters established state. |
| 268 | */ |
| 269 | |
Eric Dumazet | 6ae7053 | 2013-10-01 10:23:44 -0700 | [diff] [blame] | 270 | static void tcp_sndbuf_expand(struct sock *sk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 271 | { |
Eric Dumazet | 6ae7053 | 2013-10-01 10:23:44 -0700 | [diff] [blame] | 272 | const struct tcp_sock *tp = tcp_sk(sk); |
| 273 | int sndmem, per_mss; |
| 274 | u32 nr_segs; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 275 | |
Eric Dumazet | 6ae7053 | 2013-10-01 10:23:44 -0700 | [diff] [blame] | 276 | /* Worst case is non GSO/TSO : each frame consumes one skb |
| 277 | * and skb->head is kmalloced using power of two area of memory |
| 278 | */ |
| 279 | per_mss = max_t(u32, tp->rx_opt.mss_clamp, tp->mss_cache) + |
| 280 | MAX_TCP_HEADER + |
| 281 | SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); |
| 282 | |
| 283 | per_mss = roundup_pow_of_two(per_mss) + |
| 284 | SKB_DATA_ALIGN(sizeof(struct sk_buff)); |
| 285 | |
| 286 | nr_segs = max_t(u32, TCP_INIT_CWND, tp->snd_cwnd); |
| 287 | nr_segs = max_t(u32, nr_segs, tp->reordering + 1); |
| 288 | |
| 289 | /* Fast Recovery (RFC 5681 3.2) : |
| 290 | * Cubic needs 1.7 factor, rounded to 2 to include |
| 291 | * extra cushion (application might react slowly to POLLOUT) |
| 292 | */ |
| 293 | sndmem = 2 * nr_segs * per_mss; |
| 294 | |
Eric Dumazet | 06a59ec | 2011-10-13 18:24:42 +0000 | [diff] [blame] | 295 | if (sk->sk_sndbuf < sndmem) |
| 296 | sk->sk_sndbuf = min(sndmem, sysctl_tcp_wmem[2]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 297 | } |
| 298 | |
| 299 | /* 2. Tuning advertised window (window_clamp, rcv_ssthresh) |
| 300 | * |
| 301 | * All tcp_full_space() is split to two parts: "network" buffer, allocated |
| 302 | * forward and advertised in receiver window (tp->rcv_wnd) and |
| 303 | * "application buffer", required to isolate scheduling/application |
| 304 | * latencies from network. |
| 305 | * window_clamp is maximal advertised window. It can be less than |
| 306 | * tcp_full_space(), in this case tcp_full_space() - window_clamp |
| 307 | * is reserved for "application" buffer. The less window_clamp is |
| 308 | * the smoother our behaviour from viewpoint of network, but the lower |
| 309 | * throughput and the higher sensitivity of the connection to losses. 8) |
| 310 | * |
| 311 | * rcv_ssthresh is more strict window_clamp used at "slow start" |
| 312 | * phase to predict further behaviour of this connection. |
| 313 | * It is used for two goals: |
| 314 | * - to enforce header prediction at sender, even when application |
| 315 | * requires some significant "application buffer". It is check #1. |
| 316 | * - to prevent pruning of receive queue because of misprediction |
| 317 | * of receiver window. Check #2. |
| 318 | * |
| 319 | * The scheme does not work when sender sends good segments opening |
Stephen Hemminger | caa20d9a | 2005-11-10 17:13:47 -0800 | [diff] [blame] | 320 | * window and then starts to feed us spaghetti. But it should work |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 321 | * in common situations. Otherwise, we have to rely on queue collapsing. |
| 322 | */ |
| 323 | |
| 324 | /* Slow part of check#2. */ |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 325 | static int __tcp_grow_window(const struct sock *sk, const struct sk_buff *skb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 326 | { |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 327 | struct tcp_sock *tp = tcp_sk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 328 | /* Optimize this! */ |
Eric Dumazet | dfd4f0a | 2007-12-21 06:07:53 -0800 | [diff] [blame] | 329 | int truesize = tcp_win_from_space(skb->truesize) >> 1; |
| 330 | int window = tcp_win_from_space(sysctl_tcp_rmem[2]) >> 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 331 | |
| 332 | while (tp->rcv_ssthresh <= window) { |
| 333 | if (truesize <= skb->len) |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 334 | return 2 * inet_csk(sk)->icsk_ack.rcv_mss; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 335 | |
| 336 | truesize >>= 1; |
| 337 | window >>= 1; |
| 338 | } |
| 339 | return 0; |
| 340 | } |
| 341 | |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 342 | static void tcp_grow_window(struct sock *sk, const struct sk_buff *skb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 343 | { |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 344 | struct tcp_sock *tp = tcp_sk(sk); |
| 345 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 346 | /* Check #1 */ |
| 347 | if (tp->rcv_ssthresh < tp->window_clamp && |
| 348 | (int)tp->rcv_ssthresh < tcp_space(sk) && |
Glauber Costa | 180d8cd | 2011-12-11 21:47:02 +0000 | [diff] [blame] | 349 | !sk_under_memory_pressure(sk)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 350 | int incr; |
| 351 | |
| 352 | /* Check #2. Increase window, if skb with such overhead |
| 353 | * will fit to rcvbuf in future. |
| 354 | */ |
| 355 | if (tcp_win_from_space(skb->truesize) <= skb->len) |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 356 | incr = 2 * tp->advmss; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 357 | else |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 358 | incr = __tcp_grow_window(sk, skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 359 | |
| 360 | if (incr) { |
Eric Dumazet | 4d846f0 | 2012-04-16 23:28:07 +0000 | [diff] [blame] | 361 | incr = max_t(int, incr, 2 * skb->len); |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 362 | tp->rcv_ssthresh = min(tp->rcv_ssthresh + incr, |
| 363 | tp->window_clamp); |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 364 | inet_csk(sk)->icsk_ack.quick |= 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 365 | } |
| 366 | } |
| 367 | } |
| 368 | |
| 369 | /* 3. Tuning rcvbuf, when connection enters established state. */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 370 | static void tcp_fixup_rcvbuf(struct sock *sk) |
| 371 | { |
Eric Dumazet | e9266a0 | 2011-10-20 16:53:56 -0400 | [diff] [blame] | 372 | u32 mss = tcp_sk(sk)->advmss; |
Eric Dumazet | e9266a0 | 2011-10-20 16:53:56 -0400 | [diff] [blame] | 373 | int rcvmem; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 374 | |
Yuchung Cheng | 85f1652 | 2013-06-11 15:35:32 -0700 | [diff] [blame] | 375 | rcvmem = 2 * SKB_TRUESIZE(mss + MAX_TCP_HEADER) * |
| 376 | tcp_default_init_rwnd(mss); |
Eric Dumazet | e9266a0 | 2011-10-20 16:53:56 -0400 | [diff] [blame] | 377 | |
Eric Dumazet | b0983d3 | 2013-09-20 13:56:58 -0700 | [diff] [blame] | 378 | /* Dynamic Right Sizing (DRS) has 2 to 3 RTT latency |
| 379 | * Allow enough cushion so that sender is not limited by our window |
| 380 | */ |
| 381 | if (sysctl_tcp_moderate_rcvbuf) |
| 382 | rcvmem <<= 2; |
| 383 | |
Eric Dumazet | e9266a0 | 2011-10-20 16:53:56 -0400 | [diff] [blame] | 384 | if (sk->sk_rcvbuf < rcvmem) |
| 385 | sk->sk_rcvbuf = min(rcvmem, sysctl_tcp_rmem[2]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 386 | } |
| 387 | |
Stephen Hemminger | caa20d9a | 2005-11-10 17:13:47 -0800 | [diff] [blame] | 388 | /* 4. Try to fixup all. It is made immediately after connection enters |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 389 | * established state. |
| 390 | */ |
Jerry Chu | 1046716 | 2012-08-31 12:29:11 +0000 | [diff] [blame] | 391 | void tcp_init_buffer_space(struct sock *sk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 392 | { |
| 393 | struct tcp_sock *tp = tcp_sk(sk); |
| 394 | int maxwin; |
| 395 | |
| 396 | if (!(sk->sk_userlocks & SOCK_RCVBUF_LOCK)) |
| 397 | tcp_fixup_rcvbuf(sk); |
| 398 | if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) |
Eric Dumazet | 6ae7053 | 2013-10-01 10:23:44 -0700 | [diff] [blame] | 399 | tcp_sndbuf_expand(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 400 | |
| 401 | tp->rcvq_space.space = tp->rcv_wnd; |
Eric Dumazet | b0983d3 | 2013-09-20 13:56:58 -0700 | [diff] [blame] | 402 | tp->rcvq_space.time = tcp_time_stamp; |
| 403 | tp->rcvq_space.seq = tp->copied_seq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 404 | |
| 405 | maxwin = tcp_full_space(sk); |
| 406 | |
| 407 | if (tp->window_clamp >= maxwin) { |
| 408 | tp->window_clamp = maxwin; |
| 409 | |
| 410 | if (sysctl_tcp_app_win && maxwin > 4 * tp->advmss) |
| 411 | tp->window_clamp = max(maxwin - |
| 412 | (maxwin >> sysctl_tcp_app_win), |
| 413 | 4 * tp->advmss); |
| 414 | } |
| 415 | |
| 416 | /* Force reservation of one segment. */ |
| 417 | if (sysctl_tcp_app_win && |
| 418 | tp->window_clamp > 2 * tp->advmss && |
| 419 | tp->window_clamp + tp->advmss > maxwin) |
| 420 | tp->window_clamp = max(2 * tp->advmss, maxwin - tp->advmss); |
| 421 | |
| 422 | tp->rcv_ssthresh = min(tp->rcv_ssthresh, tp->window_clamp); |
| 423 | tp->snd_cwnd_stamp = tcp_time_stamp; |
| 424 | } |
| 425 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 426 | /* 5. Recalculate window clamp after socket hit its memory bounds. */ |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 427 | static void tcp_clamp_window(struct sock *sk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 428 | { |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 429 | struct tcp_sock *tp = tcp_sk(sk); |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 430 | struct inet_connection_sock *icsk = inet_csk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 431 | |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 432 | icsk->icsk_ack.quick = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 433 | |
John Heffner | 326f36e | 2005-11-10 17:11:48 -0800 | [diff] [blame] | 434 | if (sk->sk_rcvbuf < sysctl_tcp_rmem[2] && |
| 435 | !(sk->sk_userlocks & SOCK_RCVBUF_LOCK) && |
Glauber Costa | 180d8cd | 2011-12-11 21:47:02 +0000 | [diff] [blame] | 436 | !sk_under_memory_pressure(sk) && |
| 437 | sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)) { |
John Heffner | 326f36e | 2005-11-10 17:11:48 -0800 | [diff] [blame] | 438 | sk->sk_rcvbuf = min(atomic_read(&sk->sk_rmem_alloc), |
| 439 | sysctl_tcp_rmem[2]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 440 | } |
John Heffner | 326f36e | 2005-11-10 17:11:48 -0800 | [diff] [blame] | 441 | if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf) |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 442 | tp->rcv_ssthresh = min(tp->window_clamp, 2U * tp->advmss); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 443 | } |
| 444 | |
Stephen Hemminger | 40efc6f | 2006-01-03 16:03:49 -0800 | [diff] [blame] | 445 | /* Initialize RCV_MSS value. |
| 446 | * RCV_MSS is an our guess about MSS used by the peer. |
| 447 | * We haven't any direct information about the MSS. |
| 448 | * It's better to underestimate the RCV_MSS rather than overestimate. |
| 449 | * Overestimations make us ACKing less frequently than needed. |
| 450 | * Underestimations are more easy to detect and fix by tcp_measure_rcv_mss(). |
| 451 | */ |
| 452 | void tcp_initialize_rcv_mss(struct sock *sk) |
| 453 | { |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 454 | const struct tcp_sock *tp = tcp_sk(sk); |
Stephen Hemminger | 40efc6f | 2006-01-03 16:03:49 -0800 | [diff] [blame] | 455 | unsigned int hint = min_t(unsigned int, tp->advmss, tp->mss_cache); |
| 456 | |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 457 | hint = min(hint, tp->rcv_wnd / 2); |
William Allen Simpson | bee7ca9 | 2009-11-10 09:51:18 +0000 | [diff] [blame] | 458 | hint = min(hint, TCP_MSS_DEFAULT); |
Stephen Hemminger | 40efc6f | 2006-01-03 16:03:49 -0800 | [diff] [blame] | 459 | hint = max(hint, TCP_MIN_MSS); |
| 460 | |
| 461 | inet_csk(sk)->icsk_ack.rcv_mss = hint; |
| 462 | } |
Eric Dumazet | 4bc2f18 | 2010-07-09 21:22:10 +0000 | [diff] [blame] | 463 | EXPORT_SYMBOL(tcp_initialize_rcv_mss); |
Stephen Hemminger | 40efc6f | 2006-01-03 16:03:49 -0800 | [diff] [blame] | 464 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 465 | /* Receiver "autotuning" code. |
| 466 | * |
| 467 | * The algorithm for RTT estimation w/o timestamps is based on |
| 468 | * Dynamic Right-Sizing (DRS) by Wu Feng and Mike Fisk of LANL. |
Justin P. Mattock | 631dd1a | 2010-10-18 11:03:14 +0200 | [diff] [blame] | 469 | * <http://public.lanl.gov/radiant/pubs.html#DRS> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 470 | * |
| 471 | * More detail on this code can be found at |
Justin P. Mattock | 631dd1a | 2010-10-18 11:03:14 +0200 | [diff] [blame] | 472 | * <http://staff.psc.edu/jheffner/>, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 473 | * though this reference is out of date. A new paper |
| 474 | * is pending. |
| 475 | */ |
| 476 | static void tcp_rcv_rtt_update(struct tcp_sock *tp, u32 sample, int win_dep) |
| 477 | { |
| 478 | u32 new_sample = tp->rcv_rtt_est.rtt; |
| 479 | long m = sample; |
| 480 | |
| 481 | if (m == 0) |
| 482 | m = 1; |
| 483 | |
| 484 | if (new_sample != 0) { |
| 485 | /* If we sample in larger samples in the non-timestamp |
| 486 | * case, we could grossly overestimate the RTT especially |
| 487 | * with chatty applications or bulk transfer apps which |
| 488 | * are stalled on filesystem I/O. |
| 489 | * |
| 490 | * Also, since we are only going for a minimum in the |
Stephen Hemminger | 31f3426 | 2005-11-15 15:17:10 -0800 | [diff] [blame] | 491 | * non-timestamp case, we do not smooth things out |
Stephen Hemminger | caa20d9a | 2005-11-10 17:13:47 -0800 | [diff] [blame] | 492 | * else with timestamps disabled convergence takes too |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 493 | * long. |
| 494 | */ |
| 495 | if (!win_dep) { |
| 496 | m -= (new_sample >> 3); |
| 497 | new_sample += m; |
Neal Cardwell | 18a223e | 2012-04-10 07:59:20 +0000 | [diff] [blame] | 498 | } else { |
| 499 | m <<= 3; |
| 500 | if (m < new_sample) |
| 501 | new_sample = m; |
| 502 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 503 | } else { |
Stephen Hemminger | caa20d9a | 2005-11-10 17:13:47 -0800 | [diff] [blame] | 504 | /* No previous measure. */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 505 | new_sample = m << 3; |
| 506 | } |
| 507 | |
| 508 | if (tp->rcv_rtt_est.rtt != new_sample) |
| 509 | tp->rcv_rtt_est.rtt = new_sample; |
| 510 | } |
| 511 | |
| 512 | static inline void tcp_rcv_rtt_measure(struct tcp_sock *tp) |
| 513 | { |
| 514 | if (tp->rcv_rtt_est.time == 0) |
| 515 | goto new_measure; |
| 516 | if (before(tp->rcv_nxt, tp->rcv_rtt_est.seq)) |
| 517 | return; |
Neal Cardwell | 651913c | 2012-04-27 11:29:37 -0400 | [diff] [blame] | 518 | tcp_rcv_rtt_update(tp, tcp_time_stamp - tp->rcv_rtt_est.time, 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 519 | |
| 520 | new_measure: |
| 521 | tp->rcv_rtt_est.seq = tp->rcv_nxt + tp->rcv_wnd; |
| 522 | tp->rcv_rtt_est.time = tcp_time_stamp; |
| 523 | } |
| 524 | |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 525 | static inline void tcp_rcv_rtt_measure_ts(struct sock *sk, |
| 526 | const struct sk_buff *skb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 527 | { |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 528 | struct tcp_sock *tp = tcp_sk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 529 | if (tp->rx_opt.rcv_tsecr && |
| 530 | (TCP_SKB_CB(skb)->end_seq - |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 531 | TCP_SKB_CB(skb)->seq >= inet_csk(sk)->icsk_ack.rcv_mss)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 532 | tcp_rcv_rtt_update(tp, tcp_time_stamp - tp->rx_opt.rcv_tsecr, 0); |
| 533 | } |
| 534 | |
| 535 | /* |
| 536 | * This function should be called every time data is copied to user space. |
| 537 | * It calculates the appropriate TCP receive buffer space. |
| 538 | */ |
| 539 | void tcp_rcv_space_adjust(struct sock *sk) |
| 540 | { |
| 541 | struct tcp_sock *tp = tcp_sk(sk); |
| 542 | int time; |
Eric Dumazet | b0983d3 | 2013-09-20 13:56:58 -0700 | [diff] [blame] | 543 | int copied; |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 544 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 545 | time = tcp_time_stamp - tp->rcvq_space.time; |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 546 | if (time < (tp->rcv_rtt_est.rtt >> 3) || tp->rcv_rtt_est.rtt == 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 547 | return; |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 548 | |
Eric Dumazet | b0983d3 | 2013-09-20 13:56:58 -0700 | [diff] [blame] | 549 | /* Number of bytes copied to user in last RTT */ |
| 550 | copied = tp->copied_seq - tp->rcvq_space.seq; |
| 551 | if (copied <= tp->rcvq_space.space) |
| 552 | goto new_measure; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 553 | |
Eric Dumazet | b0983d3 | 2013-09-20 13:56:58 -0700 | [diff] [blame] | 554 | /* A bit of theory : |
| 555 | * copied = bytes received in previous RTT, our base window |
| 556 | * To cope with packet losses, we need a 2x factor |
| 557 | * To cope with slow start, and sender growing its cwin by 100 % |
| 558 | * every RTT, we need a 4x factor, because the ACK we are sending |
| 559 | * now is for the next RTT, not the current one : |
| 560 | * <prev RTT . ><current RTT .. ><next RTT .... > |
| 561 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 562 | |
Eric Dumazet | b0983d3 | 2013-09-20 13:56:58 -0700 | [diff] [blame] | 563 | if (sysctl_tcp_moderate_rcvbuf && |
| 564 | !(sk->sk_userlocks & SOCK_RCVBUF_LOCK)) { |
| 565 | int rcvwin, rcvmem, rcvbuf; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 566 | |
Eric Dumazet | b0983d3 | 2013-09-20 13:56:58 -0700 | [diff] [blame] | 567 | /* minimal window to cope with packet losses, assuming |
| 568 | * steady state. Add some cushion because of small variations. |
| 569 | */ |
| 570 | rcvwin = (copied << 1) + 16 * tp->advmss; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 571 | |
Eric Dumazet | b0983d3 | 2013-09-20 13:56:58 -0700 | [diff] [blame] | 572 | /* If rate increased by 25%, |
| 573 | * assume slow start, rcvwin = 3 * copied |
| 574 | * If rate increased by 50%, |
| 575 | * assume sender can use 2x growth, rcvwin = 4 * copied |
| 576 | */ |
| 577 | if (copied >= |
| 578 | tp->rcvq_space.space + (tp->rcvq_space.space >> 2)) { |
| 579 | if (copied >= |
| 580 | tp->rcvq_space.space + (tp->rcvq_space.space >> 1)) |
| 581 | rcvwin <<= 1; |
| 582 | else |
| 583 | rcvwin += (rcvwin >> 1); |
| 584 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 585 | |
Eric Dumazet | b0983d3 | 2013-09-20 13:56:58 -0700 | [diff] [blame] | 586 | rcvmem = SKB_TRUESIZE(tp->advmss + MAX_TCP_HEADER); |
| 587 | while (tcp_win_from_space(rcvmem) < tp->advmss) |
| 588 | rcvmem += 128; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 589 | |
Eric Dumazet | b0983d3 | 2013-09-20 13:56:58 -0700 | [diff] [blame] | 590 | rcvbuf = min(rcvwin / tp->advmss * rcvmem, sysctl_tcp_rmem[2]); |
| 591 | if (rcvbuf > sk->sk_rcvbuf) { |
| 592 | sk->sk_rcvbuf = rcvbuf; |
| 593 | |
| 594 | /* Make the window clamp follow along. */ |
| 595 | tp->window_clamp = rcvwin; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 596 | } |
| 597 | } |
Eric Dumazet | b0983d3 | 2013-09-20 13:56:58 -0700 | [diff] [blame] | 598 | tp->rcvq_space.space = copied; |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 599 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 600 | new_measure: |
| 601 | tp->rcvq_space.seq = tp->copied_seq; |
| 602 | tp->rcvq_space.time = tcp_time_stamp; |
| 603 | } |
| 604 | |
| 605 | /* There is something which you must keep in mind when you analyze the |
| 606 | * behavior of the tp->ato delayed ack timeout interval. When a |
| 607 | * connection starts up, we want to ack as quickly as possible. The |
| 608 | * problem is that "good" TCP's do slow start at the beginning of data |
| 609 | * transmission. The means that until we send the first few ACK's the |
| 610 | * sender will sit on his end and only queue most of his data, because |
| 611 | * he can only send snd_cwnd unacked packets at any given time. For |
| 612 | * each ACK we send, he increments snd_cwnd and transmits more of his |
| 613 | * queue. -DaveM |
| 614 | */ |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 615 | static void tcp_event_data_recv(struct sock *sk, struct sk_buff *skb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 616 | { |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 617 | struct tcp_sock *tp = tcp_sk(sk); |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 618 | struct inet_connection_sock *icsk = inet_csk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 619 | u32 now; |
| 620 | |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 621 | inet_csk_schedule_ack(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 622 | |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 623 | tcp_measure_rcv_mss(sk, skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 624 | |
| 625 | tcp_rcv_rtt_measure(tp); |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 626 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 627 | now = tcp_time_stamp; |
| 628 | |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 629 | if (!icsk->icsk_ack.ato) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 630 | /* The _first_ data packet received, initialize |
| 631 | * delayed ACK engine. |
| 632 | */ |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 633 | tcp_incr_quickack(sk); |
| 634 | icsk->icsk_ack.ato = TCP_ATO_MIN; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 635 | } else { |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 636 | int m = now - icsk->icsk_ack.lrcvtime; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 637 | |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 638 | if (m <= TCP_ATO_MIN / 2) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 639 | /* The fastest case is the first. */ |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 640 | icsk->icsk_ack.ato = (icsk->icsk_ack.ato >> 1) + TCP_ATO_MIN / 2; |
| 641 | } else if (m < icsk->icsk_ack.ato) { |
| 642 | icsk->icsk_ack.ato = (icsk->icsk_ack.ato >> 1) + m; |
| 643 | if (icsk->icsk_ack.ato > icsk->icsk_rto) |
| 644 | icsk->icsk_ack.ato = icsk->icsk_rto; |
| 645 | } else if (m > icsk->icsk_rto) { |
Stephen Hemminger | caa20d9a | 2005-11-10 17:13:47 -0800 | [diff] [blame] | 646 | /* Too long gap. Apparently sender failed to |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 647 | * restart window, so that we send ACKs quickly. |
| 648 | */ |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 649 | tcp_incr_quickack(sk); |
Hideo Aoki | 3ab224b | 2007-12-31 00:11:19 -0800 | [diff] [blame] | 650 | sk_mem_reclaim(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 651 | } |
| 652 | } |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 653 | icsk->icsk_ack.lrcvtime = now; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 654 | |
| 655 | TCP_ECN_check_ce(tp, skb); |
| 656 | |
| 657 | if (skb->len >= 128) |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 658 | tcp_grow_window(sk, skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 659 | } |
| 660 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 661 | /* Called to compute a smoothed rtt estimate. The data fed to this |
| 662 | * routine either comes from timestamps, or from segments that were |
| 663 | * known _not_ to have been retransmitted [see Karn/Partridge |
| 664 | * Proceedings SIGCOMM 87]. The algorithm is from the SIGCOMM 88 |
| 665 | * piece by Van Jacobson. |
| 666 | * NOTE: the next three routines used to be one big routine. |
| 667 | * To save cycles in the RFC 1323 implementation it was better to break |
| 668 | * it up into three procedures. -- erics |
| 669 | */ |
Stephen Hemminger | 2d2abba | 2005-11-10 16:56:12 -0800 | [diff] [blame] | 670 | static void tcp_rtt_estimator(struct sock *sk, const __u32 mrtt) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 671 | { |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 672 | struct tcp_sock *tp = tcp_sk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 673 | long m = mrtt; /* RTT */ |
| 674 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 675 | /* The following amusing code comes from Jacobson's |
| 676 | * article in SIGCOMM '88. Note that rtt and mdev |
| 677 | * are scaled versions of rtt and mean deviation. |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 678 | * This is designed to be as fast as possible |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 679 | * m stands for "measurement". |
| 680 | * |
| 681 | * On a 1990 paper the rto value is changed to: |
| 682 | * RTO = rtt + 4 * mdev |
| 683 | * |
| 684 | * Funny. This algorithm seems to be very broken. |
| 685 | * These formulae increase RTO, when it should be decreased, increase |
Stephen Hemminger | 31f3426 | 2005-11-15 15:17:10 -0800 | [diff] [blame] | 686 | * too slowly, when it should be increased quickly, decrease too quickly |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 687 | * etc. I guess in BSD RTO takes ONE value, so that it is absolutely |
| 688 | * does not matter how to _calculate_ it. Seems, it was trap |
| 689 | * that VJ failed to avoid. 8) |
| 690 | */ |
Stephen Hemminger | 2de979b | 2007-03-08 20:45:19 -0800 | [diff] [blame] | 691 | if (m == 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 692 | m = 1; |
| 693 | if (tp->srtt != 0) { |
| 694 | m -= (tp->srtt >> 3); /* m is now error in rtt est */ |
| 695 | tp->srtt += m; /* rtt = 7/8 rtt + 1/8 new */ |
| 696 | if (m < 0) { |
| 697 | m = -m; /* m is now abs(error) */ |
| 698 | m -= (tp->mdev >> 2); /* similar update on mdev */ |
| 699 | /* This is similar to one of Eifel findings. |
| 700 | * Eifel blocks mdev updates when rtt decreases. |
| 701 | * This solution is a bit different: we use finer gain |
| 702 | * for mdev in this case (alpha*beta). |
| 703 | * Like Eifel it also prevents growth of rto, |
| 704 | * but also it limits too fast rto decreases, |
| 705 | * happening in pure Eifel. |
| 706 | */ |
| 707 | if (m > 0) |
| 708 | m >>= 3; |
| 709 | } else { |
| 710 | m -= (tp->mdev >> 2); /* similar update on mdev */ |
| 711 | } |
| 712 | tp->mdev += m; /* mdev = 3/4 mdev + 1/4 new */ |
| 713 | if (tp->mdev > tp->mdev_max) { |
| 714 | tp->mdev_max = tp->mdev; |
| 715 | if (tp->mdev_max > tp->rttvar) |
| 716 | tp->rttvar = tp->mdev_max; |
| 717 | } |
| 718 | if (after(tp->snd_una, tp->rtt_seq)) { |
| 719 | if (tp->mdev_max < tp->rttvar) |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 720 | tp->rttvar -= (tp->rttvar - tp->mdev_max) >> 2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 721 | tp->rtt_seq = tp->snd_nxt; |
David S. Miller | 05bb1fa | 2007-08-30 22:10:28 -0700 | [diff] [blame] | 722 | tp->mdev_max = tcp_rto_min(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 723 | } |
| 724 | } else { |
| 725 | /* no previous measure. */ |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 726 | tp->srtt = m << 3; /* take the measured time to be rtt */ |
| 727 | tp->mdev = m << 1; /* make sure rto = 3*rtt */ |
David S. Miller | 05bb1fa | 2007-08-30 22:10:28 -0700 | [diff] [blame] | 728 | tp->mdev_max = tp->rttvar = max(tp->mdev, tcp_rto_min(sk)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 729 | tp->rtt_seq = tp->snd_nxt; |
| 730 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 731 | } |
| 732 | |
Eric Dumazet | 95bd09e | 2013-08-27 05:46:32 -0700 | [diff] [blame] | 733 | /* Set the sk_pacing_rate to allow proper sizing of TSO packets. |
| 734 | * Note: TCP stack does not yet implement pacing. |
| 735 | * FQ packet scheduler can be used to implement cheap but effective |
| 736 | * TCP pacing, to smooth the burst on large writes when packets |
| 737 | * in flight is significantly lower than cwnd (or rwin) |
| 738 | */ |
| 739 | static void tcp_update_pacing_rate(struct sock *sk) |
| 740 | { |
| 741 | const struct tcp_sock *tp = tcp_sk(sk); |
| 742 | u64 rate; |
| 743 | |
| 744 | /* set sk_pacing_rate to 200 % of current rate (mss * cwnd / srtt) */ |
| 745 | rate = (u64)tp->mss_cache * 2 * (HZ << 3); |
| 746 | |
| 747 | rate *= max(tp->snd_cwnd, tp->packets_out); |
| 748 | |
| 749 | /* Correction for small srtt : minimum srtt being 8 (1 jiffy << 3), |
| 750 | * be conservative and assume srtt = 1 (125 us instead of 1.25 ms) |
| 751 | * We probably need usec resolution in the future. |
| 752 | * Note: This also takes care of possible srtt=0 case, |
| 753 | * when tcp_rtt_estimator() was not yet called. |
| 754 | */ |
| 755 | if (tp->srtt > 8 + 2) |
| 756 | do_div(rate, tp->srtt); |
| 757 | |
Eric Dumazet | ba53742 | 2013-10-09 17:14:52 -0700 | [diff] [blame] | 758 | /* ACCESS_ONCE() is needed because sch_fq fetches sk_pacing_rate |
| 759 | * without any lock. We want to make sure compiler wont store |
| 760 | * intermediate values in this location. |
| 761 | */ |
| 762 | ACCESS_ONCE(sk->sk_pacing_rate) = min_t(u64, rate, |
| 763 | sk->sk_max_pacing_rate); |
Eric Dumazet | 95bd09e | 2013-08-27 05:46:32 -0700 | [diff] [blame] | 764 | } |
| 765 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 766 | /* Calculate rto without backoff. This is the second half of Van Jacobson's |
| 767 | * routine referred to above. |
| 768 | */ |
David S. Miller | 4aabd8e | 2012-07-09 16:07:30 -0700 | [diff] [blame] | 769 | void tcp_set_rto(struct sock *sk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 770 | { |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 771 | const struct tcp_sock *tp = tcp_sk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 772 | /* Old crap is replaced with new one. 8) |
| 773 | * |
| 774 | * More seriously: |
| 775 | * 1. If rtt variance happened to be less 50msec, it is hallucination. |
| 776 | * It cannot be less due to utterly erratic ACK generation made |
| 777 | * at least by solaris and freebsd. "Erratic ACKs" has _nothing_ |
| 778 | * to do with delayed acks, because at cwnd>2 true delack timeout |
| 779 | * is invisible. Actually, Linux-2.4 also generates erratic |
Stephen Hemminger | caa20d9a | 2005-11-10 17:13:47 -0800 | [diff] [blame] | 780 | * ACKs in some circumstances. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 781 | */ |
Damian Lukowski | f1ecd5d | 2009-08-26 00:16:31 +0000 | [diff] [blame] | 782 | inet_csk(sk)->icsk_rto = __tcp_set_rto(tp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 783 | |
| 784 | /* 2. Fixups made earlier cannot be right. |
| 785 | * If we do not estimate RTO correctly without them, |
| 786 | * all the algo is pure shit and should be replaced |
Stephen Hemminger | caa20d9a | 2005-11-10 17:13:47 -0800 | [diff] [blame] | 787 | * with correct one. It is exactly, which we pretend to do. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 788 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 789 | |
Ilpo Järvinen | ee6aac5 | 2008-12-05 22:43:08 -0800 | [diff] [blame] | 790 | /* NOTE: clamping at TCP_RTO_MIN is not required, current algo |
| 791 | * guarantees that rto is higher. |
| 792 | */ |
Damian Lukowski | f1ecd5d | 2009-08-26 00:16:31 +0000 | [diff] [blame] | 793 | tcp_bound_rto(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 794 | } |
| 795 | |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 796 | __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 797 | { |
| 798 | __u32 cwnd = (dst ? dst_metric(dst, RTAX_INITCWND) : 0); |
| 799 | |
Gerrit Renker | 22b71c8 | 2010-08-29 19:23:12 +0000 | [diff] [blame] | 800 | if (!cwnd) |
David S. Miller | 442b963 | 2011-02-02 17:05:11 -0800 | [diff] [blame] | 801 | cwnd = TCP_INIT_CWND; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 802 | return min_t(__u32, cwnd, tp->snd_cwnd_clamp); |
| 803 | } |
| 804 | |
Ilpo Järvinen | e60402d | 2007-08-09 15:14:46 +0300 | [diff] [blame] | 805 | /* |
| 806 | * Packet counting of FACK is based on in-order assumptions, therefore TCP |
| 807 | * disables it when reordering is detected |
| 808 | */ |
David S. Miller | 4aabd8e | 2012-07-09 16:07:30 -0700 | [diff] [blame] | 809 | void tcp_disable_fack(struct tcp_sock *tp) |
Ilpo Järvinen | e60402d | 2007-08-09 15:14:46 +0300 | [diff] [blame] | 810 | { |
Ilpo Järvinen | 85cc391 | 2007-11-15 19:39:31 -0800 | [diff] [blame] | 811 | /* RFC3517 uses different metric in lost marker => reset on change */ |
| 812 | if (tcp_is_fack(tp)) |
| 813 | tp->lost_skb_hint = NULL; |
Vijay Subramanian | ab56222 | 2011-12-20 13:23:24 +0000 | [diff] [blame] | 814 | tp->rx_opt.sack_ok &= ~TCP_FACK_ENABLED; |
Ilpo Järvinen | e60402d | 2007-08-09 15:14:46 +0300 | [diff] [blame] | 815 | } |
| 816 | |
Ryousei Takano | 564262c1 | 2007-10-25 23:03:52 -0700 | [diff] [blame] | 817 | /* Take a notice that peer is sending D-SACKs */ |
Ilpo Järvinen | e60402d | 2007-08-09 15:14:46 +0300 | [diff] [blame] | 818 | static void tcp_dsack_seen(struct tcp_sock *tp) |
| 819 | { |
Vijay Subramanian | ab56222 | 2011-12-20 13:23:24 +0000 | [diff] [blame] | 820 | tp->rx_opt.sack_ok |= TCP_DSACK_SEEN; |
Ilpo Järvinen | e60402d | 2007-08-09 15:14:46 +0300 | [diff] [blame] | 821 | } |
| 822 | |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 823 | static void tcp_update_reordering(struct sock *sk, const int metric, |
| 824 | const int ts) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 825 | { |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 826 | struct tcp_sock *tp = tcp_sk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 827 | if (metric > tp->reordering) { |
Pavel Emelyanov | 40b215e | 2008-07-03 01:05:41 -0700 | [diff] [blame] | 828 | int mib_idx; |
| 829 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 830 | tp->reordering = min(TCP_MAX_REORDERING, metric); |
| 831 | |
| 832 | /* This exciting event is worth to be remembered. 8) */ |
| 833 | if (ts) |
Pavel Emelyanov | 40b215e | 2008-07-03 01:05:41 -0700 | [diff] [blame] | 834 | mib_idx = LINUX_MIB_TCPTSREORDER; |
Ilpo Järvinen | e60402d | 2007-08-09 15:14:46 +0300 | [diff] [blame] | 835 | else if (tcp_is_reno(tp)) |
Pavel Emelyanov | 40b215e | 2008-07-03 01:05:41 -0700 | [diff] [blame] | 836 | mib_idx = LINUX_MIB_TCPRENOREORDER; |
Ilpo Järvinen | e60402d | 2007-08-09 15:14:46 +0300 | [diff] [blame] | 837 | else if (tcp_is_fack(tp)) |
Pavel Emelyanov | 40b215e | 2008-07-03 01:05:41 -0700 | [diff] [blame] | 838 | mib_idx = LINUX_MIB_TCPFACKREORDER; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 839 | else |
Pavel Emelyanov | 40b215e | 2008-07-03 01:05:41 -0700 | [diff] [blame] | 840 | mib_idx = LINUX_MIB_TCPSACKREORDER; |
| 841 | |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 842 | NET_INC_STATS_BH(sock_net(sk), mib_idx); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 843 | #if FASTRETRANS_DEBUG > 1 |
Joe Perches | 91df42b | 2012-05-15 14:11:54 +0000 | [diff] [blame] | 844 | pr_debug("Disorder%d %d %u f%u s%u rr%d\n", |
| 845 | tp->rx_opt.sack_ok, inet_csk(sk)->icsk_ca_state, |
| 846 | tp->reordering, |
| 847 | tp->fackets_out, |
| 848 | tp->sacked_out, |
| 849 | tp->undo_marker ? tp->undo_retrans : 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 850 | #endif |
Ilpo Järvinen | e60402d | 2007-08-09 15:14:46 +0300 | [diff] [blame] | 851 | tcp_disable_fack(tp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 852 | } |
Yuchung Cheng | eed530b | 2012-05-02 13:30:03 +0000 | [diff] [blame] | 853 | |
| 854 | if (metric > 0) |
| 855 | tcp_disable_early_retrans(tp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 856 | } |
| 857 | |
Ilpo Järvinen | 006f582 | 2008-09-20 21:20:20 -0700 | [diff] [blame] | 858 | /* This must be called before lost_out is incremented */ |
Ilpo Järvinen | c8c213f | 2008-09-20 21:18:55 -0700 | [diff] [blame] | 859 | static void tcp_verify_retransmit_hint(struct tcp_sock *tp, struct sk_buff *skb) |
| 860 | { |
Ilpo Järvinen | 006f582 | 2008-09-20 21:20:20 -0700 | [diff] [blame] | 861 | if ((tp->retransmit_skb_hint == NULL) || |
Ilpo Järvinen | c8c213f | 2008-09-20 21:18:55 -0700 | [diff] [blame] | 862 | before(TCP_SKB_CB(skb)->seq, |
| 863 | TCP_SKB_CB(tp->retransmit_skb_hint)->seq)) |
Ilpo Järvinen | 006f582 | 2008-09-20 21:20:20 -0700 | [diff] [blame] | 864 | tp->retransmit_skb_hint = skb; |
| 865 | |
| 866 | if (!tp->lost_out || |
| 867 | after(TCP_SKB_CB(skb)->end_seq, tp->retransmit_high)) |
| 868 | tp->retransmit_high = TCP_SKB_CB(skb)->end_seq; |
Ilpo Järvinen | c8c213f | 2008-09-20 21:18:55 -0700 | [diff] [blame] | 869 | } |
| 870 | |
Ilpo Järvinen | 41ea36e | 2008-09-20 21:19:22 -0700 | [diff] [blame] | 871 | static void tcp_skb_mark_lost(struct tcp_sock *tp, struct sk_buff *skb) |
| 872 | { |
| 873 | if (!(TCP_SKB_CB(skb)->sacked & (TCPCB_LOST|TCPCB_SACKED_ACKED))) { |
| 874 | tcp_verify_retransmit_hint(tp, skb); |
| 875 | |
| 876 | tp->lost_out += tcp_skb_pcount(skb); |
| 877 | TCP_SKB_CB(skb)->sacked |= TCPCB_LOST; |
| 878 | } |
| 879 | } |
| 880 | |
Ilpo Järvinen | e1aa680 | 2008-11-24 21:11:55 -0800 | [diff] [blame] | 881 | static void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, |
| 882 | struct sk_buff *skb) |
Ilpo Järvinen | 006f582 | 2008-09-20 21:20:20 -0700 | [diff] [blame] | 883 | { |
| 884 | tcp_verify_retransmit_hint(tp, skb); |
| 885 | |
| 886 | if (!(TCP_SKB_CB(skb)->sacked & (TCPCB_LOST|TCPCB_SACKED_ACKED))) { |
| 887 | tp->lost_out += tcp_skb_pcount(skb); |
| 888 | TCP_SKB_CB(skb)->sacked |= TCPCB_LOST; |
| 889 | } |
| 890 | } |
| 891 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 892 | /* This procedure tags the retransmission queue when SACKs arrive. |
| 893 | * |
| 894 | * We have three tag bits: SACKED(S), RETRANS(R) and LOST(L). |
| 895 | * Packets in queue with these bits set are counted in variables |
| 896 | * sacked_out, retrans_out and lost_out, correspondingly. |
| 897 | * |
| 898 | * Valid combinations are: |
| 899 | * Tag InFlight Description |
| 900 | * 0 1 - orig segment is in flight. |
| 901 | * S 0 - nothing flies, orig reached receiver. |
| 902 | * L 0 - nothing flies, orig lost by net. |
| 903 | * R 2 - both orig and retransmit are in flight. |
| 904 | * L|R 1 - orig is lost, retransmit is in flight. |
| 905 | * S|R 1 - orig reached receiver, retrans is still in flight. |
| 906 | * (L|S|R is logically valid, it could occur when L|R is sacked, |
| 907 | * but it is equivalent to plain S and code short-curcuits it to S. |
| 908 | * L|S is logically invalid, it would mean -1 packet in flight 8)) |
| 909 | * |
| 910 | * These 6 states form finite state machine, controlled by the following events: |
| 911 | * 1. New ACK (+SACK) arrives. (tcp_sacktag_write_queue()) |
| 912 | * 2. Retransmission. (tcp_retransmit_skb(), tcp_xmit_retransmit_queue()) |
Yuchung Cheng | 974c123 | 2012-01-19 14:42:21 +0000 | [diff] [blame] | 913 | * 3. Loss detection event of two flavors: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 914 | * A. Scoreboard estimator decided the packet is lost. |
| 915 | * A'. Reno "three dupacks" marks head of queue lost. |
Yuchung Cheng | 974c123 | 2012-01-19 14:42:21 +0000 | [diff] [blame] | 916 | * A''. Its FACK modification, head until snd.fack is lost. |
| 917 | * B. SACK arrives sacking SND.NXT at the moment, when the |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 918 | * segment was retransmitted. |
| 919 | * 4. D-SACK added new rule: D-SACK changes any tag to S. |
| 920 | * |
| 921 | * It is pleasant to note, that state diagram turns out to be commutative, |
| 922 | * so that we are allowed not to be bothered by order of our actions, |
| 923 | * when multiple events arrive simultaneously. (see the function below). |
| 924 | * |
| 925 | * Reordering detection. |
| 926 | * -------------------- |
| 927 | * Reordering metric is maximal distance, which a packet can be displaced |
| 928 | * in packet stream. With SACKs we can estimate it: |
| 929 | * |
| 930 | * 1. SACK fills old hole and the corresponding segment was not |
| 931 | * ever retransmitted -> reordering. Alas, we cannot use it |
| 932 | * when segment was retransmitted. |
| 933 | * 2. The last flaw is solved with D-SACK. D-SACK arrives |
| 934 | * for retransmitted and already SACKed segment -> reordering.. |
| 935 | * Both of these heuristics are not used in Loss state, when we cannot |
| 936 | * account for retransmits accurately. |
Ilpo Järvinen | 5b3c988 | 2007-08-24 22:54:44 -0700 | [diff] [blame] | 937 | * |
| 938 | * SACK block validation. |
| 939 | * ---------------------- |
| 940 | * |
| 941 | * SACK block range validation checks that the received SACK block fits to |
| 942 | * the expected sequence limits, i.e., it is between SND.UNA and SND.NXT. |
| 943 | * Note that SND.UNA is not included to the range though being valid because |
Ilpo Järvinen | 0e83533 | 2007-10-01 15:28:17 -0700 | [diff] [blame] | 944 | * it means that the receiver is rather inconsistent with itself reporting |
| 945 | * SACK reneging when it should advance SND.UNA. Such SACK block this is |
| 946 | * perfectly valid, however, in light of RFC2018 which explicitly states |
| 947 | * that "SACK block MUST reflect the newest segment. Even if the newest |
| 948 | * segment is going to be discarded ...", not that it looks very clever |
| 949 | * in case of head skb. Due to potentional receiver driven attacks, we |
| 950 | * choose to avoid immediate execution of a walk in write queue due to |
| 951 | * reneging and defer head skb's loss recovery to standard loss recovery |
| 952 | * procedure that will eventually trigger (nothing forbids us doing this). |
Ilpo Järvinen | 5b3c988 | 2007-08-24 22:54:44 -0700 | [diff] [blame] | 953 | * |
| 954 | * Implements also blockage to start_seq wrap-around. Problem lies in the |
| 955 | * fact that though start_seq (s) is before end_seq (i.e., not reversed), |
| 956 | * there's no guarantee that it will be before snd_nxt (n). The problem |
| 957 | * happens when start_seq resides between end_seq wrap (e_w) and snd_nxt |
| 958 | * wrap (s_w): |
| 959 | * |
| 960 | * <- outs wnd -> <- wrapzone -> |
| 961 | * u e n u_w e_w s n_w |
| 962 | * | | | | | | | |
| 963 | * |<------------+------+----- TCP seqno space --------------+---------->| |
| 964 | * ...-- <2^31 ->| |<--------... |
| 965 | * ...---- >2^31 ------>| |<--------... |
| 966 | * |
| 967 | * Current code wouldn't be vulnerable but it's better still to discard such |
| 968 | * crazy SACK blocks. Doing this check for start_seq alone closes somewhat |
| 969 | * similar case (end_seq after snd_nxt wrap) as earlier reversed check in |
| 970 | * snd_nxt wrap -> snd_una region will then become "well defined", i.e., |
| 971 | * equal to the ideal case (infinite seqno space without wrap caused issues). |
| 972 | * |
| 973 | * With D-SACK the lower bound is extended to cover sequence space below |
| 974 | * SND.UNA down to undo_marker, which is the last point of interest. Yet |
Ryousei Takano | 564262c1 | 2007-10-25 23:03:52 -0700 | [diff] [blame] | 975 | * again, D-SACK block must not to go across snd_una (for the same reason as |
Ilpo Järvinen | 5b3c988 | 2007-08-24 22:54:44 -0700 | [diff] [blame] | 976 | * for the normal SACK blocks, explained above). But there all simplicity |
| 977 | * ends, TCP might receive valid D-SACKs below that. As long as they reside |
| 978 | * fully below undo_marker they do not affect behavior in anyway and can |
| 979 | * therefore be safely ignored. In rare cases (which are more or less |
| 980 | * theoretical ones), the D-SACK will nicely cross that boundary due to skb |
| 981 | * fragmentation and packet reordering past skb's retransmission. To consider |
| 982 | * them correctly, the acceptable range must be extended even more though |
| 983 | * the exact amount is rather hard to quantify. However, tp->max_window can |
| 984 | * be used as an exaggerated estimate. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 985 | */ |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 986 | static bool tcp_is_sackblock_valid(struct tcp_sock *tp, bool is_dsack, |
| 987 | u32 start_seq, u32 end_seq) |
Ilpo Järvinen | 5b3c988 | 2007-08-24 22:54:44 -0700 | [diff] [blame] | 988 | { |
| 989 | /* Too far in future, or reversed (interpretation is ambiguous) */ |
| 990 | if (after(end_seq, tp->snd_nxt) || !before(start_seq, end_seq)) |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 991 | return false; |
Ilpo Järvinen | 5b3c988 | 2007-08-24 22:54:44 -0700 | [diff] [blame] | 992 | |
| 993 | /* Nasty start_seq wrap-around check (see comments above) */ |
| 994 | if (!before(start_seq, tp->snd_nxt)) |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 995 | return false; |
Ilpo Järvinen | 5b3c988 | 2007-08-24 22:54:44 -0700 | [diff] [blame] | 996 | |
Ryousei Takano | 564262c1 | 2007-10-25 23:03:52 -0700 | [diff] [blame] | 997 | /* In outstanding window? ...This is valid exit for D-SACKs too. |
Ilpo Järvinen | 5b3c988 | 2007-08-24 22:54:44 -0700 | [diff] [blame] | 998 | * start_seq == snd_una is non-sensical (see comments above) |
| 999 | */ |
| 1000 | if (after(start_seq, tp->snd_una)) |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1001 | return true; |
Ilpo Järvinen | 5b3c988 | 2007-08-24 22:54:44 -0700 | [diff] [blame] | 1002 | |
| 1003 | if (!is_dsack || !tp->undo_marker) |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1004 | return false; |
Ilpo Järvinen | 5b3c988 | 2007-08-24 22:54:44 -0700 | [diff] [blame] | 1005 | |
| 1006 | /* ...Then it's D-SACK, and must reside below snd_una completely */ |
Zheng Yan | f779b2d | 2011-09-18 22:37:34 -0400 | [diff] [blame] | 1007 | if (after(end_seq, tp->snd_una)) |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1008 | return false; |
Ilpo Järvinen | 5b3c988 | 2007-08-24 22:54:44 -0700 | [diff] [blame] | 1009 | |
| 1010 | if (!before(start_seq, tp->undo_marker)) |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1011 | return true; |
Ilpo Järvinen | 5b3c988 | 2007-08-24 22:54:44 -0700 | [diff] [blame] | 1012 | |
| 1013 | /* Too old */ |
| 1014 | if (!after(end_seq, tp->undo_marker)) |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1015 | return false; |
Ilpo Järvinen | 5b3c988 | 2007-08-24 22:54:44 -0700 | [diff] [blame] | 1016 | |
| 1017 | /* Undo_marker boundary crossing (overestimates a lot). Known already: |
| 1018 | * start_seq < undo_marker and end_seq >= undo_marker. |
| 1019 | */ |
| 1020 | return !before(start_seq, end_seq - tp->max_window); |
| 1021 | } |
| 1022 | |
Ilpo Järvinen | 1c1e87e | 2007-10-10 02:45:32 -0700 | [diff] [blame] | 1023 | /* Check for lost retransmit. This superb idea is borrowed from "ratehalving". |
Yuchung Cheng | 974c123 | 2012-01-19 14:42:21 +0000 | [diff] [blame] | 1024 | * Event "B". Later note: FACK people cheated me again 8), we have to account |
Ilpo Järvinen | 1c1e87e | 2007-10-10 02:45:32 -0700 | [diff] [blame] | 1025 | * for reordering! Ugly, but should help. |
Ilpo Järvinen | f785a8e | 2007-10-11 17:35:41 -0700 | [diff] [blame] | 1026 | * |
| 1027 | * Search retransmitted skbs from write_queue that were sent when snd_nxt was |
| 1028 | * less than what is now known to be received by the other end (derived from |
Ilpo Järvinen | 9f58f3b | 2007-11-15 19:42:54 -0800 | [diff] [blame] | 1029 | * highest SACK block). Also calculate the lowest snd_nxt among the remaining |
| 1030 | * retransmitted skbs to avoid some costly processing per ACKs. |
Ilpo Järvinen | 1c1e87e | 2007-10-10 02:45:32 -0700 | [diff] [blame] | 1031 | */ |
Ilpo Järvinen | 407ef1d | 2007-12-02 00:47:57 +0200 | [diff] [blame] | 1032 | static void tcp_mark_lost_retrans(struct sock *sk) |
Ilpo Järvinen | 1c1e87e | 2007-10-10 02:45:32 -0700 | [diff] [blame] | 1033 | { |
Ilpo Järvinen | 9f58f3b | 2007-11-15 19:42:54 -0800 | [diff] [blame] | 1034 | const struct inet_connection_sock *icsk = inet_csk(sk); |
Ilpo Järvinen | 1c1e87e | 2007-10-10 02:45:32 -0700 | [diff] [blame] | 1035 | struct tcp_sock *tp = tcp_sk(sk); |
| 1036 | struct sk_buff *skb; |
Ilpo Järvinen | f785a8e | 2007-10-11 17:35:41 -0700 | [diff] [blame] | 1037 | int cnt = 0; |
Ilpo Järvinen | df2e014 | 2007-10-18 05:07:57 -0700 | [diff] [blame] | 1038 | u32 new_low_seq = tp->snd_nxt; |
Ilpo Järvinen | 6859d49 | 2007-12-02 00:48:06 +0200 | [diff] [blame] | 1039 | u32 received_upto = tcp_highest_sack_seq(tp); |
Ilpo Järvinen | 9f58f3b | 2007-11-15 19:42:54 -0800 | [diff] [blame] | 1040 | |
| 1041 | if (!tcp_is_fack(tp) || !tp->retrans_out || |
| 1042 | !after(received_upto, tp->lost_retrans_low) || |
| 1043 | icsk->icsk_ca_state != TCP_CA_Recovery) |
Ilpo Järvinen | 407ef1d | 2007-12-02 00:47:57 +0200 | [diff] [blame] | 1044 | return; |
Ilpo Järvinen | 1c1e87e | 2007-10-10 02:45:32 -0700 | [diff] [blame] | 1045 | |
| 1046 | tcp_for_write_queue(skb, sk) { |
| 1047 | u32 ack_seq = TCP_SKB_CB(skb)->ack_seq; |
| 1048 | |
| 1049 | if (skb == tcp_send_head(sk)) |
| 1050 | break; |
Ilpo Järvinen | f785a8e | 2007-10-11 17:35:41 -0700 | [diff] [blame] | 1051 | if (cnt == tp->retrans_out) |
Ilpo Järvinen | 1c1e87e | 2007-10-10 02:45:32 -0700 | [diff] [blame] | 1052 | break; |
| 1053 | if (!after(TCP_SKB_CB(skb)->end_seq, tp->snd_una)) |
| 1054 | continue; |
| 1055 | |
Ilpo Järvinen | f785a8e | 2007-10-11 17:35:41 -0700 | [diff] [blame] | 1056 | if (!(TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_RETRANS)) |
| 1057 | continue; |
| 1058 | |
Ilpo Järvinen | d0af416 | 2009-02-28 04:44:32 +0000 | [diff] [blame] | 1059 | /* TODO: We would like to get rid of tcp_is_fack(tp) only |
| 1060 | * constraint here (see above) but figuring out that at |
| 1061 | * least tp->reordering SACK blocks reside between ack_seq |
| 1062 | * and received_upto is not easy task to do cheaply with |
| 1063 | * the available datastructures. |
| 1064 | * |
| 1065 | * Whether FACK should check here for tp->reordering segs |
| 1066 | * in-between one could argue for either way (it would be |
| 1067 | * rather simple to implement as we could count fack_count |
| 1068 | * during the walk and do tp->fackets_out - fack_count). |
| 1069 | */ |
| 1070 | if (after(received_upto, ack_seq)) { |
Ilpo Järvinen | 1c1e87e | 2007-10-10 02:45:32 -0700 | [diff] [blame] | 1071 | TCP_SKB_CB(skb)->sacked &= ~TCPCB_SACKED_RETRANS; |
| 1072 | tp->retrans_out -= tcp_skb_pcount(skb); |
| 1073 | |
Ilpo Järvinen | 006f582 | 2008-09-20 21:20:20 -0700 | [diff] [blame] | 1074 | tcp_skb_mark_lost_uncond_verify(tp, skb); |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 1075 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPLOSTRETRANSMIT); |
Ilpo Järvinen | f785a8e | 2007-10-11 17:35:41 -0700 | [diff] [blame] | 1076 | } else { |
Ilpo Järvinen | df2e014 | 2007-10-18 05:07:57 -0700 | [diff] [blame] | 1077 | if (before(ack_seq, new_low_seq)) |
Ilpo Järvinen | b08d6cb | 2007-10-11 17:36:13 -0700 | [diff] [blame] | 1078 | new_low_seq = ack_seq; |
Ilpo Järvinen | f785a8e | 2007-10-11 17:35:41 -0700 | [diff] [blame] | 1079 | cnt += tcp_skb_pcount(skb); |
Ilpo Järvinen | 1c1e87e | 2007-10-10 02:45:32 -0700 | [diff] [blame] | 1080 | } |
| 1081 | } |
Ilpo Järvinen | b08d6cb | 2007-10-11 17:36:13 -0700 | [diff] [blame] | 1082 | |
| 1083 | if (tp->retrans_out) |
| 1084 | tp->lost_retrans_low = new_low_seq; |
Ilpo Järvinen | 1c1e87e | 2007-10-10 02:45:32 -0700 | [diff] [blame] | 1085 | } |
Ilpo Järvinen | 5b3c988 | 2007-08-24 22:54:44 -0700 | [diff] [blame] | 1086 | |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1087 | static bool tcp_check_dsack(struct sock *sk, const struct sk_buff *ack_skb, |
| 1088 | struct tcp_sack_block_wire *sp, int num_sacks, |
| 1089 | u32 prior_snd_una) |
David S. Miller | d06e021 | 2007-06-18 22:43:06 -0700 | [diff] [blame] | 1090 | { |
Pavel Emelyanov | 1ed8346 | 2008-07-16 20:29:51 -0700 | [diff] [blame] | 1091 | struct tcp_sock *tp = tcp_sk(sk); |
Harvey Harrison | d3e2ce3 | 2008-05-02 16:26:16 -0700 | [diff] [blame] | 1092 | u32 start_seq_0 = get_unaligned_be32(&sp[0].start_seq); |
| 1093 | u32 end_seq_0 = get_unaligned_be32(&sp[0].end_seq); |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1094 | bool dup_sack = false; |
David S. Miller | d06e021 | 2007-06-18 22:43:06 -0700 | [diff] [blame] | 1095 | |
| 1096 | if (before(start_seq_0, TCP_SKB_CB(ack_skb)->ack_seq)) { |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1097 | dup_sack = true; |
Ilpo Järvinen | e60402d | 2007-08-09 15:14:46 +0300 | [diff] [blame] | 1098 | tcp_dsack_seen(tp); |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 1099 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPDSACKRECV); |
David S. Miller | d06e021 | 2007-06-18 22:43:06 -0700 | [diff] [blame] | 1100 | } else if (num_sacks > 1) { |
Harvey Harrison | d3e2ce3 | 2008-05-02 16:26:16 -0700 | [diff] [blame] | 1101 | u32 end_seq_1 = get_unaligned_be32(&sp[1].end_seq); |
| 1102 | u32 start_seq_1 = get_unaligned_be32(&sp[1].start_seq); |
David S. Miller | d06e021 | 2007-06-18 22:43:06 -0700 | [diff] [blame] | 1103 | |
| 1104 | if (!after(end_seq_0, end_seq_1) && |
| 1105 | !before(start_seq_0, start_seq_1)) { |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1106 | dup_sack = true; |
Ilpo Järvinen | e60402d | 2007-08-09 15:14:46 +0300 | [diff] [blame] | 1107 | tcp_dsack_seen(tp); |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 1108 | NET_INC_STATS_BH(sock_net(sk), |
| 1109 | LINUX_MIB_TCPDSACKOFORECV); |
David S. Miller | d06e021 | 2007-06-18 22:43:06 -0700 | [diff] [blame] | 1110 | } |
| 1111 | } |
| 1112 | |
| 1113 | /* D-SACK for already forgotten data... Do dumb counting. */ |
Yuchung Cheng | c24f691 | 2011-02-07 12:57:04 +0000 | [diff] [blame] | 1114 | if (dup_sack && tp->undo_marker && tp->undo_retrans && |
David S. Miller | d06e021 | 2007-06-18 22:43:06 -0700 | [diff] [blame] | 1115 | !after(end_seq_0, prior_snd_una) && |
| 1116 | after(end_seq_0, tp->undo_marker)) |
| 1117 | tp->undo_retrans--; |
| 1118 | |
| 1119 | return dup_sack; |
| 1120 | } |
| 1121 | |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1122 | struct tcp_sacktag_state { |
| 1123 | int reord; |
| 1124 | int fack_count; |
| 1125 | int flag; |
Yuchung Cheng | 59c9af4 | 2013-07-22 16:20:47 -0700 | [diff] [blame] | 1126 | s32 rtt; /* RTT measured by SACKing never-retransmitted data */ |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1127 | }; |
| 1128 | |
Ilpo Järvinen | d193594 | 2007-10-11 17:34:25 -0700 | [diff] [blame] | 1129 | /* Check if skb is fully within the SACK block. In presence of GSO skbs, |
| 1130 | * the incoming SACK may not exactly match but we can find smaller MSS |
| 1131 | * aligned portion of it that matches. Therefore we might need to fragment |
| 1132 | * which may fail and creates some hassle (caller must handle error case |
| 1133 | * returns). |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1134 | * |
| 1135 | * FIXME: this could be merged to shift decision code |
Ilpo Järvinen | d193594 | 2007-10-11 17:34:25 -0700 | [diff] [blame] | 1136 | */ |
Adrian Bunk | 0f79efd | 2007-10-26 03:57:36 -0700 | [diff] [blame] | 1137 | static int tcp_match_skb_to_sack(struct sock *sk, struct sk_buff *skb, |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1138 | u32 start_seq, u32 end_seq) |
Ilpo Järvinen | d193594 | 2007-10-11 17:34:25 -0700 | [diff] [blame] | 1139 | { |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1140 | int err; |
| 1141 | bool in_sack; |
Ilpo Järvinen | d193594 | 2007-10-11 17:34:25 -0700 | [diff] [blame] | 1142 | unsigned int pkt_len; |
Ilpo Järvinen | adb92db | 2008-11-24 21:13:50 -0800 | [diff] [blame] | 1143 | unsigned int mss; |
Ilpo Järvinen | d193594 | 2007-10-11 17:34:25 -0700 | [diff] [blame] | 1144 | |
| 1145 | in_sack = !after(start_seq, TCP_SKB_CB(skb)->seq) && |
| 1146 | !before(end_seq, TCP_SKB_CB(skb)->end_seq); |
| 1147 | |
| 1148 | if (tcp_skb_pcount(skb) > 1 && !in_sack && |
| 1149 | after(TCP_SKB_CB(skb)->end_seq, start_seq)) { |
Ilpo Järvinen | adb92db | 2008-11-24 21:13:50 -0800 | [diff] [blame] | 1150 | mss = tcp_skb_mss(skb); |
Ilpo Järvinen | d193594 | 2007-10-11 17:34:25 -0700 | [diff] [blame] | 1151 | in_sack = !after(start_seq, TCP_SKB_CB(skb)->seq); |
| 1152 | |
Ilpo Järvinen | adb92db | 2008-11-24 21:13:50 -0800 | [diff] [blame] | 1153 | if (!in_sack) { |
Ilpo Järvinen | d193594 | 2007-10-11 17:34:25 -0700 | [diff] [blame] | 1154 | pkt_len = start_seq - TCP_SKB_CB(skb)->seq; |
Ilpo Järvinen | adb92db | 2008-11-24 21:13:50 -0800 | [diff] [blame] | 1155 | if (pkt_len < mss) |
| 1156 | pkt_len = mss; |
| 1157 | } else { |
Ilpo Järvinen | d193594 | 2007-10-11 17:34:25 -0700 | [diff] [blame] | 1158 | pkt_len = end_seq - TCP_SKB_CB(skb)->seq; |
Ilpo Järvinen | adb92db | 2008-11-24 21:13:50 -0800 | [diff] [blame] | 1159 | if (pkt_len < mss) |
| 1160 | return -EINVAL; |
| 1161 | } |
| 1162 | |
| 1163 | /* Round if necessary so that SACKs cover only full MSSes |
| 1164 | * and/or the remaining small portion (if present) |
| 1165 | */ |
| 1166 | if (pkt_len > mss) { |
| 1167 | unsigned int new_len = (pkt_len / mss) * mss; |
| 1168 | if (!in_sack && new_len < pkt_len) { |
| 1169 | new_len += mss; |
| 1170 | if (new_len > skb->len) |
| 1171 | return 0; |
| 1172 | } |
| 1173 | pkt_len = new_len; |
| 1174 | } |
| 1175 | err = tcp_fragment(sk, skb, pkt_len, mss); |
Ilpo Järvinen | d193594 | 2007-10-11 17:34:25 -0700 | [diff] [blame] | 1176 | if (err < 0) |
| 1177 | return err; |
| 1178 | } |
| 1179 | |
| 1180 | return in_sack; |
| 1181 | } |
| 1182 | |
Neal Cardwell | cc9a672 | 2012-02-12 18:37:09 +0000 | [diff] [blame] | 1183 | /* Mark the given newly-SACKed range as such, adjusting counters and hints. */ |
| 1184 | static u8 tcp_sacktag_one(struct sock *sk, |
| 1185 | struct tcp_sacktag_state *state, u8 sacked, |
| 1186 | u32 start_seq, u32 end_seq, |
Yuchung Cheng | 59c9af4 | 2013-07-22 16:20:47 -0700 | [diff] [blame] | 1187 | int dup_sack, int pcount, u32 xmit_time) |
Ilpo Järvinen | 9e10c47c | 2007-11-15 19:44:56 -0800 | [diff] [blame] | 1188 | { |
Ilpo Järvinen | 6859d49 | 2007-12-02 00:48:06 +0200 | [diff] [blame] | 1189 | struct tcp_sock *tp = tcp_sk(sk); |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1190 | int fack_count = state->fack_count; |
Ilpo Järvinen | 9e10c47c | 2007-11-15 19:44:56 -0800 | [diff] [blame] | 1191 | |
| 1192 | /* Account D-SACK for retransmitted packet. */ |
| 1193 | if (dup_sack && (sacked & TCPCB_RETRANS)) { |
Yuchung Cheng | c24f691 | 2011-02-07 12:57:04 +0000 | [diff] [blame] | 1194 | if (tp->undo_marker && tp->undo_retrans && |
Neal Cardwell | cc9a672 | 2012-02-12 18:37:09 +0000 | [diff] [blame] | 1195 | after(end_seq, tp->undo_marker)) |
Ilpo Järvinen | 9e10c47c | 2007-11-15 19:44:56 -0800 | [diff] [blame] | 1196 | tp->undo_retrans--; |
Ilpo Järvinen | ede9f3b | 2007-12-02 00:47:58 +0200 | [diff] [blame] | 1197 | if (sacked & TCPCB_SACKED_ACKED) |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1198 | state->reord = min(fack_count, state->reord); |
Ilpo Järvinen | 9e10c47c | 2007-11-15 19:44:56 -0800 | [diff] [blame] | 1199 | } |
| 1200 | |
| 1201 | /* Nothing to do; acked frame is about to be dropped (was ACKed). */ |
Neal Cardwell | cc9a672 | 2012-02-12 18:37:09 +0000 | [diff] [blame] | 1202 | if (!after(end_seq, tp->snd_una)) |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1203 | return sacked; |
Ilpo Järvinen | 9e10c47c | 2007-11-15 19:44:56 -0800 | [diff] [blame] | 1204 | |
| 1205 | if (!(sacked & TCPCB_SACKED_ACKED)) { |
| 1206 | if (sacked & TCPCB_SACKED_RETRANS) { |
| 1207 | /* If the segment is not tagged as lost, |
| 1208 | * we do not clear RETRANS, believing |
| 1209 | * that retransmission is still in flight. |
| 1210 | */ |
| 1211 | if (sacked & TCPCB_LOST) { |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1212 | sacked &= ~(TCPCB_LOST|TCPCB_SACKED_RETRANS); |
Ilpo Järvinen | f58b22f | 2008-11-24 21:14:43 -0800 | [diff] [blame] | 1213 | tp->lost_out -= pcount; |
| 1214 | tp->retrans_out -= pcount; |
Ilpo Järvinen | 9e10c47c | 2007-11-15 19:44:56 -0800 | [diff] [blame] | 1215 | } |
| 1216 | } else { |
| 1217 | if (!(sacked & TCPCB_RETRANS)) { |
| 1218 | /* New sack for not retransmitted frame, |
| 1219 | * which was in hole. It is reordering. |
| 1220 | */ |
Neal Cardwell | cc9a672 | 2012-02-12 18:37:09 +0000 | [diff] [blame] | 1221 | if (before(start_seq, |
Ilpo Järvinen | 9e10c47c | 2007-11-15 19:44:56 -0800 | [diff] [blame] | 1222 | tcp_highest_sack_seq(tp))) |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1223 | state->reord = min(fack_count, |
| 1224 | state->reord); |
Yuchung Cheng | e33099f | 2013-03-20 13:33:00 +0000 | [diff] [blame] | 1225 | if (!after(end_seq, tp->high_seq)) |
| 1226 | state->flag |= FLAG_ORIG_SACK_ACKED; |
Yuchung Cheng | 59c9af4 | 2013-07-22 16:20:47 -0700 | [diff] [blame] | 1227 | /* Pick the earliest sequence sacked for RTT */ |
| 1228 | if (state->rtt < 0) |
| 1229 | state->rtt = tcp_time_stamp - xmit_time; |
Ilpo Järvinen | 9e10c47c | 2007-11-15 19:44:56 -0800 | [diff] [blame] | 1230 | } |
| 1231 | |
| 1232 | if (sacked & TCPCB_LOST) { |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1233 | sacked &= ~TCPCB_LOST; |
Ilpo Järvinen | f58b22f | 2008-11-24 21:14:43 -0800 | [diff] [blame] | 1234 | tp->lost_out -= pcount; |
Ilpo Järvinen | 9e10c47c | 2007-11-15 19:44:56 -0800 | [diff] [blame] | 1235 | } |
| 1236 | } |
| 1237 | |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1238 | sacked |= TCPCB_SACKED_ACKED; |
| 1239 | state->flag |= FLAG_DATA_SACKED; |
Ilpo Järvinen | f58b22f | 2008-11-24 21:14:43 -0800 | [diff] [blame] | 1240 | tp->sacked_out += pcount; |
Ilpo Järvinen | 9e10c47c | 2007-11-15 19:44:56 -0800 | [diff] [blame] | 1241 | |
Ilpo Järvinen | f58b22f | 2008-11-24 21:14:43 -0800 | [diff] [blame] | 1242 | fack_count += pcount; |
Ilpo Järvinen | 9e10c47c | 2007-11-15 19:44:56 -0800 | [diff] [blame] | 1243 | |
| 1244 | /* Lost marker hint past SACKed? Tweak RFC3517 cnt */ |
| 1245 | if (!tcp_is_fack(tp) && (tp->lost_skb_hint != NULL) && |
Neal Cardwell | cc9a672 | 2012-02-12 18:37:09 +0000 | [diff] [blame] | 1246 | before(start_seq, TCP_SKB_CB(tp->lost_skb_hint)->seq)) |
Ilpo Järvinen | f58b22f | 2008-11-24 21:14:43 -0800 | [diff] [blame] | 1247 | tp->lost_cnt_hint += pcount; |
Ilpo Järvinen | 9e10c47c | 2007-11-15 19:44:56 -0800 | [diff] [blame] | 1248 | |
| 1249 | if (fack_count > tp->fackets_out) |
| 1250 | tp->fackets_out = fack_count; |
Ilpo Järvinen | 9e10c47c | 2007-11-15 19:44:56 -0800 | [diff] [blame] | 1251 | } |
| 1252 | |
| 1253 | /* D-SACK. We can detect redundant retransmission in S|R and plain R |
| 1254 | * frames and clear it. undo_retrans is decreased above, L|R frames |
| 1255 | * are accounted above as well. |
| 1256 | */ |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1257 | if (dup_sack && (sacked & TCPCB_SACKED_RETRANS)) { |
| 1258 | sacked &= ~TCPCB_SACKED_RETRANS; |
Ilpo Järvinen | f58b22f | 2008-11-24 21:14:43 -0800 | [diff] [blame] | 1259 | tp->retrans_out -= pcount; |
Ilpo Järvinen | 9e10c47c | 2007-11-15 19:44:56 -0800 | [diff] [blame] | 1260 | } |
| 1261 | |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1262 | return sacked; |
Ilpo Järvinen | 9e10c47c | 2007-11-15 19:44:56 -0800 | [diff] [blame] | 1263 | } |
| 1264 | |
Neal Cardwell | daef52b | 2012-02-12 18:37:10 +0000 | [diff] [blame] | 1265 | /* Shift newly-SACKed bytes from this skb to the immediately previous |
| 1266 | * already-SACKed sk_buff. Mark the newly-SACKed bytes as such. |
| 1267 | */ |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1268 | static bool tcp_shifted_skb(struct sock *sk, struct sk_buff *skb, |
| 1269 | struct tcp_sacktag_state *state, |
| 1270 | unsigned int pcount, int shifted, int mss, |
| 1271 | bool dup_sack) |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1272 | { |
| 1273 | struct tcp_sock *tp = tcp_sk(sk); |
Ilpo Järvinen | 5013316 | 2008-12-05 22:42:41 -0800 | [diff] [blame] | 1274 | struct sk_buff *prev = tcp_write_queue_prev(sk, skb); |
Neal Cardwell | daef52b | 2012-02-12 18:37:10 +0000 | [diff] [blame] | 1275 | u32 start_seq = TCP_SKB_CB(skb)->seq; /* start of newly-SACKed */ |
| 1276 | u32 end_seq = start_seq + shifted; /* end of newly-SACKed */ |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1277 | |
| 1278 | BUG_ON(!pcount); |
| 1279 | |
Neal Cardwell | 4c90d3b | 2012-02-26 10:06:19 +0000 | [diff] [blame] | 1280 | /* Adjust counters and hints for the newly sacked sequence |
| 1281 | * range but discard the return value since prev is already |
| 1282 | * marked. We must tag the range first because the seq |
| 1283 | * advancement below implicitly advances |
| 1284 | * tcp_highest_sack_seq() when skb is highest_sack. |
| 1285 | */ |
| 1286 | tcp_sacktag_one(sk, state, TCP_SKB_CB(skb)->sacked, |
Yuchung Cheng | 59c9af4 | 2013-07-22 16:20:47 -0700 | [diff] [blame] | 1287 | start_seq, end_seq, dup_sack, pcount, |
| 1288 | TCP_SKB_CB(skb)->when); |
Neal Cardwell | 4c90d3b | 2012-02-26 10:06:19 +0000 | [diff] [blame] | 1289 | |
| 1290 | if (skb == tp->lost_skb_hint) |
Neal Cardwell | 0af2a0d | 2012-02-13 20:22:08 +0000 | [diff] [blame] | 1291 | tp->lost_cnt_hint += pcount; |
| 1292 | |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1293 | TCP_SKB_CB(prev)->end_seq += shifted; |
| 1294 | TCP_SKB_CB(skb)->seq += shifted; |
| 1295 | |
| 1296 | skb_shinfo(prev)->gso_segs += pcount; |
| 1297 | BUG_ON(skb_shinfo(skb)->gso_segs < pcount); |
| 1298 | skb_shinfo(skb)->gso_segs -= pcount; |
| 1299 | |
| 1300 | /* When we're adding to gso_segs == 1, gso_size will be zero, |
| 1301 | * in theory this shouldn't be necessary but as long as DSACK |
| 1302 | * code can come after this skb later on it's better to keep |
| 1303 | * setting gso_size to something. |
| 1304 | */ |
| 1305 | if (!skb_shinfo(prev)->gso_size) { |
| 1306 | skb_shinfo(prev)->gso_size = mss; |
Pravin B Shelar | c9af6db | 2013-02-11 09:27:41 +0000 | [diff] [blame] | 1307 | skb_shinfo(prev)->gso_type = sk->sk_gso_type; |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1308 | } |
| 1309 | |
| 1310 | /* CHECKME: To clear or not to clear? Mimics normal skb currently */ |
| 1311 | if (skb_shinfo(skb)->gso_segs <= 1) { |
| 1312 | skb_shinfo(skb)->gso_size = 0; |
Pravin B Shelar | c9af6db | 2013-02-11 09:27:41 +0000 | [diff] [blame] | 1313 | skb_shinfo(skb)->gso_type = 0; |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1314 | } |
| 1315 | |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1316 | /* Difference in this won't matter, both ACKed by the same cumul. ACK */ |
| 1317 | TCP_SKB_CB(prev)->sacked |= (TCP_SKB_CB(skb)->sacked & TCPCB_EVER_RETRANS); |
| 1318 | |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1319 | if (skb->len > 0) { |
| 1320 | BUG_ON(!tcp_skb_pcount(skb)); |
Ilpo Järvinen | 111cc8b | 2008-11-24 21:27:22 -0800 | [diff] [blame] | 1321 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SACKSHIFTED); |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1322 | return false; |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1323 | } |
| 1324 | |
| 1325 | /* Whole SKB was eaten :-) */ |
| 1326 | |
Ilpo Järvinen | 92ee76b | 2008-11-24 21:26:56 -0800 | [diff] [blame] | 1327 | if (skb == tp->retransmit_skb_hint) |
| 1328 | tp->retransmit_skb_hint = prev; |
Ilpo Järvinen | 92ee76b | 2008-11-24 21:26:56 -0800 | [diff] [blame] | 1329 | if (skb == tp->lost_skb_hint) { |
| 1330 | tp->lost_skb_hint = prev; |
| 1331 | tp->lost_cnt_hint -= tcp_skb_pcount(prev); |
| 1332 | } |
| 1333 | |
Eric Dumazet | 5e8a402f | 2013-10-04 10:31:41 -0700 | [diff] [blame] | 1334 | TCP_SKB_CB(prev)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags; |
| 1335 | if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) |
| 1336 | TCP_SKB_CB(prev)->end_seq++; |
| 1337 | |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1338 | if (skb == tcp_highest_sack(sk)) |
| 1339 | tcp_advance_highest_sack(sk, skb); |
| 1340 | |
| 1341 | tcp_unlink_write_queue(skb, sk); |
| 1342 | sk_wmem_free_skb(sk, skb); |
| 1343 | |
Ilpo Järvinen | 111cc8b | 2008-11-24 21:27:22 -0800 | [diff] [blame] | 1344 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SACKMERGED); |
| 1345 | |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1346 | return true; |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1347 | } |
| 1348 | |
| 1349 | /* I wish gso_size would have a bit more sane initialization than |
| 1350 | * something-or-zero which complicates things |
| 1351 | */ |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 1352 | static int tcp_skb_seglen(const struct sk_buff *skb) |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1353 | { |
Ilpo Järvinen | 775ffab | 2008-12-05 22:41:26 -0800 | [diff] [blame] | 1354 | return tcp_skb_pcount(skb) == 1 ? skb->len : tcp_skb_mss(skb); |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1355 | } |
| 1356 | |
| 1357 | /* Shifting pages past head area doesn't work */ |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 1358 | static int skb_can_shift(const struct sk_buff *skb) |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1359 | { |
| 1360 | return !skb_headlen(skb) && skb_is_nonlinear(skb); |
| 1361 | } |
| 1362 | |
| 1363 | /* Try collapsing SACK blocks spanning across multiple skbs to a single |
| 1364 | * skb. |
| 1365 | */ |
| 1366 | static struct sk_buff *tcp_shift_skb_data(struct sock *sk, struct sk_buff *skb, |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1367 | struct tcp_sacktag_state *state, |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1368 | u32 start_seq, u32 end_seq, |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1369 | bool dup_sack) |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1370 | { |
| 1371 | struct tcp_sock *tp = tcp_sk(sk); |
| 1372 | struct sk_buff *prev; |
| 1373 | int mss; |
| 1374 | int pcount = 0; |
| 1375 | int len; |
| 1376 | int in_sack; |
| 1377 | |
| 1378 | if (!sk_can_gso(sk)) |
| 1379 | goto fallback; |
| 1380 | |
| 1381 | /* Normally R but no L won't result in plain S */ |
| 1382 | if (!dup_sack && |
Ilpo Järvinen | 9969ca5 | 2008-12-05 22:41:06 -0800 | [diff] [blame] | 1383 | (TCP_SKB_CB(skb)->sacked & (TCPCB_LOST|TCPCB_SACKED_RETRANS)) == TCPCB_SACKED_RETRANS) |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1384 | goto fallback; |
| 1385 | if (!skb_can_shift(skb)) |
| 1386 | goto fallback; |
| 1387 | /* This frame is about to be dropped (was ACKed). */ |
| 1388 | if (!after(TCP_SKB_CB(skb)->end_seq, tp->snd_una)) |
| 1389 | goto fallback; |
| 1390 | |
| 1391 | /* Can only happen with delayed DSACK + discard craziness */ |
| 1392 | if (unlikely(skb == tcp_write_queue_head(sk))) |
| 1393 | goto fallback; |
| 1394 | prev = tcp_write_queue_prev(sk, skb); |
| 1395 | |
| 1396 | if ((TCP_SKB_CB(prev)->sacked & TCPCB_TAGBITS) != TCPCB_SACKED_ACKED) |
| 1397 | goto fallback; |
| 1398 | |
| 1399 | in_sack = !after(start_seq, TCP_SKB_CB(skb)->seq) && |
| 1400 | !before(end_seq, TCP_SKB_CB(skb)->end_seq); |
| 1401 | |
| 1402 | if (in_sack) { |
| 1403 | len = skb->len; |
| 1404 | pcount = tcp_skb_pcount(skb); |
Ilpo Järvinen | 775ffab | 2008-12-05 22:41:26 -0800 | [diff] [blame] | 1405 | mss = tcp_skb_seglen(skb); |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1406 | |
| 1407 | /* TODO: Fix DSACKs to not fragment already SACKed and we can |
| 1408 | * drop this restriction as unnecessary |
| 1409 | */ |
Ilpo Järvinen | 775ffab | 2008-12-05 22:41:26 -0800 | [diff] [blame] | 1410 | if (mss != tcp_skb_seglen(prev)) |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1411 | goto fallback; |
| 1412 | } else { |
| 1413 | if (!after(TCP_SKB_CB(skb)->end_seq, start_seq)) |
| 1414 | goto noop; |
| 1415 | /* CHECKME: This is non-MSS split case only?, this will |
| 1416 | * cause skipped skbs due to advancing loop btw, original |
| 1417 | * has that feature too |
| 1418 | */ |
| 1419 | if (tcp_skb_pcount(skb) <= 1) |
| 1420 | goto noop; |
| 1421 | |
| 1422 | in_sack = !after(start_seq, TCP_SKB_CB(skb)->seq); |
| 1423 | if (!in_sack) { |
| 1424 | /* TODO: head merge to next could be attempted here |
| 1425 | * if (!after(TCP_SKB_CB(skb)->end_seq, end_seq)), |
| 1426 | * though it might not be worth of the additional hassle |
| 1427 | * |
| 1428 | * ...we can probably just fallback to what was done |
| 1429 | * previously. We could try merging non-SACKed ones |
| 1430 | * as well but it probably isn't going to buy off |
| 1431 | * because later SACKs might again split them, and |
| 1432 | * it would make skb timestamp tracking considerably |
| 1433 | * harder problem. |
| 1434 | */ |
| 1435 | goto fallback; |
| 1436 | } |
| 1437 | |
| 1438 | len = end_seq - TCP_SKB_CB(skb)->seq; |
| 1439 | BUG_ON(len < 0); |
| 1440 | BUG_ON(len > skb->len); |
| 1441 | |
| 1442 | /* MSS boundaries should be honoured or else pcount will |
| 1443 | * severely break even though it makes things bit trickier. |
| 1444 | * Optimize common case to avoid most of the divides |
| 1445 | */ |
| 1446 | mss = tcp_skb_mss(skb); |
| 1447 | |
| 1448 | /* TODO: Fix DSACKs to not fragment already SACKed and we can |
| 1449 | * drop this restriction as unnecessary |
| 1450 | */ |
Ilpo Järvinen | 775ffab | 2008-12-05 22:41:26 -0800 | [diff] [blame] | 1451 | if (mss != tcp_skb_seglen(prev)) |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1452 | goto fallback; |
| 1453 | |
| 1454 | if (len == mss) { |
| 1455 | pcount = 1; |
| 1456 | } else if (len < mss) { |
| 1457 | goto noop; |
| 1458 | } else { |
| 1459 | pcount = len / mss; |
| 1460 | len = pcount * mss; |
| 1461 | } |
| 1462 | } |
| 1463 | |
Neal Cardwell | 4648dc9 | 2012-03-05 19:35:04 +0000 | [diff] [blame] | 1464 | /* tcp_sacktag_one() won't SACK-tag ranges below snd_una */ |
| 1465 | if (!after(TCP_SKB_CB(skb)->seq + len, tp->snd_una)) |
| 1466 | goto fallback; |
| 1467 | |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1468 | if (!skb_shift(prev, skb, len)) |
| 1469 | goto fallback; |
Ilpo Järvinen | 9ec06ff | 2009-03-01 00:21:36 -0800 | [diff] [blame] | 1470 | if (!tcp_shifted_skb(sk, skb, state, pcount, len, mss, dup_sack)) |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1471 | goto out; |
| 1472 | |
| 1473 | /* Hole filled allows collapsing with the next as well, this is very |
| 1474 | * useful when hole on every nth skb pattern happens |
| 1475 | */ |
| 1476 | if (prev == tcp_write_queue_tail(sk)) |
| 1477 | goto out; |
| 1478 | skb = tcp_write_queue_next(sk, prev); |
| 1479 | |
Ilpo Järvinen | f0bc52f | 2008-12-05 22:40:47 -0800 | [diff] [blame] | 1480 | if (!skb_can_shift(skb) || |
| 1481 | (skb == tcp_send_head(sk)) || |
| 1482 | ((TCP_SKB_CB(skb)->sacked & TCPCB_TAGBITS) != TCPCB_SACKED_ACKED) || |
Ilpo Järvinen | 775ffab | 2008-12-05 22:41:26 -0800 | [diff] [blame] | 1483 | (mss != tcp_skb_seglen(skb))) |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1484 | goto out; |
| 1485 | |
| 1486 | len = skb->len; |
| 1487 | if (skb_shift(prev, skb, len)) { |
| 1488 | pcount += tcp_skb_pcount(skb); |
Ilpo Järvinen | 9ec06ff | 2009-03-01 00:21:36 -0800 | [diff] [blame] | 1489 | tcp_shifted_skb(sk, skb, state, tcp_skb_pcount(skb), len, mss, 0); |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1490 | } |
| 1491 | |
| 1492 | out: |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1493 | state->fack_count += pcount; |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1494 | return prev; |
| 1495 | |
| 1496 | noop: |
| 1497 | return skb; |
| 1498 | |
| 1499 | fallback: |
Ilpo Järvinen | 111cc8b | 2008-11-24 21:27:22 -0800 | [diff] [blame] | 1500 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SACKSHIFTFALLBACK); |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1501 | return NULL; |
| 1502 | } |
| 1503 | |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1504 | static struct sk_buff *tcp_sacktag_walk(struct sk_buff *skb, struct sock *sk, |
| 1505 | struct tcp_sack_block *next_dup, |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1506 | struct tcp_sacktag_state *state, |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1507 | u32 start_seq, u32 end_seq, |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1508 | bool dup_sack_in) |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1509 | { |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1510 | struct tcp_sock *tp = tcp_sk(sk); |
| 1511 | struct sk_buff *tmp; |
| 1512 | |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1513 | tcp_for_write_queue_from(skb, sk) { |
| 1514 | int in_sack = 0; |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1515 | bool dup_sack = dup_sack_in; |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1516 | |
| 1517 | if (skb == tcp_send_head(sk)) |
| 1518 | break; |
| 1519 | |
| 1520 | /* queue is in-order => we can short-circuit the walk early */ |
| 1521 | if (!before(TCP_SKB_CB(skb)->seq, end_seq)) |
| 1522 | break; |
| 1523 | |
| 1524 | if ((next_dup != NULL) && |
| 1525 | before(TCP_SKB_CB(skb)->seq, next_dup->end_seq)) { |
| 1526 | in_sack = tcp_match_skb_to_sack(sk, skb, |
| 1527 | next_dup->start_seq, |
| 1528 | next_dup->end_seq); |
| 1529 | if (in_sack > 0) |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1530 | dup_sack = true; |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1531 | } |
| 1532 | |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1533 | /* skb reference here is a bit tricky to get right, since |
| 1534 | * shifting can eat and free both this skb and the next, |
| 1535 | * so not even _safe variant of the loop is enough. |
| 1536 | */ |
| 1537 | if (in_sack <= 0) { |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1538 | tmp = tcp_shift_skb_data(sk, skb, state, |
| 1539 | start_seq, end_seq, dup_sack); |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1540 | if (tmp != NULL) { |
| 1541 | if (tmp != skb) { |
| 1542 | skb = tmp; |
| 1543 | continue; |
| 1544 | } |
| 1545 | |
| 1546 | in_sack = 0; |
| 1547 | } else { |
| 1548 | in_sack = tcp_match_skb_to_sack(sk, skb, |
| 1549 | start_seq, |
| 1550 | end_seq); |
| 1551 | } |
| 1552 | } |
| 1553 | |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1554 | if (unlikely(in_sack < 0)) |
| 1555 | break; |
| 1556 | |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1557 | if (in_sack) { |
Neal Cardwell | cc9a672 | 2012-02-12 18:37:09 +0000 | [diff] [blame] | 1558 | TCP_SKB_CB(skb)->sacked = |
| 1559 | tcp_sacktag_one(sk, |
| 1560 | state, |
| 1561 | TCP_SKB_CB(skb)->sacked, |
| 1562 | TCP_SKB_CB(skb)->seq, |
| 1563 | TCP_SKB_CB(skb)->end_seq, |
| 1564 | dup_sack, |
Yuchung Cheng | 59c9af4 | 2013-07-22 16:20:47 -0700 | [diff] [blame] | 1565 | tcp_skb_pcount(skb), |
| 1566 | TCP_SKB_CB(skb)->when); |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1567 | |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1568 | if (!before(TCP_SKB_CB(skb)->seq, |
| 1569 | tcp_highest_sack_seq(tp))) |
| 1570 | tcp_advance_highest_sack(sk, skb); |
| 1571 | } |
| 1572 | |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1573 | state->fack_count += tcp_skb_pcount(skb); |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1574 | } |
| 1575 | return skb; |
| 1576 | } |
| 1577 | |
| 1578 | /* Avoid all extra work that is being done by sacktag while walking in |
| 1579 | * a normal way |
| 1580 | */ |
| 1581 | static struct sk_buff *tcp_sacktag_skip(struct sk_buff *skb, struct sock *sk, |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1582 | struct tcp_sacktag_state *state, |
| 1583 | u32 skip_to_seq) |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1584 | { |
| 1585 | tcp_for_write_queue_from(skb, sk) { |
| 1586 | if (skb == tcp_send_head(sk)) |
| 1587 | break; |
| 1588 | |
Ilpo Järvinen | e8bae27 | 2008-11-24 21:12:28 -0800 | [diff] [blame] | 1589 | if (after(TCP_SKB_CB(skb)->end_seq, skip_to_seq)) |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1590 | break; |
Ilpo Järvinen | d152a7d | 2008-03-03 12:10:16 -0800 | [diff] [blame] | 1591 | |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1592 | state->fack_count += tcp_skb_pcount(skb); |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1593 | } |
| 1594 | return skb; |
| 1595 | } |
| 1596 | |
| 1597 | static struct sk_buff *tcp_maybe_skipping_dsack(struct sk_buff *skb, |
| 1598 | struct sock *sk, |
| 1599 | struct tcp_sack_block *next_dup, |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1600 | struct tcp_sacktag_state *state, |
| 1601 | u32 skip_to_seq) |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1602 | { |
| 1603 | if (next_dup == NULL) |
| 1604 | return skb; |
| 1605 | |
| 1606 | if (before(next_dup->start_seq, skip_to_seq)) { |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1607 | skb = tcp_sacktag_skip(skb, sk, state, next_dup->start_seq); |
| 1608 | skb = tcp_sacktag_walk(skb, sk, NULL, state, |
| 1609 | next_dup->start_seq, next_dup->end_seq, |
| 1610 | 1); |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1611 | } |
| 1612 | |
| 1613 | return skb; |
| 1614 | } |
| 1615 | |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 1616 | static int tcp_sack_cache_ok(const struct tcp_sock *tp, const struct tcp_sack_block *cache) |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1617 | { |
| 1618 | return cache < tp->recv_sack_cache + ARRAY_SIZE(tp->recv_sack_cache); |
| 1619 | } |
| 1620 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1621 | static int |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 1622 | tcp_sacktag_write_queue(struct sock *sk, const struct sk_buff *ack_skb, |
Yuchung Cheng | 59c9af4 | 2013-07-22 16:20:47 -0700 | [diff] [blame] | 1623 | u32 prior_snd_una, s32 *sack_rtt) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1624 | { |
| 1625 | struct tcp_sock *tp = tcp_sk(sk); |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 1626 | const unsigned char *ptr = (skb_transport_header(ack_skb) + |
| 1627 | TCP_SKB_CB(ack_skb)->sacked); |
Ilpo Järvinen | fd6dad6 | 2007-11-15 19:49:47 -0800 | [diff] [blame] | 1628 | struct tcp_sack_block_wire *sp_wire = (struct tcp_sack_block_wire *)(ptr+2); |
Adam Langley | 4389dde | 2008-07-19 00:07:02 -0700 | [diff] [blame] | 1629 | struct tcp_sack_block sp[TCP_NUM_SACKS]; |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1630 | struct tcp_sack_block *cache; |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1631 | struct tcp_sacktag_state state; |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1632 | struct sk_buff *skb; |
Adam Langley | 4389dde | 2008-07-19 00:07:02 -0700 | [diff] [blame] | 1633 | int num_sacks = min(TCP_NUM_SACKS, (ptr[1] - TCPOLEN_SACK_BASE) >> 3); |
Ilpo Järvinen | fd6dad6 | 2007-11-15 19:49:47 -0800 | [diff] [blame] | 1634 | int used_sacks; |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1635 | bool found_dup_sack = false; |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1636 | int i, j; |
Baruch Even | fda03fb | 2007-02-04 23:35:57 -0800 | [diff] [blame] | 1637 | int first_sack_index; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1638 | |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1639 | state.flag = 0; |
| 1640 | state.reord = tp->packets_out; |
Yuchung Cheng | 59c9af4 | 2013-07-22 16:20:47 -0700 | [diff] [blame] | 1641 | state.rtt = -1; |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1642 | |
Ilpo Järvinen | d738cd8 | 2007-03-24 21:03:23 -0700 | [diff] [blame] | 1643 | if (!tp->sacked_out) { |
Ilpo Järvinen | de83c05 | 2007-10-07 23:37:55 -0700 | [diff] [blame] | 1644 | if (WARN_ON(tp->fackets_out)) |
| 1645 | tp->fackets_out = 0; |
Ilpo Järvinen | 6859d49 | 2007-12-02 00:48:06 +0200 | [diff] [blame] | 1646 | tcp_highest_sack_reset(sk); |
Ilpo Järvinen | d738cd8 | 2007-03-24 21:03:23 -0700 | [diff] [blame] | 1647 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1648 | |
Pavel Emelyanov | 1ed8346 | 2008-07-16 20:29:51 -0700 | [diff] [blame] | 1649 | found_dup_sack = tcp_check_dsack(sk, ack_skb, sp_wire, |
David S. Miller | d06e021 | 2007-06-18 22:43:06 -0700 | [diff] [blame] | 1650 | num_sacks, prior_snd_una); |
| 1651 | if (found_dup_sack) |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1652 | state.flag |= FLAG_DSACKING_ACK; |
Baruch Even | 6f74651 | 2007-02-04 23:36:42 -0800 | [diff] [blame] | 1653 | |
| 1654 | /* Eliminate too old ACKs, but take into |
| 1655 | * account more or less fresh ones, they can |
| 1656 | * contain valid SACK info. |
| 1657 | */ |
| 1658 | if (before(TCP_SKB_CB(ack_skb)->ack_seq, prior_snd_una - tp->max_window)) |
| 1659 | return 0; |
| 1660 | |
Ilpo Järvinen | 96a2d41 | 2007-11-14 15:47:18 -0800 | [diff] [blame] | 1661 | if (!tp->packets_out) |
| 1662 | goto out; |
| 1663 | |
Ilpo Järvinen | fd6dad6 | 2007-11-15 19:49:47 -0800 | [diff] [blame] | 1664 | used_sacks = 0; |
| 1665 | first_sack_index = 0; |
| 1666 | for (i = 0; i < num_sacks; i++) { |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1667 | bool dup_sack = !i && found_dup_sack; |
Ilpo Järvinen | fd6dad6 | 2007-11-15 19:49:47 -0800 | [diff] [blame] | 1668 | |
Harvey Harrison | d3e2ce3 | 2008-05-02 16:26:16 -0700 | [diff] [blame] | 1669 | sp[used_sacks].start_seq = get_unaligned_be32(&sp_wire[i].start_seq); |
| 1670 | sp[used_sacks].end_seq = get_unaligned_be32(&sp_wire[i].end_seq); |
Ilpo Järvinen | fd6dad6 | 2007-11-15 19:49:47 -0800 | [diff] [blame] | 1671 | |
| 1672 | if (!tcp_is_sackblock_valid(tp, dup_sack, |
| 1673 | sp[used_sacks].start_seq, |
| 1674 | sp[used_sacks].end_seq)) { |
Pavel Emelyanov | 40b215e | 2008-07-03 01:05:41 -0700 | [diff] [blame] | 1675 | int mib_idx; |
| 1676 | |
Ilpo Järvinen | fd6dad6 | 2007-11-15 19:49:47 -0800 | [diff] [blame] | 1677 | if (dup_sack) { |
| 1678 | if (!tp->undo_marker) |
Pavel Emelyanov | 40b215e | 2008-07-03 01:05:41 -0700 | [diff] [blame] | 1679 | mib_idx = LINUX_MIB_TCPDSACKIGNOREDNOUNDO; |
Ilpo Järvinen | fd6dad6 | 2007-11-15 19:49:47 -0800 | [diff] [blame] | 1680 | else |
Pavel Emelyanov | 40b215e | 2008-07-03 01:05:41 -0700 | [diff] [blame] | 1681 | mib_idx = LINUX_MIB_TCPDSACKIGNOREDOLD; |
Ilpo Järvinen | fd6dad6 | 2007-11-15 19:49:47 -0800 | [diff] [blame] | 1682 | } else { |
| 1683 | /* Don't count olds caused by ACK reordering */ |
| 1684 | if ((TCP_SKB_CB(ack_skb)->ack_seq != tp->snd_una) && |
| 1685 | !after(sp[used_sacks].end_seq, tp->snd_una)) |
| 1686 | continue; |
Pavel Emelyanov | 40b215e | 2008-07-03 01:05:41 -0700 | [diff] [blame] | 1687 | mib_idx = LINUX_MIB_TCPSACKDISCARD; |
Ilpo Järvinen | fd6dad6 | 2007-11-15 19:49:47 -0800 | [diff] [blame] | 1688 | } |
Pavel Emelyanov | 40b215e | 2008-07-03 01:05:41 -0700 | [diff] [blame] | 1689 | |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 1690 | NET_INC_STATS_BH(sock_net(sk), mib_idx); |
Ilpo Järvinen | fd6dad6 | 2007-11-15 19:49:47 -0800 | [diff] [blame] | 1691 | if (i == 0) |
| 1692 | first_sack_index = -1; |
| 1693 | continue; |
| 1694 | } |
| 1695 | |
| 1696 | /* Ignore very old stuff early */ |
| 1697 | if (!after(sp[used_sacks].end_seq, prior_snd_una)) |
| 1698 | continue; |
| 1699 | |
| 1700 | used_sacks++; |
| 1701 | } |
| 1702 | |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1703 | /* order SACK blocks to allow in order walk of the retrans queue */ |
| 1704 | for (i = used_sacks - 1; i > 0; i--) { |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 1705 | for (j = 0; j < i; j++) { |
| 1706 | if (after(sp[j].start_seq, sp[j + 1].start_seq)) { |
Ilpo Järvinen | a0bffff | 2009-03-21 13:36:17 -0700 | [diff] [blame] | 1707 | swap(sp[j], sp[j + 1]); |
Stephen Hemminger | 6a438bb | 2005-11-10 17:14:59 -0800 | [diff] [blame] | 1708 | |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1709 | /* Track where the first SACK block goes to */ |
| 1710 | if (j == first_sack_index) |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 1711 | first_sack_index = j + 1; |
Stephen Hemminger | 6a438bb | 2005-11-10 17:14:59 -0800 | [diff] [blame] | 1712 | } |
| 1713 | } |
| 1714 | } |
| 1715 | |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1716 | skb = tcp_write_queue_head(sk); |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1717 | state.fack_count = 0; |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1718 | i = 0; |
| 1719 | |
| 1720 | if (!tp->sacked_out) { |
| 1721 | /* It's already past, so skip checking against it */ |
| 1722 | cache = tp->recv_sack_cache + ARRAY_SIZE(tp->recv_sack_cache); |
| 1723 | } else { |
| 1724 | cache = tp->recv_sack_cache; |
| 1725 | /* Skip empty blocks in at head of the cache */ |
| 1726 | while (tcp_sack_cache_ok(tp, cache) && !cache->start_seq && |
| 1727 | !cache->end_seq) |
| 1728 | cache++; |
Baruch Even | fda03fb | 2007-02-04 23:35:57 -0800 | [diff] [blame] | 1729 | } |
| 1730 | |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1731 | while (i < used_sacks) { |
Ilpo Järvinen | fd6dad6 | 2007-11-15 19:49:47 -0800 | [diff] [blame] | 1732 | u32 start_seq = sp[i].start_seq; |
| 1733 | u32 end_seq = sp[i].end_seq; |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1734 | bool dup_sack = (found_dup_sack && (i == first_sack_index)); |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1735 | struct tcp_sack_block *next_dup = NULL; |
Ilpo Järvinen | e56d6cd | 2007-11-01 00:09:37 -0700 | [diff] [blame] | 1736 | |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1737 | if (found_dup_sack && ((i + 1) == first_sack_index)) |
| 1738 | next_dup = &sp[i + 1]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1739 | |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1740 | /* Skip too early cached blocks */ |
| 1741 | while (tcp_sack_cache_ok(tp, cache) && |
| 1742 | !before(start_seq, cache->end_seq)) |
| 1743 | cache++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1744 | |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1745 | /* Can skip some work by looking recv_sack_cache? */ |
| 1746 | if (tcp_sack_cache_ok(tp, cache) && !dup_sack && |
| 1747 | after(end_seq, cache->start_seq)) { |
David S. Miller | fe067e8 | 2007-03-07 12:12:44 -0800 | [diff] [blame] | 1748 | |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1749 | /* Head todo? */ |
| 1750 | if (before(start_seq, cache->start_seq)) { |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1751 | skb = tcp_sacktag_skip(skb, sk, &state, |
| 1752 | start_seq); |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 1753 | skb = tcp_sacktag_walk(skb, sk, next_dup, |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1754 | &state, |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 1755 | start_seq, |
| 1756 | cache->start_seq, |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1757 | dup_sack); |
Baruch Even | fda03fb | 2007-02-04 23:35:57 -0800 | [diff] [blame] | 1758 | } |
Stephen Hemminger | 6a438bb | 2005-11-10 17:14:59 -0800 | [diff] [blame] | 1759 | |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1760 | /* Rest of the block already fully processed? */ |
Ilpo Järvinen | 20de20b | 2007-11-16 16:17:05 -0800 | [diff] [blame] | 1761 | if (!after(end_seq, cache->end_seq)) |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1762 | goto advance_sp; |
Ilpo Järvinen | 20de20b | 2007-11-16 16:17:05 -0800 | [diff] [blame] | 1763 | |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 1764 | skb = tcp_maybe_skipping_dsack(skb, sk, next_dup, |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1765 | &state, |
| 1766 | cache->end_seq); |
Ilpo Järvinen | e56d6cd | 2007-11-01 00:09:37 -0700 | [diff] [blame] | 1767 | |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1768 | /* ...tail remains todo... */ |
Ilpo Järvinen | 6859d49 | 2007-12-02 00:48:06 +0200 | [diff] [blame] | 1769 | if (tcp_highest_sack_seq(tp) == cache->end_seq) { |
Ilpo Järvinen | 20de20b | 2007-11-16 16:17:05 -0800 | [diff] [blame] | 1770 | /* ...but better entrypoint exists! */ |
Ilpo Järvinen | 6859d49 | 2007-12-02 00:48:06 +0200 | [diff] [blame] | 1771 | skb = tcp_highest_sack(sk); |
| 1772 | if (skb == NULL) |
| 1773 | break; |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1774 | state.fack_count = tp->fackets_out; |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1775 | cache++; |
| 1776 | goto walk; |
| 1777 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1778 | |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1779 | skb = tcp_sacktag_skip(skb, sk, &state, cache->end_seq); |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1780 | /* Check overlap against next cached too (past this one already) */ |
| 1781 | cache++; |
| 1782 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1783 | } |
Ilpo Järvinen | fbd52eb | 2007-11-10 21:24:19 -0800 | [diff] [blame] | 1784 | |
Ilpo Järvinen | 6859d49 | 2007-12-02 00:48:06 +0200 | [diff] [blame] | 1785 | if (!before(start_seq, tcp_highest_sack_seq(tp))) { |
| 1786 | skb = tcp_highest_sack(sk); |
| 1787 | if (skb == NULL) |
| 1788 | break; |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1789 | state.fack_count = tp->fackets_out; |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1790 | } |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1791 | skb = tcp_sacktag_skip(skb, sk, &state, start_seq); |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1792 | |
| 1793 | walk: |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1794 | skb = tcp_sacktag_walk(skb, sk, next_dup, &state, |
| 1795 | start_seq, end_seq, dup_sack); |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1796 | |
| 1797 | advance_sp: |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1798 | i++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1799 | } |
| 1800 | |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1801 | /* Clear the head of the cache sack blocks so we can skip it next time */ |
| 1802 | for (i = 0; i < ARRAY_SIZE(tp->recv_sack_cache) - used_sacks; i++) { |
| 1803 | tp->recv_sack_cache[i].start_seq = 0; |
| 1804 | tp->recv_sack_cache[i].end_seq = 0; |
| 1805 | } |
| 1806 | for (j = 0; j < used_sacks; j++) |
| 1807 | tp->recv_sack_cache[i++] = sp[j]; |
| 1808 | |
Ilpo Järvinen | 407ef1d | 2007-12-02 00:47:57 +0200 | [diff] [blame] | 1809 | tcp_mark_lost_retrans(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1810 | |
Ilpo Järvinen | 86426c2 | 2007-08-09 14:45:17 +0300 | [diff] [blame] | 1811 | tcp_verify_left_out(tp); |
| 1812 | |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1813 | if ((state.reord < tp->fackets_out) && |
Yuchung Cheng | 9b44190 | 2013-03-20 13:32:58 +0000 | [diff] [blame] | 1814 | ((inet_csk(sk)->icsk_ca_state != TCP_CA_Loss) || tp->undo_marker)) |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1815 | tcp_update_reordering(sk, tp->fackets_out - state.reord, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1816 | |
Ilpo Järvinen | 96a2d41 | 2007-11-14 15:47:18 -0800 | [diff] [blame] | 1817 | out: |
| 1818 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1819 | #if FASTRETRANS_DEBUG > 0 |
Ilpo Järvinen | 547b792 | 2008-07-25 21:43:18 -0700 | [diff] [blame] | 1820 | WARN_ON((int)tp->sacked_out < 0); |
| 1821 | WARN_ON((int)tp->lost_out < 0); |
| 1822 | WARN_ON((int)tp->retrans_out < 0); |
| 1823 | WARN_ON((int)tcp_packets_in_flight(tp) < 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1824 | #endif |
Yuchung Cheng | 59c9af4 | 2013-07-22 16:20:47 -0700 | [diff] [blame] | 1825 | *sack_rtt = state.rtt; |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1826 | return state.flag; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1827 | } |
| 1828 | |
Ilpo Järvinen | 882beba | 2008-04-07 22:33:07 -0700 | [diff] [blame] | 1829 | /* Limits sacked_out so that sum with lost_out isn't ever larger than |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1830 | * packets_out. Returns false if sacked_out adjustement wasn't necessary. |
Ilpo Järvinen | 30935cf | 2007-02-21 23:01:36 -0800 | [diff] [blame] | 1831 | */ |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1832 | static bool tcp_limit_reno_sacked(struct tcp_sock *tp) |
Ilpo Järvinen | 4ddf667 | 2007-05-27 01:52:00 -0700 | [diff] [blame] | 1833 | { |
Ilpo Järvinen | 4ddf667 | 2007-05-27 01:52:00 -0700 | [diff] [blame] | 1834 | u32 holes; |
| 1835 | |
| 1836 | holes = max(tp->lost_out, 1U); |
| 1837 | holes = min(holes, tp->packets_out); |
| 1838 | |
| 1839 | if ((tp->sacked_out + holes) > tp->packets_out) { |
| 1840 | tp->sacked_out = tp->packets_out - holes; |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1841 | return true; |
Ilpo Järvinen | 4ddf667 | 2007-05-27 01:52:00 -0700 | [diff] [blame] | 1842 | } |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1843 | return false; |
Ilpo Järvinen | 882beba | 2008-04-07 22:33:07 -0700 | [diff] [blame] | 1844 | } |
| 1845 | |
| 1846 | /* If we receive more dupacks than we expected counting segments |
| 1847 | * in assumption of absent reordering, interpret this as reordering. |
| 1848 | * The only another reason could be bug in receiver TCP. |
| 1849 | */ |
| 1850 | static void tcp_check_reno_reordering(struct sock *sk, const int addend) |
| 1851 | { |
| 1852 | struct tcp_sock *tp = tcp_sk(sk); |
| 1853 | if (tcp_limit_reno_sacked(tp)) |
| 1854 | tcp_update_reordering(sk, tp->packets_out + addend, 0); |
Ilpo Järvinen | 4ddf667 | 2007-05-27 01:52:00 -0700 | [diff] [blame] | 1855 | } |
| 1856 | |
| 1857 | /* Emulate SACKs for SACKless connection: account for a new dupack. */ |
| 1858 | |
| 1859 | static void tcp_add_reno_sack(struct sock *sk) |
| 1860 | { |
| 1861 | struct tcp_sock *tp = tcp_sk(sk); |
| 1862 | tp->sacked_out++; |
| 1863 | tcp_check_reno_reordering(sk, 0); |
Ilpo Järvinen | 005903b | 2007-08-09 14:44:16 +0300 | [diff] [blame] | 1864 | tcp_verify_left_out(tp); |
Ilpo Järvinen | 4ddf667 | 2007-05-27 01:52:00 -0700 | [diff] [blame] | 1865 | } |
| 1866 | |
| 1867 | /* Account for ACK, ACKing some data in Reno Recovery phase. */ |
| 1868 | |
| 1869 | static void tcp_remove_reno_sacks(struct sock *sk, int acked) |
| 1870 | { |
| 1871 | struct tcp_sock *tp = tcp_sk(sk); |
| 1872 | |
| 1873 | if (acked > 0) { |
| 1874 | /* One ACK acked hole. The rest eat duplicate ACKs. */ |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 1875 | if (acked - 1 >= tp->sacked_out) |
Ilpo Järvinen | 4ddf667 | 2007-05-27 01:52:00 -0700 | [diff] [blame] | 1876 | tp->sacked_out = 0; |
| 1877 | else |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 1878 | tp->sacked_out -= acked - 1; |
Ilpo Järvinen | 4ddf667 | 2007-05-27 01:52:00 -0700 | [diff] [blame] | 1879 | } |
| 1880 | tcp_check_reno_reordering(sk, acked); |
Ilpo Järvinen | 005903b | 2007-08-09 14:44:16 +0300 | [diff] [blame] | 1881 | tcp_verify_left_out(tp); |
Ilpo Järvinen | 4ddf667 | 2007-05-27 01:52:00 -0700 | [diff] [blame] | 1882 | } |
| 1883 | |
| 1884 | static inline void tcp_reset_reno_sack(struct tcp_sock *tp) |
| 1885 | { |
| 1886 | tp->sacked_out = 0; |
Ilpo Järvinen | 4ddf667 | 2007-05-27 01:52:00 -0700 | [diff] [blame] | 1887 | } |
| 1888 | |
Ilpo Järvinen | 4cd8299 | 2007-10-11 17:34:57 -0700 | [diff] [blame] | 1889 | static void tcp_clear_retrans_partial(struct tcp_sock *tp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1890 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1891 | tp->retrans_out = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1892 | tp->lost_out = 0; |
| 1893 | |
| 1894 | tp->undo_marker = 0; |
| 1895 | tp->undo_retrans = 0; |
| 1896 | } |
| 1897 | |
Ilpo Järvinen | 4cd8299 | 2007-10-11 17:34:57 -0700 | [diff] [blame] | 1898 | void tcp_clear_retrans(struct tcp_sock *tp) |
| 1899 | { |
| 1900 | tcp_clear_retrans_partial(tp); |
| 1901 | |
| 1902 | tp->fackets_out = 0; |
| 1903 | tp->sacked_out = 0; |
| 1904 | } |
| 1905 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1906 | /* Enter Loss state. If "how" is not zero, forget all SACK information |
| 1907 | * and reset tags completely, otherwise preserve SACKs. If receiver |
| 1908 | * dropped its ofo queue, we will know this due to reneging detection. |
| 1909 | */ |
| 1910 | void tcp_enter_loss(struct sock *sk, int how) |
| 1911 | { |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 1912 | const struct inet_connection_sock *icsk = inet_csk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1913 | struct tcp_sock *tp = tcp_sk(sk); |
| 1914 | struct sk_buff *skb; |
Yuchung Cheng | e33099f | 2013-03-20 13:33:00 +0000 | [diff] [blame] | 1915 | bool new_recovery = false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1916 | |
| 1917 | /* Reduce ssthresh if it has not yet been made inside this window. */ |
Yuchung Cheng | e33099f | 2013-03-20 13:33:00 +0000 | [diff] [blame] | 1918 | if (icsk->icsk_ca_state <= TCP_CA_Disorder || |
| 1919 | !after(tp->high_seq, tp->snd_una) || |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 1920 | (icsk->icsk_ca_state == TCP_CA_Loss && !icsk->icsk_retransmits)) { |
Yuchung Cheng | e33099f | 2013-03-20 13:33:00 +0000 | [diff] [blame] | 1921 | new_recovery = true; |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 1922 | tp->prior_ssthresh = tcp_current_ssthresh(sk); |
| 1923 | tp->snd_ssthresh = icsk->icsk_ca_ops->ssthresh(sk); |
| 1924 | tcp_ca_event(sk, CA_EVENT_LOSS); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1925 | } |
| 1926 | tp->snd_cwnd = 1; |
| 1927 | tp->snd_cwnd_cnt = 0; |
| 1928 | tp->snd_cwnd_stamp = tcp_time_stamp; |
| 1929 | |
Ilpo Järvinen | 4cd8299 | 2007-10-11 17:34:57 -0700 | [diff] [blame] | 1930 | tcp_clear_retrans_partial(tp); |
| 1931 | |
| 1932 | if (tcp_is_reno(tp)) |
| 1933 | tcp_reset_reno_sack(tp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1934 | |
Yuchung Cheng | 7ebe183 | 2013-03-24 10:42:25 +0000 | [diff] [blame] | 1935 | tp->undo_marker = tp->snd_una; |
| 1936 | if (how) { |
Ilpo Järvinen | 4cd8299 | 2007-10-11 17:34:57 -0700 | [diff] [blame] | 1937 | tp->sacked_out = 0; |
| 1938 | tp->fackets_out = 0; |
Ilpo Järvinen | b768920 | 2007-09-20 11:37:19 -0700 | [diff] [blame] | 1939 | } |
Ilpo Järvinen | 64edc27 | 2008-09-20 21:18:32 -0700 | [diff] [blame] | 1940 | tcp_clear_all_retrans_hints(tp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1941 | |
David S. Miller | fe067e8 | 2007-03-07 12:12:44 -0800 | [diff] [blame] | 1942 | tcp_for_write_queue(skb, sk) { |
| 1943 | if (skb == tcp_send_head(sk)) |
| 1944 | break; |
Ilpo Järvinen | 4cd8299 | 2007-10-11 17:34:57 -0700 | [diff] [blame] | 1945 | |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 1946 | if (TCP_SKB_CB(skb)->sacked & TCPCB_RETRANS) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1947 | tp->undo_marker = 0; |
| 1948 | TCP_SKB_CB(skb)->sacked &= (~TCPCB_TAGBITS)|TCPCB_SACKED_ACKED; |
| 1949 | if (!(TCP_SKB_CB(skb)->sacked&TCPCB_SACKED_ACKED) || how) { |
| 1950 | TCP_SKB_CB(skb)->sacked &= ~TCPCB_SACKED_ACKED; |
| 1951 | TCP_SKB_CB(skb)->sacked |= TCPCB_LOST; |
| 1952 | tp->lost_out += tcp_skb_pcount(skb); |
Ilpo Järvinen | 006f582 | 2008-09-20 21:20:20 -0700 | [diff] [blame] | 1953 | tp->retransmit_high = TCP_SKB_CB(skb)->end_seq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1954 | } |
| 1955 | } |
Ilpo Järvinen | 005903b | 2007-08-09 14:44:16 +0300 | [diff] [blame] | 1956 | tcp_verify_left_out(tp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1957 | |
Yuchung Cheng | 74c181d | 2013-08-12 16:41:25 -0700 | [diff] [blame] | 1958 | /* Timeout in disordered state after receiving substantial DUPACKs |
| 1959 | * suggests that the degree of reordering is over-estimated. |
| 1960 | */ |
| 1961 | if (icsk->icsk_ca_state <= TCP_CA_Disorder && |
| 1962 | tp->sacked_out >= sysctl_tcp_reordering) |
| 1963 | tp->reordering = min_t(unsigned int, tp->reordering, |
| 1964 | sysctl_tcp_reordering); |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 1965 | tcp_set_ca_state(sk, TCP_CA_Loss); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1966 | tp->high_seq = tp->snd_nxt; |
| 1967 | TCP_ECN_queue_cwr(tp); |
Yuchung Cheng | e33099f | 2013-03-20 13:33:00 +0000 | [diff] [blame] | 1968 | |
| 1969 | /* F-RTO RFC5682 sec 3.1 step 1: retransmit SND.UNA if no previous |
| 1970 | * loss recovery is underway except recurring timeout(s) on |
| 1971 | * the same SND.UNA (sec 3.2). Disable F-RTO on path MTU probing |
| 1972 | */ |
| 1973 | tp->frto = sysctl_tcp_frto && |
| 1974 | (new_recovery || icsk->icsk_retransmits) && |
| 1975 | !inet_csk(sk)->icsk_mtup.probe_size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1976 | } |
| 1977 | |
Ilpo Järvinen | cadbd03 | 2007-12-31 04:49:21 -0800 | [diff] [blame] | 1978 | /* If ACK arrived pointing to a remembered SACK, it means that our |
| 1979 | * remembered SACKs do not reflect real state of receiver i.e. |
| 1980 | * receiver _host_ is heavily congested (or buggy). |
| 1981 | * |
| 1982 | * Do processing similar to RTO timeout. |
| 1983 | */ |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1984 | static bool tcp_check_sack_reneging(struct sock *sk, int flag) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1985 | { |
Ilpo Järvinen | cadbd03 | 2007-12-31 04:49:21 -0800 | [diff] [blame] | 1986 | if (flag & FLAG_SACK_RENEGING) { |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 1987 | struct inet_connection_sock *icsk = inet_csk(sk); |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 1988 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPSACKRENEGING); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1989 | |
| 1990 | tcp_enter_loss(sk, 1); |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 1991 | icsk->icsk_retransmits++; |
David S. Miller | fe067e8 | 2007-03-07 12:12:44 -0800 | [diff] [blame] | 1992 | tcp_retransmit_skb(sk, tcp_write_queue_head(sk)); |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 1993 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 1994 | icsk->icsk_rto, TCP_RTO_MAX); |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1995 | return true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1996 | } |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1997 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1998 | } |
| 1999 | |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 2000 | static inline int tcp_fackets_out(const struct tcp_sock *tp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2001 | { |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 2002 | return tcp_is_reno(tp) ? tp->sacked_out + 1 : tp->fackets_out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2003 | } |
| 2004 | |
Ilpo Järvinen | 85cc391 | 2007-11-15 19:39:31 -0800 | [diff] [blame] | 2005 | /* Heurestics to calculate number of duplicate ACKs. There's no dupACKs |
| 2006 | * counter when SACK is enabled (without SACK, sacked_out is used for |
| 2007 | * that purpose). |
| 2008 | * |
| 2009 | * Instead, with FACK TCP uses fackets_out that includes both SACKed |
| 2010 | * segments up to the highest received SACK block so far and holes in |
| 2011 | * between them. |
| 2012 | * |
| 2013 | * With reordering, holes may still be in flight, so RFC3517 recovery |
| 2014 | * uses pure sacked_out (total number of SACKed segments) even though |
| 2015 | * it violates the RFC that uses duplicate ACKs, often these are equal |
| 2016 | * but when e.g. out-of-window ACKs or packet duplication occurs, |
| 2017 | * they differ. Since neither occurs due to loss, TCP should really |
| 2018 | * ignore them. |
| 2019 | */ |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 2020 | static inline int tcp_dupack_heuristics(const struct tcp_sock *tp) |
Ilpo Järvinen | 85cc391 | 2007-11-15 19:39:31 -0800 | [diff] [blame] | 2021 | { |
| 2022 | return tcp_is_fack(tp) ? tp->fackets_out : tp->sacked_out + 1; |
| 2023 | } |
| 2024 | |
Yuchung Cheng | 750ea2b | 2012-05-02 13:30:04 +0000 | [diff] [blame] | 2025 | static bool tcp_pause_early_retransmit(struct sock *sk, int flag) |
| 2026 | { |
| 2027 | struct tcp_sock *tp = tcp_sk(sk); |
| 2028 | unsigned long delay; |
| 2029 | |
| 2030 | /* Delay early retransmit and entering fast recovery for |
| 2031 | * max(RTT/4, 2msec) unless ack has ECE mark, no RTT samples |
| 2032 | * available, or RTO is scheduled to fire first. |
| 2033 | */ |
Nandita Dukkipati | 6ba8a3b | 2013-03-11 10:00:43 +0000 | [diff] [blame] | 2034 | if (sysctl_tcp_early_retrans < 2 || sysctl_tcp_early_retrans > 3 || |
| 2035 | (flag & FLAG_ECE) || !tp->srtt) |
Yuchung Cheng | 750ea2b | 2012-05-02 13:30:04 +0000 | [diff] [blame] | 2036 | return false; |
| 2037 | |
| 2038 | delay = max_t(unsigned long, (tp->srtt >> 5), msecs_to_jiffies(2)); |
| 2039 | if (!time_after(inet_csk(sk)->icsk_timeout, (jiffies + delay))) |
| 2040 | return false; |
| 2041 | |
Nandita Dukkipati | 6ba8a3b | 2013-03-11 10:00:43 +0000 | [diff] [blame] | 2042 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_EARLY_RETRANS, delay, |
| 2043 | TCP_RTO_MAX); |
Yuchung Cheng | 750ea2b | 2012-05-02 13:30:04 +0000 | [diff] [blame] | 2044 | return true; |
| 2045 | } |
| 2046 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2047 | /* Linux NewReno/SACK/FACK/ECN state machine. |
| 2048 | * -------------------------------------- |
| 2049 | * |
| 2050 | * "Open" Normal state, no dubious events, fast path. |
| 2051 | * "Disorder" In all the respects it is "Open", |
| 2052 | * but requires a bit more attention. It is entered when |
| 2053 | * we see some SACKs or dupacks. It is split of "Open" |
| 2054 | * mainly to move some processing from fast path to slow one. |
| 2055 | * "CWR" CWND was reduced due to some Congestion Notification event. |
| 2056 | * It can be ECN, ICMP source quench, local device congestion. |
| 2057 | * "Recovery" CWND was reduced, we are fast-retransmitting. |
| 2058 | * "Loss" CWND was reduced due to RTO timeout or SACK reneging. |
| 2059 | * |
| 2060 | * tcp_fastretrans_alert() is entered: |
| 2061 | * - each incoming ACK, if state is not "Open" |
| 2062 | * - when arrived ACK is unusual, namely: |
| 2063 | * * SACK |
| 2064 | * * Duplicate ACK. |
| 2065 | * * ECN ECE. |
| 2066 | * |
| 2067 | * Counting packets in flight is pretty simple. |
| 2068 | * |
| 2069 | * in_flight = packets_out - left_out + retrans_out |
| 2070 | * |
| 2071 | * packets_out is SND.NXT-SND.UNA counted in packets. |
| 2072 | * |
| 2073 | * retrans_out is number of retransmitted segments. |
| 2074 | * |
| 2075 | * left_out is number of segments left network, but not ACKed yet. |
| 2076 | * |
| 2077 | * left_out = sacked_out + lost_out |
| 2078 | * |
| 2079 | * sacked_out: Packets, which arrived to receiver out of order |
| 2080 | * and hence not ACKed. With SACKs this number is simply |
| 2081 | * amount of SACKed data. Even without SACKs |
| 2082 | * it is easy to give pretty reliable estimate of this number, |
| 2083 | * counting duplicate ACKs. |
| 2084 | * |
| 2085 | * lost_out: Packets lost by network. TCP has no explicit |
| 2086 | * "loss notification" feedback from network (for now). |
| 2087 | * It means that this number can be only _guessed_. |
| 2088 | * Actually, it is the heuristics to predict lossage that |
| 2089 | * distinguishes different algorithms. |
| 2090 | * |
| 2091 | * F.e. after RTO, when all the queue is considered as lost, |
| 2092 | * lost_out = packets_out and in_flight = retrans_out. |
| 2093 | * |
| 2094 | * Essentially, we have now two algorithms counting |
| 2095 | * lost packets. |
| 2096 | * |
| 2097 | * FACK: It is the simplest heuristics. As soon as we decided |
| 2098 | * that something is lost, we decide that _all_ not SACKed |
| 2099 | * packets until the most forward SACK are lost. I.e. |
| 2100 | * lost_out = fackets_out - sacked_out and left_out = fackets_out. |
| 2101 | * It is absolutely correct estimate, if network does not reorder |
| 2102 | * packets. And it loses any connection to reality when reordering |
| 2103 | * takes place. We use FACK by default until reordering |
| 2104 | * is suspected on the path to this destination. |
| 2105 | * |
| 2106 | * NewReno: when Recovery is entered, we assume that one segment |
| 2107 | * is lost (classic Reno). While we are in Recovery and |
| 2108 | * a partial ACK arrives, we assume that one more packet |
| 2109 | * is lost (NewReno). This heuristics are the same in NewReno |
| 2110 | * and SACK. |
| 2111 | * |
| 2112 | * Imagine, that's all! Forget about all this shamanism about CWND inflation |
| 2113 | * deflation etc. CWND is real congestion window, never inflated, changes |
| 2114 | * only according to classic VJ rules. |
| 2115 | * |
| 2116 | * Really tricky (and requiring careful tuning) part of algorithm |
| 2117 | * is hidden in functions tcp_time_to_recover() and tcp_xmit_retransmit_queue(). |
| 2118 | * The first determines the moment _when_ we should reduce CWND and, |
| 2119 | * hence, slow down forward transmission. In fact, it determines the moment |
| 2120 | * when we decide that hole is caused by loss, rather than by a reorder. |
| 2121 | * |
| 2122 | * tcp_xmit_retransmit_queue() decides, _what_ we should retransmit to fill |
| 2123 | * holes, caused by lost packets. |
| 2124 | * |
| 2125 | * And the most logically complicated part of algorithm is undo |
| 2126 | * heuristics. We detect false retransmits due to both too early |
| 2127 | * fast retransmit (reordering) and underestimated RTO, analyzing |
| 2128 | * timestamps and D-SACKs. When we detect that some segments were |
| 2129 | * retransmitted by mistake and CWND reduction was wrong, we undo |
| 2130 | * window reduction and abort recovery phase. This logic is hidden |
| 2131 | * inside several functions named tcp_try_undo_<something>. |
| 2132 | */ |
| 2133 | |
| 2134 | /* This function decides, when we should leave Disordered state |
| 2135 | * and enter Recovery phase, reducing congestion window. |
| 2136 | * |
| 2137 | * Main question: may we further continue forward transmission |
| 2138 | * with the same cwnd? |
| 2139 | */ |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 2140 | static bool tcp_time_to_recover(struct sock *sk, int flag) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2141 | { |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 2142 | struct tcp_sock *tp = tcp_sk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2143 | __u32 packets_out; |
| 2144 | |
| 2145 | /* Trick#1: The loss is proven. */ |
| 2146 | if (tp->lost_out) |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 2147 | return true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2148 | |
| 2149 | /* Not-A-Trick#2 : Classic rule... */ |
Andreas Petlund | ea84e55 | 2009-10-27 03:27:21 +0000 | [diff] [blame] | 2150 | if (tcp_dupack_heuristics(tp) > tp->reordering) |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 2151 | return true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2152 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2153 | /* Trick#4: It is still not OK... But will it be useful to delay |
| 2154 | * recovery more? |
| 2155 | */ |
| 2156 | packets_out = tp->packets_out; |
| 2157 | if (packets_out <= tp->reordering && |
| 2158 | tp->sacked_out >= max_t(__u32, packets_out/2, sysctl_tcp_reordering) && |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 2159 | !tcp_may_send_now(sk)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2160 | /* We have nothing to send. This connection is limited |
| 2161 | * either by receiver window or by application. |
| 2162 | */ |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 2163 | return true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2164 | } |
| 2165 | |
Andreas Petlund | 7e38017 | 2010-02-18 04:48:19 +0000 | [diff] [blame] | 2166 | /* If a thin stream is detected, retransmit after first |
| 2167 | * received dupack. Employ only if SACK is supported in order |
| 2168 | * to avoid possible corner-case series of spurious retransmissions |
| 2169 | * Use only if there are no unsent data. |
| 2170 | */ |
| 2171 | if ((tp->thin_dupack || sysctl_tcp_thin_dupack) && |
| 2172 | tcp_stream_is_thin(tp) && tcp_dupack_heuristics(tp) > 1 && |
| 2173 | tcp_is_sack(tp) && !tcp_send_head(sk)) |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 2174 | return true; |
Andreas Petlund | 7e38017 | 2010-02-18 04:48:19 +0000 | [diff] [blame] | 2175 | |
Yuchung Cheng | eed530b | 2012-05-02 13:30:03 +0000 | [diff] [blame] | 2176 | /* Trick#6: TCP early retransmit, per RFC5827. To avoid spurious |
| 2177 | * retransmissions due to small network reorderings, we implement |
| 2178 | * Mitigation A.3 in the RFC and delay the retransmission for a short |
| 2179 | * interval if appropriate. |
| 2180 | */ |
| 2181 | if (tp->do_early_retrans && !tp->retrans_out && tp->sacked_out && |
Nandita Dukkipati | 6ba8a3b | 2013-03-11 10:00:43 +0000 | [diff] [blame] | 2182 | (tp->packets_out >= (tp->sacked_out + 1) && tp->packets_out < 4) && |
Yuchung Cheng | eed530b | 2012-05-02 13:30:03 +0000 | [diff] [blame] | 2183 | !tcp_may_send_now(sk)) |
Yuchung Cheng | 750ea2b | 2012-05-02 13:30:04 +0000 | [diff] [blame] | 2184 | return !tcp_pause_early_retransmit(sk, flag); |
Yuchung Cheng | eed530b | 2012-05-02 13:30:03 +0000 | [diff] [blame] | 2185 | |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 2186 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2187 | } |
| 2188 | |
Yuchung Cheng | 974c123 | 2012-01-19 14:42:21 +0000 | [diff] [blame] | 2189 | /* Detect loss in event "A" above by marking head of queue up as lost. |
| 2190 | * For FACK or non-SACK(Reno) senders, the first "packets" number of segments |
| 2191 | * are considered lost. For RFC3517 SACK, a segment is considered lost if it |
| 2192 | * has at least tp->reordering SACKed seqments above it; "packets" refers to |
| 2193 | * the maximum SACKed segments to pass before reaching this limit. |
Ilpo Järvinen | 85cc391 | 2007-11-15 19:39:31 -0800 | [diff] [blame] | 2194 | */ |
Ilpo Järvinen | 1fdb936 | 2010-10-14 01:42:30 +0000 | [diff] [blame] | 2195 | static void tcp_mark_head_lost(struct sock *sk, int packets, int mark_head) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2196 | { |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 2197 | struct tcp_sock *tp = tcp_sk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2198 | struct sk_buff *skb; |
Ilpo Järvinen | c137f3d | 2008-04-07 22:32:38 -0700 | [diff] [blame] | 2199 | int cnt, oldcnt; |
| 2200 | int err; |
| 2201 | unsigned int mss; |
Yuchung Cheng | 974c123 | 2012-01-19 14:42:21 +0000 | [diff] [blame] | 2202 | /* Use SACK to deduce losses of new sequences sent during recovery */ |
| 2203 | const u32 loss_high = tcp_is_sack(tp) ? tp->snd_nxt : tp->high_seq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2204 | |
Ilpo Järvinen | 547b792 | 2008-07-25 21:43:18 -0700 | [diff] [blame] | 2205 | WARN_ON(packets > tp->packets_out); |
Stephen Hemminger | 6a438bb | 2005-11-10 17:14:59 -0800 | [diff] [blame] | 2206 | if (tp->lost_skb_hint) { |
| 2207 | skb = tp->lost_skb_hint; |
| 2208 | cnt = tp->lost_cnt_hint; |
Ilpo Järvinen | 1fdb936 | 2010-10-14 01:42:30 +0000 | [diff] [blame] | 2209 | /* Head already handled? */ |
| 2210 | if (mark_head && skb != tcp_write_queue_head(sk)) |
| 2211 | return; |
Stephen Hemminger | 6a438bb | 2005-11-10 17:14:59 -0800 | [diff] [blame] | 2212 | } else { |
David S. Miller | fe067e8 | 2007-03-07 12:12:44 -0800 | [diff] [blame] | 2213 | skb = tcp_write_queue_head(sk); |
Stephen Hemminger | 6a438bb | 2005-11-10 17:14:59 -0800 | [diff] [blame] | 2214 | cnt = 0; |
| 2215 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2216 | |
David S. Miller | fe067e8 | 2007-03-07 12:12:44 -0800 | [diff] [blame] | 2217 | tcp_for_write_queue_from(skb, sk) { |
| 2218 | if (skb == tcp_send_head(sk)) |
| 2219 | break; |
Stephen Hemminger | 6a438bb | 2005-11-10 17:14:59 -0800 | [diff] [blame] | 2220 | /* TODO: do this better */ |
| 2221 | /* this is not the most efficient way to do this... */ |
| 2222 | tp->lost_skb_hint = skb; |
| 2223 | tp->lost_cnt_hint = cnt; |
Ilpo Järvinen | 85cc391 | 2007-11-15 19:39:31 -0800 | [diff] [blame] | 2224 | |
Yuchung Cheng | 974c123 | 2012-01-19 14:42:21 +0000 | [diff] [blame] | 2225 | if (after(TCP_SKB_CB(skb)->end_seq, loss_high)) |
Ilpo Järvinen | c137f3d | 2008-04-07 22:32:38 -0700 | [diff] [blame] | 2226 | break; |
| 2227 | |
| 2228 | oldcnt = cnt; |
Ilpo Järvinen | ad1984e | 2008-01-30 20:06:02 -0800 | [diff] [blame] | 2229 | if (tcp_is_fack(tp) || tcp_is_reno(tp) || |
Ilpo Järvinen | 85cc391 | 2007-11-15 19:39:31 -0800 | [diff] [blame] | 2230 | (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)) |
| 2231 | cnt += tcp_skb_pcount(skb); |
| 2232 | |
Ilpo Järvinen | c137f3d | 2008-04-07 22:32:38 -0700 | [diff] [blame] | 2233 | if (cnt > packets) { |
Yuchung Cheng | b3de755 | 2010-09-24 13:22:06 +0000 | [diff] [blame] | 2234 | if ((tcp_is_sack(tp) && !tcp_is_fack(tp)) || |
Neal Cardwell | c0638c2 | 2012-03-02 21:36:51 +0000 | [diff] [blame] | 2235 | (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED) || |
Yuchung Cheng | b3de755 | 2010-09-24 13:22:06 +0000 | [diff] [blame] | 2236 | (oldcnt >= packets)) |
Ilpo Järvinen | c137f3d | 2008-04-07 22:32:38 -0700 | [diff] [blame] | 2237 | break; |
| 2238 | |
| 2239 | mss = skb_shinfo(skb)->gso_size; |
| 2240 | err = tcp_fragment(sk, skb, (packets - oldcnt) * mss, mss); |
| 2241 | if (err < 0) |
| 2242 | break; |
| 2243 | cnt = packets; |
| 2244 | } |
| 2245 | |
Ilpo Järvinen | 41ea36e | 2008-09-20 21:19:22 -0700 | [diff] [blame] | 2246 | tcp_skb_mark_lost(tp, skb); |
Ilpo Järvinen | 1fdb936 | 2010-10-14 01:42:30 +0000 | [diff] [blame] | 2247 | |
| 2248 | if (mark_head) |
| 2249 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2250 | } |
Ilpo Järvinen | 005903b | 2007-08-09 14:44:16 +0300 | [diff] [blame] | 2251 | tcp_verify_left_out(tp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2252 | } |
| 2253 | |
| 2254 | /* Account newly detected lost packet(s) */ |
| 2255 | |
Ilpo Järvinen | 85cc391 | 2007-11-15 19:39:31 -0800 | [diff] [blame] | 2256 | static void tcp_update_scoreboard(struct sock *sk, int fast_rexmit) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2257 | { |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 2258 | struct tcp_sock *tp = tcp_sk(sk); |
| 2259 | |
Ilpo Järvinen | 85cc391 | 2007-11-15 19:39:31 -0800 | [diff] [blame] | 2260 | if (tcp_is_reno(tp)) { |
Ilpo Järvinen | 1fdb936 | 2010-10-14 01:42:30 +0000 | [diff] [blame] | 2261 | tcp_mark_head_lost(sk, 1, 1); |
Ilpo Järvinen | 85cc391 | 2007-11-15 19:39:31 -0800 | [diff] [blame] | 2262 | } else if (tcp_is_fack(tp)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2263 | int lost = tp->fackets_out - tp->reordering; |
| 2264 | if (lost <= 0) |
| 2265 | lost = 1; |
Ilpo Järvinen | 1fdb936 | 2010-10-14 01:42:30 +0000 | [diff] [blame] | 2266 | tcp_mark_head_lost(sk, lost, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2267 | } else { |
Ilpo Järvinen | 85cc391 | 2007-11-15 19:39:31 -0800 | [diff] [blame] | 2268 | int sacked_upto = tp->sacked_out - tp->reordering; |
Ilpo Järvinen | 1fdb936 | 2010-10-14 01:42:30 +0000 | [diff] [blame] | 2269 | if (sacked_upto >= 0) |
| 2270 | tcp_mark_head_lost(sk, sacked_upto, 0); |
| 2271 | else if (fast_rexmit) |
| 2272 | tcp_mark_head_lost(sk, 1, 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2273 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2274 | } |
| 2275 | |
| 2276 | /* CWND moderation, preventing bursts due to too big ACKs |
| 2277 | * in dubious situations. |
| 2278 | */ |
| 2279 | static inline void tcp_moderate_cwnd(struct tcp_sock *tp) |
| 2280 | { |
| 2281 | tp->snd_cwnd = min(tp->snd_cwnd, |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 2282 | tcp_packets_in_flight(tp) + tcp_max_burst(tp)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2283 | tp->snd_cwnd_stamp = tcp_time_stamp; |
| 2284 | } |
| 2285 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2286 | /* Nothing was retransmitted or returned timestamp is less |
| 2287 | * than timestamp of the first retransmission. |
| 2288 | */ |
Vijay Subramanian | 67b95bd | 2012-07-19 21:32:18 +0000 | [diff] [blame] | 2289 | static inline bool tcp_packet_delayed(const struct tcp_sock *tp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2290 | { |
| 2291 | return !tp->retrans_stamp || |
| 2292 | (tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr && |
Arnd Hannemann | d7ee147 | 2008-04-21 14:46:22 -0700 | [diff] [blame] | 2293 | before(tp->rx_opt.rcv_tsecr, tp->retrans_stamp)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2294 | } |
| 2295 | |
| 2296 | /* Undo procedures. */ |
| 2297 | |
| 2298 | #if FASTRETRANS_DEBUG > 1 |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 2299 | static void DBGUNDO(struct sock *sk, const char *msg) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2300 | { |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 2301 | struct tcp_sock *tp = tcp_sk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2302 | struct inet_sock *inet = inet_sk(sk); |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 2303 | |
YOSHIFUJI Hideaki | 569508c | 2008-04-14 04:09:36 -0700 | [diff] [blame] | 2304 | if (sk->sk_family == AF_INET) { |
Joe Perches | 91df42b | 2012-05-15 14:11:54 +0000 | [diff] [blame] | 2305 | pr_debug("Undo %s %pI4/%u c%u l%u ss%u/%u p%u\n", |
| 2306 | msg, |
| 2307 | &inet->inet_daddr, ntohs(inet->inet_dport), |
| 2308 | tp->snd_cwnd, tcp_left_out(tp), |
| 2309 | tp->snd_ssthresh, tp->prior_ssthresh, |
| 2310 | tp->packets_out); |
YOSHIFUJI Hideaki | 569508c | 2008-04-14 04:09:36 -0700 | [diff] [blame] | 2311 | } |
Eric Dumazet | dfd56b8 | 2011-12-10 09:48:31 +0000 | [diff] [blame] | 2312 | #if IS_ENABLED(CONFIG_IPV6) |
YOSHIFUJI Hideaki | 569508c | 2008-04-14 04:09:36 -0700 | [diff] [blame] | 2313 | else if (sk->sk_family == AF_INET6) { |
| 2314 | struct ipv6_pinfo *np = inet6_sk(sk); |
Joe Perches | 91df42b | 2012-05-15 14:11:54 +0000 | [diff] [blame] | 2315 | pr_debug("Undo %s %pI6/%u c%u l%u ss%u/%u p%u\n", |
| 2316 | msg, |
| 2317 | &np->daddr, ntohs(inet->inet_dport), |
| 2318 | tp->snd_cwnd, tcp_left_out(tp), |
| 2319 | tp->snd_ssthresh, tp->prior_ssthresh, |
| 2320 | tp->packets_out); |
YOSHIFUJI Hideaki | 569508c | 2008-04-14 04:09:36 -0700 | [diff] [blame] | 2321 | } |
| 2322 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2323 | } |
| 2324 | #else |
| 2325 | #define DBGUNDO(x...) do { } while (0) |
| 2326 | #endif |
| 2327 | |
Yuchung Cheng | 7026b91 | 2013-05-29 14:20:13 +0000 | [diff] [blame] | 2328 | static void tcp_undo_cwnd_reduction(struct sock *sk, bool unmark_loss) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2329 | { |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 2330 | struct tcp_sock *tp = tcp_sk(sk); |
| 2331 | |
Yuchung Cheng | 6a63df4 | 2013-05-29 14:20:12 +0000 | [diff] [blame] | 2332 | if (unmark_loss) { |
| 2333 | struct sk_buff *skb; |
| 2334 | |
| 2335 | tcp_for_write_queue(skb, sk) { |
| 2336 | if (skb == tcp_send_head(sk)) |
| 2337 | break; |
| 2338 | TCP_SKB_CB(skb)->sacked &= ~TCPCB_LOST; |
| 2339 | } |
| 2340 | tp->lost_out = 0; |
| 2341 | tcp_clear_all_retrans_hints(tp); |
| 2342 | } |
| 2343 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2344 | if (tp->prior_ssthresh) { |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 2345 | const struct inet_connection_sock *icsk = inet_csk(sk); |
| 2346 | |
| 2347 | if (icsk->icsk_ca_ops->undo_cwnd) |
| 2348 | tp->snd_cwnd = icsk->icsk_ca_ops->undo_cwnd(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2349 | else |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 2350 | tp->snd_cwnd = max(tp->snd_cwnd, tp->snd_ssthresh << 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2351 | |
Yuchung Cheng | 7026b91 | 2013-05-29 14:20:13 +0000 | [diff] [blame] | 2352 | if (tp->prior_ssthresh > tp->snd_ssthresh) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2353 | tp->snd_ssthresh = tp->prior_ssthresh; |
| 2354 | TCP_ECN_withdraw_cwr(tp); |
| 2355 | } |
| 2356 | } else { |
| 2357 | tp->snd_cwnd = max(tp->snd_cwnd, tp->snd_ssthresh); |
| 2358 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2359 | tp->snd_cwnd_stamp = tcp_time_stamp; |
Yuchung Cheng | 7026b91 | 2013-05-29 14:20:13 +0000 | [diff] [blame] | 2360 | tp->undo_marker = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2361 | } |
| 2362 | |
Vijay Subramanian | 67b95bd | 2012-07-19 21:32:18 +0000 | [diff] [blame] | 2363 | static inline bool tcp_may_undo(const struct tcp_sock *tp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2364 | { |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 2365 | return tp->undo_marker && (!tp->undo_retrans || tcp_packet_delayed(tp)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2366 | } |
| 2367 | |
| 2368 | /* People celebrate: "We love our President!" */ |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 2369 | static bool tcp_try_undo_recovery(struct sock *sk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2370 | { |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 2371 | struct tcp_sock *tp = tcp_sk(sk); |
| 2372 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2373 | if (tcp_may_undo(tp)) { |
Pavel Emelyanov | 40b215e | 2008-07-03 01:05:41 -0700 | [diff] [blame] | 2374 | int mib_idx; |
| 2375 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2376 | /* Happy end! We did not retransmit anything |
| 2377 | * or our original transmission succeeded. |
| 2378 | */ |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 2379 | DBGUNDO(sk, inet_csk(sk)->icsk_ca_state == TCP_CA_Loss ? "loss" : "retrans"); |
Yuchung Cheng | 7026b91 | 2013-05-29 14:20:13 +0000 | [diff] [blame] | 2380 | tcp_undo_cwnd_reduction(sk, false); |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 2381 | if (inet_csk(sk)->icsk_ca_state == TCP_CA_Loss) |
Pavel Emelyanov | 40b215e | 2008-07-03 01:05:41 -0700 | [diff] [blame] | 2382 | mib_idx = LINUX_MIB_TCPLOSSUNDO; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2383 | else |
Pavel Emelyanov | 40b215e | 2008-07-03 01:05:41 -0700 | [diff] [blame] | 2384 | mib_idx = LINUX_MIB_TCPFULLUNDO; |
| 2385 | |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 2386 | NET_INC_STATS_BH(sock_net(sk), mib_idx); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2387 | } |
Ilpo Järvinen | e60402d | 2007-08-09 15:14:46 +0300 | [diff] [blame] | 2388 | if (tp->snd_una == tp->high_seq && tcp_is_reno(tp)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2389 | /* Hold old state until something *above* high_seq |
| 2390 | * is ACKed. For Reno it is MUST to prevent false |
| 2391 | * fast retransmits (RFC2582). SACK TCP is safe. */ |
| 2392 | tcp_moderate_cwnd(tp); |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 2393 | return true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2394 | } |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 2395 | tcp_set_ca_state(sk, TCP_CA_Open); |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 2396 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2397 | } |
| 2398 | |
| 2399 | /* Try to undo cwnd reduction, because D-SACKs acked all retransmitted data */ |
Yuchung Cheng | c7d9d6a | 2013-05-29 14:20:14 +0000 | [diff] [blame] | 2400 | static bool tcp_try_undo_dsack(struct sock *sk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2401 | { |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 2402 | struct tcp_sock *tp = tcp_sk(sk); |
| 2403 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2404 | if (tp->undo_marker && !tp->undo_retrans) { |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 2405 | DBGUNDO(sk, "D-SACK"); |
Yuchung Cheng | 7026b91 | 2013-05-29 14:20:13 +0000 | [diff] [blame] | 2406 | tcp_undo_cwnd_reduction(sk, false); |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 2407 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPDSACKUNDO); |
Yuchung Cheng | c7d9d6a | 2013-05-29 14:20:14 +0000 | [diff] [blame] | 2408 | return true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2409 | } |
Yuchung Cheng | c7d9d6a | 2013-05-29 14:20:14 +0000 | [diff] [blame] | 2410 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2411 | } |
| 2412 | |
Ilpo Järvinen | 77722b1 | 2009-12-08 20:54:11 -0800 | [diff] [blame] | 2413 | /* We can clear retrans_stamp when there are no retransmissions in the |
| 2414 | * window. It would seem that it is trivially available for us in |
| 2415 | * tp->retrans_out, however, that kind of assumptions doesn't consider |
| 2416 | * what will happen if errors occur when sending retransmission for the |
| 2417 | * second time. ...It could the that such segment has only |
| 2418 | * TCPCB_EVER_RETRANS set at the present time. It seems that checking |
| 2419 | * the head skb is enough except for some reneging corner cases that |
| 2420 | * are not worth the effort. |
| 2421 | * |
| 2422 | * Main reason for all this complexity is the fact that connection dying |
| 2423 | * time now depends on the validity of the retrans_stamp, in particular, |
| 2424 | * that successive retransmissions of a segment must not advance |
| 2425 | * retrans_stamp under any conditions. |
| 2426 | */ |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 2427 | static bool tcp_any_retrans_done(const struct sock *sk) |
Ilpo Järvinen | 77722b1 | 2009-12-08 20:54:11 -0800 | [diff] [blame] | 2428 | { |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 2429 | const struct tcp_sock *tp = tcp_sk(sk); |
Ilpo Järvinen | 77722b1 | 2009-12-08 20:54:11 -0800 | [diff] [blame] | 2430 | struct sk_buff *skb; |
| 2431 | |
| 2432 | if (tp->retrans_out) |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 2433 | return true; |
Ilpo Järvinen | 77722b1 | 2009-12-08 20:54:11 -0800 | [diff] [blame] | 2434 | |
| 2435 | skb = tcp_write_queue_head(sk); |
| 2436 | if (unlikely(skb && TCP_SKB_CB(skb)->sacked & TCPCB_EVER_RETRANS)) |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 2437 | return true; |
Ilpo Järvinen | 77722b1 | 2009-12-08 20:54:11 -0800 | [diff] [blame] | 2438 | |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 2439 | return false; |
Ilpo Järvinen | 77722b1 | 2009-12-08 20:54:11 -0800 | [diff] [blame] | 2440 | } |
| 2441 | |
Yuchung Cheng | e33099f | 2013-03-20 13:33:00 +0000 | [diff] [blame] | 2442 | /* Undo during loss recovery after partial ACK or using F-RTO. */ |
| 2443 | static bool tcp_try_undo_loss(struct sock *sk, bool frto_undo) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2444 | { |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 2445 | struct tcp_sock *tp = tcp_sk(sk); |
| 2446 | |
Yuchung Cheng | e33099f | 2013-03-20 13:33:00 +0000 | [diff] [blame] | 2447 | if (frto_undo || tcp_may_undo(tp)) { |
Yuchung Cheng | 7026b91 | 2013-05-29 14:20:13 +0000 | [diff] [blame] | 2448 | tcp_undo_cwnd_reduction(sk, true); |
Stephen Hemminger | 6a438bb | 2005-11-10 17:14:59 -0800 | [diff] [blame] | 2449 | |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 2450 | DBGUNDO(sk, "partial loss"); |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 2451 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPLOSSUNDO); |
Yuchung Cheng | e33099f | 2013-03-20 13:33:00 +0000 | [diff] [blame] | 2452 | if (frto_undo) |
| 2453 | NET_INC_STATS_BH(sock_net(sk), |
| 2454 | LINUX_MIB_TCPSPURIOUSRTOS); |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 2455 | inet_csk(sk)->icsk_retransmits = 0; |
Yuchung Cheng | e33099f | 2013-03-20 13:33:00 +0000 | [diff] [blame] | 2456 | if (frto_undo || tcp_is_sack(tp)) |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 2457 | tcp_set_ca_state(sk, TCP_CA_Open); |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 2458 | return true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2459 | } |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 2460 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2461 | } |
| 2462 | |
Yuchung Cheng | 684bad1 | 2012-09-02 17:38:04 +0000 | [diff] [blame] | 2463 | /* The cwnd reduction in CWR and Recovery use the PRR algorithm |
| 2464 | * https://datatracker.ietf.org/doc/draft-ietf-tcpm-proportional-rate-reduction/ |
Yuchung Cheng | fb4d3d1 | 2012-09-02 17:38:03 +0000 | [diff] [blame] | 2465 | * It computes the number of packets to send (sndcnt) based on packets newly |
| 2466 | * delivered: |
| 2467 | * 1) If the packets in flight is larger than ssthresh, PRR spreads the |
| 2468 | * cwnd reductions across a full RTT. |
| 2469 | * 2) If packets in flight is lower than ssthresh (such as due to excess |
| 2470 | * losses and/or application stalls), do not perform any further cwnd |
| 2471 | * reductions, but instead slow start up to ssthresh. |
| 2472 | */ |
Yuchung Cheng | 684bad1 | 2012-09-02 17:38:04 +0000 | [diff] [blame] | 2473 | static void tcp_init_cwnd_reduction(struct sock *sk, const bool set_ssthresh) |
| 2474 | { |
| 2475 | struct tcp_sock *tp = tcp_sk(sk); |
| 2476 | |
| 2477 | tp->high_seq = tp->snd_nxt; |
Nandita Dukkipati | 9b717a8 | 2013-03-11 10:00:44 +0000 | [diff] [blame] | 2478 | tp->tlp_high_seq = 0; |
Yuchung Cheng | 684bad1 | 2012-09-02 17:38:04 +0000 | [diff] [blame] | 2479 | tp->snd_cwnd_cnt = 0; |
| 2480 | tp->prior_cwnd = tp->snd_cwnd; |
| 2481 | tp->prr_delivered = 0; |
| 2482 | tp->prr_out = 0; |
| 2483 | if (set_ssthresh) |
| 2484 | tp->snd_ssthresh = inet_csk(sk)->icsk_ca_ops->ssthresh(sk); |
| 2485 | TCP_ECN_queue_cwr(tp); |
| 2486 | } |
| 2487 | |
Yuchung Cheng | 6804973 | 2013-05-29 14:20:11 +0000 | [diff] [blame] | 2488 | static void tcp_cwnd_reduction(struct sock *sk, const int prior_unsacked, |
Yuchung Cheng | 684bad1 | 2012-09-02 17:38:04 +0000 | [diff] [blame] | 2489 | int fast_rexmit) |
Yuchung Cheng | fb4d3d1 | 2012-09-02 17:38:03 +0000 | [diff] [blame] | 2490 | { |
| 2491 | struct tcp_sock *tp = tcp_sk(sk); |
| 2492 | int sndcnt = 0; |
| 2493 | int delta = tp->snd_ssthresh - tcp_packets_in_flight(tp); |
Yuchung Cheng | 6804973 | 2013-05-29 14:20:11 +0000 | [diff] [blame] | 2494 | int newly_acked_sacked = prior_unsacked - |
| 2495 | (tp->packets_out - tp->sacked_out); |
Yuchung Cheng | fb4d3d1 | 2012-09-02 17:38:03 +0000 | [diff] [blame] | 2496 | |
Yuchung Cheng | 684bad1 | 2012-09-02 17:38:04 +0000 | [diff] [blame] | 2497 | tp->prr_delivered += newly_acked_sacked; |
Yuchung Cheng | fb4d3d1 | 2012-09-02 17:38:03 +0000 | [diff] [blame] | 2498 | if (tcp_packets_in_flight(tp) > tp->snd_ssthresh) { |
| 2499 | u64 dividend = (u64)tp->snd_ssthresh * tp->prr_delivered + |
| 2500 | tp->prior_cwnd - 1; |
| 2501 | sndcnt = div_u64(dividend, tp->prior_cwnd) - tp->prr_out; |
| 2502 | } else { |
| 2503 | sndcnt = min_t(int, delta, |
| 2504 | max_t(int, tp->prr_delivered - tp->prr_out, |
| 2505 | newly_acked_sacked) + 1); |
| 2506 | } |
| 2507 | |
| 2508 | sndcnt = max(sndcnt, (fast_rexmit ? 1 : 0)); |
| 2509 | tp->snd_cwnd = tcp_packets_in_flight(tp) + sndcnt; |
| 2510 | } |
| 2511 | |
Yuchung Cheng | 684bad1 | 2012-09-02 17:38:04 +0000 | [diff] [blame] | 2512 | static inline void tcp_end_cwnd_reduction(struct sock *sk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2513 | { |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 2514 | struct tcp_sock *tp = tcp_sk(sk); |
Nandita Dukkipati | a262f0c | 2011-08-21 20:21:57 +0000 | [diff] [blame] | 2515 | |
Yuchung Cheng | 684bad1 | 2012-09-02 17:38:04 +0000 | [diff] [blame] | 2516 | /* Reset cwnd to ssthresh in CWR or Recovery (unless it's undone) */ |
| 2517 | if (inet_csk(sk)->icsk_ca_state == TCP_CA_CWR || |
| 2518 | (tp->undo_marker && tp->snd_ssthresh < TCP_INFINITE_SSTHRESH)) { |
| 2519 | tp->snd_cwnd = tp->snd_ssthresh; |
| 2520 | tp->snd_cwnd_stamp = tcp_time_stamp; |
Yuchung Cheng | 67d4120 | 2011-03-14 10:57:03 +0000 | [diff] [blame] | 2521 | } |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 2522 | tcp_ca_event(sk, CA_EVENT_COMPLETE_CWR); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2523 | } |
| 2524 | |
Yuchung Cheng | 684bad1 | 2012-09-02 17:38:04 +0000 | [diff] [blame] | 2525 | /* Enter CWR state. Disable cwnd undo since congestion is proven with ECN */ |
Yuchung Cheng | 09484d1 | 2012-09-02 17:38:02 +0000 | [diff] [blame] | 2526 | void tcp_enter_cwr(struct sock *sk, const int set_ssthresh) |
| 2527 | { |
| 2528 | struct tcp_sock *tp = tcp_sk(sk); |
Yuchung Cheng | 09484d1 | 2012-09-02 17:38:02 +0000 | [diff] [blame] | 2529 | |
| 2530 | tp->prior_ssthresh = 0; |
Yuchung Cheng | 684bad1 | 2012-09-02 17:38:04 +0000 | [diff] [blame] | 2531 | if (inet_csk(sk)->icsk_ca_state < TCP_CA_CWR) { |
Yuchung Cheng | 09484d1 | 2012-09-02 17:38:02 +0000 | [diff] [blame] | 2532 | tp->undo_marker = 0; |
Yuchung Cheng | 684bad1 | 2012-09-02 17:38:04 +0000 | [diff] [blame] | 2533 | tcp_init_cwnd_reduction(sk, set_ssthresh); |
Yuchung Cheng | 09484d1 | 2012-09-02 17:38:02 +0000 | [diff] [blame] | 2534 | tcp_set_ca_state(sk, TCP_CA_CWR); |
| 2535 | } |
| 2536 | } |
| 2537 | |
Ilpo Järvinen | 8aca6cb | 2008-06-04 11:34:22 -0700 | [diff] [blame] | 2538 | static void tcp_try_keep_open(struct sock *sk) |
| 2539 | { |
| 2540 | struct tcp_sock *tp = tcp_sk(sk); |
| 2541 | int state = TCP_CA_Open; |
| 2542 | |
Neal Cardwell | f698204 | 2011-11-16 08:58:04 +0000 | [diff] [blame] | 2543 | if (tcp_left_out(tp) || tcp_any_retrans_done(sk)) |
Ilpo Järvinen | 8aca6cb | 2008-06-04 11:34:22 -0700 | [diff] [blame] | 2544 | state = TCP_CA_Disorder; |
| 2545 | |
| 2546 | if (inet_csk(sk)->icsk_ca_state != state) { |
| 2547 | tcp_set_ca_state(sk, state); |
| 2548 | tp->high_seq = tp->snd_nxt; |
| 2549 | } |
| 2550 | } |
| 2551 | |
Yuchung Cheng | 6804973 | 2013-05-29 14:20:11 +0000 | [diff] [blame] | 2552 | static void tcp_try_to_open(struct sock *sk, int flag, const int prior_unsacked) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2553 | { |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 2554 | struct tcp_sock *tp = tcp_sk(sk); |
| 2555 | |
Ilpo Järvinen | 86426c2 | 2007-08-09 14:45:17 +0300 | [diff] [blame] | 2556 | tcp_verify_left_out(tp); |
| 2557 | |
Yuchung Cheng | 9b44190 | 2013-03-20 13:32:58 +0000 | [diff] [blame] | 2558 | if (!tcp_any_retrans_done(sk)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2559 | tp->retrans_stamp = 0; |
| 2560 | |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 2561 | if (flag & FLAG_ECE) |
Ilpo Järvinen | 3cfe3ba | 2007-02-27 10:09:49 -0800 | [diff] [blame] | 2562 | tcp_enter_cwr(sk, 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2563 | |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 2564 | if (inet_csk(sk)->icsk_ca_state != TCP_CA_CWR) { |
Ilpo Järvinen | 8aca6cb | 2008-06-04 11:34:22 -0700 | [diff] [blame] | 2565 | tcp_try_keep_open(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2566 | } else { |
Yuchung Cheng | 6804973 | 2013-05-29 14:20:11 +0000 | [diff] [blame] | 2567 | tcp_cwnd_reduction(sk, prior_unsacked, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2568 | } |
| 2569 | } |
| 2570 | |
John Heffner | 5d424d5 | 2006-03-20 17:53:41 -0800 | [diff] [blame] | 2571 | static void tcp_mtup_probe_failed(struct sock *sk) |
| 2572 | { |
| 2573 | struct inet_connection_sock *icsk = inet_csk(sk); |
| 2574 | |
| 2575 | icsk->icsk_mtup.search_high = icsk->icsk_mtup.probe_size - 1; |
| 2576 | icsk->icsk_mtup.probe_size = 0; |
| 2577 | } |
| 2578 | |
Ilpo Järvinen | 72211e9 | 2009-03-14 14:23:04 +0000 | [diff] [blame] | 2579 | static void tcp_mtup_probe_success(struct sock *sk) |
John Heffner | 5d424d5 | 2006-03-20 17:53:41 -0800 | [diff] [blame] | 2580 | { |
| 2581 | struct tcp_sock *tp = tcp_sk(sk); |
| 2582 | struct inet_connection_sock *icsk = inet_csk(sk); |
| 2583 | |
| 2584 | /* FIXME: breaks with very large cwnd */ |
| 2585 | tp->prior_ssthresh = tcp_current_ssthresh(sk); |
| 2586 | tp->snd_cwnd = tp->snd_cwnd * |
| 2587 | tcp_mss_to_mtu(sk, tp->mss_cache) / |
| 2588 | icsk->icsk_mtup.probe_size; |
| 2589 | tp->snd_cwnd_cnt = 0; |
| 2590 | tp->snd_cwnd_stamp = tcp_time_stamp; |
John Heffner | 9c6d5e5 | 2010-10-06 21:18:02 -0700 | [diff] [blame] | 2591 | tp->snd_ssthresh = tcp_current_ssthresh(sk); |
John Heffner | 5d424d5 | 2006-03-20 17:53:41 -0800 | [diff] [blame] | 2592 | |
| 2593 | icsk->icsk_mtup.search_low = icsk->icsk_mtup.probe_size; |
| 2594 | icsk->icsk_mtup.probe_size = 0; |
| 2595 | tcp_sync_mss(sk, icsk->icsk_pmtu_cookie); |
| 2596 | } |
| 2597 | |
Ilpo Järvinen | e1aa680 | 2008-11-24 21:11:55 -0800 | [diff] [blame] | 2598 | /* Do a simple retransmit without using the backoff mechanisms in |
| 2599 | * tcp_timer. This is used for path mtu discovery. |
| 2600 | * The socket is already locked here. |
| 2601 | */ |
| 2602 | void tcp_simple_retransmit(struct sock *sk) |
| 2603 | { |
| 2604 | const struct inet_connection_sock *icsk = inet_csk(sk); |
| 2605 | struct tcp_sock *tp = tcp_sk(sk); |
| 2606 | struct sk_buff *skb; |
Ilpo Järvinen | 0c54b85 | 2009-03-14 14:23:05 +0000 | [diff] [blame] | 2607 | unsigned int mss = tcp_current_mss(sk); |
Ilpo Järvinen | e1aa680 | 2008-11-24 21:11:55 -0800 | [diff] [blame] | 2608 | u32 prior_lost = tp->lost_out; |
| 2609 | |
| 2610 | tcp_for_write_queue(skb, sk) { |
| 2611 | if (skb == tcp_send_head(sk)) |
| 2612 | break; |
Ilpo Järvinen | 775ffab | 2008-12-05 22:41:26 -0800 | [diff] [blame] | 2613 | if (tcp_skb_seglen(skb) > mss && |
Ilpo Järvinen | e1aa680 | 2008-11-24 21:11:55 -0800 | [diff] [blame] | 2614 | !(TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)) { |
| 2615 | if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_RETRANS) { |
| 2616 | TCP_SKB_CB(skb)->sacked &= ~TCPCB_SACKED_RETRANS; |
| 2617 | tp->retrans_out -= tcp_skb_pcount(skb); |
| 2618 | } |
| 2619 | tcp_skb_mark_lost_uncond_verify(tp, skb); |
| 2620 | } |
| 2621 | } |
| 2622 | |
| 2623 | tcp_clear_retrans_hints_partial(tp); |
| 2624 | |
| 2625 | if (prior_lost == tp->lost_out) |
| 2626 | return; |
| 2627 | |
| 2628 | if (tcp_is_reno(tp)) |
| 2629 | tcp_limit_reno_sacked(tp); |
| 2630 | |
| 2631 | tcp_verify_left_out(tp); |
| 2632 | |
| 2633 | /* Don't muck with the congestion window here. |
| 2634 | * Reason is that we do not increase amount of _data_ |
| 2635 | * in network, but units changed and effective |
| 2636 | * cwnd/ssthresh really reduced now. |
| 2637 | */ |
| 2638 | if (icsk->icsk_ca_state != TCP_CA_Loss) { |
| 2639 | tp->high_seq = tp->snd_nxt; |
| 2640 | tp->snd_ssthresh = tcp_current_ssthresh(sk); |
| 2641 | tp->prior_ssthresh = 0; |
| 2642 | tp->undo_marker = 0; |
| 2643 | tcp_set_ca_state(sk, TCP_CA_Loss); |
| 2644 | } |
| 2645 | tcp_xmit_retransmit_queue(sk); |
| 2646 | } |
Eric Dumazet | 4bc2f18 | 2010-07-09 21:22:10 +0000 | [diff] [blame] | 2647 | EXPORT_SYMBOL(tcp_simple_retransmit); |
Ilpo Järvinen | e1aa680 | 2008-11-24 21:11:55 -0800 | [diff] [blame] | 2648 | |
Yuchung Cheng | 1fbc340 | 2012-05-02 13:30:02 +0000 | [diff] [blame] | 2649 | static void tcp_enter_recovery(struct sock *sk, bool ece_ack) |
| 2650 | { |
| 2651 | struct tcp_sock *tp = tcp_sk(sk); |
| 2652 | int mib_idx; |
| 2653 | |
| 2654 | if (tcp_is_reno(tp)) |
| 2655 | mib_idx = LINUX_MIB_TCPRENORECOVERY; |
| 2656 | else |
| 2657 | mib_idx = LINUX_MIB_TCPSACKRECOVERY; |
| 2658 | |
| 2659 | NET_INC_STATS_BH(sock_net(sk), mib_idx); |
| 2660 | |
Yuchung Cheng | 1fbc340 | 2012-05-02 13:30:02 +0000 | [diff] [blame] | 2661 | tp->prior_ssthresh = 0; |
| 2662 | tp->undo_marker = tp->snd_una; |
| 2663 | tp->undo_retrans = tp->retrans_out; |
| 2664 | |
| 2665 | if (inet_csk(sk)->icsk_ca_state < TCP_CA_CWR) { |
| 2666 | if (!ece_ack) |
| 2667 | tp->prior_ssthresh = tcp_current_ssthresh(sk); |
Yuchung Cheng | 684bad1 | 2012-09-02 17:38:04 +0000 | [diff] [blame] | 2668 | tcp_init_cwnd_reduction(sk, true); |
Yuchung Cheng | 1fbc340 | 2012-05-02 13:30:02 +0000 | [diff] [blame] | 2669 | } |
Yuchung Cheng | 1fbc340 | 2012-05-02 13:30:02 +0000 | [diff] [blame] | 2670 | tcp_set_ca_state(sk, TCP_CA_Recovery); |
| 2671 | } |
| 2672 | |
Yuchung Cheng | ab42d9e | 2013-03-20 13:32:59 +0000 | [diff] [blame] | 2673 | /* Process an ACK in CA_Loss state. Move to CA_Open if lost data are |
| 2674 | * recovered or spurious. Otherwise retransmits more on partial ACKs. |
| 2675 | */ |
Yuchung Cheng | e33099f | 2013-03-20 13:33:00 +0000 | [diff] [blame] | 2676 | static void tcp_process_loss(struct sock *sk, int flag, bool is_dupack) |
Yuchung Cheng | ab42d9e | 2013-03-20 13:32:59 +0000 | [diff] [blame] | 2677 | { |
| 2678 | struct inet_connection_sock *icsk = inet_csk(sk); |
| 2679 | struct tcp_sock *tp = tcp_sk(sk); |
Yuchung Cheng | e33099f | 2013-03-20 13:33:00 +0000 | [diff] [blame] | 2680 | bool recovered = !before(tp->snd_una, tp->high_seq); |
Yuchung Cheng | ab42d9e | 2013-03-20 13:32:59 +0000 | [diff] [blame] | 2681 | |
Yuchung Cheng | e33099f | 2013-03-20 13:33:00 +0000 | [diff] [blame] | 2682 | if (tp->frto) { /* F-RTO RFC5682 sec 3.1 (sack enhanced version). */ |
| 2683 | if (flag & FLAG_ORIG_SACK_ACKED) { |
| 2684 | /* Step 3.b. A timeout is spurious if not all data are |
| 2685 | * lost, i.e., never-retransmitted data are (s)acked. |
| 2686 | */ |
| 2687 | tcp_try_undo_loss(sk, true); |
| 2688 | return; |
| 2689 | } |
| 2690 | if (after(tp->snd_nxt, tp->high_seq) && |
| 2691 | (flag & FLAG_DATA_SACKED || is_dupack)) { |
| 2692 | tp->frto = 0; /* Loss was real: 2nd part of step 3.a */ |
| 2693 | } else if (flag & FLAG_SND_UNA_ADVANCED && !recovered) { |
| 2694 | tp->high_seq = tp->snd_nxt; |
| 2695 | __tcp_push_pending_frames(sk, tcp_current_mss(sk), |
| 2696 | TCP_NAGLE_OFF); |
| 2697 | if (after(tp->snd_nxt, tp->high_seq)) |
| 2698 | return; /* Step 2.b */ |
| 2699 | tp->frto = 0; |
| 2700 | } |
| 2701 | } |
| 2702 | |
| 2703 | if (recovered) { |
| 2704 | /* F-RTO RFC5682 sec 3.1 step 2.a and 1st part of step 3.a */ |
Yuchung Cheng | ab42d9e | 2013-03-20 13:32:59 +0000 | [diff] [blame] | 2705 | icsk->icsk_retransmits = 0; |
| 2706 | tcp_try_undo_recovery(sk); |
| 2707 | return; |
| 2708 | } |
Yuchung Cheng | ab42d9e | 2013-03-20 13:32:59 +0000 | [diff] [blame] | 2709 | if (flag & FLAG_DATA_ACKED) |
| 2710 | icsk->icsk_retransmits = 0; |
Yuchung Cheng | e33099f | 2013-03-20 13:33:00 +0000 | [diff] [blame] | 2711 | if (tcp_is_reno(tp)) { |
| 2712 | /* A Reno DUPACK means new data in F-RTO step 2.b above are |
| 2713 | * delivered. Lower inflight to clock out (re)tranmissions. |
| 2714 | */ |
| 2715 | if (after(tp->snd_nxt, tp->high_seq) && is_dupack) |
| 2716 | tcp_add_reno_sack(sk); |
| 2717 | else if (flag & FLAG_SND_UNA_ADVANCED) |
| 2718 | tcp_reset_reno_sack(tp); |
| 2719 | } |
| 2720 | if (tcp_try_undo_loss(sk, false)) |
Yuchung Cheng | ab42d9e | 2013-03-20 13:32:59 +0000 | [diff] [blame] | 2721 | return; |
Yuchung Cheng | ab42d9e | 2013-03-20 13:32:59 +0000 | [diff] [blame] | 2722 | tcp_xmit_retransmit_queue(sk); |
| 2723 | } |
| 2724 | |
Yuchung Cheng | 6a63df4 | 2013-05-29 14:20:12 +0000 | [diff] [blame] | 2725 | /* Undo during fast recovery after partial ACK. */ |
Yuchung Cheng | 7026b91 | 2013-05-29 14:20:13 +0000 | [diff] [blame] | 2726 | static bool tcp_try_undo_partial(struct sock *sk, const int acked, |
| 2727 | const int prior_unsacked) |
Yuchung Cheng | 6a63df4 | 2013-05-29 14:20:12 +0000 | [diff] [blame] | 2728 | { |
| 2729 | struct tcp_sock *tp = tcp_sk(sk); |
Yuchung Cheng | 6a63df4 | 2013-05-29 14:20:12 +0000 | [diff] [blame] | 2730 | |
Yuchung Cheng | 7026b91 | 2013-05-29 14:20:13 +0000 | [diff] [blame] | 2731 | if (tp->undo_marker && tcp_packet_delayed(tp)) { |
Yuchung Cheng | 6a63df4 | 2013-05-29 14:20:12 +0000 | [diff] [blame] | 2732 | /* Plain luck! Hole if filled with delayed |
| 2733 | * packet, rather than with a retransmit. |
| 2734 | */ |
Yuchung Cheng | 7026b91 | 2013-05-29 14:20:13 +0000 | [diff] [blame] | 2735 | tcp_update_reordering(sk, tcp_fackets_out(tp) + acked, 1); |
| 2736 | |
| 2737 | /* We are getting evidence that the reordering degree is higher |
| 2738 | * than we realized. If there are no retransmits out then we |
| 2739 | * can undo. Otherwise we clock out new packets but do not |
| 2740 | * mark more packets lost or retransmit more. |
| 2741 | */ |
| 2742 | if (tp->retrans_out) { |
| 2743 | tcp_cwnd_reduction(sk, prior_unsacked, 0); |
| 2744 | return true; |
| 2745 | } |
| 2746 | |
Yuchung Cheng | 6a63df4 | 2013-05-29 14:20:12 +0000 | [diff] [blame] | 2747 | if (!tcp_any_retrans_done(sk)) |
| 2748 | tp->retrans_stamp = 0; |
| 2749 | |
Yuchung Cheng | 7026b91 | 2013-05-29 14:20:13 +0000 | [diff] [blame] | 2750 | DBGUNDO(sk, "partial recovery"); |
| 2751 | tcp_undo_cwnd_reduction(sk, true); |
Yuchung Cheng | 6a63df4 | 2013-05-29 14:20:12 +0000 | [diff] [blame] | 2752 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPPARTIALUNDO); |
Yuchung Cheng | 7026b91 | 2013-05-29 14:20:13 +0000 | [diff] [blame] | 2753 | tcp_try_keep_open(sk); |
| 2754 | return true; |
Yuchung Cheng | 6a63df4 | 2013-05-29 14:20:12 +0000 | [diff] [blame] | 2755 | } |
Yuchung Cheng | 7026b91 | 2013-05-29 14:20:13 +0000 | [diff] [blame] | 2756 | return false; |
Yuchung Cheng | 6a63df4 | 2013-05-29 14:20:12 +0000 | [diff] [blame] | 2757 | } |
| 2758 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2759 | /* Process an event, which can update packets-in-flight not trivially. |
| 2760 | * Main goal of this function is to calculate new estimate for left_out, |
| 2761 | * taking into account both packets sitting in receiver's buffer and |
| 2762 | * packets lost by network. |
| 2763 | * |
| 2764 | * Besides that it does CWND reduction, when packet loss is detected |
| 2765 | * and changes state of machine. |
| 2766 | * |
| 2767 | * It does _not_ decide what to send, it is made in function |
| 2768 | * tcp_xmit_retransmit_queue(). |
| 2769 | */ |
Yuchung Cheng | 6804973 | 2013-05-29 14:20:11 +0000 | [diff] [blame] | 2770 | static void tcp_fastretrans_alert(struct sock *sk, const int acked, |
| 2771 | const int prior_unsacked, |
Nandita Dukkipati | 35f079e | 2013-05-21 15:12:07 +0000 | [diff] [blame] | 2772 | bool is_dupack, int flag) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2773 | { |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 2774 | struct inet_connection_sock *icsk = inet_csk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2775 | struct tcp_sock *tp = tcp_sk(sk); |
Yuchung Cheng | 6a63df4 | 2013-05-29 14:20:12 +0000 | [diff] [blame] | 2776 | bool do_lost = is_dupack || ((flag & FLAG_DATA_SACKED) && |
Ilpo Järvinen | 85cc391 | 2007-11-15 19:39:31 -0800 | [diff] [blame] | 2777 | (tcp_fackets_out(tp) > tp->reordering)); |
Yuchung Cheng | 1fbc340 | 2012-05-02 13:30:02 +0000 | [diff] [blame] | 2778 | int fast_rexmit = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2779 | |
Ilpo Järvinen | 3ccd313 | 2007-12-31 04:43:32 -0800 | [diff] [blame] | 2780 | if (WARN_ON(!tp->packets_out && tp->sacked_out)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2781 | tp->sacked_out = 0; |
Ilpo Järvinen | 91fed7a | 2007-10-09 01:24:15 -0700 | [diff] [blame] | 2782 | if (WARN_ON(!tp->sacked_out && tp->fackets_out)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2783 | tp->fackets_out = 0; |
| 2784 | |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 2785 | /* Now state machine starts. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2786 | * A. ECE, hence prohibit cwnd undoing, the reduction is required. */ |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 2787 | if (flag & FLAG_ECE) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2788 | tp->prior_ssthresh = 0; |
| 2789 | |
| 2790 | /* B. In all the states check for reneging SACKs. */ |
Ilpo Järvinen | cadbd03 | 2007-12-31 04:49:21 -0800 | [diff] [blame] | 2791 | if (tcp_check_sack_reneging(sk, flag)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2792 | return; |
| 2793 | |
Yuchung Cheng | 974c123 | 2012-01-19 14:42:21 +0000 | [diff] [blame] | 2794 | /* C. Check consistency of the current state. */ |
Ilpo Järvinen | 005903b | 2007-08-09 14:44:16 +0300 | [diff] [blame] | 2795 | tcp_verify_left_out(tp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2796 | |
Yuchung Cheng | 974c123 | 2012-01-19 14:42:21 +0000 | [diff] [blame] | 2797 | /* D. Check state exit conditions. State can be terminated |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2798 | * when high_seq is ACKed. */ |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 2799 | if (icsk->icsk_ca_state == TCP_CA_Open) { |
Ilpo Järvinen | 547b792 | 2008-07-25 21:43:18 -0700 | [diff] [blame] | 2800 | WARN_ON(tp->retrans_out != 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2801 | tp->retrans_stamp = 0; |
| 2802 | } else if (!before(tp->snd_una, tp->high_seq)) { |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 2803 | switch (icsk->icsk_ca_state) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2804 | case TCP_CA_CWR: |
| 2805 | /* CWR is to be held something *above* high_seq |
| 2806 | * is ACKed for CWR bit to reach receiver. */ |
| 2807 | if (tp->snd_una != tp->high_seq) { |
Yuchung Cheng | 684bad1 | 2012-09-02 17:38:04 +0000 | [diff] [blame] | 2808 | tcp_end_cwnd_reduction(sk); |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 2809 | tcp_set_ca_state(sk, TCP_CA_Open); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2810 | } |
| 2811 | break; |
| 2812 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2813 | case TCP_CA_Recovery: |
Ilpo Järvinen | e60402d | 2007-08-09 15:14:46 +0300 | [diff] [blame] | 2814 | if (tcp_is_reno(tp)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2815 | tcp_reset_reno_sack(tp); |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 2816 | if (tcp_try_undo_recovery(sk)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2817 | return; |
Yuchung Cheng | 684bad1 | 2012-09-02 17:38:04 +0000 | [diff] [blame] | 2818 | tcp_end_cwnd_reduction(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2819 | break; |
| 2820 | } |
| 2821 | } |
| 2822 | |
Yuchung Cheng | 974c123 | 2012-01-19 14:42:21 +0000 | [diff] [blame] | 2823 | /* E. Process state. */ |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 2824 | switch (icsk->icsk_ca_state) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2825 | case TCP_CA_Recovery: |
Ilpo Järvinen | 2e60529 | 2007-08-02 19:46:58 -0700 | [diff] [blame] | 2826 | if (!(flag & FLAG_SND_UNA_ADVANCED)) { |
Ilpo Järvinen | e60402d | 2007-08-09 15:14:46 +0300 | [diff] [blame] | 2827 | if (tcp_is_reno(tp) && is_dupack) |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 2828 | tcp_add_reno_sack(sk); |
Yuchung Cheng | 7026b91 | 2013-05-29 14:20:13 +0000 | [diff] [blame] | 2829 | } else { |
| 2830 | if (tcp_try_undo_partial(sk, acked, prior_unsacked)) |
| 2831 | return; |
| 2832 | /* Partial ACK arrived. Force fast retransmit. */ |
| 2833 | do_lost = tcp_is_reno(tp) || |
| 2834 | tcp_fackets_out(tp) > tp->reordering; |
| 2835 | } |
Yuchung Cheng | c7d9d6a | 2013-05-29 14:20:14 +0000 | [diff] [blame] | 2836 | if (tcp_try_undo_dsack(sk)) { |
| 2837 | tcp_try_keep_open(sk); |
| 2838 | return; |
| 2839 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2840 | break; |
| 2841 | case TCP_CA_Loss: |
Yuchung Cheng | e33099f | 2013-03-20 13:33:00 +0000 | [diff] [blame] | 2842 | tcp_process_loss(sk, flag, is_dupack); |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 2843 | if (icsk->icsk_ca_state != TCP_CA_Open) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2844 | return; |
Yuchung Cheng | ab42d9e | 2013-03-20 13:32:59 +0000 | [diff] [blame] | 2845 | /* Fall through to processing in Open state. */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2846 | default: |
Ilpo Järvinen | e60402d | 2007-08-09 15:14:46 +0300 | [diff] [blame] | 2847 | if (tcp_is_reno(tp)) { |
Ilpo Järvinen | 2e60529 | 2007-08-02 19:46:58 -0700 | [diff] [blame] | 2848 | if (flag & FLAG_SND_UNA_ADVANCED) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2849 | tcp_reset_reno_sack(tp); |
| 2850 | if (is_dupack) |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 2851 | tcp_add_reno_sack(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2852 | } |
| 2853 | |
Neal Cardwell | f698204 | 2011-11-16 08:58:04 +0000 | [diff] [blame] | 2854 | if (icsk->icsk_ca_state <= TCP_CA_Disorder) |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 2855 | tcp_try_undo_dsack(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2856 | |
Yuchung Cheng | 750ea2b | 2012-05-02 13:30:04 +0000 | [diff] [blame] | 2857 | if (!tcp_time_to_recover(sk, flag)) { |
Yuchung Cheng | 6804973 | 2013-05-29 14:20:11 +0000 | [diff] [blame] | 2858 | tcp_try_to_open(sk, flag, prior_unsacked); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2859 | return; |
| 2860 | } |
| 2861 | |
John Heffner | 5d424d5 | 2006-03-20 17:53:41 -0800 | [diff] [blame] | 2862 | /* MTU probe failure: don't reduce cwnd */ |
| 2863 | if (icsk->icsk_ca_state < TCP_CA_CWR && |
| 2864 | icsk->icsk_mtup.probe_size && |
John Heffner | 0e7b136 | 2006-03-20 21:32:58 -0800 | [diff] [blame] | 2865 | tp->snd_una == tp->mtu_probe.probe_seq_start) { |
John Heffner | 5d424d5 | 2006-03-20 17:53:41 -0800 | [diff] [blame] | 2866 | tcp_mtup_probe_failed(sk); |
| 2867 | /* Restores the reduction we did in tcp_mtup_probe() */ |
| 2868 | tp->snd_cwnd++; |
| 2869 | tcp_simple_retransmit(sk); |
| 2870 | return; |
| 2871 | } |
| 2872 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2873 | /* Otherwise enter Recovery state */ |
Yuchung Cheng | 1fbc340 | 2012-05-02 13:30:02 +0000 | [diff] [blame] | 2874 | tcp_enter_recovery(sk, (flag & FLAG_ECE)); |
Ilpo Järvinen | 85cc391 | 2007-11-15 19:39:31 -0800 | [diff] [blame] | 2875 | fast_rexmit = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2876 | } |
| 2877 | |
Yuchung Cheng | 3e59cb0 | 2013-05-17 13:45:05 +0000 | [diff] [blame] | 2878 | if (do_lost) |
Ilpo Järvinen | 85cc391 | 2007-11-15 19:39:31 -0800 | [diff] [blame] | 2879 | tcp_update_scoreboard(sk, fast_rexmit); |
Yuchung Cheng | 6804973 | 2013-05-29 14:20:11 +0000 | [diff] [blame] | 2880 | tcp_cwnd_reduction(sk, prior_unsacked, fast_rexmit); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2881 | tcp_xmit_retransmit_queue(sk); |
| 2882 | } |
| 2883 | |
Yuchung Cheng | ed08495 | 2013-07-22 16:20:48 -0700 | [diff] [blame] | 2884 | static inline bool tcp_ack_update_rtt(struct sock *sk, const int flag, |
| 2885 | s32 seq_rtt, s32 sack_rtt) |
Ilpo Järvinen | 41834b7 | 2008-12-05 22:43:26 -0800 | [diff] [blame] | 2886 | { |
Yuchung Cheng | 5b08e47 | 2013-07-22 16:20:46 -0700 | [diff] [blame] | 2887 | const struct tcp_sock *tp = tcp_sk(sk); |
Ilpo Järvinen | 41834b7 | 2008-12-05 22:43:26 -0800 | [diff] [blame] | 2888 | |
Yuchung Cheng | 5b08e47 | 2013-07-22 16:20:46 -0700 | [diff] [blame] | 2889 | /* Prefer RTT measured from ACK's timing to TS-ECR. This is because |
| 2890 | * broken middle-boxes or peers may corrupt TS-ECR fields. But |
| 2891 | * Karn's algorithm forbids taking RTT if some retransmitted data |
| 2892 | * is acked (RFC6298). |
| 2893 | */ |
| 2894 | if (flag & FLAG_RETRANS_DATA_ACKED) |
| 2895 | seq_rtt = -1; |
| 2896 | |
Yuchung Cheng | ed08495 | 2013-07-22 16:20:48 -0700 | [diff] [blame] | 2897 | if (seq_rtt < 0) |
| 2898 | seq_rtt = sack_rtt; |
| 2899 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2900 | /* RTTM Rule: A TSecr value received in a segment is used to |
| 2901 | * update the averaged RTT measurement only if the segment |
| 2902 | * acknowledges some new data, i.e., only if it advances the |
| 2903 | * left edge of the send window. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2904 | * See draft-ietf-tcplw-high-performance-00, section 3.3. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2905 | */ |
Yuchung Cheng | 5b08e47 | 2013-07-22 16:20:46 -0700 | [diff] [blame] | 2906 | if (seq_rtt < 0 && tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr) |
| 2907 | seq_rtt = tcp_time_stamp - tp->rx_opt.rcv_tsecr; |
Ilpo Järvinen | 41834b7 | 2008-12-05 22:43:26 -0800 | [diff] [blame] | 2908 | |
Yuchung Cheng | 5b08e47 | 2013-07-22 16:20:46 -0700 | [diff] [blame] | 2909 | if (seq_rtt < 0) |
Yuchung Cheng | ed08495 | 2013-07-22 16:20:48 -0700 | [diff] [blame] | 2910 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2911 | |
Yuchung Cheng | 5b08e47 | 2013-07-22 16:20:46 -0700 | [diff] [blame] | 2912 | tcp_rtt_estimator(sk, seq_rtt); |
| 2913 | tcp_set_rto(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2914 | |
Yuchung Cheng | 5b08e47 | 2013-07-22 16:20:46 -0700 | [diff] [blame] | 2915 | /* RFC6298: only reset backoff on valid RTT measurement. */ |
| 2916 | inet_csk(sk)->icsk_backoff = 0; |
Yuchung Cheng | ed08495 | 2013-07-22 16:20:48 -0700 | [diff] [blame] | 2917 | return true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2918 | } |
| 2919 | |
Yuchung Cheng | 375fe02 | 2013-07-22 16:20:45 -0700 | [diff] [blame] | 2920 | /* Compute time elapsed between (last) SYNACK and the ACK completing 3WHS. */ |
| 2921 | static void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req) |
| 2922 | { |
| 2923 | struct tcp_sock *tp = tcp_sk(sk); |
| 2924 | s32 seq_rtt = -1; |
| 2925 | |
| 2926 | if (tp->lsndtime && !tp->total_retrans) |
| 2927 | seq_rtt = tcp_time_stamp - tp->lsndtime; |
Yuchung Cheng | ed08495 | 2013-07-22 16:20:48 -0700 | [diff] [blame] | 2928 | tcp_ack_update_rtt(sk, FLAG_SYN_ACKED, seq_rtt, -1); |
Yuchung Cheng | 375fe02 | 2013-07-22 16:20:45 -0700 | [diff] [blame] | 2929 | } |
| 2930 | |
Ilpo Järvinen | c3a05c6 | 2007-12-02 00:47:59 +0200 | [diff] [blame] | 2931 | static void tcp_cong_avoid(struct sock *sk, u32 ack, u32 in_flight) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2932 | { |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 2933 | const struct inet_connection_sock *icsk = inet_csk(sk); |
Ilpo Järvinen | c3a05c6 | 2007-12-02 00:47:59 +0200 | [diff] [blame] | 2934 | icsk->icsk_ca_ops->cong_avoid(sk, ack, in_flight); |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 2935 | tcp_sk(sk)->snd_cwnd_stamp = tcp_time_stamp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2936 | } |
| 2937 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2938 | /* Restart timer after forward progress on connection. |
| 2939 | * RFC2988 recommends to restart timer to now+rto. |
| 2940 | */ |
Yuchung Cheng | 750ea2b | 2012-05-02 13:30:04 +0000 | [diff] [blame] | 2941 | void tcp_rearm_rto(struct sock *sk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2942 | { |
Nandita Dukkipati | 6ba8a3b | 2013-03-11 10:00:43 +0000 | [diff] [blame] | 2943 | const struct inet_connection_sock *icsk = inet_csk(sk); |
Yuchung Cheng | 750ea2b | 2012-05-02 13:30:04 +0000 | [diff] [blame] | 2944 | struct tcp_sock *tp = tcp_sk(sk); |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 2945 | |
Jerry Chu | 168a8f5 | 2012-08-31 12:29:13 +0000 | [diff] [blame] | 2946 | /* If the retrans timer is currently being used by Fast Open |
| 2947 | * for SYN-ACK retrans purpose, stay put. |
| 2948 | */ |
| 2949 | if (tp->fastopen_rsk) |
| 2950 | return; |
| 2951 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2952 | if (!tp->packets_out) { |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 2953 | inet_csk_clear_xmit_timer(sk, ICSK_TIME_RETRANS); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2954 | } else { |
Yuchung Cheng | 750ea2b | 2012-05-02 13:30:04 +0000 | [diff] [blame] | 2955 | u32 rto = inet_csk(sk)->icsk_rto; |
| 2956 | /* Offset the time elapsed after installing regular RTO */ |
Nandita Dukkipati | 6ba8a3b | 2013-03-11 10:00:43 +0000 | [diff] [blame] | 2957 | if (icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS || |
| 2958 | icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) { |
Yuchung Cheng | 750ea2b | 2012-05-02 13:30:04 +0000 | [diff] [blame] | 2959 | struct sk_buff *skb = tcp_write_queue_head(sk); |
| 2960 | const u32 rto_time_stamp = TCP_SKB_CB(skb)->when + rto; |
| 2961 | s32 delta = (s32)(rto_time_stamp - tcp_time_stamp); |
| 2962 | /* delta may not be positive if the socket is locked |
Nandita Dukkipati | 6ba8a3b | 2013-03-11 10:00:43 +0000 | [diff] [blame] | 2963 | * when the retrans timer fires and is rescheduled. |
Yuchung Cheng | 750ea2b | 2012-05-02 13:30:04 +0000 | [diff] [blame] | 2964 | */ |
| 2965 | if (delta > 0) |
| 2966 | rto = delta; |
| 2967 | } |
| 2968 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, rto, |
| 2969 | TCP_RTO_MAX); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2970 | } |
Yuchung Cheng | 750ea2b | 2012-05-02 13:30:04 +0000 | [diff] [blame] | 2971 | } |
| 2972 | |
| 2973 | /* This function is called when the delayed ER timer fires. TCP enters |
| 2974 | * fast recovery and performs fast-retransmit. |
| 2975 | */ |
| 2976 | void tcp_resume_early_retransmit(struct sock *sk) |
| 2977 | { |
| 2978 | struct tcp_sock *tp = tcp_sk(sk); |
| 2979 | |
| 2980 | tcp_rearm_rto(sk); |
| 2981 | |
| 2982 | /* Stop if ER is disabled after the delayed ER timer is scheduled */ |
| 2983 | if (!tp->do_early_retrans) |
| 2984 | return; |
| 2985 | |
| 2986 | tcp_enter_recovery(sk, false); |
| 2987 | tcp_update_scoreboard(sk, 1); |
| 2988 | tcp_xmit_retransmit_queue(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2989 | } |
| 2990 | |
Ilpo Järvinen | 7c46a03 | 2007-09-20 11:33:43 -0700 | [diff] [blame] | 2991 | /* If we get here, the whole TSO packet has not been acked. */ |
Ilpo Järvinen | 13fcf85 | 2007-10-09 01:28:45 -0700 | [diff] [blame] | 2992 | static u32 tcp_tso_acked(struct sock *sk, struct sk_buff *skb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2993 | { |
| 2994 | struct tcp_sock *tp = tcp_sk(sk); |
Ilpo Järvinen | 7c46a03 | 2007-09-20 11:33:43 -0700 | [diff] [blame] | 2995 | u32 packets_acked; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2996 | |
Ilpo Järvinen | 7c46a03 | 2007-09-20 11:33:43 -0700 | [diff] [blame] | 2997 | BUG_ON(!after(TCP_SKB_CB(skb)->end_seq, tp->snd_una)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2998 | |
| 2999 | packets_acked = tcp_skb_pcount(skb); |
Ilpo Järvinen | 7c46a03 | 2007-09-20 11:33:43 -0700 | [diff] [blame] | 3000 | if (tcp_trim_head(sk, skb, tp->snd_una - TCP_SKB_CB(skb)->seq)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3001 | return 0; |
| 3002 | packets_acked -= tcp_skb_pcount(skb); |
| 3003 | |
| 3004 | if (packets_acked) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3005 | BUG_ON(tcp_skb_pcount(skb) == 0); |
Ilpo Järvinen | 7c46a03 | 2007-09-20 11:33:43 -0700 | [diff] [blame] | 3006 | BUG_ON(!before(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3007 | } |
| 3008 | |
Ilpo Järvinen | 13fcf85 | 2007-10-09 01:28:45 -0700 | [diff] [blame] | 3009 | return packets_acked; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3010 | } |
| 3011 | |
Ilpo Järvinen | 7c46a03 | 2007-09-20 11:33:43 -0700 | [diff] [blame] | 3012 | /* Remove acknowledged frames from the retransmission queue. If our packet |
| 3013 | * is before the ack sequence we can discard it as it's confirmed to have |
| 3014 | * arrived at the other end. |
| 3015 | */ |
Ilpo Järvinen | 33f5f57 | 2008-10-07 14:43:06 -0700 | [diff] [blame] | 3016 | static int tcp_clean_rtx_queue(struct sock *sk, int prior_fackets, |
Yuchung Cheng | ed08495 | 2013-07-22 16:20:48 -0700 | [diff] [blame] | 3017 | u32 prior_snd_una, s32 sack_rtt) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3018 | { |
| 3019 | struct tcp_sock *tp = tcp_sk(sk); |
Stephen Hemminger | 2d2abba | 2005-11-10 16:56:12 -0800 | [diff] [blame] | 3020 | const struct inet_connection_sock *icsk = inet_csk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3021 | struct sk_buff *skb; |
Ilpo Järvinen | 7c46a03 | 2007-09-20 11:33:43 -0700 | [diff] [blame] | 3022 | u32 now = tcp_time_stamp; |
Peter Senna Tschudin | 34a6eda | 2013-10-02 14:19:51 +0200 | [diff] [blame] | 3023 | bool fully_acked = true; |
Ilpo Järvinen | 7c46a03 | 2007-09-20 11:33:43 -0700 | [diff] [blame] | 3024 | int flag = 0; |
Ilpo Järvinen | 7201883 | 2007-12-30 04:37:55 -0800 | [diff] [blame] | 3025 | u32 pkts_acked = 0; |
Ilpo Järvinen | c7caf8d | 2007-11-10 21:22:18 -0800 | [diff] [blame] | 3026 | u32 reord = tp->packets_out; |
Ilpo Järvinen | 90638a0 | 2008-09-20 21:25:52 -0700 | [diff] [blame] | 3027 | u32 prior_sacked = tp->sacked_out; |
Ilpo Järvinen | 7c46a03 | 2007-09-20 11:33:43 -0700 | [diff] [blame] | 3028 | s32 seq_rtt = -1; |
Gavin McCullagh | 2072c22 | 2007-12-29 19:11:21 -0800 | [diff] [blame] | 3029 | s32 ca_seq_rtt = -1; |
Ilpo Järvinen | b9ce204 | 2007-06-15 15:08:43 -0700 | [diff] [blame] | 3030 | ktime_t last_ackt = net_invalid_timestamp(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3031 | |
Ilpo Järvinen | 7c46a03 | 2007-09-20 11:33:43 -0700 | [diff] [blame] | 3032 | while ((skb = tcp_write_queue_head(sk)) && skb != tcp_send_head(sk)) { |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 3033 | struct tcp_skb_cb *scb = TCP_SKB_CB(skb); |
Ilpo Järvinen | 7201883 | 2007-12-30 04:37:55 -0800 | [diff] [blame] | 3034 | u32 acked_pcount; |
Ilpo Järvinen | 7c46a03 | 2007-09-20 11:33:43 -0700 | [diff] [blame] | 3035 | u8 sacked = scb->sacked; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3036 | |
Gavin McCullagh | 2072c22 | 2007-12-29 19:11:21 -0800 | [diff] [blame] | 3037 | /* Determine how many packets and what bytes were acked, tso and else */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3038 | if (after(scb->end_seq, tp->snd_una)) { |
Ilpo Järvinen | 13fcf85 | 2007-10-09 01:28:45 -0700 | [diff] [blame] | 3039 | if (tcp_skb_pcount(skb) == 1 || |
| 3040 | !after(tp->snd_una, scb->seq)) |
| 3041 | break; |
| 3042 | |
Ilpo Järvinen | 7201883 | 2007-12-30 04:37:55 -0800 | [diff] [blame] | 3043 | acked_pcount = tcp_tso_acked(sk, skb); |
| 3044 | if (!acked_pcount) |
Ilpo Järvinen | 13fcf85 | 2007-10-09 01:28:45 -0700 | [diff] [blame] | 3045 | break; |
| 3046 | |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 3047 | fully_acked = false; |
Ilpo Järvinen | 13fcf85 | 2007-10-09 01:28:45 -0700 | [diff] [blame] | 3048 | } else { |
Ilpo Järvinen | 7201883 | 2007-12-30 04:37:55 -0800 | [diff] [blame] | 3049 | acked_pcount = tcp_skb_pcount(skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3050 | } |
| 3051 | |
Ilpo Järvinen | 89d478f | 2007-12-30 04:35:27 -0800 | [diff] [blame] | 3052 | if (sacked & TCPCB_RETRANS) { |
| 3053 | if (sacked & TCPCB_SACKED_RETRANS) |
Ilpo Järvinen | 7201883 | 2007-12-30 04:37:55 -0800 | [diff] [blame] | 3054 | tp->retrans_out -= acked_pcount; |
Ilpo Järvinen | 89d478f | 2007-12-30 04:35:27 -0800 | [diff] [blame] | 3055 | flag |= FLAG_RETRANS_DATA_ACKED; |
Ilpo Järvinen | c7caf8d | 2007-11-10 21:22:18 -0800 | [diff] [blame] | 3056 | } else { |
Gavin McCullagh | 2072c22 | 2007-12-29 19:11:21 -0800 | [diff] [blame] | 3057 | ca_seq_rtt = now - scb->when; |
| 3058 | last_ackt = skb->tstamp; |
Ilpo Järvinen | c7caf8d | 2007-11-10 21:22:18 -0800 | [diff] [blame] | 3059 | if (seq_rtt < 0) { |
Gavin McCullagh | 2072c22 | 2007-12-29 19:11:21 -0800 | [diff] [blame] | 3060 | seq_rtt = ca_seq_rtt; |
Ilpo Järvinen | c7caf8d | 2007-11-10 21:22:18 -0800 | [diff] [blame] | 3061 | } |
Ilpo Järvinen | 89d478f | 2007-12-30 04:35:27 -0800 | [diff] [blame] | 3062 | if (!(sacked & TCPCB_SACKED_ACKED)) |
Ilpo Järvinen | 7201883 | 2007-12-30 04:37:55 -0800 | [diff] [blame] | 3063 | reord = min(pkts_acked, reord); |
Yuchung Cheng | e33099f | 2013-03-20 13:33:00 +0000 | [diff] [blame] | 3064 | if (!after(scb->end_seq, tp->high_seq)) |
| 3065 | flag |= FLAG_ORIG_SACK_ACKED; |
Stephen Hemminger | 2d2abba | 2005-11-10 16:56:12 -0800 | [diff] [blame] | 3066 | } |
Ilpo Järvinen | 89d478f | 2007-12-30 04:35:27 -0800 | [diff] [blame] | 3067 | |
| 3068 | if (sacked & TCPCB_SACKED_ACKED) |
Ilpo Järvinen | 7201883 | 2007-12-30 04:37:55 -0800 | [diff] [blame] | 3069 | tp->sacked_out -= acked_pcount; |
Ilpo Järvinen | 89d478f | 2007-12-30 04:35:27 -0800 | [diff] [blame] | 3070 | if (sacked & TCPCB_LOST) |
Ilpo Järvinen | 7201883 | 2007-12-30 04:37:55 -0800 | [diff] [blame] | 3071 | tp->lost_out -= acked_pcount; |
Ilpo Järvinen | 89d478f | 2007-12-30 04:35:27 -0800 | [diff] [blame] | 3072 | |
Ilpo Järvinen | 7201883 | 2007-12-30 04:37:55 -0800 | [diff] [blame] | 3073 | tp->packets_out -= acked_pcount; |
| 3074 | pkts_acked += acked_pcount; |
Ilpo Järvinen | 13fcf85 | 2007-10-09 01:28:45 -0700 | [diff] [blame] | 3075 | |
Ilpo Järvinen | 009a2e3 | 2007-09-20 11:34:38 -0700 | [diff] [blame] | 3076 | /* Initial outgoing SYN's get put onto the write_queue |
| 3077 | * just like anything else we transmit. It is not |
| 3078 | * true data, and if we misinform our callers that |
| 3079 | * this ACK acks real data, we will erroneously exit |
| 3080 | * connection startup slow start one packet too |
| 3081 | * quickly. This is severely frowned upon behavior. |
| 3082 | */ |
Eric Dumazet | 4de075e | 2011-09-27 13:25:05 -0400 | [diff] [blame] | 3083 | if (!(scb->tcp_flags & TCPHDR_SYN)) { |
Ilpo Järvinen | 009a2e3 | 2007-09-20 11:34:38 -0700 | [diff] [blame] | 3084 | flag |= FLAG_DATA_ACKED; |
| 3085 | } else { |
| 3086 | flag |= FLAG_SYN_ACKED; |
| 3087 | tp->retrans_stamp = 0; |
| 3088 | } |
| 3089 | |
Ilpo Järvinen | 13fcf85 | 2007-10-09 01:28:45 -0700 | [diff] [blame] | 3090 | if (!fully_acked) |
| 3091 | break; |
| 3092 | |
David S. Miller | fe067e8 | 2007-03-07 12:12:44 -0800 | [diff] [blame] | 3093 | tcp_unlink_write_queue(skb, sk); |
Hideo Aoki | 3ab224b | 2007-12-31 00:11:19 -0800 | [diff] [blame] | 3094 | sk_wmem_free_skb(sk, skb); |
Ilpo Järvinen | ef9da47 | 2008-09-20 21:25:15 -0700 | [diff] [blame] | 3095 | if (skb == tp->retransmit_skb_hint) |
| 3096 | tp->retransmit_skb_hint = NULL; |
Ilpo Järvinen | 90638a0 | 2008-09-20 21:25:52 -0700 | [diff] [blame] | 3097 | if (skb == tp->lost_skb_hint) |
| 3098 | tp->lost_skb_hint = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3099 | } |
| 3100 | |
Ilpo Järvinen | 33f5f57 | 2008-10-07 14:43:06 -0700 | [diff] [blame] | 3101 | if (likely(between(tp->snd_up, prior_snd_una, tp->snd_una))) |
| 3102 | tp->snd_up = tp->snd_una; |
| 3103 | |
Ilpo Järvinen | cadbd03 | 2007-12-31 04:49:21 -0800 | [diff] [blame] | 3104 | if (skb && (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)) |
| 3105 | flag |= FLAG_SACK_RENEGING; |
| 3106 | |
Yuchung Cheng | ed08495 | 2013-07-22 16:20:48 -0700 | [diff] [blame] | 3107 | if (tcp_ack_update_rtt(sk, flag, seq_rtt, sack_rtt) || |
| 3108 | (flag & FLAG_ACKED)) |
| 3109 | tcp_rearm_rto(sk); |
| 3110 | |
Ilpo Järvinen | 7c46a03 | 2007-09-20 11:33:43 -0700 | [diff] [blame] | 3111 | if (flag & FLAG_ACKED) { |
Stephen Hemminger | 164891a | 2007-04-23 22:26:16 -0700 | [diff] [blame] | 3112 | const struct tcp_congestion_ops *ca_ops |
| 3113 | = inet_csk(sk)->icsk_ca_ops; |
| 3114 | |
Ilpo Järvinen | 72211e9 | 2009-03-14 14:23:04 +0000 | [diff] [blame] | 3115 | if (unlikely(icsk->icsk_mtup.probe_size && |
| 3116 | !after(tp->mtu_probe.probe_seq_end, tp->snd_una))) { |
| 3117 | tcp_mtup_probe_success(sk); |
| 3118 | } |
| 3119 | |
Ilpo Järvinen | c7caf8d | 2007-11-10 21:22:18 -0800 | [diff] [blame] | 3120 | if (tcp_is_reno(tp)) { |
| 3121 | tcp_remove_reno_sacks(sk, pkts_acked); |
| 3122 | } else { |
Ilpo Järvinen | 59a08cb | 2009-02-28 04:44:28 +0000 | [diff] [blame] | 3123 | int delta; |
| 3124 | |
Ilpo Järvinen | c7caf8d | 2007-11-10 21:22:18 -0800 | [diff] [blame] | 3125 | /* Non-retransmitted hole got filled? That's reordering */ |
| 3126 | if (reord < prior_fackets) |
| 3127 | tcp_update_reordering(sk, tp->fackets_out - reord, 0); |
Ilpo Järvinen | 90638a0 | 2008-09-20 21:25:52 -0700 | [diff] [blame] | 3128 | |
Ilpo Järvinen | 59a08cb | 2009-02-28 04:44:28 +0000 | [diff] [blame] | 3129 | delta = tcp_is_fack(tp) ? pkts_acked : |
| 3130 | prior_sacked - tp->sacked_out; |
| 3131 | tp->lost_cnt_hint -= min(tp->lost_cnt_hint, delta); |
Ilpo Järvinen | c7caf8d | 2007-11-10 21:22:18 -0800 | [diff] [blame] | 3132 | } |
| 3133 | |
Ilpo Järvinen | 91fed7a | 2007-10-09 01:24:15 -0700 | [diff] [blame] | 3134 | tp->fackets_out -= min(pkts_acked, tp->fackets_out); |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 3135 | |
Stephen Hemminger | 30cfd0b | 2007-07-25 23:49:34 -0700 | [diff] [blame] | 3136 | if (ca_ops->pkts_acked) { |
| 3137 | s32 rtt_us = -1; |
Ilpo Järvinen | b9ce204 | 2007-06-15 15:08:43 -0700 | [diff] [blame] | 3138 | |
Stephen Hemminger | 30cfd0b | 2007-07-25 23:49:34 -0700 | [diff] [blame] | 3139 | /* Is the ACK triggering packet unambiguous? */ |
Ilpo Järvinen | 7c46a03 | 2007-09-20 11:33:43 -0700 | [diff] [blame] | 3140 | if (!(flag & FLAG_RETRANS_DATA_ACKED)) { |
Stephen Hemminger | 30cfd0b | 2007-07-25 23:49:34 -0700 | [diff] [blame] | 3141 | /* High resolution needed and available? */ |
| 3142 | if (ca_ops->flags & TCP_CONG_RTT_STAMP && |
| 3143 | !ktime_equal(last_ackt, |
| 3144 | net_invalid_timestamp())) |
| 3145 | rtt_us = ktime_us_delta(ktime_get_real(), |
| 3146 | last_ackt); |
stephen hemminger | febf081 | 2011-03-14 07:52:12 +0000 | [diff] [blame] | 3147 | else if (ca_seq_rtt >= 0) |
Gavin McCullagh | 2072c22 | 2007-12-29 19:11:21 -0800 | [diff] [blame] | 3148 | rtt_us = jiffies_to_usecs(ca_seq_rtt); |
Stephen Hemminger | 30cfd0b | 2007-07-25 23:49:34 -0700 | [diff] [blame] | 3149 | } |
| 3150 | |
| 3151 | ca_ops->pkts_acked(sk, pkts_acked, rtt_us); |
| 3152 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3153 | } |
| 3154 | |
| 3155 | #if FASTRETRANS_DEBUG > 0 |
Ilpo Järvinen | 547b792 | 2008-07-25 21:43:18 -0700 | [diff] [blame] | 3156 | WARN_ON((int)tp->sacked_out < 0); |
| 3157 | WARN_ON((int)tp->lost_out < 0); |
| 3158 | WARN_ON((int)tp->retrans_out < 0); |
Ilpo Järvinen | e60402d | 2007-08-09 15:14:46 +0300 | [diff] [blame] | 3159 | if (!tp->packets_out && tcp_is_sack(tp)) { |
Stephen Hemminger | cfcabdc | 2007-10-09 01:59:42 -0700 | [diff] [blame] | 3160 | icsk = inet_csk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3161 | if (tp->lost_out) { |
Joe Perches | 91df42b | 2012-05-15 14:11:54 +0000 | [diff] [blame] | 3162 | pr_debug("Leak l=%u %d\n", |
| 3163 | tp->lost_out, icsk->icsk_ca_state); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3164 | tp->lost_out = 0; |
| 3165 | } |
| 3166 | if (tp->sacked_out) { |
Joe Perches | 91df42b | 2012-05-15 14:11:54 +0000 | [diff] [blame] | 3167 | pr_debug("Leak s=%u %d\n", |
| 3168 | tp->sacked_out, icsk->icsk_ca_state); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3169 | tp->sacked_out = 0; |
| 3170 | } |
| 3171 | if (tp->retrans_out) { |
Joe Perches | 91df42b | 2012-05-15 14:11:54 +0000 | [diff] [blame] | 3172 | pr_debug("Leak r=%u %d\n", |
| 3173 | tp->retrans_out, icsk->icsk_ca_state); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3174 | tp->retrans_out = 0; |
| 3175 | } |
| 3176 | } |
| 3177 | #endif |
Ilpo Järvinen | 7c46a03 | 2007-09-20 11:33:43 -0700 | [diff] [blame] | 3178 | return flag; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3179 | } |
| 3180 | |
| 3181 | static void tcp_ack_probe(struct sock *sk) |
| 3182 | { |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 3183 | const struct tcp_sock *tp = tcp_sk(sk); |
| 3184 | struct inet_connection_sock *icsk = inet_csk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3185 | |
| 3186 | /* Was it a usable window open? */ |
| 3187 | |
Ilpo Järvinen | 90840de | 2007-12-31 04:48:41 -0800 | [diff] [blame] | 3188 | if (!after(TCP_SKB_CB(tcp_send_head(sk))->end_seq, tcp_wnd_end(tp))) { |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 3189 | icsk->icsk_backoff = 0; |
| 3190 | inet_csk_clear_xmit_timer(sk, ICSK_TIME_PROBE0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3191 | /* Socket must be waked up by subsequent tcp_data_snd_check(). |
| 3192 | * This function is not for random using! |
| 3193 | */ |
| 3194 | } else { |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 3195 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, |
Arnaldo Carvalho de Melo | 3f421ba | 2005-08-09 20:11:08 -0700 | [diff] [blame] | 3196 | min(icsk->icsk_rto << icsk->icsk_backoff, TCP_RTO_MAX), |
| 3197 | TCP_RTO_MAX); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3198 | } |
| 3199 | } |
| 3200 | |
Vijay Subramanian | 67b95bd | 2012-07-19 21:32:18 +0000 | [diff] [blame] | 3201 | static inline bool tcp_ack_is_dubious(const struct sock *sk, const int flag) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3202 | { |
Eric Dumazet | a02cec2 | 2010-09-22 20:43:57 +0000 | [diff] [blame] | 3203 | return !(flag & FLAG_NOT_DUP) || (flag & FLAG_CA_ALERT) || |
| 3204 | inet_csk(sk)->icsk_ca_state != TCP_CA_Open; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3205 | } |
| 3206 | |
Yuchung Cheng | 0f7cc9a | 2013-08-21 17:29:23 -0700 | [diff] [blame] | 3207 | /* Decide wheather to run the increase function of congestion control. */ |
Vijay Subramanian | 67b95bd | 2012-07-19 21:32:18 +0000 | [diff] [blame] | 3208 | static inline bool tcp_may_raise_cwnd(const struct sock *sk, const int flag) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3209 | { |
Yuchung Cheng | 0f7cc9a | 2013-08-21 17:29:23 -0700 | [diff] [blame] | 3210 | if (tcp_in_cwnd_reduction(sk)) |
| 3211 | return false; |
| 3212 | |
| 3213 | /* If reordering is high then always grow cwnd whenever data is |
| 3214 | * delivered regardless of its ordering. Otherwise stay conservative |
Yuchung Cheng | 16edfe7 | 2013-09-05 15:36:09 -0700 | [diff] [blame] | 3215 | * and only grow cwnd on in-order delivery (RFC5681). A stretched ACK w/ |
Yuchung Cheng | 0f7cc9a | 2013-08-21 17:29:23 -0700 | [diff] [blame] | 3216 | * new SACK or ECE mark may first advance cwnd here and later reduce |
| 3217 | * cwnd in tcp_fastretrans_alert() based on more states. |
| 3218 | */ |
| 3219 | if (tcp_sk(sk)->reordering > sysctl_tcp_reordering) |
| 3220 | return flag & FLAG_FORWARD_PROGRESS; |
| 3221 | |
Yuchung Cheng | 16edfe7 | 2013-09-05 15:36:09 -0700 | [diff] [blame] | 3222 | return flag & FLAG_DATA_ACKED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3223 | } |
| 3224 | |
| 3225 | /* Check that window update is acceptable. |
| 3226 | * The function assumes that snd_una<=ack<=snd_next. |
| 3227 | */ |
Vijay Subramanian | 67b95bd | 2012-07-19 21:32:18 +0000 | [diff] [blame] | 3228 | static inline bool tcp_may_update_window(const struct tcp_sock *tp, |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 3229 | const u32 ack, const u32 ack_seq, |
| 3230 | const u32 nwin) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3231 | { |
Eric Dumazet | a02cec2 | 2010-09-22 20:43:57 +0000 | [diff] [blame] | 3232 | return after(ack, tp->snd_una) || |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3233 | after(ack_seq, tp->snd_wl1) || |
Eric Dumazet | a02cec2 | 2010-09-22 20:43:57 +0000 | [diff] [blame] | 3234 | (ack_seq == tp->snd_wl1 && nwin > tp->snd_wnd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3235 | } |
| 3236 | |
| 3237 | /* Update our send window. |
| 3238 | * |
| 3239 | * Window update algorithm, described in RFC793/RFC1122 (used in linux-2.2 |
| 3240 | * and in FreeBSD. NetBSD's one is even worse.) is wrong. |
| 3241 | */ |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 3242 | static int tcp_ack_update_window(struct sock *sk, const struct sk_buff *skb, u32 ack, |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 3243 | u32 ack_seq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3244 | { |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 3245 | struct tcp_sock *tp = tcp_sk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3246 | int flag = 0; |
Arnaldo Carvalho de Melo | aa8223c | 2007-04-10 21:04:22 -0700 | [diff] [blame] | 3247 | u32 nwin = ntohs(tcp_hdr(skb)->window); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3248 | |
Arnaldo Carvalho de Melo | aa8223c | 2007-04-10 21:04:22 -0700 | [diff] [blame] | 3249 | if (likely(!tcp_hdr(skb)->syn)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3250 | nwin <<= tp->rx_opt.snd_wscale; |
| 3251 | |
| 3252 | if (tcp_may_update_window(tp, ack, ack_seq, nwin)) { |
| 3253 | flag |= FLAG_WIN_UPDATE; |
Hantzis Fotis | ee7537b | 2009-03-02 22:42:02 -0800 | [diff] [blame] | 3254 | tcp_update_wl(tp, ack_seq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3255 | |
| 3256 | if (tp->snd_wnd != nwin) { |
| 3257 | tp->snd_wnd = nwin; |
| 3258 | |
| 3259 | /* Note, it is the only place, where |
| 3260 | * fast path is recovered for sending TCP. |
| 3261 | */ |
Herbert Xu | 2ad4106 | 2005-10-27 18:47:46 +1000 | [diff] [blame] | 3262 | tp->pred_flags = 0; |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 3263 | tcp_fast_path_check(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3264 | |
| 3265 | if (nwin > tp->max_window) { |
| 3266 | tp->max_window = nwin; |
Arnaldo Carvalho de Melo | d83d846 | 2005-12-13 23:26:10 -0800 | [diff] [blame] | 3267 | tcp_sync_mss(sk, inet_csk(sk)->icsk_pmtu_cookie); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3268 | } |
| 3269 | } |
| 3270 | } |
| 3271 | |
| 3272 | tp->snd_una = ack; |
| 3273 | |
| 3274 | return flag; |
| 3275 | } |
| 3276 | |
Eric Dumazet | 354e4aa | 2012-10-21 19:57:11 +0000 | [diff] [blame] | 3277 | /* RFC 5961 7 [ACK Throttling] */ |
| 3278 | static void tcp_send_challenge_ack(struct sock *sk) |
| 3279 | { |
| 3280 | /* unprotected vars, we dont care of overwrites */ |
| 3281 | static u32 challenge_timestamp; |
| 3282 | static unsigned int challenge_count; |
| 3283 | u32 now = jiffies / HZ; |
| 3284 | |
| 3285 | if (now != challenge_timestamp) { |
| 3286 | challenge_timestamp = now; |
| 3287 | challenge_count = 0; |
| 3288 | } |
| 3289 | if (++challenge_count <= sysctl_tcp_challenge_ack_limit) { |
| 3290 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPCHALLENGEACK); |
| 3291 | tcp_send_ack(sk); |
| 3292 | } |
| 3293 | } |
| 3294 | |
Eric Dumazet | 12fb3dd | 2013-04-19 07:19:48 +0000 | [diff] [blame] | 3295 | static void tcp_store_ts_recent(struct tcp_sock *tp) |
| 3296 | { |
| 3297 | tp->rx_opt.ts_recent = tp->rx_opt.rcv_tsval; |
| 3298 | tp->rx_opt.ts_recent_stamp = get_seconds(); |
| 3299 | } |
| 3300 | |
| 3301 | static void tcp_replace_ts_recent(struct tcp_sock *tp, u32 seq) |
| 3302 | { |
| 3303 | if (tp->rx_opt.saw_tstamp && !after(seq, tp->rcv_wup)) { |
| 3304 | /* PAWS bug workaround wrt. ACK frames, the PAWS discard |
| 3305 | * extra check below makes sure this can only happen |
| 3306 | * for pure ACK frames. -DaveM |
| 3307 | * |
| 3308 | * Not only, also it occurs for expired timestamps. |
| 3309 | */ |
| 3310 | |
| 3311 | if (tcp_paws_check(&tp->rx_opt, 0)) |
| 3312 | tcp_store_ts_recent(tp); |
| 3313 | } |
| 3314 | } |
| 3315 | |
Nandita Dukkipati | 9b717a8 | 2013-03-11 10:00:44 +0000 | [diff] [blame] | 3316 | /* This routine deals with acks during a TLP episode. |
| 3317 | * Ref: loss detection algorithm in draft-dukkipati-tcpm-tcp-loss-probe. |
| 3318 | */ |
| 3319 | static void tcp_process_tlp_ack(struct sock *sk, u32 ack, int flag) |
| 3320 | { |
| 3321 | struct tcp_sock *tp = tcp_sk(sk); |
| 3322 | bool is_tlp_dupack = (ack == tp->tlp_high_seq) && |
| 3323 | !(flag & (FLAG_SND_UNA_ADVANCED | |
| 3324 | FLAG_NOT_DUP | FLAG_DATA_SACKED)); |
| 3325 | |
| 3326 | /* Mark the end of TLP episode on receiving TLP dupack or when |
| 3327 | * ack is after tlp_high_seq. |
| 3328 | */ |
| 3329 | if (is_tlp_dupack) { |
| 3330 | tp->tlp_high_seq = 0; |
| 3331 | return; |
| 3332 | } |
| 3333 | |
| 3334 | if (after(ack, tp->tlp_high_seq)) { |
| 3335 | tp->tlp_high_seq = 0; |
| 3336 | /* Don't reduce cwnd if DSACK arrives for TLP retrans. */ |
| 3337 | if (!(flag & FLAG_DSACKING_ACK)) { |
| 3338 | tcp_init_cwnd_reduction(sk, true); |
| 3339 | tcp_set_ca_state(sk, TCP_CA_CWR); |
| 3340 | tcp_end_cwnd_reduction(sk); |
Yuchung Cheng | 031afe4 | 2013-10-12 10:16:27 -0700 | [diff] [blame] | 3341 | tcp_try_keep_open(sk); |
Nandita Dukkipati | 9b717a8 | 2013-03-11 10:00:44 +0000 | [diff] [blame] | 3342 | NET_INC_STATS_BH(sock_net(sk), |
| 3343 | LINUX_MIB_TCPLOSSPROBERECOVERY); |
| 3344 | } |
| 3345 | } |
| 3346 | } |
| 3347 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3348 | /* This routine deals with incoming acks, but not outgoing ones. */ |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 3349 | static int tcp_ack(struct sock *sk, const struct sk_buff *skb, int flag) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3350 | { |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 3351 | struct inet_connection_sock *icsk = inet_csk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3352 | struct tcp_sock *tp = tcp_sk(sk); |
| 3353 | u32 prior_snd_una = tp->snd_una; |
| 3354 | u32 ack_seq = TCP_SKB_CB(skb)->seq; |
| 3355 | u32 ack = TCP_SKB_CB(skb)->ack_seq; |
Neal Cardwell | 7d2b55f | 2011-11-16 08:58:01 +0000 | [diff] [blame] | 3356 | bool is_dupack = false; |
Eric Dumazet | 95bd09e | 2013-08-27 05:46:32 -0700 | [diff] [blame] | 3357 | u32 prior_in_flight, prior_cwnd = tp->snd_cwnd, prior_rtt = tp->srtt; |
Ilpo Järvinen | c7caf8d | 2007-11-10 21:22:18 -0800 | [diff] [blame] | 3358 | u32 prior_fackets; |
Nandita Dukkipati | 35f079e | 2013-05-21 15:12:07 +0000 | [diff] [blame] | 3359 | int prior_packets = tp->packets_out; |
Yuchung Cheng | 6804973 | 2013-05-29 14:20:11 +0000 | [diff] [blame] | 3360 | const int prior_unsacked = tp->packets_out - tp->sacked_out; |
| 3361 | int acked = 0; /* Number of packets newly acked */ |
Yuchung Cheng | 59c9af4 | 2013-07-22 16:20:47 -0700 | [diff] [blame] | 3362 | s32 sack_rtt = -1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3363 | |
John Dykstra | 96e0bf4 | 2009-03-22 21:49:57 -0700 | [diff] [blame] | 3364 | /* If the ack is older than previous acks |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3365 | * then we can probably ignore it. |
| 3366 | */ |
Eric Dumazet | 354e4aa | 2012-10-21 19:57:11 +0000 | [diff] [blame] | 3367 | if (before(ack, prior_snd_una)) { |
| 3368 | /* RFC 5961 5.2 [Blind Data Injection Attack].[Mitigation] */ |
| 3369 | if (before(ack, prior_snd_una - tp->max_window)) { |
| 3370 | tcp_send_challenge_ack(sk); |
| 3371 | return -1; |
| 3372 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3373 | goto old_ack; |
Eric Dumazet | 354e4aa | 2012-10-21 19:57:11 +0000 | [diff] [blame] | 3374 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3375 | |
John Dykstra | 96e0bf4 | 2009-03-22 21:49:57 -0700 | [diff] [blame] | 3376 | /* If the ack includes data we haven't sent yet, discard |
| 3377 | * this segment (RFC793 Section 3.9). |
| 3378 | */ |
| 3379 | if (after(ack, tp->snd_nxt)) |
| 3380 | goto invalid_ack; |
| 3381 | |
Nandita Dukkipati | 6ba8a3b | 2013-03-11 10:00:43 +0000 | [diff] [blame] | 3382 | if (icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS || |
| 3383 | icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) |
Yuchung Cheng | 750ea2b | 2012-05-02 13:30:04 +0000 | [diff] [blame] | 3384 | tcp_rearm_rto(sk); |
| 3385 | |
Ilpo Järvinen | 2e60529 | 2007-08-02 19:46:58 -0700 | [diff] [blame] | 3386 | if (after(ack, prior_snd_una)) |
| 3387 | flag |= FLAG_SND_UNA_ADVANCED; |
| 3388 | |
Ilpo Järvinen | c7caf8d | 2007-11-10 21:22:18 -0800 | [diff] [blame] | 3389 | prior_fackets = tp->fackets_out; |
Ilpo Järvinen | 52d3408 | 2007-12-05 02:21:35 -0800 | [diff] [blame] | 3390 | prior_in_flight = tcp_packets_in_flight(tp); |
Ilpo Järvinen | c7caf8d | 2007-11-10 21:22:18 -0800 | [diff] [blame] | 3391 | |
Eric Dumazet | 12fb3dd | 2013-04-19 07:19:48 +0000 | [diff] [blame] | 3392 | /* ts_recent update must be made after we are sure that the packet |
| 3393 | * is in window. |
| 3394 | */ |
| 3395 | if (flag & FLAG_UPDATE_TS_RECENT) |
| 3396 | tcp_replace_ts_recent(tp, TCP_SKB_CB(skb)->seq); |
| 3397 | |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 3398 | if (!(flag & FLAG_SLOWPATH) && after(ack, prior_snd_una)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3399 | /* Window is constant, pure forward advance. |
| 3400 | * No more checks are required. |
| 3401 | * Note, we use the fact that SND.UNA>=SND.WL2. |
| 3402 | */ |
Hantzis Fotis | ee7537b | 2009-03-02 22:42:02 -0800 | [diff] [blame] | 3403 | tcp_update_wl(tp, ack_seq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3404 | tp->snd_una = ack; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3405 | flag |= FLAG_WIN_UPDATE; |
| 3406 | |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 3407 | tcp_ca_event(sk, CA_EVENT_FAST_ACK); |
Stephen Hemminger | 317a76f | 2005-06-23 12:19:55 -0700 | [diff] [blame] | 3408 | |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 3409 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPHPACKS); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3410 | } else { |
| 3411 | if (ack_seq != TCP_SKB_CB(skb)->end_seq) |
| 3412 | flag |= FLAG_DATA; |
| 3413 | else |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 3414 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPPUREACKS); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3415 | |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 3416 | flag |= tcp_ack_update_window(sk, skb, ack, ack_seq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3417 | |
| 3418 | if (TCP_SKB_CB(skb)->sacked) |
Yuchung Cheng | 59c9af4 | 2013-07-22 16:20:47 -0700 | [diff] [blame] | 3419 | flag |= tcp_sacktag_write_queue(sk, skb, prior_snd_una, |
| 3420 | &sack_rtt); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3421 | |
Arnaldo Carvalho de Melo | aa8223c | 2007-04-10 21:04:22 -0700 | [diff] [blame] | 3422 | if (TCP_ECN_rcv_ecn_echo(tp, tcp_hdr(skb))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3423 | flag |= FLAG_ECE; |
| 3424 | |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 3425 | tcp_ca_event(sk, CA_EVENT_SLOW_ACK); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3426 | } |
| 3427 | |
| 3428 | /* We passed data and got it acked, remove any soft error |
| 3429 | * log. Something worked... |
| 3430 | */ |
| 3431 | sk->sk_err_soft = 0; |
David S. Miller | 4b53fb6 | 2008-07-23 16:38:45 -0700 | [diff] [blame] | 3432 | icsk->icsk_probes_out = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3433 | tp->rcv_tstamp = tcp_time_stamp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3434 | if (!prior_packets) |
| 3435 | goto no_queue; |
| 3436 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3437 | /* See if we can take anything off of the retransmit queue. */ |
Yuchung Cheng | 6804973 | 2013-05-29 14:20:11 +0000 | [diff] [blame] | 3438 | acked = tp->packets_out; |
Yuchung Cheng | ed08495 | 2013-07-22 16:20:48 -0700 | [diff] [blame] | 3439 | flag |= tcp_clean_rtx_queue(sk, prior_fackets, prior_snd_una, sack_rtt); |
Yuchung Cheng | 6804973 | 2013-05-29 14:20:11 +0000 | [diff] [blame] | 3440 | acked -= tp->packets_out; |
Nandita Dukkipati | a262f0c | 2011-08-21 20:21:57 +0000 | [diff] [blame] | 3441 | |
Yuchung Cheng | 0f7cc9a | 2013-08-21 17:29:23 -0700 | [diff] [blame] | 3442 | /* Advance cwnd if state allows */ |
| 3443 | if (tcp_may_raise_cwnd(sk, flag)) |
| 3444 | tcp_cong_avoid(sk, ack, prior_in_flight); |
| 3445 | |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 3446 | if (tcp_ack_is_dubious(sk, flag)) { |
Neal Cardwell | 7d2b55f | 2011-11-16 08:58:01 +0000 | [diff] [blame] | 3447 | is_dupack = !(flag & (FLAG_SND_UNA_ADVANCED | FLAG_NOT_DUP)); |
Yuchung Cheng | 6804973 | 2013-05-29 14:20:11 +0000 | [diff] [blame] | 3448 | tcp_fastretrans_alert(sk, acked, prior_unsacked, |
| 3449 | is_dupack, flag); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3450 | } |
Nandita Dukkipati | 9b717a8 | 2013-03-11 10:00:44 +0000 | [diff] [blame] | 3451 | if (tp->tlp_high_seq) |
| 3452 | tcp_process_tlp_ack(sk, ack, flag); |
| 3453 | |
David S. Miller | 5110effe | 2012-07-02 02:21:03 -0700 | [diff] [blame] | 3454 | if ((flag & FLAG_FORWARD_PROGRESS) || !(flag & FLAG_NOT_DUP)) { |
| 3455 | struct dst_entry *dst = __sk_dst_get(sk); |
| 3456 | if (dst) |
| 3457 | dst_confirm(dst); |
| 3458 | } |
Nandita Dukkipati | 6ba8a3b | 2013-03-11 10:00:43 +0000 | [diff] [blame] | 3459 | |
| 3460 | if (icsk->icsk_pending == ICSK_TIME_RETRANS) |
| 3461 | tcp_schedule_loss_probe(sk); |
Eric Dumazet | 95bd09e | 2013-08-27 05:46:32 -0700 | [diff] [blame] | 3462 | if (tp->srtt != prior_rtt || tp->snd_cwnd != prior_cwnd) |
| 3463 | tcp_update_pacing_rate(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3464 | return 1; |
| 3465 | |
| 3466 | no_queue: |
Neal Cardwell | 5628adf | 2011-11-16 08:58:02 +0000 | [diff] [blame] | 3467 | /* If data was DSACKed, see if we can undo a cwnd reduction. */ |
| 3468 | if (flag & FLAG_DSACKING_ACK) |
Yuchung Cheng | 6804973 | 2013-05-29 14:20:11 +0000 | [diff] [blame] | 3469 | tcp_fastretrans_alert(sk, acked, prior_unsacked, |
| 3470 | is_dupack, flag); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3471 | /* If this ack opens up a zero window, clear backoff. It was |
| 3472 | * being used to time the probes, and is probably far higher than |
| 3473 | * it needs to be for normal retransmission. |
| 3474 | */ |
David S. Miller | fe067e8 | 2007-03-07 12:12:44 -0800 | [diff] [blame] | 3475 | if (tcp_send_head(sk)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3476 | tcp_ack_probe(sk); |
Nandita Dukkipati | 9b717a8 | 2013-03-11 10:00:44 +0000 | [diff] [blame] | 3477 | |
| 3478 | if (tp->tlp_high_seq) |
| 3479 | tcp_process_tlp_ack(sk, ack, flag); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3480 | return 1; |
| 3481 | |
John Dykstra | 96e0bf4 | 2009-03-22 21:49:57 -0700 | [diff] [blame] | 3482 | invalid_ack: |
| 3483 | SOCK_DEBUG(sk, "Ack %u after %u:%u\n", ack, tp->snd_una, tp->snd_nxt); |
| 3484 | return -1; |
| 3485 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3486 | old_ack: |
Neal Cardwell | e95ae2f | 2011-11-16 08:58:03 +0000 | [diff] [blame] | 3487 | /* If data was SACKed, tag it and see if we should send more data. |
| 3488 | * If data was DSACKed, see if we can undo a cwnd reduction. |
| 3489 | */ |
Ilpo Järvinen | 8aca6cb | 2008-06-04 11:34:22 -0700 | [diff] [blame] | 3490 | if (TCP_SKB_CB(skb)->sacked) { |
Yuchung Cheng | 59c9af4 | 2013-07-22 16:20:47 -0700 | [diff] [blame] | 3491 | flag |= tcp_sacktag_write_queue(sk, skb, prior_snd_una, |
| 3492 | &sack_rtt); |
Yuchung Cheng | 6804973 | 2013-05-29 14:20:11 +0000 | [diff] [blame] | 3493 | tcp_fastretrans_alert(sk, acked, prior_unsacked, |
| 3494 | is_dupack, flag); |
Ilpo Järvinen | 8aca6cb | 2008-06-04 11:34:22 -0700 | [diff] [blame] | 3495 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3496 | |
John Dykstra | 96e0bf4 | 2009-03-22 21:49:57 -0700 | [diff] [blame] | 3497 | SOCK_DEBUG(sk, "Ack %u before %u:%u\n", ack, tp->snd_una, tp->snd_nxt); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3498 | return 0; |
| 3499 | } |
| 3500 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3501 | /* Look for tcp options. Normally only called on SYN and SYNACK packets. |
| 3502 | * But, this can also be called on packets in the established flow when |
| 3503 | * the fast version below fails. |
| 3504 | */ |
Christoph Paasch | 1a2c618 | 2013-03-17 08:23:34 +0000 | [diff] [blame] | 3505 | void tcp_parse_options(const struct sk_buff *skb, |
| 3506 | struct tcp_options_received *opt_rx, int estab, |
Yuchung Cheng | 2100c8d | 2012-07-19 06:43:05 +0000 | [diff] [blame] | 3507 | struct tcp_fastopen_cookie *foc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3508 | { |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 3509 | const unsigned char *ptr; |
| 3510 | const struct tcphdr *th = tcp_hdr(skb); |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 3511 | int length = (th->doff * 4) - sizeof(struct tcphdr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3512 | |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 3513 | ptr = (const unsigned char *)(th + 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3514 | opt_rx->saw_tstamp = 0; |
| 3515 | |
Stephen Hemminger | 2de979b | 2007-03-08 20:45:19 -0800 | [diff] [blame] | 3516 | while (length > 0) { |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 3517 | int opcode = *ptr++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3518 | int opsize; |
| 3519 | |
| 3520 | switch (opcode) { |
Ilpo Järvinen | f038ac8 | 2008-01-03 20:36:55 -0800 | [diff] [blame] | 3521 | case TCPOPT_EOL: |
| 3522 | return; |
| 3523 | case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */ |
| 3524 | length--; |
| 3525 | continue; |
| 3526 | default: |
| 3527 | opsize = *ptr++; |
| 3528 | if (opsize < 2) /* "silly options" */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3529 | return; |
Ilpo Järvinen | f038ac8 | 2008-01-03 20:36:55 -0800 | [diff] [blame] | 3530 | if (opsize > length) |
| 3531 | return; /* don't parse partial options */ |
| 3532 | switch (opcode) { |
| 3533 | case TCPOPT_MSS: |
| 3534 | if (opsize == TCPOLEN_MSS && th->syn && !estab) { |
Harvey Harrison | d3e2ce3 | 2008-05-02 16:26:16 -0700 | [diff] [blame] | 3535 | u16 in_mss = get_unaligned_be16(ptr); |
Ilpo Järvinen | f038ac8 | 2008-01-03 20:36:55 -0800 | [diff] [blame] | 3536 | if (in_mss) { |
| 3537 | if (opt_rx->user_mss && |
| 3538 | opt_rx->user_mss < in_mss) |
| 3539 | in_mss = opt_rx->user_mss; |
| 3540 | opt_rx->mss_clamp = in_mss; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3541 | } |
Stephen Hemminger | 3ff50b7 | 2007-04-20 17:09:22 -0700 | [diff] [blame] | 3542 | } |
Ilpo Järvinen | f038ac8 | 2008-01-03 20:36:55 -0800 | [diff] [blame] | 3543 | break; |
| 3544 | case TCPOPT_WINDOW: |
| 3545 | if (opsize == TCPOLEN_WINDOW && th->syn && |
David S. Miller | bb5b7c1 | 2009-12-15 20:56:42 -0800 | [diff] [blame] | 3546 | !estab && sysctl_tcp_window_scaling) { |
Ilpo Järvinen | f038ac8 | 2008-01-03 20:36:55 -0800 | [diff] [blame] | 3547 | __u8 snd_wscale = *(__u8 *)ptr; |
| 3548 | opt_rx->wscale_ok = 1; |
| 3549 | if (snd_wscale > 14) { |
Joe Perches | e87cc47 | 2012-05-13 21:56:26 +0000 | [diff] [blame] | 3550 | net_info_ratelimited("%s: Illegal window scaling value %d >14 received\n", |
| 3551 | __func__, |
| 3552 | snd_wscale); |
Ilpo Järvinen | f038ac8 | 2008-01-03 20:36:55 -0800 | [diff] [blame] | 3553 | snd_wscale = 14; |
| 3554 | } |
| 3555 | opt_rx->snd_wscale = snd_wscale; |
| 3556 | } |
| 3557 | break; |
| 3558 | case TCPOPT_TIMESTAMP: |
| 3559 | if ((opsize == TCPOLEN_TIMESTAMP) && |
| 3560 | ((estab && opt_rx->tstamp_ok) || |
David S. Miller | bb5b7c1 | 2009-12-15 20:56:42 -0800 | [diff] [blame] | 3561 | (!estab && sysctl_tcp_timestamps))) { |
Ilpo Järvinen | f038ac8 | 2008-01-03 20:36:55 -0800 | [diff] [blame] | 3562 | opt_rx->saw_tstamp = 1; |
Harvey Harrison | d3e2ce3 | 2008-05-02 16:26:16 -0700 | [diff] [blame] | 3563 | opt_rx->rcv_tsval = get_unaligned_be32(ptr); |
| 3564 | opt_rx->rcv_tsecr = get_unaligned_be32(ptr + 4); |
Ilpo Järvinen | f038ac8 | 2008-01-03 20:36:55 -0800 | [diff] [blame] | 3565 | } |
| 3566 | break; |
| 3567 | case TCPOPT_SACK_PERM: |
| 3568 | if (opsize == TCPOLEN_SACK_PERM && th->syn && |
David S. Miller | bb5b7c1 | 2009-12-15 20:56:42 -0800 | [diff] [blame] | 3569 | !estab && sysctl_tcp_sack) { |
Vijay Subramanian | ab56222 | 2011-12-20 13:23:24 +0000 | [diff] [blame] | 3570 | opt_rx->sack_ok = TCP_SACK_SEEN; |
Ilpo Järvinen | f038ac8 | 2008-01-03 20:36:55 -0800 | [diff] [blame] | 3571 | tcp_sack_reset(opt_rx); |
| 3572 | } |
| 3573 | break; |
Stephen Hemminger | 3ff50b7 | 2007-04-20 17:09:22 -0700 | [diff] [blame] | 3574 | |
Ilpo Järvinen | f038ac8 | 2008-01-03 20:36:55 -0800 | [diff] [blame] | 3575 | case TCPOPT_SACK: |
| 3576 | if ((opsize >= (TCPOLEN_SACK_BASE + TCPOLEN_SACK_PERBLOCK)) && |
| 3577 | !((opsize - TCPOLEN_SACK_BASE) % TCPOLEN_SACK_PERBLOCK) && |
| 3578 | opt_rx->sack_ok) { |
| 3579 | TCP_SKB_CB(skb)->sacked = (ptr - 2) - (unsigned char *)th; |
| 3580 | } |
| 3581 | break; |
| 3582 | #ifdef CONFIG_TCP_MD5SIG |
| 3583 | case TCPOPT_MD5SIG: |
| 3584 | /* |
| 3585 | * The MD5 Hash has already been |
| 3586 | * checked (see tcp_v{4,6}_do_rcv()). |
| 3587 | */ |
| 3588 | break; |
| 3589 | #endif |
Yuchung Cheng | 2100c8d | 2012-07-19 06:43:05 +0000 | [diff] [blame] | 3590 | case TCPOPT_EXP: |
| 3591 | /* Fast Open option shares code 254 using a |
| 3592 | * 16 bits magic number. It's valid only in |
| 3593 | * SYN or SYN-ACK with an even size. |
| 3594 | */ |
| 3595 | if (opsize < TCPOLEN_EXP_FASTOPEN_BASE || |
| 3596 | get_unaligned_be16(ptr) != TCPOPT_FASTOPEN_MAGIC || |
| 3597 | foc == NULL || !th->syn || (opsize & 1)) |
| 3598 | break; |
| 3599 | foc->len = opsize - TCPOLEN_EXP_FASTOPEN_BASE; |
| 3600 | if (foc->len >= TCP_FASTOPEN_COOKIE_MIN && |
| 3601 | foc->len <= TCP_FASTOPEN_COOKIE_MAX) |
| 3602 | memcpy(foc->val, ptr + 2, foc->len); |
| 3603 | else if (foc->len != 0) |
| 3604 | foc->len = -1; |
| 3605 | break; |
| 3606 | |
| 3607 | } |
Ilpo Järvinen | f038ac8 | 2008-01-03 20:36:55 -0800 | [diff] [blame] | 3608 | ptr += opsize-2; |
| 3609 | length -= opsize; |
Stephen Hemminger | 3ff50b7 | 2007-04-20 17:09:22 -0700 | [diff] [blame] | 3610 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3611 | } |
| 3612 | } |
Eric Dumazet | 4bc2f18 | 2010-07-09 21:22:10 +0000 | [diff] [blame] | 3613 | EXPORT_SYMBOL(tcp_parse_options); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3614 | |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 3615 | static bool tcp_parse_aligned_timestamp(struct tcp_sock *tp, const struct tcphdr *th) |
Ilpo Järvinen | a4356b2 | 2008-08-23 05:12:29 -0700 | [diff] [blame] | 3616 | { |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 3617 | const __be32 *ptr = (const __be32 *)(th + 1); |
Ilpo Järvinen | a4356b2 | 2008-08-23 05:12:29 -0700 | [diff] [blame] | 3618 | |
| 3619 | if (*ptr == htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
| 3620 | | (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP)) { |
| 3621 | tp->rx_opt.saw_tstamp = 1; |
| 3622 | ++ptr; |
| 3623 | tp->rx_opt.rcv_tsval = ntohl(*ptr); |
| 3624 | ++ptr; |
Andrew Vagin | e3e1202 | 2013-08-27 12:21:55 +0400 | [diff] [blame] | 3625 | if (*ptr) |
| 3626 | tp->rx_opt.rcv_tsecr = ntohl(*ptr) - tp->tsoffset; |
| 3627 | else |
| 3628 | tp->rx_opt.rcv_tsecr = 0; |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 3629 | return true; |
Ilpo Järvinen | a4356b2 | 2008-08-23 05:12:29 -0700 | [diff] [blame] | 3630 | } |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 3631 | return false; |
Ilpo Järvinen | a4356b2 | 2008-08-23 05:12:29 -0700 | [diff] [blame] | 3632 | } |
| 3633 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3634 | /* Fast parse options. This hopes to only see timestamps. |
| 3635 | * If it is wrong it falls back on tcp_parse_options(). |
| 3636 | */ |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 3637 | static bool tcp_fast_parse_options(const struct sk_buff *skb, |
Christoph Paasch | 1a2c618 | 2013-03-17 08:23:34 +0000 | [diff] [blame] | 3638 | const struct tcphdr *th, struct tcp_sock *tp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3639 | { |
William Allen Simpson | 4957faade | 2009-12-02 18:25:27 +0000 | [diff] [blame] | 3640 | /* In the spirit of fast parsing, compare doff directly to constant |
| 3641 | * values. Because equality is used, short doff can be ignored here. |
| 3642 | */ |
| 3643 | if (th->doff == (sizeof(*th) / 4)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3644 | tp->rx_opt.saw_tstamp = 0; |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 3645 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3646 | } else if (tp->rx_opt.tstamp_ok && |
William Allen Simpson | 4957faade | 2009-12-02 18:25:27 +0000 | [diff] [blame] | 3647 | th->doff == ((sizeof(*th) + TCPOLEN_TSTAMP_ALIGNED) / 4)) { |
Ilpo Järvinen | a4356b2 | 2008-08-23 05:12:29 -0700 | [diff] [blame] | 3648 | if (tcp_parse_aligned_timestamp(tp, th)) |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 3649 | return true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3650 | } |
Andrey Vagin | ee684b6 | 2013-02-11 05:50:19 +0000 | [diff] [blame] | 3651 | |
Christoph Paasch | 1a2c618 | 2013-03-17 08:23:34 +0000 | [diff] [blame] | 3652 | tcp_parse_options(skb, &tp->rx_opt, 1, NULL); |
Andrew Vagin | e3e1202 | 2013-08-27 12:21:55 +0400 | [diff] [blame] | 3653 | if (tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr) |
Andrey Vagin | ee684b6 | 2013-02-11 05:50:19 +0000 | [diff] [blame] | 3654 | tp->rx_opt.rcv_tsecr -= tp->tsoffset; |
| 3655 | |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 3656 | return true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3657 | } |
| 3658 | |
YOSHIFUJI Hideaki | 7d5d552 | 2008-04-17 12:29:53 +0900 | [diff] [blame] | 3659 | #ifdef CONFIG_TCP_MD5SIG |
| 3660 | /* |
| 3661 | * Parse MD5 Signature option |
| 3662 | */ |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 3663 | const u8 *tcp_parse_md5sig_option(const struct tcphdr *th) |
YOSHIFUJI Hideaki | 7d5d552 | 2008-04-17 12:29:53 +0900 | [diff] [blame] | 3664 | { |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 3665 | int length = (th->doff << 2) - sizeof(*th); |
| 3666 | const u8 *ptr = (const u8 *)(th + 1); |
YOSHIFUJI Hideaki | 7d5d552 | 2008-04-17 12:29:53 +0900 | [diff] [blame] | 3667 | |
| 3668 | /* If the TCP option is too short, we can short cut */ |
| 3669 | if (length < TCPOLEN_MD5SIG) |
| 3670 | return NULL; |
| 3671 | |
| 3672 | while (length > 0) { |
| 3673 | int opcode = *ptr++; |
| 3674 | int opsize; |
| 3675 | |
| 3676 | switch(opcode) { |
| 3677 | case TCPOPT_EOL: |
| 3678 | return NULL; |
| 3679 | case TCPOPT_NOP: |
| 3680 | length--; |
| 3681 | continue; |
| 3682 | default: |
| 3683 | opsize = *ptr++; |
| 3684 | if (opsize < 2 || opsize > length) |
| 3685 | return NULL; |
| 3686 | if (opcode == TCPOPT_MD5SIG) |
Dmitry Popov | ba78e2d | 2010-08-07 20:24:28 -0700 | [diff] [blame] | 3687 | return opsize == TCPOLEN_MD5SIG ? ptr : NULL; |
YOSHIFUJI Hideaki | 7d5d552 | 2008-04-17 12:29:53 +0900 | [diff] [blame] | 3688 | } |
| 3689 | ptr += opsize - 2; |
| 3690 | length -= opsize; |
| 3691 | } |
| 3692 | return NULL; |
| 3693 | } |
Eric Dumazet | 4bc2f18 | 2010-07-09 21:22:10 +0000 | [diff] [blame] | 3694 | EXPORT_SYMBOL(tcp_parse_md5sig_option); |
YOSHIFUJI Hideaki | 7d5d552 | 2008-04-17 12:29:53 +0900 | [diff] [blame] | 3695 | #endif |
| 3696 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3697 | /* Sorry, PAWS as specified is broken wrt. pure-ACKs -DaveM |
| 3698 | * |
| 3699 | * It is not fatal. If this ACK does _not_ change critical state (seqs, window) |
| 3700 | * it can pass through stack. So, the following predicate verifies that |
| 3701 | * this segment is not used for anything but congestion avoidance or |
| 3702 | * fast retransmit. Moreover, we even are able to eliminate most of such |
| 3703 | * second order effects, if we apply some small "replay" window (~RTO) |
| 3704 | * to timestamp space. |
| 3705 | * |
| 3706 | * All these measures still do not guarantee that we reject wrapped ACKs |
| 3707 | * on networks with high bandwidth, when sequence space is recycled fastly, |
| 3708 | * but it guarantees that such events will be very rare and do not affect |
| 3709 | * connection seriously. This doesn't look nice, but alas, PAWS is really |
| 3710 | * buggy extension. |
| 3711 | * |
| 3712 | * [ Later note. Even worse! It is buggy for segments _with_ data. RFC |
| 3713 | * states that events when retransmit arrives after original data are rare. |
| 3714 | * It is a blatant lie. VJ forgot about fast retransmit! 8)8) It is |
| 3715 | * the biggest problem on large power networks even with minor reordering. |
| 3716 | * OK, let's give it small replay window. If peer clock is even 1hz, it is safe |
| 3717 | * up to bandwidth of 18Gigabit/sec. 8) ] |
| 3718 | */ |
| 3719 | |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 3720 | static int tcp_disordered_ack(const struct sock *sk, const struct sk_buff *skb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3721 | { |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 3722 | const struct tcp_sock *tp = tcp_sk(sk); |
| 3723 | const struct tcphdr *th = tcp_hdr(skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3724 | u32 seq = TCP_SKB_CB(skb)->seq; |
| 3725 | u32 ack = TCP_SKB_CB(skb)->ack_seq; |
| 3726 | |
| 3727 | return (/* 1. Pure ACK with correct sequence number. */ |
| 3728 | (th->ack && seq == TCP_SKB_CB(skb)->end_seq && seq == tp->rcv_nxt) && |
| 3729 | |
| 3730 | /* 2. ... and duplicate ACK. */ |
| 3731 | ack == tp->snd_una && |
| 3732 | |
| 3733 | /* 3. ... and does not update window. */ |
| 3734 | !tcp_may_update_window(tp, ack, seq, ntohs(th->window) << tp->rx_opt.snd_wscale) && |
| 3735 | |
| 3736 | /* 4. ... and sits in replay window. */ |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 3737 | (s32)(tp->rx_opt.ts_recent - tp->rx_opt.rcv_tsval) <= (inet_csk(sk)->icsk_rto * 1024) / HZ); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3738 | } |
| 3739 | |
Vijay Subramanian | 67b95bd | 2012-07-19 21:32:18 +0000 | [diff] [blame] | 3740 | static inline bool tcp_paws_discard(const struct sock *sk, |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 3741 | const struct sk_buff *skb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3742 | { |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 3743 | const struct tcp_sock *tp = tcp_sk(sk); |
Ilpo Järvinen | c887e6d | 2009-03-14 14:23:03 +0000 | [diff] [blame] | 3744 | |
| 3745 | return !tcp_paws_check(&tp->rx_opt, TCP_PAWS_WINDOW) && |
| 3746 | !tcp_disordered_ack(sk, skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3747 | } |
| 3748 | |
| 3749 | /* Check segment sequence number for validity. |
| 3750 | * |
| 3751 | * Segment controls are considered valid, if the segment |
| 3752 | * fits to the window after truncation to the window. Acceptability |
| 3753 | * of data (and SYN, FIN, of course) is checked separately. |
| 3754 | * See tcp_data_queue(), for example. |
| 3755 | * |
| 3756 | * Also, controls (RST is main one) are accepted using RCV.WUP instead |
| 3757 | * of RCV.NXT. Peer still did not advance his SND.UNA when we |
| 3758 | * delayed ACK, so that hisSND.UNA<=ourRCV.WUP. |
| 3759 | * (borrowed from freebsd) |
| 3760 | */ |
| 3761 | |
Vijay Subramanian | 67b95bd | 2012-07-19 21:32:18 +0000 | [diff] [blame] | 3762 | static inline bool tcp_sequence(const struct tcp_sock *tp, u32 seq, u32 end_seq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3763 | { |
| 3764 | return !before(end_seq, tp->rcv_wup) && |
| 3765 | !after(seq, tp->rcv_nxt + tcp_receive_window(tp)); |
| 3766 | } |
| 3767 | |
| 3768 | /* When we get a reset we do this. */ |
Jerry Chu | 1046716 | 2012-08-31 12:29:11 +0000 | [diff] [blame] | 3769 | void tcp_reset(struct sock *sk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3770 | { |
| 3771 | /* We want the right error as BSD sees it (and indeed as we do). */ |
| 3772 | switch (sk->sk_state) { |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 3773 | case TCP_SYN_SENT: |
| 3774 | sk->sk_err = ECONNREFUSED; |
| 3775 | break; |
| 3776 | case TCP_CLOSE_WAIT: |
| 3777 | sk->sk_err = EPIPE; |
| 3778 | break; |
| 3779 | case TCP_CLOSE: |
| 3780 | return; |
| 3781 | default: |
| 3782 | sk->sk_err = ECONNRESET; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3783 | } |
Tom Marshall | a4d2580 | 2010-09-20 15:42:05 -0700 | [diff] [blame] | 3784 | /* This barrier is coupled with smp_rmb() in tcp_poll() */ |
| 3785 | smp_wmb(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3786 | |
| 3787 | if (!sock_flag(sk, SOCK_DEAD)) |
| 3788 | sk->sk_error_report(sk); |
| 3789 | |
| 3790 | tcp_done(sk); |
| 3791 | } |
| 3792 | |
| 3793 | /* |
| 3794 | * Process the FIN bit. This now behaves as it is supposed to work |
| 3795 | * and the FIN takes effect when it is validly part of sequence |
| 3796 | * space. Not before when we get holes. |
| 3797 | * |
| 3798 | * If we are ESTABLISHED, a received fin moves us to CLOSE-WAIT |
| 3799 | * (and thence onto LAST-ACK and finally, CLOSE, we never enter |
| 3800 | * TIME-WAIT) |
| 3801 | * |
| 3802 | * If we are in FINWAIT-1, a received FIN indicates simultaneous |
| 3803 | * close and we go into CLOSING (and later onto TIME-WAIT) |
| 3804 | * |
| 3805 | * If we are in FINWAIT-2, a received FIN moves us to TIME-WAIT. |
| 3806 | */ |
Eric Dumazet | 20c4cb7 | 2011-10-20 17:44:03 -0400 | [diff] [blame] | 3807 | static void tcp_fin(struct sock *sk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3808 | { |
| 3809 | struct tcp_sock *tp = tcp_sk(sk); |
Cong Wang | bcefe17 | 2013-06-15 09:39:18 +0800 | [diff] [blame] | 3810 | const struct dst_entry *dst; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3811 | |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 3812 | inet_csk_schedule_ack(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3813 | |
| 3814 | sk->sk_shutdown |= RCV_SHUTDOWN; |
| 3815 | sock_set_flag(sk, SOCK_DONE); |
| 3816 | |
| 3817 | switch (sk->sk_state) { |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 3818 | case TCP_SYN_RECV: |
| 3819 | case TCP_ESTABLISHED: |
| 3820 | /* Move to CLOSE_WAIT */ |
| 3821 | tcp_set_state(sk, TCP_CLOSE_WAIT); |
Cong Wang | bcefe17 | 2013-06-15 09:39:18 +0800 | [diff] [blame] | 3822 | dst = __sk_dst_get(sk); |
| 3823 | if (!dst || !dst_metric(dst, RTAX_QUICKACK)) |
| 3824 | inet_csk(sk)->icsk_ack.pingpong = 1; |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 3825 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3826 | |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 3827 | case TCP_CLOSE_WAIT: |
| 3828 | case TCP_CLOSING: |
| 3829 | /* Received a retransmission of the FIN, do |
| 3830 | * nothing. |
| 3831 | */ |
| 3832 | break; |
| 3833 | case TCP_LAST_ACK: |
| 3834 | /* RFC793: Remain in the LAST-ACK state. */ |
| 3835 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3836 | |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 3837 | case TCP_FIN_WAIT1: |
| 3838 | /* This case occurs when a simultaneous close |
| 3839 | * happens, we must ack the received FIN and |
| 3840 | * enter the CLOSING state. |
| 3841 | */ |
| 3842 | tcp_send_ack(sk); |
| 3843 | tcp_set_state(sk, TCP_CLOSING); |
| 3844 | break; |
| 3845 | case TCP_FIN_WAIT2: |
| 3846 | /* Received a FIN -- send ACK and enter TIME_WAIT. */ |
| 3847 | tcp_send_ack(sk); |
| 3848 | tcp_time_wait(sk, TCP_TIME_WAIT, 0); |
| 3849 | break; |
| 3850 | default: |
| 3851 | /* Only TCP_LISTEN and TCP_CLOSE are left, in these |
| 3852 | * cases we should never reach this piece of code. |
| 3853 | */ |
Joe Perches | 058bd4d | 2012-03-11 18:36:11 +0000 | [diff] [blame] | 3854 | pr_err("%s: Impossible, sk->sk_state=%d\n", |
Harvey Harrison | 0dc4787 | 2008-03-05 20:47:47 -0800 | [diff] [blame] | 3855 | __func__, sk->sk_state); |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 3856 | break; |
Stephen Hemminger | 3ff50b7 | 2007-04-20 17:09:22 -0700 | [diff] [blame] | 3857 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3858 | |
| 3859 | /* It _is_ possible, that we have something out-of-order _after_ FIN. |
| 3860 | * Probably, we should reset in this case. For now drop them. |
| 3861 | */ |
| 3862 | __skb_queue_purge(&tp->out_of_order_queue); |
Ilpo Järvinen | e60402d | 2007-08-09 15:14:46 +0300 | [diff] [blame] | 3863 | if (tcp_is_sack(tp)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3864 | tcp_sack_reset(&tp->rx_opt); |
Hideo Aoki | 3ab224b | 2007-12-31 00:11:19 -0800 | [diff] [blame] | 3865 | sk_mem_reclaim(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3866 | |
| 3867 | if (!sock_flag(sk, SOCK_DEAD)) { |
| 3868 | sk->sk_state_change(sk); |
| 3869 | |
| 3870 | /* Do not send POLL_HUP for half duplex close. */ |
| 3871 | if (sk->sk_shutdown == SHUTDOWN_MASK || |
| 3872 | sk->sk_state == TCP_CLOSE) |
Pavel Emelyanov | 8d8ad9d | 2007-11-26 20:10:50 +0800 | [diff] [blame] | 3873 | sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_HUP); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3874 | else |
Pavel Emelyanov | 8d8ad9d | 2007-11-26 20:10:50 +0800 | [diff] [blame] | 3875 | sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3876 | } |
| 3877 | } |
| 3878 | |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 3879 | static inline bool tcp_sack_extend(struct tcp_sack_block *sp, u32 seq, |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 3880 | u32 end_seq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3881 | { |
| 3882 | if (!after(seq, sp->end_seq) && !after(sp->start_seq, end_seq)) { |
| 3883 | if (before(seq, sp->start_seq)) |
| 3884 | sp->start_seq = seq; |
| 3885 | if (after(end_seq, sp->end_seq)) |
| 3886 | sp->end_seq = end_seq; |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 3887 | return true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3888 | } |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 3889 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3890 | } |
| 3891 | |
Pavel Emelyanov | 1ed8346 | 2008-07-16 20:29:51 -0700 | [diff] [blame] | 3892 | static void tcp_dsack_set(struct sock *sk, u32 seq, u32 end_seq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3893 | { |
Pavel Emelyanov | 1ed8346 | 2008-07-16 20:29:51 -0700 | [diff] [blame] | 3894 | struct tcp_sock *tp = tcp_sk(sk); |
| 3895 | |
David S. Miller | bb5b7c1 | 2009-12-15 20:56:42 -0800 | [diff] [blame] | 3896 | if (tcp_is_sack(tp) && sysctl_tcp_dsack) { |
Pavel Emelyanov | 40b215e | 2008-07-03 01:05:41 -0700 | [diff] [blame] | 3897 | int mib_idx; |
| 3898 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3899 | if (before(seq, tp->rcv_nxt)) |
Pavel Emelyanov | 40b215e | 2008-07-03 01:05:41 -0700 | [diff] [blame] | 3900 | mib_idx = LINUX_MIB_TCPDSACKOLDSENT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3901 | else |
Pavel Emelyanov | 40b215e | 2008-07-03 01:05:41 -0700 | [diff] [blame] | 3902 | mib_idx = LINUX_MIB_TCPDSACKOFOSENT; |
| 3903 | |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 3904 | NET_INC_STATS_BH(sock_net(sk), mib_idx); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3905 | |
| 3906 | tp->rx_opt.dsack = 1; |
| 3907 | tp->duplicate_sack[0].start_seq = seq; |
| 3908 | tp->duplicate_sack[0].end_seq = end_seq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3909 | } |
| 3910 | } |
| 3911 | |
Pavel Emelyanov | 1ed8346 | 2008-07-16 20:29:51 -0700 | [diff] [blame] | 3912 | static void tcp_dsack_extend(struct sock *sk, u32 seq, u32 end_seq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3913 | { |
Pavel Emelyanov | 1ed8346 | 2008-07-16 20:29:51 -0700 | [diff] [blame] | 3914 | struct tcp_sock *tp = tcp_sk(sk); |
| 3915 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3916 | if (!tp->rx_opt.dsack) |
Pavel Emelyanov | 1ed8346 | 2008-07-16 20:29:51 -0700 | [diff] [blame] | 3917 | tcp_dsack_set(sk, seq, end_seq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3918 | else |
| 3919 | tcp_sack_extend(tp->duplicate_sack, seq, end_seq); |
| 3920 | } |
| 3921 | |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 3922 | static void tcp_send_dupack(struct sock *sk, const struct sk_buff *skb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3923 | { |
| 3924 | struct tcp_sock *tp = tcp_sk(sk); |
| 3925 | |
| 3926 | if (TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(skb)->seq && |
| 3927 | before(TCP_SKB_CB(skb)->seq, tp->rcv_nxt)) { |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 3928 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_DELAYEDACKLOST); |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 3929 | tcp_enter_quickack_mode(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3930 | |
David S. Miller | bb5b7c1 | 2009-12-15 20:56:42 -0800 | [diff] [blame] | 3931 | if (tcp_is_sack(tp) && sysctl_tcp_dsack) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3932 | u32 end_seq = TCP_SKB_CB(skb)->end_seq; |
| 3933 | |
| 3934 | if (after(TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt)) |
| 3935 | end_seq = tp->rcv_nxt; |
Pavel Emelyanov | 1ed8346 | 2008-07-16 20:29:51 -0700 | [diff] [blame] | 3936 | tcp_dsack_set(sk, TCP_SKB_CB(skb)->seq, end_seq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3937 | } |
| 3938 | } |
| 3939 | |
| 3940 | tcp_send_ack(sk); |
| 3941 | } |
| 3942 | |
| 3943 | /* These routines update the SACK block as out-of-order packets arrive or |
| 3944 | * in-order packets close up the sequence space. |
| 3945 | */ |
| 3946 | static void tcp_sack_maybe_coalesce(struct tcp_sock *tp) |
| 3947 | { |
| 3948 | int this_sack; |
| 3949 | struct tcp_sack_block *sp = &tp->selective_acks[0]; |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 3950 | struct tcp_sack_block *swalk = sp + 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3951 | |
| 3952 | /* See if the recent change to the first SACK eats into |
| 3953 | * or hits the sequence space of other SACK blocks, if so coalesce. |
| 3954 | */ |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 3955 | for (this_sack = 1; this_sack < tp->rx_opt.num_sacks;) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3956 | if (tcp_sack_extend(sp, swalk->start_seq, swalk->end_seq)) { |
| 3957 | int i; |
| 3958 | |
| 3959 | /* Zap SWALK, by moving every further SACK up by one slot. |
| 3960 | * Decrease num_sacks. |
| 3961 | */ |
| 3962 | tp->rx_opt.num_sacks--; |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 3963 | for (i = this_sack; i < tp->rx_opt.num_sacks; i++) |
| 3964 | sp[i] = sp[i + 1]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3965 | continue; |
| 3966 | } |
| 3967 | this_sack++, swalk++; |
| 3968 | } |
| 3969 | } |
| 3970 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3971 | static void tcp_sack_new_ofo_skb(struct sock *sk, u32 seq, u32 end_seq) |
| 3972 | { |
| 3973 | struct tcp_sock *tp = tcp_sk(sk); |
| 3974 | struct tcp_sack_block *sp = &tp->selective_acks[0]; |
| 3975 | int cur_sacks = tp->rx_opt.num_sacks; |
| 3976 | int this_sack; |
| 3977 | |
| 3978 | if (!cur_sacks) |
| 3979 | goto new_sack; |
| 3980 | |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 3981 | for (this_sack = 0; this_sack < cur_sacks; this_sack++, sp++) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3982 | if (tcp_sack_extend(sp, seq, end_seq)) { |
| 3983 | /* Rotate this_sack to the first one. */ |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 3984 | for (; this_sack > 0; this_sack--, sp--) |
Ilpo Järvinen | a0bffff | 2009-03-21 13:36:17 -0700 | [diff] [blame] | 3985 | swap(*sp, *(sp - 1)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3986 | if (cur_sacks > 1) |
| 3987 | tcp_sack_maybe_coalesce(tp); |
| 3988 | return; |
| 3989 | } |
| 3990 | } |
| 3991 | |
| 3992 | /* Could not find an adjacent existing SACK, build a new one, |
| 3993 | * put it at the front, and shift everyone else down. We |
| 3994 | * always know there is at least one SACK present already here. |
| 3995 | * |
| 3996 | * If the sack array is full, forget about the last one. |
| 3997 | */ |
Adam Langley | 4389dde | 2008-07-19 00:07:02 -0700 | [diff] [blame] | 3998 | if (this_sack >= TCP_NUM_SACKS) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3999 | this_sack--; |
| 4000 | tp->rx_opt.num_sacks--; |
| 4001 | sp--; |
| 4002 | } |
Stephen Hemminger | 2de979b | 2007-03-08 20:45:19 -0800 | [diff] [blame] | 4003 | for (; this_sack > 0; this_sack--, sp--) |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 4004 | *sp = *(sp - 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4005 | |
| 4006 | new_sack: |
| 4007 | /* Build the new head SACK, and we're done. */ |
| 4008 | sp->start_seq = seq; |
| 4009 | sp->end_seq = end_seq; |
| 4010 | tp->rx_opt.num_sacks++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4011 | } |
| 4012 | |
| 4013 | /* RCV.NXT advances, some SACKs should be eaten. */ |
| 4014 | |
| 4015 | static void tcp_sack_remove(struct tcp_sock *tp) |
| 4016 | { |
| 4017 | struct tcp_sack_block *sp = &tp->selective_acks[0]; |
| 4018 | int num_sacks = tp->rx_opt.num_sacks; |
| 4019 | int this_sack; |
| 4020 | |
| 4021 | /* Empty ofo queue, hence, all the SACKs are eaten. Clear. */ |
David S. Miller | b03efcf | 2005-07-08 14:57:23 -0700 | [diff] [blame] | 4022 | if (skb_queue_empty(&tp->out_of_order_queue)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4023 | tp->rx_opt.num_sacks = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4024 | return; |
| 4025 | } |
| 4026 | |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 4027 | for (this_sack = 0; this_sack < num_sacks;) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4028 | /* Check if the start of the sack is covered by RCV.NXT. */ |
| 4029 | if (!before(tp->rcv_nxt, sp->start_seq)) { |
| 4030 | int i; |
| 4031 | |
| 4032 | /* RCV.NXT must cover all the block! */ |
Ilpo Järvinen | 547b792 | 2008-07-25 21:43:18 -0700 | [diff] [blame] | 4033 | WARN_ON(before(tp->rcv_nxt, sp->end_seq)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4034 | |
| 4035 | /* Zap this SACK, by moving forward any other SACKS. */ |
| 4036 | for (i=this_sack+1; i < num_sacks; i++) |
| 4037 | tp->selective_acks[i-1] = tp->selective_acks[i]; |
| 4038 | num_sacks--; |
| 4039 | continue; |
| 4040 | } |
| 4041 | this_sack++; |
| 4042 | sp++; |
| 4043 | } |
Ilpo Järvinen | 5861f8e | 2009-03-14 14:23:01 +0000 | [diff] [blame] | 4044 | tp->rx_opt.num_sacks = num_sacks; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4045 | } |
| 4046 | |
| 4047 | /* This one checks to see if we can put data from the |
| 4048 | * out_of_order queue into the receive_queue. |
| 4049 | */ |
| 4050 | static void tcp_ofo_queue(struct sock *sk) |
| 4051 | { |
| 4052 | struct tcp_sock *tp = tcp_sk(sk); |
| 4053 | __u32 dsack_high = tp->rcv_nxt; |
| 4054 | struct sk_buff *skb; |
| 4055 | |
| 4056 | while ((skb = skb_peek(&tp->out_of_order_queue)) != NULL) { |
| 4057 | if (after(TCP_SKB_CB(skb)->seq, tp->rcv_nxt)) |
| 4058 | break; |
| 4059 | |
| 4060 | if (before(TCP_SKB_CB(skb)->seq, dsack_high)) { |
| 4061 | __u32 dsack = dsack_high; |
| 4062 | if (before(TCP_SKB_CB(skb)->end_seq, dsack_high)) |
| 4063 | dsack_high = TCP_SKB_CB(skb)->end_seq; |
Pavel Emelyanov | 1ed8346 | 2008-07-16 20:29:51 -0700 | [diff] [blame] | 4064 | tcp_dsack_extend(sk, TCP_SKB_CB(skb)->seq, dsack); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4065 | } |
| 4066 | |
| 4067 | if (!after(TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt)) { |
Frans Pop | b138338 | 2010-03-24 07:57:28 +0000 | [diff] [blame] | 4068 | SOCK_DEBUG(sk, "ofo packet was already received\n"); |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 4069 | __skb_unlink(skb, &tp->out_of_order_queue); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4070 | __kfree_skb(skb); |
| 4071 | continue; |
| 4072 | } |
| 4073 | SOCK_DEBUG(sk, "ofo requeuing : rcv_next %X seq %X - %X\n", |
| 4074 | tp->rcv_nxt, TCP_SKB_CB(skb)->seq, |
| 4075 | TCP_SKB_CB(skb)->end_seq); |
| 4076 | |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 4077 | __skb_unlink(skb, &tp->out_of_order_queue); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4078 | __skb_queue_tail(&sk->sk_receive_queue, skb); |
| 4079 | tp->rcv_nxt = TCP_SKB_CB(skb)->end_seq; |
Arnaldo Carvalho de Melo | aa8223c | 2007-04-10 21:04:22 -0700 | [diff] [blame] | 4080 | if (tcp_hdr(skb)->fin) |
Eric Dumazet | 20c4cb7 | 2011-10-20 17:44:03 -0400 | [diff] [blame] | 4081 | tcp_fin(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4082 | } |
| 4083 | } |
| 4084 | |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 4085 | static bool tcp_prune_ofo_queue(struct sock *sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4086 | static int tcp_prune_queue(struct sock *sk); |
| 4087 | |
Mel Gorman | c76562b | 2012-07-31 16:44:41 -0700 | [diff] [blame] | 4088 | static int tcp_try_rmem_schedule(struct sock *sk, struct sk_buff *skb, |
| 4089 | unsigned int size) |
Vitaliy Gusev | b000cd3 | 2008-04-15 00:33:38 -0700 | [diff] [blame] | 4090 | { |
| 4091 | if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf || |
Mel Gorman | c76562b | 2012-07-31 16:44:41 -0700 | [diff] [blame] | 4092 | !sk_rmem_schedule(sk, skb, size)) { |
Vitaliy Gusev | b000cd3 | 2008-04-15 00:33:38 -0700 | [diff] [blame] | 4093 | |
| 4094 | if (tcp_prune_queue(sk) < 0) |
| 4095 | return -1; |
| 4096 | |
Mel Gorman | c76562b | 2012-07-31 16:44:41 -0700 | [diff] [blame] | 4097 | if (!sk_rmem_schedule(sk, skb, size)) { |
Vitaliy Gusev | 56f367b | 2008-04-15 20:26:34 -0700 | [diff] [blame] | 4098 | if (!tcp_prune_ofo_queue(sk)) |
| 4099 | return -1; |
| 4100 | |
Mel Gorman | c76562b | 2012-07-31 16:44:41 -0700 | [diff] [blame] | 4101 | if (!sk_rmem_schedule(sk, skb, size)) |
Vitaliy Gusev | b000cd3 | 2008-04-15 00:33:38 -0700 | [diff] [blame] | 4102 | return -1; |
| 4103 | } |
| 4104 | } |
| 4105 | return 0; |
| 4106 | } |
| 4107 | |
Eric Dumazet | 1402d36 | 2012-04-23 07:11:42 +0000 | [diff] [blame] | 4108 | /** |
| 4109 | * tcp_try_coalesce - try to merge skb to prior one |
| 4110 | * @sk: socket |
| 4111 | * @to: prior buffer |
| 4112 | * @from: buffer to add in queue |
Eric Dumazet | 923dd34 | 2012-05-02 07:55:58 +0000 | [diff] [blame] | 4113 | * @fragstolen: pointer to boolean |
Eric Dumazet | 1402d36 | 2012-04-23 07:11:42 +0000 | [diff] [blame] | 4114 | * |
| 4115 | * Before queueing skb @from after @to, try to merge them |
| 4116 | * to reduce overall memory use and queue lengths, if cost is small. |
| 4117 | * Packets in ofo or receive queues can stay a long time. |
| 4118 | * Better try to coalesce them right now to avoid future collapses. |
Eric Dumazet | 783c175 | 2012-04-23 17:34:36 +0000 | [diff] [blame] | 4119 | * Returns true if caller should free @from instead of queueing it |
Eric Dumazet | 1402d36 | 2012-04-23 07:11:42 +0000 | [diff] [blame] | 4120 | */ |
Eric Dumazet | 783c175 | 2012-04-23 17:34:36 +0000 | [diff] [blame] | 4121 | static bool tcp_try_coalesce(struct sock *sk, |
| 4122 | struct sk_buff *to, |
Eric Dumazet | 329033f | 2012-04-27 00:38:33 +0000 | [diff] [blame] | 4123 | struct sk_buff *from, |
| 4124 | bool *fragstolen) |
Eric Dumazet | 1402d36 | 2012-04-23 07:11:42 +0000 | [diff] [blame] | 4125 | { |
Eric Dumazet | bad43ca | 2012-05-19 03:02:02 +0000 | [diff] [blame] | 4126 | int delta; |
Eric Dumazet | 1402d36 | 2012-04-23 07:11:42 +0000 | [diff] [blame] | 4127 | |
Eric Dumazet | 329033f | 2012-04-27 00:38:33 +0000 | [diff] [blame] | 4128 | *fragstolen = false; |
Alexander Duyck | 34a802a | 2012-05-02 21:19:09 +0000 | [diff] [blame] | 4129 | |
Eric Dumazet | bad43ca | 2012-05-19 03:02:02 +0000 | [diff] [blame] | 4130 | if (tcp_hdr(from)->fin) |
| 4131 | return false; |
Eric Dumazet | 1ca7ee3 | 2012-05-23 17:51:37 +0000 | [diff] [blame] | 4132 | |
| 4133 | /* Its possible this segment overlaps with prior segment in queue */ |
| 4134 | if (TCP_SKB_CB(from)->seq != TCP_SKB_CB(to)->end_seq) |
| 4135 | return false; |
| 4136 | |
Eric Dumazet | bad43ca | 2012-05-19 03:02:02 +0000 | [diff] [blame] | 4137 | if (!skb_try_coalesce(to, from, fragstolen, &delta)) |
Eric Dumazet | 783c175 | 2012-04-23 17:34:36 +0000 | [diff] [blame] | 4138 | return false; |
Alexander Duyck | 34a802a | 2012-05-02 21:19:09 +0000 | [diff] [blame] | 4139 | |
Alexander Duyck | 34a802a | 2012-05-02 21:19:09 +0000 | [diff] [blame] | 4140 | atomic_add(delta, &sk->sk_rmem_alloc); |
| 4141 | sk_mem_charge(sk, delta); |
Alexander Duyck | 34a802a | 2012-05-02 21:19:09 +0000 | [diff] [blame] | 4142 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPRCVCOALESCE); |
| 4143 | TCP_SKB_CB(to)->end_seq = TCP_SKB_CB(from)->end_seq; |
| 4144 | TCP_SKB_CB(to)->ack_seq = TCP_SKB_CB(from)->ack_seq; |
| 4145 | return true; |
Eric Dumazet | 1402d36 | 2012-04-23 07:11:42 +0000 | [diff] [blame] | 4146 | } |
| 4147 | |
Eric Dumazet | e86b2919 | 2012-03-18 11:06:44 +0000 | [diff] [blame] | 4148 | static void tcp_data_queue_ofo(struct sock *sk, struct sk_buff *skb) |
| 4149 | { |
| 4150 | struct tcp_sock *tp = tcp_sk(sk); |
| 4151 | struct sk_buff *skb1; |
| 4152 | u32 seq, end_seq; |
| 4153 | |
| 4154 | TCP_ECN_check_ce(tp, skb); |
| 4155 | |
Mel Gorman | c76562b | 2012-07-31 16:44:41 -0700 | [diff] [blame] | 4156 | if (unlikely(tcp_try_rmem_schedule(sk, skb, skb->truesize))) { |
Eric Dumazet | a6df1ae | 2012-07-16 01:41:36 +0000 | [diff] [blame] | 4157 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPOFODROP); |
Eric Dumazet | e86b2919 | 2012-03-18 11:06:44 +0000 | [diff] [blame] | 4158 | __kfree_skb(skb); |
| 4159 | return; |
| 4160 | } |
| 4161 | |
| 4162 | /* Disable header prediction. */ |
| 4163 | tp->pred_flags = 0; |
| 4164 | inet_csk_schedule_ack(sk); |
| 4165 | |
Eric Dumazet | a6df1ae | 2012-07-16 01:41:36 +0000 | [diff] [blame] | 4166 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPOFOQUEUE); |
Eric Dumazet | e86b2919 | 2012-03-18 11:06:44 +0000 | [diff] [blame] | 4167 | SOCK_DEBUG(sk, "out of order segment: rcv_next %X seq %X - %X\n", |
| 4168 | tp->rcv_nxt, TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq); |
| 4169 | |
| 4170 | skb1 = skb_peek_tail(&tp->out_of_order_queue); |
| 4171 | if (!skb1) { |
| 4172 | /* Initial out of order segment, build 1 SACK. */ |
| 4173 | if (tcp_is_sack(tp)) { |
| 4174 | tp->rx_opt.num_sacks = 1; |
| 4175 | tp->selective_acks[0].start_seq = TCP_SKB_CB(skb)->seq; |
| 4176 | tp->selective_acks[0].end_seq = |
| 4177 | TCP_SKB_CB(skb)->end_seq; |
| 4178 | } |
| 4179 | __skb_queue_head(&tp->out_of_order_queue, skb); |
| 4180 | goto end; |
| 4181 | } |
| 4182 | |
| 4183 | seq = TCP_SKB_CB(skb)->seq; |
| 4184 | end_seq = TCP_SKB_CB(skb)->end_seq; |
| 4185 | |
| 4186 | if (seq == TCP_SKB_CB(skb1)->end_seq) { |
Eric Dumazet | 329033f | 2012-04-27 00:38:33 +0000 | [diff] [blame] | 4187 | bool fragstolen; |
| 4188 | |
| 4189 | if (!tcp_try_coalesce(sk, skb1, skb, &fragstolen)) { |
Eric Dumazet | 1402d36 | 2012-04-23 07:11:42 +0000 | [diff] [blame] | 4190 | __skb_queue_after(&tp->out_of_order_queue, skb1, skb); |
| 4191 | } else { |
Eric Dumazet | 4e4f1fc | 2013-09-06 10:35:58 -0700 | [diff] [blame] | 4192 | tcp_grow_window(sk, skb); |
Eric Dumazet | 923dd34 | 2012-05-02 07:55:58 +0000 | [diff] [blame] | 4193 | kfree_skb_partial(skb, fragstolen); |
Eric Dumazet | c862815 | 2012-03-18 11:07:47 +0000 | [diff] [blame] | 4194 | skb = NULL; |
Eric Dumazet | c862815 | 2012-03-18 11:07:47 +0000 | [diff] [blame] | 4195 | } |
Eric Dumazet | e86b2919 | 2012-03-18 11:06:44 +0000 | [diff] [blame] | 4196 | |
| 4197 | if (!tp->rx_opt.num_sacks || |
| 4198 | tp->selective_acks[0].end_seq != seq) |
| 4199 | goto add_sack; |
| 4200 | |
| 4201 | /* Common case: data arrive in order after hole. */ |
| 4202 | tp->selective_acks[0].end_seq = end_seq; |
| 4203 | goto end; |
| 4204 | } |
| 4205 | |
| 4206 | /* Find place to insert this segment. */ |
| 4207 | while (1) { |
| 4208 | if (!after(TCP_SKB_CB(skb1)->seq, seq)) |
| 4209 | break; |
| 4210 | if (skb_queue_is_first(&tp->out_of_order_queue, skb1)) { |
| 4211 | skb1 = NULL; |
| 4212 | break; |
| 4213 | } |
| 4214 | skb1 = skb_queue_prev(&tp->out_of_order_queue, skb1); |
| 4215 | } |
| 4216 | |
| 4217 | /* Do skb overlap to previous one? */ |
| 4218 | if (skb1 && before(seq, TCP_SKB_CB(skb1)->end_seq)) { |
| 4219 | if (!after(end_seq, TCP_SKB_CB(skb1)->end_seq)) { |
| 4220 | /* All the bits are present. Drop. */ |
Eric Dumazet | a6df1ae | 2012-07-16 01:41:36 +0000 | [diff] [blame] | 4221 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPOFOMERGE); |
Eric Dumazet | e86b2919 | 2012-03-18 11:06:44 +0000 | [diff] [blame] | 4222 | __kfree_skb(skb); |
| 4223 | skb = NULL; |
| 4224 | tcp_dsack_set(sk, seq, end_seq); |
| 4225 | goto add_sack; |
| 4226 | } |
| 4227 | if (after(seq, TCP_SKB_CB(skb1)->seq)) { |
| 4228 | /* Partial overlap. */ |
| 4229 | tcp_dsack_set(sk, seq, |
| 4230 | TCP_SKB_CB(skb1)->end_seq); |
| 4231 | } else { |
| 4232 | if (skb_queue_is_first(&tp->out_of_order_queue, |
| 4233 | skb1)) |
| 4234 | skb1 = NULL; |
| 4235 | else |
| 4236 | skb1 = skb_queue_prev( |
| 4237 | &tp->out_of_order_queue, |
| 4238 | skb1); |
| 4239 | } |
| 4240 | } |
| 4241 | if (!skb1) |
| 4242 | __skb_queue_head(&tp->out_of_order_queue, skb); |
| 4243 | else |
| 4244 | __skb_queue_after(&tp->out_of_order_queue, skb1, skb); |
| 4245 | |
| 4246 | /* And clean segments covered by new one as whole. */ |
| 4247 | while (!skb_queue_is_last(&tp->out_of_order_queue, skb)) { |
| 4248 | skb1 = skb_queue_next(&tp->out_of_order_queue, skb); |
| 4249 | |
| 4250 | if (!after(end_seq, TCP_SKB_CB(skb1)->seq)) |
| 4251 | break; |
| 4252 | if (before(end_seq, TCP_SKB_CB(skb1)->end_seq)) { |
| 4253 | tcp_dsack_extend(sk, TCP_SKB_CB(skb1)->seq, |
| 4254 | end_seq); |
| 4255 | break; |
| 4256 | } |
| 4257 | __skb_unlink(skb1, &tp->out_of_order_queue); |
| 4258 | tcp_dsack_extend(sk, TCP_SKB_CB(skb1)->seq, |
| 4259 | TCP_SKB_CB(skb1)->end_seq); |
Eric Dumazet | a6df1ae | 2012-07-16 01:41:36 +0000 | [diff] [blame] | 4260 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPOFOMERGE); |
Eric Dumazet | e86b2919 | 2012-03-18 11:06:44 +0000 | [diff] [blame] | 4261 | __kfree_skb(skb1); |
| 4262 | } |
| 4263 | |
| 4264 | add_sack: |
| 4265 | if (tcp_is_sack(tp)) |
| 4266 | tcp_sack_new_ofo_skb(sk, seq, end_seq); |
| 4267 | end: |
Eric Dumazet | 4e4f1fc | 2013-09-06 10:35:58 -0700 | [diff] [blame] | 4268 | if (skb) { |
| 4269 | tcp_grow_window(sk, skb); |
Eric Dumazet | e86b2919 | 2012-03-18 11:06:44 +0000 | [diff] [blame] | 4270 | skb_set_owner_r(skb, sk); |
Eric Dumazet | 4e4f1fc | 2013-09-06 10:35:58 -0700 | [diff] [blame] | 4271 | } |
Eric Dumazet | e86b2919 | 2012-03-18 11:06:44 +0000 | [diff] [blame] | 4272 | } |
| 4273 | |
Pavel Emelyanov | 292e8d8 | 2012-05-10 01:49:41 +0000 | [diff] [blame] | 4274 | static int __must_check tcp_queue_rcv(struct sock *sk, struct sk_buff *skb, int hdrlen, |
Eric Dumazet | b081f85 | 2012-05-02 09:58:29 +0000 | [diff] [blame] | 4275 | bool *fragstolen) |
| 4276 | { |
| 4277 | int eaten; |
| 4278 | struct sk_buff *tail = skb_peek_tail(&sk->sk_receive_queue); |
| 4279 | |
| 4280 | __skb_pull(skb, hdrlen); |
| 4281 | eaten = (tail && |
| 4282 | tcp_try_coalesce(sk, tail, skb, fragstolen)) ? 1 : 0; |
| 4283 | tcp_sk(sk)->rcv_nxt = TCP_SKB_CB(skb)->end_seq; |
| 4284 | if (!eaten) { |
| 4285 | __skb_queue_tail(&sk->sk_receive_queue, skb); |
| 4286 | skb_set_owner_r(skb, sk); |
| 4287 | } |
| 4288 | return eaten; |
| 4289 | } |
Eric Dumazet | e86b2919 | 2012-03-18 11:06:44 +0000 | [diff] [blame] | 4290 | |
Pavel Emelyanov | 292e8d8 | 2012-05-10 01:49:41 +0000 | [diff] [blame] | 4291 | int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size) |
| 4292 | { |
Mel Gorman | c76562b | 2012-07-31 16:44:41 -0700 | [diff] [blame] | 4293 | struct sk_buff *skb = NULL; |
Pavel Emelyanov | 292e8d8 | 2012-05-10 01:49:41 +0000 | [diff] [blame] | 4294 | struct tcphdr *th; |
| 4295 | bool fragstolen; |
| 4296 | |
Pavel Emelyanov | c454e61 | 2012-10-29 05:05:33 +0000 | [diff] [blame] | 4297 | if (size == 0) |
| 4298 | return 0; |
| 4299 | |
Pavel Emelyanov | 292e8d8 | 2012-05-10 01:49:41 +0000 | [diff] [blame] | 4300 | skb = alloc_skb(size + sizeof(*th), sk->sk_allocation); |
| 4301 | if (!skb) |
| 4302 | goto err; |
| 4303 | |
Mel Gorman | c76562b | 2012-07-31 16:44:41 -0700 | [diff] [blame] | 4304 | if (tcp_try_rmem_schedule(sk, skb, size + sizeof(*th))) |
| 4305 | goto err_free; |
| 4306 | |
Pavel Emelyanov | 292e8d8 | 2012-05-10 01:49:41 +0000 | [diff] [blame] | 4307 | th = (struct tcphdr *)skb_put(skb, sizeof(*th)); |
| 4308 | skb_reset_transport_header(skb); |
| 4309 | memset(th, 0, sizeof(*th)); |
| 4310 | |
| 4311 | if (memcpy_fromiovec(skb_put(skb, size), msg->msg_iov, size)) |
| 4312 | goto err_free; |
| 4313 | |
| 4314 | TCP_SKB_CB(skb)->seq = tcp_sk(sk)->rcv_nxt; |
| 4315 | TCP_SKB_CB(skb)->end_seq = TCP_SKB_CB(skb)->seq + size; |
| 4316 | TCP_SKB_CB(skb)->ack_seq = tcp_sk(sk)->snd_una - 1; |
| 4317 | |
| 4318 | if (tcp_queue_rcv(sk, skb, sizeof(*th), &fragstolen)) { |
| 4319 | WARN_ON_ONCE(fragstolen); /* should not happen */ |
| 4320 | __kfree_skb(skb); |
| 4321 | } |
| 4322 | return size; |
| 4323 | |
| 4324 | err_free: |
| 4325 | kfree_skb(skb); |
| 4326 | err: |
| 4327 | return -ENOMEM; |
| 4328 | } |
| 4329 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4330 | static void tcp_data_queue(struct sock *sk, struct sk_buff *skb) |
| 4331 | { |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 4332 | const struct tcphdr *th = tcp_hdr(skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4333 | struct tcp_sock *tp = tcp_sk(sk); |
| 4334 | int eaten = -1; |
Eric Dumazet | 329033f | 2012-04-27 00:38:33 +0000 | [diff] [blame] | 4335 | bool fragstolen = false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4336 | |
| 4337 | if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) |
| 4338 | goto drop; |
| 4339 | |
Eric Dumazet | f84af32 | 2010-04-28 15:31:51 -0700 | [diff] [blame] | 4340 | skb_dst_drop(skb); |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 4341 | __skb_pull(skb, th->doff * 4); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4342 | |
| 4343 | TCP_ECN_accept_cwr(tp, skb); |
| 4344 | |
Ilpo Järvinen | 5861f8e | 2009-03-14 14:23:01 +0000 | [diff] [blame] | 4345 | tp->rx_opt.dsack = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4346 | |
| 4347 | /* Queue data for delivery to the user. |
| 4348 | * Packets in sequence go to the receive queue. |
| 4349 | * Out of sequence packets to the out_of_order_queue. |
| 4350 | */ |
| 4351 | if (TCP_SKB_CB(skb)->seq == tp->rcv_nxt) { |
| 4352 | if (tcp_receive_window(tp) == 0) |
| 4353 | goto out_of_window; |
| 4354 | |
| 4355 | /* Ok. In sequence. In window. */ |
| 4356 | if (tp->ucopy.task == current && |
| 4357 | tp->copied_seq == tp->rcv_nxt && tp->ucopy.len && |
| 4358 | sock_owned_by_user(sk) && !tp->urg_data) { |
| 4359 | int chunk = min_t(unsigned int, skb->len, |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 4360 | tp->ucopy.len); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4361 | |
| 4362 | __set_current_state(TASK_RUNNING); |
| 4363 | |
| 4364 | local_bh_enable(); |
| 4365 | if (!skb_copy_datagram_iovec(skb, 0, tp->ucopy.iov, chunk)) { |
| 4366 | tp->ucopy.len -= chunk; |
| 4367 | tp->copied_seq += chunk; |
Jerry Chu | 44f5324 | 2011-01-25 13:46:30 -0800 | [diff] [blame] | 4368 | eaten = (chunk == skb->len); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4369 | tcp_rcv_space_adjust(sk); |
| 4370 | } |
| 4371 | local_bh_disable(); |
| 4372 | } |
| 4373 | |
| 4374 | if (eaten <= 0) { |
| 4375 | queue_and_out: |
| 4376 | if (eaten < 0 && |
Mel Gorman | c76562b | 2012-07-31 16:44:41 -0700 | [diff] [blame] | 4377 | tcp_try_rmem_schedule(sk, skb, skb->truesize)) |
Vitaliy Gusev | b000cd3 | 2008-04-15 00:33:38 -0700 | [diff] [blame] | 4378 | goto drop; |
| 4379 | |
Eric Dumazet | b081f85 | 2012-05-02 09:58:29 +0000 | [diff] [blame] | 4380 | eaten = tcp_queue_rcv(sk, skb, 0, &fragstolen); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4381 | } |
| 4382 | tp->rcv_nxt = TCP_SKB_CB(skb)->end_seq; |
Stephen Hemminger | 2de979b | 2007-03-08 20:45:19 -0800 | [diff] [blame] | 4383 | if (skb->len) |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 4384 | tcp_event_data_recv(sk, skb); |
Stephen Hemminger | 2de979b | 2007-03-08 20:45:19 -0800 | [diff] [blame] | 4385 | if (th->fin) |
Eric Dumazet | 20c4cb7 | 2011-10-20 17:44:03 -0400 | [diff] [blame] | 4386 | tcp_fin(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4387 | |
David S. Miller | b03efcf | 2005-07-08 14:57:23 -0700 | [diff] [blame] | 4388 | if (!skb_queue_empty(&tp->out_of_order_queue)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4389 | tcp_ofo_queue(sk); |
| 4390 | |
| 4391 | /* RFC2581. 4.2. SHOULD send immediate ACK, when |
| 4392 | * gap in queue is filled. |
| 4393 | */ |
David S. Miller | b03efcf | 2005-07-08 14:57:23 -0700 | [diff] [blame] | 4394 | if (skb_queue_empty(&tp->out_of_order_queue)) |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 4395 | inet_csk(sk)->icsk_ack.pingpong = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4396 | } |
| 4397 | |
| 4398 | if (tp->rx_opt.num_sacks) |
| 4399 | tcp_sack_remove(tp); |
| 4400 | |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 4401 | tcp_fast_path_check(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4402 | |
Eric Dumazet | 923dd34 | 2012-05-02 07:55:58 +0000 | [diff] [blame] | 4403 | if (eaten > 0) |
| 4404 | kfree_skb_partial(skb, fragstolen); |
Eric Dumazet | 1d57f19 | 2012-09-17 12:51:39 +0000 | [diff] [blame] | 4405 | if (!sock_flag(sk, SOCK_DEAD)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4406 | sk->sk_data_ready(sk, 0); |
| 4407 | return; |
| 4408 | } |
| 4409 | |
| 4410 | if (!after(TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt)) { |
| 4411 | /* A retransmit, 2nd most common case. Force an immediate ack. */ |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 4412 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_DELAYEDACKLOST); |
Pavel Emelyanov | 1ed8346 | 2008-07-16 20:29:51 -0700 | [diff] [blame] | 4413 | tcp_dsack_set(sk, TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4414 | |
| 4415 | out_of_window: |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 4416 | tcp_enter_quickack_mode(sk); |
| 4417 | inet_csk_schedule_ack(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4418 | drop: |
| 4419 | __kfree_skb(skb); |
| 4420 | return; |
| 4421 | } |
| 4422 | |
| 4423 | /* Out of window. F.e. zero window probe. */ |
| 4424 | if (!before(TCP_SKB_CB(skb)->seq, tp->rcv_nxt + tcp_receive_window(tp))) |
| 4425 | goto out_of_window; |
| 4426 | |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 4427 | tcp_enter_quickack_mode(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4428 | |
| 4429 | if (before(TCP_SKB_CB(skb)->seq, tp->rcv_nxt)) { |
| 4430 | /* Partial packet, seq < rcv_next < end_seq */ |
| 4431 | SOCK_DEBUG(sk, "partial packet: rcv_next %X seq %X - %X\n", |
| 4432 | tp->rcv_nxt, TCP_SKB_CB(skb)->seq, |
| 4433 | TCP_SKB_CB(skb)->end_seq); |
| 4434 | |
Pavel Emelyanov | 1ed8346 | 2008-07-16 20:29:51 -0700 | [diff] [blame] | 4435 | tcp_dsack_set(sk, TCP_SKB_CB(skb)->seq, tp->rcv_nxt); |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 4436 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4437 | /* If window is closed, drop tail of packet. But after |
| 4438 | * remembering D-SACK for its head made in previous line. |
| 4439 | */ |
| 4440 | if (!tcp_receive_window(tp)) |
| 4441 | goto out_of_window; |
| 4442 | goto queue_and_out; |
| 4443 | } |
| 4444 | |
Eric Dumazet | e86b2919 | 2012-03-18 11:06:44 +0000 | [diff] [blame] | 4445 | tcp_data_queue_ofo(sk, skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4446 | } |
| 4447 | |
Ilpo Järvinen | 2cf4663 | 2008-08-23 05:11:41 -0700 | [diff] [blame] | 4448 | static struct sk_buff *tcp_collapse_one(struct sock *sk, struct sk_buff *skb, |
| 4449 | struct sk_buff_head *list) |
| 4450 | { |
David S. Miller | 9152194 | 2009-05-28 21:35:47 -0700 | [diff] [blame] | 4451 | struct sk_buff *next = NULL; |
| 4452 | |
| 4453 | if (!skb_queue_is_last(list, skb)) |
| 4454 | next = skb_queue_next(list, skb); |
Ilpo Järvinen | 2cf4663 | 2008-08-23 05:11:41 -0700 | [diff] [blame] | 4455 | |
| 4456 | __skb_unlink(skb, list); |
| 4457 | __kfree_skb(skb); |
| 4458 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPRCVCOLLAPSED); |
| 4459 | |
| 4460 | return next; |
| 4461 | } |
| 4462 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4463 | /* Collapse contiguous sequence of skbs head..tail with |
| 4464 | * sequence numbers start..end. |
David S. Miller | 9152194 | 2009-05-28 21:35:47 -0700 | [diff] [blame] | 4465 | * |
| 4466 | * If tail is NULL, this means until the end of the list. |
| 4467 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4468 | * Segments with FIN/SYN are not collapsed (only because this |
| 4469 | * simplifies code) |
| 4470 | */ |
| 4471 | static void |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 4472 | tcp_collapse(struct sock *sk, struct sk_buff_head *list, |
| 4473 | struct sk_buff *head, struct sk_buff *tail, |
| 4474 | u32 start, u32 end) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4475 | { |
David S. Miller | 9152194 | 2009-05-28 21:35:47 -0700 | [diff] [blame] | 4476 | struct sk_buff *skb, *n; |
| 4477 | bool end_of_skbs; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4478 | |
Stephen Hemminger | caa20d9a | 2005-11-10 17:13:47 -0800 | [diff] [blame] | 4479 | /* First, check that queue is collapsible and find |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4480 | * the point where collapsing can be useful. */ |
David S. Miller | 9152194 | 2009-05-28 21:35:47 -0700 | [diff] [blame] | 4481 | skb = head; |
| 4482 | restart: |
| 4483 | end_of_skbs = true; |
| 4484 | skb_queue_walk_from_safe(list, skb, n) { |
| 4485 | if (skb == tail) |
| 4486 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4487 | /* No new bits? It is possible on ofo queue. */ |
| 4488 | if (!before(start, TCP_SKB_CB(skb)->end_seq)) { |
Ilpo Järvinen | 2cf4663 | 2008-08-23 05:11:41 -0700 | [diff] [blame] | 4489 | skb = tcp_collapse_one(sk, skb, list); |
David S. Miller | 9152194 | 2009-05-28 21:35:47 -0700 | [diff] [blame] | 4490 | if (!skb) |
| 4491 | break; |
| 4492 | goto restart; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4493 | } |
| 4494 | |
| 4495 | /* The first skb to collapse is: |
| 4496 | * - not SYN/FIN and |
| 4497 | * - bloated or contains data before "start" or |
| 4498 | * overlaps to the next one. |
| 4499 | */ |
Arnaldo Carvalho de Melo | aa8223c | 2007-04-10 21:04:22 -0700 | [diff] [blame] | 4500 | if (!tcp_hdr(skb)->syn && !tcp_hdr(skb)->fin && |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4501 | (tcp_win_from_space(skb->truesize) > skb->len || |
David S. Miller | 9152194 | 2009-05-28 21:35:47 -0700 | [diff] [blame] | 4502 | before(TCP_SKB_CB(skb)->seq, start))) { |
| 4503 | end_of_skbs = false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4504 | break; |
David S. Miller | 9152194 | 2009-05-28 21:35:47 -0700 | [diff] [blame] | 4505 | } |
| 4506 | |
| 4507 | if (!skb_queue_is_last(list, skb)) { |
| 4508 | struct sk_buff *next = skb_queue_next(list, skb); |
| 4509 | if (next != tail && |
| 4510 | TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(next)->seq) { |
| 4511 | end_of_skbs = false; |
| 4512 | break; |
| 4513 | } |
| 4514 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4515 | |
| 4516 | /* Decided to skip this, advance start seq. */ |
| 4517 | start = TCP_SKB_CB(skb)->end_seq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4518 | } |
David S. Miller | 9152194 | 2009-05-28 21:35:47 -0700 | [diff] [blame] | 4519 | if (end_of_skbs || tcp_hdr(skb)->syn || tcp_hdr(skb)->fin) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4520 | return; |
| 4521 | |
| 4522 | while (before(start, end)) { |
| 4523 | struct sk_buff *nskb; |
Chuck Lever | c2636b4 | 2007-10-23 21:07:32 -0700 | [diff] [blame] | 4524 | unsigned int header = skb_headroom(skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4525 | int copy = SKB_MAX_ORDER(header, 0); |
| 4526 | |
| 4527 | /* Too big header? This can happen with IPv6. */ |
| 4528 | if (copy < 0) |
| 4529 | return; |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 4530 | if (end - start < copy) |
| 4531 | copy = end - start; |
| 4532 | nskb = alloc_skb(copy + header, GFP_ATOMIC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4533 | if (!nskb) |
| 4534 | return; |
Arnaldo Carvalho de Melo | c51957d | 2007-03-10 12:47:22 -0300 | [diff] [blame] | 4535 | |
Arnaldo Carvalho de Melo | 98e399f | 2007-03-19 15:33:04 -0700 | [diff] [blame] | 4536 | skb_set_mac_header(nskb, skb_mac_header(skb) - skb->head); |
Arnaldo Carvalho de Melo | 9c70220 | 2007-04-25 18:04:18 -0700 | [diff] [blame] | 4537 | skb_set_network_header(nskb, (skb_network_header(skb) - |
| 4538 | skb->head)); |
| 4539 | skb_set_transport_header(nskb, (skb_transport_header(skb) - |
| 4540 | skb->head)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4541 | skb_reserve(nskb, header); |
| 4542 | memcpy(nskb->head, skb->head, header); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4543 | memcpy(nskb->cb, skb->cb, sizeof(skb->cb)); |
| 4544 | TCP_SKB_CB(nskb)->seq = TCP_SKB_CB(nskb)->end_seq = start; |
David S. Miller | 43f59c8 | 2008-09-21 21:28:51 -0700 | [diff] [blame] | 4545 | __skb_queue_before(list, skb, nskb); |
Hideo Aoki | 3ab224b | 2007-12-31 00:11:19 -0800 | [diff] [blame] | 4546 | skb_set_owner_r(nskb, sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4547 | |
| 4548 | /* Copy data, releasing collapsed skbs. */ |
| 4549 | while (copy > 0) { |
| 4550 | int offset = start - TCP_SKB_CB(skb)->seq; |
| 4551 | int size = TCP_SKB_CB(skb)->end_seq - start; |
| 4552 | |
Kris Katterjohn | 09a6266 | 2006-01-08 22:24:28 -0800 | [diff] [blame] | 4553 | BUG_ON(offset < 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4554 | if (size > 0) { |
| 4555 | size = min(copy, size); |
| 4556 | if (skb_copy_bits(skb, offset, skb_put(nskb, size), size)) |
| 4557 | BUG(); |
| 4558 | TCP_SKB_CB(nskb)->end_seq += size; |
| 4559 | copy -= size; |
| 4560 | start += size; |
| 4561 | } |
| 4562 | if (!before(start, TCP_SKB_CB(skb)->end_seq)) { |
Ilpo Järvinen | 2cf4663 | 2008-08-23 05:11:41 -0700 | [diff] [blame] | 4563 | skb = tcp_collapse_one(sk, skb, list); |
David S. Miller | 9152194 | 2009-05-28 21:35:47 -0700 | [diff] [blame] | 4564 | if (!skb || |
| 4565 | skb == tail || |
Arnaldo Carvalho de Melo | aa8223c | 2007-04-10 21:04:22 -0700 | [diff] [blame] | 4566 | tcp_hdr(skb)->syn || |
| 4567 | tcp_hdr(skb)->fin) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4568 | return; |
| 4569 | } |
| 4570 | } |
| 4571 | } |
| 4572 | } |
| 4573 | |
| 4574 | /* Collapse ofo queue. Algorithm: select contiguous sequence of skbs |
| 4575 | * and tcp_collapse() them until all the queue is collapsed. |
| 4576 | */ |
| 4577 | static void tcp_collapse_ofo_queue(struct sock *sk) |
| 4578 | { |
| 4579 | struct tcp_sock *tp = tcp_sk(sk); |
| 4580 | struct sk_buff *skb = skb_peek(&tp->out_of_order_queue); |
| 4581 | struct sk_buff *head; |
| 4582 | u32 start, end; |
| 4583 | |
| 4584 | if (skb == NULL) |
| 4585 | return; |
| 4586 | |
| 4587 | start = TCP_SKB_CB(skb)->seq; |
| 4588 | end = TCP_SKB_CB(skb)->end_seq; |
| 4589 | head = skb; |
| 4590 | |
| 4591 | for (;;) { |
David S. Miller | 9152194 | 2009-05-28 21:35:47 -0700 | [diff] [blame] | 4592 | struct sk_buff *next = NULL; |
| 4593 | |
| 4594 | if (!skb_queue_is_last(&tp->out_of_order_queue, skb)) |
| 4595 | next = skb_queue_next(&tp->out_of_order_queue, skb); |
| 4596 | skb = next; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4597 | |
| 4598 | /* Segment is terminated when we see gap or when |
| 4599 | * we are at the end of all the queue. */ |
David S. Miller | 9152194 | 2009-05-28 21:35:47 -0700 | [diff] [blame] | 4600 | if (!skb || |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4601 | after(TCP_SKB_CB(skb)->seq, end) || |
| 4602 | before(TCP_SKB_CB(skb)->end_seq, start)) { |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 4603 | tcp_collapse(sk, &tp->out_of_order_queue, |
| 4604 | head, skb, start, end); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4605 | head = skb; |
David S. Miller | 9152194 | 2009-05-28 21:35:47 -0700 | [diff] [blame] | 4606 | if (!skb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4607 | break; |
| 4608 | /* Start new segment */ |
| 4609 | start = TCP_SKB_CB(skb)->seq; |
| 4610 | end = TCP_SKB_CB(skb)->end_seq; |
| 4611 | } else { |
| 4612 | if (before(TCP_SKB_CB(skb)->seq, start)) |
| 4613 | start = TCP_SKB_CB(skb)->seq; |
| 4614 | if (after(TCP_SKB_CB(skb)->end_seq, end)) |
| 4615 | end = TCP_SKB_CB(skb)->end_seq; |
| 4616 | } |
| 4617 | } |
| 4618 | } |
| 4619 | |
Vitaliy Gusev | b000cd3 | 2008-04-15 00:33:38 -0700 | [diff] [blame] | 4620 | /* |
| 4621 | * Purge the out-of-order queue. |
Vitaliy Gusev | 56f367b | 2008-04-15 20:26:34 -0700 | [diff] [blame] | 4622 | * Return true if queue was pruned. |
Vitaliy Gusev | b000cd3 | 2008-04-15 00:33:38 -0700 | [diff] [blame] | 4623 | */ |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 4624 | static bool tcp_prune_ofo_queue(struct sock *sk) |
Vitaliy Gusev | b000cd3 | 2008-04-15 00:33:38 -0700 | [diff] [blame] | 4625 | { |
| 4626 | struct tcp_sock *tp = tcp_sk(sk); |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 4627 | bool res = false; |
Vitaliy Gusev | b000cd3 | 2008-04-15 00:33:38 -0700 | [diff] [blame] | 4628 | |
| 4629 | if (!skb_queue_empty(&tp->out_of_order_queue)) { |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 4630 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_OFOPRUNED); |
Vitaliy Gusev | b000cd3 | 2008-04-15 00:33:38 -0700 | [diff] [blame] | 4631 | __skb_queue_purge(&tp->out_of_order_queue); |
| 4632 | |
| 4633 | /* Reset SACK state. A conforming SACK implementation will |
| 4634 | * do the same at a timeout based retransmit. When a connection |
| 4635 | * is in a sad state like this, we care only about integrity |
| 4636 | * of the connection not performance. |
| 4637 | */ |
| 4638 | if (tp->rx_opt.sack_ok) |
| 4639 | tcp_sack_reset(&tp->rx_opt); |
| 4640 | sk_mem_reclaim(sk); |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 4641 | res = true; |
Vitaliy Gusev | b000cd3 | 2008-04-15 00:33:38 -0700 | [diff] [blame] | 4642 | } |
Vitaliy Gusev | 56f367b | 2008-04-15 20:26:34 -0700 | [diff] [blame] | 4643 | return res; |
Vitaliy Gusev | b000cd3 | 2008-04-15 00:33:38 -0700 | [diff] [blame] | 4644 | } |
| 4645 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4646 | /* Reduce allocated memory if we can, trying to get |
| 4647 | * the socket within its memory limits again. |
| 4648 | * |
| 4649 | * Return less than zero if we should start dropping frames |
| 4650 | * until the socket owning process reads some of the data |
| 4651 | * to stabilize the situation. |
| 4652 | */ |
| 4653 | static int tcp_prune_queue(struct sock *sk) |
| 4654 | { |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 4655 | struct tcp_sock *tp = tcp_sk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4656 | |
| 4657 | SOCK_DEBUG(sk, "prune_queue: c=%x\n", tp->copied_seq); |
| 4658 | |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 4659 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PRUNECALLED); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4660 | |
| 4661 | if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 4662 | tcp_clamp_window(sk); |
Glauber Costa | 180d8cd | 2011-12-11 21:47:02 +0000 | [diff] [blame] | 4663 | else if (sk_under_memory_pressure(sk)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4664 | tp->rcv_ssthresh = min(tp->rcv_ssthresh, 4U * tp->advmss); |
| 4665 | |
| 4666 | tcp_collapse_ofo_queue(sk); |
David S. Miller | 9152194 | 2009-05-28 21:35:47 -0700 | [diff] [blame] | 4667 | if (!skb_queue_empty(&sk->sk_receive_queue)) |
| 4668 | tcp_collapse(sk, &sk->sk_receive_queue, |
| 4669 | skb_peek(&sk->sk_receive_queue), |
| 4670 | NULL, |
| 4671 | tp->copied_seq, tp->rcv_nxt); |
Hideo Aoki | 3ab224b | 2007-12-31 00:11:19 -0800 | [diff] [blame] | 4672 | sk_mem_reclaim(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4673 | |
| 4674 | if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf) |
| 4675 | return 0; |
| 4676 | |
| 4677 | /* Collapsing did not help, destructive actions follow. |
| 4678 | * This must not ever occur. */ |
| 4679 | |
Vitaliy Gusev | b000cd3 | 2008-04-15 00:33:38 -0700 | [diff] [blame] | 4680 | tcp_prune_ofo_queue(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4681 | |
| 4682 | if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf) |
| 4683 | return 0; |
| 4684 | |
| 4685 | /* If we are really being abused, tell the caller to silently |
| 4686 | * drop receive data on the floor. It will get retransmitted |
| 4687 | * and hopefully then we'll have sufficient space. |
| 4688 | */ |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 4689 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_RCVPRUNED); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4690 | |
| 4691 | /* Massive buffer overcommit. */ |
| 4692 | tp->pred_flags = 0; |
| 4693 | return -1; |
| 4694 | } |
| 4695 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4696 | /* RFC2861, slow part. Adjust cwnd, after it was not full during one rto. |
| 4697 | * As additional protections, we do not touch cwnd in retransmission phases, |
| 4698 | * and if application hit its sndbuf limit recently. |
| 4699 | */ |
| 4700 | void tcp_cwnd_application_limited(struct sock *sk) |
| 4701 | { |
| 4702 | struct tcp_sock *tp = tcp_sk(sk); |
| 4703 | |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 4704 | if (inet_csk(sk)->icsk_ca_state == TCP_CA_Open && |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4705 | sk->sk_socket && !test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { |
| 4706 | /* Limited by application or receiver window. */ |
Ilpo Järvinen | d254bcd | 2006-08-04 16:57:42 -0700 | [diff] [blame] | 4707 | u32 init_win = tcp_init_cwnd(tp, __sk_dst_get(sk)); |
| 4708 | u32 win_used = max(tp->snd_cwnd_used, init_win); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4709 | if (win_used < tp->snd_cwnd) { |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 4710 | tp->snd_ssthresh = tcp_current_ssthresh(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4711 | tp->snd_cwnd = (tp->snd_cwnd + win_used) >> 1; |
| 4712 | } |
| 4713 | tp->snd_cwnd_used = 0; |
| 4714 | } |
| 4715 | tp->snd_cwnd_stamp = tcp_time_stamp; |
| 4716 | } |
| 4717 | |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 4718 | static bool tcp_should_expand_sndbuf(const struct sock *sk) |
David S. Miller | 0d9901d | 2005-07-05 15:21:10 -0700 | [diff] [blame] | 4719 | { |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 4720 | const struct tcp_sock *tp = tcp_sk(sk); |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 4721 | |
David S. Miller | 0d9901d | 2005-07-05 15:21:10 -0700 | [diff] [blame] | 4722 | /* If the user specified a specific send buffer setting, do |
| 4723 | * not modify it. |
| 4724 | */ |
| 4725 | if (sk->sk_userlocks & SOCK_SNDBUF_LOCK) |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 4726 | return false; |
David S. Miller | 0d9901d | 2005-07-05 15:21:10 -0700 | [diff] [blame] | 4727 | |
| 4728 | /* If we are under global TCP memory pressure, do not expand. */ |
Glauber Costa | 180d8cd | 2011-12-11 21:47:02 +0000 | [diff] [blame] | 4729 | if (sk_under_memory_pressure(sk)) |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 4730 | return false; |
David S. Miller | 0d9901d | 2005-07-05 15:21:10 -0700 | [diff] [blame] | 4731 | |
| 4732 | /* If we are under soft global TCP memory pressure, do not expand. */ |
Glauber Costa | 180d8cd | 2011-12-11 21:47:02 +0000 | [diff] [blame] | 4733 | if (sk_memory_allocated(sk) >= sk_prot_mem_limits(sk, 0)) |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 4734 | return false; |
David S. Miller | 0d9901d | 2005-07-05 15:21:10 -0700 | [diff] [blame] | 4735 | |
| 4736 | /* If we filled the congestion window, do not expand. */ |
| 4737 | if (tp->packets_out >= tp->snd_cwnd) |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 4738 | return false; |
David S. Miller | 0d9901d | 2005-07-05 15:21:10 -0700 | [diff] [blame] | 4739 | |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 4740 | return true; |
David S. Miller | 0d9901d | 2005-07-05 15:21:10 -0700 | [diff] [blame] | 4741 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4742 | |
| 4743 | /* When incoming ACK allowed to free some skb from write_queue, |
| 4744 | * we remember this event in flag SOCK_QUEUE_SHRUNK and wake up socket |
| 4745 | * on the exit from tcp input handler. |
| 4746 | * |
| 4747 | * PROBLEM: sndbuf expansion does not work well with largesend. |
| 4748 | */ |
| 4749 | static void tcp_new_space(struct sock *sk) |
| 4750 | { |
| 4751 | struct tcp_sock *tp = tcp_sk(sk); |
| 4752 | |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 4753 | if (tcp_should_expand_sndbuf(sk)) { |
Eric Dumazet | 6ae7053 | 2013-10-01 10:23:44 -0700 | [diff] [blame] | 4754 | tcp_sndbuf_expand(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4755 | tp->snd_cwnd_stamp = tcp_time_stamp; |
| 4756 | } |
| 4757 | |
| 4758 | sk->sk_write_space(sk); |
| 4759 | } |
| 4760 | |
Stephen Hemminger | 40efc6f | 2006-01-03 16:03:49 -0800 | [diff] [blame] | 4761 | static void tcp_check_space(struct sock *sk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4762 | { |
| 4763 | if (sock_flag(sk, SOCK_QUEUE_SHRUNK)) { |
| 4764 | sock_reset_flag(sk, SOCK_QUEUE_SHRUNK); |
| 4765 | if (sk->sk_socket && |
| 4766 | test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) |
| 4767 | tcp_new_space(sk); |
| 4768 | } |
| 4769 | } |
| 4770 | |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 4771 | static inline void tcp_data_snd_check(struct sock *sk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4772 | { |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 4773 | tcp_push_pending_frames(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4774 | tcp_check_space(sk); |
| 4775 | } |
| 4776 | |
| 4777 | /* |
| 4778 | * Check if sending an ack is needed. |
| 4779 | */ |
| 4780 | static void __tcp_ack_snd_check(struct sock *sk, int ofo_possible) |
| 4781 | { |
| 4782 | struct tcp_sock *tp = tcp_sk(sk); |
| 4783 | |
| 4784 | /* More than one full frame received... */ |
Joe Perches | 9d4fb27 | 2009-11-23 10:41:23 -0800 | [diff] [blame] | 4785 | if (((tp->rcv_nxt - tp->rcv_wup) > inet_csk(sk)->icsk_ack.rcv_mss && |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4786 | /* ... and right edge of window advances far enough. |
| 4787 | * (tcp_recvmsg() will send ACK otherwise). Or... |
| 4788 | */ |
Joe Perches | 9d4fb27 | 2009-11-23 10:41:23 -0800 | [diff] [blame] | 4789 | __tcp_select_window(sk) >= tp->rcv_wnd) || |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4790 | /* We ACK each frame or... */ |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 4791 | tcp_in_quickack_mode(sk) || |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4792 | /* We have out of order data. */ |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 4793 | (ofo_possible && skb_peek(&tp->out_of_order_queue))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4794 | /* Then ack it now */ |
| 4795 | tcp_send_ack(sk); |
| 4796 | } else { |
| 4797 | /* Else, send delayed ack. */ |
| 4798 | tcp_send_delayed_ack(sk); |
| 4799 | } |
| 4800 | } |
| 4801 | |
Stephen Hemminger | 40efc6f | 2006-01-03 16:03:49 -0800 | [diff] [blame] | 4802 | static inline void tcp_ack_snd_check(struct sock *sk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4803 | { |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 4804 | if (!inet_csk_ack_scheduled(sk)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4805 | /* We sent a data segment already. */ |
| 4806 | return; |
| 4807 | } |
| 4808 | __tcp_ack_snd_check(sk, 1); |
| 4809 | } |
| 4810 | |
| 4811 | /* |
| 4812 | * This routine is only called when we have urgent data |
Stephen Hemminger | caa20d9a | 2005-11-10 17:13:47 -0800 | [diff] [blame] | 4813 | * signaled. Its the 'slow' part of tcp_urg. It could be |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4814 | * moved inline now as tcp_urg is only called from one |
| 4815 | * place. We handle URGent data wrong. We have to - as |
| 4816 | * BSD still doesn't use the correction from RFC961. |
| 4817 | * For 1003.1g we should support a new option TCP_STDURG to permit |
| 4818 | * either form (or just set the sysctl tcp_stdurg). |
| 4819 | */ |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 4820 | |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 4821 | static void tcp_check_urg(struct sock *sk, const struct tcphdr *th) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4822 | { |
| 4823 | struct tcp_sock *tp = tcp_sk(sk); |
| 4824 | u32 ptr = ntohs(th->urg_ptr); |
| 4825 | |
| 4826 | if (ptr && !sysctl_tcp_stdurg) |
| 4827 | ptr--; |
| 4828 | ptr += ntohl(th->seq); |
| 4829 | |
| 4830 | /* Ignore urgent data that we've already seen and read. */ |
| 4831 | if (after(tp->copied_seq, ptr)) |
| 4832 | return; |
| 4833 | |
| 4834 | /* Do not replay urg ptr. |
| 4835 | * |
| 4836 | * NOTE: interesting situation not covered by specs. |
| 4837 | * Misbehaving sender may send urg ptr, pointing to segment, |
| 4838 | * which we already have in ofo queue. We are not able to fetch |
| 4839 | * such data and will stay in TCP_URG_NOTYET until will be eaten |
| 4840 | * by recvmsg(). Seems, we are not obliged to handle such wicked |
| 4841 | * situations. But it is worth to think about possibility of some |
| 4842 | * DoSes using some hypothetical application level deadlock. |
| 4843 | */ |
| 4844 | if (before(ptr, tp->rcv_nxt)) |
| 4845 | return; |
| 4846 | |
| 4847 | /* Do we already have a newer (or duplicate) urgent pointer? */ |
| 4848 | if (tp->urg_data && !after(ptr, tp->urg_seq)) |
| 4849 | return; |
| 4850 | |
| 4851 | /* Tell the world about our new urgent pointer. */ |
| 4852 | sk_send_sigurg(sk); |
| 4853 | |
| 4854 | /* We may be adding urgent data when the last byte read was |
| 4855 | * urgent. To do this requires some care. We cannot just ignore |
| 4856 | * tp->copied_seq since we would read the last urgent byte again |
| 4857 | * as data, nor can we alter copied_seq until this data arrives |
Stephen Hemminger | caa20d9a | 2005-11-10 17:13:47 -0800 | [diff] [blame] | 4858 | * or we break the semantics of SIOCATMARK (and thus sockatmark()) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4859 | * |
| 4860 | * NOTE. Double Dutch. Rendering to plain English: author of comment |
| 4861 | * above did something sort of send("A", MSG_OOB); send("B", MSG_OOB); |
| 4862 | * and expect that both A and B disappear from stream. This is _wrong_. |
| 4863 | * Though this happens in BSD with high probability, this is occasional. |
| 4864 | * Any application relying on this is buggy. Note also, that fix "works" |
| 4865 | * only in this artificial test. Insert some normal data between A and B and we will |
| 4866 | * decline of BSD again. Verdict: it is better to remove to trap |
| 4867 | * buggy users. |
| 4868 | */ |
| 4869 | if (tp->urg_seq == tp->copied_seq && tp->urg_data && |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 4870 | !sock_flag(sk, SOCK_URGINLINE) && tp->copied_seq != tp->rcv_nxt) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4871 | struct sk_buff *skb = skb_peek(&sk->sk_receive_queue); |
| 4872 | tp->copied_seq++; |
| 4873 | if (skb && !before(tp->copied_seq, TCP_SKB_CB(skb)->end_seq)) { |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 4874 | __skb_unlink(skb, &sk->sk_receive_queue); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4875 | __kfree_skb(skb); |
| 4876 | } |
| 4877 | } |
| 4878 | |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 4879 | tp->urg_data = TCP_URG_NOTYET; |
| 4880 | tp->urg_seq = ptr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4881 | |
| 4882 | /* Disable header prediction. */ |
| 4883 | tp->pred_flags = 0; |
| 4884 | } |
| 4885 | |
| 4886 | /* This is the 'fast' part of urgent handling. */ |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 4887 | static void tcp_urg(struct sock *sk, struct sk_buff *skb, const struct tcphdr *th) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4888 | { |
| 4889 | struct tcp_sock *tp = tcp_sk(sk); |
| 4890 | |
| 4891 | /* Check if we get a new urgent pointer - normally not. */ |
| 4892 | if (th->urg) |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 4893 | tcp_check_urg(sk, th); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4894 | |
| 4895 | /* Do we wait for any urgent data? - normally not... */ |
| 4896 | if (tp->urg_data == TCP_URG_NOTYET) { |
| 4897 | u32 ptr = tp->urg_seq - ntohl(th->seq) + (th->doff * 4) - |
| 4898 | th->syn; |
| 4899 | |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 4900 | /* Is the urgent pointer pointing into this packet? */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4901 | if (ptr < skb->len) { |
| 4902 | u8 tmp; |
| 4903 | if (skb_copy_bits(skb, ptr, &tmp, 1)) |
| 4904 | BUG(); |
| 4905 | tp->urg_data = TCP_URG_VALID | tmp; |
| 4906 | if (!sock_flag(sk, SOCK_DEAD)) |
| 4907 | sk->sk_data_ready(sk, 0); |
| 4908 | } |
| 4909 | } |
| 4910 | } |
| 4911 | |
| 4912 | static int tcp_copy_to_iovec(struct sock *sk, struct sk_buff *skb, int hlen) |
| 4913 | { |
| 4914 | struct tcp_sock *tp = tcp_sk(sk); |
| 4915 | int chunk = skb->len - hlen; |
| 4916 | int err; |
| 4917 | |
| 4918 | local_bh_enable(); |
Herbert Xu | 6047637 | 2007-04-09 11:59:39 -0700 | [diff] [blame] | 4919 | if (skb_csum_unnecessary(skb)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4920 | err = skb_copy_datagram_iovec(skb, hlen, tp->ucopy.iov, chunk); |
| 4921 | else |
| 4922 | err = skb_copy_and_csum_datagram_iovec(skb, hlen, |
| 4923 | tp->ucopy.iov); |
| 4924 | |
| 4925 | if (!err) { |
| 4926 | tp->ucopy.len -= chunk; |
| 4927 | tp->copied_seq += chunk; |
| 4928 | tcp_rcv_space_adjust(sk); |
| 4929 | } |
| 4930 | |
| 4931 | local_bh_disable(); |
| 4932 | return err; |
| 4933 | } |
| 4934 | |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 4935 | static __sum16 __tcp_checksum_complete_user(struct sock *sk, |
| 4936 | struct sk_buff *skb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4937 | { |
Al Viro | b51655b | 2006-11-14 21:40:42 -0800 | [diff] [blame] | 4938 | __sum16 result; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4939 | |
| 4940 | if (sock_owned_by_user(sk)) { |
| 4941 | local_bh_enable(); |
| 4942 | result = __tcp_checksum_complete(skb); |
| 4943 | local_bh_disable(); |
| 4944 | } else { |
| 4945 | result = __tcp_checksum_complete(skb); |
| 4946 | } |
| 4947 | return result; |
| 4948 | } |
| 4949 | |
Vijay Subramanian | 67b95bd | 2012-07-19 21:32:18 +0000 | [diff] [blame] | 4950 | static inline bool tcp_checksum_complete_user(struct sock *sk, |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 4951 | struct sk_buff *skb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4952 | { |
Herbert Xu | 6047637 | 2007-04-09 11:59:39 -0700 | [diff] [blame] | 4953 | return !skb_csum_unnecessary(skb) && |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 4954 | __tcp_checksum_complete_user(sk, skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4955 | } |
| 4956 | |
Chris Leech | 1a2449a | 2006-05-23 18:05:53 -0700 | [diff] [blame] | 4957 | #ifdef CONFIG_NET_DMA |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 4958 | static bool tcp_dma_try_early_copy(struct sock *sk, struct sk_buff *skb, |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 4959 | int hlen) |
Chris Leech | 1a2449a | 2006-05-23 18:05:53 -0700 | [diff] [blame] | 4960 | { |
| 4961 | struct tcp_sock *tp = tcp_sk(sk); |
| 4962 | int chunk = skb->len - hlen; |
| 4963 | int dma_cookie; |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 4964 | bool copied_early = false; |
Chris Leech | 1a2449a | 2006-05-23 18:05:53 -0700 | [diff] [blame] | 4965 | |
| 4966 | if (tp->ucopy.wakeup) |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 4967 | return false; |
Chris Leech | 1a2449a | 2006-05-23 18:05:53 -0700 | [diff] [blame] | 4968 | |
| 4969 | if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list) |
Dave Jiang | a2bd114 | 2012-04-04 16:10:46 -0700 | [diff] [blame] | 4970 | tp->ucopy.dma_chan = net_dma_find_channel(); |
Chris Leech | 1a2449a | 2006-05-23 18:05:53 -0700 | [diff] [blame] | 4971 | |
Herbert Xu | 6047637 | 2007-04-09 11:59:39 -0700 | [diff] [blame] | 4972 | if (tp->ucopy.dma_chan && skb_csum_unnecessary(skb)) { |
Chris Leech | 1a2449a | 2006-05-23 18:05:53 -0700 | [diff] [blame] | 4973 | |
| 4974 | dma_cookie = dma_skb_copy_datagram_iovec(tp->ucopy.dma_chan, |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 4975 | skb, hlen, |
| 4976 | tp->ucopy.iov, chunk, |
| 4977 | tp->ucopy.pinned_list); |
Chris Leech | 1a2449a | 2006-05-23 18:05:53 -0700 | [diff] [blame] | 4978 | |
| 4979 | if (dma_cookie < 0) |
| 4980 | goto out; |
| 4981 | |
| 4982 | tp->ucopy.dma_cookie = dma_cookie; |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 4983 | copied_early = true; |
Chris Leech | 1a2449a | 2006-05-23 18:05:53 -0700 | [diff] [blame] | 4984 | |
| 4985 | tp->ucopy.len -= chunk; |
| 4986 | tp->copied_seq += chunk; |
| 4987 | tcp_rcv_space_adjust(sk); |
| 4988 | |
| 4989 | if ((tp->ucopy.len == 0) || |
Arnaldo Carvalho de Melo | aa8223c | 2007-04-10 21:04:22 -0700 | [diff] [blame] | 4990 | (tcp_flag_word(tcp_hdr(skb)) & TCP_FLAG_PSH) || |
Chris Leech | 1a2449a | 2006-05-23 18:05:53 -0700 | [diff] [blame] | 4991 | (atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1))) { |
| 4992 | tp->ucopy.wakeup = 1; |
| 4993 | sk->sk_data_ready(sk, 0); |
| 4994 | } |
| 4995 | } else if (chunk > 0) { |
| 4996 | tp->ucopy.wakeup = 1; |
| 4997 | sk->sk_data_ready(sk, 0); |
| 4998 | } |
| 4999 | out: |
| 5000 | return copied_early; |
| 5001 | } |
| 5002 | #endif /* CONFIG_NET_DMA */ |
| 5003 | |
Ilpo Järvinen | cbe2d12 | 2008-08-23 05:10:12 -0700 | [diff] [blame] | 5004 | /* Does PAWS and seqno based validation of an incoming segment, flags will |
| 5005 | * play significant role here. |
| 5006 | */ |
Eric Dumazet | 0c24604 | 2012-07-17 01:41:30 +0000 | [diff] [blame] | 5007 | static bool tcp_validate_incoming(struct sock *sk, struct sk_buff *skb, |
| 5008 | const struct tcphdr *th, int syn_inerr) |
Ilpo Järvinen | cbe2d12 | 2008-08-23 05:10:12 -0700 | [diff] [blame] | 5009 | { |
| 5010 | struct tcp_sock *tp = tcp_sk(sk); |
| 5011 | |
| 5012 | /* RFC1323: H1. Apply PAWS check first. */ |
Christoph Paasch | 1a2c618 | 2013-03-17 08:23:34 +0000 | [diff] [blame] | 5013 | if (tcp_fast_parse_options(skb, th, tp) && tp->rx_opt.saw_tstamp && |
Ilpo Järvinen | cbe2d12 | 2008-08-23 05:10:12 -0700 | [diff] [blame] | 5014 | tcp_paws_discard(sk, skb)) { |
| 5015 | if (!th->rst) { |
| 5016 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSESTABREJECTED); |
| 5017 | tcp_send_dupack(sk, skb); |
| 5018 | goto discard; |
| 5019 | } |
| 5020 | /* Reset is accepted even if it did not pass PAWS. */ |
| 5021 | } |
| 5022 | |
| 5023 | /* Step 1: check sequence number */ |
| 5024 | if (!tcp_sequence(tp, TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq)) { |
| 5025 | /* RFC793, page 37: "In all states except SYN-SENT, all reset |
| 5026 | * (RST) segments are validated by checking their SEQ-fields." |
| 5027 | * And page 69: "If an incoming segment is not acceptable, |
| 5028 | * an acknowledgment should be sent in reply (unless the RST |
| 5029 | * bit is set, if so drop the segment and return)". |
| 5030 | */ |
Eric Dumazet | e371589 | 2012-07-17 12:29:30 +0000 | [diff] [blame] | 5031 | if (!th->rst) { |
| 5032 | if (th->syn) |
| 5033 | goto syn_challenge; |
Ilpo Järvinen | cbe2d12 | 2008-08-23 05:10:12 -0700 | [diff] [blame] | 5034 | tcp_send_dupack(sk, skb); |
Eric Dumazet | e371589 | 2012-07-17 12:29:30 +0000 | [diff] [blame] | 5035 | } |
Ilpo Järvinen | cbe2d12 | 2008-08-23 05:10:12 -0700 | [diff] [blame] | 5036 | goto discard; |
| 5037 | } |
| 5038 | |
| 5039 | /* Step 2: check RST bit */ |
| 5040 | if (th->rst) { |
Eric Dumazet | 282f23c | 2012-07-17 10:13:05 +0200 | [diff] [blame] | 5041 | /* RFC 5961 3.2 : |
| 5042 | * If sequence number exactly matches RCV.NXT, then |
| 5043 | * RESET the connection |
| 5044 | * else |
| 5045 | * Send a challenge ACK |
| 5046 | */ |
| 5047 | if (TCP_SKB_CB(skb)->seq == tp->rcv_nxt) |
| 5048 | tcp_reset(sk); |
| 5049 | else |
| 5050 | tcp_send_challenge_ack(sk); |
Ilpo Järvinen | cbe2d12 | 2008-08-23 05:10:12 -0700 | [diff] [blame] | 5051 | goto discard; |
| 5052 | } |
| 5053 | |
Ilpo Järvinen | cbe2d12 | 2008-08-23 05:10:12 -0700 | [diff] [blame] | 5054 | /* step 3: check security and precedence [ignored] */ |
| 5055 | |
Eric Dumazet | 0c24604 | 2012-07-17 01:41:30 +0000 | [diff] [blame] | 5056 | /* step 4: Check for a SYN |
| 5057 | * RFC 5691 4.2 : Send a challenge ack |
| 5058 | */ |
| 5059 | if (th->syn) { |
Eric Dumazet | e371589 | 2012-07-17 12:29:30 +0000 | [diff] [blame] | 5060 | syn_challenge: |
Ilpo Järvinen | cbe2d12 | 2008-08-23 05:10:12 -0700 | [diff] [blame] | 5061 | if (syn_inerr) |
| 5062 | TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS); |
Eric Dumazet | 0c24604 | 2012-07-17 01:41:30 +0000 | [diff] [blame] | 5063 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPSYNCHALLENGE); |
| 5064 | tcp_send_challenge_ack(sk); |
| 5065 | goto discard; |
Ilpo Järvinen | cbe2d12 | 2008-08-23 05:10:12 -0700 | [diff] [blame] | 5066 | } |
| 5067 | |
Eric Dumazet | 0c24604 | 2012-07-17 01:41:30 +0000 | [diff] [blame] | 5068 | return true; |
Ilpo Järvinen | cbe2d12 | 2008-08-23 05:10:12 -0700 | [diff] [blame] | 5069 | |
| 5070 | discard: |
| 5071 | __kfree_skb(skb); |
Eric Dumazet | 0c24604 | 2012-07-17 01:41:30 +0000 | [diff] [blame] | 5072 | return false; |
Ilpo Järvinen | cbe2d12 | 2008-08-23 05:10:12 -0700 | [diff] [blame] | 5073 | } |
| 5074 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5075 | /* |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5076 | * TCP receive function for the ESTABLISHED state. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5077 | * |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5078 | * It is split into a fast path and a slow path. The fast path is |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5079 | * disabled when: |
| 5080 | * - A zero window was announced from us - zero window probing |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5081 | * is only handled properly in the slow path. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5082 | * - Out of order segments arrived. |
| 5083 | * - Urgent data is expected. |
| 5084 | * - There is no buffer space left |
| 5085 | * - Unexpected TCP flags/window values/header lengths are received |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5086 | * (detected by checking the TCP header against pred_flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5087 | * - Data is sent in both directions. Fast path only supports pure senders |
| 5088 | * or pure receivers (this means either the sequence number or the ack |
| 5089 | * value must stay constant) |
| 5090 | * - Unexpected TCP option. |
| 5091 | * |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5092 | * When these conditions are not satisfied it drops into a standard |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5093 | * receive procedure patterned after RFC793 to handle all cases. |
| 5094 | * The first three cases are guaranteed by proper pred_flags setting, |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5095 | * the rest is checked inline. Fast processing is turned on in |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5096 | * tcp_data_queue when everything is OK. |
| 5097 | */ |
Vijay Subramanian | c995ae2 | 2013-09-03 12:23:22 -0700 | [diff] [blame] | 5098 | void tcp_rcv_established(struct sock *sk, struct sk_buff *skb, |
| 5099 | const struct tcphdr *th, unsigned int len) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5100 | { |
| 5101 | struct tcp_sock *tp = tcp_sk(sk); |
| 5102 | |
Eric Dumazet | 5d299f3 | 2012-08-06 05:09:33 +0000 | [diff] [blame] | 5103 | if (unlikely(sk->sk_rx_dst == NULL)) |
| 5104 | inet_csk(sk)->icsk_af_ops->sk_rx_dst_set(sk, skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5105 | /* |
| 5106 | * Header prediction. |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5107 | * The code loosely follows the one in the famous |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5108 | * "30 instruction TCP receive" Van Jacobson mail. |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5109 | * |
| 5110 | * Van's trick is to deposit buffers into socket queue |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5111 | * on a device interrupt, to call tcp_recv function |
| 5112 | * on the receive process context and checksum and copy |
| 5113 | * the buffer to user space. smart... |
| 5114 | * |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5115 | * Our current scheme is not silly either but we take the |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5116 | * extra cost of the net_bh soft interrupt processing... |
| 5117 | * We do checksum and copy also but from device to kernel. |
| 5118 | */ |
| 5119 | |
| 5120 | tp->rx_opt.saw_tstamp = 0; |
| 5121 | |
| 5122 | /* pred_flags is 0xS?10 << 16 + snd_wnd |
Stephen Hemminger | caa20d9a | 2005-11-10 17:13:47 -0800 | [diff] [blame] | 5123 | * if header_prediction is to be made |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5124 | * 'S' will always be tp->tcp_header_len >> 2 |
| 5125 | * '?' will be 0 for the fast path, otherwise pred_flags is 0 to |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5126 | * turn it off (when there are holes in the receive |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5127 | * space for instance) |
| 5128 | * PSH flag is ignored. |
| 5129 | */ |
| 5130 | |
| 5131 | if ((tcp_flag_word(th) & TCP_HP_BITS) == tp->pred_flags && |
John Dykstra | 96e0bf4 | 2009-03-22 21:49:57 -0700 | [diff] [blame] | 5132 | TCP_SKB_CB(skb)->seq == tp->rcv_nxt && |
| 5133 | !after(TCP_SKB_CB(skb)->ack_seq, tp->snd_nxt)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5134 | int tcp_header_len = tp->tcp_header_len; |
| 5135 | |
| 5136 | /* Timestamp header prediction: tcp_header_len |
| 5137 | * is automatically equal to th->doff*4 due to pred_flags |
| 5138 | * match. |
| 5139 | */ |
| 5140 | |
| 5141 | /* Check timestamp */ |
| 5142 | if (tcp_header_len == sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5143 | /* No? Slow path! */ |
Ilpo Järvinen | a4356b2 | 2008-08-23 05:12:29 -0700 | [diff] [blame] | 5144 | if (!tcp_parse_aligned_timestamp(tp, th)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5145 | goto slow_path; |
| 5146 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5147 | /* If PAWS failed, check it more carefully in slow path */ |
| 5148 | if ((s32)(tp->rx_opt.rcv_tsval - tp->rx_opt.ts_recent) < 0) |
| 5149 | goto slow_path; |
| 5150 | |
| 5151 | /* DO NOT update ts_recent here, if checksum fails |
| 5152 | * and timestamp was corrupted part, it will result |
| 5153 | * in a hung connection since we will drop all |
| 5154 | * future packets due to the PAWS test. |
| 5155 | */ |
| 5156 | } |
| 5157 | |
| 5158 | if (len <= tcp_header_len) { |
| 5159 | /* Bulk data transfer: sender */ |
| 5160 | if (len == tcp_header_len) { |
| 5161 | /* Predicted packet is in window by definition. |
| 5162 | * seq == rcv_nxt and rcv_wup <= rcv_nxt. |
| 5163 | * Hence, check seq<=rcv_wup reduces to: |
| 5164 | */ |
| 5165 | if (tcp_header_len == |
| 5166 | (sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED) && |
| 5167 | tp->rcv_nxt == tp->rcv_wup) |
| 5168 | tcp_store_ts_recent(tp); |
| 5169 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5170 | /* We know that such packets are checksummed |
| 5171 | * on entry. |
| 5172 | */ |
| 5173 | tcp_ack(sk, skb, 0); |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5174 | __kfree_skb(skb); |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 5175 | tcp_data_snd_check(sk); |
Vijay Subramanian | c995ae2 | 2013-09-03 12:23:22 -0700 | [diff] [blame] | 5176 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5177 | } else { /* Header too small */ |
Pavel Emelyanov | 63231bd | 2008-07-16 20:22:25 -0700 | [diff] [blame] | 5178 | TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5179 | goto discard; |
| 5180 | } |
| 5181 | } else { |
| 5182 | int eaten = 0; |
Chris Leech | 1a2449a | 2006-05-23 18:05:53 -0700 | [diff] [blame] | 5183 | int copied_early = 0; |
Eric Dumazet | b081f85 | 2012-05-02 09:58:29 +0000 | [diff] [blame] | 5184 | bool fragstolen = false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5185 | |
Chris Leech | 1a2449a | 2006-05-23 18:05:53 -0700 | [diff] [blame] | 5186 | if (tp->copied_seq == tp->rcv_nxt && |
| 5187 | len - tcp_header_len <= tp->ucopy.len) { |
| 5188 | #ifdef CONFIG_NET_DMA |
Jiri Kosina | 59ea33a | 2012-07-27 10:38:50 +0000 | [diff] [blame] | 5189 | if (tp->ucopy.task == current && |
| 5190 | sock_owned_by_user(sk) && |
| 5191 | tcp_dma_try_early_copy(sk, skb, tcp_header_len)) { |
Chris Leech | 1a2449a | 2006-05-23 18:05:53 -0700 | [diff] [blame] | 5192 | copied_early = 1; |
| 5193 | eaten = 1; |
| 5194 | } |
| 5195 | #endif |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 5196 | if (tp->ucopy.task == current && |
| 5197 | sock_owned_by_user(sk) && !copied_early) { |
Chris Leech | 1a2449a | 2006-05-23 18:05:53 -0700 | [diff] [blame] | 5198 | __set_current_state(TASK_RUNNING); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5199 | |
Chris Leech | 1a2449a | 2006-05-23 18:05:53 -0700 | [diff] [blame] | 5200 | if (!tcp_copy_to_iovec(sk, skb, tcp_header_len)) |
| 5201 | eaten = 1; |
| 5202 | } |
| 5203 | if (eaten) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5204 | /* Predicted packet is in window by definition. |
| 5205 | * seq == rcv_nxt and rcv_wup <= rcv_nxt. |
| 5206 | * Hence, check seq<=rcv_wup reduces to: |
| 5207 | */ |
| 5208 | if (tcp_header_len == |
| 5209 | (sizeof(struct tcphdr) + |
| 5210 | TCPOLEN_TSTAMP_ALIGNED) && |
| 5211 | tp->rcv_nxt == tp->rcv_wup) |
| 5212 | tcp_store_ts_recent(tp); |
| 5213 | |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 5214 | tcp_rcv_rtt_measure_ts(sk, skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5215 | |
| 5216 | __skb_pull(skb, tcp_header_len); |
| 5217 | tp->rcv_nxt = TCP_SKB_CB(skb)->end_seq; |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 5218 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPHPHITSTOUSER); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5219 | } |
Chris Leech | 1a2449a | 2006-05-23 18:05:53 -0700 | [diff] [blame] | 5220 | if (copied_early) |
| 5221 | tcp_cleanup_rbuf(sk, skb->len); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5222 | } |
| 5223 | if (!eaten) { |
| 5224 | if (tcp_checksum_complete_user(sk, skb)) |
| 5225 | goto csum_error; |
| 5226 | |
Neal Cardwell | aab2b4b | 2013-03-04 06:23:05 +0000 | [diff] [blame] | 5227 | if ((int)skb->truesize > sk->sk_forward_alloc) |
| 5228 | goto step5; |
| 5229 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5230 | /* Predicted packet is in window by definition. |
| 5231 | * seq == rcv_nxt and rcv_wup <= rcv_nxt. |
| 5232 | * Hence, check seq<=rcv_wup reduces to: |
| 5233 | */ |
| 5234 | if (tcp_header_len == |
| 5235 | (sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED) && |
| 5236 | tp->rcv_nxt == tp->rcv_wup) |
| 5237 | tcp_store_ts_recent(tp); |
| 5238 | |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 5239 | tcp_rcv_rtt_measure_ts(sk, skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5240 | |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 5241 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPHPHITS); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5242 | |
| 5243 | /* Bulk data transfer: receiver */ |
Eric Dumazet | b081f85 | 2012-05-02 09:58:29 +0000 | [diff] [blame] | 5244 | eaten = tcp_queue_rcv(sk, skb, tcp_header_len, |
| 5245 | &fragstolen); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5246 | } |
| 5247 | |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 5248 | tcp_event_data_recv(sk, skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5249 | |
| 5250 | if (TCP_SKB_CB(skb)->ack_seq != tp->snd_una) { |
| 5251 | /* Well, only one small jumplet in fast path... */ |
| 5252 | tcp_ack(sk, skb, FLAG_DATA); |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 5253 | tcp_data_snd_check(sk); |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 5254 | if (!inet_csk_ack_scheduled(sk)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5255 | goto no_ack; |
| 5256 | } |
| 5257 | |
Ali Saidi | 53240c2 | 2008-10-07 15:31:19 -0700 | [diff] [blame] | 5258 | if (!copied_early || tp->rcv_nxt != tp->rcv_wup) |
| 5259 | __tcp_ack_snd_check(sk, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5260 | no_ack: |
Chris Leech | 1a2449a | 2006-05-23 18:05:53 -0700 | [diff] [blame] | 5261 | #ifdef CONFIG_NET_DMA |
| 5262 | if (copied_early) |
| 5263 | __skb_queue_tail(&sk->sk_async_wait_queue, skb); |
| 5264 | else |
| 5265 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5266 | if (eaten) |
Eric Dumazet | b081f85 | 2012-05-02 09:58:29 +0000 | [diff] [blame] | 5267 | kfree_skb_partial(skb, fragstolen); |
Eric Dumazet | 1d57f19 | 2012-09-17 12:51:39 +0000 | [diff] [blame] | 5268 | sk->sk_data_ready(sk, 0); |
Vijay Subramanian | c995ae2 | 2013-09-03 12:23:22 -0700 | [diff] [blame] | 5269 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5270 | } |
| 5271 | } |
| 5272 | |
| 5273 | slow_path: |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 5274 | if (len < (th->doff << 2) || tcp_checksum_complete_user(sk, skb)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5275 | goto csum_error; |
| 5276 | |
Eric Dumazet | 7b514a8 | 2013-01-10 16:18:47 +0000 | [diff] [blame] | 5277 | if (!th->ack && !th->rst) |
Eric Dumazet | c3ae62af | 2012-12-26 12:44:34 +0000 | [diff] [blame] | 5278 | goto discard; |
| 5279 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5280 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5281 | * Standard slow path. |
| 5282 | */ |
| 5283 | |
Eric Dumazet | 0c24604 | 2012-07-17 01:41:30 +0000 | [diff] [blame] | 5284 | if (!tcp_validate_incoming(sk, skb, th, 1)) |
Vijay Subramanian | c995ae2 | 2013-09-03 12:23:22 -0700 | [diff] [blame] | 5285 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5286 | |
| 5287 | step5: |
Eric Dumazet | 12fb3dd | 2013-04-19 07:19:48 +0000 | [diff] [blame] | 5288 | if (tcp_ack(sk, skb, FLAG_SLOWPATH | FLAG_UPDATE_TS_RECENT) < 0) |
John Dykstra | 96e0bf4 | 2009-03-22 21:49:57 -0700 | [diff] [blame] | 5289 | goto discard; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5290 | |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 5291 | tcp_rcv_rtt_measure_ts(sk, skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5292 | |
| 5293 | /* Process urgent data. */ |
| 5294 | tcp_urg(sk, skb, th); |
| 5295 | |
| 5296 | /* step 7: process the segment text */ |
| 5297 | tcp_data_queue(sk, skb); |
| 5298 | |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 5299 | tcp_data_snd_check(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5300 | tcp_ack_snd_check(sk); |
Vijay Subramanian | c995ae2 | 2013-09-03 12:23:22 -0700 | [diff] [blame] | 5301 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5302 | |
| 5303 | csum_error: |
Eric Dumazet | 6a5dc9e | 2013-04-29 08:39:56 +0000 | [diff] [blame] | 5304 | TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS); |
Pavel Emelyanov | 63231bd | 2008-07-16 20:22:25 -0700 | [diff] [blame] | 5305 | TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5306 | |
| 5307 | discard: |
| 5308 | __kfree_skb(skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5309 | } |
Eric Dumazet | 4bc2f18 | 2010-07-09 21:22:10 +0000 | [diff] [blame] | 5310 | EXPORT_SYMBOL(tcp_rcv_established); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5311 | |
Pavel Emelyanov | 370816a | 2012-04-19 03:40:01 +0000 | [diff] [blame] | 5312 | void tcp_finish_connect(struct sock *sk, struct sk_buff *skb) |
| 5313 | { |
| 5314 | struct tcp_sock *tp = tcp_sk(sk); |
| 5315 | struct inet_connection_sock *icsk = inet_csk(sk); |
| 5316 | |
| 5317 | tcp_set_state(sk, TCP_ESTABLISHED); |
| 5318 | |
David S. Miller | 41063e9 | 2012-06-19 21:22:05 -0700 | [diff] [blame] | 5319 | if (skb != NULL) { |
Eric Dumazet | 5d299f3 | 2012-08-06 05:09:33 +0000 | [diff] [blame] | 5320 | icsk->icsk_af_ops->sk_rx_dst_set(sk, skb); |
Pavel Emelyanov | 370816a | 2012-04-19 03:40:01 +0000 | [diff] [blame] | 5321 | security_inet_conn_established(sk, skb); |
David S. Miller | 41063e9 | 2012-06-19 21:22:05 -0700 | [diff] [blame] | 5322 | } |
Pavel Emelyanov | 370816a | 2012-04-19 03:40:01 +0000 | [diff] [blame] | 5323 | |
| 5324 | /* Make sure socket is routed, for correct metrics. */ |
| 5325 | icsk->icsk_af_ops->rebuild_header(sk); |
| 5326 | |
| 5327 | tcp_init_metrics(sk); |
| 5328 | |
| 5329 | tcp_init_congestion_control(sk); |
| 5330 | |
| 5331 | /* Prevent spurious tcp_cwnd_restart() on first data |
| 5332 | * packet. |
| 5333 | */ |
| 5334 | tp->lsndtime = tcp_time_stamp; |
| 5335 | |
| 5336 | tcp_init_buffer_space(sk); |
| 5337 | |
| 5338 | if (sock_flag(sk, SOCK_KEEPOPEN)) |
| 5339 | inet_csk_reset_keepalive_timer(sk, keepalive_time_when(tp)); |
| 5340 | |
| 5341 | if (!tp->rx_opt.snd_wscale) |
| 5342 | __tcp_fast_path_on(tp, tp->snd_wnd); |
| 5343 | else |
| 5344 | tp->pred_flags = 0; |
| 5345 | |
| 5346 | if (!sock_flag(sk, SOCK_DEAD)) { |
| 5347 | sk->sk_state_change(sk); |
| 5348 | sk_wake_async(sk, SOCK_WAKE_IO, POLL_OUT); |
| 5349 | } |
| 5350 | } |
| 5351 | |
Yuchung Cheng | 8e4178c | 2012-07-19 06:43:08 +0000 | [diff] [blame] | 5352 | static bool tcp_rcv_fastopen_synack(struct sock *sk, struct sk_buff *synack, |
| 5353 | struct tcp_fastopen_cookie *cookie) |
| 5354 | { |
| 5355 | struct tcp_sock *tp = tcp_sk(sk); |
Yuchung Cheng | 67da22d | 2012-07-19 06:43:11 +0000 | [diff] [blame] | 5356 | struct sk_buff *data = tp->syn_data ? tcp_write_queue_head(sk) : NULL; |
Yuchung Cheng | 8e4178c | 2012-07-19 06:43:08 +0000 | [diff] [blame] | 5357 | u16 mss = tp->rx_opt.mss_clamp; |
Yuchung Cheng | aab4874 | 2012-07-19 06:43:10 +0000 | [diff] [blame] | 5358 | bool syn_drop; |
Yuchung Cheng | 8e4178c | 2012-07-19 06:43:08 +0000 | [diff] [blame] | 5359 | |
| 5360 | if (mss == tp->rx_opt.user_mss) { |
| 5361 | struct tcp_options_received opt; |
Yuchung Cheng | 8e4178c | 2012-07-19 06:43:08 +0000 | [diff] [blame] | 5362 | |
| 5363 | /* Get original SYNACK MSS value if user MSS sets mss_clamp */ |
| 5364 | tcp_clear_options(&opt); |
| 5365 | opt.user_mss = opt.mss_clamp = 0; |
Christoph Paasch | 1a2c618 | 2013-03-17 08:23:34 +0000 | [diff] [blame] | 5366 | tcp_parse_options(synack, &opt, 0, NULL); |
Yuchung Cheng | 8e4178c | 2012-07-19 06:43:08 +0000 | [diff] [blame] | 5367 | mss = opt.mss_clamp; |
| 5368 | } |
| 5369 | |
Yuchung Cheng | 67da22d | 2012-07-19 06:43:11 +0000 | [diff] [blame] | 5370 | if (!tp->syn_fastopen) /* Ignore an unsolicited cookie */ |
| 5371 | cookie->len = -1; |
| 5372 | |
Yuchung Cheng | aab4874 | 2012-07-19 06:43:10 +0000 | [diff] [blame] | 5373 | /* The SYN-ACK neither has cookie nor acknowledges the data. Presumably |
| 5374 | * the remote receives only the retransmitted (regular) SYNs: either |
| 5375 | * the original SYN-data or the corresponding SYN-ACK is lost. |
| 5376 | */ |
Yuchung Cheng | 66555e9 | 2013-01-31 11:16:46 -0800 | [diff] [blame] | 5377 | syn_drop = (cookie->len <= 0 && data && tp->total_retrans); |
Yuchung Cheng | aab4874 | 2012-07-19 06:43:10 +0000 | [diff] [blame] | 5378 | |
| 5379 | tcp_fastopen_cache_set(sk, mss, cookie, syn_drop); |
Yuchung Cheng | 8e4178c | 2012-07-19 06:43:08 +0000 | [diff] [blame] | 5380 | |
| 5381 | if (data) { /* Retransmit unacked data in SYN */ |
Yuchung Cheng | 93b174a | 2012-12-06 08:45:32 +0000 | [diff] [blame] | 5382 | tcp_for_write_queue_from(data, sk) { |
| 5383 | if (data == tcp_send_head(sk) || |
| 5384 | __tcp_retransmit_skb(sk, data)) |
| 5385 | break; |
| 5386 | } |
Yuchung Cheng | 8e4178c | 2012-07-19 06:43:08 +0000 | [diff] [blame] | 5387 | tcp_rearm_rto(sk); |
| 5388 | return true; |
| 5389 | } |
Yuchung Cheng | 6f73601 | 2012-10-19 15:14:44 +0000 | [diff] [blame] | 5390 | tp->syn_data_acked = tp->syn_data; |
Yuchung Cheng | 8e4178c | 2012-07-19 06:43:08 +0000 | [diff] [blame] | 5391 | return false; |
| 5392 | } |
| 5393 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5394 | static int tcp_rcv_synsent_state_process(struct sock *sk, struct sk_buff *skb, |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 5395 | const struct tcphdr *th, unsigned int len) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5396 | { |
Arnaldo Carvalho de Melo | d83d846 | 2005-12-13 23:26:10 -0800 | [diff] [blame] | 5397 | struct inet_connection_sock *icsk = inet_csk(sk); |
William Allen Simpson | 4957faade | 2009-12-02 18:25:27 +0000 | [diff] [blame] | 5398 | struct tcp_sock *tp = tcp_sk(sk); |
Yuchung Cheng | 8e4178c | 2012-07-19 06:43:08 +0000 | [diff] [blame] | 5399 | struct tcp_fastopen_cookie foc = { .len = -1 }; |
William Allen Simpson | 4957faade | 2009-12-02 18:25:27 +0000 | [diff] [blame] | 5400 | int saved_clamp = tp->rx_opt.mss_clamp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5401 | |
Christoph Paasch | 1a2c618 | 2013-03-17 08:23:34 +0000 | [diff] [blame] | 5402 | tcp_parse_options(skb, &tp->rx_opt, 0, &foc); |
Andrew Vagin | e3e1202 | 2013-08-27 12:21:55 +0400 | [diff] [blame] | 5403 | if (tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr) |
Andrey Vagin | ee684b6 | 2013-02-11 05:50:19 +0000 | [diff] [blame] | 5404 | tp->rx_opt.rcv_tsecr -= tp->tsoffset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5405 | |
| 5406 | if (th->ack) { |
| 5407 | /* rfc793: |
| 5408 | * "If the state is SYN-SENT then |
| 5409 | * first check the ACK bit |
| 5410 | * If the ACK bit is set |
| 5411 | * If SEG.ACK =< ISS, or SEG.ACK > SND.NXT, send |
| 5412 | * a reset (unless the RST bit is set, if so drop |
| 5413 | * the segment and return)" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5414 | */ |
Yuchung Cheng | 8e4178c | 2012-07-19 06:43:08 +0000 | [diff] [blame] | 5415 | if (!after(TCP_SKB_CB(skb)->ack_seq, tp->snd_una) || |
| 5416 | after(TCP_SKB_CB(skb)->ack_seq, tp->snd_nxt)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5417 | goto reset_and_undo; |
| 5418 | |
| 5419 | if (tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr && |
| 5420 | !between(tp->rx_opt.rcv_tsecr, tp->retrans_stamp, |
| 5421 | tcp_time_stamp)) { |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 5422 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSACTIVEREJECTED); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5423 | goto reset_and_undo; |
| 5424 | } |
| 5425 | |
| 5426 | /* Now ACK is acceptable. |
| 5427 | * |
| 5428 | * "If the RST bit is set |
| 5429 | * If the ACK was acceptable then signal the user "error: |
| 5430 | * connection reset", drop the segment, enter CLOSED state, |
| 5431 | * delete TCB, and return." |
| 5432 | */ |
| 5433 | |
| 5434 | if (th->rst) { |
| 5435 | tcp_reset(sk); |
| 5436 | goto discard; |
| 5437 | } |
| 5438 | |
| 5439 | /* rfc793: |
| 5440 | * "fifth, if neither of the SYN or RST bits is set then |
| 5441 | * drop the segment and return." |
| 5442 | * |
| 5443 | * See note below! |
| 5444 | * --ANK(990513) |
| 5445 | */ |
| 5446 | if (!th->syn) |
| 5447 | goto discard_and_undo; |
| 5448 | |
| 5449 | /* rfc793: |
| 5450 | * "If the SYN bit is on ... |
| 5451 | * are acceptable then ... |
| 5452 | * (our SYN has been ACKed), change the connection |
| 5453 | * state to ESTABLISHED..." |
| 5454 | */ |
| 5455 | |
| 5456 | TCP_ECN_rcv_synack(tp, th); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5457 | |
Razvan Ghitulete | 1b3a692 | 2012-08-14 16:30:20 +0300 | [diff] [blame] | 5458 | tcp_init_wl(tp, TCP_SKB_CB(skb)->seq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5459 | tcp_ack(sk, skb, FLAG_SLOWPATH); |
| 5460 | |
| 5461 | /* Ok.. it's good. Set up sequence numbers and |
| 5462 | * move to established. |
| 5463 | */ |
| 5464 | tp->rcv_nxt = TCP_SKB_CB(skb)->seq + 1; |
| 5465 | tp->rcv_wup = TCP_SKB_CB(skb)->seq + 1; |
| 5466 | |
| 5467 | /* RFC1323: The window in SYN & SYN/ACK segments is |
| 5468 | * never scaled. |
| 5469 | */ |
| 5470 | tp->snd_wnd = ntohs(th->window); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5471 | |
| 5472 | if (!tp->rx_opt.wscale_ok) { |
| 5473 | tp->rx_opt.snd_wscale = tp->rx_opt.rcv_wscale = 0; |
| 5474 | tp->window_clamp = min(tp->window_clamp, 65535U); |
| 5475 | } |
| 5476 | |
| 5477 | if (tp->rx_opt.saw_tstamp) { |
| 5478 | tp->rx_opt.tstamp_ok = 1; |
| 5479 | tp->tcp_header_len = |
| 5480 | sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED; |
| 5481 | tp->advmss -= TCPOLEN_TSTAMP_ALIGNED; |
| 5482 | tcp_store_ts_recent(tp); |
| 5483 | } else { |
| 5484 | tp->tcp_header_len = sizeof(struct tcphdr); |
| 5485 | } |
| 5486 | |
Ilpo Järvinen | e60402d | 2007-08-09 15:14:46 +0300 | [diff] [blame] | 5487 | if (tcp_is_sack(tp) && sysctl_tcp_fack) |
| 5488 | tcp_enable_fack(tp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5489 | |
John Heffner | 5d424d5 | 2006-03-20 17:53:41 -0800 | [diff] [blame] | 5490 | tcp_mtup_init(sk); |
Arnaldo Carvalho de Melo | d83d846 | 2005-12-13 23:26:10 -0800 | [diff] [blame] | 5491 | tcp_sync_mss(sk, icsk->icsk_pmtu_cookie); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5492 | tcp_initialize_rcv_mss(sk); |
| 5493 | |
| 5494 | /* Remember, tcp_poll() does not lock socket! |
| 5495 | * Change state from SYN-SENT only after copied_seq |
| 5496 | * is initialized. */ |
| 5497 | tp->copied_seq = tp->rcv_nxt; |
William Allen Simpson | 4957faade | 2009-12-02 18:25:27 +0000 | [diff] [blame] | 5498 | |
Ralf Baechle | e16aa20 | 2006-12-07 00:11:33 -0800 | [diff] [blame] | 5499 | smp_mb(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5500 | |
Pavel Emelyanov | 370816a | 2012-04-19 03:40:01 +0000 | [diff] [blame] | 5501 | tcp_finish_connect(sk, skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5502 | |
Yuchung Cheng | 67da22d | 2012-07-19 06:43:11 +0000 | [diff] [blame] | 5503 | if ((tp->syn_fastopen || tp->syn_data) && |
| 5504 | tcp_rcv_fastopen_synack(sk, skb, &foc)) |
Yuchung Cheng | 8e4178c | 2012-07-19 06:43:08 +0000 | [diff] [blame] | 5505 | return -1; |
| 5506 | |
Arnaldo Carvalho de Melo | 295f732 | 2005-08-09 20:11:56 -0700 | [diff] [blame] | 5507 | if (sk->sk_write_pending || |
| 5508 | icsk->icsk_accept_queue.rskq_defer_accept || |
| 5509 | icsk->icsk_ack.pingpong) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5510 | /* Save one ACK. Data will be ready after |
| 5511 | * several ticks, if write_pending is set. |
| 5512 | * |
| 5513 | * It may be deleted, but with this feature tcpdumps |
| 5514 | * look so _wonderfully_ clever, that I was not able |
| 5515 | * to stand against the temptation 8) --ANK |
| 5516 | */ |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 5517 | inet_csk_schedule_ack(sk); |
Arnaldo Carvalho de Melo | 295f732 | 2005-08-09 20:11:56 -0700 | [diff] [blame] | 5518 | icsk->icsk_ack.lrcvtime = tcp_time_stamp; |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 5519 | tcp_enter_quickack_mode(sk); |
Arnaldo Carvalho de Melo | 3f421ba | 2005-08-09 20:11:08 -0700 | [diff] [blame] | 5520 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, |
| 5521 | TCP_DELACK_MAX, TCP_RTO_MAX); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5522 | |
| 5523 | discard: |
| 5524 | __kfree_skb(skb); |
| 5525 | return 0; |
| 5526 | } else { |
| 5527 | tcp_send_ack(sk); |
| 5528 | } |
| 5529 | return -1; |
| 5530 | } |
| 5531 | |
| 5532 | /* No ACK in the segment */ |
| 5533 | |
| 5534 | if (th->rst) { |
| 5535 | /* rfc793: |
| 5536 | * "If the RST bit is set |
| 5537 | * |
| 5538 | * Otherwise (no ACK) drop the segment and return." |
| 5539 | */ |
| 5540 | |
| 5541 | goto discard_and_undo; |
| 5542 | } |
| 5543 | |
| 5544 | /* PAWS check. */ |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 5545 | if (tp->rx_opt.ts_recent_stamp && tp->rx_opt.saw_tstamp && |
Ilpo Järvinen | c887e6d | 2009-03-14 14:23:03 +0000 | [diff] [blame] | 5546 | tcp_paws_reject(&tp->rx_opt, 0)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5547 | goto discard_and_undo; |
| 5548 | |
| 5549 | if (th->syn) { |
| 5550 | /* We see SYN without ACK. It is attempt of |
| 5551 | * simultaneous connect with crossed SYNs. |
| 5552 | * Particularly, it can be connect to self. |
| 5553 | */ |
| 5554 | tcp_set_state(sk, TCP_SYN_RECV); |
| 5555 | |
| 5556 | if (tp->rx_opt.saw_tstamp) { |
| 5557 | tp->rx_opt.tstamp_ok = 1; |
| 5558 | tcp_store_ts_recent(tp); |
| 5559 | tp->tcp_header_len = |
| 5560 | sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED; |
| 5561 | } else { |
| 5562 | tp->tcp_header_len = sizeof(struct tcphdr); |
| 5563 | } |
| 5564 | |
| 5565 | tp->rcv_nxt = TCP_SKB_CB(skb)->seq + 1; |
| 5566 | tp->rcv_wup = TCP_SKB_CB(skb)->seq + 1; |
| 5567 | |
| 5568 | /* RFC1323: The window in SYN & SYN/ACK segments is |
| 5569 | * never scaled. |
| 5570 | */ |
| 5571 | tp->snd_wnd = ntohs(th->window); |
| 5572 | tp->snd_wl1 = TCP_SKB_CB(skb)->seq; |
| 5573 | tp->max_window = tp->snd_wnd; |
| 5574 | |
| 5575 | TCP_ECN_rcv_syn(tp, th); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5576 | |
John Heffner | 5d424d5 | 2006-03-20 17:53:41 -0800 | [diff] [blame] | 5577 | tcp_mtup_init(sk); |
Arnaldo Carvalho de Melo | d83d846 | 2005-12-13 23:26:10 -0800 | [diff] [blame] | 5578 | tcp_sync_mss(sk, icsk->icsk_pmtu_cookie); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5579 | tcp_initialize_rcv_mss(sk); |
| 5580 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5581 | tcp_send_synack(sk); |
| 5582 | #if 0 |
| 5583 | /* Note, we could accept data and URG from this segment. |
Jerry Chu | 168a8f5 | 2012-08-31 12:29:13 +0000 | [diff] [blame] | 5584 | * There are no obstacles to make this (except that we must |
| 5585 | * either change tcp_recvmsg() to prevent it from returning data |
| 5586 | * before 3WHS completes per RFC793, or employ TCP Fast Open). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5587 | * |
| 5588 | * However, if we ignore data in ACKless segments sometimes, |
| 5589 | * we have no reasons to accept it sometimes. |
| 5590 | * Also, seems the code doing it in step6 of tcp_rcv_state_process |
| 5591 | * is not flawless. So, discard packet for sanity. |
| 5592 | * Uncomment this return to process the data. |
| 5593 | */ |
| 5594 | return -1; |
| 5595 | #else |
| 5596 | goto discard; |
| 5597 | #endif |
| 5598 | } |
| 5599 | /* "fifth, if neither of the SYN or RST bits is set then |
| 5600 | * drop the segment and return." |
| 5601 | */ |
| 5602 | |
| 5603 | discard_and_undo: |
| 5604 | tcp_clear_options(&tp->rx_opt); |
| 5605 | tp->rx_opt.mss_clamp = saved_clamp; |
| 5606 | goto discard; |
| 5607 | |
| 5608 | reset_and_undo: |
| 5609 | tcp_clear_options(&tp->rx_opt); |
| 5610 | tp->rx_opt.mss_clamp = saved_clamp; |
| 5611 | return 1; |
| 5612 | } |
| 5613 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5614 | /* |
| 5615 | * This function implements the receiving procedure of RFC 793 for |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5616 | * all states except ESTABLISHED and TIME_WAIT. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5617 | * It's called from both tcp_v4_rcv and tcp_v6_rcv and should be |
| 5618 | * address independent. |
| 5619 | */ |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5620 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5621 | int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb, |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 5622 | const struct tcphdr *th, unsigned int len) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5623 | { |
| 5624 | struct tcp_sock *tp = tcp_sk(sk); |
Arnaldo Carvalho de Melo | 8292a17 | 2005-12-13 23:15:52 -0800 | [diff] [blame] | 5625 | struct inet_connection_sock *icsk = inet_csk(sk); |
Jerry Chu | 168a8f5 | 2012-08-31 12:29:13 +0000 | [diff] [blame] | 5626 | struct request_sock *req; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5627 | int queued = 0; |
Eric Dumazet | 1f6afc8 | 2013-05-24 15:03:54 +0000 | [diff] [blame] | 5628 | bool acceptable; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5629 | |
| 5630 | tp->rx_opt.saw_tstamp = 0; |
| 5631 | |
| 5632 | switch (sk->sk_state) { |
| 5633 | case TCP_CLOSE: |
| 5634 | goto discard; |
| 5635 | |
| 5636 | case TCP_LISTEN: |
Stephen Hemminger | 2de979b | 2007-03-08 20:45:19 -0800 | [diff] [blame] | 5637 | if (th->ack) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5638 | return 1; |
| 5639 | |
Stephen Hemminger | 2de979b | 2007-03-08 20:45:19 -0800 | [diff] [blame] | 5640 | if (th->rst) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5641 | goto discard; |
| 5642 | |
Stephen Hemminger | 2de979b | 2007-03-08 20:45:19 -0800 | [diff] [blame] | 5643 | if (th->syn) { |
Eric Dumazet | fdf5af0 | 2011-12-02 23:41:42 +0000 | [diff] [blame] | 5644 | if (th->fin) |
| 5645 | goto discard; |
Arnaldo Carvalho de Melo | 8292a17 | 2005-12-13 23:15:52 -0800 | [diff] [blame] | 5646 | if (icsk->icsk_af_ops->conn_request(sk, skb) < 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5647 | return 1; |
| 5648 | |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5649 | /* Now we have several options: In theory there is |
| 5650 | * nothing else in the frame. KA9Q has an option to |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5651 | * send data with the syn, BSD accepts data with the |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5652 | * syn up to the [to be] advertised window and |
| 5653 | * Solaris 2.1 gives you a protocol error. For now |
| 5654 | * we just ignore it, that fits the spec precisely |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5655 | * and avoids incompatibilities. It would be nice in |
| 5656 | * future to drop through and process the data. |
| 5657 | * |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5658 | * Now that TTCP is starting to be used we ought to |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5659 | * queue this data. |
| 5660 | * But, this leaves one open to an easy denial of |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5661 | * service attack, and SYN cookies can't defend |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5662 | * against this problem. So, we drop the data |
Masayuki Nakagawa | fb7e239 | 2007-01-23 20:15:06 -0800 | [diff] [blame] | 5663 | * in the interest of security over speed unless |
| 5664 | * it's still in use. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5665 | */ |
Masayuki Nakagawa | fb7e239 | 2007-01-23 20:15:06 -0800 | [diff] [blame] | 5666 | kfree_skb(skb); |
| 5667 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5668 | } |
| 5669 | goto discard; |
| 5670 | |
| 5671 | case TCP_SYN_SENT: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5672 | queued = tcp_rcv_synsent_state_process(sk, skb, th, len); |
| 5673 | if (queued >= 0) |
| 5674 | return queued; |
| 5675 | |
| 5676 | /* Do step6 onward by hand. */ |
| 5677 | tcp_urg(sk, skb, th); |
| 5678 | __kfree_skb(skb); |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 5679 | tcp_data_snd_check(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5680 | return 0; |
| 5681 | } |
| 5682 | |
Jerry Chu | 168a8f5 | 2012-08-31 12:29:13 +0000 | [diff] [blame] | 5683 | req = tp->fastopen_rsk; |
| 5684 | if (req != NULL) { |
Jerry Chu | 37561f6 | 2012-10-22 11:26:36 +0000 | [diff] [blame] | 5685 | WARN_ON_ONCE(sk->sk_state != TCP_SYN_RECV && |
Jerry Chu | 168a8f5 | 2012-08-31 12:29:13 +0000 | [diff] [blame] | 5686 | sk->sk_state != TCP_FIN_WAIT1); |
| 5687 | |
| 5688 | if (tcp_check_req(sk, skb, req, NULL, true) == NULL) |
| 5689 | goto discard; |
Neal Cardwell | e69bebd | 2012-09-22 04:18:57 +0000 | [diff] [blame] | 5690 | } |
Eric Dumazet | c3ae62af | 2012-12-26 12:44:34 +0000 | [diff] [blame] | 5691 | |
Eric Dumazet | 7b514a8 | 2013-01-10 16:18:47 +0000 | [diff] [blame] | 5692 | if (!th->ack && !th->rst) |
Eric Dumazet | c3ae62af | 2012-12-26 12:44:34 +0000 | [diff] [blame] | 5693 | goto discard; |
| 5694 | |
Neal Cardwell | e69bebd | 2012-09-22 04:18:57 +0000 | [diff] [blame] | 5695 | if (!tcp_validate_incoming(sk, skb, th, 0)) |
Eric Dumazet | 0c24604 | 2012-07-17 01:41:30 +0000 | [diff] [blame] | 5696 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5697 | |
| 5698 | /* step 5: check the ACK field */ |
Eric Dumazet | 1f6afc8 | 2013-05-24 15:03:54 +0000 | [diff] [blame] | 5699 | acceptable = tcp_ack(sk, skb, FLAG_SLOWPATH | |
| 5700 | FLAG_UPDATE_TS_RECENT) > 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5701 | |
Eric Dumazet | 1f6afc8 | 2013-05-24 15:03:54 +0000 | [diff] [blame] | 5702 | switch (sk->sk_state) { |
| 5703 | case TCP_SYN_RECV: |
Joe Perches | 61eb9003 | 2013-05-24 18:36:13 +0000 | [diff] [blame] | 5704 | if (!acceptable) |
Eric Dumazet | 1f6afc8 | 2013-05-24 15:03:54 +0000 | [diff] [blame] | 5705 | return 1; |
Joe Perches | 61eb9003 | 2013-05-24 18:36:13 +0000 | [diff] [blame] | 5706 | |
| 5707 | /* Once we leave TCP_SYN_RECV, we no longer need req |
| 5708 | * so release it. |
| 5709 | */ |
| 5710 | if (req) { |
Joe Perches | 61eb9003 | 2013-05-24 18:36:13 +0000 | [diff] [blame] | 5711 | tp->total_retrans = req->num_retrans; |
Joe Perches | 61eb9003 | 2013-05-24 18:36:13 +0000 | [diff] [blame] | 5712 | reqsk_fastopen_remove(sk, req, false); |
| 5713 | } else { |
| 5714 | /* Make sure socket is routed, for correct metrics. */ |
| 5715 | icsk->icsk_af_ops->rebuild_header(sk); |
| 5716 | tcp_init_congestion_control(sk); |
| 5717 | |
| 5718 | tcp_mtup_init(sk); |
Joe Perches | 61eb9003 | 2013-05-24 18:36:13 +0000 | [diff] [blame] | 5719 | tp->copied_seq = tp->rcv_nxt; |
Eric Dumazet | b0983d3 | 2013-09-20 13:56:58 -0700 | [diff] [blame] | 5720 | tcp_init_buffer_space(sk); |
Eric Dumazet | 1f6afc8 | 2013-05-24 15:03:54 +0000 | [diff] [blame] | 5721 | } |
Joe Perches | 61eb9003 | 2013-05-24 18:36:13 +0000 | [diff] [blame] | 5722 | smp_mb(); |
| 5723 | tcp_set_state(sk, TCP_ESTABLISHED); |
| 5724 | sk->sk_state_change(sk); |
| 5725 | |
| 5726 | /* Note, that this wakeup is only for marginal crossed SYN case. |
| 5727 | * Passively open sockets are not waked up, because |
| 5728 | * sk->sk_sleep == NULL and sk->sk_socket == NULL. |
| 5729 | */ |
| 5730 | if (sk->sk_socket) |
| 5731 | sk_wake_async(sk, SOCK_WAKE_IO, POLL_OUT); |
| 5732 | |
| 5733 | tp->snd_una = TCP_SKB_CB(skb)->ack_seq; |
| 5734 | tp->snd_wnd = ntohs(th->window) << tp->rx_opt.snd_wscale; |
| 5735 | tcp_init_wl(tp, TCP_SKB_CB(skb)->seq); |
Yuchung Cheng | 375fe02 | 2013-07-22 16:20:45 -0700 | [diff] [blame] | 5736 | tcp_synack_rtt_meas(sk, req); |
Joe Perches | 61eb9003 | 2013-05-24 18:36:13 +0000 | [diff] [blame] | 5737 | |
| 5738 | if (tp->rx_opt.tstamp_ok) |
| 5739 | tp->advmss -= TCPOLEN_TSTAMP_ALIGNED; |
| 5740 | |
| 5741 | if (req) { |
| 5742 | /* Re-arm the timer because data may have been sent out. |
| 5743 | * This is similar to the regular data transmission case |
| 5744 | * when new data has just been ack'ed. |
| 5745 | * |
| 5746 | * (TFO) - we could try to be more aggressive and |
| 5747 | * retransmitting any data sooner based on when they |
| 5748 | * are sent out. |
| 5749 | */ |
| 5750 | tcp_rearm_rto(sk); |
| 5751 | } else |
| 5752 | tcp_init_metrics(sk); |
| 5753 | |
Neal Cardwell | 02cf4eb | 2013-10-21 15:40:19 -0400 | [diff] [blame] | 5754 | tcp_update_pacing_rate(sk); |
| 5755 | |
Joe Perches | 61eb9003 | 2013-05-24 18:36:13 +0000 | [diff] [blame] | 5756 | /* Prevent spurious tcp_cwnd_restart() on first data packet */ |
| 5757 | tp->lsndtime = tcp_time_stamp; |
| 5758 | |
| 5759 | tcp_initialize_rcv_mss(sk); |
| 5760 | tcp_fast_path_on(tp); |
Eric Dumazet | 1f6afc8 | 2013-05-24 15:03:54 +0000 | [diff] [blame] | 5761 | break; |
| 5762 | |
Joe Perches | c48b22d | 2013-05-24 18:06:58 +0000 | [diff] [blame] | 5763 | case TCP_FIN_WAIT1: { |
| 5764 | struct dst_entry *dst; |
| 5765 | int tmo; |
| 5766 | |
Eric Dumazet | 1f6afc8 | 2013-05-24 15:03:54 +0000 | [diff] [blame] | 5767 | /* If we enter the TCP_FIN_WAIT1 state and we are a |
| 5768 | * Fast Open socket and this is the first acceptable |
| 5769 | * ACK we have received, this would have acknowledged |
| 5770 | * our SYNACK so stop the SYNACK timer. |
| 5771 | */ |
| 5772 | if (req != NULL) { |
| 5773 | /* Return RST if ack_seq is invalid. |
| 5774 | * Note that RFC793 only says to generate a |
| 5775 | * DUPACK for it but for TCP Fast Open it seems |
| 5776 | * better to treat this case like TCP_SYN_RECV |
| 5777 | * above. |
| 5778 | */ |
| 5779 | if (!acceptable) |
| 5780 | return 1; |
| 5781 | /* We no longer need the request sock. */ |
| 5782 | reqsk_fastopen_remove(sk, req, false); |
| 5783 | tcp_rearm_rto(sk); |
| 5784 | } |
Joe Perches | c48b22d | 2013-05-24 18:06:58 +0000 | [diff] [blame] | 5785 | if (tp->snd_una != tp->write_seq) |
| 5786 | break; |
Eric Dumazet | 1f6afc8 | 2013-05-24 15:03:54 +0000 | [diff] [blame] | 5787 | |
Joe Perches | c48b22d | 2013-05-24 18:06:58 +0000 | [diff] [blame] | 5788 | tcp_set_state(sk, TCP_FIN_WAIT2); |
| 5789 | sk->sk_shutdown |= SEND_SHUTDOWN; |
Eric Dumazet | 1f6afc8 | 2013-05-24 15:03:54 +0000 | [diff] [blame] | 5790 | |
Joe Perches | c48b22d | 2013-05-24 18:06:58 +0000 | [diff] [blame] | 5791 | dst = __sk_dst_get(sk); |
| 5792 | if (dst) |
| 5793 | dst_confirm(dst); |
Eric Dumazet | 1f6afc8 | 2013-05-24 15:03:54 +0000 | [diff] [blame] | 5794 | |
Joe Perches | c48b22d | 2013-05-24 18:06:58 +0000 | [diff] [blame] | 5795 | if (!sock_flag(sk, SOCK_DEAD)) { |
| 5796 | /* Wake up lingering close() */ |
| 5797 | sk->sk_state_change(sk); |
| 5798 | break; |
| 5799 | } |
Eric Dumazet | 1f6afc8 | 2013-05-24 15:03:54 +0000 | [diff] [blame] | 5800 | |
Joe Perches | c48b22d | 2013-05-24 18:06:58 +0000 | [diff] [blame] | 5801 | if (tp->linger2 < 0 || |
| 5802 | (TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(skb)->seq && |
| 5803 | after(TCP_SKB_CB(skb)->end_seq - th->fin, tp->rcv_nxt))) { |
| 5804 | tcp_done(sk); |
| 5805 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPABORTONDATA); |
| 5806 | return 1; |
| 5807 | } |
Eric Dumazet | 1f6afc8 | 2013-05-24 15:03:54 +0000 | [diff] [blame] | 5808 | |
Joe Perches | c48b22d | 2013-05-24 18:06:58 +0000 | [diff] [blame] | 5809 | tmo = tcp_fin_time(sk); |
| 5810 | if (tmo > TCP_TIMEWAIT_LEN) { |
| 5811 | inet_csk_reset_keepalive_timer(sk, tmo - TCP_TIMEWAIT_LEN); |
| 5812 | } else if (th->fin || sock_owned_by_user(sk)) { |
| 5813 | /* Bad case. We could lose such FIN otherwise. |
| 5814 | * It is not a big problem, but it looks confusing |
| 5815 | * and not so rare event. We still can lose it now, |
| 5816 | * if it spins in bh_lock_sock(), but it is really |
| 5817 | * marginal case. |
| 5818 | */ |
| 5819 | inet_csk_reset_keepalive_timer(sk, tmo); |
| 5820 | } else { |
| 5821 | tcp_time_wait(sk, TCP_FIN_WAIT2, tmo); |
| 5822 | goto discard; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5823 | } |
Eric Dumazet | 1f6afc8 | 2013-05-24 15:03:54 +0000 | [diff] [blame] | 5824 | break; |
Joe Perches | c48b22d | 2013-05-24 18:06:58 +0000 | [diff] [blame] | 5825 | } |
Eric Dumazet | 1f6afc8 | 2013-05-24 15:03:54 +0000 | [diff] [blame] | 5826 | |
| 5827 | case TCP_CLOSING: |
| 5828 | if (tp->snd_una == tp->write_seq) { |
| 5829 | tcp_time_wait(sk, TCP_TIME_WAIT, 0); |
| 5830 | goto discard; |
| 5831 | } |
| 5832 | break; |
| 5833 | |
| 5834 | case TCP_LAST_ACK: |
| 5835 | if (tp->snd_una == tp->write_seq) { |
| 5836 | tcp_update_metrics(sk); |
| 5837 | tcp_done(sk); |
| 5838 | goto discard; |
| 5839 | } |
| 5840 | break; |
Eric Dumazet | c3ae62af | 2012-12-26 12:44:34 +0000 | [diff] [blame] | 5841 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5842 | |
| 5843 | /* step 6: check the URG bit */ |
| 5844 | tcp_urg(sk, skb, th); |
| 5845 | |
| 5846 | /* step 7: process the segment text */ |
| 5847 | switch (sk->sk_state) { |
| 5848 | case TCP_CLOSE_WAIT: |
| 5849 | case TCP_CLOSING: |
| 5850 | case TCP_LAST_ACK: |
| 5851 | if (!before(TCP_SKB_CB(skb)->seq, tp->rcv_nxt)) |
| 5852 | break; |
| 5853 | case TCP_FIN_WAIT1: |
| 5854 | case TCP_FIN_WAIT2: |
| 5855 | /* RFC 793 says to queue data in these states, |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5856 | * RFC 1122 says we MUST send a reset. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5857 | * BSD 4.4 also does reset. |
| 5858 | */ |
| 5859 | if (sk->sk_shutdown & RCV_SHUTDOWN) { |
| 5860 | if (TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(skb)->seq && |
| 5861 | after(TCP_SKB_CB(skb)->end_seq - th->fin, tp->rcv_nxt)) { |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 5862 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPABORTONDATA); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5863 | tcp_reset(sk); |
| 5864 | return 1; |
| 5865 | } |
| 5866 | } |
| 5867 | /* Fall through */ |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5868 | case TCP_ESTABLISHED: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5869 | tcp_data_queue(sk, skb); |
| 5870 | queued = 1; |
| 5871 | break; |
| 5872 | } |
| 5873 | |
| 5874 | /* tcp_data could move socket to TIME-WAIT */ |
| 5875 | if (sk->sk_state != TCP_CLOSE) { |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 5876 | tcp_data_snd_check(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5877 | tcp_ack_snd_check(sk); |
| 5878 | } |
| 5879 | |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5880 | if (!queued) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5881 | discard: |
| 5882 | __kfree_skb(skb); |
| 5883 | } |
| 5884 | return 0; |
| 5885 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5886 | EXPORT_SYMBOL(tcp_rcv_state_process); |