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> |
Eric Dumazet | ad971f6 | 2014-10-11 15:17:29 -0700 | [diff] [blame] | 71 | #include <linux/prefetch.h> |
Satoru SATOH | 5ffc02a | 2008-05-04 22:14:42 -0700 | [diff] [blame] | 72 | #include <net/dst.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 73 | #include <net/tcp.h> |
| 74 | #include <net/inet_common.h> |
| 75 | #include <linux/ipsec.h> |
| 76 | #include <asm/unaligned.h> |
Willem de Bruijn | e1c8a60 | 2014-08-04 22:11:50 -0400 | [diff] [blame] | 77 | #include <linux/errqueue.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 78 | |
Brian Haley | ab32ea5 | 2006-09-22 14:15:41 -0700 | [diff] [blame] | 79 | int sysctl_tcp_timestamps __read_mostly = 1; |
| 80 | int sysctl_tcp_window_scaling __read_mostly = 1; |
| 81 | int sysctl_tcp_sack __read_mostly = 1; |
| 82 | int sysctl_tcp_fack __read_mostly = 1; |
| 83 | int sysctl_tcp_reordering __read_mostly = TCP_FASTRETRANS_THRESH; |
Eric Dumazet | dca145f | 2014-10-27 21:45:24 -0700 | [diff] [blame] | 84 | int sysctl_tcp_max_reordering __read_mostly = 300; |
Eric Dumazet | 4bc2f18 | 2010-07-09 21:22:10 +0000 | [diff] [blame] | 85 | EXPORT_SYMBOL(sysctl_tcp_reordering); |
Brian Haley | ab32ea5 | 2006-09-22 14:15:41 -0700 | [diff] [blame] | 86 | int sysctl_tcp_dsack __read_mostly = 1; |
| 87 | int sysctl_tcp_app_win __read_mostly = 31; |
Eric Dumazet | b49960a | 2012-05-02 02:28:41 +0000 | [diff] [blame] | 88 | int sysctl_tcp_adv_win_scale __read_mostly = 1; |
Eric Dumazet | 4bc2f18 | 2010-07-09 21:22:10 +0000 | [diff] [blame] | 89 | EXPORT_SYMBOL(sysctl_tcp_adv_win_scale); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 90 | |
Eric Dumazet | 282f23c | 2012-07-17 10:13:05 +0200 | [diff] [blame] | 91 | /* rfc5961 challenge ack rate limiting */ |
| 92 | int sysctl_tcp_challenge_ack_limit = 100; |
| 93 | |
Brian Haley | ab32ea5 | 2006-09-22 14:15:41 -0700 | [diff] [blame] | 94 | int sysctl_tcp_stdurg __read_mostly; |
| 95 | int sysctl_tcp_rfc1337 __read_mostly; |
| 96 | int sysctl_tcp_max_orphans __read_mostly = NR_FILE; |
Ilpo Järvinen | c96fd3d | 2007-09-20 11:36:37 -0700 | [diff] [blame] | 97 | int sysctl_tcp_frto __read_mostly = 2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 98 | |
Andreas Petlund | 7e38017 | 2010-02-18 04:48:19 +0000 | [diff] [blame] | 99 | int sysctl_tcp_thin_dupack __read_mostly; |
| 100 | |
Brian Haley | ab32ea5 | 2006-09-22 14:15:41 -0700 | [diff] [blame] | 101 | int sysctl_tcp_moderate_rcvbuf __read_mostly = 1; |
Nandita Dukkipati | 6ba8a3b | 2013-03-11 10:00:43 +0000 | [diff] [blame] | 102 | int sysctl_tcp_early_retrans __read_mostly = 3; |
Neal Cardwell | 032ee42 | 2015-02-06 16:04:38 -0500 | [diff] [blame] | 103 | int sysctl_tcp_invalid_ratelimit __read_mostly = HZ/2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 104 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 105 | #define FLAG_DATA 0x01 /* Incoming frame contained data. */ |
| 106 | #define FLAG_WIN_UPDATE 0x02 /* Incoming ACK was a window update. */ |
| 107 | #define FLAG_DATA_ACKED 0x04 /* This ACK acknowledged new data. */ |
| 108 | #define FLAG_RETRANS_DATA_ACKED 0x08 /* "" "" some of which was retransmitted. */ |
| 109 | #define FLAG_SYN_ACKED 0x10 /* This ACK acknowledged SYN. */ |
| 110 | #define FLAG_DATA_SACKED 0x20 /* New SACK. */ |
| 111 | #define FLAG_ECE 0x40 /* ECE in this ACK */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 112 | #define FLAG_SLOWPATH 0x100 /* Do not skip RFC checks for window update.*/ |
Yuchung Cheng | e33099f | 2013-03-20 13:33:00 +0000 | [diff] [blame] | 113 | #define FLAG_ORIG_SACK_ACKED 0x200 /* Never retransmitted data are (s)acked */ |
Ilpo Järvinen | 2e60529 | 2007-08-02 19:46:58 -0700 | [diff] [blame] | 114 | #define FLAG_SND_UNA_ADVANCED 0x400 /* Snd_una was changed (!= FLAG_DATA_ACKED) */ |
Ryousei Takano | 564262c | 2007-10-25 23:03:52 -0700 | [diff] [blame] | 115 | #define FLAG_DSACKING_ACK 0x800 /* SACK blocks contained D-SACK info */ |
Ilpo Järvinen | cadbd03 | 2007-12-31 04:49:21 -0800 | [diff] [blame] | 116 | #define FLAG_SACK_RENEGING 0x2000 /* snd_una advanced to a sacked seq */ |
Eric Dumazet | 12fb3dd | 2013-04-19 07:19:48 +0000 | [diff] [blame] | 117 | #define FLAG_UPDATE_TS_RECENT 0x4000 /* tcp_replace_ts_recent() */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 118 | |
| 119 | #define FLAG_ACKED (FLAG_DATA_ACKED|FLAG_SYN_ACKED) |
| 120 | #define FLAG_NOT_DUP (FLAG_DATA|FLAG_WIN_UPDATE|FLAG_ACKED) |
| 121 | #define FLAG_CA_ALERT (FLAG_DATA_SACKED|FLAG_ECE) |
| 122 | #define FLAG_FORWARD_PROGRESS (FLAG_ACKED|FLAG_DATA_SACKED) |
| 123 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 124 | #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] | 125 | #define TCP_HP_BITS (~(TCP_RESERVED_BITS|TCP_FLAG_PSH)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 126 | |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 127 | /* Adapt the MSS value used to make delayed ack decision to the |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 128 | * real world. |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 129 | */ |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 130 | 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] | 131 | { |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 132 | struct inet_connection_sock *icsk = inet_csk(sk); |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 133 | const unsigned int lss = icsk->icsk_ack.last_seg_size; |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 134 | unsigned int len; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 135 | |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 136 | icsk->icsk_ack.last_seg_size = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 137 | |
| 138 | /* skb->len may jitter because of SACKs, even if peer |
| 139 | * sends good full-sized frames. |
| 140 | */ |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 141 | len = skb_shinfo(skb)->gso_size ? : skb->len; |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 142 | if (len >= icsk->icsk_ack.rcv_mss) { |
| 143 | icsk->icsk_ack.rcv_mss = len; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 144 | } else { |
| 145 | /* Otherwise, we make more careful check taking into account, |
| 146 | * that SACKs block is variable. |
| 147 | * |
| 148 | * "len" is invariant segment length, including TCP header. |
| 149 | */ |
Arnaldo Carvalho de Melo | 9c70220 | 2007-04-25 18:04:18 -0700 | [diff] [blame] | 150 | len += skb->data - skb_transport_header(skb); |
William Allen Simpson | bee7ca9 | 2009-11-10 09:51:18 +0000 | [diff] [blame] | 151 | if (len >= TCP_MSS_DEFAULT + sizeof(struct tcphdr) || |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 152 | /* If PSH is not set, packet should be |
| 153 | * full sized, provided peer TCP is not badly broken. |
| 154 | * This observation (if it is correct 8)) allows |
| 155 | * to handle super-low mtu links fairly. |
| 156 | */ |
| 157 | (len >= TCP_MIN_MSS + sizeof(struct tcphdr) && |
Arnaldo Carvalho de Melo | aa8223c | 2007-04-10 21:04:22 -0700 | [diff] [blame] | 158 | !(tcp_flag_word(tcp_hdr(skb)) & TCP_REMNANT))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 159 | /* Subtract also invariant (if peer is RFC compliant), |
| 160 | * tcp header plus fixed timestamp option length. |
| 161 | * Resulting "len" is MSS free of SACK jitter. |
| 162 | */ |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 163 | len -= tcp_sk(sk)->tcp_header_len; |
| 164 | icsk->icsk_ack.last_seg_size = len; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 165 | if (len == lss) { |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 166 | icsk->icsk_ack.rcv_mss = len; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 167 | return; |
| 168 | } |
| 169 | } |
Alexey Kuznetsov | 1ef9696 | 2006-09-19 12:52:50 -0700 | [diff] [blame] | 170 | if (icsk->icsk_ack.pending & ICSK_ACK_PUSHED) |
| 171 | icsk->icsk_ack.pending |= ICSK_ACK_PUSHED2; |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 172 | icsk->icsk_ack.pending |= ICSK_ACK_PUSHED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 173 | } |
| 174 | } |
| 175 | |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 176 | static void tcp_incr_quickack(struct sock *sk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 177 | { |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 178 | struct inet_connection_sock *icsk = inet_csk(sk); |
Eric Dumazet | 95c9617 | 2012-04-15 05:58:06 +0000 | [diff] [blame] | 179 | 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] | 180 | |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 181 | if (quickacks == 0) |
| 182 | quickacks = 2; |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 183 | if (quickacks > icsk->icsk_ack.quick) |
| 184 | icsk->icsk_ack.quick = min(quickacks, TCP_MAX_QUICKACKS); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 185 | } |
| 186 | |
stephen hemminger | 1b9f409 | 2010-09-28 19:30:14 +0000 | [diff] [blame] | 187 | static void tcp_enter_quickack_mode(struct sock *sk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 188 | { |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 189 | struct inet_connection_sock *icsk = inet_csk(sk); |
| 190 | tcp_incr_quickack(sk); |
| 191 | icsk->icsk_ack.pingpong = 0; |
| 192 | icsk->icsk_ack.ato = TCP_ATO_MIN; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 193 | } |
| 194 | |
| 195 | /* Send ACKs quickly, if "quick" count is not exhausted |
| 196 | * and the session is not interactive. |
| 197 | */ |
| 198 | |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 199 | static inline bool tcp_in_quickack_mode(const struct sock *sk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 200 | { |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 201 | const struct inet_connection_sock *icsk = inet_csk(sk); |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 202 | |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 203 | return icsk->icsk_ack.quick && !icsk->icsk_ack.pingpong; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 204 | } |
| 205 | |
Florian Westphal | 735d383 | 2014-09-29 13:08:30 +0200 | [diff] [blame] | 206 | static void tcp_ecn_queue_cwr(struct tcp_sock *tp) |
Ilpo Järvinen | bdf1ee5 | 2007-05-27 02:04:16 -0700 | [diff] [blame] | 207 | { |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 208 | if (tp->ecn_flags & TCP_ECN_OK) |
Ilpo Järvinen | bdf1ee5 | 2007-05-27 02:04:16 -0700 | [diff] [blame] | 209 | tp->ecn_flags |= TCP_ECN_QUEUE_CWR; |
| 210 | } |
| 211 | |
Florian Westphal | 735d383 | 2014-09-29 13:08:30 +0200 | [diff] [blame] | 212 | static 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] | 213 | { |
| 214 | if (tcp_hdr(skb)->cwr) |
| 215 | tp->ecn_flags &= ~TCP_ECN_DEMAND_CWR; |
| 216 | } |
| 217 | |
Florian Westphal | 735d383 | 2014-09-29 13:08:30 +0200 | [diff] [blame] | 218 | static void tcp_ecn_withdraw_cwr(struct tcp_sock *tp) |
Ilpo Järvinen | bdf1ee5 | 2007-05-27 02:04:16 -0700 | [diff] [blame] | 219 | { |
| 220 | tp->ecn_flags &= ~TCP_ECN_DEMAND_CWR; |
| 221 | } |
| 222 | |
Florian Westphal | 735d383 | 2014-09-29 13:08:30 +0200 | [diff] [blame] | 223 | static 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] | 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: |
Florian Westphal | 9890092 | 2014-09-26 22:37:35 +0200 | [diff] [blame] | 235 | if (tcp_ca_needs_ecn((struct sock *)tp)) |
| 236 | tcp_ca_event((struct sock *)tp, CA_EVENT_ECN_IS_CE); |
| 237 | |
Eric Dumazet | aae06bf | 2012-08-06 11:04:43 +0000 | [diff] [blame] | 238 | if (!(tp->ecn_flags & TCP_ECN_DEMAND_CWR)) { |
| 239 | /* Better not delay acks, sender can have a very low cwnd */ |
| 240 | tcp_enter_quickack_mode((struct sock *)tp); |
| 241 | tp->ecn_flags |= TCP_ECN_DEMAND_CWR; |
| 242 | } |
Eric Dumazet | 7a269ff | 2011-09-22 20:02:19 +0000 | [diff] [blame] | 243 | tp->ecn_flags |= TCP_ECN_SEEN; |
Florian Westphal | 9890092 | 2014-09-26 22:37:35 +0200 | [diff] [blame] | 244 | break; |
| 245 | default: |
| 246 | if (tcp_ca_needs_ecn((struct sock *)tp)) |
| 247 | tcp_ca_event((struct sock *)tp, CA_EVENT_ECN_NO_CE); |
| 248 | tp->ecn_flags |= TCP_ECN_SEEN; |
| 249 | break; |
Ilpo Järvinen | bdf1ee5 | 2007-05-27 02:04:16 -0700 | [diff] [blame] | 250 | } |
| 251 | } |
| 252 | |
Florian Westphal | 735d383 | 2014-09-29 13:08:30 +0200 | [diff] [blame] | 253 | static void tcp_ecn_check_ce(struct tcp_sock *tp, const struct sk_buff *skb) |
| 254 | { |
| 255 | if (tp->ecn_flags & TCP_ECN_OK) |
| 256 | __tcp_ecn_check_ce(tp, skb); |
| 257 | } |
| 258 | |
| 259 | static 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] | 260 | { |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 261 | if ((tp->ecn_flags & TCP_ECN_OK) && (!th->ece || th->cwr)) |
Ilpo Järvinen | bdf1ee5 | 2007-05-27 02:04:16 -0700 | [diff] [blame] | 262 | tp->ecn_flags &= ~TCP_ECN_OK; |
| 263 | } |
| 264 | |
Florian Westphal | 735d383 | 2014-09-29 13:08:30 +0200 | [diff] [blame] | 265 | static 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] | 266 | { |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 267 | if ((tp->ecn_flags & TCP_ECN_OK) && (!th->ece || !th->cwr)) |
Ilpo Järvinen | bdf1ee5 | 2007-05-27 02:04:16 -0700 | [diff] [blame] | 268 | tp->ecn_flags &= ~TCP_ECN_OK; |
| 269 | } |
| 270 | |
Florian Westphal | 735d383 | 2014-09-29 13:08:30 +0200 | [diff] [blame] | 271 | 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] | 272 | { |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 273 | if (th->ece && !th->syn && (tp->ecn_flags & TCP_ECN_OK)) |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 274 | return true; |
| 275 | return false; |
Ilpo Järvinen | bdf1ee5 | 2007-05-27 02:04:16 -0700 | [diff] [blame] | 276 | } |
| 277 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 278 | /* Buffer size and advertised window tuning. |
| 279 | * |
| 280 | * 1. Tuning sk->sk_sndbuf, when connection enters established state. |
| 281 | */ |
| 282 | |
Eric Dumazet | 6ae7053 | 2013-10-01 10:23:44 -0700 | [diff] [blame] | 283 | static void tcp_sndbuf_expand(struct sock *sk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 284 | { |
Eric Dumazet | 6ae7053 | 2013-10-01 10:23:44 -0700 | [diff] [blame] | 285 | const struct tcp_sock *tp = tcp_sk(sk); |
| 286 | int sndmem, per_mss; |
| 287 | u32 nr_segs; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 288 | |
Eric Dumazet | 6ae7053 | 2013-10-01 10:23:44 -0700 | [diff] [blame] | 289 | /* Worst case is non GSO/TSO : each frame consumes one skb |
| 290 | * and skb->head is kmalloced using power of two area of memory |
| 291 | */ |
| 292 | per_mss = max_t(u32, tp->rx_opt.mss_clamp, tp->mss_cache) + |
| 293 | MAX_TCP_HEADER + |
| 294 | SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); |
| 295 | |
| 296 | per_mss = roundup_pow_of_two(per_mss) + |
| 297 | SKB_DATA_ALIGN(sizeof(struct sk_buff)); |
| 298 | |
| 299 | nr_segs = max_t(u32, TCP_INIT_CWND, tp->snd_cwnd); |
| 300 | nr_segs = max_t(u32, nr_segs, tp->reordering + 1); |
| 301 | |
| 302 | /* Fast Recovery (RFC 5681 3.2) : |
| 303 | * Cubic needs 1.7 factor, rounded to 2 to include |
| 304 | * extra cushion (application might react slowly to POLLOUT) |
| 305 | */ |
| 306 | sndmem = 2 * nr_segs * per_mss; |
| 307 | |
Eric Dumazet | 06a59ec | 2011-10-13 18:24:42 +0000 | [diff] [blame] | 308 | if (sk->sk_sndbuf < sndmem) |
| 309 | sk->sk_sndbuf = min(sndmem, sysctl_tcp_wmem[2]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 310 | } |
| 311 | |
| 312 | /* 2. Tuning advertised window (window_clamp, rcv_ssthresh) |
| 313 | * |
| 314 | * All tcp_full_space() is split to two parts: "network" buffer, allocated |
| 315 | * forward and advertised in receiver window (tp->rcv_wnd) and |
| 316 | * "application buffer", required to isolate scheduling/application |
| 317 | * latencies from network. |
| 318 | * window_clamp is maximal advertised window. It can be less than |
| 319 | * tcp_full_space(), in this case tcp_full_space() - window_clamp |
| 320 | * is reserved for "application" buffer. The less window_clamp is |
| 321 | * the smoother our behaviour from viewpoint of network, but the lower |
| 322 | * throughput and the higher sensitivity of the connection to losses. 8) |
| 323 | * |
| 324 | * rcv_ssthresh is more strict window_clamp used at "slow start" |
| 325 | * phase to predict further behaviour of this connection. |
| 326 | * It is used for two goals: |
| 327 | * - to enforce header prediction at sender, even when application |
| 328 | * requires some significant "application buffer". It is check #1. |
| 329 | * - to prevent pruning of receive queue because of misprediction |
| 330 | * of receiver window. Check #2. |
| 331 | * |
| 332 | * The scheme does not work when sender sends good segments opening |
Stephen Hemminger | caa20d9a | 2005-11-10 17:13:47 -0800 | [diff] [blame] | 333 | * window and then starts to feed us spaghetti. But it should work |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 334 | * in common situations. Otherwise, we have to rely on queue collapsing. |
| 335 | */ |
| 336 | |
| 337 | /* Slow part of check#2. */ |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 338 | 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] | 339 | { |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 340 | struct tcp_sock *tp = tcp_sk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 341 | /* Optimize this! */ |
Eric Dumazet | dfd4f0a | 2007-12-21 06:07:53 -0800 | [diff] [blame] | 342 | int truesize = tcp_win_from_space(skb->truesize) >> 1; |
| 343 | int window = tcp_win_from_space(sysctl_tcp_rmem[2]) >> 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 344 | |
| 345 | while (tp->rcv_ssthresh <= window) { |
| 346 | if (truesize <= skb->len) |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 347 | return 2 * inet_csk(sk)->icsk_ack.rcv_mss; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 348 | |
| 349 | truesize >>= 1; |
| 350 | window >>= 1; |
| 351 | } |
| 352 | return 0; |
| 353 | } |
| 354 | |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 355 | 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] | 356 | { |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 357 | struct tcp_sock *tp = tcp_sk(sk); |
| 358 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 359 | /* Check #1 */ |
| 360 | if (tp->rcv_ssthresh < tp->window_clamp && |
| 361 | (int)tp->rcv_ssthresh < tcp_space(sk) && |
Glauber Costa | 180d8cd | 2011-12-11 21:47:02 +0000 | [diff] [blame] | 362 | !sk_under_memory_pressure(sk)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 363 | int incr; |
| 364 | |
| 365 | /* Check #2. Increase window, if skb with such overhead |
| 366 | * will fit to rcvbuf in future. |
| 367 | */ |
| 368 | if (tcp_win_from_space(skb->truesize) <= skb->len) |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 369 | incr = 2 * tp->advmss; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 370 | else |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 371 | incr = __tcp_grow_window(sk, skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 372 | |
| 373 | if (incr) { |
Eric Dumazet | 4d846f0 | 2012-04-16 23:28:07 +0000 | [diff] [blame] | 374 | incr = max_t(int, incr, 2 * skb->len); |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 375 | tp->rcv_ssthresh = min(tp->rcv_ssthresh + incr, |
| 376 | tp->window_clamp); |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 377 | inet_csk(sk)->icsk_ack.quick |= 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 378 | } |
| 379 | } |
| 380 | } |
| 381 | |
| 382 | /* 3. Tuning rcvbuf, when connection enters established state. */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 383 | static void tcp_fixup_rcvbuf(struct sock *sk) |
| 384 | { |
Eric Dumazet | e9266a0 | 2011-10-20 16:53:56 -0400 | [diff] [blame] | 385 | u32 mss = tcp_sk(sk)->advmss; |
Eric Dumazet | e9266a0 | 2011-10-20 16:53:56 -0400 | [diff] [blame] | 386 | int rcvmem; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 387 | |
Yuchung Cheng | 85f1652 | 2013-06-11 15:35:32 -0700 | [diff] [blame] | 388 | rcvmem = 2 * SKB_TRUESIZE(mss + MAX_TCP_HEADER) * |
| 389 | tcp_default_init_rwnd(mss); |
Eric Dumazet | e9266a0 | 2011-10-20 16:53:56 -0400 | [diff] [blame] | 390 | |
Eric Dumazet | b0983d3 | 2013-09-20 13:56:58 -0700 | [diff] [blame] | 391 | /* Dynamic Right Sizing (DRS) has 2 to 3 RTT latency |
| 392 | * Allow enough cushion so that sender is not limited by our window |
| 393 | */ |
| 394 | if (sysctl_tcp_moderate_rcvbuf) |
| 395 | rcvmem <<= 2; |
| 396 | |
Eric Dumazet | e9266a0 | 2011-10-20 16:53:56 -0400 | [diff] [blame] | 397 | if (sk->sk_rcvbuf < rcvmem) |
| 398 | sk->sk_rcvbuf = min(rcvmem, sysctl_tcp_rmem[2]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 399 | } |
| 400 | |
Stephen Hemminger | caa20d9a | 2005-11-10 17:13:47 -0800 | [diff] [blame] | 401 | /* 4. Try to fixup all. It is made immediately after connection enters |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 402 | * established state. |
| 403 | */ |
Jerry Chu | 1046716 | 2012-08-31 12:29:11 +0000 | [diff] [blame] | 404 | void tcp_init_buffer_space(struct sock *sk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 405 | { |
| 406 | struct tcp_sock *tp = tcp_sk(sk); |
| 407 | int maxwin; |
| 408 | |
| 409 | if (!(sk->sk_userlocks & SOCK_RCVBUF_LOCK)) |
| 410 | tcp_fixup_rcvbuf(sk); |
| 411 | if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) |
Eric Dumazet | 6ae7053 | 2013-10-01 10:23:44 -0700 | [diff] [blame] | 412 | tcp_sndbuf_expand(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 413 | |
| 414 | tp->rcvq_space.space = tp->rcv_wnd; |
Eric Dumazet | b0983d3 | 2013-09-20 13:56:58 -0700 | [diff] [blame] | 415 | tp->rcvq_space.time = tcp_time_stamp; |
| 416 | tp->rcvq_space.seq = tp->copied_seq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 417 | |
| 418 | maxwin = tcp_full_space(sk); |
| 419 | |
| 420 | if (tp->window_clamp >= maxwin) { |
| 421 | tp->window_clamp = maxwin; |
| 422 | |
| 423 | if (sysctl_tcp_app_win && maxwin > 4 * tp->advmss) |
| 424 | tp->window_clamp = max(maxwin - |
| 425 | (maxwin >> sysctl_tcp_app_win), |
| 426 | 4 * tp->advmss); |
| 427 | } |
| 428 | |
| 429 | /* Force reservation of one segment. */ |
| 430 | if (sysctl_tcp_app_win && |
| 431 | tp->window_clamp > 2 * tp->advmss && |
| 432 | tp->window_clamp + tp->advmss > maxwin) |
| 433 | tp->window_clamp = max(2 * tp->advmss, maxwin - tp->advmss); |
| 434 | |
| 435 | tp->rcv_ssthresh = min(tp->rcv_ssthresh, tp->window_clamp); |
| 436 | tp->snd_cwnd_stamp = tcp_time_stamp; |
| 437 | } |
| 438 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 439 | /* 5. Recalculate window clamp after socket hit its memory bounds. */ |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 440 | static void tcp_clamp_window(struct sock *sk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 441 | { |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 442 | struct tcp_sock *tp = tcp_sk(sk); |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 443 | struct inet_connection_sock *icsk = inet_csk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 444 | |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 445 | icsk->icsk_ack.quick = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 446 | |
John Heffner | 326f36e | 2005-11-10 17:11:48 -0800 | [diff] [blame] | 447 | if (sk->sk_rcvbuf < sysctl_tcp_rmem[2] && |
| 448 | !(sk->sk_userlocks & SOCK_RCVBUF_LOCK) && |
Glauber Costa | 180d8cd | 2011-12-11 21:47:02 +0000 | [diff] [blame] | 449 | !sk_under_memory_pressure(sk) && |
| 450 | sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)) { |
John Heffner | 326f36e | 2005-11-10 17:11:48 -0800 | [diff] [blame] | 451 | sk->sk_rcvbuf = min(atomic_read(&sk->sk_rmem_alloc), |
| 452 | sysctl_tcp_rmem[2]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 453 | } |
John Heffner | 326f36e | 2005-11-10 17:11:48 -0800 | [diff] [blame] | 454 | if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf) |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 455 | tp->rcv_ssthresh = min(tp->window_clamp, 2U * tp->advmss); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 456 | } |
| 457 | |
Stephen Hemminger | 40efc6f | 2006-01-03 16:03:49 -0800 | [diff] [blame] | 458 | /* Initialize RCV_MSS value. |
| 459 | * RCV_MSS is an our guess about MSS used by the peer. |
| 460 | * We haven't any direct information about the MSS. |
| 461 | * It's better to underestimate the RCV_MSS rather than overestimate. |
| 462 | * Overestimations make us ACKing less frequently than needed. |
| 463 | * Underestimations are more easy to detect and fix by tcp_measure_rcv_mss(). |
| 464 | */ |
| 465 | void tcp_initialize_rcv_mss(struct sock *sk) |
| 466 | { |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 467 | const struct tcp_sock *tp = tcp_sk(sk); |
Stephen Hemminger | 40efc6f | 2006-01-03 16:03:49 -0800 | [diff] [blame] | 468 | unsigned int hint = min_t(unsigned int, tp->advmss, tp->mss_cache); |
| 469 | |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 470 | hint = min(hint, tp->rcv_wnd / 2); |
William Allen Simpson | bee7ca9 | 2009-11-10 09:51:18 +0000 | [diff] [blame] | 471 | hint = min(hint, TCP_MSS_DEFAULT); |
Stephen Hemminger | 40efc6f | 2006-01-03 16:03:49 -0800 | [diff] [blame] | 472 | hint = max(hint, TCP_MIN_MSS); |
| 473 | |
| 474 | inet_csk(sk)->icsk_ack.rcv_mss = hint; |
| 475 | } |
Eric Dumazet | 4bc2f18 | 2010-07-09 21:22:10 +0000 | [diff] [blame] | 476 | EXPORT_SYMBOL(tcp_initialize_rcv_mss); |
Stephen Hemminger | 40efc6f | 2006-01-03 16:03:49 -0800 | [diff] [blame] | 477 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 478 | /* Receiver "autotuning" code. |
| 479 | * |
| 480 | * The algorithm for RTT estimation w/o timestamps is based on |
| 481 | * 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] | 482 | * <http://public.lanl.gov/radiant/pubs.html#DRS> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 483 | * |
| 484 | * More detail on this code can be found at |
Justin P. Mattock | 631dd1a | 2010-10-18 11:03:14 +0200 | [diff] [blame] | 485 | * <http://staff.psc.edu/jheffner/>, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 486 | * though this reference is out of date. A new paper |
| 487 | * is pending. |
| 488 | */ |
| 489 | static void tcp_rcv_rtt_update(struct tcp_sock *tp, u32 sample, int win_dep) |
| 490 | { |
| 491 | u32 new_sample = tp->rcv_rtt_est.rtt; |
| 492 | long m = sample; |
| 493 | |
| 494 | if (m == 0) |
| 495 | m = 1; |
| 496 | |
| 497 | if (new_sample != 0) { |
| 498 | /* If we sample in larger samples in the non-timestamp |
| 499 | * case, we could grossly overestimate the RTT especially |
| 500 | * with chatty applications or bulk transfer apps which |
| 501 | * are stalled on filesystem I/O. |
| 502 | * |
| 503 | * Also, since we are only going for a minimum in the |
Stephen Hemminger | 31f3426 | 2005-11-15 15:17:10 -0800 | [diff] [blame] | 504 | * non-timestamp case, we do not smooth things out |
Stephen Hemminger | caa20d9a | 2005-11-10 17:13:47 -0800 | [diff] [blame] | 505 | * else with timestamps disabled convergence takes too |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 506 | * long. |
| 507 | */ |
| 508 | if (!win_dep) { |
| 509 | m -= (new_sample >> 3); |
| 510 | new_sample += m; |
Neal Cardwell | 18a223e | 2012-04-10 07:59:20 +0000 | [diff] [blame] | 511 | } else { |
| 512 | m <<= 3; |
| 513 | if (m < new_sample) |
| 514 | new_sample = m; |
| 515 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 516 | } else { |
Stephen Hemminger | caa20d9a | 2005-11-10 17:13:47 -0800 | [diff] [blame] | 517 | /* No previous measure. */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 518 | new_sample = m << 3; |
| 519 | } |
| 520 | |
| 521 | if (tp->rcv_rtt_est.rtt != new_sample) |
| 522 | tp->rcv_rtt_est.rtt = new_sample; |
| 523 | } |
| 524 | |
| 525 | static inline void tcp_rcv_rtt_measure(struct tcp_sock *tp) |
| 526 | { |
| 527 | if (tp->rcv_rtt_est.time == 0) |
| 528 | goto new_measure; |
| 529 | if (before(tp->rcv_nxt, tp->rcv_rtt_est.seq)) |
| 530 | return; |
Neal Cardwell | 651913c | 2012-04-27 11:29:37 -0400 | [diff] [blame] | 531 | 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] | 532 | |
| 533 | new_measure: |
| 534 | tp->rcv_rtt_est.seq = tp->rcv_nxt + tp->rcv_wnd; |
| 535 | tp->rcv_rtt_est.time = tcp_time_stamp; |
| 536 | } |
| 537 | |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 538 | static inline void tcp_rcv_rtt_measure_ts(struct sock *sk, |
| 539 | const struct sk_buff *skb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 540 | { |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 541 | struct tcp_sock *tp = tcp_sk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 542 | if (tp->rx_opt.rcv_tsecr && |
| 543 | (TCP_SKB_CB(skb)->end_seq - |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 544 | TCP_SKB_CB(skb)->seq >= inet_csk(sk)->icsk_ack.rcv_mss)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 545 | tcp_rcv_rtt_update(tp, tcp_time_stamp - tp->rx_opt.rcv_tsecr, 0); |
| 546 | } |
| 547 | |
| 548 | /* |
| 549 | * This function should be called every time data is copied to user space. |
| 550 | * It calculates the appropriate TCP receive buffer space. |
| 551 | */ |
| 552 | void tcp_rcv_space_adjust(struct sock *sk) |
| 553 | { |
| 554 | struct tcp_sock *tp = tcp_sk(sk); |
| 555 | int time; |
Eric Dumazet | b0983d3 | 2013-09-20 13:56:58 -0700 | [diff] [blame] | 556 | int copied; |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 557 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 558 | time = tcp_time_stamp - tp->rcvq_space.time; |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 559 | 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] | 560 | return; |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 561 | |
Eric Dumazet | b0983d3 | 2013-09-20 13:56:58 -0700 | [diff] [blame] | 562 | /* Number of bytes copied to user in last RTT */ |
| 563 | copied = tp->copied_seq - tp->rcvq_space.seq; |
| 564 | if (copied <= tp->rcvq_space.space) |
| 565 | goto new_measure; |
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 | /* A bit of theory : |
| 568 | * copied = bytes received in previous RTT, our base window |
| 569 | * To cope with packet losses, we need a 2x factor |
| 570 | * To cope with slow start, and sender growing its cwin by 100 % |
| 571 | * every RTT, we need a 4x factor, because the ACK we are sending |
| 572 | * now is for the next RTT, not the current one : |
| 573 | * <prev RTT . ><current RTT .. ><next RTT .... > |
| 574 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 575 | |
Eric Dumazet | b0983d3 | 2013-09-20 13:56:58 -0700 | [diff] [blame] | 576 | if (sysctl_tcp_moderate_rcvbuf && |
| 577 | !(sk->sk_userlocks & SOCK_RCVBUF_LOCK)) { |
| 578 | int rcvwin, rcvmem, rcvbuf; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 579 | |
Eric Dumazet | b0983d3 | 2013-09-20 13:56:58 -0700 | [diff] [blame] | 580 | /* minimal window to cope with packet losses, assuming |
| 581 | * steady state. Add some cushion because of small variations. |
| 582 | */ |
| 583 | rcvwin = (copied << 1) + 16 * tp->advmss; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 584 | |
Eric Dumazet | b0983d3 | 2013-09-20 13:56:58 -0700 | [diff] [blame] | 585 | /* If rate increased by 25%, |
| 586 | * assume slow start, rcvwin = 3 * copied |
| 587 | * If rate increased by 50%, |
| 588 | * assume sender can use 2x growth, rcvwin = 4 * copied |
| 589 | */ |
| 590 | if (copied >= |
| 591 | tp->rcvq_space.space + (tp->rcvq_space.space >> 2)) { |
| 592 | if (copied >= |
| 593 | tp->rcvq_space.space + (tp->rcvq_space.space >> 1)) |
| 594 | rcvwin <<= 1; |
| 595 | else |
| 596 | rcvwin += (rcvwin >> 1); |
| 597 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 598 | |
Eric Dumazet | b0983d3 | 2013-09-20 13:56:58 -0700 | [diff] [blame] | 599 | rcvmem = SKB_TRUESIZE(tp->advmss + MAX_TCP_HEADER); |
| 600 | while (tcp_win_from_space(rcvmem) < tp->advmss) |
| 601 | rcvmem += 128; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 602 | |
Eric Dumazet | b0983d3 | 2013-09-20 13:56:58 -0700 | [diff] [blame] | 603 | rcvbuf = min(rcvwin / tp->advmss * rcvmem, sysctl_tcp_rmem[2]); |
| 604 | if (rcvbuf > sk->sk_rcvbuf) { |
| 605 | sk->sk_rcvbuf = rcvbuf; |
| 606 | |
| 607 | /* Make the window clamp follow along. */ |
| 608 | tp->window_clamp = rcvwin; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 609 | } |
| 610 | } |
Eric Dumazet | b0983d3 | 2013-09-20 13:56:58 -0700 | [diff] [blame] | 611 | tp->rcvq_space.space = copied; |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 612 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 613 | new_measure: |
| 614 | tp->rcvq_space.seq = tp->copied_seq; |
| 615 | tp->rcvq_space.time = tcp_time_stamp; |
| 616 | } |
| 617 | |
| 618 | /* There is something which you must keep in mind when you analyze the |
| 619 | * behavior of the tp->ato delayed ack timeout interval. When a |
| 620 | * connection starts up, we want to ack as quickly as possible. The |
| 621 | * problem is that "good" TCP's do slow start at the beginning of data |
| 622 | * transmission. The means that until we send the first few ACK's the |
| 623 | * sender will sit on his end and only queue most of his data, because |
| 624 | * he can only send snd_cwnd unacked packets at any given time. For |
| 625 | * each ACK we send, he increments snd_cwnd and transmits more of his |
| 626 | * queue. -DaveM |
| 627 | */ |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 628 | 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] | 629 | { |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 630 | struct tcp_sock *tp = tcp_sk(sk); |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 631 | struct inet_connection_sock *icsk = inet_csk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 632 | u32 now; |
| 633 | |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 634 | inet_csk_schedule_ack(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 635 | |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 636 | tcp_measure_rcv_mss(sk, skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 637 | |
| 638 | tcp_rcv_rtt_measure(tp); |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 639 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 640 | now = tcp_time_stamp; |
| 641 | |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 642 | if (!icsk->icsk_ack.ato) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 643 | /* The _first_ data packet received, initialize |
| 644 | * delayed ACK engine. |
| 645 | */ |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 646 | tcp_incr_quickack(sk); |
| 647 | icsk->icsk_ack.ato = TCP_ATO_MIN; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 648 | } else { |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 649 | int m = now - icsk->icsk_ack.lrcvtime; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 650 | |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 651 | if (m <= TCP_ATO_MIN / 2) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 652 | /* The fastest case is the first. */ |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 653 | icsk->icsk_ack.ato = (icsk->icsk_ack.ato >> 1) + TCP_ATO_MIN / 2; |
| 654 | } else if (m < icsk->icsk_ack.ato) { |
| 655 | icsk->icsk_ack.ato = (icsk->icsk_ack.ato >> 1) + m; |
| 656 | if (icsk->icsk_ack.ato > icsk->icsk_rto) |
| 657 | icsk->icsk_ack.ato = icsk->icsk_rto; |
| 658 | } else if (m > icsk->icsk_rto) { |
Stephen Hemminger | caa20d9a | 2005-11-10 17:13:47 -0800 | [diff] [blame] | 659 | /* Too long gap. Apparently sender failed to |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 660 | * restart window, so that we send ACKs quickly. |
| 661 | */ |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 662 | tcp_incr_quickack(sk); |
Hideo Aoki | 3ab224b | 2007-12-31 00:11:19 -0800 | [diff] [blame] | 663 | sk_mem_reclaim(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 664 | } |
| 665 | } |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 666 | icsk->icsk_ack.lrcvtime = now; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 667 | |
Florian Westphal | 735d383 | 2014-09-29 13:08:30 +0200 | [diff] [blame] | 668 | tcp_ecn_check_ce(tp, skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 669 | |
| 670 | if (skb->len >= 128) |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 671 | tcp_grow_window(sk, skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 672 | } |
| 673 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 674 | /* Called to compute a smoothed rtt estimate. The data fed to this |
| 675 | * routine either comes from timestamps, or from segments that were |
| 676 | * known _not_ to have been retransmitted [see Karn/Partridge |
| 677 | * Proceedings SIGCOMM 87]. The algorithm is from the SIGCOMM 88 |
| 678 | * piece by Van Jacobson. |
| 679 | * NOTE: the next three routines used to be one big routine. |
| 680 | * To save cycles in the RFC 1323 implementation it was better to break |
| 681 | * it up into three procedures. -- erics |
| 682 | */ |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 683 | static void tcp_rtt_estimator(struct sock *sk, long mrtt_us) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 684 | { |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 685 | struct tcp_sock *tp = tcp_sk(sk); |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 686 | long m = mrtt_us; /* RTT */ |
| 687 | u32 srtt = tp->srtt_us; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 688 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 689 | /* The following amusing code comes from Jacobson's |
| 690 | * article in SIGCOMM '88. Note that rtt and mdev |
| 691 | * are scaled versions of rtt and mean deviation. |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 692 | * This is designed to be as fast as possible |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 693 | * m stands for "measurement". |
| 694 | * |
| 695 | * On a 1990 paper the rto value is changed to: |
| 696 | * RTO = rtt + 4 * mdev |
| 697 | * |
| 698 | * Funny. This algorithm seems to be very broken. |
| 699 | * These formulae increase RTO, when it should be decreased, increase |
Stephen Hemminger | 31f3426 | 2005-11-15 15:17:10 -0800 | [diff] [blame] | 700 | * too slowly, when it should be increased quickly, decrease too quickly |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 701 | * etc. I guess in BSD RTO takes ONE value, so that it is absolutely |
| 702 | * does not matter how to _calculate_ it. Seems, it was trap |
| 703 | * that VJ failed to avoid. 8) |
| 704 | */ |
Eric Dumazet | 4a5ab4e | 2014-02-06 15:57:10 -0800 | [diff] [blame] | 705 | if (srtt != 0) { |
| 706 | m -= (srtt >> 3); /* m is now error in rtt est */ |
| 707 | srtt += m; /* rtt = 7/8 rtt + 1/8 new */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 708 | if (m < 0) { |
| 709 | m = -m; /* m is now abs(error) */ |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 710 | m -= (tp->mdev_us >> 2); /* similar update on mdev */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 711 | /* This is similar to one of Eifel findings. |
| 712 | * Eifel blocks mdev updates when rtt decreases. |
| 713 | * This solution is a bit different: we use finer gain |
| 714 | * for mdev in this case (alpha*beta). |
| 715 | * Like Eifel it also prevents growth of rto, |
| 716 | * but also it limits too fast rto decreases, |
| 717 | * happening in pure Eifel. |
| 718 | */ |
| 719 | if (m > 0) |
| 720 | m >>= 3; |
| 721 | } else { |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 722 | m -= (tp->mdev_us >> 2); /* similar update on mdev */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 723 | } |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 724 | tp->mdev_us += m; /* mdev = 3/4 mdev + 1/4 new */ |
| 725 | if (tp->mdev_us > tp->mdev_max_us) { |
| 726 | tp->mdev_max_us = tp->mdev_us; |
| 727 | if (tp->mdev_max_us > tp->rttvar_us) |
| 728 | tp->rttvar_us = tp->mdev_max_us; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 729 | } |
| 730 | if (after(tp->snd_una, tp->rtt_seq)) { |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 731 | if (tp->mdev_max_us < tp->rttvar_us) |
| 732 | tp->rttvar_us -= (tp->rttvar_us - tp->mdev_max_us) >> 2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 733 | tp->rtt_seq = tp->snd_nxt; |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 734 | tp->mdev_max_us = tcp_rto_min_us(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 735 | } |
| 736 | } else { |
| 737 | /* no previous measure. */ |
Eric Dumazet | 4a5ab4e | 2014-02-06 15:57:10 -0800 | [diff] [blame] | 738 | srtt = m << 3; /* take the measured time to be rtt */ |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 739 | tp->mdev_us = m << 1; /* make sure rto = 3*rtt */ |
| 740 | tp->rttvar_us = max(tp->mdev_us, tcp_rto_min_us(sk)); |
| 741 | tp->mdev_max_us = tp->rttvar_us; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 742 | tp->rtt_seq = tp->snd_nxt; |
| 743 | } |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 744 | tp->srtt_us = max(1U, srtt); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 745 | } |
| 746 | |
Eric Dumazet | 95bd09e | 2013-08-27 05:46:32 -0700 | [diff] [blame] | 747 | /* Set the sk_pacing_rate to allow proper sizing of TSO packets. |
| 748 | * Note: TCP stack does not yet implement pacing. |
| 749 | * FQ packet scheduler can be used to implement cheap but effective |
| 750 | * TCP pacing, to smooth the burst on large writes when packets |
| 751 | * in flight is significantly lower than cwnd (or rwin) |
| 752 | */ |
| 753 | static void tcp_update_pacing_rate(struct sock *sk) |
| 754 | { |
| 755 | const struct tcp_sock *tp = tcp_sk(sk); |
| 756 | u64 rate; |
| 757 | |
| 758 | /* set sk_pacing_rate to 200 % of current rate (mss * cwnd / srtt) */ |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 759 | rate = (u64)tp->mss_cache * 2 * (USEC_PER_SEC << 3); |
Eric Dumazet | 95bd09e | 2013-08-27 05:46:32 -0700 | [diff] [blame] | 760 | |
| 761 | rate *= max(tp->snd_cwnd, tp->packets_out); |
| 762 | |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 763 | if (likely(tp->srtt_us)) |
| 764 | do_div(rate, tp->srtt_us); |
Eric Dumazet | 95bd09e | 2013-08-27 05:46:32 -0700 | [diff] [blame] | 765 | |
Eric Dumazet | ba53742 | 2013-10-09 17:14:52 -0700 | [diff] [blame] | 766 | /* ACCESS_ONCE() is needed because sch_fq fetches sk_pacing_rate |
| 767 | * without any lock. We want to make sure compiler wont store |
| 768 | * intermediate values in this location. |
| 769 | */ |
| 770 | ACCESS_ONCE(sk->sk_pacing_rate) = min_t(u64, rate, |
| 771 | sk->sk_max_pacing_rate); |
Eric Dumazet | 95bd09e | 2013-08-27 05:46:32 -0700 | [diff] [blame] | 772 | } |
| 773 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 774 | /* Calculate rto without backoff. This is the second half of Van Jacobson's |
| 775 | * routine referred to above. |
| 776 | */ |
stephen hemminger | f7e56a7 | 2013-12-29 11:39:51 -0800 | [diff] [blame] | 777 | static void tcp_set_rto(struct sock *sk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 778 | { |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 779 | const struct tcp_sock *tp = tcp_sk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 780 | /* Old crap is replaced with new one. 8) |
| 781 | * |
| 782 | * More seriously: |
| 783 | * 1. If rtt variance happened to be less 50msec, it is hallucination. |
| 784 | * It cannot be less due to utterly erratic ACK generation made |
| 785 | * at least by solaris and freebsd. "Erratic ACKs" has _nothing_ |
| 786 | * to do with delayed acks, because at cwnd>2 true delack timeout |
| 787 | * is invisible. Actually, Linux-2.4 also generates erratic |
Stephen Hemminger | caa20d9a | 2005-11-10 17:13:47 -0800 | [diff] [blame] | 788 | * ACKs in some circumstances. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 789 | */ |
Damian Lukowski | f1ecd5d | 2009-08-26 00:16:31 +0000 | [diff] [blame] | 790 | inet_csk(sk)->icsk_rto = __tcp_set_rto(tp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 791 | |
| 792 | /* 2. Fixups made earlier cannot be right. |
| 793 | * If we do not estimate RTO correctly without them, |
| 794 | * all the algo is pure shit and should be replaced |
Stephen Hemminger | caa20d9a | 2005-11-10 17:13:47 -0800 | [diff] [blame] | 795 | * with correct one. It is exactly, which we pretend to do. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 796 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 797 | |
Ilpo Järvinen | ee6aac5 | 2008-12-05 22:43:08 -0800 | [diff] [blame] | 798 | /* NOTE: clamping at TCP_RTO_MIN is not required, current algo |
| 799 | * guarantees that rto is higher. |
| 800 | */ |
Damian Lukowski | f1ecd5d | 2009-08-26 00:16:31 +0000 | [diff] [blame] | 801 | tcp_bound_rto(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 802 | } |
| 803 | |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 804 | __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] | 805 | { |
| 806 | __u32 cwnd = (dst ? dst_metric(dst, RTAX_INITCWND) : 0); |
| 807 | |
Gerrit Renker | 22b71c8 | 2010-08-29 19:23:12 +0000 | [diff] [blame] | 808 | if (!cwnd) |
David S. Miller | 442b963 | 2011-02-02 17:05:11 -0800 | [diff] [blame] | 809 | cwnd = TCP_INIT_CWND; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 810 | return min_t(__u32, cwnd, tp->snd_cwnd_clamp); |
| 811 | } |
| 812 | |
Ilpo Järvinen | e60402d | 2007-08-09 15:14:46 +0300 | [diff] [blame] | 813 | /* |
| 814 | * Packet counting of FACK is based on in-order assumptions, therefore TCP |
| 815 | * disables it when reordering is detected |
| 816 | */ |
David S. Miller | 4aabd8e | 2012-07-09 16:07:30 -0700 | [diff] [blame] | 817 | void tcp_disable_fack(struct tcp_sock *tp) |
Ilpo Järvinen | e60402d | 2007-08-09 15:14:46 +0300 | [diff] [blame] | 818 | { |
Ilpo Järvinen | 85cc391 | 2007-11-15 19:39:31 -0800 | [diff] [blame] | 819 | /* RFC3517 uses different metric in lost marker => reset on change */ |
| 820 | if (tcp_is_fack(tp)) |
| 821 | tp->lost_skb_hint = NULL; |
Vijay Subramanian | ab56222 | 2011-12-20 13:23:24 +0000 | [diff] [blame] | 822 | tp->rx_opt.sack_ok &= ~TCP_FACK_ENABLED; |
Ilpo Järvinen | e60402d | 2007-08-09 15:14:46 +0300 | [diff] [blame] | 823 | } |
| 824 | |
Ryousei Takano | 564262c | 2007-10-25 23:03:52 -0700 | [diff] [blame] | 825 | /* Take a notice that peer is sending D-SACKs */ |
Ilpo Järvinen | e60402d | 2007-08-09 15:14:46 +0300 | [diff] [blame] | 826 | static void tcp_dsack_seen(struct tcp_sock *tp) |
| 827 | { |
Vijay Subramanian | ab56222 | 2011-12-20 13:23:24 +0000 | [diff] [blame] | 828 | tp->rx_opt.sack_ok |= TCP_DSACK_SEEN; |
Ilpo Järvinen | e60402d | 2007-08-09 15:14:46 +0300 | [diff] [blame] | 829 | } |
| 830 | |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 831 | static void tcp_update_reordering(struct sock *sk, const int metric, |
| 832 | const int ts) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 833 | { |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 834 | struct tcp_sock *tp = tcp_sk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 835 | if (metric > tp->reordering) { |
Pavel Emelyanov | 40b215e | 2008-07-03 01:05:41 -0700 | [diff] [blame] | 836 | int mib_idx; |
| 837 | |
Eric Dumazet | dca145f | 2014-10-27 21:45:24 -0700 | [diff] [blame] | 838 | tp->reordering = min(sysctl_tcp_max_reordering, metric); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 839 | |
| 840 | /* This exciting event is worth to be remembered. 8) */ |
| 841 | if (ts) |
Pavel Emelyanov | 40b215e | 2008-07-03 01:05:41 -0700 | [diff] [blame] | 842 | mib_idx = LINUX_MIB_TCPTSREORDER; |
Ilpo Järvinen | e60402d | 2007-08-09 15:14:46 +0300 | [diff] [blame] | 843 | else if (tcp_is_reno(tp)) |
Pavel Emelyanov | 40b215e | 2008-07-03 01:05:41 -0700 | [diff] [blame] | 844 | mib_idx = LINUX_MIB_TCPRENOREORDER; |
Ilpo Järvinen | e60402d | 2007-08-09 15:14:46 +0300 | [diff] [blame] | 845 | else if (tcp_is_fack(tp)) |
Pavel Emelyanov | 40b215e | 2008-07-03 01:05:41 -0700 | [diff] [blame] | 846 | mib_idx = LINUX_MIB_TCPFACKREORDER; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 847 | else |
Pavel Emelyanov | 40b215e | 2008-07-03 01:05:41 -0700 | [diff] [blame] | 848 | mib_idx = LINUX_MIB_TCPSACKREORDER; |
| 849 | |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 850 | NET_INC_STATS_BH(sock_net(sk), mib_idx); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 851 | #if FASTRETRANS_DEBUG > 1 |
Joe Perches | 91df42b | 2012-05-15 14:11:54 +0000 | [diff] [blame] | 852 | pr_debug("Disorder%d %d %u f%u s%u rr%d\n", |
| 853 | tp->rx_opt.sack_ok, inet_csk(sk)->icsk_ca_state, |
| 854 | tp->reordering, |
| 855 | tp->fackets_out, |
| 856 | tp->sacked_out, |
| 857 | tp->undo_marker ? tp->undo_retrans : 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 858 | #endif |
Ilpo Järvinen | e60402d | 2007-08-09 15:14:46 +0300 | [diff] [blame] | 859 | tcp_disable_fack(tp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 860 | } |
Yuchung Cheng | eed530b | 2012-05-02 13:30:03 +0000 | [diff] [blame] | 861 | |
| 862 | if (metric > 0) |
| 863 | tcp_disable_early_retrans(tp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 864 | } |
| 865 | |
Ilpo Järvinen | 006f582 | 2008-09-20 21:20:20 -0700 | [diff] [blame] | 866 | /* This must be called before lost_out is incremented */ |
Ilpo Järvinen | c8c213f | 2008-09-20 21:18:55 -0700 | [diff] [blame] | 867 | static void tcp_verify_retransmit_hint(struct tcp_sock *tp, struct sk_buff *skb) |
| 868 | { |
Ilpo Järvinen | 006f582 | 2008-09-20 21:20:20 -0700 | [diff] [blame] | 869 | if ((tp->retransmit_skb_hint == NULL) || |
Ilpo Järvinen | c8c213f | 2008-09-20 21:18:55 -0700 | [diff] [blame] | 870 | before(TCP_SKB_CB(skb)->seq, |
| 871 | TCP_SKB_CB(tp->retransmit_skb_hint)->seq)) |
Ilpo Järvinen | 006f582 | 2008-09-20 21:20:20 -0700 | [diff] [blame] | 872 | tp->retransmit_skb_hint = skb; |
| 873 | |
| 874 | if (!tp->lost_out || |
| 875 | after(TCP_SKB_CB(skb)->end_seq, tp->retransmit_high)) |
| 876 | tp->retransmit_high = TCP_SKB_CB(skb)->end_seq; |
Ilpo Järvinen | c8c213f | 2008-09-20 21:18:55 -0700 | [diff] [blame] | 877 | } |
| 878 | |
Ilpo Järvinen | 41ea36e | 2008-09-20 21:19:22 -0700 | [diff] [blame] | 879 | static void tcp_skb_mark_lost(struct tcp_sock *tp, struct sk_buff *skb) |
| 880 | { |
| 881 | if (!(TCP_SKB_CB(skb)->sacked & (TCPCB_LOST|TCPCB_SACKED_ACKED))) { |
| 882 | tcp_verify_retransmit_hint(tp, skb); |
| 883 | |
| 884 | tp->lost_out += tcp_skb_pcount(skb); |
| 885 | TCP_SKB_CB(skb)->sacked |= TCPCB_LOST; |
| 886 | } |
| 887 | } |
| 888 | |
Ilpo Järvinen | e1aa680 | 2008-11-24 21:11:55 -0800 | [diff] [blame] | 889 | static void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, |
| 890 | struct sk_buff *skb) |
Ilpo Järvinen | 006f582 | 2008-09-20 21:20:20 -0700 | [diff] [blame] | 891 | { |
| 892 | tcp_verify_retransmit_hint(tp, skb); |
| 893 | |
| 894 | if (!(TCP_SKB_CB(skb)->sacked & (TCPCB_LOST|TCPCB_SACKED_ACKED))) { |
| 895 | tp->lost_out += tcp_skb_pcount(skb); |
| 896 | TCP_SKB_CB(skb)->sacked |= TCPCB_LOST; |
| 897 | } |
| 898 | } |
| 899 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 900 | /* This procedure tags the retransmission queue when SACKs arrive. |
| 901 | * |
| 902 | * We have three tag bits: SACKED(S), RETRANS(R) and LOST(L). |
| 903 | * Packets in queue with these bits set are counted in variables |
| 904 | * sacked_out, retrans_out and lost_out, correspondingly. |
| 905 | * |
| 906 | * Valid combinations are: |
| 907 | * Tag InFlight Description |
| 908 | * 0 1 - orig segment is in flight. |
| 909 | * S 0 - nothing flies, orig reached receiver. |
| 910 | * L 0 - nothing flies, orig lost by net. |
| 911 | * R 2 - both orig and retransmit are in flight. |
| 912 | * L|R 1 - orig is lost, retransmit is in flight. |
| 913 | * S|R 1 - orig reached receiver, retrans is still in flight. |
| 914 | * (L|S|R is logically valid, it could occur when L|R is sacked, |
| 915 | * but it is equivalent to plain S and code short-curcuits it to S. |
| 916 | * L|S is logically invalid, it would mean -1 packet in flight 8)) |
| 917 | * |
| 918 | * These 6 states form finite state machine, controlled by the following events: |
| 919 | * 1. New ACK (+SACK) arrives. (tcp_sacktag_write_queue()) |
| 920 | * 2. Retransmission. (tcp_retransmit_skb(), tcp_xmit_retransmit_queue()) |
Yuchung Cheng | 974c123 | 2012-01-19 14:42:21 +0000 | [diff] [blame] | 921 | * 3. Loss detection event of two flavors: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 922 | * A. Scoreboard estimator decided the packet is lost. |
| 923 | * A'. Reno "three dupacks" marks head of queue lost. |
Yuchung Cheng | 974c123 | 2012-01-19 14:42:21 +0000 | [diff] [blame] | 924 | * A''. Its FACK modification, head until snd.fack is lost. |
| 925 | * B. SACK arrives sacking SND.NXT at the moment, when the |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 926 | * segment was retransmitted. |
| 927 | * 4. D-SACK added new rule: D-SACK changes any tag to S. |
| 928 | * |
| 929 | * It is pleasant to note, that state diagram turns out to be commutative, |
| 930 | * so that we are allowed not to be bothered by order of our actions, |
| 931 | * when multiple events arrive simultaneously. (see the function below). |
| 932 | * |
| 933 | * Reordering detection. |
| 934 | * -------------------- |
| 935 | * Reordering metric is maximal distance, which a packet can be displaced |
| 936 | * in packet stream. With SACKs we can estimate it: |
| 937 | * |
| 938 | * 1. SACK fills old hole and the corresponding segment was not |
| 939 | * ever retransmitted -> reordering. Alas, we cannot use it |
| 940 | * when segment was retransmitted. |
| 941 | * 2. The last flaw is solved with D-SACK. D-SACK arrives |
| 942 | * for retransmitted and already SACKed segment -> reordering.. |
| 943 | * Both of these heuristics are not used in Loss state, when we cannot |
| 944 | * account for retransmits accurately. |
Ilpo Järvinen | 5b3c988 | 2007-08-24 22:54:44 -0700 | [diff] [blame] | 945 | * |
| 946 | * SACK block validation. |
| 947 | * ---------------------- |
| 948 | * |
| 949 | * SACK block range validation checks that the received SACK block fits to |
| 950 | * the expected sequence limits, i.e., it is between SND.UNA and SND.NXT. |
| 951 | * 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] | 952 | * it means that the receiver is rather inconsistent with itself reporting |
| 953 | * SACK reneging when it should advance SND.UNA. Such SACK block this is |
| 954 | * perfectly valid, however, in light of RFC2018 which explicitly states |
| 955 | * that "SACK block MUST reflect the newest segment. Even if the newest |
| 956 | * segment is going to be discarded ...", not that it looks very clever |
| 957 | * in case of head skb. Due to potentional receiver driven attacks, we |
| 958 | * choose to avoid immediate execution of a walk in write queue due to |
| 959 | * reneging and defer head skb's loss recovery to standard loss recovery |
| 960 | * procedure that will eventually trigger (nothing forbids us doing this). |
Ilpo Järvinen | 5b3c988 | 2007-08-24 22:54:44 -0700 | [diff] [blame] | 961 | * |
| 962 | * Implements also blockage to start_seq wrap-around. Problem lies in the |
| 963 | * fact that though start_seq (s) is before end_seq (i.e., not reversed), |
| 964 | * there's no guarantee that it will be before snd_nxt (n). The problem |
| 965 | * happens when start_seq resides between end_seq wrap (e_w) and snd_nxt |
| 966 | * wrap (s_w): |
| 967 | * |
| 968 | * <- outs wnd -> <- wrapzone -> |
| 969 | * u e n u_w e_w s n_w |
| 970 | * | | | | | | | |
| 971 | * |<------------+------+----- TCP seqno space --------------+---------->| |
| 972 | * ...-- <2^31 ->| |<--------... |
| 973 | * ...---- >2^31 ------>| |<--------... |
| 974 | * |
| 975 | * Current code wouldn't be vulnerable but it's better still to discard such |
| 976 | * crazy SACK blocks. Doing this check for start_seq alone closes somewhat |
| 977 | * similar case (end_seq after snd_nxt wrap) as earlier reversed check in |
| 978 | * snd_nxt wrap -> snd_una region will then become "well defined", i.e., |
| 979 | * equal to the ideal case (infinite seqno space without wrap caused issues). |
| 980 | * |
| 981 | * With D-SACK the lower bound is extended to cover sequence space below |
| 982 | * SND.UNA down to undo_marker, which is the last point of interest. Yet |
Ryousei Takano | 564262c | 2007-10-25 23:03:52 -0700 | [diff] [blame] | 983 | * 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] | 984 | * for the normal SACK blocks, explained above). But there all simplicity |
| 985 | * ends, TCP might receive valid D-SACKs below that. As long as they reside |
| 986 | * fully below undo_marker they do not affect behavior in anyway and can |
| 987 | * therefore be safely ignored. In rare cases (which are more or less |
| 988 | * theoretical ones), the D-SACK will nicely cross that boundary due to skb |
| 989 | * fragmentation and packet reordering past skb's retransmission. To consider |
| 990 | * them correctly, the acceptable range must be extended even more though |
| 991 | * the exact amount is rather hard to quantify. However, tp->max_window can |
| 992 | * be used as an exaggerated estimate. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 993 | */ |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 994 | static bool tcp_is_sackblock_valid(struct tcp_sock *tp, bool is_dsack, |
| 995 | u32 start_seq, u32 end_seq) |
Ilpo Järvinen | 5b3c988 | 2007-08-24 22:54:44 -0700 | [diff] [blame] | 996 | { |
| 997 | /* Too far in future, or reversed (interpretation is ambiguous) */ |
| 998 | if (after(end_seq, tp->snd_nxt) || !before(start_seq, end_seq)) |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 999 | return false; |
Ilpo Järvinen | 5b3c988 | 2007-08-24 22:54:44 -0700 | [diff] [blame] | 1000 | |
| 1001 | /* Nasty start_seq wrap-around check (see comments above) */ |
| 1002 | if (!before(start_seq, tp->snd_nxt)) |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1003 | return false; |
Ilpo Järvinen | 5b3c988 | 2007-08-24 22:54:44 -0700 | [diff] [blame] | 1004 | |
Ryousei Takano | 564262c | 2007-10-25 23:03:52 -0700 | [diff] [blame] | 1005 | /* In outstanding window? ...This is valid exit for D-SACKs too. |
Ilpo Järvinen | 5b3c988 | 2007-08-24 22:54:44 -0700 | [diff] [blame] | 1006 | * start_seq == snd_una is non-sensical (see comments above) |
| 1007 | */ |
| 1008 | if (after(start_seq, tp->snd_una)) |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1009 | return true; |
Ilpo Järvinen | 5b3c988 | 2007-08-24 22:54:44 -0700 | [diff] [blame] | 1010 | |
| 1011 | if (!is_dsack || !tp->undo_marker) |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1012 | return false; |
Ilpo Järvinen | 5b3c988 | 2007-08-24 22:54:44 -0700 | [diff] [blame] | 1013 | |
| 1014 | /* ...Then it's D-SACK, and must reside below snd_una completely */ |
Zheng Yan | f779b2d | 2011-09-18 22:37:34 -0400 | [diff] [blame] | 1015 | if (after(end_seq, tp->snd_una)) |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1016 | return false; |
Ilpo Järvinen | 5b3c988 | 2007-08-24 22:54:44 -0700 | [diff] [blame] | 1017 | |
| 1018 | if (!before(start_seq, tp->undo_marker)) |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1019 | return true; |
Ilpo Järvinen | 5b3c988 | 2007-08-24 22:54:44 -0700 | [diff] [blame] | 1020 | |
| 1021 | /* Too old */ |
| 1022 | if (!after(end_seq, tp->undo_marker)) |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1023 | return false; |
Ilpo Järvinen | 5b3c988 | 2007-08-24 22:54:44 -0700 | [diff] [blame] | 1024 | |
| 1025 | /* Undo_marker boundary crossing (overestimates a lot). Known already: |
| 1026 | * start_seq < undo_marker and end_seq >= undo_marker. |
| 1027 | */ |
| 1028 | return !before(start_seq, end_seq - tp->max_window); |
| 1029 | } |
| 1030 | |
Ilpo Järvinen | 1c1e87e | 2007-10-10 02:45:32 -0700 | [diff] [blame] | 1031 | /* Check for lost retransmit. This superb idea is borrowed from "ratehalving". |
Yuchung Cheng | 974c123 | 2012-01-19 14:42:21 +0000 | [diff] [blame] | 1032 | * 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] | 1033 | * for reordering! Ugly, but should help. |
Ilpo Järvinen | f785a8e | 2007-10-11 17:35:41 -0700 | [diff] [blame] | 1034 | * |
| 1035 | * Search retransmitted skbs from write_queue that were sent when snd_nxt was |
| 1036 | * 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] | 1037 | * highest SACK block). Also calculate the lowest snd_nxt among the remaining |
| 1038 | * retransmitted skbs to avoid some costly processing per ACKs. |
Ilpo Järvinen | 1c1e87e | 2007-10-10 02:45:32 -0700 | [diff] [blame] | 1039 | */ |
Ilpo Järvinen | 407ef1d | 2007-12-02 00:47:57 +0200 | [diff] [blame] | 1040 | static void tcp_mark_lost_retrans(struct sock *sk) |
Ilpo Järvinen | 1c1e87e | 2007-10-10 02:45:32 -0700 | [diff] [blame] | 1041 | { |
Ilpo Järvinen | 9f58f3b | 2007-11-15 19:42:54 -0800 | [diff] [blame] | 1042 | const struct inet_connection_sock *icsk = inet_csk(sk); |
Ilpo Järvinen | 1c1e87e | 2007-10-10 02:45:32 -0700 | [diff] [blame] | 1043 | struct tcp_sock *tp = tcp_sk(sk); |
| 1044 | struct sk_buff *skb; |
Ilpo Järvinen | f785a8e | 2007-10-11 17:35:41 -0700 | [diff] [blame] | 1045 | int cnt = 0; |
Ilpo Järvinen | df2e014 | 2007-10-18 05:07:57 -0700 | [diff] [blame] | 1046 | u32 new_low_seq = tp->snd_nxt; |
Ilpo Järvinen | 6859d49 | 2007-12-02 00:48:06 +0200 | [diff] [blame] | 1047 | u32 received_upto = tcp_highest_sack_seq(tp); |
Ilpo Järvinen | 9f58f3b | 2007-11-15 19:42:54 -0800 | [diff] [blame] | 1048 | |
| 1049 | if (!tcp_is_fack(tp) || !tp->retrans_out || |
| 1050 | !after(received_upto, tp->lost_retrans_low) || |
| 1051 | icsk->icsk_ca_state != TCP_CA_Recovery) |
Ilpo Järvinen | 407ef1d | 2007-12-02 00:47:57 +0200 | [diff] [blame] | 1052 | return; |
Ilpo Järvinen | 1c1e87e | 2007-10-10 02:45:32 -0700 | [diff] [blame] | 1053 | |
| 1054 | tcp_for_write_queue(skb, sk) { |
| 1055 | u32 ack_seq = TCP_SKB_CB(skb)->ack_seq; |
| 1056 | |
| 1057 | if (skb == tcp_send_head(sk)) |
| 1058 | break; |
Ilpo Järvinen | f785a8e | 2007-10-11 17:35:41 -0700 | [diff] [blame] | 1059 | if (cnt == tp->retrans_out) |
Ilpo Järvinen | 1c1e87e | 2007-10-10 02:45:32 -0700 | [diff] [blame] | 1060 | break; |
| 1061 | if (!after(TCP_SKB_CB(skb)->end_seq, tp->snd_una)) |
| 1062 | continue; |
| 1063 | |
Ilpo Järvinen | f785a8e | 2007-10-11 17:35:41 -0700 | [diff] [blame] | 1064 | if (!(TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_RETRANS)) |
| 1065 | continue; |
| 1066 | |
Ilpo Järvinen | d0af416 | 2009-02-28 04:44:32 +0000 | [diff] [blame] | 1067 | /* TODO: We would like to get rid of tcp_is_fack(tp) only |
| 1068 | * constraint here (see above) but figuring out that at |
| 1069 | * least tp->reordering SACK blocks reside between ack_seq |
| 1070 | * and received_upto is not easy task to do cheaply with |
| 1071 | * the available datastructures. |
| 1072 | * |
| 1073 | * Whether FACK should check here for tp->reordering segs |
| 1074 | * in-between one could argue for either way (it would be |
| 1075 | * rather simple to implement as we could count fack_count |
| 1076 | * during the walk and do tp->fackets_out - fack_count). |
| 1077 | */ |
| 1078 | if (after(received_upto, ack_seq)) { |
Ilpo Järvinen | 1c1e87e | 2007-10-10 02:45:32 -0700 | [diff] [blame] | 1079 | TCP_SKB_CB(skb)->sacked &= ~TCPCB_SACKED_RETRANS; |
| 1080 | tp->retrans_out -= tcp_skb_pcount(skb); |
| 1081 | |
Ilpo Järvinen | 006f582 | 2008-09-20 21:20:20 -0700 | [diff] [blame] | 1082 | tcp_skb_mark_lost_uncond_verify(tp, skb); |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 1083 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPLOSTRETRANSMIT); |
Ilpo Järvinen | f785a8e | 2007-10-11 17:35:41 -0700 | [diff] [blame] | 1084 | } else { |
Ilpo Järvinen | df2e014 | 2007-10-18 05:07:57 -0700 | [diff] [blame] | 1085 | if (before(ack_seq, new_low_seq)) |
Ilpo Järvinen | b08d6cb | 2007-10-11 17:36:13 -0700 | [diff] [blame] | 1086 | new_low_seq = ack_seq; |
Ilpo Järvinen | f785a8e | 2007-10-11 17:35:41 -0700 | [diff] [blame] | 1087 | cnt += tcp_skb_pcount(skb); |
Ilpo Järvinen | 1c1e87e | 2007-10-10 02:45:32 -0700 | [diff] [blame] | 1088 | } |
| 1089 | } |
Ilpo Järvinen | b08d6cb | 2007-10-11 17:36:13 -0700 | [diff] [blame] | 1090 | |
| 1091 | if (tp->retrans_out) |
| 1092 | tp->lost_retrans_low = new_low_seq; |
Ilpo Järvinen | 1c1e87e | 2007-10-10 02:45:32 -0700 | [diff] [blame] | 1093 | } |
Ilpo Järvinen | 5b3c988 | 2007-08-24 22:54:44 -0700 | [diff] [blame] | 1094 | |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1095 | static bool tcp_check_dsack(struct sock *sk, const struct sk_buff *ack_skb, |
| 1096 | struct tcp_sack_block_wire *sp, int num_sacks, |
| 1097 | u32 prior_snd_una) |
David S. Miller | d06e021 | 2007-06-18 22:43:06 -0700 | [diff] [blame] | 1098 | { |
Pavel Emelyanov | 1ed8346 | 2008-07-16 20:29:51 -0700 | [diff] [blame] | 1099 | struct tcp_sock *tp = tcp_sk(sk); |
Harvey Harrison | d3e2ce3 | 2008-05-02 16:26:16 -0700 | [diff] [blame] | 1100 | u32 start_seq_0 = get_unaligned_be32(&sp[0].start_seq); |
| 1101 | u32 end_seq_0 = get_unaligned_be32(&sp[0].end_seq); |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1102 | bool dup_sack = false; |
David S. Miller | d06e021 | 2007-06-18 22:43:06 -0700 | [diff] [blame] | 1103 | |
| 1104 | if (before(start_seq_0, TCP_SKB_CB(ack_skb)->ack_seq)) { |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1105 | dup_sack = true; |
Ilpo Järvinen | e60402d | 2007-08-09 15:14:46 +0300 | [diff] [blame] | 1106 | tcp_dsack_seen(tp); |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 1107 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPDSACKRECV); |
David S. Miller | d06e021 | 2007-06-18 22:43:06 -0700 | [diff] [blame] | 1108 | } else if (num_sacks > 1) { |
Harvey Harrison | d3e2ce3 | 2008-05-02 16:26:16 -0700 | [diff] [blame] | 1109 | u32 end_seq_1 = get_unaligned_be32(&sp[1].end_seq); |
| 1110 | u32 start_seq_1 = get_unaligned_be32(&sp[1].start_seq); |
David S. Miller | d06e021 | 2007-06-18 22:43:06 -0700 | [diff] [blame] | 1111 | |
| 1112 | if (!after(end_seq_0, end_seq_1) && |
| 1113 | !before(start_seq_0, start_seq_1)) { |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1114 | dup_sack = true; |
Ilpo Järvinen | e60402d | 2007-08-09 15:14:46 +0300 | [diff] [blame] | 1115 | tcp_dsack_seen(tp); |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 1116 | NET_INC_STATS_BH(sock_net(sk), |
| 1117 | LINUX_MIB_TCPDSACKOFORECV); |
David S. Miller | d06e021 | 2007-06-18 22:43:06 -0700 | [diff] [blame] | 1118 | } |
| 1119 | } |
| 1120 | |
| 1121 | /* D-SACK for already forgotten data... Do dumb counting. */ |
Yuchung Cheng | 6e08d5e | 2014-07-02 12:07:16 -0700 | [diff] [blame] | 1122 | if (dup_sack && tp->undo_marker && tp->undo_retrans > 0 && |
David S. Miller | d06e021 | 2007-06-18 22:43:06 -0700 | [diff] [blame] | 1123 | !after(end_seq_0, prior_snd_una) && |
| 1124 | after(end_seq_0, tp->undo_marker)) |
| 1125 | tp->undo_retrans--; |
| 1126 | |
| 1127 | return dup_sack; |
| 1128 | } |
| 1129 | |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1130 | struct tcp_sacktag_state { |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 1131 | int reord; |
| 1132 | int fack_count; |
| 1133 | long rtt_us; /* RTT measured by SACKing never-retransmitted data */ |
| 1134 | int flag; |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1135 | }; |
| 1136 | |
Ilpo Järvinen | d193594 | 2007-10-11 17:34:25 -0700 | [diff] [blame] | 1137 | /* Check if skb is fully within the SACK block. In presence of GSO skbs, |
| 1138 | * the incoming SACK may not exactly match but we can find smaller MSS |
| 1139 | * aligned portion of it that matches. Therefore we might need to fragment |
| 1140 | * which may fail and creates some hassle (caller must handle error case |
| 1141 | * returns). |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1142 | * |
| 1143 | * FIXME: this could be merged to shift decision code |
Ilpo Järvinen | d193594 | 2007-10-11 17:34:25 -0700 | [diff] [blame] | 1144 | */ |
Adrian Bunk | 0f79efd | 2007-10-26 03:57:36 -0700 | [diff] [blame] | 1145 | 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] | 1146 | u32 start_seq, u32 end_seq) |
Ilpo Järvinen | d193594 | 2007-10-11 17:34:25 -0700 | [diff] [blame] | 1147 | { |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1148 | int err; |
| 1149 | bool in_sack; |
Ilpo Järvinen | d193594 | 2007-10-11 17:34:25 -0700 | [diff] [blame] | 1150 | unsigned int pkt_len; |
Ilpo Järvinen | adb92db | 2008-11-24 21:13:50 -0800 | [diff] [blame] | 1151 | unsigned int mss; |
Ilpo Järvinen | d193594 | 2007-10-11 17:34:25 -0700 | [diff] [blame] | 1152 | |
| 1153 | in_sack = !after(start_seq, TCP_SKB_CB(skb)->seq) && |
| 1154 | !before(end_seq, TCP_SKB_CB(skb)->end_seq); |
| 1155 | |
| 1156 | if (tcp_skb_pcount(skb) > 1 && !in_sack && |
| 1157 | after(TCP_SKB_CB(skb)->end_seq, start_seq)) { |
Ilpo Järvinen | adb92db | 2008-11-24 21:13:50 -0800 | [diff] [blame] | 1158 | mss = tcp_skb_mss(skb); |
Ilpo Järvinen | d193594 | 2007-10-11 17:34:25 -0700 | [diff] [blame] | 1159 | in_sack = !after(start_seq, TCP_SKB_CB(skb)->seq); |
| 1160 | |
Ilpo Järvinen | adb92db | 2008-11-24 21:13:50 -0800 | [diff] [blame] | 1161 | if (!in_sack) { |
Ilpo Järvinen | d193594 | 2007-10-11 17:34:25 -0700 | [diff] [blame] | 1162 | pkt_len = start_seq - TCP_SKB_CB(skb)->seq; |
Ilpo Järvinen | adb92db | 2008-11-24 21:13:50 -0800 | [diff] [blame] | 1163 | if (pkt_len < mss) |
| 1164 | pkt_len = mss; |
| 1165 | } else { |
Ilpo Järvinen | d193594 | 2007-10-11 17:34:25 -0700 | [diff] [blame] | 1166 | pkt_len = end_seq - TCP_SKB_CB(skb)->seq; |
Ilpo Järvinen | adb92db | 2008-11-24 21:13:50 -0800 | [diff] [blame] | 1167 | if (pkt_len < mss) |
| 1168 | return -EINVAL; |
| 1169 | } |
| 1170 | |
| 1171 | /* Round if necessary so that SACKs cover only full MSSes |
| 1172 | * and/or the remaining small portion (if present) |
| 1173 | */ |
| 1174 | if (pkt_len > mss) { |
| 1175 | unsigned int new_len = (pkt_len / mss) * mss; |
| 1176 | if (!in_sack && new_len < pkt_len) { |
| 1177 | new_len += mss; |
Neal Cardwell | 2cd0d74 | 2014-06-18 21:15:03 -0400 | [diff] [blame] | 1178 | if (new_len >= skb->len) |
Ilpo Järvinen | adb92db | 2008-11-24 21:13:50 -0800 | [diff] [blame] | 1179 | return 0; |
| 1180 | } |
| 1181 | pkt_len = new_len; |
| 1182 | } |
Octavian Purdila | 6cc55e0 | 2014-06-06 17:32:37 +0300 | [diff] [blame] | 1183 | err = tcp_fragment(sk, skb, pkt_len, mss, GFP_ATOMIC); |
Ilpo Järvinen | d193594 | 2007-10-11 17:34:25 -0700 | [diff] [blame] | 1184 | if (err < 0) |
| 1185 | return err; |
| 1186 | } |
| 1187 | |
| 1188 | return in_sack; |
| 1189 | } |
| 1190 | |
Neal Cardwell | cc9a672 | 2012-02-12 18:37:09 +0000 | [diff] [blame] | 1191 | /* Mark the given newly-SACKed range as such, adjusting counters and hints. */ |
| 1192 | static u8 tcp_sacktag_one(struct sock *sk, |
| 1193 | struct tcp_sacktag_state *state, u8 sacked, |
| 1194 | u32 start_seq, u32 end_seq, |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 1195 | int dup_sack, int pcount, |
| 1196 | const struct skb_mstamp *xmit_time) |
Ilpo Järvinen | 9e10c47c | 2007-11-15 19:44:56 -0800 | [diff] [blame] | 1197 | { |
Ilpo Järvinen | 6859d49 | 2007-12-02 00:48:06 +0200 | [diff] [blame] | 1198 | struct tcp_sock *tp = tcp_sk(sk); |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1199 | int fack_count = state->fack_count; |
Ilpo Järvinen | 9e10c47c | 2007-11-15 19:44:56 -0800 | [diff] [blame] | 1200 | |
| 1201 | /* Account D-SACK for retransmitted packet. */ |
| 1202 | if (dup_sack && (sacked & TCPCB_RETRANS)) { |
Yuchung Cheng | 6e08d5e | 2014-07-02 12:07:16 -0700 | [diff] [blame] | 1203 | if (tp->undo_marker && tp->undo_retrans > 0 && |
Neal Cardwell | cc9a672 | 2012-02-12 18:37:09 +0000 | [diff] [blame] | 1204 | after(end_seq, tp->undo_marker)) |
Ilpo Järvinen | 9e10c47c | 2007-11-15 19:44:56 -0800 | [diff] [blame] | 1205 | tp->undo_retrans--; |
Ilpo Järvinen | ede9f3b | 2007-12-02 00:47:58 +0200 | [diff] [blame] | 1206 | if (sacked & TCPCB_SACKED_ACKED) |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1207 | state->reord = min(fack_count, state->reord); |
Ilpo Järvinen | 9e10c47c | 2007-11-15 19:44:56 -0800 | [diff] [blame] | 1208 | } |
| 1209 | |
| 1210 | /* Nothing to do; acked frame is about to be dropped (was ACKed). */ |
Neal Cardwell | cc9a672 | 2012-02-12 18:37:09 +0000 | [diff] [blame] | 1211 | if (!after(end_seq, tp->snd_una)) |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1212 | return sacked; |
Ilpo Järvinen | 9e10c47c | 2007-11-15 19:44:56 -0800 | [diff] [blame] | 1213 | |
| 1214 | if (!(sacked & TCPCB_SACKED_ACKED)) { |
| 1215 | if (sacked & TCPCB_SACKED_RETRANS) { |
| 1216 | /* If the segment is not tagged as lost, |
| 1217 | * we do not clear RETRANS, believing |
| 1218 | * that retransmission is still in flight. |
| 1219 | */ |
| 1220 | if (sacked & TCPCB_LOST) { |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1221 | sacked &= ~(TCPCB_LOST|TCPCB_SACKED_RETRANS); |
Ilpo Järvinen | f58b22f | 2008-11-24 21:14:43 -0800 | [diff] [blame] | 1222 | tp->lost_out -= pcount; |
| 1223 | tp->retrans_out -= pcount; |
Ilpo Järvinen | 9e10c47c | 2007-11-15 19:44:56 -0800 | [diff] [blame] | 1224 | } |
| 1225 | } else { |
| 1226 | if (!(sacked & TCPCB_RETRANS)) { |
| 1227 | /* New sack for not retransmitted frame, |
| 1228 | * which was in hole. It is reordering. |
| 1229 | */ |
Neal Cardwell | cc9a672 | 2012-02-12 18:37:09 +0000 | [diff] [blame] | 1230 | if (before(start_seq, |
Ilpo Järvinen | 9e10c47c | 2007-11-15 19:44:56 -0800 | [diff] [blame] | 1231 | tcp_highest_sack_seq(tp))) |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1232 | state->reord = min(fack_count, |
| 1233 | state->reord); |
Yuchung Cheng | e33099f | 2013-03-20 13:33:00 +0000 | [diff] [blame] | 1234 | if (!after(end_seq, tp->high_seq)) |
| 1235 | state->flag |= FLAG_ORIG_SACK_ACKED; |
Yuchung Cheng | 59c9af4 | 2013-07-22 16:20:47 -0700 | [diff] [blame] | 1236 | /* Pick the earliest sequence sacked for RTT */ |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 1237 | if (state->rtt_us < 0) { |
| 1238 | struct skb_mstamp now; |
| 1239 | |
| 1240 | skb_mstamp_get(&now); |
| 1241 | state->rtt_us = skb_mstamp_us_delta(&now, |
| 1242 | xmit_time); |
| 1243 | } |
Ilpo Järvinen | 9e10c47c | 2007-11-15 19:44:56 -0800 | [diff] [blame] | 1244 | } |
| 1245 | |
| 1246 | if (sacked & TCPCB_LOST) { |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1247 | sacked &= ~TCPCB_LOST; |
Ilpo Järvinen | f58b22f | 2008-11-24 21:14:43 -0800 | [diff] [blame] | 1248 | tp->lost_out -= pcount; |
Ilpo Järvinen | 9e10c47c | 2007-11-15 19:44:56 -0800 | [diff] [blame] | 1249 | } |
| 1250 | } |
| 1251 | |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1252 | sacked |= TCPCB_SACKED_ACKED; |
| 1253 | state->flag |= FLAG_DATA_SACKED; |
Ilpo Järvinen | f58b22f | 2008-11-24 21:14:43 -0800 | [diff] [blame] | 1254 | tp->sacked_out += pcount; |
Ilpo Järvinen | 9e10c47c | 2007-11-15 19:44:56 -0800 | [diff] [blame] | 1255 | |
Ilpo Järvinen | f58b22f | 2008-11-24 21:14:43 -0800 | [diff] [blame] | 1256 | fack_count += pcount; |
Ilpo Järvinen | 9e10c47c | 2007-11-15 19:44:56 -0800 | [diff] [blame] | 1257 | |
| 1258 | /* Lost marker hint past SACKed? Tweak RFC3517 cnt */ |
| 1259 | if (!tcp_is_fack(tp) && (tp->lost_skb_hint != NULL) && |
Neal Cardwell | cc9a672 | 2012-02-12 18:37:09 +0000 | [diff] [blame] | 1260 | before(start_seq, TCP_SKB_CB(tp->lost_skb_hint)->seq)) |
Ilpo Järvinen | f58b22f | 2008-11-24 21:14:43 -0800 | [diff] [blame] | 1261 | tp->lost_cnt_hint += pcount; |
Ilpo Järvinen | 9e10c47c | 2007-11-15 19:44:56 -0800 | [diff] [blame] | 1262 | |
| 1263 | if (fack_count > tp->fackets_out) |
| 1264 | tp->fackets_out = fack_count; |
Ilpo Järvinen | 9e10c47c | 2007-11-15 19:44:56 -0800 | [diff] [blame] | 1265 | } |
| 1266 | |
| 1267 | /* D-SACK. We can detect redundant retransmission in S|R and plain R |
| 1268 | * frames and clear it. undo_retrans is decreased above, L|R frames |
| 1269 | * are accounted above as well. |
| 1270 | */ |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1271 | if (dup_sack && (sacked & TCPCB_SACKED_RETRANS)) { |
| 1272 | sacked &= ~TCPCB_SACKED_RETRANS; |
Ilpo Järvinen | f58b22f | 2008-11-24 21:14:43 -0800 | [diff] [blame] | 1273 | tp->retrans_out -= pcount; |
Ilpo Järvinen | 9e10c47c | 2007-11-15 19:44:56 -0800 | [diff] [blame] | 1274 | } |
| 1275 | |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1276 | return sacked; |
Ilpo Järvinen | 9e10c47c | 2007-11-15 19:44:56 -0800 | [diff] [blame] | 1277 | } |
| 1278 | |
Neal Cardwell | daef52b | 2012-02-12 18:37:10 +0000 | [diff] [blame] | 1279 | /* Shift newly-SACKed bytes from this skb to the immediately previous |
| 1280 | * already-SACKed sk_buff. Mark the newly-SACKed bytes as such. |
| 1281 | */ |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1282 | static bool tcp_shifted_skb(struct sock *sk, struct sk_buff *skb, |
| 1283 | struct tcp_sacktag_state *state, |
| 1284 | unsigned int pcount, int shifted, int mss, |
| 1285 | bool dup_sack) |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1286 | { |
| 1287 | struct tcp_sock *tp = tcp_sk(sk); |
Ilpo Järvinen | 5013316 | 2008-12-05 22:42:41 -0800 | [diff] [blame] | 1288 | struct sk_buff *prev = tcp_write_queue_prev(sk, skb); |
Neal Cardwell | daef52b | 2012-02-12 18:37:10 +0000 | [diff] [blame] | 1289 | u32 start_seq = TCP_SKB_CB(skb)->seq; /* start of newly-SACKed */ |
| 1290 | u32 end_seq = start_seq + shifted; /* end of newly-SACKed */ |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1291 | |
| 1292 | BUG_ON(!pcount); |
| 1293 | |
Neal Cardwell | 4c90d3b | 2012-02-26 10:06:19 +0000 | [diff] [blame] | 1294 | /* Adjust counters and hints for the newly sacked sequence |
| 1295 | * range but discard the return value since prev is already |
| 1296 | * marked. We must tag the range first because the seq |
| 1297 | * advancement below implicitly advances |
| 1298 | * tcp_highest_sack_seq() when skb is highest_sack. |
| 1299 | */ |
| 1300 | tcp_sacktag_one(sk, state, TCP_SKB_CB(skb)->sacked, |
Yuchung Cheng | 59c9af4 | 2013-07-22 16:20:47 -0700 | [diff] [blame] | 1301 | start_seq, end_seq, dup_sack, pcount, |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 1302 | &skb->skb_mstamp); |
Neal Cardwell | 4c90d3b | 2012-02-26 10:06:19 +0000 | [diff] [blame] | 1303 | |
| 1304 | if (skb == tp->lost_skb_hint) |
Neal Cardwell | 0af2a0d | 2012-02-13 20:22:08 +0000 | [diff] [blame] | 1305 | tp->lost_cnt_hint += pcount; |
| 1306 | |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1307 | TCP_SKB_CB(prev)->end_seq += shifted; |
| 1308 | TCP_SKB_CB(skb)->seq += shifted; |
| 1309 | |
Eric Dumazet | cd7d849 | 2014-09-24 04:11:22 -0700 | [diff] [blame] | 1310 | tcp_skb_pcount_add(prev, pcount); |
| 1311 | BUG_ON(tcp_skb_pcount(skb) < pcount); |
| 1312 | tcp_skb_pcount_add(skb, -pcount); |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1313 | |
| 1314 | /* When we're adding to gso_segs == 1, gso_size will be zero, |
| 1315 | * in theory this shouldn't be necessary but as long as DSACK |
| 1316 | * code can come after this skb later on it's better to keep |
| 1317 | * setting gso_size to something. |
| 1318 | */ |
| 1319 | if (!skb_shinfo(prev)->gso_size) { |
| 1320 | skb_shinfo(prev)->gso_size = mss; |
Pravin B Shelar | c9af6db | 2013-02-11 09:27:41 +0000 | [diff] [blame] | 1321 | skb_shinfo(prev)->gso_type = sk->sk_gso_type; |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1322 | } |
| 1323 | |
| 1324 | /* CHECKME: To clear or not to clear? Mimics normal skb currently */ |
Eric Dumazet | cd7d849 | 2014-09-24 04:11:22 -0700 | [diff] [blame] | 1325 | if (tcp_skb_pcount(skb) <= 1) { |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1326 | skb_shinfo(skb)->gso_size = 0; |
Pravin B Shelar | c9af6db | 2013-02-11 09:27:41 +0000 | [diff] [blame] | 1327 | skb_shinfo(skb)->gso_type = 0; |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1328 | } |
| 1329 | |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1330 | /* Difference in this won't matter, both ACKed by the same cumul. ACK */ |
| 1331 | TCP_SKB_CB(prev)->sacked |= (TCP_SKB_CB(skb)->sacked & TCPCB_EVER_RETRANS); |
| 1332 | |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1333 | if (skb->len > 0) { |
| 1334 | BUG_ON(!tcp_skb_pcount(skb)); |
Ilpo Järvinen | 111cc8b | 2008-11-24 21:27:22 -0800 | [diff] [blame] | 1335 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SACKSHIFTED); |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1336 | return false; |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1337 | } |
| 1338 | |
| 1339 | /* Whole SKB was eaten :-) */ |
| 1340 | |
Ilpo Järvinen | 92ee76b | 2008-11-24 21:26:56 -0800 | [diff] [blame] | 1341 | if (skb == tp->retransmit_skb_hint) |
| 1342 | tp->retransmit_skb_hint = prev; |
Ilpo Järvinen | 92ee76b | 2008-11-24 21:26:56 -0800 | [diff] [blame] | 1343 | if (skb == tp->lost_skb_hint) { |
| 1344 | tp->lost_skb_hint = prev; |
| 1345 | tp->lost_cnt_hint -= tcp_skb_pcount(prev); |
| 1346 | } |
| 1347 | |
Eric Dumazet | 5e8a402f | 2013-10-04 10:31:41 -0700 | [diff] [blame] | 1348 | TCP_SKB_CB(prev)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags; |
| 1349 | if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) |
| 1350 | TCP_SKB_CB(prev)->end_seq++; |
| 1351 | |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1352 | if (skb == tcp_highest_sack(sk)) |
| 1353 | tcp_advance_highest_sack(sk, skb); |
| 1354 | |
| 1355 | tcp_unlink_write_queue(skb, sk); |
| 1356 | sk_wmem_free_skb(sk, skb); |
| 1357 | |
Ilpo Järvinen | 111cc8b | 2008-11-24 21:27:22 -0800 | [diff] [blame] | 1358 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SACKMERGED); |
| 1359 | |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1360 | return true; |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1361 | } |
| 1362 | |
| 1363 | /* I wish gso_size would have a bit more sane initialization than |
| 1364 | * something-or-zero which complicates things |
| 1365 | */ |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 1366 | static int tcp_skb_seglen(const struct sk_buff *skb) |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1367 | { |
Ilpo Järvinen | 775ffab | 2008-12-05 22:41:26 -0800 | [diff] [blame] | 1368 | return tcp_skb_pcount(skb) == 1 ? skb->len : tcp_skb_mss(skb); |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1369 | } |
| 1370 | |
| 1371 | /* Shifting pages past head area doesn't work */ |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 1372 | static int skb_can_shift(const struct sk_buff *skb) |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1373 | { |
| 1374 | return !skb_headlen(skb) && skb_is_nonlinear(skb); |
| 1375 | } |
| 1376 | |
| 1377 | /* Try collapsing SACK blocks spanning across multiple skbs to a single |
| 1378 | * skb. |
| 1379 | */ |
| 1380 | 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] | 1381 | struct tcp_sacktag_state *state, |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1382 | u32 start_seq, u32 end_seq, |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1383 | bool dup_sack) |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1384 | { |
| 1385 | struct tcp_sock *tp = tcp_sk(sk); |
| 1386 | struct sk_buff *prev; |
| 1387 | int mss; |
| 1388 | int pcount = 0; |
| 1389 | int len; |
| 1390 | int in_sack; |
| 1391 | |
| 1392 | if (!sk_can_gso(sk)) |
| 1393 | goto fallback; |
| 1394 | |
| 1395 | /* Normally R but no L won't result in plain S */ |
| 1396 | if (!dup_sack && |
Ilpo Järvinen | 9969ca5 | 2008-12-05 22:41:06 -0800 | [diff] [blame] | 1397 | (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] | 1398 | goto fallback; |
| 1399 | if (!skb_can_shift(skb)) |
| 1400 | goto fallback; |
| 1401 | /* This frame is about to be dropped (was ACKed). */ |
| 1402 | if (!after(TCP_SKB_CB(skb)->end_seq, tp->snd_una)) |
| 1403 | goto fallback; |
| 1404 | |
| 1405 | /* Can only happen with delayed DSACK + discard craziness */ |
| 1406 | if (unlikely(skb == tcp_write_queue_head(sk))) |
| 1407 | goto fallback; |
| 1408 | prev = tcp_write_queue_prev(sk, skb); |
| 1409 | |
| 1410 | if ((TCP_SKB_CB(prev)->sacked & TCPCB_TAGBITS) != TCPCB_SACKED_ACKED) |
| 1411 | goto fallback; |
| 1412 | |
| 1413 | in_sack = !after(start_seq, TCP_SKB_CB(skb)->seq) && |
| 1414 | !before(end_seq, TCP_SKB_CB(skb)->end_seq); |
| 1415 | |
| 1416 | if (in_sack) { |
| 1417 | len = skb->len; |
| 1418 | pcount = tcp_skb_pcount(skb); |
Ilpo Järvinen | 775ffab | 2008-12-05 22:41:26 -0800 | [diff] [blame] | 1419 | mss = tcp_skb_seglen(skb); |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1420 | |
| 1421 | /* TODO: Fix DSACKs to not fragment already SACKed and we can |
| 1422 | * drop this restriction as unnecessary |
| 1423 | */ |
Ilpo Järvinen | 775ffab | 2008-12-05 22:41:26 -0800 | [diff] [blame] | 1424 | if (mss != tcp_skb_seglen(prev)) |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1425 | goto fallback; |
| 1426 | } else { |
| 1427 | if (!after(TCP_SKB_CB(skb)->end_seq, start_seq)) |
| 1428 | goto noop; |
| 1429 | /* CHECKME: This is non-MSS split case only?, this will |
| 1430 | * cause skipped skbs due to advancing loop btw, original |
| 1431 | * has that feature too |
| 1432 | */ |
| 1433 | if (tcp_skb_pcount(skb) <= 1) |
| 1434 | goto noop; |
| 1435 | |
| 1436 | in_sack = !after(start_seq, TCP_SKB_CB(skb)->seq); |
| 1437 | if (!in_sack) { |
| 1438 | /* TODO: head merge to next could be attempted here |
| 1439 | * if (!after(TCP_SKB_CB(skb)->end_seq, end_seq)), |
| 1440 | * though it might not be worth of the additional hassle |
| 1441 | * |
| 1442 | * ...we can probably just fallback to what was done |
| 1443 | * previously. We could try merging non-SACKed ones |
| 1444 | * as well but it probably isn't going to buy off |
| 1445 | * because later SACKs might again split them, and |
| 1446 | * it would make skb timestamp tracking considerably |
| 1447 | * harder problem. |
| 1448 | */ |
| 1449 | goto fallback; |
| 1450 | } |
| 1451 | |
| 1452 | len = end_seq - TCP_SKB_CB(skb)->seq; |
| 1453 | BUG_ON(len < 0); |
| 1454 | BUG_ON(len > skb->len); |
| 1455 | |
| 1456 | /* MSS boundaries should be honoured or else pcount will |
| 1457 | * severely break even though it makes things bit trickier. |
| 1458 | * Optimize common case to avoid most of the divides |
| 1459 | */ |
| 1460 | mss = tcp_skb_mss(skb); |
| 1461 | |
| 1462 | /* TODO: Fix DSACKs to not fragment already SACKed and we can |
| 1463 | * drop this restriction as unnecessary |
| 1464 | */ |
Ilpo Järvinen | 775ffab | 2008-12-05 22:41:26 -0800 | [diff] [blame] | 1465 | if (mss != tcp_skb_seglen(prev)) |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1466 | goto fallback; |
| 1467 | |
| 1468 | if (len == mss) { |
| 1469 | pcount = 1; |
| 1470 | } else if (len < mss) { |
| 1471 | goto noop; |
| 1472 | } else { |
| 1473 | pcount = len / mss; |
| 1474 | len = pcount * mss; |
| 1475 | } |
| 1476 | } |
| 1477 | |
Neal Cardwell | 4648dc9 | 2012-03-05 19:35:04 +0000 | [diff] [blame] | 1478 | /* tcp_sacktag_one() won't SACK-tag ranges below snd_una */ |
| 1479 | if (!after(TCP_SKB_CB(skb)->seq + len, tp->snd_una)) |
| 1480 | goto fallback; |
| 1481 | |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1482 | if (!skb_shift(prev, skb, len)) |
| 1483 | goto fallback; |
Ilpo Järvinen | 9ec06ff | 2009-03-01 00:21:36 -0800 | [diff] [blame] | 1484 | if (!tcp_shifted_skb(sk, skb, state, pcount, len, mss, dup_sack)) |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1485 | goto out; |
| 1486 | |
| 1487 | /* Hole filled allows collapsing with the next as well, this is very |
| 1488 | * useful when hole on every nth skb pattern happens |
| 1489 | */ |
| 1490 | if (prev == tcp_write_queue_tail(sk)) |
| 1491 | goto out; |
| 1492 | skb = tcp_write_queue_next(sk, prev); |
| 1493 | |
Ilpo Järvinen | f0bc52f | 2008-12-05 22:40:47 -0800 | [diff] [blame] | 1494 | if (!skb_can_shift(skb) || |
| 1495 | (skb == tcp_send_head(sk)) || |
| 1496 | ((TCP_SKB_CB(skb)->sacked & TCPCB_TAGBITS) != TCPCB_SACKED_ACKED) || |
Ilpo Järvinen | 775ffab | 2008-12-05 22:41:26 -0800 | [diff] [blame] | 1497 | (mss != tcp_skb_seglen(skb))) |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1498 | goto out; |
| 1499 | |
| 1500 | len = skb->len; |
| 1501 | if (skb_shift(prev, skb, len)) { |
| 1502 | pcount += tcp_skb_pcount(skb); |
Ilpo Järvinen | 9ec06ff | 2009-03-01 00:21:36 -0800 | [diff] [blame] | 1503 | 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] | 1504 | } |
| 1505 | |
| 1506 | out: |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1507 | state->fack_count += pcount; |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1508 | return prev; |
| 1509 | |
| 1510 | noop: |
| 1511 | return skb; |
| 1512 | |
| 1513 | fallback: |
Ilpo Järvinen | 111cc8b | 2008-11-24 21:27:22 -0800 | [diff] [blame] | 1514 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SACKSHIFTFALLBACK); |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1515 | return NULL; |
| 1516 | } |
| 1517 | |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1518 | static struct sk_buff *tcp_sacktag_walk(struct sk_buff *skb, struct sock *sk, |
| 1519 | struct tcp_sack_block *next_dup, |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1520 | struct tcp_sacktag_state *state, |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1521 | u32 start_seq, u32 end_seq, |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1522 | bool dup_sack_in) |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1523 | { |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1524 | struct tcp_sock *tp = tcp_sk(sk); |
| 1525 | struct sk_buff *tmp; |
| 1526 | |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1527 | tcp_for_write_queue_from(skb, sk) { |
| 1528 | int in_sack = 0; |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1529 | bool dup_sack = dup_sack_in; |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1530 | |
| 1531 | if (skb == tcp_send_head(sk)) |
| 1532 | break; |
| 1533 | |
| 1534 | /* queue is in-order => we can short-circuit the walk early */ |
| 1535 | if (!before(TCP_SKB_CB(skb)->seq, end_seq)) |
| 1536 | break; |
| 1537 | |
| 1538 | if ((next_dup != NULL) && |
| 1539 | before(TCP_SKB_CB(skb)->seq, next_dup->end_seq)) { |
| 1540 | in_sack = tcp_match_skb_to_sack(sk, skb, |
| 1541 | next_dup->start_seq, |
| 1542 | next_dup->end_seq); |
| 1543 | if (in_sack > 0) |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1544 | dup_sack = true; |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1545 | } |
| 1546 | |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1547 | /* skb reference here is a bit tricky to get right, since |
| 1548 | * shifting can eat and free both this skb and the next, |
| 1549 | * so not even _safe variant of the loop is enough. |
| 1550 | */ |
| 1551 | if (in_sack <= 0) { |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1552 | tmp = tcp_shift_skb_data(sk, skb, state, |
| 1553 | start_seq, end_seq, dup_sack); |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1554 | if (tmp != NULL) { |
| 1555 | if (tmp != skb) { |
| 1556 | skb = tmp; |
| 1557 | continue; |
| 1558 | } |
| 1559 | |
| 1560 | in_sack = 0; |
| 1561 | } else { |
| 1562 | in_sack = tcp_match_skb_to_sack(sk, skb, |
| 1563 | start_seq, |
| 1564 | end_seq); |
| 1565 | } |
| 1566 | } |
| 1567 | |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1568 | if (unlikely(in_sack < 0)) |
| 1569 | break; |
| 1570 | |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1571 | if (in_sack) { |
Neal Cardwell | cc9a672 | 2012-02-12 18:37:09 +0000 | [diff] [blame] | 1572 | TCP_SKB_CB(skb)->sacked = |
| 1573 | tcp_sacktag_one(sk, |
| 1574 | state, |
| 1575 | TCP_SKB_CB(skb)->sacked, |
| 1576 | TCP_SKB_CB(skb)->seq, |
| 1577 | TCP_SKB_CB(skb)->end_seq, |
| 1578 | dup_sack, |
Yuchung Cheng | 59c9af4 | 2013-07-22 16:20:47 -0700 | [diff] [blame] | 1579 | tcp_skb_pcount(skb), |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 1580 | &skb->skb_mstamp); |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1581 | |
Ilpo Järvinen | 832d11c | 2008-11-24 21:20:15 -0800 | [diff] [blame] | 1582 | if (!before(TCP_SKB_CB(skb)->seq, |
| 1583 | tcp_highest_sack_seq(tp))) |
| 1584 | tcp_advance_highest_sack(sk, skb); |
| 1585 | } |
| 1586 | |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1587 | state->fack_count += tcp_skb_pcount(skb); |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1588 | } |
| 1589 | return skb; |
| 1590 | } |
| 1591 | |
| 1592 | /* Avoid all extra work that is being done by sacktag while walking in |
| 1593 | * a normal way |
| 1594 | */ |
| 1595 | 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] | 1596 | struct tcp_sacktag_state *state, |
| 1597 | u32 skip_to_seq) |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1598 | { |
| 1599 | tcp_for_write_queue_from(skb, sk) { |
| 1600 | if (skb == tcp_send_head(sk)) |
| 1601 | break; |
| 1602 | |
Ilpo Järvinen | e8bae27 | 2008-11-24 21:12:28 -0800 | [diff] [blame] | 1603 | if (after(TCP_SKB_CB(skb)->end_seq, skip_to_seq)) |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1604 | break; |
Ilpo Järvinen | d152a7d | 2008-03-03 12:10:16 -0800 | [diff] [blame] | 1605 | |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1606 | state->fack_count += tcp_skb_pcount(skb); |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1607 | } |
| 1608 | return skb; |
| 1609 | } |
| 1610 | |
| 1611 | static struct sk_buff *tcp_maybe_skipping_dsack(struct sk_buff *skb, |
| 1612 | struct sock *sk, |
| 1613 | struct tcp_sack_block *next_dup, |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1614 | struct tcp_sacktag_state *state, |
| 1615 | u32 skip_to_seq) |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1616 | { |
| 1617 | if (next_dup == NULL) |
| 1618 | return skb; |
| 1619 | |
| 1620 | if (before(next_dup->start_seq, skip_to_seq)) { |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1621 | skb = tcp_sacktag_skip(skb, sk, state, next_dup->start_seq); |
| 1622 | skb = tcp_sacktag_walk(skb, sk, NULL, state, |
| 1623 | next_dup->start_seq, next_dup->end_seq, |
| 1624 | 1); |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1625 | } |
| 1626 | |
| 1627 | return skb; |
| 1628 | } |
| 1629 | |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 1630 | 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] | 1631 | { |
| 1632 | return cache < tp->recv_sack_cache + ARRAY_SIZE(tp->recv_sack_cache); |
| 1633 | } |
| 1634 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1635 | static int |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 1636 | tcp_sacktag_write_queue(struct sock *sk, const struct sk_buff *ack_skb, |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 1637 | u32 prior_snd_una, long *sack_rtt_us) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1638 | { |
| 1639 | struct tcp_sock *tp = tcp_sk(sk); |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 1640 | const unsigned char *ptr = (skb_transport_header(ack_skb) + |
| 1641 | TCP_SKB_CB(ack_skb)->sacked); |
Ilpo Järvinen | fd6dad6 | 2007-11-15 19:49:47 -0800 | [diff] [blame] | 1642 | 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] | 1643 | struct tcp_sack_block sp[TCP_NUM_SACKS]; |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1644 | struct tcp_sack_block *cache; |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1645 | struct tcp_sacktag_state state; |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1646 | struct sk_buff *skb; |
Adam Langley | 4389dde | 2008-07-19 00:07:02 -0700 | [diff] [blame] | 1647 | 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] | 1648 | int used_sacks; |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1649 | bool found_dup_sack = false; |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1650 | int i, j; |
Baruch Even | fda03fb | 2007-02-04 23:35:57 -0800 | [diff] [blame] | 1651 | int first_sack_index; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1652 | |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1653 | state.flag = 0; |
| 1654 | state.reord = tp->packets_out; |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 1655 | state.rtt_us = -1L; |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1656 | |
Ilpo Järvinen | d738cd8 | 2007-03-24 21:03:23 -0700 | [diff] [blame] | 1657 | if (!tp->sacked_out) { |
Ilpo Järvinen | de83c05 | 2007-10-07 23:37:55 -0700 | [diff] [blame] | 1658 | if (WARN_ON(tp->fackets_out)) |
| 1659 | tp->fackets_out = 0; |
Ilpo Järvinen | 6859d49 | 2007-12-02 00:48:06 +0200 | [diff] [blame] | 1660 | tcp_highest_sack_reset(sk); |
Ilpo Järvinen | d738cd8 | 2007-03-24 21:03:23 -0700 | [diff] [blame] | 1661 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1662 | |
Pavel Emelyanov | 1ed8346 | 2008-07-16 20:29:51 -0700 | [diff] [blame] | 1663 | found_dup_sack = tcp_check_dsack(sk, ack_skb, sp_wire, |
David S. Miller | d06e021 | 2007-06-18 22:43:06 -0700 | [diff] [blame] | 1664 | num_sacks, prior_snd_una); |
| 1665 | if (found_dup_sack) |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1666 | state.flag |= FLAG_DSACKING_ACK; |
Baruch Even | 6f74651 | 2007-02-04 23:36:42 -0800 | [diff] [blame] | 1667 | |
| 1668 | /* Eliminate too old ACKs, but take into |
| 1669 | * account more or less fresh ones, they can |
| 1670 | * contain valid SACK info. |
| 1671 | */ |
| 1672 | if (before(TCP_SKB_CB(ack_skb)->ack_seq, prior_snd_una - tp->max_window)) |
| 1673 | return 0; |
| 1674 | |
Ilpo Järvinen | 96a2d41 | 2007-11-14 15:47:18 -0800 | [diff] [blame] | 1675 | if (!tp->packets_out) |
| 1676 | goto out; |
| 1677 | |
Ilpo Järvinen | fd6dad6 | 2007-11-15 19:49:47 -0800 | [diff] [blame] | 1678 | used_sacks = 0; |
| 1679 | first_sack_index = 0; |
| 1680 | for (i = 0; i < num_sacks; i++) { |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1681 | bool dup_sack = !i && found_dup_sack; |
Ilpo Järvinen | fd6dad6 | 2007-11-15 19:49:47 -0800 | [diff] [blame] | 1682 | |
Harvey Harrison | d3e2ce3 | 2008-05-02 16:26:16 -0700 | [diff] [blame] | 1683 | sp[used_sacks].start_seq = get_unaligned_be32(&sp_wire[i].start_seq); |
| 1684 | 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] | 1685 | |
| 1686 | if (!tcp_is_sackblock_valid(tp, dup_sack, |
| 1687 | sp[used_sacks].start_seq, |
| 1688 | sp[used_sacks].end_seq)) { |
Pavel Emelyanov | 40b215e | 2008-07-03 01:05:41 -0700 | [diff] [blame] | 1689 | int mib_idx; |
| 1690 | |
Ilpo Järvinen | fd6dad6 | 2007-11-15 19:49:47 -0800 | [diff] [blame] | 1691 | if (dup_sack) { |
| 1692 | if (!tp->undo_marker) |
Pavel Emelyanov | 40b215e | 2008-07-03 01:05:41 -0700 | [diff] [blame] | 1693 | mib_idx = LINUX_MIB_TCPDSACKIGNOREDNOUNDO; |
Ilpo Järvinen | fd6dad6 | 2007-11-15 19:49:47 -0800 | [diff] [blame] | 1694 | else |
Pavel Emelyanov | 40b215e | 2008-07-03 01:05:41 -0700 | [diff] [blame] | 1695 | mib_idx = LINUX_MIB_TCPDSACKIGNOREDOLD; |
Ilpo Järvinen | fd6dad6 | 2007-11-15 19:49:47 -0800 | [diff] [blame] | 1696 | } else { |
| 1697 | /* Don't count olds caused by ACK reordering */ |
| 1698 | if ((TCP_SKB_CB(ack_skb)->ack_seq != tp->snd_una) && |
| 1699 | !after(sp[used_sacks].end_seq, tp->snd_una)) |
| 1700 | continue; |
Pavel Emelyanov | 40b215e | 2008-07-03 01:05:41 -0700 | [diff] [blame] | 1701 | mib_idx = LINUX_MIB_TCPSACKDISCARD; |
Ilpo Järvinen | fd6dad6 | 2007-11-15 19:49:47 -0800 | [diff] [blame] | 1702 | } |
Pavel Emelyanov | 40b215e | 2008-07-03 01:05:41 -0700 | [diff] [blame] | 1703 | |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 1704 | NET_INC_STATS_BH(sock_net(sk), mib_idx); |
Ilpo Järvinen | fd6dad6 | 2007-11-15 19:49:47 -0800 | [diff] [blame] | 1705 | if (i == 0) |
| 1706 | first_sack_index = -1; |
| 1707 | continue; |
| 1708 | } |
| 1709 | |
| 1710 | /* Ignore very old stuff early */ |
| 1711 | if (!after(sp[used_sacks].end_seq, prior_snd_una)) |
| 1712 | continue; |
| 1713 | |
| 1714 | used_sacks++; |
| 1715 | } |
| 1716 | |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1717 | /* order SACK blocks to allow in order walk of the retrans queue */ |
| 1718 | for (i = used_sacks - 1; i > 0; i--) { |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 1719 | for (j = 0; j < i; j++) { |
| 1720 | if (after(sp[j].start_seq, sp[j + 1].start_seq)) { |
Ilpo Järvinen | a0bffff | 2009-03-21 13:36:17 -0700 | [diff] [blame] | 1721 | swap(sp[j], sp[j + 1]); |
Stephen Hemminger | 6a438bb | 2005-11-10 17:14:59 -0800 | [diff] [blame] | 1722 | |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1723 | /* Track where the first SACK block goes to */ |
| 1724 | if (j == first_sack_index) |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 1725 | first_sack_index = j + 1; |
Stephen Hemminger | 6a438bb | 2005-11-10 17:14:59 -0800 | [diff] [blame] | 1726 | } |
| 1727 | } |
| 1728 | } |
| 1729 | |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1730 | skb = tcp_write_queue_head(sk); |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1731 | state.fack_count = 0; |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1732 | i = 0; |
| 1733 | |
| 1734 | if (!tp->sacked_out) { |
| 1735 | /* It's already past, so skip checking against it */ |
| 1736 | cache = tp->recv_sack_cache + ARRAY_SIZE(tp->recv_sack_cache); |
| 1737 | } else { |
| 1738 | cache = tp->recv_sack_cache; |
| 1739 | /* Skip empty blocks in at head of the cache */ |
| 1740 | while (tcp_sack_cache_ok(tp, cache) && !cache->start_seq && |
| 1741 | !cache->end_seq) |
| 1742 | cache++; |
Baruch Even | fda03fb | 2007-02-04 23:35:57 -0800 | [diff] [blame] | 1743 | } |
| 1744 | |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1745 | while (i < used_sacks) { |
Ilpo Järvinen | fd6dad6 | 2007-11-15 19:49:47 -0800 | [diff] [blame] | 1746 | u32 start_seq = sp[i].start_seq; |
| 1747 | u32 end_seq = sp[i].end_seq; |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1748 | bool dup_sack = (found_dup_sack && (i == first_sack_index)); |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1749 | struct tcp_sack_block *next_dup = NULL; |
Ilpo Järvinen | e56d6cd | 2007-11-01 00:09:37 -0700 | [diff] [blame] | 1750 | |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1751 | if (found_dup_sack && ((i + 1) == first_sack_index)) |
| 1752 | next_dup = &sp[i + 1]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1753 | |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1754 | /* Skip too early cached blocks */ |
| 1755 | while (tcp_sack_cache_ok(tp, cache) && |
| 1756 | !before(start_seq, cache->end_seq)) |
| 1757 | cache++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1758 | |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1759 | /* Can skip some work by looking recv_sack_cache? */ |
| 1760 | if (tcp_sack_cache_ok(tp, cache) && !dup_sack && |
| 1761 | after(end_seq, cache->start_seq)) { |
David S. Miller | fe067e8 | 2007-03-07 12:12:44 -0800 | [diff] [blame] | 1762 | |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1763 | /* Head todo? */ |
| 1764 | if (before(start_seq, cache->start_seq)) { |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1765 | skb = tcp_sacktag_skip(skb, sk, &state, |
| 1766 | start_seq); |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 1767 | skb = tcp_sacktag_walk(skb, sk, next_dup, |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1768 | &state, |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 1769 | start_seq, |
| 1770 | cache->start_seq, |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1771 | dup_sack); |
Baruch Even | fda03fb | 2007-02-04 23:35:57 -0800 | [diff] [blame] | 1772 | } |
Stephen Hemminger | 6a438bb | 2005-11-10 17:14:59 -0800 | [diff] [blame] | 1773 | |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1774 | /* Rest of the block already fully processed? */ |
Ilpo Järvinen | 20de20b | 2007-11-16 16:17:05 -0800 | [diff] [blame] | 1775 | if (!after(end_seq, cache->end_seq)) |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1776 | goto advance_sp; |
Ilpo Järvinen | 20de20b | 2007-11-16 16:17:05 -0800 | [diff] [blame] | 1777 | |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 1778 | skb = tcp_maybe_skipping_dsack(skb, sk, next_dup, |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1779 | &state, |
| 1780 | cache->end_seq); |
Ilpo Järvinen | e56d6cd | 2007-11-01 00:09:37 -0700 | [diff] [blame] | 1781 | |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1782 | /* ...tail remains todo... */ |
Ilpo Järvinen | 6859d49 | 2007-12-02 00:48:06 +0200 | [diff] [blame] | 1783 | if (tcp_highest_sack_seq(tp) == cache->end_seq) { |
Ilpo Järvinen | 20de20b | 2007-11-16 16:17:05 -0800 | [diff] [blame] | 1784 | /* ...but better entrypoint exists! */ |
Ilpo Järvinen | 6859d49 | 2007-12-02 00:48:06 +0200 | [diff] [blame] | 1785 | skb = tcp_highest_sack(sk); |
| 1786 | if (skb == NULL) |
| 1787 | break; |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1788 | state.fack_count = tp->fackets_out; |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1789 | cache++; |
| 1790 | goto walk; |
| 1791 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1792 | |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1793 | skb = tcp_sacktag_skip(skb, sk, &state, cache->end_seq); |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1794 | /* Check overlap against next cached too (past this one already) */ |
| 1795 | cache++; |
| 1796 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1797 | } |
Ilpo Järvinen | fbd52eb | 2007-11-10 21:24:19 -0800 | [diff] [blame] | 1798 | |
Ilpo Järvinen | 6859d49 | 2007-12-02 00:48:06 +0200 | [diff] [blame] | 1799 | if (!before(start_seq, tcp_highest_sack_seq(tp))) { |
| 1800 | skb = tcp_highest_sack(sk); |
| 1801 | if (skb == NULL) |
| 1802 | break; |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1803 | state.fack_count = tp->fackets_out; |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1804 | } |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1805 | skb = tcp_sacktag_skip(skb, sk, &state, start_seq); |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1806 | |
| 1807 | walk: |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1808 | skb = tcp_sacktag_walk(skb, sk, next_dup, &state, |
| 1809 | start_seq, end_seq, dup_sack); |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1810 | |
| 1811 | advance_sp: |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1812 | i++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1813 | } |
| 1814 | |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 1815 | /* Clear the head of the cache sack blocks so we can skip it next time */ |
| 1816 | for (i = 0; i < ARRAY_SIZE(tp->recv_sack_cache) - used_sacks; i++) { |
| 1817 | tp->recv_sack_cache[i].start_seq = 0; |
| 1818 | tp->recv_sack_cache[i].end_seq = 0; |
| 1819 | } |
| 1820 | for (j = 0; j < used_sacks; j++) |
| 1821 | tp->recv_sack_cache[i++] = sp[j]; |
| 1822 | |
Ilpo Järvinen | 407ef1d | 2007-12-02 00:47:57 +0200 | [diff] [blame] | 1823 | tcp_mark_lost_retrans(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1824 | |
Ilpo Järvinen | 86426c2 | 2007-08-09 14:45:17 +0300 | [diff] [blame] | 1825 | tcp_verify_left_out(tp); |
| 1826 | |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1827 | if ((state.reord < tp->fackets_out) && |
Yuchung Cheng | 9b44190 | 2013-03-20 13:32:58 +0000 | [diff] [blame] | 1828 | ((inet_csk(sk)->icsk_ca_state != TCP_CA_Loss) || tp->undo_marker)) |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1829 | tcp_update_reordering(sk, tp->fackets_out - state.reord, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1830 | |
Ilpo Järvinen | 96a2d41 | 2007-11-14 15:47:18 -0800 | [diff] [blame] | 1831 | out: |
| 1832 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1833 | #if FASTRETRANS_DEBUG > 0 |
Ilpo Järvinen | 547b792 | 2008-07-25 21:43:18 -0700 | [diff] [blame] | 1834 | WARN_ON((int)tp->sacked_out < 0); |
| 1835 | WARN_ON((int)tp->lost_out < 0); |
| 1836 | WARN_ON((int)tp->retrans_out < 0); |
| 1837 | WARN_ON((int)tcp_packets_in_flight(tp) < 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1838 | #endif |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 1839 | *sack_rtt_us = state.rtt_us; |
Ilpo Järvinen | a1197f5 | 2008-12-05 22:42:22 -0800 | [diff] [blame] | 1840 | return state.flag; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1841 | } |
| 1842 | |
Ilpo Järvinen | 882beba | 2008-04-07 22:33:07 -0700 | [diff] [blame] | 1843 | /* 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] | 1844 | * packets_out. Returns false if sacked_out adjustement wasn't necessary. |
Ilpo Järvinen | 30935cf | 2007-02-21 23:01:36 -0800 | [diff] [blame] | 1845 | */ |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1846 | static bool tcp_limit_reno_sacked(struct tcp_sock *tp) |
Ilpo Järvinen | 4ddf667 | 2007-05-27 01:52:00 -0700 | [diff] [blame] | 1847 | { |
Ilpo Järvinen | 4ddf667 | 2007-05-27 01:52:00 -0700 | [diff] [blame] | 1848 | u32 holes; |
| 1849 | |
| 1850 | holes = max(tp->lost_out, 1U); |
| 1851 | holes = min(holes, tp->packets_out); |
| 1852 | |
| 1853 | if ((tp->sacked_out + holes) > tp->packets_out) { |
| 1854 | tp->sacked_out = tp->packets_out - holes; |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1855 | return true; |
Ilpo Järvinen | 4ddf667 | 2007-05-27 01:52:00 -0700 | [diff] [blame] | 1856 | } |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 1857 | return false; |
Ilpo Järvinen | 882beba | 2008-04-07 22:33:07 -0700 | [diff] [blame] | 1858 | } |
| 1859 | |
| 1860 | /* If we receive more dupacks than we expected counting segments |
| 1861 | * in assumption of absent reordering, interpret this as reordering. |
| 1862 | * The only another reason could be bug in receiver TCP. |
| 1863 | */ |
| 1864 | static void tcp_check_reno_reordering(struct sock *sk, const int addend) |
| 1865 | { |
| 1866 | struct tcp_sock *tp = tcp_sk(sk); |
| 1867 | if (tcp_limit_reno_sacked(tp)) |
| 1868 | tcp_update_reordering(sk, tp->packets_out + addend, 0); |
Ilpo Järvinen | 4ddf667 | 2007-05-27 01:52:00 -0700 | [diff] [blame] | 1869 | } |
| 1870 | |
| 1871 | /* Emulate SACKs for SACKless connection: account for a new dupack. */ |
| 1872 | |
| 1873 | static void tcp_add_reno_sack(struct sock *sk) |
| 1874 | { |
| 1875 | struct tcp_sock *tp = tcp_sk(sk); |
| 1876 | tp->sacked_out++; |
| 1877 | tcp_check_reno_reordering(sk, 0); |
Ilpo Järvinen | 005903b | 2007-08-09 14:44:16 +0300 | [diff] [blame] | 1878 | tcp_verify_left_out(tp); |
Ilpo Järvinen | 4ddf667 | 2007-05-27 01:52:00 -0700 | [diff] [blame] | 1879 | } |
| 1880 | |
| 1881 | /* Account for ACK, ACKing some data in Reno Recovery phase. */ |
| 1882 | |
| 1883 | static void tcp_remove_reno_sacks(struct sock *sk, int acked) |
| 1884 | { |
| 1885 | struct tcp_sock *tp = tcp_sk(sk); |
| 1886 | |
| 1887 | if (acked > 0) { |
| 1888 | /* One ACK acked hole. The rest eat duplicate ACKs. */ |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 1889 | if (acked - 1 >= tp->sacked_out) |
Ilpo Järvinen | 4ddf667 | 2007-05-27 01:52:00 -0700 | [diff] [blame] | 1890 | tp->sacked_out = 0; |
| 1891 | else |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 1892 | tp->sacked_out -= acked - 1; |
Ilpo Järvinen | 4ddf667 | 2007-05-27 01:52:00 -0700 | [diff] [blame] | 1893 | } |
| 1894 | tcp_check_reno_reordering(sk, acked); |
Ilpo Järvinen | 005903b | 2007-08-09 14:44:16 +0300 | [diff] [blame] | 1895 | tcp_verify_left_out(tp); |
Ilpo Järvinen | 4ddf667 | 2007-05-27 01:52:00 -0700 | [diff] [blame] | 1896 | } |
| 1897 | |
| 1898 | static inline void tcp_reset_reno_sack(struct tcp_sock *tp) |
| 1899 | { |
| 1900 | tp->sacked_out = 0; |
Ilpo Järvinen | 4ddf667 | 2007-05-27 01:52:00 -0700 | [diff] [blame] | 1901 | } |
| 1902 | |
Yuchung Cheng | 989e04c | 2014-08-22 14:15:22 -0700 | [diff] [blame] | 1903 | void tcp_clear_retrans(struct tcp_sock *tp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1904 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1905 | tp->retrans_out = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1906 | tp->lost_out = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1907 | tp->undo_marker = 0; |
Yuchung Cheng | 6e08d5e | 2014-07-02 12:07:16 -0700 | [diff] [blame] | 1908 | tp->undo_retrans = -1; |
Ilpo Järvinen | 4cd8299 | 2007-10-11 17:34:57 -0700 | [diff] [blame] | 1909 | tp->fackets_out = 0; |
| 1910 | tp->sacked_out = 0; |
| 1911 | } |
| 1912 | |
Yuchung Cheng | 989e04c | 2014-08-22 14:15:22 -0700 | [diff] [blame] | 1913 | static inline void tcp_init_undo(struct tcp_sock *tp) |
| 1914 | { |
| 1915 | tp->undo_marker = tp->snd_una; |
| 1916 | /* Retransmission still in flight may cause DSACKs later. */ |
| 1917 | tp->undo_retrans = tp->retrans_out ? : -1; |
| 1918 | } |
| 1919 | |
Neal Cardwell | 5ae344c | 2014-08-04 19:12:29 -0400 | [diff] [blame] | 1920 | /* Enter Loss state. If we detect SACK reneging, forget all SACK information |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1921 | * and reset tags completely, otherwise preserve SACKs. If receiver |
| 1922 | * dropped its ofo queue, we will know this due to reneging detection. |
| 1923 | */ |
Neal Cardwell | 5ae344c | 2014-08-04 19:12:29 -0400 | [diff] [blame] | 1924 | void tcp_enter_loss(struct sock *sk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1925 | { |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 1926 | const struct inet_connection_sock *icsk = inet_csk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1927 | struct tcp_sock *tp = tcp_sk(sk); |
| 1928 | struct sk_buff *skb; |
Yuchung Cheng | e33099f | 2013-03-20 13:33:00 +0000 | [diff] [blame] | 1929 | bool new_recovery = false; |
Neal Cardwell | 5ae344c | 2014-08-04 19:12:29 -0400 | [diff] [blame] | 1930 | bool is_reneg; /* is receiver reneging on SACKs? */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1931 | |
| 1932 | /* Reduce ssthresh if it has not yet been made inside this window. */ |
Yuchung Cheng | e33099f | 2013-03-20 13:33:00 +0000 | [diff] [blame] | 1933 | if (icsk->icsk_ca_state <= TCP_CA_Disorder || |
| 1934 | !after(tp->high_seq, tp->snd_una) || |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 1935 | (icsk->icsk_ca_state == TCP_CA_Loss && !icsk->icsk_retransmits)) { |
Yuchung Cheng | e33099f | 2013-03-20 13:33:00 +0000 | [diff] [blame] | 1936 | new_recovery = true; |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 1937 | tp->prior_ssthresh = tcp_current_ssthresh(sk); |
| 1938 | tp->snd_ssthresh = icsk->icsk_ca_ops->ssthresh(sk); |
| 1939 | tcp_ca_event(sk, CA_EVENT_LOSS); |
Yuchung Cheng | 989e04c | 2014-08-22 14:15:22 -0700 | [diff] [blame] | 1940 | tcp_init_undo(tp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1941 | } |
| 1942 | tp->snd_cwnd = 1; |
| 1943 | tp->snd_cwnd_cnt = 0; |
| 1944 | tp->snd_cwnd_stamp = tcp_time_stamp; |
| 1945 | |
Yuchung Cheng | 989e04c | 2014-08-22 14:15:22 -0700 | [diff] [blame] | 1946 | tp->retrans_out = 0; |
| 1947 | tp->lost_out = 0; |
Ilpo Järvinen | 4cd8299 | 2007-10-11 17:34:57 -0700 | [diff] [blame] | 1948 | |
| 1949 | if (tcp_is_reno(tp)) |
| 1950 | tcp_reset_reno_sack(tp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1951 | |
Neal Cardwell | 5ae344c | 2014-08-04 19:12:29 -0400 | [diff] [blame] | 1952 | skb = tcp_write_queue_head(sk); |
| 1953 | is_reneg = skb && (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED); |
| 1954 | if (is_reneg) { |
| 1955 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPSACKRENEGING); |
Ilpo Järvinen | 4cd8299 | 2007-10-11 17:34:57 -0700 | [diff] [blame] | 1956 | tp->sacked_out = 0; |
| 1957 | tp->fackets_out = 0; |
Ilpo Järvinen | b768920 | 2007-09-20 11:37:19 -0700 | [diff] [blame] | 1958 | } |
Ilpo Järvinen | 64edc27 | 2008-09-20 21:18:32 -0700 | [diff] [blame] | 1959 | tcp_clear_all_retrans_hints(tp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1960 | |
David S. Miller | fe067e8 | 2007-03-07 12:12:44 -0800 | [diff] [blame] | 1961 | tcp_for_write_queue(skb, sk) { |
| 1962 | if (skb == tcp_send_head(sk)) |
| 1963 | break; |
Ilpo Järvinen | 4cd8299 | 2007-10-11 17:34:57 -0700 | [diff] [blame] | 1964 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1965 | TCP_SKB_CB(skb)->sacked &= (~TCPCB_TAGBITS)|TCPCB_SACKED_ACKED; |
Neal Cardwell | 5ae344c | 2014-08-04 19:12:29 -0400 | [diff] [blame] | 1966 | if (!(TCP_SKB_CB(skb)->sacked&TCPCB_SACKED_ACKED) || is_reneg) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1967 | TCP_SKB_CB(skb)->sacked &= ~TCPCB_SACKED_ACKED; |
| 1968 | TCP_SKB_CB(skb)->sacked |= TCPCB_LOST; |
| 1969 | tp->lost_out += tcp_skb_pcount(skb); |
Ilpo Järvinen | 006f582 | 2008-09-20 21:20:20 -0700 | [diff] [blame] | 1970 | tp->retransmit_high = TCP_SKB_CB(skb)->end_seq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1971 | } |
| 1972 | } |
Ilpo Järvinen | 005903b | 2007-08-09 14:44:16 +0300 | [diff] [blame] | 1973 | tcp_verify_left_out(tp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1974 | |
Yuchung Cheng | 74c181d | 2013-08-12 16:41:25 -0700 | [diff] [blame] | 1975 | /* Timeout in disordered state after receiving substantial DUPACKs |
| 1976 | * suggests that the degree of reordering is over-estimated. |
| 1977 | */ |
| 1978 | if (icsk->icsk_ca_state <= TCP_CA_Disorder && |
| 1979 | tp->sacked_out >= sysctl_tcp_reordering) |
| 1980 | tp->reordering = min_t(unsigned int, tp->reordering, |
| 1981 | sysctl_tcp_reordering); |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 1982 | tcp_set_ca_state(sk, TCP_CA_Loss); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1983 | tp->high_seq = tp->snd_nxt; |
Florian Westphal | 735d383 | 2014-09-29 13:08:30 +0200 | [diff] [blame] | 1984 | tcp_ecn_queue_cwr(tp); |
Yuchung Cheng | e33099f | 2013-03-20 13:33:00 +0000 | [diff] [blame] | 1985 | |
| 1986 | /* F-RTO RFC5682 sec 3.1 step 1: retransmit SND.UNA if no previous |
| 1987 | * loss recovery is underway except recurring timeout(s) on |
| 1988 | * the same SND.UNA (sec 3.2). Disable F-RTO on path MTU probing |
| 1989 | */ |
| 1990 | tp->frto = sysctl_tcp_frto && |
| 1991 | (new_recovery || icsk->icsk_retransmits) && |
| 1992 | !inet_csk(sk)->icsk_mtup.probe_size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1993 | } |
| 1994 | |
Ilpo Järvinen | cadbd03 | 2007-12-31 04:49:21 -0800 | [diff] [blame] | 1995 | /* If ACK arrived pointing to a remembered SACK, it means that our |
| 1996 | * remembered SACKs do not reflect real state of receiver i.e. |
| 1997 | * receiver _host_ is heavily congested (or buggy). |
| 1998 | * |
Neal Cardwell | 5ae344c | 2014-08-04 19:12:29 -0400 | [diff] [blame] | 1999 | * To avoid big spurious retransmission bursts due to transient SACK |
| 2000 | * scoreboard oddities that look like reneging, we give the receiver a |
| 2001 | * little time (max(RTT/2, 10ms)) to send us some more ACKs that will |
| 2002 | * restore sanity to the SACK scoreboard. If the apparent reneging |
| 2003 | * persists until this RTO then we'll clear the SACK scoreboard. |
Ilpo Järvinen | cadbd03 | 2007-12-31 04:49:21 -0800 | [diff] [blame] | 2004 | */ |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 2005 | static bool tcp_check_sack_reneging(struct sock *sk, int flag) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2006 | { |
Ilpo Järvinen | cadbd03 | 2007-12-31 04:49:21 -0800 | [diff] [blame] | 2007 | if (flag & FLAG_SACK_RENEGING) { |
Neal Cardwell | 5ae344c | 2014-08-04 19:12:29 -0400 | [diff] [blame] | 2008 | struct tcp_sock *tp = tcp_sk(sk); |
| 2009 | unsigned long delay = max(usecs_to_jiffies(tp->srtt_us >> 4), |
| 2010 | msecs_to_jiffies(10)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2011 | |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 2012 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, |
Neal Cardwell | 5ae344c | 2014-08-04 19:12:29 -0400 | [diff] [blame] | 2013 | delay, TCP_RTO_MAX); |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 2014 | return true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2015 | } |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 2016 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2017 | } |
| 2018 | |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 2019 | static inline int tcp_fackets_out(const struct tcp_sock *tp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2020 | { |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 2021 | return tcp_is_reno(tp) ? tp->sacked_out + 1 : tp->fackets_out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2022 | } |
| 2023 | |
Ilpo Järvinen | 85cc391 | 2007-11-15 19:39:31 -0800 | [diff] [blame] | 2024 | /* Heurestics to calculate number of duplicate ACKs. There's no dupACKs |
| 2025 | * counter when SACK is enabled (without SACK, sacked_out is used for |
| 2026 | * that purpose). |
| 2027 | * |
| 2028 | * Instead, with FACK TCP uses fackets_out that includes both SACKed |
| 2029 | * segments up to the highest received SACK block so far and holes in |
| 2030 | * between them. |
| 2031 | * |
| 2032 | * With reordering, holes may still be in flight, so RFC3517 recovery |
| 2033 | * uses pure sacked_out (total number of SACKed segments) even though |
| 2034 | * it violates the RFC that uses duplicate ACKs, often these are equal |
| 2035 | * but when e.g. out-of-window ACKs or packet duplication occurs, |
| 2036 | * they differ. Since neither occurs due to loss, TCP should really |
| 2037 | * ignore them. |
| 2038 | */ |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 2039 | static inline int tcp_dupack_heuristics(const struct tcp_sock *tp) |
Ilpo Järvinen | 85cc391 | 2007-11-15 19:39:31 -0800 | [diff] [blame] | 2040 | { |
| 2041 | return tcp_is_fack(tp) ? tp->fackets_out : tp->sacked_out + 1; |
| 2042 | } |
| 2043 | |
Yuchung Cheng | 750ea2b | 2012-05-02 13:30:04 +0000 | [diff] [blame] | 2044 | static bool tcp_pause_early_retransmit(struct sock *sk, int flag) |
| 2045 | { |
| 2046 | struct tcp_sock *tp = tcp_sk(sk); |
| 2047 | unsigned long delay; |
| 2048 | |
| 2049 | /* Delay early retransmit and entering fast recovery for |
| 2050 | * max(RTT/4, 2msec) unless ack has ECE mark, no RTT samples |
| 2051 | * available, or RTO is scheduled to fire first. |
| 2052 | */ |
Nandita Dukkipati | 6ba8a3b | 2013-03-11 10:00:43 +0000 | [diff] [blame] | 2053 | if (sysctl_tcp_early_retrans < 2 || sysctl_tcp_early_retrans > 3 || |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 2054 | (flag & FLAG_ECE) || !tp->srtt_us) |
Yuchung Cheng | 750ea2b | 2012-05-02 13:30:04 +0000 | [diff] [blame] | 2055 | return false; |
| 2056 | |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 2057 | delay = max(usecs_to_jiffies(tp->srtt_us >> 5), |
| 2058 | msecs_to_jiffies(2)); |
| 2059 | |
Yuchung Cheng | 750ea2b | 2012-05-02 13:30:04 +0000 | [diff] [blame] | 2060 | if (!time_after(inet_csk(sk)->icsk_timeout, (jiffies + delay))) |
| 2061 | return false; |
| 2062 | |
Nandita Dukkipati | 6ba8a3b | 2013-03-11 10:00:43 +0000 | [diff] [blame] | 2063 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_EARLY_RETRANS, delay, |
| 2064 | TCP_RTO_MAX); |
Yuchung Cheng | 750ea2b | 2012-05-02 13:30:04 +0000 | [diff] [blame] | 2065 | return true; |
| 2066 | } |
| 2067 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2068 | /* Linux NewReno/SACK/FACK/ECN state machine. |
| 2069 | * -------------------------------------- |
| 2070 | * |
| 2071 | * "Open" Normal state, no dubious events, fast path. |
| 2072 | * "Disorder" In all the respects it is "Open", |
| 2073 | * but requires a bit more attention. It is entered when |
| 2074 | * we see some SACKs or dupacks. It is split of "Open" |
| 2075 | * mainly to move some processing from fast path to slow one. |
| 2076 | * "CWR" CWND was reduced due to some Congestion Notification event. |
| 2077 | * It can be ECN, ICMP source quench, local device congestion. |
| 2078 | * "Recovery" CWND was reduced, we are fast-retransmitting. |
| 2079 | * "Loss" CWND was reduced due to RTO timeout or SACK reneging. |
| 2080 | * |
| 2081 | * tcp_fastretrans_alert() is entered: |
| 2082 | * - each incoming ACK, if state is not "Open" |
| 2083 | * - when arrived ACK is unusual, namely: |
| 2084 | * * SACK |
| 2085 | * * Duplicate ACK. |
| 2086 | * * ECN ECE. |
| 2087 | * |
| 2088 | * Counting packets in flight is pretty simple. |
| 2089 | * |
| 2090 | * in_flight = packets_out - left_out + retrans_out |
| 2091 | * |
| 2092 | * packets_out is SND.NXT-SND.UNA counted in packets. |
| 2093 | * |
| 2094 | * retrans_out is number of retransmitted segments. |
| 2095 | * |
| 2096 | * left_out is number of segments left network, but not ACKed yet. |
| 2097 | * |
| 2098 | * left_out = sacked_out + lost_out |
| 2099 | * |
| 2100 | * sacked_out: Packets, which arrived to receiver out of order |
| 2101 | * and hence not ACKed. With SACKs this number is simply |
| 2102 | * amount of SACKed data. Even without SACKs |
| 2103 | * it is easy to give pretty reliable estimate of this number, |
| 2104 | * counting duplicate ACKs. |
| 2105 | * |
| 2106 | * lost_out: Packets lost by network. TCP has no explicit |
| 2107 | * "loss notification" feedback from network (for now). |
| 2108 | * It means that this number can be only _guessed_. |
| 2109 | * Actually, it is the heuristics to predict lossage that |
| 2110 | * distinguishes different algorithms. |
| 2111 | * |
| 2112 | * F.e. after RTO, when all the queue is considered as lost, |
| 2113 | * lost_out = packets_out and in_flight = retrans_out. |
| 2114 | * |
| 2115 | * Essentially, we have now two algorithms counting |
| 2116 | * lost packets. |
| 2117 | * |
| 2118 | * FACK: It is the simplest heuristics. As soon as we decided |
| 2119 | * that something is lost, we decide that _all_ not SACKed |
| 2120 | * packets until the most forward SACK are lost. I.e. |
| 2121 | * lost_out = fackets_out - sacked_out and left_out = fackets_out. |
| 2122 | * It is absolutely correct estimate, if network does not reorder |
| 2123 | * packets. And it loses any connection to reality when reordering |
| 2124 | * takes place. We use FACK by default until reordering |
| 2125 | * is suspected on the path to this destination. |
| 2126 | * |
| 2127 | * NewReno: when Recovery is entered, we assume that one segment |
| 2128 | * is lost (classic Reno). While we are in Recovery and |
| 2129 | * a partial ACK arrives, we assume that one more packet |
| 2130 | * is lost (NewReno). This heuristics are the same in NewReno |
| 2131 | * and SACK. |
| 2132 | * |
| 2133 | * Imagine, that's all! Forget about all this shamanism about CWND inflation |
| 2134 | * deflation etc. CWND is real congestion window, never inflated, changes |
| 2135 | * only according to classic VJ rules. |
| 2136 | * |
| 2137 | * Really tricky (and requiring careful tuning) part of algorithm |
| 2138 | * is hidden in functions tcp_time_to_recover() and tcp_xmit_retransmit_queue(). |
| 2139 | * The first determines the moment _when_ we should reduce CWND and, |
| 2140 | * hence, slow down forward transmission. In fact, it determines the moment |
| 2141 | * when we decide that hole is caused by loss, rather than by a reorder. |
| 2142 | * |
| 2143 | * tcp_xmit_retransmit_queue() decides, _what_ we should retransmit to fill |
| 2144 | * holes, caused by lost packets. |
| 2145 | * |
| 2146 | * And the most logically complicated part of algorithm is undo |
| 2147 | * heuristics. We detect false retransmits due to both too early |
| 2148 | * fast retransmit (reordering) and underestimated RTO, analyzing |
| 2149 | * timestamps and D-SACKs. When we detect that some segments were |
| 2150 | * retransmitted by mistake and CWND reduction was wrong, we undo |
| 2151 | * window reduction and abort recovery phase. This logic is hidden |
| 2152 | * inside several functions named tcp_try_undo_<something>. |
| 2153 | */ |
| 2154 | |
| 2155 | /* This function decides, when we should leave Disordered state |
| 2156 | * and enter Recovery phase, reducing congestion window. |
| 2157 | * |
| 2158 | * Main question: may we further continue forward transmission |
| 2159 | * with the same cwnd? |
| 2160 | */ |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 2161 | static bool tcp_time_to_recover(struct sock *sk, int flag) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2162 | { |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 2163 | struct tcp_sock *tp = tcp_sk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2164 | __u32 packets_out; |
| 2165 | |
| 2166 | /* Trick#1: The loss is proven. */ |
| 2167 | if (tp->lost_out) |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 2168 | return true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2169 | |
| 2170 | /* Not-A-Trick#2 : Classic rule... */ |
Andreas Petlund | ea84e55 | 2009-10-27 03:27:21 +0000 | [diff] [blame] | 2171 | if (tcp_dupack_heuristics(tp) > tp->reordering) |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 2172 | return true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2173 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2174 | /* Trick#4: It is still not OK... But will it be useful to delay |
| 2175 | * recovery more? |
| 2176 | */ |
| 2177 | packets_out = tp->packets_out; |
| 2178 | if (packets_out <= tp->reordering && |
| 2179 | tp->sacked_out >= max_t(__u32, packets_out/2, sysctl_tcp_reordering) && |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 2180 | !tcp_may_send_now(sk)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2181 | /* We have nothing to send. This connection is limited |
| 2182 | * either by receiver window or by application. |
| 2183 | */ |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 2184 | return true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2185 | } |
| 2186 | |
Andreas Petlund | 7e38017 | 2010-02-18 04:48:19 +0000 | [diff] [blame] | 2187 | /* If a thin stream is detected, retransmit after first |
| 2188 | * received dupack. Employ only if SACK is supported in order |
| 2189 | * to avoid possible corner-case series of spurious retransmissions |
| 2190 | * Use only if there are no unsent data. |
| 2191 | */ |
| 2192 | if ((tp->thin_dupack || sysctl_tcp_thin_dupack) && |
| 2193 | tcp_stream_is_thin(tp) && tcp_dupack_heuristics(tp) > 1 && |
| 2194 | tcp_is_sack(tp) && !tcp_send_head(sk)) |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 2195 | return true; |
Andreas Petlund | 7e38017 | 2010-02-18 04:48:19 +0000 | [diff] [blame] | 2196 | |
Yuchung Cheng | eed530b | 2012-05-02 13:30:03 +0000 | [diff] [blame] | 2197 | /* Trick#6: TCP early retransmit, per RFC5827. To avoid spurious |
| 2198 | * retransmissions due to small network reorderings, we implement |
| 2199 | * Mitigation A.3 in the RFC and delay the retransmission for a short |
| 2200 | * interval if appropriate. |
| 2201 | */ |
| 2202 | if (tp->do_early_retrans && !tp->retrans_out && tp->sacked_out && |
Nandita Dukkipati | 6ba8a3b | 2013-03-11 10:00:43 +0000 | [diff] [blame] | 2203 | (tp->packets_out >= (tp->sacked_out + 1) && tp->packets_out < 4) && |
Yuchung Cheng | eed530b | 2012-05-02 13:30:03 +0000 | [diff] [blame] | 2204 | !tcp_may_send_now(sk)) |
Yuchung Cheng | 750ea2b | 2012-05-02 13:30:04 +0000 | [diff] [blame] | 2205 | return !tcp_pause_early_retransmit(sk, flag); |
Yuchung Cheng | eed530b | 2012-05-02 13:30:03 +0000 | [diff] [blame] | 2206 | |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 2207 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2208 | } |
| 2209 | |
Yuchung Cheng | 974c123 | 2012-01-19 14:42:21 +0000 | [diff] [blame] | 2210 | /* Detect loss in event "A" above by marking head of queue up as lost. |
| 2211 | * For FACK or non-SACK(Reno) senders, the first "packets" number of segments |
| 2212 | * are considered lost. For RFC3517 SACK, a segment is considered lost if it |
| 2213 | * has at least tp->reordering SACKed seqments above it; "packets" refers to |
| 2214 | * the maximum SACKed segments to pass before reaching this limit. |
Ilpo Järvinen | 85cc391 | 2007-11-15 19:39:31 -0800 | [diff] [blame] | 2215 | */ |
Ilpo Järvinen | 1fdb936 | 2010-10-14 01:42:30 +0000 | [diff] [blame] | 2216 | 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] | 2217 | { |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 2218 | struct tcp_sock *tp = tcp_sk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2219 | struct sk_buff *skb; |
Ilpo Järvinen | c137f3d | 2008-04-07 22:32:38 -0700 | [diff] [blame] | 2220 | int cnt, oldcnt; |
| 2221 | int err; |
| 2222 | unsigned int mss; |
Yuchung Cheng | 974c123 | 2012-01-19 14:42:21 +0000 | [diff] [blame] | 2223 | /* Use SACK to deduce losses of new sequences sent during recovery */ |
| 2224 | 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] | 2225 | |
Ilpo Järvinen | 547b792 | 2008-07-25 21:43:18 -0700 | [diff] [blame] | 2226 | WARN_ON(packets > tp->packets_out); |
Stephen Hemminger | 6a438bb | 2005-11-10 17:14:59 -0800 | [diff] [blame] | 2227 | if (tp->lost_skb_hint) { |
| 2228 | skb = tp->lost_skb_hint; |
| 2229 | cnt = tp->lost_cnt_hint; |
Ilpo Järvinen | 1fdb936 | 2010-10-14 01:42:30 +0000 | [diff] [blame] | 2230 | /* Head already handled? */ |
| 2231 | if (mark_head && skb != tcp_write_queue_head(sk)) |
| 2232 | return; |
Stephen Hemminger | 6a438bb | 2005-11-10 17:14:59 -0800 | [diff] [blame] | 2233 | } else { |
David S. Miller | fe067e8 | 2007-03-07 12:12:44 -0800 | [diff] [blame] | 2234 | skb = tcp_write_queue_head(sk); |
Stephen Hemminger | 6a438bb | 2005-11-10 17:14:59 -0800 | [diff] [blame] | 2235 | cnt = 0; |
| 2236 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2237 | |
David S. Miller | fe067e8 | 2007-03-07 12:12:44 -0800 | [diff] [blame] | 2238 | tcp_for_write_queue_from(skb, sk) { |
| 2239 | if (skb == tcp_send_head(sk)) |
| 2240 | break; |
Stephen Hemminger | 6a438bb | 2005-11-10 17:14:59 -0800 | [diff] [blame] | 2241 | /* TODO: do this better */ |
| 2242 | /* this is not the most efficient way to do this... */ |
| 2243 | tp->lost_skb_hint = skb; |
| 2244 | tp->lost_cnt_hint = cnt; |
Ilpo Järvinen | 85cc391 | 2007-11-15 19:39:31 -0800 | [diff] [blame] | 2245 | |
Yuchung Cheng | 974c123 | 2012-01-19 14:42:21 +0000 | [diff] [blame] | 2246 | if (after(TCP_SKB_CB(skb)->end_seq, loss_high)) |
Ilpo Järvinen | c137f3d | 2008-04-07 22:32:38 -0700 | [diff] [blame] | 2247 | break; |
| 2248 | |
| 2249 | oldcnt = cnt; |
Ilpo Järvinen | ad1984e | 2008-01-30 20:06:02 -0800 | [diff] [blame] | 2250 | if (tcp_is_fack(tp) || tcp_is_reno(tp) || |
Ilpo Järvinen | 85cc391 | 2007-11-15 19:39:31 -0800 | [diff] [blame] | 2251 | (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)) |
| 2252 | cnt += tcp_skb_pcount(skb); |
| 2253 | |
Ilpo Järvinen | c137f3d | 2008-04-07 22:32:38 -0700 | [diff] [blame] | 2254 | if (cnt > packets) { |
Yuchung Cheng | b3de755 | 2010-09-24 13:22:06 +0000 | [diff] [blame] | 2255 | if ((tcp_is_sack(tp) && !tcp_is_fack(tp)) || |
Neal Cardwell | c0638c2 | 2012-03-02 21:36:51 +0000 | [diff] [blame] | 2256 | (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED) || |
Yuchung Cheng | b3de755 | 2010-09-24 13:22:06 +0000 | [diff] [blame] | 2257 | (oldcnt >= packets)) |
Ilpo Järvinen | c137f3d | 2008-04-07 22:32:38 -0700 | [diff] [blame] | 2258 | break; |
| 2259 | |
| 2260 | mss = skb_shinfo(skb)->gso_size; |
Octavian Purdila | 6cc55e0 | 2014-06-06 17:32:37 +0300 | [diff] [blame] | 2261 | err = tcp_fragment(sk, skb, (packets - oldcnt) * mss, |
| 2262 | mss, GFP_ATOMIC); |
Ilpo Järvinen | c137f3d | 2008-04-07 22:32:38 -0700 | [diff] [blame] | 2263 | if (err < 0) |
| 2264 | break; |
| 2265 | cnt = packets; |
| 2266 | } |
| 2267 | |
Ilpo Järvinen | 41ea36e | 2008-09-20 21:19:22 -0700 | [diff] [blame] | 2268 | tcp_skb_mark_lost(tp, skb); |
Ilpo Järvinen | 1fdb936 | 2010-10-14 01:42:30 +0000 | [diff] [blame] | 2269 | |
| 2270 | if (mark_head) |
| 2271 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2272 | } |
Ilpo Järvinen | 005903b | 2007-08-09 14:44:16 +0300 | [diff] [blame] | 2273 | tcp_verify_left_out(tp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2274 | } |
| 2275 | |
| 2276 | /* Account newly detected lost packet(s) */ |
| 2277 | |
Ilpo Järvinen | 85cc391 | 2007-11-15 19:39:31 -0800 | [diff] [blame] | 2278 | static void tcp_update_scoreboard(struct sock *sk, int fast_rexmit) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2279 | { |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 2280 | struct tcp_sock *tp = tcp_sk(sk); |
| 2281 | |
Ilpo Järvinen | 85cc391 | 2007-11-15 19:39:31 -0800 | [diff] [blame] | 2282 | if (tcp_is_reno(tp)) { |
Ilpo Järvinen | 1fdb936 | 2010-10-14 01:42:30 +0000 | [diff] [blame] | 2283 | tcp_mark_head_lost(sk, 1, 1); |
Ilpo Järvinen | 85cc391 | 2007-11-15 19:39:31 -0800 | [diff] [blame] | 2284 | } else if (tcp_is_fack(tp)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2285 | int lost = tp->fackets_out - tp->reordering; |
| 2286 | if (lost <= 0) |
| 2287 | lost = 1; |
Ilpo Järvinen | 1fdb936 | 2010-10-14 01:42:30 +0000 | [diff] [blame] | 2288 | tcp_mark_head_lost(sk, lost, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2289 | } else { |
Ilpo Järvinen | 85cc391 | 2007-11-15 19:39:31 -0800 | [diff] [blame] | 2290 | int sacked_upto = tp->sacked_out - tp->reordering; |
Ilpo Järvinen | 1fdb936 | 2010-10-14 01:42:30 +0000 | [diff] [blame] | 2291 | if (sacked_upto >= 0) |
| 2292 | tcp_mark_head_lost(sk, sacked_upto, 0); |
| 2293 | else if (fast_rexmit) |
| 2294 | tcp_mark_head_lost(sk, 1, 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2295 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2296 | } |
| 2297 | |
| 2298 | /* CWND moderation, preventing bursts due to too big ACKs |
| 2299 | * in dubious situations. |
| 2300 | */ |
| 2301 | static inline void tcp_moderate_cwnd(struct tcp_sock *tp) |
| 2302 | { |
| 2303 | tp->snd_cwnd = min(tp->snd_cwnd, |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 2304 | tcp_packets_in_flight(tp) + tcp_max_burst(tp)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2305 | tp->snd_cwnd_stamp = tcp_time_stamp; |
| 2306 | } |
| 2307 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2308 | /* Nothing was retransmitted or returned timestamp is less |
| 2309 | * than timestamp of the first retransmission. |
| 2310 | */ |
Vijay Subramanian | 67b95bd | 2012-07-19 21:32:18 +0000 | [diff] [blame] | 2311 | static inline bool tcp_packet_delayed(const struct tcp_sock *tp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2312 | { |
| 2313 | return !tp->retrans_stamp || |
| 2314 | (tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr && |
Arnd Hannemann | d7ee147 | 2008-04-21 14:46:22 -0700 | [diff] [blame] | 2315 | before(tp->rx_opt.rcv_tsecr, tp->retrans_stamp)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2316 | } |
| 2317 | |
| 2318 | /* Undo procedures. */ |
| 2319 | |
Marcelo Leitner | 1f37bf8 | 2014-11-04 17:15:08 -0200 | [diff] [blame] | 2320 | /* We can clear retrans_stamp when there are no retransmissions in the |
| 2321 | * window. It would seem that it is trivially available for us in |
| 2322 | * tp->retrans_out, however, that kind of assumptions doesn't consider |
| 2323 | * what will happen if errors occur when sending retransmission for the |
| 2324 | * second time. ...It could the that such segment has only |
| 2325 | * TCPCB_EVER_RETRANS set at the present time. It seems that checking |
| 2326 | * the head skb is enough except for some reneging corner cases that |
| 2327 | * are not worth the effort. |
| 2328 | * |
| 2329 | * Main reason for all this complexity is the fact that connection dying |
| 2330 | * time now depends on the validity of the retrans_stamp, in particular, |
| 2331 | * that successive retransmissions of a segment must not advance |
| 2332 | * retrans_stamp under any conditions. |
| 2333 | */ |
| 2334 | static bool tcp_any_retrans_done(const struct sock *sk) |
| 2335 | { |
| 2336 | const struct tcp_sock *tp = tcp_sk(sk); |
| 2337 | struct sk_buff *skb; |
| 2338 | |
| 2339 | if (tp->retrans_out) |
| 2340 | return true; |
| 2341 | |
| 2342 | skb = tcp_write_queue_head(sk); |
| 2343 | if (unlikely(skb && TCP_SKB_CB(skb)->sacked & TCPCB_EVER_RETRANS)) |
| 2344 | return true; |
| 2345 | |
| 2346 | return false; |
| 2347 | } |
| 2348 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2349 | #if FASTRETRANS_DEBUG > 1 |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 2350 | static void DBGUNDO(struct sock *sk, const char *msg) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2351 | { |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 2352 | struct tcp_sock *tp = tcp_sk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2353 | struct inet_sock *inet = inet_sk(sk); |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 2354 | |
YOSHIFUJI Hideaki | 569508c | 2008-04-14 04:09:36 -0700 | [diff] [blame] | 2355 | if (sk->sk_family == AF_INET) { |
Joe Perches | 91df42b | 2012-05-15 14:11:54 +0000 | [diff] [blame] | 2356 | pr_debug("Undo %s %pI4/%u c%u l%u ss%u/%u p%u\n", |
| 2357 | msg, |
| 2358 | &inet->inet_daddr, ntohs(inet->inet_dport), |
| 2359 | tp->snd_cwnd, tcp_left_out(tp), |
| 2360 | tp->snd_ssthresh, tp->prior_ssthresh, |
| 2361 | tp->packets_out); |
YOSHIFUJI Hideaki | 569508c | 2008-04-14 04:09:36 -0700 | [diff] [blame] | 2362 | } |
Eric Dumazet | dfd56b8 | 2011-12-10 09:48:31 +0000 | [diff] [blame] | 2363 | #if IS_ENABLED(CONFIG_IPV6) |
YOSHIFUJI Hideaki | 569508c | 2008-04-14 04:09:36 -0700 | [diff] [blame] | 2364 | else if (sk->sk_family == AF_INET6) { |
| 2365 | struct ipv6_pinfo *np = inet6_sk(sk); |
Joe Perches | 91df42b | 2012-05-15 14:11:54 +0000 | [diff] [blame] | 2366 | pr_debug("Undo %s %pI6/%u c%u l%u ss%u/%u p%u\n", |
| 2367 | msg, |
| 2368 | &np->daddr, ntohs(inet->inet_dport), |
| 2369 | tp->snd_cwnd, tcp_left_out(tp), |
| 2370 | tp->snd_ssthresh, tp->prior_ssthresh, |
| 2371 | tp->packets_out); |
YOSHIFUJI Hideaki | 569508c | 2008-04-14 04:09:36 -0700 | [diff] [blame] | 2372 | } |
| 2373 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2374 | } |
| 2375 | #else |
| 2376 | #define DBGUNDO(x...) do { } while (0) |
| 2377 | #endif |
| 2378 | |
Yuchung Cheng | 7026b91 | 2013-05-29 14:20:13 +0000 | [diff] [blame] | 2379 | static void tcp_undo_cwnd_reduction(struct sock *sk, bool unmark_loss) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2380 | { |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 2381 | struct tcp_sock *tp = tcp_sk(sk); |
| 2382 | |
Yuchung Cheng | 6a63df4 | 2013-05-29 14:20:12 +0000 | [diff] [blame] | 2383 | if (unmark_loss) { |
| 2384 | struct sk_buff *skb; |
| 2385 | |
| 2386 | tcp_for_write_queue(skb, sk) { |
| 2387 | if (skb == tcp_send_head(sk)) |
| 2388 | break; |
| 2389 | TCP_SKB_CB(skb)->sacked &= ~TCPCB_LOST; |
| 2390 | } |
| 2391 | tp->lost_out = 0; |
| 2392 | tcp_clear_all_retrans_hints(tp); |
| 2393 | } |
| 2394 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2395 | if (tp->prior_ssthresh) { |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 2396 | const struct inet_connection_sock *icsk = inet_csk(sk); |
| 2397 | |
| 2398 | if (icsk->icsk_ca_ops->undo_cwnd) |
| 2399 | tp->snd_cwnd = icsk->icsk_ca_ops->undo_cwnd(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2400 | else |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 2401 | tp->snd_cwnd = max(tp->snd_cwnd, tp->snd_ssthresh << 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2402 | |
Yuchung Cheng | 7026b91 | 2013-05-29 14:20:13 +0000 | [diff] [blame] | 2403 | if (tp->prior_ssthresh > tp->snd_ssthresh) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2404 | tp->snd_ssthresh = tp->prior_ssthresh; |
Florian Westphal | 735d383 | 2014-09-29 13:08:30 +0200 | [diff] [blame] | 2405 | tcp_ecn_withdraw_cwr(tp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2406 | } |
| 2407 | } else { |
| 2408 | tp->snd_cwnd = max(tp->snd_cwnd, tp->snd_ssthresh); |
| 2409 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2410 | tp->snd_cwnd_stamp = tcp_time_stamp; |
Yuchung Cheng | 7026b91 | 2013-05-29 14:20:13 +0000 | [diff] [blame] | 2411 | tp->undo_marker = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2412 | } |
| 2413 | |
Vijay Subramanian | 67b95bd | 2012-07-19 21:32:18 +0000 | [diff] [blame] | 2414 | static inline bool tcp_may_undo(const struct tcp_sock *tp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2415 | { |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 2416 | return tp->undo_marker && (!tp->undo_retrans || tcp_packet_delayed(tp)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2417 | } |
| 2418 | |
| 2419 | /* People celebrate: "We love our President!" */ |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 2420 | static bool tcp_try_undo_recovery(struct sock *sk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2421 | { |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 2422 | struct tcp_sock *tp = tcp_sk(sk); |
| 2423 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2424 | if (tcp_may_undo(tp)) { |
Pavel Emelyanov | 40b215e | 2008-07-03 01:05:41 -0700 | [diff] [blame] | 2425 | int mib_idx; |
| 2426 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2427 | /* Happy end! We did not retransmit anything |
| 2428 | * or our original transmission succeeded. |
| 2429 | */ |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 2430 | 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] | 2431 | tcp_undo_cwnd_reduction(sk, false); |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 2432 | if (inet_csk(sk)->icsk_ca_state == TCP_CA_Loss) |
Pavel Emelyanov | 40b215e | 2008-07-03 01:05:41 -0700 | [diff] [blame] | 2433 | mib_idx = LINUX_MIB_TCPLOSSUNDO; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2434 | else |
Pavel Emelyanov | 40b215e | 2008-07-03 01:05:41 -0700 | [diff] [blame] | 2435 | mib_idx = LINUX_MIB_TCPFULLUNDO; |
| 2436 | |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 2437 | NET_INC_STATS_BH(sock_net(sk), mib_idx); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2438 | } |
Ilpo Järvinen | e60402d | 2007-08-09 15:14:46 +0300 | [diff] [blame] | 2439 | if (tp->snd_una == tp->high_seq && tcp_is_reno(tp)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2440 | /* Hold old state until something *above* high_seq |
| 2441 | * is ACKed. For Reno it is MUST to prevent false |
| 2442 | * fast retransmits (RFC2582). SACK TCP is safe. */ |
| 2443 | tcp_moderate_cwnd(tp); |
Marcelo Leitner | 1f37bf8 | 2014-11-04 17:15:08 -0200 | [diff] [blame] | 2444 | if (!tcp_any_retrans_done(sk)) |
| 2445 | tp->retrans_stamp = 0; |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 2446 | return true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2447 | } |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 2448 | tcp_set_ca_state(sk, TCP_CA_Open); |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 2449 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2450 | } |
| 2451 | |
| 2452 | /* Try to undo cwnd reduction, because D-SACKs acked all retransmitted data */ |
Yuchung Cheng | c7d9d6a | 2013-05-29 14:20:14 +0000 | [diff] [blame] | 2453 | static bool tcp_try_undo_dsack(struct sock *sk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2454 | { |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 2455 | struct tcp_sock *tp = tcp_sk(sk); |
| 2456 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2457 | if (tp->undo_marker && !tp->undo_retrans) { |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 2458 | DBGUNDO(sk, "D-SACK"); |
Yuchung Cheng | 7026b91 | 2013-05-29 14:20:13 +0000 | [diff] [blame] | 2459 | tcp_undo_cwnd_reduction(sk, false); |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 2460 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPDSACKUNDO); |
Yuchung Cheng | c7d9d6a | 2013-05-29 14:20:14 +0000 | [diff] [blame] | 2461 | return true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2462 | } |
Yuchung Cheng | c7d9d6a | 2013-05-29 14:20:14 +0000 | [diff] [blame] | 2463 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2464 | } |
| 2465 | |
Yuchung Cheng | e33099f | 2013-03-20 13:33:00 +0000 | [diff] [blame] | 2466 | /* Undo during loss recovery after partial ACK or using F-RTO. */ |
| 2467 | static bool tcp_try_undo_loss(struct sock *sk, bool frto_undo) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2468 | { |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 2469 | struct tcp_sock *tp = tcp_sk(sk); |
| 2470 | |
Yuchung Cheng | e33099f | 2013-03-20 13:33:00 +0000 | [diff] [blame] | 2471 | if (frto_undo || tcp_may_undo(tp)) { |
Yuchung Cheng | 7026b91 | 2013-05-29 14:20:13 +0000 | [diff] [blame] | 2472 | tcp_undo_cwnd_reduction(sk, true); |
Stephen Hemminger | 6a438bb | 2005-11-10 17:14:59 -0800 | [diff] [blame] | 2473 | |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 2474 | DBGUNDO(sk, "partial loss"); |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 2475 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPLOSSUNDO); |
Yuchung Cheng | e33099f | 2013-03-20 13:33:00 +0000 | [diff] [blame] | 2476 | if (frto_undo) |
| 2477 | NET_INC_STATS_BH(sock_net(sk), |
| 2478 | LINUX_MIB_TCPSPURIOUSRTOS); |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 2479 | inet_csk(sk)->icsk_retransmits = 0; |
Yuchung Cheng | e33099f | 2013-03-20 13:33:00 +0000 | [diff] [blame] | 2480 | if (frto_undo || tcp_is_sack(tp)) |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 2481 | tcp_set_ca_state(sk, TCP_CA_Open); |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 2482 | return true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2483 | } |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 2484 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2485 | } |
| 2486 | |
Yuchung Cheng | 684bad1 | 2012-09-02 17:38:04 +0000 | [diff] [blame] | 2487 | /* The cwnd reduction in CWR and Recovery use the PRR algorithm |
| 2488 | * https://datatracker.ietf.org/doc/draft-ietf-tcpm-proportional-rate-reduction/ |
Yuchung Cheng | fb4d3d1 | 2012-09-02 17:38:03 +0000 | [diff] [blame] | 2489 | * It computes the number of packets to send (sndcnt) based on packets newly |
| 2490 | * delivered: |
| 2491 | * 1) If the packets in flight is larger than ssthresh, PRR spreads the |
| 2492 | * cwnd reductions across a full RTT. |
| 2493 | * 2) If packets in flight is lower than ssthresh (such as due to excess |
| 2494 | * losses and/or application stalls), do not perform any further cwnd |
| 2495 | * reductions, but instead slow start up to ssthresh. |
| 2496 | */ |
Christoph Paasch | 5ee2c94 | 2014-07-14 16:58:32 +0200 | [diff] [blame] | 2497 | static void tcp_init_cwnd_reduction(struct sock *sk) |
Yuchung Cheng | 684bad1 | 2012-09-02 17:38:04 +0000 | [diff] [blame] | 2498 | { |
| 2499 | struct tcp_sock *tp = tcp_sk(sk); |
| 2500 | |
| 2501 | tp->high_seq = tp->snd_nxt; |
Nandita Dukkipati | 9b717a8 | 2013-03-11 10:00:44 +0000 | [diff] [blame] | 2502 | tp->tlp_high_seq = 0; |
Yuchung Cheng | 684bad1 | 2012-09-02 17:38:04 +0000 | [diff] [blame] | 2503 | tp->snd_cwnd_cnt = 0; |
| 2504 | tp->prior_cwnd = tp->snd_cwnd; |
| 2505 | tp->prr_delivered = 0; |
| 2506 | tp->prr_out = 0; |
Christoph Paasch | 5ee2c94 | 2014-07-14 16:58:32 +0200 | [diff] [blame] | 2507 | tp->snd_ssthresh = inet_csk(sk)->icsk_ca_ops->ssthresh(sk); |
Florian Westphal | 735d383 | 2014-09-29 13:08:30 +0200 | [diff] [blame] | 2508 | tcp_ecn_queue_cwr(tp); |
Yuchung Cheng | 684bad1 | 2012-09-02 17:38:04 +0000 | [diff] [blame] | 2509 | } |
| 2510 | |
Yuchung Cheng | 6804973 | 2013-05-29 14:20:11 +0000 | [diff] [blame] | 2511 | static void tcp_cwnd_reduction(struct sock *sk, const int prior_unsacked, |
Yuchung Cheng | 684bad1 | 2012-09-02 17:38:04 +0000 | [diff] [blame] | 2512 | int fast_rexmit) |
Yuchung Cheng | fb4d3d1 | 2012-09-02 17:38:03 +0000 | [diff] [blame] | 2513 | { |
| 2514 | struct tcp_sock *tp = tcp_sk(sk); |
| 2515 | int sndcnt = 0; |
| 2516 | int delta = tp->snd_ssthresh - tcp_packets_in_flight(tp); |
Yuchung Cheng | 6804973 | 2013-05-29 14:20:11 +0000 | [diff] [blame] | 2517 | int newly_acked_sacked = prior_unsacked - |
| 2518 | (tp->packets_out - tp->sacked_out); |
Yuchung Cheng | fb4d3d1 | 2012-09-02 17:38:03 +0000 | [diff] [blame] | 2519 | |
Yuchung Cheng | 684bad1 | 2012-09-02 17:38:04 +0000 | [diff] [blame] | 2520 | tp->prr_delivered += newly_acked_sacked; |
Yuchung Cheng | fb4d3d1 | 2012-09-02 17:38:03 +0000 | [diff] [blame] | 2521 | if (tcp_packets_in_flight(tp) > tp->snd_ssthresh) { |
| 2522 | u64 dividend = (u64)tp->snd_ssthresh * tp->prr_delivered + |
| 2523 | tp->prior_cwnd - 1; |
| 2524 | sndcnt = div_u64(dividend, tp->prior_cwnd) - tp->prr_out; |
| 2525 | } else { |
| 2526 | sndcnt = min_t(int, delta, |
| 2527 | max_t(int, tp->prr_delivered - tp->prr_out, |
| 2528 | newly_acked_sacked) + 1); |
| 2529 | } |
| 2530 | |
| 2531 | sndcnt = max(sndcnt, (fast_rexmit ? 1 : 0)); |
| 2532 | tp->snd_cwnd = tcp_packets_in_flight(tp) + sndcnt; |
| 2533 | } |
| 2534 | |
Yuchung Cheng | 684bad1 | 2012-09-02 17:38:04 +0000 | [diff] [blame] | 2535 | static inline void tcp_end_cwnd_reduction(struct sock *sk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2536 | { |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 2537 | struct tcp_sock *tp = tcp_sk(sk); |
Nandita Dukkipati | a262f0c | 2011-08-21 20:21:57 +0000 | [diff] [blame] | 2538 | |
Yuchung Cheng | 684bad1 | 2012-09-02 17:38:04 +0000 | [diff] [blame] | 2539 | /* Reset cwnd to ssthresh in CWR or Recovery (unless it's undone) */ |
| 2540 | if (inet_csk(sk)->icsk_ca_state == TCP_CA_CWR || |
| 2541 | (tp->undo_marker && tp->snd_ssthresh < TCP_INFINITE_SSTHRESH)) { |
| 2542 | tp->snd_cwnd = tp->snd_ssthresh; |
| 2543 | tp->snd_cwnd_stamp = tcp_time_stamp; |
Yuchung Cheng | 67d4120 | 2011-03-14 10:57:03 +0000 | [diff] [blame] | 2544 | } |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 2545 | tcp_ca_event(sk, CA_EVENT_COMPLETE_CWR); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2546 | } |
| 2547 | |
Yuchung Cheng | 684bad1 | 2012-09-02 17:38:04 +0000 | [diff] [blame] | 2548 | /* Enter CWR state. Disable cwnd undo since congestion is proven with ECN */ |
Christoph Paasch | 5ee2c94 | 2014-07-14 16:58:32 +0200 | [diff] [blame] | 2549 | void tcp_enter_cwr(struct sock *sk) |
Yuchung Cheng | 09484d1 | 2012-09-02 17:38:02 +0000 | [diff] [blame] | 2550 | { |
| 2551 | struct tcp_sock *tp = tcp_sk(sk); |
Yuchung Cheng | 09484d1 | 2012-09-02 17:38:02 +0000 | [diff] [blame] | 2552 | |
| 2553 | tp->prior_ssthresh = 0; |
Yuchung Cheng | 684bad1 | 2012-09-02 17:38:04 +0000 | [diff] [blame] | 2554 | if (inet_csk(sk)->icsk_ca_state < TCP_CA_CWR) { |
Yuchung Cheng | 09484d1 | 2012-09-02 17:38:02 +0000 | [diff] [blame] | 2555 | tp->undo_marker = 0; |
Christoph Paasch | 5ee2c94 | 2014-07-14 16:58:32 +0200 | [diff] [blame] | 2556 | tcp_init_cwnd_reduction(sk); |
Yuchung Cheng | 09484d1 | 2012-09-02 17:38:02 +0000 | [diff] [blame] | 2557 | tcp_set_ca_state(sk, TCP_CA_CWR); |
| 2558 | } |
| 2559 | } |
| 2560 | |
Ilpo Järvinen | 8aca6cb | 2008-06-04 11:34:22 -0700 | [diff] [blame] | 2561 | static void tcp_try_keep_open(struct sock *sk) |
| 2562 | { |
| 2563 | struct tcp_sock *tp = tcp_sk(sk); |
| 2564 | int state = TCP_CA_Open; |
| 2565 | |
Neal Cardwell | f698204 | 2011-11-16 08:58:04 +0000 | [diff] [blame] | 2566 | if (tcp_left_out(tp) || tcp_any_retrans_done(sk)) |
Ilpo Järvinen | 8aca6cb | 2008-06-04 11:34:22 -0700 | [diff] [blame] | 2567 | state = TCP_CA_Disorder; |
| 2568 | |
| 2569 | if (inet_csk(sk)->icsk_ca_state != state) { |
| 2570 | tcp_set_ca_state(sk, state); |
| 2571 | tp->high_seq = tp->snd_nxt; |
| 2572 | } |
| 2573 | } |
| 2574 | |
Yuchung Cheng | 6804973 | 2013-05-29 14:20:11 +0000 | [diff] [blame] | 2575 | 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] | 2576 | { |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 2577 | struct tcp_sock *tp = tcp_sk(sk); |
| 2578 | |
Ilpo Järvinen | 86426c2 | 2007-08-09 14:45:17 +0300 | [diff] [blame] | 2579 | tcp_verify_left_out(tp); |
| 2580 | |
Yuchung Cheng | 9b44190 | 2013-03-20 13:32:58 +0000 | [diff] [blame] | 2581 | if (!tcp_any_retrans_done(sk)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2582 | tp->retrans_stamp = 0; |
| 2583 | |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 2584 | if (flag & FLAG_ECE) |
Christoph Paasch | 5ee2c94 | 2014-07-14 16:58:32 +0200 | [diff] [blame] | 2585 | tcp_enter_cwr(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2586 | |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 2587 | if (inet_csk(sk)->icsk_ca_state != TCP_CA_CWR) { |
Ilpo Järvinen | 8aca6cb | 2008-06-04 11:34:22 -0700 | [diff] [blame] | 2588 | tcp_try_keep_open(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2589 | } else { |
Yuchung Cheng | 6804973 | 2013-05-29 14:20:11 +0000 | [diff] [blame] | 2590 | tcp_cwnd_reduction(sk, prior_unsacked, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2591 | } |
| 2592 | } |
| 2593 | |
John Heffner | 5d424d5 | 2006-03-20 17:53:41 -0800 | [diff] [blame] | 2594 | static void tcp_mtup_probe_failed(struct sock *sk) |
| 2595 | { |
| 2596 | struct inet_connection_sock *icsk = inet_csk(sk); |
| 2597 | |
| 2598 | icsk->icsk_mtup.search_high = icsk->icsk_mtup.probe_size - 1; |
| 2599 | icsk->icsk_mtup.probe_size = 0; |
| 2600 | } |
| 2601 | |
Ilpo Järvinen | 72211e9 | 2009-03-14 14:23:04 +0000 | [diff] [blame] | 2602 | static void tcp_mtup_probe_success(struct sock *sk) |
John Heffner | 5d424d5 | 2006-03-20 17:53:41 -0800 | [diff] [blame] | 2603 | { |
| 2604 | struct tcp_sock *tp = tcp_sk(sk); |
| 2605 | struct inet_connection_sock *icsk = inet_csk(sk); |
| 2606 | |
| 2607 | /* FIXME: breaks with very large cwnd */ |
| 2608 | tp->prior_ssthresh = tcp_current_ssthresh(sk); |
| 2609 | tp->snd_cwnd = tp->snd_cwnd * |
| 2610 | tcp_mss_to_mtu(sk, tp->mss_cache) / |
| 2611 | icsk->icsk_mtup.probe_size; |
| 2612 | tp->snd_cwnd_cnt = 0; |
| 2613 | tp->snd_cwnd_stamp = tcp_time_stamp; |
John Heffner | 9c6d5e5 | 2010-10-06 21:18:02 -0700 | [diff] [blame] | 2614 | tp->snd_ssthresh = tcp_current_ssthresh(sk); |
John Heffner | 5d424d5 | 2006-03-20 17:53:41 -0800 | [diff] [blame] | 2615 | |
| 2616 | icsk->icsk_mtup.search_low = icsk->icsk_mtup.probe_size; |
| 2617 | icsk->icsk_mtup.probe_size = 0; |
| 2618 | tcp_sync_mss(sk, icsk->icsk_pmtu_cookie); |
| 2619 | } |
| 2620 | |
Ilpo Järvinen | e1aa680 | 2008-11-24 21:11:55 -0800 | [diff] [blame] | 2621 | /* Do a simple retransmit without using the backoff mechanisms in |
| 2622 | * tcp_timer. This is used for path mtu discovery. |
| 2623 | * The socket is already locked here. |
| 2624 | */ |
| 2625 | void tcp_simple_retransmit(struct sock *sk) |
| 2626 | { |
| 2627 | const struct inet_connection_sock *icsk = inet_csk(sk); |
| 2628 | struct tcp_sock *tp = tcp_sk(sk); |
| 2629 | struct sk_buff *skb; |
Ilpo Järvinen | 0c54b85 | 2009-03-14 14:23:05 +0000 | [diff] [blame] | 2630 | unsigned int mss = tcp_current_mss(sk); |
Ilpo Järvinen | e1aa680 | 2008-11-24 21:11:55 -0800 | [diff] [blame] | 2631 | u32 prior_lost = tp->lost_out; |
| 2632 | |
| 2633 | tcp_for_write_queue(skb, sk) { |
| 2634 | if (skb == tcp_send_head(sk)) |
| 2635 | break; |
Ilpo Järvinen | 775ffab | 2008-12-05 22:41:26 -0800 | [diff] [blame] | 2636 | if (tcp_skb_seglen(skb) > mss && |
Ilpo Järvinen | e1aa680 | 2008-11-24 21:11:55 -0800 | [diff] [blame] | 2637 | !(TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)) { |
| 2638 | if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_RETRANS) { |
| 2639 | TCP_SKB_CB(skb)->sacked &= ~TCPCB_SACKED_RETRANS; |
| 2640 | tp->retrans_out -= tcp_skb_pcount(skb); |
| 2641 | } |
| 2642 | tcp_skb_mark_lost_uncond_verify(tp, skb); |
| 2643 | } |
| 2644 | } |
| 2645 | |
| 2646 | tcp_clear_retrans_hints_partial(tp); |
| 2647 | |
| 2648 | if (prior_lost == tp->lost_out) |
| 2649 | return; |
| 2650 | |
| 2651 | if (tcp_is_reno(tp)) |
| 2652 | tcp_limit_reno_sacked(tp); |
| 2653 | |
| 2654 | tcp_verify_left_out(tp); |
| 2655 | |
| 2656 | /* Don't muck with the congestion window here. |
| 2657 | * Reason is that we do not increase amount of _data_ |
| 2658 | * in network, but units changed and effective |
| 2659 | * cwnd/ssthresh really reduced now. |
| 2660 | */ |
| 2661 | if (icsk->icsk_ca_state != TCP_CA_Loss) { |
| 2662 | tp->high_seq = tp->snd_nxt; |
| 2663 | tp->snd_ssthresh = tcp_current_ssthresh(sk); |
| 2664 | tp->prior_ssthresh = 0; |
| 2665 | tp->undo_marker = 0; |
| 2666 | tcp_set_ca_state(sk, TCP_CA_Loss); |
| 2667 | } |
| 2668 | tcp_xmit_retransmit_queue(sk); |
| 2669 | } |
Eric Dumazet | 4bc2f18 | 2010-07-09 21:22:10 +0000 | [diff] [blame] | 2670 | EXPORT_SYMBOL(tcp_simple_retransmit); |
Ilpo Järvinen | e1aa680 | 2008-11-24 21:11:55 -0800 | [diff] [blame] | 2671 | |
Yuchung Cheng | 1fbc340 | 2012-05-02 13:30:02 +0000 | [diff] [blame] | 2672 | static void tcp_enter_recovery(struct sock *sk, bool ece_ack) |
| 2673 | { |
| 2674 | struct tcp_sock *tp = tcp_sk(sk); |
| 2675 | int mib_idx; |
| 2676 | |
| 2677 | if (tcp_is_reno(tp)) |
| 2678 | mib_idx = LINUX_MIB_TCPRENORECOVERY; |
| 2679 | else |
| 2680 | mib_idx = LINUX_MIB_TCPSACKRECOVERY; |
| 2681 | |
| 2682 | NET_INC_STATS_BH(sock_net(sk), mib_idx); |
| 2683 | |
Yuchung Cheng | 1fbc340 | 2012-05-02 13:30:02 +0000 | [diff] [blame] | 2684 | tp->prior_ssthresh = 0; |
Yuchung Cheng | 989e04c | 2014-08-22 14:15:22 -0700 | [diff] [blame] | 2685 | tcp_init_undo(tp); |
Yuchung Cheng | 1fbc340 | 2012-05-02 13:30:02 +0000 | [diff] [blame] | 2686 | |
| 2687 | if (inet_csk(sk)->icsk_ca_state < TCP_CA_CWR) { |
| 2688 | if (!ece_ack) |
| 2689 | tp->prior_ssthresh = tcp_current_ssthresh(sk); |
Christoph Paasch | 5ee2c94 | 2014-07-14 16:58:32 +0200 | [diff] [blame] | 2690 | tcp_init_cwnd_reduction(sk); |
Yuchung Cheng | 1fbc340 | 2012-05-02 13:30:02 +0000 | [diff] [blame] | 2691 | } |
Yuchung Cheng | 1fbc340 | 2012-05-02 13:30:02 +0000 | [diff] [blame] | 2692 | tcp_set_ca_state(sk, TCP_CA_Recovery); |
| 2693 | } |
| 2694 | |
Yuchung Cheng | ab42d9e | 2013-03-20 13:32:59 +0000 | [diff] [blame] | 2695 | /* Process an ACK in CA_Loss state. Move to CA_Open if lost data are |
| 2696 | * recovered or spurious. Otherwise retransmits more on partial ACKs. |
| 2697 | */ |
Yuchung Cheng | e33099f | 2013-03-20 13:33:00 +0000 | [diff] [blame] | 2698 | 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] | 2699 | { |
Yuchung Cheng | ab42d9e | 2013-03-20 13:32:59 +0000 | [diff] [blame] | 2700 | struct tcp_sock *tp = tcp_sk(sk); |
Yuchung Cheng | e33099f | 2013-03-20 13:33:00 +0000 | [diff] [blame] | 2701 | bool recovered = !before(tp->snd_una, tp->high_seq); |
Yuchung Cheng | ab42d9e | 2013-03-20 13:32:59 +0000 | [diff] [blame] | 2702 | |
Yuchung Cheng | e33099f | 2013-03-20 13:33:00 +0000 | [diff] [blame] | 2703 | if (tp->frto) { /* F-RTO RFC5682 sec 3.1 (sack enhanced version). */ |
Yuchung Cheng | 0cfa5c0 | 2014-05-30 15:25:59 -0700 | [diff] [blame] | 2704 | /* Step 3.b. A timeout is spurious if not all data are |
| 2705 | * lost, i.e., never-retransmitted data are (s)acked. |
| 2706 | */ |
| 2707 | if (tcp_try_undo_loss(sk, flag & FLAG_ORIG_SACK_ACKED)) |
Yuchung Cheng | e33099f | 2013-03-20 13:33:00 +0000 | [diff] [blame] | 2708 | return; |
Yuchung Cheng | 0cfa5c0 | 2014-05-30 15:25:59 -0700 | [diff] [blame] | 2709 | |
Yuchung Cheng | e33099f | 2013-03-20 13:33:00 +0000 | [diff] [blame] | 2710 | if (after(tp->snd_nxt, tp->high_seq) && |
| 2711 | (flag & FLAG_DATA_SACKED || is_dupack)) { |
| 2712 | tp->frto = 0; /* Loss was real: 2nd part of step 3.a */ |
| 2713 | } else if (flag & FLAG_SND_UNA_ADVANCED && !recovered) { |
| 2714 | tp->high_seq = tp->snd_nxt; |
| 2715 | __tcp_push_pending_frames(sk, tcp_current_mss(sk), |
| 2716 | TCP_NAGLE_OFF); |
| 2717 | if (after(tp->snd_nxt, tp->high_seq)) |
| 2718 | return; /* Step 2.b */ |
| 2719 | tp->frto = 0; |
| 2720 | } |
| 2721 | } |
| 2722 | |
| 2723 | if (recovered) { |
| 2724 | /* 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] | 2725 | tcp_try_undo_recovery(sk); |
| 2726 | return; |
| 2727 | } |
Yuchung Cheng | e33099f | 2013-03-20 13:33:00 +0000 | [diff] [blame] | 2728 | if (tcp_is_reno(tp)) { |
| 2729 | /* A Reno DUPACK means new data in F-RTO step 2.b above are |
| 2730 | * delivered. Lower inflight to clock out (re)tranmissions. |
| 2731 | */ |
| 2732 | if (after(tp->snd_nxt, tp->high_seq) && is_dupack) |
| 2733 | tcp_add_reno_sack(sk); |
| 2734 | else if (flag & FLAG_SND_UNA_ADVANCED) |
| 2735 | tcp_reset_reno_sack(tp); |
| 2736 | } |
| 2737 | if (tcp_try_undo_loss(sk, false)) |
Yuchung Cheng | ab42d9e | 2013-03-20 13:32:59 +0000 | [diff] [blame] | 2738 | return; |
Yuchung Cheng | ab42d9e | 2013-03-20 13:32:59 +0000 | [diff] [blame] | 2739 | tcp_xmit_retransmit_queue(sk); |
| 2740 | } |
| 2741 | |
Yuchung Cheng | 6a63df4 | 2013-05-29 14:20:12 +0000 | [diff] [blame] | 2742 | /* Undo during fast recovery after partial ACK. */ |
Yuchung Cheng | 7026b91 | 2013-05-29 14:20:13 +0000 | [diff] [blame] | 2743 | static bool tcp_try_undo_partial(struct sock *sk, const int acked, |
| 2744 | const int prior_unsacked) |
Yuchung Cheng | 6a63df4 | 2013-05-29 14:20:12 +0000 | [diff] [blame] | 2745 | { |
| 2746 | struct tcp_sock *tp = tcp_sk(sk); |
Yuchung Cheng | 6a63df4 | 2013-05-29 14:20:12 +0000 | [diff] [blame] | 2747 | |
Yuchung Cheng | 7026b91 | 2013-05-29 14:20:13 +0000 | [diff] [blame] | 2748 | if (tp->undo_marker && tcp_packet_delayed(tp)) { |
Yuchung Cheng | 6a63df4 | 2013-05-29 14:20:12 +0000 | [diff] [blame] | 2749 | /* Plain luck! Hole if filled with delayed |
| 2750 | * packet, rather than with a retransmit. |
| 2751 | */ |
Yuchung Cheng | 7026b91 | 2013-05-29 14:20:13 +0000 | [diff] [blame] | 2752 | tcp_update_reordering(sk, tcp_fackets_out(tp) + acked, 1); |
| 2753 | |
| 2754 | /* We are getting evidence that the reordering degree is higher |
| 2755 | * than we realized. If there are no retransmits out then we |
| 2756 | * can undo. Otherwise we clock out new packets but do not |
| 2757 | * mark more packets lost or retransmit more. |
| 2758 | */ |
| 2759 | if (tp->retrans_out) { |
| 2760 | tcp_cwnd_reduction(sk, prior_unsacked, 0); |
| 2761 | return true; |
| 2762 | } |
| 2763 | |
Yuchung Cheng | 6a63df4 | 2013-05-29 14:20:12 +0000 | [diff] [blame] | 2764 | if (!tcp_any_retrans_done(sk)) |
| 2765 | tp->retrans_stamp = 0; |
| 2766 | |
Yuchung Cheng | 7026b91 | 2013-05-29 14:20:13 +0000 | [diff] [blame] | 2767 | DBGUNDO(sk, "partial recovery"); |
| 2768 | tcp_undo_cwnd_reduction(sk, true); |
Yuchung Cheng | 6a63df4 | 2013-05-29 14:20:12 +0000 | [diff] [blame] | 2769 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPPARTIALUNDO); |
Yuchung Cheng | 7026b91 | 2013-05-29 14:20:13 +0000 | [diff] [blame] | 2770 | tcp_try_keep_open(sk); |
| 2771 | return true; |
Yuchung Cheng | 6a63df4 | 2013-05-29 14:20:12 +0000 | [diff] [blame] | 2772 | } |
Yuchung Cheng | 7026b91 | 2013-05-29 14:20:13 +0000 | [diff] [blame] | 2773 | return false; |
Yuchung Cheng | 6a63df4 | 2013-05-29 14:20:12 +0000 | [diff] [blame] | 2774 | } |
| 2775 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2776 | /* Process an event, which can update packets-in-flight not trivially. |
| 2777 | * Main goal of this function is to calculate new estimate for left_out, |
| 2778 | * taking into account both packets sitting in receiver's buffer and |
| 2779 | * packets lost by network. |
| 2780 | * |
| 2781 | * Besides that it does CWND reduction, when packet loss is detected |
| 2782 | * and changes state of machine. |
| 2783 | * |
| 2784 | * It does _not_ decide what to send, it is made in function |
| 2785 | * tcp_xmit_retransmit_queue(). |
| 2786 | */ |
Yuchung Cheng | 6804973 | 2013-05-29 14:20:11 +0000 | [diff] [blame] | 2787 | static void tcp_fastretrans_alert(struct sock *sk, const int acked, |
| 2788 | const int prior_unsacked, |
Nandita Dukkipati | 35f079e | 2013-05-21 15:12:07 +0000 | [diff] [blame] | 2789 | bool is_dupack, int flag) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2790 | { |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 2791 | struct inet_connection_sock *icsk = inet_csk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2792 | struct tcp_sock *tp = tcp_sk(sk); |
Yuchung Cheng | 6a63df4 | 2013-05-29 14:20:12 +0000 | [diff] [blame] | 2793 | bool do_lost = is_dupack || ((flag & FLAG_DATA_SACKED) && |
Ilpo Järvinen | 85cc391 | 2007-11-15 19:39:31 -0800 | [diff] [blame] | 2794 | (tcp_fackets_out(tp) > tp->reordering)); |
Yuchung Cheng | 1fbc340 | 2012-05-02 13:30:02 +0000 | [diff] [blame] | 2795 | int fast_rexmit = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2796 | |
Ilpo Järvinen | 3ccd313 | 2007-12-31 04:43:32 -0800 | [diff] [blame] | 2797 | if (WARN_ON(!tp->packets_out && tp->sacked_out)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2798 | tp->sacked_out = 0; |
Ilpo Järvinen | 91fed7a | 2007-10-09 01:24:15 -0700 | [diff] [blame] | 2799 | if (WARN_ON(!tp->sacked_out && tp->fackets_out)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2800 | tp->fackets_out = 0; |
| 2801 | |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 2802 | /* Now state machine starts. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2803 | * A. ECE, hence prohibit cwnd undoing, the reduction is required. */ |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 2804 | if (flag & FLAG_ECE) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2805 | tp->prior_ssthresh = 0; |
| 2806 | |
| 2807 | /* B. In all the states check for reneging SACKs. */ |
Ilpo Järvinen | cadbd03 | 2007-12-31 04:49:21 -0800 | [diff] [blame] | 2808 | if (tcp_check_sack_reneging(sk, flag)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2809 | return; |
| 2810 | |
Yuchung Cheng | 974c123 | 2012-01-19 14:42:21 +0000 | [diff] [blame] | 2811 | /* C. Check consistency of the current state. */ |
Ilpo Järvinen | 005903b | 2007-08-09 14:44:16 +0300 | [diff] [blame] | 2812 | tcp_verify_left_out(tp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2813 | |
Yuchung Cheng | 974c123 | 2012-01-19 14:42:21 +0000 | [diff] [blame] | 2814 | /* D. Check state exit conditions. State can be terminated |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2815 | * when high_seq is ACKed. */ |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 2816 | if (icsk->icsk_ca_state == TCP_CA_Open) { |
Ilpo Järvinen | 547b792 | 2008-07-25 21:43:18 -0700 | [diff] [blame] | 2817 | WARN_ON(tp->retrans_out != 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2818 | tp->retrans_stamp = 0; |
| 2819 | } else if (!before(tp->snd_una, tp->high_seq)) { |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 2820 | switch (icsk->icsk_ca_state) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2821 | case TCP_CA_CWR: |
| 2822 | /* CWR is to be held something *above* high_seq |
| 2823 | * is ACKed for CWR bit to reach receiver. */ |
| 2824 | if (tp->snd_una != tp->high_seq) { |
Yuchung Cheng | 684bad1 | 2012-09-02 17:38:04 +0000 | [diff] [blame] | 2825 | tcp_end_cwnd_reduction(sk); |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 2826 | tcp_set_ca_state(sk, TCP_CA_Open); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2827 | } |
| 2828 | break; |
| 2829 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2830 | case TCP_CA_Recovery: |
Ilpo Järvinen | e60402d | 2007-08-09 15:14:46 +0300 | [diff] [blame] | 2831 | if (tcp_is_reno(tp)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2832 | tcp_reset_reno_sack(tp); |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 2833 | if (tcp_try_undo_recovery(sk)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2834 | return; |
Yuchung Cheng | 684bad1 | 2012-09-02 17:38:04 +0000 | [diff] [blame] | 2835 | tcp_end_cwnd_reduction(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2836 | break; |
| 2837 | } |
| 2838 | } |
| 2839 | |
Yuchung Cheng | 974c123 | 2012-01-19 14:42:21 +0000 | [diff] [blame] | 2840 | /* E. Process state. */ |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 2841 | switch (icsk->icsk_ca_state) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2842 | case TCP_CA_Recovery: |
Ilpo Järvinen | 2e60529 | 2007-08-02 19:46:58 -0700 | [diff] [blame] | 2843 | if (!(flag & FLAG_SND_UNA_ADVANCED)) { |
Ilpo Järvinen | e60402d | 2007-08-09 15:14:46 +0300 | [diff] [blame] | 2844 | if (tcp_is_reno(tp) && is_dupack) |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 2845 | tcp_add_reno_sack(sk); |
Yuchung Cheng | 7026b91 | 2013-05-29 14:20:13 +0000 | [diff] [blame] | 2846 | } else { |
| 2847 | if (tcp_try_undo_partial(sk, acked, prior_unsacked)) |
| 2848 | return; |
| 2849 | /* Partial ACK arrived. Force fast retransmit. */ |
| 2850 | do_lost = tcp_is_reno(tp) || |
| 2851 | tcp_fackets_out(tp) > tp->reordering; |
| 2852 | } |
Yuchung Cheng | c7d9d6a | 2013-05-29 14:20:14 +0000 | [diff] [blame] | 2853 | if (tcp_try_undo_dsack(sk)) { |
| 2854 | tcp_try_keep_open(sk); |
| 2855 | return; |
| 2856 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2857 | break; |
| 2858 | case TCP_CA_Loss: |
Yuchung Cheng | e33099f | 2013-03-20 13:33:00 +0000 | [diff] [blame] | 2859 | tcp_process_loss(sk, flag, is_dupack); |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 2860 | if (icsk->icsk_ca_state != TCP_CA_Open) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2861 | return; |
Yuchung Cheng | ab42d9e | 2013-03-20 13:32:59 +0000 | [diff] [blame] | 2862 | /* Fall through to processing in Open state. */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2863 | default: |
Ilpo Järvinen | e60402d | 2007-08-09 15:14:46 +0300 | [diff] [blame] | 2864 | if (tcp_is_reno(tp)) { |
Ilpo Järvinen | 2e60529 | 2007-08-02 19:46:58 -0700 | [diff] [blame] | 2865 | if (flag & FLAG_SND_UNA_ADVANCED) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2866 | tcp_reset_reno_sack(tp); |
| 2867 | if (is_dupack) |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 2868 | tcp_add_reno_sack(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2869 | } |
| 2870 | |
Neal Cardwell | f698204 | 2011-11-16 08:58:04 +0000 | [diff] [blame] | 2871 | if (icsk->icsk_ca_state <= TCP_CA_Disorder) |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 2872 | tcp_try_undo_dsack(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2873 | |
Yuchung Cheng | 750ea2b | 2012-05-02 13:30:04 +0000 | [diff] [blame] | 2874 | if (!tcp_time_to_recover(sk, flag)) { |
Yuchung Cheng | 6804973 | 2013-05-29 14:20:11 +0000 | [diff] [blame] | 2875 | tcp_try_to_open(sk, flag, prior_unsacked); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2876 | return; |
| 2877 | } |
| 2878 | |
John Heffner | 5d424d5 | 2006-03-20 17:53:41 -0800 | [diff] [blame] | 2879 | /* MTU probe failure: don't reduce cwnd */ |
| 2880 | if (icsk->icsk_ca_state < TCP_CA_CWR && |
| 2881 | icsk->icsk_mtup.probe_size && |
John Heffner | 0e7b136 | 2006-03-20 21:32:58 -0800 | [diff] [blame] | 2882 | tp->snd_una == tp->mtu_probe.probe_seq_start) { |
John Heffner | 5d424d5 | 2006-03-20 17:53:41 -0800 | [diff] [blame] | 2883 | tcp_mtup_probe_failed(sk); |
| 2884 | /* Restores the reduction we did in tcp_mtup_probe() */ |
| 2885 | tp->snd_cwnd++; |
| 2886 | tcp_simple_retransmit(sk); |
| 2887 | return; |
| 2888 | } |
| 2889 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2890 | /* Otherwise enter Recovery state */ |
Yuchung Cheng | 1fbc340 | 2012-05-02 13:30:02 +0000 | [diff] [blame] | 2891 | tcp_enter_recovery(sk, (flag & FLAG_ECE)); |
Ilpo Järvinen | 85cc391 | 2007-11-15 19:39:31 -0800 | [diff] [blame] | 2892 | fast_rexmit = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2893 | } |
| 2894 | |
Yuchung Cheng | 3e59cb0 | 2013-05-17 13:45:05 +0000 | [diff] [blame] | 2895 | if (do_lost) |
Ilpo Järvinen | 85cc391 | 2007-11-15 19:39:31 -0800 | [diff] [blame] | 2896 | tcp_update_scoreboard(sk, fast_rexmit); |
Yuchung Cheng | 6804973 | 2013-05-29 14:20:11 +0000 | [diff] [blame] | 2897 | tcp_cwnd_reduction(sk, prior_unsacked, fast_rexmit); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2898 | tcp_xmit_retransmit_queue(sk); |
| 2899 | } |
| 2900 | |
Yuchung Cheng | ed08495 | 2013-07-22 16:20:48 -0700 | [diff] [blame] | 2901 | static inline bool tcp_ack_update_rtt(struct sock *sk, const int flag, |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 2902 | long seq_rtt_us, long sack_rtt_us) |
Ilpo Järvinen | 41834b7 | 2008-12-05 22:43:26 -0800 | [diff] [blame] | 2903 | { |
Yuchung Cheng | 5b08e47 | 2013-07-22 16:20:46 -0700 | [diff] [blame] | 2904 | const struct tcp_sock *tp = tcp_sk(sk); |
Ilpo Järvinen | 41834b7 | 2008-12-05 22:43:26 -0800 | [diff] [blame] | 2905 | |
Yuchung Cheng | 5b08e47 | 2013-07-22 16:20:46 -0700 | [diff] [blame] | 2906 | /* Prefer RTT measured from ACK's timing to TS-ECR. This is because |
| 2907 | * broken middle-boxes or peers may corrupt TS-ECR fields. But |
| 2908 | * Karn's algorithm forbids taking RTT if some retransmitted data |
| 2909 | * is acked (RFC6298). |
| 2910 | */ |
| 2911 | if (flag & FLAG_RETRANS_DATA_ACKED) |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 2912 | seq_rtt_us = -1L; |
Yuchung Cheng | 5b08e47 | 2013-07-22 16:20:46 -0700 | [diff] [blame] | 2913 | |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 2914 | if (seq_rtt_us < 0) |
| 2915 | seq_rtt_us = sack_rtt_us; |
Yuchung Cheng | ed08495 | 2013-07-22 16:20:48 -0700 | [diff] [blame] | 2916 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2917 | /* RTTM Rule: A TSecr value received in a segment is used to |
| 2918 | * update the averaged RTT measurement only if the segment |
| 2919 | * acknowledges some new data, i.e., only if it advances the |
| 2920 | * left edge of the send window. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2921 | * See draft-ietf-tcplw-high-performance-00, section 3.3. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2922 | */ |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 2923 | if (seq_rtt_us < 0 && tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr && |
Yuchung Cheng | 2909d87 | 2013-10-24 08:55:25 -0700 | [diff] [blame] | 2924 | flag & FLAG_ACKED) |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 2925 | seq_rtt_us = jiffies_to_usecs(tcp_time_stamp - tp->rx_opt.rcv_tsecr); |
Ilpo Järvinen | 41834b7 | 2008-12-05 22:43:26 -0800 | [diff] [blame] | 2926 | |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 2927 | if (seq_rtt_us < 0) |
Yuchung Cheng | ed08495 | 2013-07-22 16:20:48 -0700 | [diff] [blame] | 2928 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2929 | |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 2930 | tcp_rtt_estimator(sk, seq_rtt_us); |
Yuchung Cheng | 5b08e47 | 2013-07-22 16:20:46 -0700 | [diff] [blame] | 2931 | tcp_set_rto(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2932 | |
Yuchung Cheng | 5b08e47 | 2013-07-22 16:20:46 -0700 | [diff] [blame] | 2933 | /* RFC6298: only reset backoff on valid RTT measurement. */ |
| 2934 | inet_csk(sk)->icsk_backoff = 0; |
Yuchung Cheng | ed08495 | 2013-07-22 16:20:48 -0700 | [diff] [blame] | 2935 | return true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2936 | } |
| 2937 | |
Yuchung Cheng | 375fe02 | 2013-07-22 16:20:45 -0700 | [diff] [blame] | 2938 | /* Compute time elapsed between (last) SYNACK and the ACK completing 3WHS. */ |
Yuchung Cheng | bc15afa | 2013-10-24 08:44:25 -0700 | [diff] [blame] | 2939 | static void tcp_synack_rtt_meas(struct sock *sk, const u32 synack_stamp) |
Yuchung Cheng | 375fe02 | 2013-07-22 16:20:45 -0700 | [diff] [blame] | 2940 | { |
| 2941 | struct tcp_sock *tp = tcp_sk(sk); |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 2942 | long seq_rtt_us = -1L; |
Yuchung Cheng | 375fe02 | 2013-07-22 16:20:45 -0700 | [diff] [blame] | 2943 | |
Yuchung Cheng | bc15afa | 2013-10-24 08:44:25 -0700 | [diff] [blame] | 2944 | if (synack_stamp && !tp->total_retrans) |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 2945 | seq_rtt_us = jiffies_to_usecs(tcp_time_stamp - synack_stamp); |
Yuchung Cheng | bc15afa | 2013-10-24 08:44:25 -0700 | [diff] [blame] | 2946 | |
| 2947 | /* If the ACK acks both the SYNACK and the (Fast Open'd) data packets |
| 2948 | * sent in SYN_RECV, SYNACK RTT is the smooth RTT computed in tcp_ack() |
| 2949 | */ |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 2950 | if (!tp->srtt_us) |
| 2951 | tcp_ack_update_rtt(sk, FLAG_SYN_ACKED, seq_rtt_us, -1L); |
Yuchung Cheng | 375fe02 | 2013-07-22 16:20:45 -0700 | [diff] [blame] | 2952 | } |
| 2953 | |
Eric Dumazet | 2490155 | 2014-05-02 21:18:05 -0700 | [diff] [blame] | 2954 | static void tcp_cong_avoid(struct sock *sk, u32 ack, u32 acked) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2955 | { |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 2956 | const struct inet_connection_sock *icsk = inet_csk(sk); |
Eric Dumazet | 2490155 | 2014-05-02 21:18:05 -0700 | [diff] [blame] | 2957 | |
| 2958 | icsk->icsk_ca_ops->cong_avoid(sk, ack, acked); |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 2959 | tcp_sk(sk)->snd_cwnd_stamp = tcp_time_stamp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2960 | } |
| 2961 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2962 | /* Restart timer after forward progress on connection. |
| 2963 | * RFC2988 recommends to restart timer to now+rto. |
| 2964 | */ |
Yuchung Cheng | 750ea2b | 2012-05-02 13:30:04 +0000 | [diff] [blame] | 2965 | void tcp_rearm_rto(struct sock *sk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2966 | { |
Nandita Dukkipati | 6ba8a3b | 2013-03-11 10:00:43 +0000 | [diff] [blame] | 2967 | const struct inet_connection_sock *icsk = inet_csk(sk); |
Yuchung Cheng | 750ea2b | 2012-05-02 13:30:04 +0000 | [diff] [blame] | 2968 | struct tcp_sock *tp = tcp_sk(sk); |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 2969 | |
Jerry Chu | 168a8f5 | 2012-08-31 12:29:13 +0000 | [diff] [blame] | 2970 | /* If the retrans timer is currently being used by Fast Open |
| 2971 | * for SYN-ACK retrans purpose, stay put. |
| 2972 | */ |
| 2973 | if (tp->fastopen_rsk) |
| 2974 | return; |
| 2975 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2976 | if (!tp->packets_out) { |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 2977 | inet_csk_clear_xmit_timer(sk, ICSK_TIME_RETRANS); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2978 | } else { |
Yuchung Cheng | 750ea2b | 2012-05-02 13:30:04 +0000 | [diff] [blame] | 2979 | u32 rto = inet_csk(sk)->icsk_rto; |
| 2980 | /* Offset the time elapsed after installing regular RTO */ |
Nandita Dukkipati | 6ba8a3b | 2013-03-11 10:00:43 +0000 | [diff] [blame] | 2981 | if (icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS || |
| 2982 | icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) { |
Yuchung Cheng | 750ea2b | 2012-05-02 13:30:04 +0000 | [diff] [blame] | 2983 | struct sk_buff *skb = tcp_write_queue_head(sk); |
Eric Dumazet | 7faee5c | 2014-09-05 15:33:33 -0700 | [diff] [blame] | 2984 | const u32 rto_time_stamp = |
| 2985 | tcp_skb_timestamp(skb) + rto; |
Yuchung Cheng | 750ea2b | 2012-05-02 13:30:04 +0000 | [diff] [blame] | 2986 | s32 delta = (s32)(rto_time_stamp - tcp_time_stamp); |
| 2987 | /* delta may not be positive if the socket is locked |
Nandita Dukkipati | 6ba8a3b | 2013-03-11 10:00:43 +0000 | [diff] [blame] | 2988 | * when the retrans timer fires and is rescheduled. |
Yuchung Cheng | 750ea2b | 2012-05-02 13:30:04 +0000 | [diff] [blame] | 2989 | */ |
| 2990 | if (delta > 0) |
| 2991 | rto = delta; |
| 2992 | } |
| 2993 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, rto, |
| 2994 | TCP_RTO_MAX); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2995 | } |
Yuchung Cheng | 750ea2b | 2012-05-02 13:30:04 +0000 | [diff] [blame] | 2996 | } |
| 2997 | |
| 2998 | /* This function is called when the delayed ER timer fires. TCP enters |
| 2999 | * fast recovery and performs fast-retransmit. |
| 3000 | */ |
| 3001 | void tcp_resume_early_retransmit(struct sock *sk) |
| 3002 | { |
| 3003 | struct tcp_sock *tp = tcp_sk(sk); |
| 3004 | |
| 3005 | tcp_rearm_rto(sk); |
| 3006 | |
| 3007 | /* Stop if ER is disabled after the delayed ER timer is scheduled */ |
| 3008 | if (!tp->do_early_retrans) |
| 3009 | return; |
| 3010 | |
| 3011 | tcp_enter_recovery(sk, false); |
| 3012 | tcp_update_scoreboard(sk, 1); |
| 3013 | tcp_xmit_retransmit_queue(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3014 | } |
| 3015 | |
Ilpo Järvinen | 7c46a03 | 2007-09-20 11:33:43 -0700 | [diff] [blame] | 3016 | /* If we get here, the whole TSO packet has not been acked. */ |
Ilpo Järvinen | 13fcf85 | 2007-10-09 01:28:45 -0700 | [diff] [blame] | 3017 | static u32 tcp_tso_acked(struct sock *sk, struct sk_buff *skb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3018 | { |
| 3019 | struct tcp_sock *tp = tcp_sk(sk); |
Ilpo Järvinen | 7c46a03 | 2007-09-20 11:33:43 -0700 | [diff] [blame] | 3020 | u32 packets_acked; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3021 | |
Ilpo Järvinen | 7c46a03 | 2007-09-20 11:33:43 -0700 | [diff] [blame] | 3022 | BUG_ON(!after(TCP_SKB_CB(skb)->end_seq, tp->snd_una)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3023 | |
| 3024 | packets_acked = tcp_skb_pcount(skb); |
Ilpo Järvinen | 7c46a03 | 2007-09-20 11:33:43 -0700 | [diff] [blame] | 3025 | 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] | 3026 | return 0; |
| 3027 | packets_acked -= tcp_skb_pcount(skb); |
| 3028 | |
| 3029 | if (packets_acked) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3030 | BUG_ON(tcp_skb_pcount(skb) == 0); |
Ilpo Järvinen | 7c46a03 | 2007-09-20 11:33:43 -0700 | [diff] [blame] | 3031 | 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] | 3032 | } |
| 3033 | |
Ilpo Järvinen | 13fcf85 | 2007-10-09 01:28:45 -0700 | [diff] [blame] | 3034 | return packets_acked; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3035 | } |
| 3036 | |
Eric Dumazet | ad971f6 | 2014-10-11 15:17:29 -0700 | [diff] [blame] | 3037 | static void tcp_ack_tstamp(struct sock *sk, struct sk_buff *skb, |
| 3038 | u32 prior_snd_una) |
| 3039 | { |
| 3040 | const struct skb_shared_info *shinfo; |
| 3041 | |
| 3042 | /* Avoid cache line misses to get skb_shinfo() and shinfo->tx_flags */ |
| 3043 | if (likely(!(sk->sk_tsflags & SOF_TIMESTAMPING_TX_ACK))) |
| 3044 | return; |
| 3045 | |
| 3046 | shinfo = skb_shinfo(skb); |
| 3047 | if ((shinfo->tx_flags & SKBTX_ACK_TSTAMP) && |
| 3048 | between(shinfo->tskey, prior_snd_una, tcp_sk(sk)->snd_una - 1)) |
| 3049 | __skb_tstamp_tx(skb, NULL, sk, SCM_TSTAMP_ACK); |
| 3050 | } |
| 3051 | |
Ilpo Järvinen | 7c46a03 | 2007-09-20 11:33:43 -0700 | [diff] [blame] | 3052 | /* Remove acknowledged frames from the retransmission queue. If our packet |
| 3053 | * is before the ack sequence we can discard it as it's confirmed to have |
| 3054 | * arrived at the other end. |
| 3055 | */ |
Ilpo Järvinen | 33f5f57 | 2008-10-07 14:43:06 -0700 | [diff] [blame] | 3056 | static int tcp_clean_rtx_queue(struct sock *sk, int prior_fackets, |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 3057 | u32 prior_snd_una, long sack_rtt_us) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3058 | { |
Stephen Hemminger | 2d2abba | 2005-11-10 16:56:12 -0800 | [diff] [blame] | 3059 | const struct inet_connection_sock *icsk = inet_csk(sk); |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 3060 | struct skb_mstamp first_ackt, last_ackt, now; |
| 3061 | struct tcp_sock *tp = tcp_sk(sk); |
Ilpo Järvinen | 90638a0 | 2008-09-20 21:25:52 -0700 | [diff] [blame] | 3062 | u32 prior_sacked = tp->sacked_out; |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 3063 | u32 reord = tp->packets_out; |
| 3064 | bool fully_acked = true; |
| 3065 | long ca_seq_rtt_us = -1L; |
| 3066 | long seq_rtt_us = -1L; |
| 3067 | struct sk_buff *skb; |
| 3068 | u32 pkts_acked = 0; |
Yuchung Cheng | 2f715c1 | 2013-10-24 08:59:27 -0700 | [diff] [blame] | 3069 | bool rtt_update; |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 3070 | int flag = 0; |
| 3071 | |
| 3072 | first_ackt.v64 = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3073 | |
Ilpo Järvinen | 7c46a03 | 2007-09-20 11:33:43 -0700 | [diff] [blame] | 3074 | while ((skb = tcp_write_queue_head(sk)) && skb != tcp_send_head(sk)) { |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 3075 | struct tcp_skb_cb *scb = TCP_SKB_CB(skb); |
Ilpo Järvinen | 7c46a03 | 2007-09-20 11:33:43 -0700 | [diff] [blame] | 3076 | u8 sacked = scb->sacked; |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 3077 | u32 acked_pcount; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3078 | |
Eric Dumazet | ad971f6 | 2014-10-11 15:17:29 -0700 | [diff] [blame] | 3079 | tcp_ack_tstamp(sk, skb, prior_snd_una); |
Willem de Bruijn | 712a722 | 2014-08-12 14:53:16 -0400 | [diff] [blame] | 3080 | |
Gavin McCullagh | 2072c22 | 2007-12-29 19:11:21 -0800 | [diff] [blame] | 3081 | /* Determine how many packets and what bytes were acked, tso and else */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3082 | if (after(scb->end_seq, tp->snd_una)) { |
Ilpo Järvinen | 13fcf85 | 2007-10-09 01:28:45 -0700 | [diff] [blame] | 3083 | if (tcp_skb_pcount(skb) == 1 || |
| 3084 | !after(tp->snd_una, scb->seq)) |
| 3085 | break; |
| 3086 | |
Ilpo Järvinen | 7201883 | 2007-12-30 04:37:55 -0800 | [diff] [blame] | 3087 | acked_pcount = tcp_tso_acked(sk, skb); |
| 3088 | if (!acked_pcount) |
Ilpo Järvinen | 13fcf85 | 2007-10-09 01:28:45 -0700 | [diff] [blame] | 3089 | break; |
| 3090 | |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 3091 | fully_acked = false; |
Ilpo Järvinen | 13fcf85 | 2007-10-09 01:28:45 -0700 | [diff] [blame] | 3092 | } else { |
Eric Dumazet | ad971f6 | 2014-10-11 15:17:29 -0700 | [diff] [blame] | 3093 | /* Speedup tcp_unlink_write_queue() and next loop */ |
| 3094 | prefetchw(skb->next); |
Ilpo Järvinen | 7201883 | 2007-12-30 04:37:55 -0800 | [diff] [blame] | 3095 | acked_pcount = tcp_skb_pcount(skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3096 | } |
| 3097 | |
Eric Dumazet | ad971f6 | 2014-10-11 15:17:29 -0700 | [diff] [blame] | 3098 | if (unlikely(sacked & TCPCB_RETRANS)) { |
Ilpo Järvinen | 89d478f | 2007-12-30 04:35:27 -0800 | [diff] [blame] | 3099 | if (sacked & TCPCB_SACKED_RETRANS) |
Ilpo Järvinen | 7201883 | 2007-12-30 04:37:55 -0800 | [diff] [blame] | 3100 | tp->retrans_out -= acked_pcount; |
Ilpo Järvinen | 89d478f | 2007-12-30 04:35:27 -0800 | [diff] [blame] | 3101 | flag |= FLAG_RETRANS_DATA_ACKED; |
Ilpo Järvinen | c7caf8d | 2007-11-10 21:22:18 -0800 | [diff] [blame] | 3102 | } else { |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 3103 | last_ackt = skb->skb_mstamp; |
Eric Dumazet | 431a912 | 2014-03-09 17:36:02 -0700 | [diff] [blame] | 3104 | WARN_ON_ONCE(last_ackt.v64 == 0); |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 3105 | if (!first_ackt.v64) |
| 3106 | first_ackt = last_ackt; |
| 3107 | |
Ilpo Järvinen | 89d478f | 2007-12-30 04:35:27 -0800 | [diff] [blame] | 3108 | if (!(sacked & TCPCB_SACKED_ACKED)) |
Ilpo Järvinen | 7201883 | 2007-12-30 04:37:55 -0800 | [diff] [blame] | 3109 | reord = min(pkts_acked, reord); |
Yuchung Cheng | e33099f | 2013-03-20 13:33:00 +0000 | [diff] [blame] | 3110 | if (!after(scb->end_seq, tp->high_seq)) |
| 3111 | flag |= FLAG_ORIG_SACK_ACKED; |
Stephen Hemminger | 2d2abba | 2005-11-10 16:56:12 -0800 | [diff] [blame] | 3112 | } |
Ilpo Järvinen | 89d478f | 2007-12-30 04:35:27 -0800 | [diff] [blame] | 3113 | |
| 3114 | if (sacked & TCPCB_SACKED_ACKED) |
Ilpo Järvinen | 7201883 | 2007-12-30 04:37:55 -0800 | [diff] [blame] | 3115 | tp->sacked_out -= acked_pcount; |
Ilpo Järvinen | 89d478f | 2007-12-30 04:35:27 -0800 | [diff] [blame] | 3116 | if (sacked & TCPCB_LOST) |
Ilpo Järvinen | 7201883 | 2007-12-30 04:37:55 -0800 | [diff] [blame] | 3117 | tp->lost_out -= acked_pcount; |
Ilpo Järvinen | 89d478f | 2007-12-30 04:35:27 -0800 | [diff] [blame] | 3118 | |
Ilpo Järvinen | 7201883 | 2007-12-30 04:37:55 -0800 | [diff] [blame] | 3119 | tp->packets_out -= acked_pcount; |
| 3120 | pkts_acked += acked_pcount; |
Ilpo Järvinen | 13fcf85 | 2007-10-09 01:28:45 -0700 | [diff] [blame] | 3121 | |
Ilpo Järvinen | 009a2e3 | 2007-09-20 11:34:38 -0700 | [diff] [blame] | 3122 | /* Initial outgoing SYN's get put onto the write_queue |
| 3123 | * just like anything else we transmit. It is not |
| 3124 | * true data, and if we misinform our callers that |
| 3125 | * this ACK acks real data, we will erroneously exit |
| 3126 | * connection startup slow start one packet too |
| 3127 | * quickly. This is severely frowned upon behavior. |
| 3128 | */ |
Eric Dumazet | ad971f6 | 2014-10-11 15:17:29 -0700 | [diff] [blame] | 3129 | if (likely(!(scb->tcp_flags & TCPHDR_SYN))) { |
Ilpo Järvinen | 009a2e3 | 2007-09-20 11:34:38 -0700 | [diff] [blame] | 3130 | flag |= FLAG_DATA_ACKED; |
| 3131 | } else { |
| 3132 | flag |= FLAG_SYN_ACKED; |
| 3133 | tp->retrans_stamp = 0; |
| 3134 | } |
| 3135 | |
Ilpo Järvinen | 13fcf85 | 2007-10-09 01:28:45 -0700 | [diff] [blame] | 3136 | if (!fully_acked) |
| 3137 | break; |
| 3138 | |
David S. Miller | fe067e8 | 2007-03-07 12:12:44 -0800 | [diff] [blame] | 3139 | tcp_unlink_write_queue(skb, sk); |
Hideo Aoki | 3ab224b | 2007-12-31 00:11:19 -0800 | [diff] [blame] | 3140 | sk_wmem_free_skb(sk, skb); |
Eric Dumazet | ad971f6 | 2014-10-11 15:17:29 -0700 | [diff] [blame] | 3141 | if (unlikely(skb == tp->retransmit_skb_hint)) |
Ilpo Järvinen | ef9da47 | 2008-09-20 21:25:15 -0700 | [diff] [blame] | 3142 | tp->retransmit_skb_hint = NULL; |
Eric Dumazet | ad971f6 | 2014-10-11 15:17:29 -0700 | [diff] [blame] | 3143 | if (unlikely(skb == tp->lost_skb_hint)) |
Ilpo Järvinen | 90638a0 | 2008-09-20 21:25:52 -0700 | [diff] [blame] | 3144 | tp->lost_skb_hint = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3145 | } |
| 3146 | |
Ilpo Järvinen | 33f5f57 | 2008-10-07 14:43:06 -0700 | [diff] [blame] | 3147 | if (likely(between(tp->snd_up, prior_snd_una, tp->snd_una))) |
| 3148 | tp->snd_up = tp->snd_una; |
| 3149 | |
Ilpo Järvinen | cadbd03 | 2007-12-31 04:49:21 -0800 | [diff] [blame] | 3150 | if (skb && (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)) |
| 3151 | flag |= FLAG_SACK_RENEGING; |
| 3152 | |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 3153 | skb_mstamp_get(&now); |
Eric Dumazet | ad971f6 | 2014-10-11 15:17:29 -0700 | [diff] [blame] | 3154 | if (likely(first_ackt.v64)) { |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 3155 | seq_rtt_us = skb_mstamp_us_delta(&now, &first_ackt); |
| 3156 | ca_seq_rtt_us = skb_mstamp_us_delta(&now, &last_ackt); |
| 3157 | } |
| 3158 | |
| 3159 | rtt_update = tcp_ack_update_rtt(sk, flag, seq_rtt_us, sack_rtt_us); |
Yuchung Cheng | ed08495 | 2013-07-22 16:20:48 -0700 | [diff] [blame] | 3160 | |
Ilpo Järvinen | 7c46a03 | 2007-09-20 11:33:43 -0700 | [diff] [blame] | 3161 | if (flag & FLAG_ACKED) { |
Stephen Hemminger | 164891a | 2007-04-23 22:26:16 -0700 | [diff] [blame] | 3162 | const struct tcp_congestion_ops *ca_ops |
| 3163 | = inet_csk(sk)->icsk_ca_ops; |
| 3164 | |
Yuchung Cheng | 2f715c1 | 2013-10-24 08:59:27 -0700 | [diff] [blame] | 3165 | tcp_rearm_rto(sk); |
Ilpo Järvinen | 72211e9 | 2009-03-14 14:23:04 +0000 | [diff] [blame] | 3166 | if (unlikely(icsk->icsk_mtup.probe_size && |
| 3167 | !after(tp->mtu_probe.probe_seq_end, tp->snd_una))) { |
| 3168 | tcp_mtup_probe_success(sk); |
| 3169 | } |
| 3170 | |
Ilpo Järvinen | c7caf8d | 2007-11-10 21:22:18 -0800 | [diff] [blame] | 3171 | if (tcp_is_reno(tp)) { |
| 3172 | tcp_remove_reno_sacks(sk, pkts_acked); |
| 3173 | } else { |
Ilpo Järvinen | 59a08cb | 2009-02-28 04:44:28 +0000 | [diff] [blame] | 3174 | int delta; |
| 3175 | |
Ilpo Järvinen | c7caf8d | 2007-11-10 21:22:18 -0800 | [diff] [blame] | 3176 | /* Non-retransmitted hole got filled? That's reordering */ |
| 3177 | if (reord < prior_fackets) |
| 3178 | tcp_update_reordering(sk, tp->fackets_out - reord, 0); |
Ilpo Järvinen | 90638a0 | 2008-09-20 21:25:52 -0700 | [diff] [blame] | 3179 | |
Ilpo Järvinen | 59a08cb | 2009-02-28 04:44:28 +0000 | [diff] [blame] | 3180 | delta = tcp_is_fack(tp) ? pkts_acked : |
| 3181 | prior_sacked - tp->sacked_out; |
| 3182 | tp->lost_cnt_hint -= min(tp->lost_cnt_hint, delta); |
Ilpo Järvinen | c7caf8d | 2007-11-10 21:22:18 -0800 | [diff] [blame] | 3183 | } |
| 3184 | |
Ilpo Järvinen | 91fed7a | 2007-10-09 01:24:15 -0700 | [diff] [blame] | 3185 | tp->fackets_out -= min(pkts_acked, tp->fackets_out); |
Ilpo Järvinen | 68f8353 | 2007-11-15 19:50:37 -0800 | [diff] [blame] | 3186 | |
Kenneth Klette Jonassen | 932eb76 | 2015-01-29 20:08:03 +0100 | [diff] [blame] | 3187 | if (ca_ops->pkts_acked) { |
| 3188 | long rtt_us = min_t(ulong, ca_seq_rtt_us, sack_rtt_us); |
| 3189 | ca_ops->pkts_acked(sk, pkts_acked, rtt_us); |
| 3190 | } |
Ilpo Järvinen | b9ce204 | 2007-06-15 15:08:43 -0700 | [diff] [blame] | 3191 | |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 3192 | } else if (skb && rtt_update && sack_rtt_us >= 0 && |
| 3193 | sack_rtt_us > skb_mstamp_us_delta(&now, &skb->skb_mstamp)) { |
Yuchung Cheng | 2f715c1 | 2013-10-24 08:59:27 -0700 | [diff] [blame] | 3194 | /* Do not re-arm RTO if the sack RTT is measured from data sent |
| 3195 | * after when the head was last (re)transmitted. Otherwise the |
| 3196 | * timeout may continue to extend in loss recovery. |
| 3197 | */ |
| 3198 | tcp_rearm_rto(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3199 | } |
| 3200 | |
| 3201 | #if FASTRETRANS_DEBUG > 0 |
Ilpo Järvinen | 547b792 | 2008-07-25 21:43:18 -0700 | [diff] [blame] | 3202 | WARN_ON((int)tp->sacked_out < 0); |
| 3203 | WARN_ON((int)tp->lost_out < 0); |
| 3204 | WARN_ON((int)tp->retrans_out < 0); |
Ilpo Järvinen | e60402d | 2007-08-09 15:14:46 +0300 | [diff] [blame] | 3205 | if (!tp->packets_out && tcp_is_sack(tp)) { |
Stephen Hemminger | cfcabdc | 2007-10-09 01:59:42 -0700 | [diff] [blame] | 3206 | icsk = inet_csk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3207 | if (tp->lost_out) { |
Joe Perches | 91df42b | 2012-05-15 14:11:54 +0000 | [diff] [blame] | 3208 | pr_debug("Leak l=%u %d\n", |
| 3209 | tp->lost_out, icsk->icsk_ca_state); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3210 | tp->lost_out = 0; |
| 3211 | } |
| 3212 | if (tp->sacked_out) { |
Joe Perches | 91df42b | 2012-05-15 14:11:54 +0000 | [diff] [blame] | 3213 | pr_debug("Leak s=%u %d\n", |
| 3214 | tp->sacked_out, icsk->icsk_ca_state); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3215 | tp->sacked_out = 0; |
| 3216 | } |
| 3217 | if (tp->retrans_out) { |
Joe Perches | 91df42b | 2012-05-15 14:11:54 +0000 | [diff] [blame] | 3218 | pr_debug("Leak r=%u %d\n", |
| 3219 | tp->retrans_out, icsk->icsk_ca_state); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3220 | tp->retrans_out = 0; |
| 3221 | } |
| 3222 | } |
| 3223 | #endif |
Ilpo Järvinen | 7c46a03 | 2007-09-20 11:33:43 -0700 | [diff] [blame] | 3224 | return flag; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3225 | } |
| 3226 | |
| 3227 | static void tcp_ack_probe(struct sock *sk) |
| 3228 | { |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 3229 | const struct tcp_sock *tp = tcp_sk(sk); |
| 3230 | struct inet_connection_sock *icsk = inet_csk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3231 | |
| 3232 | /* Was it a usable window open? */ |
| 3233 | |
Ilpo Järvinen | 90840de | 2007-12-31 04:48:41 -0800 | [diff] [blame] | 3234 | 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] | 3235 | icsk->icsk_backoff = 0; |
| 3236 | inet_csk_clear_xmit_timer(sk, ICSK_TIME_PROBE0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3237 | /* Socket must be waked up by subsequent tcp_data_snd_check(). |
| 3238 | * This function is not for random using! |
| 3239 | */ |
| 3240 | } else { |
Eric Dumazet | fcdd1cf | 2014-09-22 13:19:44 -0700 | [diff] [blame] | 3241 | unsigned long when = inet_csk_rto_backoff(icsk, TCP_RTO_MAX); |
| 3242 | |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 3243 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, |
Eric Dumazet | fcdd1cf | 2014-09-22 13:19:44 -0700 | [diff] [blame] | 3244 | when, TCP_RTO_MAX); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3245 | } |
| 3246 | } |
| 3247 | |
Vijay Subramanian | 67b95bd | 2012-07-19 21:32:18 +0000 | [diff] [blame] | 3248 | 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] | 3249 | { |
Eric Dumazet | a02cec2 | 2010-09-22 20:43:57 +0000 | [diff] [blame] | 3250 | return !(flag & FLAG_NOT_DUP) || (flag & FLAG_CA_ALERT) || |
| 3251 | inet_csk(sk)->icsk_ca_state != TCP_CA_Open; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3252 | } |
| 3253 | |
Yuchung Cheng | 0f7cc9a | 2013-08-21 17:29:23 -0700 | [diff] [blame] | 3254 | /* Decide wheather to run the increase function of congestion control. */ |
Vijay Subramanian | 67b95bd | 2012-07-19 21:32:18 +0000 | [diff] [blame] | 3255 | 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] | 3256 | { |
Yuchung Cheng | 0f7cc9a | 2013-08-21 17:29:23 -0700 | [diff] [blame] | 3257 | if (tcp_in_cwnd_reduction(sk)) |
| 3258 | return false; |
| 3259 | |
| 3260 | /* If reordering is high then always grow cwnd whenever data is |
| 3261 | * delivered regardless of its ordering. Otherwise stay conservative |
Yuchung Cheng | 16edfe7 | 2013-09-05 15:36:09 -0700 | [diff] [blame] | 3262 | * 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] | 3263 | * new SACK or ECE mark may first advance cwnd here and later reduce |
| 3264 | * cwnd in tcp_fastretrans_alert() based on more states. |
| 3265 | */ |
| 3266 | if (tcp_sk(sk)->reordering > sysctl_tcp_reordering) |
| 3267 | return flag & FLAG_FORWARD_PROGRESS; |
| 3268 | |
Yuchung Cheng | 16edfe7 | 2013-09-05 15:36:09 -0700 | [diff] [blame] | 3269 | return flag & FLAG_DATA_ACKED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3270 | } |
| 3271 | |
| 3272 | /* Check that window update is acceptable. |
| 3273 | * The function assumes that snd_una<=ack<=snd_next. |
| 3274 | */ |
Vijay Subramanian | 67b95bd | 2012-07-19 21:32:18 +0000 | [diff] [blame] | 3275 | static inline bool tcp_may_update_window(const struct tcp_sock *tp, |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 3276 | const u32 ack, const u32 ack_seq, |
| 3277 | const u32 nwin) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3278 | { |
Eric Dumazet | a02cec2 | 2010-09-22 20:43:57 +0000 | [diff] [blame] | 3279 | return after(ack, tp->snd_una) || |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3280 | after(ack_seq, tp->snd_wl1) || |
Eric Dumazet | a02cec2 | 2010-09-22 20:43:57 +0000 | [diff] [blame] | 3281 | (ack_seq == tp->snd_wl1 && nwin > tp->snd_wnd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3282 | } |
| 3283 | |
| 3284 | /* Update our send window. |
| 3285 | * |
| 3286 | * Window update algorithm, described in RFC793/RFC1122 (used in linux-2.2 |
| 3287 | * and in FreeBSD. NetBSD's one is even worse.) is wrong. |
| 3288 | */ |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 3289 | 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] | 3290 | u32 ack_seq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3291 | { |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 3292 | struct tcp_sock *tp = tcp_sk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3293 | int flag = 0; |
Arnaldo Carvalho de Melo | aa8223c | 2007-04-10 21:04:22 -0700 | [diff] [blame] | 3294 | u32 nwin = ntohs(tcp_hdr(skb)->window); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3295 | |
Arnaldo Carvalho de Melo | aa8223c | 2007-04-10 21:04:22 -0700 | [diff] [blame] | 3296 | if (likely(!tcp_hdr(skb)->syn)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3297 | nwin <<= tp->rx_opt.snd_wscale; |
| 3298 | |
| 3299 | if (tcp_may_update_window(tp, ack, ack_seq, nwin)) { |
| 3300 | flag |= FLAG_WIN_UPDATE; |
Hantzis Fotis | ee7537b | 2009-03-02 22:42:02 -0800 | [diff] [blame] | 3301 | tcp_update_wl(tp, ack_seq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3302 | |
| 3303 | if (tp->snd_wnd != nwin) { |
| 3304 | tp->snd_wnd = nwin; |
| 3305 | |
| 3306 | /* Note, it is the only place, where |
| 3307 | * fast path is recovered for sending TCP. |
| 3308 | */ |
Herbert Xu | 2ad4106 | 2005-10-27 18:47:46 +1000 | [diff] [blame] | 3309 | tp->pred_flags = 0; |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 3310 | tcp_fast_path_check(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3311 | |
| 3312 | if (nwin > tp->max_window) { |
| 3313 | tp->max_window = nwin; |
Arnaldo Carvalho de Melo | d83d846 | 2005-12-13 23:26:10 -0800 | [diff] [blame] | 3314 | tcp_sync_mss(sk, inet_csk(sk)->icsk_pmtu_cookie); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3315 | } |
| 3316 | } |
| 3317 | } |
| 3318 | |
| 3319 | tp->snd_una = ack; |
| 3320 | |
| 3321 | return flag; |
| 3322 | } |
| 3323 | |
Eric Dumazet | 354e4aa | 2012-10-21 19:57:11 +0000 | [diff] [blame] | 3324 | /* RFC 5961 7 [ACK Throttling] */ |
Neal Cardwell | f2b2c58 | 2015-02-06 16:04:40 -0500 | [diff] [blame] | 3325 | static void tcp_send_challenge_ack(struct sock *sk, const struct sk_buff *skb) |
Eric Dumazet | 354e4aa | 2012-10-21 19:57:11 +0000 | [diff] [blame] | 3326 | { |
| 3327 | /* unprotected vars, we dont care of overwrites */ |
| 3328 | static u32 challenge_timestamp; |
| 3329 | static unsigned int challenge_count; |
Neal Cardwell | f2b2c58 | 2015-02-06 16:04:40 -0500 | [diff] [blame] | 3330 | struct tcp_sock *tp = tcp_sk(sk); |
| 3331 | u32 now; |
Eric Dumazet | 354e4aa | 2012-10-21 19:57:11 +0000 | [diff] [blame] | 3332 | |
Neal Cardwell | f2b2c58 | 2015-02-06 16:04:40 -0500 | [diff] [blame] | 3333 | /* First check our per-socket dupack rate limit. */ |
| 3334 | if (tcp_oow_rate_limited(sock_net(sk), skb, |
| 3335 | LINUX_MIB_TCPACKSKIPPEDCHALLENGE, |
| 3336 | &tp->last_oow_ack_time)) |
| 3337 | return; |
| 3338 | |
| 3339 | /* Then check the check host-wide RFC 5961 rate limit. */ |
| 3340 | now = jiffies / HZ; |
Eric Dumazet | 354e4aa | 2012-10-21 19:57:11 +0000 | [diff] [blame] | 3341 | if (now != challenge_timestamp) { |
| 3342 | challenge_timestamp = now; |
| 3343 | challenge_count = 0; |
| 3344 | } |
| 3345 | if (++challenge_count <= sysctl_tcp_challenge_ack_limit) { |
| 3346 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPCHALLENGEACK); |
| 3347 | tcp_send_ack(sk); |
| 3348 | } |
| 3349 | } |
| 3350 | |
Eric Dumazet | 12fb3dd | 2013-04-19 07:19:48 +0000 | [diff] [blame] | 3351 | static void tcp_store_ts_recent(struct tcp_sock *tp) |
| 3352 | { |
| 3353 | tp->rx_opt.ts_recent = tp->rx_opt.rcv_tsval; |
| 3354 | tp->rx_opt.ts_recent_stamp = get_seconds(); |
| 3355 | } |
| 3356 | |
| 3357 | static void tcp_replace_ts_recent(struct tcp_sock *tp, u32 seq) |
| 3358 | { |
| 3359 | if (tp->rx_opt.saw_tstamp && !after(seq, tp->rcv_wup)) { |
| 3360 | /* PAWS bug workaround wrt. ACK frames, the PAWS discard |
| 3361 | * extra check below makes sure this can only happen |
| 3362 | * for pure ACK frames. -DaveM |
| 3363 | * |
| 3364 | * Not only, also it occurs for expired timestamps. |
| 3365 | */ |
| 3366 | |
| 3367 | if (tcp_paws_check(&tp->rx_opt, 0)) |
| 3368 | tcp_store_ts_recent(tp); |
| 3369 | } |
| 3370 | } |
| 3371 | |
Nandita Dukkipati | 9b717a8 | 2013-03-11 10:00:44 +0000 | [diff] [blame] | 3372 | /* This routine deals with acks during a TLP episode. |
Sébastien Barré | 08abdff | 2015-01-12 10:30:40 +0100 | [diff] [blame] | 3373 | * We mark the end of a TLP episode on receiving TLP dupack or when |
| 3374 | * ack is after tlp_high_seq. |
Nandita Dukkipati | 9b717a8 | 2013-03-11 10:00:44 +0000 | [diff] [blame] | 3375 | * Ref: loss detection algorithm in draft-dukkipati-tcpm-tcp-loss-probe. |
| 3376 | */ |
| 3377 | static void tcp_process_tlp_ack(struct sock *sk, u32 ack, int flag) |
| 3378 | { |
| 3379 | struct tcp_sock *tp = tcp_sk(sk); |
Nandita Dukkipati | 9b717a8 | 2013-03-11 10:00:44 +0000 | [diff] [blame] | 3380 | |
Sébastien Barré | 08abdff | 2015-01-12 10:30:40 +0100 | [diff] [blame] | 3381 | if (before(ack, tp->tlp_high_seq)) |
Nandita Dukkipati | 9b717a8 | 2013-03-11 10:00:44 +0000 | [diff] [blame] | 3382 | return; |
Nandita Dukkipati | 9b717a8 | 2013-03-11 10:00:44 +0000 | [diff] [blame] | 3383 | |
Sébastien Barré | 08abdff | 2015-01-12 10:30:40 +0100 | [diff] [blame] | 3384 | if (flag & FLAG_DSACKING_ACK) { |
| 3385 | /* This DSACK means original and TLP probe arrived; no loss */ |
Nandita Dukkipati | 9b717a8 | 2013-03-11 10:00:44 +0000 | [diff] [blame] | 3386 | tp->tlp_high_seq = 0; |
Sébastien Barré | 08abdff | 2015-01-12 10:30:40 +0100 | [diff] [blame] | 3387 | } else if (after(ack, tp->tlp_high_seq)) { |
| 3388 | /* ACK advances: there was a loss, so reduce cwnd. Reset |
| 3389 | * tlp_high_seq in tcp_init_cwnd_reduction() |
| 3390 | */ |
| 3391 | tcp_init_cwnd_reduction(sk); |
| 3392 | tcp_set_ca_state(sk, TCP_CA_CWR); |
| 3393 | tcp_end_cwnd_reduction(sk); |
| 3394 | tcp_try_keep_open(sk); |
| 3395 | NET_INC_STATS_BH(sock_net(sk), |
| 3396 | LINUX_MIB_TCPLOSSPROBERECOVERY); |
| 3397 | } else if (!(flag & (FLAG_SND_UNA_ADVANCED | |
| 3398 | FLAG_NOT_DUP | FLAG_DATA_SACKED))) { |
| 3399 | /* Pure dupack: original and TLP probe arrived; no loss */ |
| 3400 | tp->tlp_high_seq = 0; |
Nandita Dukkipati | 9b717a8 | 2013-03-11 10:00:44 +0000 | [diff] [blame] | 3401 | } |
| 3402 | } |
| 3403 | |
Florian Westphal | 7354c8c | 2014-09-26 22:37:34 +0200 | [diff] [blame] | 3404 | static inline void tcp_in_ack_event(struct sock *sk, u32 flags) |
| 3405 | { |
| 3406 | const struct inet_connection_sock *icsk = inet_csk(sk); |
| 3407 | |
| 3408 | if (icsk->icsk_ca_ops->in_ack_event) |
| 3409 | icsk->icsk_ca_ops->in_ack_event(sk, flags); |
| 3410 | } |
| 3411 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3412 | /* This routine deals with incoming acks, but not outgoing ones. */ |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 3413 | 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] | 3414 | { |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 3415 | struct inet_connection_sock *icsk = inet_csk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3416 | struct tcp_sock *tp = tcp_sk(sk); |
| 3417 | u32 prior_snd_una = tp->snd_una; |
| 3418 | u32 ack_seq = TCP_SKB_CB(skb)->seq; |
| 3419 | u32 ack = TCP_SKB_CB(skb)->ack_seq; |
Neal Cardwell | 7d2b55f | 2011-11-16 08:58:01 +0000 | [diff] [blame] | 3420 | bool is_dupack = false; |
Ilpo Järvinen | c7caf8d | 2007-11-10 21:22:18 -0800 | [diff] [blame] | 3421 | u32 prior_fackets; |
Nandita Dukkipati | 35f079e | 2013-05-21 15:12:07 +0000 | [diff] [blame] | 3422 | int prior_packets = tp->packets_out; |
Yuchung Cheng | 6804973 | 2013-05-29 14:20:11 +0000 | [diff] [blame] | 3423 | const int prior_unsacked = tp->packets_out - tp->sacked_out; |
| 3424 | int acked = 0; /* Number of packets newly acked */ |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 3425 | long sack_rtt_us = -1L; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3426 | |
Eric Dumazet | ad971f6 | 2014-10-11 15:17:29 -0700 | [diff] [blame] | 3427 | /* We very likely will need to access write queue head. */ |
| 3428 | prefetchw(sk->sk_write_queue.next); |
| 3429 | |
John Dykstra | 96e0bf4 | 2009-03-22 21:49:57 -0700 | [diff] [blame] | 3430 | /* If the ack is older than previous acks |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3431 | * then we can probably ignore it. |
| 3432 | */ |
Eric Dumazet | 354e4aa | 2012-10-21 19:57:11 +0000 | [diff] [blame] | 3433 | if (before(ack, prior_snd_una)) { |
| 3434 | /* RFC 5961 5.2 [Blind Data Injection Attack].[Mitigation] */ |
| 3435 | if (before(ack, prior_snd_una - tp->max_window)) { |
Neal Cardwell | f2b2c58 | 2015-02-06 16:04:40 -0500 | [diff] [blame] | 3436 | tcp_send_challenge_ack(sk, skb); |
Eric Dumazet | 354e4aa | 2012-10-21 19:57:11 +0000 | [diff] [blame] | 3437 | return -1; |
| 3438 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3439 | goto old_ack; |
Eric Dumazet | 354e4aa | 2012-10-21 19:57:11 +0000 | [diff] [blame] | 3440 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3441 | |
John Dykstra | 96e0bf4 | 2009-03-22 21:49:57 -0700 | [diff] [blame] | 3442 | /* If the ack includes data we haven't sent yet, discard |
| 3443 | * this segment (RFC793 Section 3.9). |
| 3444 | */ |
| 3445 | if (after(ack, tp->snd_nxt)) |
| 3446 | goto invalid_ack; |
| 3447 | |
Nandita Dukkipati | 6ba8a3b | 2013-03-11 10:00:43 +0000 | [diff] [blame] | 3448 | if (icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS || |
| 3449 | icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) |
Yuchung Cheng | 750ea2b | 2012-05-02 13:30:04 +0000 | [diff] [blame] | 3450 | tcp_rearm_rto(sk); |
| 3451 | |
Neal Cardwell | 0c9ab09 | 2014-08-14 16:13:07 -0400 | [diff] [blame] | 3452 | if (after(ack, prior_snd_una)) { |
Ilpo Järvinen | 2e60529 | 2007-08-02 19:46:58 -0700 | [diff] [blame] | 3453 | flag |= FLAG_SND_UNA_ADVANCED; |
Neal Cardwell | 0c9ab09 | 2014-08-14 16:13:07 -0400 | [diff] [blame] | 3454 | icsk->icsk_retransmits = 0; |
| 3455 | } |
Ilpo Järvinen | 2e60529 | 2007-08-02 19:46:58 -0700 | [diff] [blame] | 3456 | |
Ilpo Järvinen | c7caf8d | 2007-11-10 21:22:18 -0800 | [diff] [blame] | 3457 | prior_fackets = tp->fackets_out; |
| 3458 | |
Eric Dumazet | 12fb3dd | 2013-04-19 07:19:48 +0000 | [diff] [blame] | 3459 | /* ts_recent update must be made after we are sure that the packet |
| 3460 | * is in window. |
| 3461 | */ |
| 3462 | if (flag & FLAG_UPDATE_TS_RECENT) |
| 3463 | tcp_replace_ts_recent(tp, TCP_SKB_CB(skb)->seq); |
| 3464 | |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 3465 | if (!(flag & FLAG_SLOWPATH) && after(ack, prior_snd_una)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3466 | /* Window is constant, pure forward advance. |
| 3467 | * No more checks are required. |
| 3468 | * Note, we use the fact that SND.UNA>=SND.WL2. |
| 3469 | */ |
Hantzis Fotis | ee7537b | 2009-03-02 22:42:02 -0800 | [diff] [blame] | 3470 | tcp_update_wl(tp, ack_seq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3471 | tp->snd_una = ack; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3472 | flag |= FLAG_WIN_UPDATE; |
| 3473 | |
Florian Westphal | 9890092 | 2014-09-26 22:37:35 +0200 | [diff] [blame] | 3474 | tcp_in_ack_event(sk, CA_ACK_WIN_UPDATE); |
Stephen Hemminger | 317a76f | 2005-06-23 12:19:55 -0700 | [diff] [blame] | 3475 | |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 3476 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPHPACKS); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3477 | } else { |
Florian Westphal | 9890092 | 2014-09-26 22:37:35 +0200 | [diff] [blame] | 3478 | u32 ack_ev_flags = CA_ACK_SLOWPATH; |
| 3479 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3480 | if (ack_seq != TCP_SKB_CB(skb)->end_seq) |
| 3481 | flag |= FLAG_DATA; |
| 3482 | else |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 3483 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPPUREACKS); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3484 | |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 3485 | flag |= tcp_ack_update_window(sk, skb, ack, ack_seq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3486 | |
| 3487 | if (TCP_SKB_CB(skb)->sacked) |
Yuchung Cheng | 59c9af4 | 2013-07-22 16:20:47 -0700 | [diff] [blame] | 3488 | flag |= tcp_sacktag_write_queue(sk, skb, prior_snd_una, |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 3489 | &sack_rtt_us); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3490 | |
Florian Westphal | 735d383 | 2014-09-29 13:08:30 +0200 | [diff] [blame] | 3491 | if (tcp_ecn_rcv_ecn_echo(tp, tcp_hdr(skb))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3492 | flag |= FLAG_ECE; |
Florian Westphal | 9890092 | 2014-09-26 22:37:35 +0200 | [diff] [blame] | 3493 | ack_ev_flags |= CA_ACK_ECE; |
| 3494 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3495 | |
Florian Westphal | 9890092 | 2014-09-26 22:37:35 +0200 | [diff] [blame] | 3496 | if (flag & FLAG_WIN_UPDATE) |
| 3497 | ack_ev_flags |= CA_ACK_WIN_UPDATE; |
| 3498 | |
| 3499 | tcp_in_ack_event(sk, ack_ev_flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3500 | } |
| 3501 | |
| 3502 | /* We passed data and got it acked, remove any soft error |
| 3503 | * log. Something worked... |
| 3504 | */ |
| 3505 | sk->sk_err_soft = 0; |
David S. Miller | 4b53fb6 | 2008-07-23 16:38:45 -0700 | [diff] [blame] | 3506 | icsk->icsk_probes_out = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3507 | tp->rcv_tstamp = tcp_time_stamp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3508 | if (!prior_packets) |
| 3509 | goto no_queue; |
| 3510 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3511 | /* See if we can take anything off of the retransmit queue. */ |
Yuchung Cheng | 6804973 | 2013-05-29 14:20:11 +0000 | [diff] [blame] | 3512 | acked = tp->packets_out; |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 3513 | flag |= tcp_clean_rtx_queue(sk, prior_fackets, prior_snd_una, |
| 3514 | sack_rtt_us); |
Yuchung Cheng | 6804973 | 2013-05-29 14:20:11 +0000 | [diff] [blame] | 3515 | acked -= tp->packets_out; |
Nandita Dukkipati | a262f0c | 2011-08-21 20:21:57 +0000 | [diff] [blame] | 3516 | |
Yuchung Cheng | 0f7cc9a | 2013-08-21 17:29:23 -0700 | [diff] [blame] | 3517 | /* Advance cwnd if state allows */ |
| 3518 | if (tcp_may_raise_cwnd(sk, flag)) |
Eric Dumazet | 2490155 | 2014-05-02 21:18:05 -0700 | [diff] [blame] | 3519 | tcp_cong_avoid(sk, ack, acked); |
Yuchung Cheng | 0f7cc9a | 2013-08-21 17:29:23 -0700 | [diff] [blame] | 3520 | |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 3521 | if (tcp_ack_is_dubious(sk, flag)) { |
Neal Cardwell | 7d2b55f | 2011-11-16 08:58:01 +0000 | [diff] [blame] | 3522 | is_dupack = !(flag & (FLAG_SND_UNA_ADVANCED | FLAG_NOT_DUP)); |
Yuchung Cheng | 6804973 | 2013-05-29 14:20:11 +0000 | [diff] [blame] | 3523 | tcp_fastretrans_alert(sk, acked, prior_unsacked, |
| 3524 | is_dupack, flag); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3525 | } |
Nandita Dukkipati | 9b717a8 | 2013-03-11 10:00:44 +0000 | [diff] [blame] | 3526 | if (tp->tlp_high_seq) |
| 3527 | tcp_process_tlp_ack(sk, ack, flag); |
| 3528 | |
David S. Miller | 5110effe | 2012-07-02 02:21:03 -0700 | [diff] [blame] | 3529 | if ((flag & FLAG_FORWARD_PROGRESS) || !(flag & FLAG_NOT_DUP)) { |
| 3530 | struct dst_entry *dst = __sk_dst_get(sk); |
| 3531 | if (dst) |
| 3532 | dst_confirm(dst); |
| 3533 | } |
Nandita Dukkipati | 6ba8a3b | 2013-03-11 10:00:43 +0000 | [diff] [blame] | 3534 | |
| 3535 | if (icsk->icsk_pending == ICSK_TIME_RETRANS) |
| 3536 | tcp_schedule_loss_probe(sk); |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 3537 | tcp_update_pacing_rate(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3538 | return 1; |
| 3539 | |
| 3540 | no_queue: |
Neal Cardwell | 5628adf | 2011-11-16 08:58:02 +0000 | [diff] [blame] | 3541 | /* If data was DSACKed, see if we can undo a cwnd reduction. */ |
| 3542 | if (flag & FLAG_DSACKING_ACK) |
Yuchung Cheng | 6804973 | 2013-05-29 14:20:11 +0000 | [diff] [blame] | 3543 | tcp_fastretrans_alert(sk, acked, prior_unsacked, |
| 3544 | is_dupack, flag); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3545 | /* If this ack opens up a zero window, clear backoff. It was |
| 3546 | * being used to time the probes, and is probably far higher than |
| 3547 | * it needs to be for normal retransmission. |
| 3548 | */ |
David S. Miller | fe067e8 | 2007-03-07 12:12:44 -0800 | [diff] [blame] | 3549 | if (tcp_send_head(sk)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3550 | tcp_ack_probe(sk); |
Nandita Dukkipati | 9b717a8 | 2013-03-11 10:00:44 +0000 | [diff] [blame] | 3551 | |
| 3552 | if (tp->tlp_high_seq) |
| 3553 | tcp_process_tlp_ack(sk, ack, flag); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3554 | return 1; |
| 3555 | |
John Dykstra | 96e0bf4 | 2009-03-22 21:49:57 -0700 | [diff] [blame] | 3556 | invalid_ack: |
| 3557 | SOCK_DEBUG(sk, "Ack %u after %u:%u\n", ack, tp->snd_una, tp->snd_nxt); |
| 3558 | return -1; |
| 3559 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3560 | old_ack: |
Neal Cardwell | e95ae2f | 2011-11-16 08:58:03 +0000 | [diff] [blame] | 3561 | /* If data was SACKed, tag it and see if we should send more data. |
| 3562 | * If data was DSACKed, see if we can undo a cwnd reduction. |
| 3563 | */ |
Ilpo Järvinen | 8aca6cb | 2008-06-04 11:34:22 -0700 | [diff] [blame] | 3564 | if (TCP_SKB_CB(skb)->sacked) { |
Yuchung Cheng | 59c9af4 | 2013-07-22 16:20:47 -0700 | [diff] [blame] | 3565 | flag |= tcp_sacktag_write_queue(sk, skb, prior_snd_una, |
Eric Dumazet | 740b0f1 | 2014-02-26 14:02:48 -0800 | [diff] [blame] | 3566 | &sack_rtt_us); |
Yuchung Cheng | 6804973 | 2013-05-29 14:20:11 +0000 | [diff] [blame] | 3567 | tcp_fastretrans_alert(sk, acked, prior_unsacked, |
| 3568 | is_dupack, flag); |
Ilpo Järvinen | 8aca6cb | 2008-06-04 11:34:22 -0700 | [diff] [blame] | 3569 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3570 | |
John Dykstra | 96e0bf4 | 2009-03-22 21:49:57 -0700 | [diff] [blame] | 3571 | 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] | 3572 | return 0; |
| 3573 | } |
| 3574 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3575 | /* Look for tcp options. Normally only called on SYN and SYNACK packets. |
| 3576 | * But, this can also be called on packets in the established flow when |
| 3577 | * the fast version below fails. |
| 3578 | */ |
Christoph Paasch | 1a2c618 | 2013-03-17 08:23:34 +0000 | [diff] [blame] | 3579 | void tcp_parse_options(const struct sk_buff *skb, |
| 3580 | struct tcp_options_received *opt_rx, int estab, |
Yuchung Cheng | 2100c8d | 2012-07-19 06:43:05 +0000 | [diff] [blame] | 3581 | struct tcp_fastopen_cookie *foc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3582 | { |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 3583 | const unsigned char *ptr; |
| 3584 | const struct tcphdr *th = tcp_hdr(skb); |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 3585 | int length = (th->doff * 4) - sizeof(struct tcphdr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3586 | |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 3587 | ptr = (const unsigned char *)(th + 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3588 | opt_rx->saw_tstamp = 0; |
| 3589 | |
Stephen Hemminger | 2de979b | 2007-03-08 20:45:19 -0800 | [diff] [blame] | 3590 | while (length > 0) { |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 3591 | int opcode = *ptr++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3592 | int opsize; |
| 3593 | |
| 3594 | switch (opcode) { |
Ilpo Järvinen | f038ac8 | 2008-01-03 20:36:55 -0800 | [diff] [blame] | 3595 | case TCPOPT_EOL: |
| 3596 | return; |
| 3597 | case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */ |
| 3598 | length--; |
| 3599 | continue; |
| 3600 | default: |
| 3601 | opsize = *ptr++; |
| 3602 | if (opsize < 2) /* "silly options" */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3603 | return; |
Ilpo Järvinen | f038ac8 | 2008-01-03 20:36:55 -0800 | [diff] [blame] | 3604 | if (opsize > length) |
| 3605 | return; /* don't parse partial options */ |
| 3606 | switch (opcode) { |
| 3607 | case TCPOPT_MSS: |
| 3608 | if (opsize == TCPOLEN_MSS && th->syn && !estab) { |
Harvey Harrison | d3e2ce3 | 2008-05-02 16:26:16 -0700 | [diff] [blame] | 3609 | u16 in_mss = get_unaligned_be16(ptr); |
Ilpo Järvinen | f038ac8 | 2008-01-03 20:36:55 -0800 | [diff] [blame] | 3610 | if (in_mss) { |
| 3611 | if (opt_rx->user_mss && |
| 3612 | opt_rx->user_mss < in_mss) |
| 3613 | in_mss = opt_rx->user_mss; |
| 3614 | opt_rx->mss_clamp = in_mss; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3615 | } |
Stephen Hemminger | 3ff50b7 | 2007-04-20 17:09:22 -0700 | [diff] [blame] | 3616 | } |
Ilpo Järvinen | f038ac8 | 2008-01-03 20:36:55 -0800 | [diff] [blame] | 3617 | break; |
| 3618 | case TCPOPT_WINDOW: |
| 3619 | if (opsize == TCPOLEN_WINDOW && th->syn && |
David S. Miller | bb5b7c1 | 2009-12-15 20:56:42 -0800 | [diff] [blame] | 3620 | !estab && sysctl_tcp_window_scaling) { |
Ilpo Järvinen | f038ac8 | 2008-01-03 20:36:55 -0800 | [diff] [blame] | 3621 | __u8 snd_wscale = *(__u8 *)ptr; |
| 3622 | opt_rx->wscale_ok = 1; |
| 3623 | if (snd_wscale > 14) { |
Joe Perches | e87cc47 | 2012-05-13 21:56:26 +0000 | [diff] [blame] | 3624 | net_info_ratelimited("%s: Illegal window scaling value %d >14 received\n", |
| 3625 | __func__, |
| 3626 | snd_wscale); |
Ilpo Järvinen | f038ac8 | 2008-01-03 20:36:55 -0800 | [diff] [blame] | 3627 | snd_wscale = 14; |
| 3628 | } |
| 3629 | opt_rx->snd_wscale = snd_wscale; |
| 3630 | } |
| 3631 | break; |
| 3632 | case TCPOPT_TIMESTAMP: |
| 3633 | if ((opsize == TCPOLEN_TIMESTAMP) && |
| 3634 | ((estab && opt_rx->tstamp_ok) || |
David S. Miller | bb5b7c1 | 2009-12-15 20:56:42 -0800 | [diff] [blame] | 3635 | (!estab && sysctl_tcp_timestamps))) { |
Ilpo Järvinen | f038ac8 | 2008-01-03 20:36:55 -0800 | [diff] [blame] | 3636 | opt_rx->saw_tstamp = 1; |
Harvey Harrison | d3e2ce3 | 2008-05-02 16:26:16 -0700 | [diff] [blame] | 3637 | opt_rx->rcv_tsval = get_unaligned_be32(ptr); |
| 3638 | opt_rx->rcv_tsecr = get_unaligned_be32(ptr + 4); |
Ilpo Järvinen | f038ac8 | 2008-01-03 20:36:55 -0800 | [diff] [blame] | 3639 | } |
| 3640 | break; |
| 3641 | case TCPOPT_SACK_PERM: |
| 3642 | if (opsize == TCPOLEN_SACK_PERM && th->syn && |
David S. Miller | bb5b7c1 | 2009-12-15 20:56:42 -0800 | [diff] [blame] | 3643 | !estab && sysctl_tcp_sack) { |
Vijay Subramanian | ab56222 | 2011-12-20 13:23:24 +0000 | [diff] [blame] | 3644 | opt_rx->sack_ok = TCP_SACK_SEEN; |
Ilpo Järvinen | f038ac8 | 2008-01-03 20:36:55 -0800 | [diff] [blame] | 3645 | tcp_sack_reset(opt_rx); |
| 3646 | } |
| 3647 | break; |
Stephen Hemminger | 3ff50b7 | 2007-04-20 17:09:22 -0700 | [diff] [blame] | 3648 | |
Ilpo Järvinen | f038ac8 | 2008-01-03 20:36:55 -0800 | [diff] [blame] | 3649 | case TCPOPT_SACK: |
| 3650 | if ((opsize >= (TCPOLEN_SACK_BASE + TCPOLEN_SACK_PERBLOCK)) && |
| 3651 | !((opsize - TCPOLEN_SACK_BASE) % TCPOLEN_SACK_PERBLOCK) && |
| 3652 | opt_rx->sack_ok) { |
| 3653 | TCP_SKB_CB(skb)->sacked = (ptr - 2) - (unsigned char *)th; |
| 3654 | } |
| 3655 | break; |
| 3656 | #ifdef CONFIG_TCP_MD5SIG |
| 3657 | case TCPOPT_MD5SIG: |
| 3658 | /* |
| 3659 | * The MD5 Hash has already been |
| 3660 | * checked (see tcp_v{4,6}_do_rcv()). |
| 3661 | */ |
| 3662 | break; |
| 3663 | #endif |
Yuchung Cheng | 2100c8d | 2012-07-19 06:43:05 +0000 | [diff] [blame] | 3664 | case TCPOPT_EXP: |
| 3665 | /* Fast Open option shares code 254 using a |
| 3666 | * 16 bits magic number. It's valid only in |
| 3667 | * SYN or SYN-ACK with an even size. |
| 3668 | */ |
| 3669 | if (opsize < TCPOLEN_EXP_FASTOPEN_BASE || |
| 3670 | get_unaligned_be16(ptr) != TCPOPT_FASTOPEN_MAGIC || |
| 3671 | foc == NULL || !th->syn || (opsize & 1)) |
| 3672 | break; |
| 3673 | foc->len = opsize - TCPOLEN_EXP_FASTOPEN_BASE; |
| 3674 | if (foc->len >= TCP_FASTOPEN_COOKIE_MIN && |
| 3675 | foc->len <= TCP_FASTOPEN_COOKIE_MAX) |
| 3676 | memcpy(foc->val, ptr + 2, foc->len); |
| 3677 | else if (foc->len != 0) |
| 3678 | foc->len = -1; |
| 3679 | break; |
| 3680 | |
| 3681 | } |
Ilpo Järvinen | f038ac8 | 2008-01-03 20:36:55 -0800 | [diff] [blame] | 3682 | ptr += opsize-2; |
| 3683 | length -= opsize; |
Stephen Hemminger | 3ff50b7 | 2007-04-20 17:09:22 -0700 | [diff] [blame] | 3684 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3685 | } |
| 3686 | } |
Eric Dumazet | 4bc2f18 | 2010-07-09 21:22:10 +0000 | [diff] [blame] | 3687 | EXPORT_SYMBOL(tcp_parse_options); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3688 | |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 3689 | 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] | 3690 | { |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 3691 | const __be32 *ptr = (const __be32 *)(th + 1); |
Ilpo Järvinen | a4356b2 | 2008-08-23 05:12:29 -0700 | [diff] [blame] | 3692 | |
| 3693 | if (*ptr == htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
| 3694 | | (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP)) { |
| 3695 | tp->rx_opt.saw_tstamp = 1; |
| 3696 | ++ptr; |
| 3697 | tp->rx_opt.rcv_tsval = ntohl(*ptr); |
| 3698 | ++ptr; |
Andrew Vagin | e3e1202 | 2013-08-27 12:21:55 +0400 | [diff] [blame] | 3699 | if (*ptr) |
| 3700 | tp->rx_opt.rcv_tsecr = ntohl(*ptr) - tp->tsoffset; |
| 3701 | else |
| 3702 | tp->rx_opt.rcv_tsecr = 0; |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 3703 | return true; |
Ilpo Järvinen | a4356b2 | 2008-08-23 05:12:29 -0700 | [diff] [blame] | 3704 | } |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 3705 | return false; |
Ilpo Järvinen | a4356b2 | 2008-08-23 05:12:29 -0700 | [diff] [blame] | 3706 | } |
| 3707 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3708 | /* Fast parse options. This hopes to only see timestamps. |
| 3709 | * If it is wrong it falls back on tcp_parse_options(). |
| 3710 | */ |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 3711 | static bool tcp_fast_parse_options(const struct sk_buff *skb, |
Christoph Paasch | 1a2c618 | 2013-03-17 08:23:34 +0000 | [diff] [blame] | 3712 | const struct tcphdr *th, struct tcp_sock *tp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3713 | { |
William Allen Simpson | 4957faade | 2009-12-02 18:25:27 +0000 | [diff] [blame] | 3714 | /* In the spirit of fast parsing, compare doff directly to constant |
| 3715 | * values. Because equality is used, short doff can be ignored here. |
| 3716 | */ |
| 3717 | if (th->doff == (sizeof(*th) / 4)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3718 | tp->rx_opt.saw_tstamp = 0; |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 3719 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3720 | } else if (tp->rx_opt.tstamp_ok && |
William Allen Simpson | 4957faade | 2009-12-02 18:25:27 +0000 | [diff] [blame] | 3721 | th->doff == ((sizeof(*th) + TCPOLEN_TSTAMP_ALIGNED) / 4)) { |
Ilpo Järvinen | a4356b2 | 2008-08-23 05:12:29 -0700 | [diff] [blame] | 3722 | if (tcp_parse_aligned_timestamp(tp, th)) |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 3723 | return true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3724 | } |
Andrey Vagin | ee684b6 | 2013-02-11 05:50:19 +0000 | [diff] [blame] | 3725 | |
Christoph Paasch | 1a2c618 | 2013-03-17 08:23:34 +0000 | [diff] [blame] | 3726 | tcp_parse_options(skb, &tp->rx_opt, 1, NULL); |
Andrew Vagin | e3e1202 | 2013-08-27 12:21:55 +0400 | [diff] [blame] | 3727 | if (tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr) |
Andrey Vagin | ee684b6 | 2013-02-11 05:50:19 +0000 | [diff] [blame] | 3728 | tp->rx_opt.rcv_tsecr -= tp->tsoffset; |
| 3729 | |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 3730 | return true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3731 | } |
| 3732 | |
YOSHIFUJI Hideaki | 7d5d552 | 2008-04-17 12:29:53 +0900 | [diff] [blame] | 3733 | #ifdef CONFIG_TCP_MD5SIG |
| 3734 | /* |
| 3735 | * Parse MD5 Signature option |
| 3736 | */ |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 3737 | const u8 *tcp_parse_md5sig_option(const struct tcphdr *th) |
YOSHIFUJI Hideaki | 7d5d552 | 2008-04-17 12:29:53 +0900 | [diff] [blame] | 3738 | { |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 3739 | int length = (th->doff << 2) - sizeof(*th); |
| 3740 | const u8 *ptr = (const u8 *)(th + 1); |
YOSHIFUJI Hideaki | 7d5d552 | 2008-04-17 12:29:53 +0900 | [diff] [blame] | 3741 | |
| 3742 | /* If the TCP option is too short, we can short cut */ |
| 3743 | if (length < TCPOLEN_MD5SIG) |
| 3744 | return NULL; |
| 3745 | |
| 3746 | while (length > 0) { |
| 3747 | int opcode = *ptr++; |
| 3748 | int opsize; |
| 3749 | |
Weilong Chen | a22318e | 2013-12-23 14:37:26 +0800 | [diff] [blame] | 3750 | switch (opcode) { |
YOSHIFUJI Hideaki | 7d5d552 | 2008-04-17 12:29:53 +0900 | [diff] [blame] | 3751 | case TCPOPT_EOL: |
| 3752 | return NULL; |
| 3753 | case TCPOPT_NOP: |
| 3754 | length--; |
| 3755 | continue; |
| 3756 | default: |
| 3757 | opsize = *ptr++; |
| 3758 | if (opsize < 2 || opsize > length) |
| 3759 | return NULL; |
| 3760 | if (opcode == TCPOPT_MD5SIG) |
Dmitry Popov | ba78e2d | 2010-08-07 20:24:28 -0700 | [diff] [blame] | 3761 | return opsize == TCPOLEN_MD5SIG ? ptr : NULL; |
YOSHIFUJI Hideaki | 7d5d552 | 2008-04-17 12:29:53 +0900 | [diff] [blame] | 3762 | } |
| 3763 | ptr += opsize - 2; |
| 3764 | length -= opsize; |
| 3765 | } |
| 3766 | return NULL; |
| 3767 | } |
Eric Dumazet | 4bc2f18 | 2010-07-09 21:22:10 +0000 | [diff] [blame] | 3768 | EXPORT_SYMBOL(tcp_parse_md5sig_option); |
YOSHIFUJI Hideaki | 7d5d552 | 2008-04-17 12:29:53 +0900 | [diff] [blame] | 3769 | #endif |
| 3770 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3771 | /* Sorry, PAWS as specified is broken wrt. pure-ACKs -DaveM |
| 3772 | * |
| 3773 | * It is not fatal. If this ACK does _not_ change critical state (seqs, window) |
| 3774 | * it can pass through stack. So, the following predicate verifies that |
| 3775 | * this segment is not used for anything but congestion avoidance or |
| 3776 | * fast retransmit. Moreover, we even are able to eliminate most of such |
| 3777 | * second order effects, if we apply some small "replay" window (~RTO) |
| 3778 | * to timestamp space. |
| 3779 | * |
| 3780 | * All these measures still do not guarantee that we reject wrapped ACKs |
| 3781 | * on networks with high bandwidth, when sequence space is recycled fastly, |
| 3782 | * but it guarantees that such events will be very rare and do not affect |
| 3783 | * connection seriously. This doesn't look nice, but alas, PAWS is really |
| 3784 | * buggy extension. |
| 3785 | * |
| 3786 | * [ Later note. Even worse! It is buggy for segments _with_ data. RFC |
| 3787 | * states that events when retransmit arrives after original data are rare. |
| 3788 | * It is a blatant lie. VJ forgot about fast retransmit! 8)8) It is |
| 3789 | * the biggest problem on large power networks even with minor reordering. |
| 3790 | * OK, let's give it small replay window. If peer clock is even 1hz, it is safe |
| 3791 | * up to bandwidth of 18Gigabit/sec. 8) ] |
| 3792 | */ |
| 3793 | |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 3794 | 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] | 3795 | { |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 3796 | const struct tcp_sock *tp = tcp_sk(sk); |
| 3797 | const struct tcphdr *th = tcp_hdr(skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3798 | u32 seq = TCP_SKB_CB(skb)->seq; |
| 3799 | u32 ack = TCP_SKB_CB(skb)->ack_seq; |
| 3800 | |
| 3801 | return (/* 1. Pure ACK with correct sequence number. */ |
| 3802 | (th->ack && seq == TCP_SKB_CB(skb)->end_seq && seq == tp->rcv_nxt) && |
| 3803 | |
| 3804 | /* 2. ... and duplicate ACK. */ |
| 3805 | ack == tp->snd_una && |
| 3806 | |
| 3807 | /* 3. ... and does not update window. */ |
| 3808 | !tcp_may_update_window(tp, ack, seq, ntohs(th->window) << tp->rx_opt.snd_wscale) && |
| 3809 | |
| 3810 | /* 4. ... and sits in replay window. */ |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 3811 | (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] | 3812 | } |
| 3813 | |
Vijay Subramanian | 67b95bd | 2012-07-19 21:32:18 +0000 | [diff] [blame] | 3814 | static inline bool tcp_paws_discard(const struct sock *sk, |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 3815 | const struct sk_buff *skb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3816 | { |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 3817 | const struct tcp_sock *tp = tcp_sk(sk); |
Ilpo Järvinen | c887e6d | 2009-03-14 14:23:03 +0000 | [diff] [blame] | 3818 | |
| 3819 | return !tcp_paws_check(&tp->rx_opt, TCP_PAWS_WINDOW) && |
| 3820 | !tcp_disordered_ack(sk, skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3821 | } |
| 3822 | |
| 3823 | /* Check segment sequence number for validity. |
| 3824 | * |
| 3825 | * Segment controls are considered valid, if the segment |
| 3826 | * fits to the window after truncation to the window. Acceptability |
| 3827 | * of data (and SYN, FIN, of course) is checked separately. |
| 3828 | * See tcp_data_queue(), for example. |
| 3829 | * |
| 3830 | * Also, controls (RST is main one) are accepted using RCV.WUP instead |
| 3831 | * of RCV.NXT. Peer still did not advance his SND.UNA when we |
| 3832 | * delayed ACK, so that hisSND.UNA<=ourRCV.WUP. |
| 3833 | * (borrowed from freebsd) |
| 3834 | */ |
| 3835 | |
Vijay Subramanian | 67b95bd | 2012-07-19 21:32:18 +0000 | [diff] [blame] | 3836 | 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] | 3837 | { |
| 3838 | return !before(end_seq, tp->rcv_wup) && |
| 3839 | !after(seq, tp->rcv_nxt + tcp_receive_window(tp)); |
| 3840 | } |
| 3841 | |
| 3842 | /* When we get a reset we do this. */ |
Jerry Chu | 1046716 | 2012-08-31 12:29:11 +0000 | [diff] [blame] | 3843 | void tcp_reset(struct sock *sk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3844 | { |
| 3845 | /* We want the right error as BSD sees it (and indeed as we do). */ |
| 3846 | switch (sk->sk_state) { |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 3847 | case TCP_SYN_SENT: |
| 3848 | sk->sk_err = ECONNREFUSED; |
| 3849 | break; |
| 3850 | case TCP_CLOSE_WAIT: |
| 3851 | sk->sk_err = EPIPE; |
| 3852 | break; |
| 3853 | case TCP_CLOSE: |
| 3854 | return; |
| 3855 | default: |
| 3856 | sk->sk_err = ECONNRESET; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3857 | } |
Tom Marshall | a4d2580 | 2010-09-20 15:42:05 -0700 | [diff] [blame] | 3858 | /* This barrier is coupled with smp_rmb() in tcp_poll() */ |
| 3859 | smp_wmb(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3860 | |
| 3861 | if (!sock_flag(sk, SOCK_DEAD)) |
| 3862 | sk->sk_error_report(sk); |
| 3863 | |
| 3864 | tcp_done(sk); |
| 3865 | } |
| 3866 | |
| 3867 | /* |
| 3868 | * Process the FIN bit. This now behaves as it is supposed to work |
| 3869 | * and the FIN takes effect when it is validly part of sequence |
| 3870 | * space. Not before when we get holes. |
| 3871 | * |
| 3872 | * If we are ESTABLISHED, a received fin moves us to CLOSE-WAIT |
| 3873 | * (and thence onto LAST-ACK and finally, CLOSE, we never enter |
| 3874 | * TIME-WAIT) |
| 3875 | * |
| 3876 | * If we are in FINWAIT-1, a received FIN indicates simultaneous |
| 3877 | * close and we go into CLOSING (and later onto TIME-WAIT) |
| 3878 | * |
| 3879 | * If we are in FINWAIT-2, a received FIN moves us to TIME-WAIT. |
| 3880 | */ |
Eric Dumazet | 20c4cb7 | 2011-10-20 17:44:03 -0400 | [diff] [blame] | 3881 | static void tcp_fin(struct sock *sk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3882 | { |
| 3883 | struct tcp_sock *tp = tcp_sk(sk); |
Cong Wang | bcefe17 | 2013-06-15 09:39:18 +0800 | [diff] [blame] | 3884 | const struct dst_entry *dst; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3885 | |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 3886 | inet_csk_schedule_ack(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3887 | |
| 3888 | sk->sk_shutdown |= RCV_SHUTDOWN; |
| 3889 | sock_set_flag(sk, SOCK_DONE); |
| 3890 | |
| 3891 | switch (sk->sk_state) { |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 3892 | case TCP_SYN_RECV: |
| 3893 | case TCP_ESTABLISHED: |
| 3894 | /* Move to CLOSE_WAIT */ |
| 3895 | tcp_set_state(sk, TCP_CLOSE_WAIT); |
Cong Wang | bcefe17 | 2013-06-15 09:39:18 +0800 | [diff] [blame] | 3896 | dst = __sk_dst_get(sk); |
| 3897 | if (!dst || !dst_metric(dst, RTAX_QUICKACK)) |
| 3898 | inet_csk(sk)->icsk_ack.pingpong = 1; |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 3899 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3900 | |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 3901 | case TCP_CLOSE_WAIT: |
| 3902 | case TCP_CLOSING: |
| 3903 | /* Received a retransmission of the FIN, do |
| 3904 | * nothing. |
| 3905 | */ |
| 3906 | break; |
| 3907 | case TCP_LAST_ACK: |
| 3908 | /* RFC793: Remain in the LAST-ACK state. */ |
| 3909 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3910 | |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 3911 | case TCP_FIN_WAIT1: |
| 3912 | /* This case occurs when a simultaneous close |
| 3913 | * happens, we must ack the received FIN and |
| 3914 | * enter the CLOSING state. |
| 3915 | */ |
| 3916 | tcp_send_ack(sk); |
| 3917 | tcp_set_state(sk, TCP_CLOSING); |
| 3918 | break; |
| 3919 | case TCP_FIN_WAIT2: |
| 3920 | /* Received a FIN -- send ACK and enter TIME_WAIT. */ |
| 3921 | tcp_send_ack(sk); |
| 3922 | tcp_time_wait(sk, TCP_TIME_WAIT, 0); |
| 3923 | break; |
| 3924 | default: |
| 3925 | /* Only TCP_LISTEN and TCP_CLOSE are left, in these |
| 3926 | * cases we should never reach this piece of code. |
| 3927 | */ |
Joe Perches | 058bd4d | 2012-03-11 18:36:11 +0000 | [diff] [blame] | 3928 | pr_err("%s: Impossible, sk->sk_state=%d\n", |
Harvey Harrison | 0dc4787 | 2008-03-05 20:47:47 -0800 | [diff] [blame] | 3929 | __func__, sk->sk_state); |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 3930 | break; |
Stephen Hemminger | 3ff50b7 | 2007-04-20 17:09:22 -0700 | [diff] [blame] | 3931 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3932 | |
| 3933 | /* It _is_ possible, that we have something out-of-order _after_ FIN. |
| 3934 | * Probably, we should reset in this case. For now drop them. |
| 3935 | */ |
| 3936 | __skb_queue_purge(&tp->out_of_order_queue); |
Ilpo Järvinen | e60402d | 2007-08-09 15:14:46 +0300 | [diff] [blame] | 3937 | if (tcp_is_sack(tp)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3938 | tcp_sack_reset(&tp->rx_opt); |
Hideo Aoki | 3ab224b | 2007-12-31 00:11:19 -0800 | [diff] [blame] | 3939 | sk_mem_reclaim(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3940 | |
| 3941 | if (!sock_flag(sk, SOCK_DEAD)) { |
| 3942 | sk->sk_state_change(sk); |
| 3943 | |
| 3944 | /* Do not send POLL_HUP for half duplex close. */ |
| 3945 | if (sk->sk_shutdown == SHUTDOWN_MASK || |
| 3946 | sk->sk_state == TCP_CLOSE) |
Pavel Emelyanov | 8d8ad9d | 2007-11-26 20:10:50 +0800 | [diff] [blame] | 3947 | sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_HUP); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3948 | else |
Pavel Emelyanov | 8d8ad9d | 2007-11-26 20:10:50 +0800 | [diff] [blame] | 3949 | sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3950 | } |
| 3951 | } |
| 3952 | |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 3953 | 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] | 3954 | u32 end_seq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3955 | { |
| 3956 | if (!after(seq, sp->end_seq) && !after(sp->start_seq, end_seq)) { |
| 3957 | if (before(seq, sp->start_seq)) |
| 3958 | sp->start_seq = seq; |
| 3959 | if (after(end_seq, sp->end_seq)) |
| 3960 | sp->end_seq = end_seq; |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 3961 | return true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3962 | } |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 3963 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3964 | } |
| 3965 | |
Pavel Emelyanov | 1ed8346 | 2008-07-16 20:29:51 -0700 | [diff] [blame] | 3966 | 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] | 3967 | { |
Pavel Emelyanov | 1ed8346 | 2008-07-16 20:29:51 -0700 | [diff] [blame] | 3968 | struct tcp_sock *tp = tcp_sk(sk); |
| 3969 | |
David S. Miller | bb5b7c1 | 2009-12-15 20:56:42 -0800 | [diff] [blame] | 3970 | if (tcp_is_sack(tp) && sysctl_tcp_dsack) { |
Pavel Emelyanov | 40b215e | 2008-07-03 01:05:41 -0700 | [diff] [blame] | 3971 | int mib_idx; |
| 3972 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3973 | if (before(seq, tp->rcv_nxt)) |
Pavel Emelyanov | 40b215e | 2008-07-03 01:05:41 -0700 | [diff] [blame] | 3974 | mib_idx = LINUX_MIB_TCPDSACKOLDSENT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3975 | else |
Pavel Emelyanov | 40b215e | 2008-07-03 01:05:41 -0700 | [diff] [blame] | 3976 | mib_idx = LINUX_MIB_TCPDSACKOFOSENT; |
| 3977 | |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 3978 | NET_INC_STATS_BH(sock_net(sk), mib_idx); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3979 | |
| 3980 | tp->rx_opt.dsack = 1; |
| 3981 | tp->duplicate_sack[0].start_seq = seq; |
| 3982 | tp->duplicate_sack[0].end_seq = end_seq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3983 | } |
| 3984 | } |
| 3985 | |
Pavel Emelyanov | 1ed8346 | 2008-07-16 20:29:51 -0700 | [diff] [blame] | 3986 | 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] | 3987 | { |
Pavel Emelyanov | 1ed8346 | 2008-07-16 20:29:51 -0700 | [diff] [blame] | 3988 | struct tcp_sock *tp = tcp_sk(sk); |
| 3989 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3990 | if (!tp->rx_opt.dsack) |
Pavel Emelyanov | 1ed8346 | 2008-07-16 20:29:51 -0700 | [diff] [blame] | 3991 | tcp_dsack_set(sk, seq, end_seq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3992 | else |
| 3993 | tcp_sack_extend(tp->duplicate_sack, seq, end_seq); |
| 3994 | } |
| 3995 | |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 3996 | 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] | 3997 | { |
| 3998 | struct tcp_sock *tp = tcp_sk(sk); |
| 3999 | |
| 4000 | if (TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(skb)->seq && |
| 4001 | before(TCP_SKB_CB(skb)->seq, tp->rcv_nxt)) { |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 4002 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_DELAYEDACKLOST); |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 4003 | tcp_enter_quickack_mode(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4004 | |
David S. Miller | bb5b7c1 | 2009-12-15 20:56:42 -0800 | [diff] [blame] | 4005 | if (tcp_is_sack(tp) && sysctl_tcp_dsack) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4006 | u32 end_seq = TCP_SKB_CB(skb)->end_seq; |
| 4007 | |
| 4008 | if (after(TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt)) |
| 4009 | end_seq = tp->rcv_nxt; |
Pavel Emelyanov | 1ed8346 | 2008-07-16 20:29:51 -0700 | [diff] [blame] | 4010 | tcp_dsack_set(sk, TCP_SKB_CB(skb)->seq, end_seq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4011 | } |
| 4012 | } |
| 4013 | |
| 4014 | tcp_send_ack(sk); |
| 4015 | } |
| 4016 | |
| 4017 | /* These routines update the SACK block as out-of-order packets arrive or |
| 4018 | * in-order packets close up the sequence space. |
| 4019 | */ |
| 4020 | static void tcp_sack_maybe_coalesce(struct tcp_sock *tp) |
| 4021 | { |
| 4022 | int this_sack; |
| 4023 | struct tcp_sack_block *sp = &tp->selective_acks[0]; |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 4024 | struct tcp_sack_block *swalk = sp + 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4025 | |
| 4026 | /* See if the recent change to the first SACK eats into |
| 4027 | * or hits the sequence space of other SACK blocks, if so coalesce. |
| 4028 | */ |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 4029 | for (this_sack = 1; this_sack < tp->rx_opt.num_sacks;) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4030 | if (tcp_sack_extend(sp, swalk->start_seq, swalk->end_seq)) { |
| 4031 | int i; |
| 4032 | |
| 4033 | /* Zap SWALK, by moving every further SACK up by one slot. |
| 4034 | * Decrease num_sacks. |
| 4035 | */ |
| 4036 | tp->rx_opt.num_sacks--; |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 4037 | for (i = this_sack; i < tp->rx_opt.num_sacks; i++) |
| 4038 | sp[i] = sp[i + 1]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4039 | continue; |
| 4040 | } |
| 4041 | this_sack++, swalk++; |
| 4042 | } |
| 4043 | } |
| 4044 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4045 | static void tcp_sack_new_ofo_skb(struct sock *sk, u32 seq, u32 end_seq) |
| 4046 | { |
| 4047 | struct tcp_sock *tp = tcp_sk(sk); |
| 4048 | struct tcp_sack_block *sp = &tp->selective_acks[0]; |
| 4049 | int cur_sacks = tp->rx_opt.num_sacks; |
| 4050 | int this_sack; |
| 4051 | |
| 4052 | if (!cur_sacks) |
| 4053 | goto new_sack; |
| 4054 | |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 4055 | for (this_sack = 0; this_sack < cur_sacks; this_sack++, sp++) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4056 | if (tcp_sack_extend(sp, seq, end_seq)) { |
| 4057 | /* Rotate this_sack to the first one. */ |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 4058 | for (; this_sack > 0; this_sack--, sp--) |
Ilpo Järvinen | a0bffff | 2009-03-21 13:36:17 -0700 | [diff] [blame] | 4059 | swap(*sp, *(sp - 1)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4060 | if (cur_sacks > 1) |
| 4061 | tcp_sack_maybe_coalesce(tp); |
| 4062 | return; |
| 4063 | } |
| 4064 | } |
| 4065 | |
| 4066 | /* Could not find an adjacent existing SACK, build a new one, |
| 4067 | * put it at the front, and shift everyone else down. We |
| 4068 | * always know there is at least one SACK present already here. |
| 4069 | * |
| 4070 | * If the sack array is full, forget about the last one. |
| 4071 | */ |
Adam Langley | 4389dde | 2008-07-19 00:07:02 -0700 | [diff] [blame] | 4072 | if (this_sack >= TCP_NUM_SACKS) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4073 | this_sack--; |
| 4074 | tp->rx_opt.num_sacks--; |
| 4075 | sp--; |
| 4076 | } |
Stephen Hemminger | 2de979b | 2007-03-08 20:45:19 -0800 | [diff] [blame] | 4077 | for (; this_sack > 0; this_sack--, sp--) |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 4078 | *sp = *(sp - 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4079 | |
| 4080 | new_sack: |
| 4081 | /* Build the new head SACK, and we're done. */ |
| 4082 | sp->start_seq = seq; |
| 4083 | sp->end_seq = end_seq; |
| 4084 | tp->rx_opt.num_sacks++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4085 | } |
| 4086 | |
| 4087 | /* RCV.NXT advances, some SACKs should be eaten. */ |
| 4088 | |
| 4089 | static void tcp_sack_remove(struct tcp_sock *tp) |
| 4090 | { |
| 4091 | struct tcp_sack_block *sp = &tp->selective_acks[0]; |
| 4092 | int num_sacks = tp->rx_opt.num_sacks; |
| 4093 | int this_sack; |
| 4094 | |
| 4095 | /* Empty ofo queue, hence, all the SACKs are eaten. Clear. */ |
David S. Miller | b03efcf | 2005-07-08 14:57:23 -0700 | [diff] [blame] | 4096 | if (skb_queue_empty(&tp->out_of_order_queue)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4097 | tp->rx_opt.num_sacks = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4098 | return; |
| 4099 | } |
| 4100 | |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 4101 | for (this_sack = 0; this_sack < num_sacks;) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4102 | /* Check if the start of the sack is covered by RCV.NXT. */ |
| 4103 | if (!before(tp->rcv_nxt, sp->start_seq)) { |
| 4104 | int i; |
| 4105 | |
| 4106 | /* RCV.NXT must cover all the block! */ |
Ilpo Järvinen | 547b792 | 2008-07-25 21:43:18 -0700 | [diff] [blame] | 4107 | WARN_ON(before(tp->rcv_nxt, sp->end_seq)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4108 | |
| 4109 | /* Zap this SACK, by moving forward any other SACKS. */ |
Weilong Chen | a22318e | 2013-12-23 14:37:26 +0800 | [diff] [blame] | 4110 | for (i = this_sack+1; i < num_sacks; i++) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4111 | tp->selective_acks[i-1] = tp->selective_acks[i]; |
| 4112 | num_sacks--; |
| 4113 | continue; |
| 4114 | } |
| 4115 | this_sack++; |
| 4116 | sp++; |
| 4117 | } |
Ilpo Järvinen | 5861f8e | 2009-03-14 14:23:01 +0000 | [diff] [blame] | 4118 | tp->rx_opt.num_sacks = num_sacks; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4119 | } |
| 4120 | |
Eric Dumazet | 1402d36 | 2012-04-23 07:11:42 +0000 | [diff] [blame] | 4121 | /** |
| 4122 | * tcp_try_coalesce - try to merge skb to prior one |
| 4123 | * @sk: socket |
| 4124 | * @to: prior buffer |
| 4125 | * @from: buffer to add in queue |
Eric Dumazet | 923dd34 | 2012-05-02 07:55:58 +0000 | [diff] [blame] | 4126 | * @fragstolen: pointer to boolean |
Eric Dumazet | 1402d36 | 2012-04-23 07:11:42 +0000 | [diff] [blame] | 4127 | * |
| 4128 | * Before queueing skb @from after @to, try to merge them |
| 4129 | * to reduce overall memory use and queue lengths, if cost is small. |
| 4130 | * Packets in ofo or receive queues can stay a long time. |
| 4131 | * Better try to coalesce them right now to avoid future collapses. |
Eric Dumazet | 783c175 | 2012-04-23 17:34:36 +0000 | [diff] [blame] | 4132 | * Returns true if caller should free @from instead of queueing it |
Eric Dumazet | 1402d36 | 2012-04-23 07:11:42 +0000 | [diff] [blame] | 4133 | */ |
Eric Dumazet | 783c175 | 2012-04-23 17:34:36 +0000 | [diff] [blame] | 4134 | static bool tcp_try_coalesce(struct sock *sk, |
| 4135 | struct sk_buff *to, |
Eric Dumazet | 329033f | 2012-04-27 00:38:33 +0000 | [diff] [blame] | 4136 | struct sk_buff *from, |
| 4137 | bool *fragstolen) |
Eric Dumazet | 1402d36 | 2012-04-23 07:11:42 +0000 | [diff] [blame] | 4138 | { |
Eric Dumazet | bad43ca | 2012-05-19 03:02:02 +0000 | [diff] [blame] | 4139 | int delta; |
Eric Dumazet | 1402d36 | 2012-04-23 07:11:42 +0000 | [diff] [blame] | 4140 | |
Eric Dumazet | 329033f | 2012-04-27 00:38:33 +0000 | [diff] [blame] | 4141 | *fragstolen = false; |
Alexander Duyck | 34a802a | 2012-05-02 21:19:09 +0000 | [diff] [blame] | 4142 | |
Eric Dumazet | 1ca7ee3 | 2012-05-23 17:51:37 +0000 | [diff] [blame] | 4143 | /* Its possible this segment overlaps with prior segment in queue */ |
| 4144 | if (TCP_SKB_CB(from)->seq != TCP_SKB_CB(to)->end_seq) |
| 4145 | return false; |
| 4146 | |
Eric Dumazet | bad43ca | 2012-05-19 03:02:02 +0000 | [diff] [blame] | 4147 | if (!skb_try_coalesce(to, from, fragstolen, &delta)) |
Eric Dumazet | 783c175 | 2012-04-23 17:34:36 +0000 | [diff] [blame] | 4148 | return false; |
Alexander Duyck | 34a802a | 2012-05-02 21:19:09 +0000 | [diff] [blame] | 4149 | |
Alexander Duyck | 34a802a | 2012-05-02 21:19:09 +0000 | [diff] [blame] | 4150 | atomic_add(delta, &sk->sk_rmem_alloc); |
| 4151 | sk_mem_charge(sk, delta); |
Alexander Duyck | 34a802a | 2012-05-02 21:19:09 +0000 | [diff] [blame] | 4152 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPRCVCOALESCE); |
| 4153 | TCP_SKB_CB(to)->end_seq = TCP_SKB_CB(from)->end_seq; |
| 4154 | TCP_SKB_CB(to)->ack_seq = TCP_SKB_CB(from)->ack_seq; |
Eric Dumazet | e93a043 | 2014-09-15 04:19:52 -0700 | [diff] [blame] | 4155 | TCP_SKB_CB(to)->tcp_flags |= TCP_SKB_CB(from)->tcp_flags; |
Alexander Duyck | 34a802a | 2012-05-02 21:19:09 +0000 | [diff] [blame] | 4156 | return true; |
Eric Dumazet | 1402d36 | 2012-04-23 07:11:42 +0000 | [diff] [blame] | 4157 | } |
| 4158 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4159 | /* This one checks to see if we can put data from the |
| 4160 | * out_of_order queue into the receive_queue. |
| 4161 | */ |
| 4162 | static void tcp_ofo_queue(struct sock *sk) |
| 4163 | { |
| 4164 | struct tcp_sock *tp = tcp_sk(sk); |
| 4165 | __u32 dsack_high = tp->rcv_nxt; |
Eric Dumazet | bd1e75a | 2014-09-19 08:26:20 -0700 | [diff] [blame] | 4166 | struct sk_buff *skb, *tail; |
| 4167 | bool fragstolen, eaten; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4168 | |
| 4169 | while ((skb = skb_peek(&tp->out_of_order_queue)) != NULL) { |
| 4170 | if (after(TCP_SKB_CB(skb)->seq, tp->rcv_nxt)) |
| 4171 | break; |
| 4172 | |
| 4173 | if (before(TCP_SKB_CB(skb)->seq, dsack_high)) { |
| 4174 | __u32 dsack = dsack_high; |
| 4175 | if (before(TCP_SKB_CB(skb)->end_seq, dsack_high)) |
| 4176 | dsack_high = TCP_SKB_CB(skb)->end_seq; |
| 4177 | tcp_dsack_extend(sk, TCP_SKB_CB(skb)->seq, dsack); |
| 4178 | } |
| 4179 | |
Eric Dumazet | bd1e75a | 2014-09-19 08:26:20 -0700 | [diff] [blame] | 4180 | __skb_unlink(skb, &tp->out_of_order_queue); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4181 | if (!after(TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt)) { |
| 4182 | SOCK_DEBUG(sk, "ofo packet was already received\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4183 | __kfree_skb(skb); |
| 4184 | continue; |
| 4185 | } |
| 4186 | SOCK_DEBUG(sk, "ofo requeuing : rcv_next %X seq %X - %X\n", |
| 4187 | tp->rcv_nxt, TCP_SKB_CB(skb)->seq, |
| 4188 | TCP_SKB_CB(skb)->end_seq); |
| 4189 | |
Eric Dumazet | bd1e75a | 2014-09-19 08:26:20 -0700 | [diff] [blame] | 4190 | tail = skb_peek_tail(&sk->sk_receive_queue); |
| 4191 | eaten = tail && tcp_try_coalesce(sk, tail, skb, &fragstolen); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4192 | tp->rcv_nxt = TCP_SKB_CB(skb)->end_seq; |
Eric Dumazet | bd1e75a | 2014-09-19 08:26:20 -0700 | [diff] [blame] | 4193 | if (!eaten) |
| 4194 | __skb_queue_tail(&sk->sk_receive_queue, skb); |
| 4195 | if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4196 | tcp_fin(sk); |
Eric Dumazet | bd1e75a | 2014-09-19 08:26:20 -0700 | [diff] [blame] | 4197 | if (eaten) |
| 4198 | kfree_skb_partial(skb, fragstolen); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4199 | } |
| 4200 | } |
| 4201 | |
| 4202 | static bool tcp_prune_ofo_queue(struct sock *sk); |
| 4203 | static int tcp_prune_queue(struct sock *sk); |
| 4204 | |
| 4205 | static int tcp_try_rmem_schedule(struct sock *sk, struct sk_buff *skb, |
| 4206 | unsigned int size) |
| 4207 | { |
| 4208 | if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf || |
| 4209 | !sk_rmem_schedule(sk, skb, size)) { |
| 4210 | |
| 4211 | if (tcp_prune_queue(sk) < 0) |
| 4212 | return -1; |
Pavel Emelyanov | 1ed8346 | 2008-07-16 20:29:51 -0700 | [diff] [blame] | 4213 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4214 | if (!sk_rmem_schedule(sk, skb, size)) { |
| 4215 | if (!tcp_prune_ofo_queue(sk)) |
| 4216 | return -1; |
Frans Pop | b138338 | 2010-03-24 07:57:28 +0000 | [diff] [blame] | 4217 | |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 4218 | if (!sk_rmem_schedule(sk, skb, size)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4219 | return -1; |
| 4220 | } |
| 4221 | } |
| 4222 | return 0; |
| 4223 | } |
| 4224 | |
Eric Dumazet | e86b2919 | 2012-03-18 11:06:44 +0000 | [diff] [blame] | 4225 | static void tcp_data_queue_ofo(struct sock *sk, struct sk_buff *skb) |
| 4226 | { |
| 4227 | struct tcp_sock *tp = tcp_sk(sk); |
| 4228 | struct sk_buff *skb1; |
| 4229 | u32 seq, end_seq; |
| 4230 | |
Florian Westphal | 735d383 | 2014-09-29 13:08:30 +0200 | [diff] [blame] | 4231 | tcp_ecn_check_ce(tp, skb); |
Eric Dumazet | e86b2919 | 2012-03-18 11:06:44 +0000 | [diff] [blame] | 4232 | |
Mel Gorman | c76562b | 2012-07-31 16:44:41 -0700 | [diff] [blame] | 4233 | if (unlikely(tcp_try_rmem_schedule(sk, skb, skb->truesize))) { |
Eric Dumazet | a6df1ae | 2012-07-16 01:41:36 +0000 | [diff] [blame] | 4234 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPOFODROP); |
Eric Dumazet | e86b2919 | 2012-03-18 11:06:44 +0000 | [diff] [blame] | 4235 | __kfree_skb(skb); |
| 4236 | return; |
| 4237 | } |
| 4238 | |
| 4239 | /* Disable header prediction. */ |
| 4240 | tp->pred_flags = 0; |
| 4241 | inet_csk_schedule_ack(sk); |
| 4242 | |
Eric Dumazet | a6df1ae | 2012-07-16 01:41:36 +0000 | [diff] [blame] | 4243 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPOFOQUEUE); |
Eric Dumazet | e86b2919 | 2012-03-18 11:06:44 +0000 | [diff] [blame] | 4244 | SOCK_DEBUG(sk, "out of order segment: rcv_next %X seq %X - %X\n", |
| 4245 | tp->rcv_nxt, TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq); |
| 4246 | |
| 4247 | skb1 = skb_peek_tail(&tp->out_of_order_queue); |
| 4248 | if (!skb1) { |
| 4249 | /* Initial out of order segment, build 1 SACK. */ |
| 4250 | if (tcp_is_sack(tp)) { |
| 4251 | tp->rx_opt.num_sacks = 1; |
| 4252 | tp->selective_acks[0].start_seq = TCP_SKB_CB(skb)->seq; |
| 4253 | tp->selective_acks[0].end_seq = |
| 4254 | TCP_SKB_CB(skb)->end_seq; |
| 4255 | } |
| 4256 | __skb_queue_head(&tp->out_of_order_queue, skb); |
| 4257 | goto end; |
| 4258 | } |
| 4259 | |
| 4260 | seq = TCP_SKB_CB(skb)->seq; |
| 4261 | end_seq = TCP_SKB_CB(skb)->end_seq; |
| 4262 | |
| 4263 | if (seq == TCP_SKB_CB(skb1)->end_seq) { |
Eric Dumazet | 329033f | 2012-04-27 00:38:33 +0000 | [diff] [blame] | 4264 | bool fragstolen; |
| 4265 | |
| 4266 | if (!tcp_try_coalesce(sk, skb1, skb, &fragstolen)) { |
Eric Dumazet | 1402d36 | 2012-04-23 07:11:42 +0000 | [diff] [blame] | 4267 | __skb_queue_after(&tp->out_of_order_queue, skb1, skb); |
| 4268 | } else { |
Eric Dumazet | 4e4f1fc | 2013-09-06 10:35:58 -0700 | [diff] [blame] | 4269 | tcp_grow_window(sk, skb); |
Eric Dumazet | 923dd34 | 2012-05-02 07:55:58 +0000 | [diff] [blame] | 4270 | kfree_skb_partial(skb, fragstolen); |
Eric Dumazet | c862815 | 2012-03-18 11:07:47 +0000 | [diff] [blame] | 4271 | skb = NULL; |
Eric Dumazet | c862815 | 2012-03-18 11:07:47 +0000 | [diff] [blame] | 4272 | } |
Eric Dumazet | e86b2919 | 2012-03-18 11:06:44 +0000 | [diff] [blame] | 4273 | |
| 4274 | if (!tp->rx_opt.num_sacks || |
| 4275 | tp->selective_acks[0].end_seq != seq) |
| 4276 | goto add_sack; |
| 4277 | |
| 4278 | /* Common case: data arrive in order after hole. */ |
| 4279 | tp->selective_acks[0].end_seq = end_seq; |
| 4280 | goto end; |
| 4281 | } |
| 4282 | |
| 4283 | /* Find place to insert this segment. */ |
| 4284 | while (1) { |
| 4285 | if (!after(TCP_SKB_CB(skb1)->seq, seq)) |
| 4286 | break; |
| 4287 | if (skb_queue_is_first(&tp->out_of_order_queue, skb1)) { |
| 4288 | skb1 = NULL; |
| 4289 | break; |
| 4290 | } |
| 4291 | skb1 = skb_queue_prev(&tp->out_of_order_queue, skb1); |
| 4292 | } |
| 4293 | |
| 4294 | /* Do skb overlap to previous one? */ |
| 4295 | if (skb1 && before(seq, TCP_SKB_CB(skb1)->end_seq)) { |
| 4296 | if (!after(end_seq, TCP_SKB_CB(skb1)->end_seq)) { |
| 4297 | /* All the bits are present. Drop. */ |
Eric Dumazet | a6df1ae | 2012-07-16 01:41:36 +0000 | [diff] [blame] | 4298 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPOFOMERGE); |
Eric Dumazet | e86b2919 | 2012-03-18 11:06:44 +0000 | [diff] [blame] | 4299 | __kfree_skb(skb); |
| 4300 | skb = NULL; |
| 4301 | tcp_dsack_set(sk, seq, end_seq); |
| 4302 | goto add_sack; |
| 4303 | } |
| 4304 | if (after(seq, TCP_SKB_CB(skb1)->seq)) { |
| 4305 | /* Partial overlap. */ |
| 4306 | tcp_dsack_set(sk, seq, |
| 4307 | TCP_SKB_CB(skb1)->end_seq); |
| 4308 | } else { |
| 4309 | if (skb_queue_is_first(&tp->out_of_order_queue, |
| 4310 | skb1)) |
| 4311 | skb1 = NULL; |
| 4312 | else |
| 4313 | skb1 = skb_queue_prev( |
| 4314 | &tp->out_of_order_queue, |
| 4315 | skb1); |
| 4316 | } |
| 4317 | } |
| 4318 | if (!skb1) |
| 4319 | __skb_queue_head(&tp->out_of_order_queue, skb); |
| 4320 | else |
| 4321 | __skb_queue_after(&tp->out_of_order_queue, skb1, skb); |
| 4322 | |
| 4323 | /* And clean segments covered by new one as whole. */ |
| 4324 | while (!skb_queue_is_last(&tp->out_of_order_queue, skb)) { |
| 4325 | skb1 = skb_queue_next(&tp->out_of_order_queue, skb); |
| 4326 | |
| 4327 | if (!after(end_seq, TCP_SKB_CB(skb1)->seq)) |
| 4328 | break; |
| 4329 | if (before(end_seq, TCP_SKB_CB(skb1)->end_seq)) { |
| 4330 | tcp_dsack_extend(sk, TCP_SKB_CB(skb1)->seq, |
| 4331 | end_seq); |
| 4332 | break; |
| 4333 | } |
| 4334 | __skb_unlink(skb1, &tp->out_of_order_queue); |
| 4335 | tcp_dsack_extend(sk, TCP_SKB_CB(skb1)->seq, |
| 4336 | TCP_SKB_CB(skb1)->end_seq); |
Eric Dumazet | a6df1ae | 2012-07-16 01:41:36 +0000 | [diff] [blame] | 4337 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPOFOMERGE); |
Eric Dumazet | e86b2919 | 2012-03-18 11:06:44 +0000 | [diff] [blame] | 4338 | __kfree_skb(skb1); |
| 4339 | } |
| 4340 | |
| 4341 | add_sack: |
| 4342 | if (tcp_is_sack(tp)) |
| 4343 | tcp_sack_new_ofo_skb(sk, seq, end_seq); |
| 4344 | end: |
Eric Dumazet | 4e4f1fc | 2013-09-06 10:35:58 -0700 | [diff] [blame] | 4345 | if (skb) { |
| 4346 | tcp_grow_window(sk, skb); |
Eric Dumazet | e86b2919 | 2012-03-18 11:06:44 +0000 | [diff] [blame] | 4347 | skb_set_owner_r(skb, sk); |
Eric Dumazet | 4e4f1fc | 2013-09-06 10:35:58 -0700 | [diff] [blame] | 4348 | } |
Eric Dumazet | e86b2919 | 2012-03-18 11:06:44 +0000 | [diff] [blame] | 4349 | } |
| 4350 | |
Pavel Emelyanov | 292e8d8 | 2012-05-10 01:49:41 +0000 | [diff] [blame] | 4351 | 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] | 4352 | bool *fragstolen) |
| 4353 | { |
| 4354 | int eaten; |
| 4355 | struct sk_buff *tail = skb_peek_tail(&sk->sk_receive_queue); |
| 4356 | |
| 4357 | __skb_pull(skb, hdrlen); |
| 4358 | eaten = (tail && |
| 4359 | tcp_try_coalesce(sk, tail, skb, fragstolen)) ? 1 : 0; |
| 4360 | tcp_sk(sk)->rcv_nxt = TCP_SKB_CB(skb)->end_seq; |
| 4361 | if (!eaten) { |
| 4362 | __skb_queue_tail(&sk->sk_receive_queue, skb); |
| 4363 | skb_set_owner_r(skb, sk); |
| 4364 | } |
| 4365 | return eaten; |
| 4366 | } |
Eric Dumazet | e86b2919 | 2012-03-18 11:06:44 +0000 | [diff] [blame] | 4367 | |
Pavel Emelyanov | 292e8d8 | 2012-05-10 01:49:41 +0000 | [diff] [blame] | 4368 | int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size) |
| 4369 | { |
Eric Dumazet | cb93471 | 2014-09-17 03:14:42 -0700 | [diff] [blame] | 4370 | struct sk_buff *skb; |
Pavel Emelyanov | 292e8d8 | 2012-05-10 01:49:41 +0000 | [diff] [blame] | 4371 | bool fragstolen; |
| 4372 | |
Pavel Emelyanov | c454e61 | 2012-10-29 05:05:33 +0000 | [diff] [blame] | 4373 | if (size == 0) |
| 4374 | return 0; |
| 4375 | |
Eric Dumazet | cb93471 | 2014-09-17 03:14:42 -0700 | [diff] [blame] | 4376 | skb = alloc_skb(size, sk->sk_allocation); |
Pavel Emelyanov | 292e8d8 | 2012-05-10 01:49:41 +0000 | [diff] [blame] | 4377 | if (!skb) |
| 4378 | goto err; |
| 4379 | |
Eric Dumazet | cb93471 | 2014-09-17 03:14:42 -0700 | [diff] [blame] | 4380 | if (tcp_try_rmem_schedule(sk, skb, skb->truesize)) |
Mel Gorman | c76562b | 2012-07-31 16:44:41 -0700 | [diff] [blame] | 4381 | goto err_free; |
| 4382 | |
Al Viro | 6ce8e9c | 2014-04-06 21:25:44 -0400 | [diff] [blame] | 4383 | if (memcpy_from_msg(skb_put(skb, size), msg, size)) |
Pavel Emelyanov | 292e8d8 | 2012-05-10 01:49:41 +0000 | [diff] [blame] | 4384 | goto err_free; |
| 4385 | |
| 4386 | TCP_SKB_CB(skb)->seq = tcp_sk(sk)->rcv_nxt; |
| 4387 | TCP_SKB_CB(skb)->end_seq = TCP_SKB_CB(skb)->seq + size; |
| 4388 | TCP_SKB_CB(skb)->ack_seq = tcp_sk(sk)->snd_una - 1; |
| 4389 | |
Eric Dumazet | cb93471 | 2014-09-17 03:14:42 -0700 | [diff] [blame] | 4390 | if (tcp_queue_rcv(sk, skb, 0, &fragstolen)) { |
Pavel Emelyanov | 292e8d8 | 2012-05-10 01:49:41 +0000 | [diff] [blame] | 4391 | WARN_ON_ONCE(fragstolen); /* should not happen */ |
| 4392 | __kfree_skb(skb); |
| 4393 | } |
| 4394 | return size; |
| 4395 | |
| 4396 | err_free: |
| 4397 | kfree_skb(skb); |
| 4398 | err: |
| 4399 | return -ENOMEM; |
| 4400 | } |
| 4401 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4402 | static void tcp_data_queue(struct sock *sk, struct sk_buff *skb) |
| 4403 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4404 | struct tcp_sock *tp = tcp_sk(sk); |
| 4405 | int eaten = -1; |
Eric Dumazet | 329033f | 2012-04-27 00:38:33 +0000 | [diff] [blame] | 4406 | bool fragstolen = false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4407 | |
| 4408 | if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) |
| 4409 | goto drop; |
| 4410 | |
Eric Dumazet | f84af32 | 2010-04-28 15:31:51 -0700 | [diff] [blame] | 4411 | skb_dst_drop(skb); |
Peter Pan(潘卫平) | 155c6e1 | 2014-09-24 22:17:02 +0800 | [diff] [blame] | 4412 | __skb_pull(skb, tcp_hdr(skb)->doff * 4); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4413 | |
Florian Westphal | 735d383 | 2014-09-29 13:08:30 +0200 | [diff] [blame] | 4414 | tcp_ecn_accept_cwr(tp, skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4415 | |
Ilpo Järvinen | 5861f8e | 2009-03-14 14:23:01 +0000 | [diff] [blame] | 4416 | tp->rx_opt.dsack = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4417 | |
| 4418 | /* Queue data for delivery to the user. |
| 4419 | * Packets in sequence go to the receive queue. |
| 4420 | * Out of sequence packets to the out_of_order_queue. |
| 4421 | */ |
| 4422 | if (TCP_SKB_CB(skb)->seq == tp->rcv_nxt) { |
| 4423 | if (tcp_receive_window(tp) == 0) |
| 4424 | goto out_of_window; |
| 4425 | |
| 4426 | /* Ok. In sequence. In window. */ |
| 4427 | if (tp->ucopy.task == current && |
| 4428 | tp->copied_seq == tp->rcv_nxt && tp->ucopy.len && |
| 4429 | sock_owned_by_user(sk) && !tp->urg_data) { |
| 4430 | int chunk = min_t(unsigned int, skb->len, |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 4431 | tp->ucopy.len); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4432 | |
| 4433 | __set_current_state(TASK_RUNNING); |
| 4434 | |
| 4435 | local_bh_enable(); |
Al Viro | f4362a2 | 2014-11-24 13:26:06 -0500 | [diff] [blame] | 4436 | if (!skb_copy_datagram_msg(skb, 0, tp->ucopy.msg, chunk)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4437 | tp->ucopy.len -= chunk; |
| 4438 | tp->copied_seq += chunk; |
Jerry Chu | 44f5324 | 2011-01-25 13:46:30 -0800 | [diff] [blame] | 4439 | eaten = (chunk == skb->len); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4440 | tcp_rcv_space_adjust(sk); |
| 4441 | } |
| 4442 | local_bh_disable(); |
| 4443 | } |
| 4444 | |
| 4445 | if (eaten <= 0) { |
| 4446 | queue_and_out: |
| 4447 | if (eaten < 0 && |
Mel Gorman | c76562b | 2012-07-31 16:44:41 -0700 | [diff] [blame] | 4448 | tcp_try_rmem_schedule(sk, skb, skb->truesize)) |
Vitaliy Gusev | b000cd3 | 2008-04-15 00:33:38 -0700 | [diff] [blame] | 4449 | goto drop; |
| 4450 | |
Eric Dumazet | b081f85 | 2012-05-02 09:58:29 +0000 | [diff] [blame] | 4451 | eaten = tcp_queue_rcv(sk, skb, 0, &fragstolen); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4452 | } |
| 4453 | tp->rcv_nxt = TCP_SKB_CB(skb)->end_seq; |
Stephen Hemminger | 2de979b | 2007-03-08 20:45:19 -0800 | [diff] [blame] | 4454 | if (skb->len) |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 4455 | tcp_event_data_recv(sk, skb); |
Peter Pan(潘卫平) | 155c6e1 | 2014-09-24 22:17:02 +0800 | [diff] [blame] | 4456 | if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) |
Eric Dumazet | 20c4cb7 | 2011-10-20 17:44:03 -0400 | [diff] [blame] | 4457 | tcp_fin(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4458 | |
David S. Miller | b03efcf | 2005-07-08 14:57:23 -0700 | [diff] [blame] | 4459 | if (!skb_queue_empty(&tp->out_of_order_queue)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4460 | tcp_ofo_queue(sk); |
| 4461 | |
| 4462 | /* RFC2581. 4.2. SHOULD send immediate ACK, when |
| 4463 | * gap in queue is filled. |
| 4464 | */ |
David S. Miller | b03efcf | 2005-07-08 14:57:23 -0700 | [diff] [blame] | 4465 | if (skb_queue_empty(&tp->out_of_order_queue)) |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 4466 | inet_csk(sk)->icsk_ack.pingpong = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4467 | } |
| 4468 | |
| 4469 | if (tp->rx_opt.num_sacks) |
| 4470 | tcp_sack_remove(tp); |
| 4471 | |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 4472 | tcp_fast_path_check(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4473 | |
Eric Dumazet | 923dd34 | 2012-05-02 07:55:58 +0000 | [diff] [blame] | 4474 | if (eaten > 0) |
| 4475 | kfree_skb_partial(skb, fragstolen); |
Eric Dumazet | 1d57f19 | 2012-09-17 12:51:39 +0000 | [diff] [blame] | 4476 | if (!sock_flag(sk, SOCK_DEAD)) |
David S. Miller | 676d236 | 2014-04-11 16:15:36 -0400 | [diff] [blame] | 4477 | sk->sk_data_ready(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4478 | return; |
| 4479 | } |
| 4480 | |
| 4481 | if (!after(TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt)) { |
| 4482 | /* A retransmit, 2nd most common case. Force an immediate ack. */ |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 4483 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_DELAYEDACKLOST); |
Pavel Emelyanov | 1ed8346 | 2008-07-16 20:29:51 -0700 | [diff] [blame] | 4484 | 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] | 4485 | |
| 4486 | out_of_window: |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 4487 | tcp_enter_quickack_mode(sk); |
| 4488 | inet_csk_schedule_ack(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4489 | drop: |
| 4490 | __kfree_skb(skb); |
| 4491 | return; |
| 4492 | } |
| 4493 | |
| 4494 | /* Out of window. F.e. zero window probe. */ |
| 4495 | if (!before(TCP_SKB_CB(skb)->seq, tp->rcv_nxt + tcp_receive_window(tp))) |
| 4496 | goto out_of_window; |
| 4497 | |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 4498 | tcp_enter_quickack_mode(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4499 | |
| 4500 | if (before(TCP_SKB_CB(skb)->seq, tp->rcv_nxt)) { |
| 4501 | /* Partial packet, seq < rcv_next < end_seq */ |
| 4502 | SOCK_DEBUG(sk, "partial packet: rcv_next %X seq %X - %X\n", |
| 4503 | tp->rcv_nxt, TCP_SKB_CB(skb)->seq, |
| 4504 | TCP_SKB_CB(skb)->end_seq); |
| 4505 | |
Pavel Emelyanov | 1ed8346 | 2008-07-16 20:29:51 -0700 | [diff] [blame] | 4506 | tcp_dsack_set(sk, TCP_SKB_CB(skb)->seq, tp->rcv_nxt); |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 4507 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4508 | /* If window is closed, drop tail of packet. But after |
| 4509 | * remembering D-SACK for its head made in previous line. |
| 4510 | */ |
| 4511 | if (!tcp_receive_window(tp)) |
| 4512 | goto out_of_window; |
| 4513 | goto queue_and_out; |
| 4514 | } |
| 4515 | |
Eric Dumazet | e86b2919 | 2012-03-18 11:06:44 +0000 | [diff] [blame] | 4516 | tcp_data_queue_ofo(sk, skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4517 | } |
| 4518 | |
Ilpo Järvinen | 2cf4663 | 2008-08-23 05:11:41 -0700 | [diff] [blame] | 4519 | static struct sk_buff *tcp_collapse_one(struct sock *sk, struct sk_buff *skb, |
| 4520 | struct sk_buff_head *list) |
| 4521 | { |
David S. Miller | 9152194 | 2009-05-28 21:35:47 -0700 | [diff] [blame] | 4522 | struct sk_buff *next = NULL; |
| 4523 | |
| 4524 | if (!skb_queue_is_last(list, skb)) |
| 4525 | next = skb_queue_next(list, skb); |
Ilpo Järvinen | 2cf4663 | 2008-08-23 05:11:41 -0700 | [diff] [blame] | 4526 | |
| 4527 | __skb_unlink(skb, list); |
| 4528 | __kfree_skb(skb); |
| 4529 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPRCVCOLLAPSED); |
| 4530 | |
| 4531 | return next; |
| 4532 | } |
| 4533 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4534 | /* Collapse contiguous sequence of skbs head..tail with |
| 4535 | * sequence numbers start..end. |
David S. Miller | 9152194 | 2009-05-28 21:35:47 -0700 | [diff] [blame] | 4536 | * |
| 4537 | * If tail is NULL, this means until the end of the list. |
| 4538 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4539 | * Segments with FIN/SYN are not collapsed (only because this |
| 4540 | * simplifies code) |
| 4541 | */ |
| 4542 | static void |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 4543 | tcp_collapse(struct sock *sk, struct sk_buff_head *list, |
| 4544 | struct sk_buff *head, struct sk_buff *tail, |
| 4545 | u32 start, u32 end) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4546 | { |
David S. Miller | 9152194 | 2009-05-28 21:35:47 -0700 | [diff] [blame] | 4547 | struct sk_buff *skb, *n; |
| 4548 | bool end_of_skbs; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4549 | |
Stephen Hemminger | caa20d9a | 2005-11-10 17:13:47 -0800 | [diff] [blame] | 4550 | /* First, check that queue is collapsible and find |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4551 | * the point where collapsing can be useful. */ |
David S. Miller | 9152194 | 2009-05-28 21:35:47 -0700 | [diff] [blame] | 4552 | skb = head; |
| 4553 | restart: |
| 4554 | end_of_skbs = true; |
| 4555 | skb_queue_walk_from_safe(list, skb, n) { |
| 4556 | if (skb == tail) |
| 4557 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4558 | /* No new bits? It is possible on ofo queue. */ |
| 4559 | if (!before(start, TCP_SKB_CB(skb)->end_seq)) { |
Ilpo Järvinen | 2cf4663 | 2008-08-23 05:11:41 -0700 | [diff] [blame] | 4560 | skb = tcp_collapse_one(sk, skb, list); |
David S. Miller | 9152194 | 2009-05-28 21:35:47 -0700 | [diff] [blame] | 4561 | if (!skb) |
| 4562 | break; |
| 4563 | goto restart; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4564 | } |
| 4565 | |
| 4566 | /* The first skb to collapse is: |
| 4567 | * - not SYN/FIN and |
| 4568 | * - bloated or contains data before "start" or |
| 4569 | * overlaps to the next one. |
| 4570 | */ |
Eric Dumazet | e11ecdd | 2014-09-15 04:19:51 -0700 | [diff] [blame] | 4571 | if (!(TCP_SKB_CB(skb)->tcp_flags & (TCPHDR_SYN | TCPHDR_FIN)) && |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4572 | (tcp_win_from_space(skb->truesize) > skb->len || |
David S. Miller | 9152194 | 2009-05-28 21:35:47 -0700 | [diff] [blame] | 4573 | before(TCP_SKB_CB(skb)->seq, start))) { |
| 4574 | end_of_skbs = false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4575 | break; |
David S. Miller | 9152194 | 2009-05-28 21:35:47 -0700 | [diff] [blame] | 4576 | } |
| 4577 | |
| 4578 | if (!skb_queue_is_last(list, skb)) { |
| 4579 | struct sk_buff *next = skb_queue_next(list, skb); |
| 4580 | if (next != tail && |
| 4581 | TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(next)->seq) { |
| 4582 | end_of_skbs = false; |
| 4583 | break; |
| 4584 | } |
| 4585 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4586 | |
| 4587 | /* Decided to skip this, advance start seq. */ |
| 4588 | start = TCP_SKB_CB(skb)->end_seq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4589 | } |
Eric Dumazet | e11ecdd | 2014-09-15 04:19:51 -0700 | [diff] [blame] | 4590 | if (end_of_skbs || |
| 4591 | (TCP_SKB_CB(skb)->tcp_flags & (TCPHDR_SYN | TCPHDR_FIN))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4592 | return; |
| 4593 | |
| 4594 | while (before(start, end)) { |
Eric Dumazet | b3d6cb9 | 2014-09-15 04:19:53 -0700 | [diff] [blame] | 4595 | int copy = min_t(int, SKB_MAX_ORDER(0, 0), end - start); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4596 | struct sk_buff *nskb; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4597 | |
Eric Dumazet | b3d6cb9 | 2014-09-15 04:19:53 -0700 | [diff] [blame] | 4598 | nskb = alloc_skb(copy, GFP_ATOMIC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4599 | if (!nskb) |
| 4600 | return; |
Arnaldo Carvalho de Melo | c51957d | 2007-03-10 12:47:22 -0300 | [diff] [blame] | 4601 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4602 | memcpy(nskb->cb, skb->cb, sizeof(skb->cb)); |
| 4603 | 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] | 4604 | __skb_queue_before(list, skb, nskb); |
Hideo Aoki | 3ab224b | 2007-12-31 00:11:19 -0800 | [diff] [blame] | 4605 | skb_set_owner_r(nskb, sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4606 | |
| 4607 | /* Copy data, releasing collapsed skbs. */ |
| 4608 | while (copy > 0) { |
| 4609 | int offset = start - TCP_SKB_CB(skb)->seq; |
| 4610 | int size = TCP_SKB_CB(skb)->end_seq - start; |
| 4611 | |
Kris Katterjohn | 09a6266 | 2006-01-08 22:24:28 -0800 | [diff] [blame] | 4612 | BUG_ON(offset < 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4613 | if (size > 0) { |
| 4614 | size = min(copy, size); |
| 4615 | if (skb_copy_bits(skb, offset, skb_put(nskb, size), size)) |
| 4616 | BUG(); |
| 4617 | TCP_SKB_CB(nskb)->end_seq += size; |
| 4618 | copy -= size; |
| 4619 | start += size; |
| 4620 | } |
| 4621 | if (!before(start, TCP_SKB_CB(skb)->end_seq)) { |
Ilpo Järvinen | 2cf4663 | 2008-08-23 05:11:41 -0700 | [diff] [blame] | 4622 | skb = tcp_collapse_one(sk, skb, list); |
David S. Miller | 9152194 | 2009-05-28 21:35:47 -0700 | [diff] [blame] | 4623 | if (!skb || |
| 4624 | skb == tail || |
Eric Dumazet | e11ecdd | 2014-09-15 04:19:51 -0700 | [diff] [blame] | 4625 | (TCP_SKB_CB(skb)->tcp_flags & (TCPHDR_SYN | TCPHDR_FIN))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4626 | return; |
| 4627 | } |
| 4628 | } |
| 4629 | } |
| 4630 | } |
| 4631 | |
| 4632 | /* Collapse ofo queue. Algorithm: select contiguous sequence of skbs |
| 4633 | * and tcp_collapse() them until all the queue is collapsed. |
| 4634 | */ |
| 4635 | static void tcp_collapse_ofo_queue(struct sock *sk) |
| 4636 | { |
| 4637 | struct tcp_sock *tp = tcp_sk(sk); |
| 4638 | struct sk_buff *skb = skb_peek(&tp->out_of_order_queue); |
| 4639 | struct sk_buff *head; |
| 4640 | u32 start, end; |
| 4641 | |
| 4642 | if (skb == NULL) |
| 4643 | return; |
| 4644 | |
| 4645 | start = TCP_SKB_CB(skb)->seq; |
| 4646 | end = TCP_SKB_CB(skb)->end_seq; |
| 4647 | head = skb; |
| 4648 | |
| 4649 | for (;;) { |
David S. Miller | 9152194 | 2009-05-28 21:35:47 -0700 | [diff] [blame] | 4650 | struct sk_buff *next = NULL; |
| 4651 | |
| 4652 | if (!skb_queue_is_last(&tp->out_of_order_queue, skb)) |
| 4653 | next = skb_queue_next(&tp->out_of_order_queue, skb); |
| 4654 | skb = next; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4655 | |
| 4656 | /* Segment is terminated when we see gap or when |
| 4657 | * we are at the end of all the queue. */ |
David S. Miller | 9152194 | 2009-05-28 21:35:47 -0700 | [diff] [blame] | 4658 | if (!skb || |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4659 | after(TCP_SKB_CB(skb)->seq, end) || |
| 4660 | before(TCP_SKB_CB(skb)->end_seq, start)) { |
David S. Miller | 8728b83 | 2005-08-09 19:25:21 -0700 | [diff] [blame] | 4661 | tcp_collapse(sk, &tp->out_of_order_queue, |
| 4662 | head, skb, start, end); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4663 | head = skb; |
David S. Miller | 9152194 | 2009-05-28 21:35:47 -0700 | [diff] [blame] | 4664 | if (!skb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4665 | break; |
| 4666 | /* Start new segment */ |
| 4667 | start = TCP_SKB_CB(skb)->seq; |
| 4668 | end = TCP_SKB_CB(skb)->end_seq; |
| 4669 | } else { |
| 4670 | if (before(TCP_SKB_CB(skb)->seq, start)) |
| 4671 | start = TCP_SKB_CB(skb)->seq; |
| 4672 | if (after(TCP_SKB_CB(skb)->end_seq, end)) |
| 4673 | end = TCP_SKB_CB(skb)->end_seq; |
| 4674 | } |
| 4675 | } |
| 4676 | } |
| 4677 | |
Vitaliy Gusev | b000cd3 | 2008-04-15 00:33:38 -0700 | [diff] [blame] | 4678 | /* |
| 4679 | * Purge the out-of-order queue. |
Vitaliy Gusev | 56f367b | 2008-04-15 20:26:34 -0700 | [diff] [blame] | 4680 | * Return true if queue was pruned. |
Vitaliy Gusev | b000cd3 | 2008-04-15 00:33:38 -0700 | [diff] [blame] | 4681 | */ |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 4682 | static bool tcp_prune_ofo_queue(struct sock *sk) |
Vitaliy Gusev | b000cd3 | 2008-04-15 00:33:38 -0700 | [diff] [blame] | 4683 | { |
| 4684 | struct tcp_sock *tp = tcp_sk(sk); |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 4685 | bool res = false; |
Vitaliy Gusev | b000cd3 | 2008-04-15 00:33:38 -0700 | [diff] [blame] | 4686 | |
| 4687 | if (!skb_queue_empty(&tp->out_of_order_queue)) { |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 4688 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_OFOPRUNED); |
Vitaliy Gusev | b000cd3 | 2008-04-15 00:33:38 -0700 | [diff] [blame] | 4689 | __skb_queue_purge(&tp->out_of_order_queue); |
| 4690 | |
| 4691 | /* Reset SACK state. A conforming SACK implementation will |
| 4692 | * do the same at a timeout based retransmit. When a connection |
| 4693 | * is in a sad state like this, we care only about integrity |
| 4694 | * of the connection not performance. |
| 4695 | */ |
| 4696 | if (tp->rx_opt.sack_ok) |
| 4697 | tcp_sack_reset(&tp->rx_opt); |
| 4698 | sk_mem_reclaim(sk); |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 4699 | res = true; |
Vitaliy Gusev | b000cd3 | 2008-04-15 00:33:38 -0700 | [diff] [blame] | 4700 | } |
Vitaliy Gusev | 56f367b | 2008-04-15 20:26:34 -0700 | [diff] [blame] | 4701 | return res; |
Vitaliy Gusev | b000cd3 | 2008-04-15 00:33:38 -0700 | [diff] [blame] | 4702 | } |
| 4703 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4704 | /* Reduce allocated memory if we can, trying to get |
| 4705 | * the socket within its memory limits again. |
| 4706 | * |
| 4707 | * Return less than zero if we should start dropping frames |
| 4708 | * until the socket owning process reads some of the data |
| 4709 | * to stabilize the situation. |
| 4710 | */ |
| 4711 | static int tcp_prune_queue(struct sock *sk) |
| 4712 | { |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 4713 | struct tcp_sock *tp = tcp_sk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4714 | |
| 4715 | SOCK_DEBUG(sk, "prune_queue: c=%x\n", tp->copied_seq); |
| 4716 | |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 4717 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PRUNECALLED); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4718 | |
| 4719 | if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 4720 | tcp_clamp_window(sk); |
Glauber Costa | 180d8cd | 2011-12-11 21:47:02 +0000 | [diff] [blame] | 4721 | else if (sk_under_memory_pressure(sk)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4722 | tp->rcv_ssthresh = min(tp->rcv_ssthresh, 4U * tp->advmss); |
| 4723 | |
| 4724 | tcp_collapse_ofo_queue(sk); |
David S. Miller | 9152194 | 2009-05-28 21:35:47 -0700 | [diff] [blame] | 4725 | if (!skb_queue_empty(&sk->sk_receive_queue)) |
| 4726 | tcp_collapse(sk, &sk->sk_receive_queue, |
| 4727 | skb_peek(&sk->sk_receive_queue), |
| 4728 | NULL, |
| 4729 | tp->copied_seq, tp->rcv_nxt); |
Hideo Aoki | 3ab224b | 2007-12-31 00:11:19 -0800 | [diff] [blame] | 4730 | sk_mem_reclaim(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4731 | |
| 4732 | if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf) |
| 4733 | return 0; |
| 4734 | |
| 4735 | /* Collapsing did not help, destructive actions follow. |
| 4736 | * This must not ever occur. */ |
| 4737 | |
Vitaliy Gusev | b000cd3 | 2008-04-15 00:33:38 -0700 | [diff] [blame] | 4738 | tcp_prune_ofo_queue(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4739 | |
| 4740 | if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf) |
| 4741 | return 0; |
| 4742 | |
| 4743 | /* If we are really being abused, tell the caller to silently |
| 4744 | * drop receive data on the floor. It will get retransmitted |
| 4745 | * and hopefully then we'll have sufficient space. |
| 4746 | */ |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 4747 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_RCVPRUNED); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4748 | |
| 4749 | /* Massive buffer overcommit. */ |
| 4750 | tp->pred_flags = 0; |
| 4751 | return -1; |
| 4752 | } |
| 4753 | |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 4754 | static bool tcp_should_expand_sndbuf(const struct sock *sk) |
David S. Miller | 0d9901d | 2005-07-05 15:21:10 -0700 | [diff] [blame] | 4755 | { |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 4756 | const struct tcp_sock *tp = tcp_sk(sk); |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 4757 | |
David S. Miller | 0d9901d | 2005-07-05 15:21:10 -0700 | [diff] [blame] | 4758 | /* If the user specified a specific send buffer setting, do |
| 4759 | * not modify it. |
| 4760 | */ |
| 4761 | if (sk->sk_userlocks & SOCK_SNDBUF_LOCK) |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 4762 | return false; |
David S. Miller | 0d9901d | 2005-07-05 15:21:10 -0700 | [diff] [blame] | 4763 | |
| 4764 | /* If we are under global TCP memory pressure, do not expand. */ |
Glauber Costa | 180d8cd | 2011-12-11 21:47:02 +0000 | [diff] [blame] | 4765 | if (sk_under_memory_pressure(sk)) |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 4766 | return false; |
David S. Miller | 0d9901d | 2005-07-05 15:21:10 -0700 | [diff] [blame] | 4767 | |
| 4768 | /* If we are under soft global TCP memory pressure, do not expand. */ |
Glauber Costa | 180d8cd | 2011-12-11 21:47:02 +0000 | [diff] [blame] | 4769 | if (sk_memory_allocated(sk) >= sk_prot_mem_limits(sk, 0)) |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 4770 | return false; |
David S. Miller | 0d9901d | 2005-07-05 15:21:10 -0700 | [diff] [blame] | 4771 | |
| 4772 | /* If we filled the congestion window, do not expand. */ |
Neal Cardwell | 6514890 | 2015-02-20 13:33:16 -0500 | [diff] [blame] | 4773 | if (tcp_packets_in_flight(tp) >= tp->snd_cwnd) |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 4774 | return false; |
David S. Miller | 0d9901d | 2005-07-05 15:21:10 -0700 | [diff] [blame] | 4775 | |
Eric Dumazet | a2a385d | 2012-05-16 23:15:34 +0000 | [diff] [blame] | 4776 | return true; |
David S. Miller | 0d9901d | 2005-07-05 15:21:10 -0700 | [diff] [blame] | 4777 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4778 | |
| 4779 | /* When incoming ACK allowed to free some skb from write_queue, |
| 4780 | * we remember this event in flag SOCK_QUEUE_SHRUNK and wake up socket |
| 4781 | * on the exit from tcp input handler. |
| 4782 | * |
| 4783 | * PROBLEM: sndbuf expansion does not work well with largesend. |
| 4784 | */ |
| 4785 | static void tcp_new_space(struct sock *sk) |
| 4786 | { |
| 4787 | struct tcp_sock *tp = tcp_sk(sk); |
| 4788 | |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 4789 | if (tcp_should_expand_sndbuf(sk)) { |
Eric Dumazet | 6ae7053 | 2013-10-01 10:23:44 -0700 | [diff] [blame] | 4790 | tcp_sndbuf_expand(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4791 | tp->snd_cwnd_stamp = tcp_time_stamp; |
| 4792 | } |
| 4793 | |
| 4794 | sk->sk_write_space(sk); |
| 4795 | } |
| 4796 | |
Stephen Hemminger | 40efc6f | 2006-01-03 16:03:49 -0800 | [diff] [blame] | 4797 | static void tcp_check_space(struct sock *sk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4798 | { |
| 4799 | if (sock_flag(sk, SOCK_QUEUE_SHRUNK)) { |
| 4800 | sock_reset_flag(sk, SOCK_QUEUE_SHRUNK); |
| 4801 | if (sk->sk_socket && |
| 4802 | test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) |
| 4803 | tcp_new_space(sk); |
| 4804 | } |
| 4805 | } |
| 4806 | |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 4807 | static inline void tcp_data_snd_check(struct sock *sk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4808 | { |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 4809 | tcp_push_pending_frames(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4810 | tcp_check_space(sk); |
| 4811 | } |
| 4812 | |
| 4813 | /* |
| 4814 | * Check if sending an ack is needed. |
| 4815 | */ |
| 4816 | static void __tcp_ack_snd_check(struct sock *sk, int ofo_possible) |
| 4817 | { |
| 4818 | struct tcp_sock *tp = tcp_sk(sk); |
| 4819 | |
| 4820 | /* More than one full frame received... */ |
Joe Perches | 9d4fb27 | 2009-11-23 10:41:23 -0800 | [diff] [blame] | 4821 | 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] | 4822 | /* ... and right edge of window advances far enough. |
| 4823 | * (tcp_recvmsg() will send ACK otherwise). Or... |
| 4824 | */ |
Joe Perches | 9d4fb27 | 2009-11-23 10:41:23 -0800 | [diff] [blame] | 4825 | __tcp_select_window(sk) >= tp->rcv_wnd) || |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4826 | /* We ACK each frame or... */ |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 4827 | tcp_in_quickack_mode(sk) || |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4828 | /* We have out of order data. */ |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 4829 | (ofo_possible && skb_peek(&tp->out_of_order_queue))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4830 | /* Then ack it now */ |
| 4831 | tcp_send_ack(sk); |
| 4832 | } else { |
| 4833 | /* Else, send delayed ack. */ |
| 4834 | tcp_send_delayed_ack(sk); |
| 4835 | } |
| 4836 | } |
| 4837 | |
Stephen Hemminger | 40efc6f | 2006-01-03 16:03:49 -0800 | [diff] [blame] | 4838 | static inline void tcp_ack_snd_check(struct sock *sk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4839 | { |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 4840 | if (!inet_csk_ack_scheduled(sk)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4841 | /* We sent a data segment already. */ |
| 4842 | return; |
| 4843 | } |
| 4844 | __tcp_ack_snd_check(sk, 1); |
| 4845 | } |
| 4846 | |
| 4847 | /* |
| 4848 | * This routine is only called when we have urgent data |
Stephen Hemminger | caa20d9a | 2005-11-10 17:13:47 -0800 | [diff] [blame] | 4849 | * signaled. Its the 'slow' part of tcp_urg. It could be |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4850 | * moved inline now as tcp_urg is only called from one |
| 4851 | * place. We handle URGent data wrong. We have to - as |
| 4852 | * BSD still doesn't use the correction from RFC961. |
| 4853 | * For 1003.1g we should support a new option TCP_STDURG to permit |
| 4854 | * either form (or just set the sysctl tcp_stdurg). |
| 4855 | */ |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 4856 | |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 4857 | static void tcp_check_urg(struct sock *sk, const struct tcphdr *th) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4858 | { |
| 4859 | struct tcp_sock *tp = tcp_sk(sk); |
| 4860 | u32 ptr = ntohs(th->urg_ptr); |
| 4861 | |
| 4862 | if (ptr && !sysctl_tcp_stdurg) |
| 4863 | ptr--; |
| 4864 | ptr += ntohl(th->seq); |
| 4865 | |
| 4866 | /* Ignore urgent data that we've already seen and read. */ |
| 4867 | if (after(tp->copied_seq, ptr)) |
| 4868 | return; |
| 4869 | |
| 4870 | /* Do not replay urg ptr. |
| 4871 | * |
| 4872 | * NOTE: interesting situation not covered by specs. |
| 4873 | * Misbehaving sender may send urg ptr, pointing to segment, |
| 4874 | * which we already have in ofo queue. We are not able to fetch |
| 4875 | * such data and will stay in TCP_URG_NOTYET until will be eaten |
| 4876 | * by recvmsg(). Seems, we are not obliged to handle such wicked |
| 4877 | * situations. But it is worth to think about possibility of some |
| 4878 | * DoSes using some hypothetical application level deadlock. |
| 4879 | */ |
| 4880 | if (before(ptr, tp->rcv_nxt)) |
| 4881 | return; |
| 4882 | |
| 4883 | /* Do we already have a newer (or duplicate) urgent pointer? */ |
| 4884 | if (tp->urg_data && !after(ptr, tp->urg_seq)) |
| 4885 | return; |
| 4886 | |
| 4887 | /* Tell the world about our new urgent pointer. */ |
| 4888 | sk_send_sigurg(sk); |
| 4889 | |
| 4890 | /* We may be adding urgent data when the last byte read was |
| 4891 | * urgent. To do this requires some care. We cannot just ignore |
| 4892 | * tp->copied_seq since we would read the last urgent byte again |
| 4893 | * as data, nor can we alter copied_seq until this data arrives |
Stephen Hemminger | caa20d9a | 2005-11-10 17:13:47 -0800 | [diff] [blame] | 4894 | * or we break the semantics of SIOCATMARK (and thus sockatmark()) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4895 | * |
| 4896 | * NOTE. Double Dutch. Rendering to plain English: author of comment |
| 4897 | * above did something sort of send("A", MSG_OOB); send("B", MSG_OOB); |
| 4898 | * and expect that both A and B disappear from stream. This is _wrong_. |
| 4899 | * Though this happens in BSD with high probability, this is occasional. |
| 4900 | * Any application relying on this is buggy. Note also, that fix "works" |
| 4901 | * only in this artificial test. Insert some normal data between A and B and we will |
| 4902 | * decline of BSD again. Verdict: it is better to remove to trap |
| 4903 | * buggy users. |
| 4904 | */ |
| 4905 | if (tp->urg_seq == tp->copied_seq && tp->urg_data && |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 4906 | !sock_flag(sk, SOCK_URGINLINE) && tp->copied_seq != tp->rcv_nxt) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4907 | struct sk_buff *skb = skb_peek(&sk->sk_receive_queue); |
| 4908 | tp->copied_seq++; |
| 4909 | 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] | 4910 | __skb_unlink(skb, &sk->sk_receive_queue); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4911 | __kfree_skb(skb); |
| 4912 | } |
| 4913 | } |
| 4914 | |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 4915 | tp->urg_data = TCP_URG_NOTYET; |
| 4916 | tp->urg_seq = ptr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4917 | |
| 4918 | /* Disable header prediction. */ |
| 4919 | tp->pred_flags = 0; |
| 4920 | } |
| 4921 | |
| 4922 | /* This is the 'fast' part of urgent handling. */ |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 4923 | 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] | 4924 | { |
| 4925 | struct tcp_sock *tp = tcp_sk(sk); |
| 4926 | |
| 4927 | /* Check if we get a new urgent pointer - normally not. */ |
| 4928 | if (th->urg) |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 4929 | tcp_check_urg(sk, th); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4930 | |
| 4931 | /* Do we wait for any urgent data? - normally not... */ |
| 4932 | if (tp->urg_data == TCP_URG_NOTYET) { |
| 4933 | u32 ptr = tp->urg_seq - ntohl(th->seq) + (th->doff * 4) - |
| 4934 | th->syn; |
| 4935 | |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 4936 | /* Is the urgent pointer pointing into this packet? */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4937 | if (ptr < skb->len) { |
| 4938 | u8 tmp; |
| 4939 | if (skb_copy_bits(skb, ptr, &tmp, 1)) |
| 4940 | BUG(); |
| 4941 | tp->urg_data = TCP_URG_VALID | tmp; |
| 4942 | if (!sock_flag(sk, SOCK_DEAD)) |
David S. Miller | 676d236 | 2014-04-11 16:15:36 -0400 | [diff] [blame] | 4943 | sk->sk_data_ready(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4944 | } |
| 4945 | } |
| 4946 | } |
| 4947 | |
| 4948 | static int tcp_copy_to_iovec(struct sock *sk, struct sk_buff *skb, int hlen) |
| 4949 | { |
| 4950 | struct tcp_sock *tp = tcp_sk(sk); |
| 4951 | int chunk = skb->len - hlen; |
| 4952 | int err; |
| 4953 | |
| 4954 | local_bh_enable(); |
Herbert Xu | 6047637 | 2007-04-09 11:59:39 -0700 | [diff] [blame] | 4955 | if (skb_csum_unnecessary(skb)) |
Al Viro | f4362a2 | 2014-11-24 13:26:06 -0500 | [diff] [blame] | 4956 | err = skb_copy_datagram_msg(skb, hlen, tp->ucopy.msg, chunk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4957 | else |
Al Viro | f4362a2 | 2014-11-24 13:26:06 -0500 | [diff] [blame] | 4958 | err = skb_copy_and_csum_datagram_msg(skb, hlen, tp->ucopy.msg); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4959 | |
| 4960 | if (!err) { |
| 4961 | tp->ucopy.len -= chunk; |
| 4962 | tp->copied_seq += chunk; |
| 4963 | tcp_rcv_space_adjust(sk); |
| 4964 | } |
| 4965 | |
| 4966 | local_bh_disable(); |
| 4967 | return err; |
| 4968 | } |
| 4969 | |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 4970 | static __sum16 __tcp_checksum_complete_user(struct sock *sk, |
| 4971 | struct sk_buff *skb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4972 | { |
Al Viro | b51655b | 2006-11-14 21:40:42 -0800 | [diff] [blame] | 4973 | __sum16 result; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4974 | |
| 4975 | if (sock_owned_by_user(sk)) { |
| 4976 | local_bh_enable(); |
| 4977 | result = __tcp_checksum_complete(skb); |
| 4978 | local_bh_disable(); |
| 4979 | } else { |
| 4980 | result = __tcp_checksum_complete(skb); |
| 4981 | } |
| 4982 | return result; |
| 4983 | } |
| 4984 | |
Vijay Subramanian | 67b95bd | 2012-07-19 21:32:18 +0000 | [diff] [blame] | 4985 | static inline bool tcp_checksum_complete_user(struct sock *sk, |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 4986 | struct sk_buff *skb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4987 | { |
Herbert Xu | 6047637 | 2007-04-09 11:59:39 -0700 | [diff] [blame] | 4988 | return !skb_csum_unnecessary(skb) && |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 4989 | __tcp_checksum_complete_user(sk, skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4990 | } |
| 4991 | |
Ilpo Järvinen | cbe2d12 | 2008-08-23 05:10:12 -0700 | [diff] [blame] | 4992 | /* Does PAWS and seqno based validation of an incoming segment, flags will |
| 4993 | * play significant role here. |
| 4994 | */ |
Eric Dumazet | 0c24604 | 2012-07-17 01:41:30 +0000 | [diff] [blame] | 4995 | static bool tcp_validate_incoming(struct sock *sk, struct sk_buff *skb, |
| 4996 | const struct tcphdr *th, int syn_inerr) |
Ilpo Järvinen | cbe2d12 | 2008-08-23 05:10:12 -0700 | [diff] [blame] | 4997 | { |
| 4998 | struct tcp_sock *tp = tcp_sk(sk); |
| 4999 | |
| 5000 | /* RFC1323: H1. Apply PAWS check first. */ |
Christoph Paasch | 1a2c618 | 2013-03-17 08:23:34 +0000 | [diff] [blame] | 5001 | 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] | 5002 | tcp_paws_discard(sk, skb)) { |
| 5003 | if (!th->rst) { |
| 5004 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSESTABREJECTED); |
Neal Cardwell | f2b2c58 | 2015-02-06 16:04:40 -0500 | [diff] [blame] | 5005 | if (!tcp_oow_rate_limited(sock_net(sk), skb, |
| 5006 | LINUX_MIB_TCPACKSKIPPEDPAWS, |
| 5007 | &tp->last_oow_ack_time)) |
| 5008 | tcp_send_dupack(sk, skb); |
Ilpo Järvinen | cbe2d12 | 2008-08-23 05:10:12 -0700 | [diff] [blame] | 5009 | goto discard; |
| 5010 | } |
| 5011 | /* Reset is accepted even if it did not pass PAWS. */ |
| 5012 | } |
| 5013 | |
| 5014 | /* Step 1: check sequence number */ |
| 5015 | if (!tcp_sequence(tp, TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq)) { |
| 5016 | /* RFC793, page 37: "In all states except SYN-SENT, all reset |
| 5017 | * (RST) segments are validated by checking their SEQ-fields." |
| 5018 | * And page 69: "If an incoming segment is not acceptable, |
| 5019 | * an acknowledgment should be sent in reply (unless the RST |
| 5020 | * bit is set, if so drop the segment and return)". |
| 5021 | */ |
Eric Dumazet | e371589 | 2012-07-17 12:29:30 +0000 | [diff] [blame] | 5022 | if (!th->rst) { |
| 5023 | if (th->syn) |
| 5024 | goto syn_challenge; |
Neal Cardwell | f2b2c58 | 2015-02-06 16:04:40 -0500 | [diff] [blame] | 5025 | if (!tcp_oow_rate_limited(sock_net(sk), skb, |
| 5026 | LINUX_MIB_TCPACKSKIPPEDSEQ, |
| 5027 | &tp->last_oow_ack_time)) |
| 5028 | tcp_send_dupack(sk, skb); |
Eric Dumazet | e371589 | 2012-07-17 12:29:30 +0000 | [diff] [blame] | 5029 | } |
Ilpo Järvinen | cbe2d12 | 2008-08-23 05:10:12 -0700 | [diff] [blame] | 5030 | goto discard; |
| 5031 | } |
| 5032 | |
| 5033 | /* Step 2: check RST bit */ |
| 5034 | if (th->rst) { |
Eric Dumazet | 282f23c | 2012-07-17 10:13:05 +0200 | [diff] [blame] | 5035 | /* RFC 5961 3.2 : |
| 5036 | * If sequence number exactly matches RCV.NXT, then |
| 5037 | * RESET the connection |
| 5038 | * else |
| 5039 | * Send a challenge ACK |
| 5040 | */ |
| 5041 | if (TCP_SKB_CB(skb)->seq == tp->rcv_nxt) |
| 5042 | tcp_reset(sk); |
| 5043 | else |
Neal Cardwell | f2b2c58 | 2015-02-06 16:04:40 -0500 | [diff] [blame] | 5044 | tcp_send_challenge_ack(sk, skb); |
Ilpo Järvinen | cbe2d12 | 2008-08-23 05:10:12 -0700 | [diff] [blame] | 5045 | goto discard; |
| 5046 | } |
| 5047 | |
Ilpo Järvinen | cbe2d12 | 2008-08-23 05:10:12 -0700 | [diff] [blame] | 5048 | /* step 3: check security and precedence [ignored] */ |
| 5049 | |
Eric Dumazet | 0c24604 | 2012-07-17 01:41:30 +0000 | [diff] [blame] | 5050 | /* step 4: Check for a SYN |
Sowmini Varadhan | cd21453 | 2014-10-30 12:48:08 -0400 | [diff] [blame] | 5051 | * RFC 5961 4.2 : Send a challenge ack |
Eric Dumazet | 0c24604 | 2012-07-17 01:41:30 +0000 | [diff] [blame] | 5052 | */ |
| 5053 | if (th->syn) { |
Eric Dumazet | e371589 | 2012-07-17 12:29:30 +0000 | [diff] [blame] | 5054 | syn_challenge: |
Ilpo Järvinen | cbe2d12 | 2008-08-23 05:10:12 -0700 | [diff] [blame] | 5055 | if (syn_inerr) |
| 5056 | TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS); |
Eric Dumazet | 0c24604 | 2012-07-17 01:41:30 +0000 | [diff] [blame] | 5057 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPSYNCHALLENGE); |
Neal Cardwell | f2b2c58 | 2015-02-06 16:04:40 -0500 | [diff] [blame] | 5058 | tcp_send_challenge_ack(sk, skb); |
Eric Dumazet | 0c24604 | 2012-07-17 01:41:30 +0000 | [diff] [blame] | 5059 | goto discard; |
Ilpo Järvinen | cbe2d12 | 2008-08-23 05:10:12 -0700 | [diff] [blame] | 5060 | } |
| 5061 | |
Eric Dumazet | 0c24604 | 2012-07-17 01:41:30 +0000 | [diff] [blame] | 5062 | return true; |
Ilpo Järvinen | cbe2d12 | 2008-08-23 05:10:12 -0700 | [diff] [blame] | 5063 | |
| 5064 | discard: |
| 5065 | __kfree_skb(skb); |
Eric Dumazet | 0c24604 | 2012-07-17 01:41:30 +0000 | [diff] [blame] | 5066 | return false; |
Ilpo Järvinen | cbe2d12 | 2008-08-23 05:10:12 -0700 | [diff] [blame] | 5067 | } |
| 5068 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5069 | /* |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5070 | * TCP receive function for the ESTABLISHED state. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5071 | * |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5072 | * 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] | 5073 | * disabled when: |
| 5074 | * - A zero window was announced from us - zero window probing |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5075 | * is only handled properly in the slow path. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5076 | * - Out of order segments arrived. |
| 5077 | * - Urgent data is expected. |
| 5078 | * - There is no buffer space left |
| 5079 | * - Unexpected TCP flags/window values/header lengths are received |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5080 | * (detected by checking the TCP header against pred_flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5081 | * - Data is sent in both directions. Fast path only supports pure senders |
| 5082 | * or pure receivers (this means either the sequence number or the ack |
| 5083 | * value must stay constant) |
| 5084 | * - Unexpected TCP option. |
| 5085 | * |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5086 | * When these conditions are not satisfied it drops into a standard |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5087 | * receive procedure patterned after RFC793 to handle all cases. |
| 5088 | * The first three cases are guaranteed by proper pred_flags setting, |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5089 | * the rest is checked inline. Fast processing is turned on in |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5090 | * tcp_data_queue when everything is OK. |
| 5091 | */ |
Vijay Subramanian | c995ae2 | 2013-09-03 12:23:22 -0700 | [diff] [blame] | 5092 | void tcp_rcv_established(struct sock *sk, struct sk_buff *skb, |
| 5093 | const struct tcphdr *th, unsigned int len) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5094 | { |
| 5095 | struct tcp_sock *tp = tcp_sk(sk); |
| 5096 | |
Eric Dumazet | 5d299f3 | 2012-08-06 05:09:33 +0000 | [diff] [blame] | 5097 | if (unlikely(sk->sk_rx_dst == NULL)) |
| 5098 | inet_csk(sk)->icsk_af_ops->sk_rx_dst_set(sk, skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5099 | /* |
| 5100 | * Header prediction. |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5101 | * The code loosely follows the one in the famous |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5102 | * "30 instruction TCP receive" Van Jacobson mail. |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5103 | * |
| 5104 | * Van's trick is to deposit buffers into socket queue |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5105 | * on a device interrupt, to call tcp_recv function |
| 5106 | * on the receive process context and checksum and copy |
| 5107 | * the buffer to user space. smart... |
| 5108 | * |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5109 | * Our current scheme is not silly either but we take the |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5110 | * extra cost of the net_bh soft interrupt processing... |
| 5111 | * We do checksum and copy also but from device to kernel. |
| 5112 | */ |
| 5113 | |
| 5114 | tp->rx_opt.saw_tstamp = 0; |
| 5115 | |
| 5116 | /* pred_flags is 0xS?10 << 16 + snd_wnd |
Stephen Hemminger | caa20d9a | 2005-11-10 17:13:47 -0800 | [diff] [blame] | 5117 | * if header_prediction is to be made |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5118 | * 'S' will always be tp->tcp_header_len >> 2 |
| 5119 | * '?' 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] | 5120 | * turn it off (when there are holes in the receive |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5121 | * space for instance) |
| 5122 | * PSH flag is ignored. |
| 5123 | */ |
| 5124 | |
| 5125 | if ((tcp_flag_word(th) & TCP_HP_BITS) == tp->pred_flags && |
John Dykstra | 96e0bf4 | 2009-03-22 21:49:57 -0700 | [diff] [blame] | 5126 | TCP_SKB_CB(skb)->seq == tp->rcv_nxt && |
| 5127 | !after(TCP_SKB_CB(skb)->ack_seq, tp->snd_nxt)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5128 | int tcp_header_len = tp->tcp_header_len; |
| 5129 | |
| 5130 | /* Timestamp header prediction: tcp_header_len |
| 5131 | * is automatically equal to th->doff*4 due to pred_flags |
| 5132 | * match. |
| 5133 | */ |
| 5134 | |
| 5135 | /* Check timestamp */ |
| 5136 | if (tcp_header_len == sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5137 | /* No? Slow path! */ |
Ilpo Järvinen | a4356b2 | 2008-08-23 05:12:29 -0700 | [diff] [blame] | 5138 | if (!tcp_parse_aligned_timestamp(tp, th)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5139 | goto slow_path; |
| 5140 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5141 | /* If PAWS failed, check it more carefully in slow path */ |
| 5142 | if ((s32)(tp->rx_opt.rcv_tsval - tp->rx_opt.ts_recent) < 0) |
| 5143 | goto slow_path; |
| 5144 | |
| 5145 | /* DO NOT update ts_recent here, if checksum fails |
| 5146 | * and timestamp was corrupted part, it will result |
| 5147 | * in a hung connection since we will drop all |
| 5148 | * future packets due to the PAWS test. |
| 5149 | */ |
| 5150 | } |
| 5151 | |
| 5152 | if (len <= tcp_header_len) { |
| 5153 | /* Bulk data transfer: sender */ |
| 5154 | if (len == tcp_header_len) { |
| 5155 | /* Predicted packet is in window by definition. |
| 5156 | * seq == rcv_nxt and rcv_wup <= rcv_nxt. |
| 5157 | * Hence, check seq<=rcv_wup reduces to: |
| 5158 | */ |
| 5159 | if (tcp_header_len == |
| 5160 | (sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED) && |
| 5161 | tp->rcv_nxt == tp->rcv_wup) |
| 5162 | tcp_store_ts_recent(tp); |
| 5163 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5164 | /* We know that such packets are checksummed |
| 5165 | * on entry. |
| 5166 | */ |
| 5167 | tcp_ack(sk, skb, 0); |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5168 | __kfree_skb(skb); |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 5169 | tcp_data_snd_check(sk); |
Vijay Subramanian | c995ae2 | 2013-09-03 12:23:22 -0700 | [diff] [blame] | 5170 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5171 | } else { /* Header too small */ |
Pavel Emelyanov | 63231bd | 2008-07-16 20:22:25 -0700 | [diff] [blame] | 5172 | TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5173 | goto discard; |
| 5174 | } |
| 5175 | } else { |
| 5176 | int eaten = 0; |
Eric Dumazet | b081f85 | 2012-05-02 09:58:29 +0000 | [diff] [blame] | 5177 | bool fragstolen = false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5178 | |
Dan Williams | d27f9bc | 2013-12-30 11:37:15 -0800 | [diff] [blame] | 5179 | if (tp->ucopy.task == current && |
| 5180 | tp->copied_seq == tp->rcv_nxt && |
| 5181 | len - tcp_header_len <= tp->ucopy.len && |
| 5182 | sock_owned_by_user(sk)) { |
| 5183 | __set_current_state(TASK_RUNNING); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5184 | |
Dan Williams | d27f9bc | 2013-12-30 11:37:15 -0800 | [diff] [blame] | 5185 | if (!tcp_copy_to_iovec(sk, skb, tcp_header_len)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5186 | /* Predicted packet is in window by definition. |
| 5187 | * seq == rcv_nxt and rcv_wup <= rcv_nxt. |
| 5188 | * Hence, check seq<=rcv_wup reduces to: |
| 5189 | */ |
| 5190 | if (tcp_header_len == |
| 5191 | (sizeof(struct tcphdr) + |
| 5192 | TCPOLEN_TSTAMP_ALIGNED) && |
| 5193 | tp->rcv_nxt == tp->rcv_wup) |
| 5194 | tcp_store_ts_recent(tp); |
| 5195 | |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 5196 | tcp_rcv_rtt_measure_ts(sk, skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5197 | |
| 5198 | __skb_pull(skb, tcp_header_len); |
| 5199 | tp->rcv_nxt = TCP_SKB_CB(skb)->end_seq; |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 5200 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPHPHITSTOUSER); |
Dan Williams | d27f9bc | 2013-12-30 11:37:15 -0800 | [diff] [blame] | 5201 | eaten = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5202 | } |
| 5203 | } |
| 5204 | if (!eaten) { |
| 5205 | if (tcp_checksum_complete_user(sk, skb)) |
| 5206 | goto csum_error; |
| 5207 | |
Neal Cardwell | aab2b4b | 2013-03-04 06:23:05 +0000 | [diff] [blame] | 5208 | if ((int)skb->truesize > sk->sk_forward_alloc) |
| 5209 | goto step5; |
| 5210 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5211 | /* Predicted packet is in window by definition. |
| 5212 | * seq == rcv_nxt and rcv_wup <= rcv_nxt. |
| 5213 | * Hence, check seq<=rcv_wup reduces to: |
| 5214 | */ |
| 5215 | if (tcp_header_len == |
| 5216 | (sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED) && |
| 5217 | tp->rcv_nxt == tp->rcv_wup) |
| 5218 | tcp_store_ts_recent(tp); |
| 5219 | |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 5220 | tcp_rcv_rtt_measure_ts(sk, skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5221 | |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 5222 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPHPHITS); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5223 | |
| 5224 | /* Bulk data transfer: receiver */ |
Eric Dumazet | b081f85 | 2012-05-02 09:58:29 +0000 | [diff] [blame] | 5225 | eaten = tcp_queue_rcv(sk, skb, tcp_header_len, |
| 5226 | &fragstolen); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5227 | } |
| 5228 | |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 5229 | tcp_event_data_recv(sk, skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5230 | |
| 5231 | if (TCP_SKB_CB(skb)->ack_seq != tp->snd_una) { |
| 5232 | /* Well, only one small jumplet in fast path... */ |
| 5233 | tcp_ack(sk, skb, FLAG_DATA); |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 5234 | tcp_data_snd_check(sk); |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 5235 | if (!inet_csk_ack_scheduled(sk)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5236 | goto no_ack; |
| 5237 | } |
| 5238 | |
Dan Williams | d27f9bc | 2013-12-30 11:37:15 -0800 | [diff] [blame] | 5239 | __tcp_ack_snd_check(sk, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5240 | no_ack: |
| 5241 | if (eaten) |
Eric Dumazet | b081f85 | 2012-05-02 09:58:29 +0000 | [diff] [blame] | 5242 | kfree_skb_partial(skb, fragstolen); |
David S. Miller | 676d236 | 2014-04-11 16:15:36 -0400 | [diff] [blame] | 5243 | sk->sk_data_ready(sk); |
Vijay Subramanian | c995ae2 | 2013-09-03 12:23:22 -0700 | [diff] [blame] | 5244 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5245 | } |
| 5246 | } |
| 5247 | |
| 5248 | slow_path: |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 5249 | if (len < (th->doff << 2) || tcp_checksum_complete_user(sk, skb)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5250 | goto csum_error; |
| 5251 | |
Calvin Owens | 0c228e8 | 2014-11-20 15:09:53 -0800 | [diff] [blame] | 5252 | if (!th->ack && !th->rst && !th->syn) |
Eric Dumazet | c3ae62af | 2012-12-26 12:44:34 +0000 | [diff] [blame] | 5253 | goto discard; |
| 5254 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5255 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5256 | * Standard slow path. |
| 5257 | */ |
| 5258 | |
Eric Dumazet | 0c24604 | 2012-07-17 01:41:30 +0000 | [diff] [blame] | 5259 | if (!tcp_validate_incoming(sk, skb, th, 1)) |
Vijay Subramanian | c995ae2 | 2013-09-03 12:23:22 -0700 | [diff] [blame] | 5260 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5261 | |
| 5262 | step5: |
Eric Dumazet | 12fb3dd | 2013-04-19 07:19:48 +0000 | [diff] [blame] | 5263 | if (tcp_ack(sk, skb, FLAG_SLOWPATH | FLAG_UPDATE_TS_RECENT) < 0) |
John Dykstra | 96e0bf4 | 2009-03-22 21:49:57 -0700 | [diff] [blame] | 5264 | goto discard; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5265 | |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 5266 | tcp_rcv_rtt_measure_ts(sk, skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5267 | |
| 5268 | /* Process urgent data. */ |
| 5269 | tcp_urg(sk, skb, th); |
| 5270 | |
| 5271 | /* step 7: process the segment text */ |
| 5272 | tcp_data_queue(sk, skb); |
| 5273 | |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 5274 | tcp_data_snd_check(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5275 | tcp_ack_snd_check(sk); |
Vijay Subramanian | c995ae2 | 2013-09-03 12:23:22 -0700 | [diff] [blame] | 5276 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5277 | |
| 5278 | csum_error: |
Eric Dumazet | 6a5dc9e | 2013-04-29 08:39:56 +0000 | [diff] [blame] | 5279 | TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS); |
Pavel Emelyanov | 63231bd | 2008-07-16 20:22:25 -0700 | [diff] [blame] | 5280 | TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5281 | |
| 5282 | discard: |
| 5283 | __kfree_skb(skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5284 | } |
Eric Dumazet | 4bc2f18 | 2010-07-09 21:22:10 +0000 | [diff] [blame] | 5285 | EXPORT_SYMBOL(tcp_rcv_established); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5286 | |
Pavel Emelyanov | 370816a | 2012-04-19 03:40:01 +0000 | [diff] [blame] | 5287 | void tcp_finish_connect(struct sock *sk, struct sk_buff *skb) |
| 5288 | { |
| 5289 | struct tcp_sock *tp = tcp_sk(sk); |
| 5290 | struct inet_connection_sock *icsk = inet_csk(sk); |
| 5291 | |
| 5292 | tcp_set_state(sk, TCP_ESTABLISHED); |
| 5293 | |
David S. Miller | 41063e9 | 2012-06-19 21:22:05 -0700 | [diff] [blame] | 5294 | if (skb != NULL) { |
Eric Dumazet | 5d299f3 | 2012-08-06 05:09:33 +0000 | [diff] [blame] | 5295 | icsk->icsk_af_ops->sk_rx_dst_set(sk, skb); |
Pavel Emelyanov | 370816a | 2012-04-19 03:40:01 +0000 | [diff] [blame] | 5296 | security_inet_conn_established(sk, skb); |
David S. Miller | 41063e9 | 2012-06-19 21:22:05 -0700 | [diff] [blame] | 5297 | } |
Pavel Emelyanov | 370816a | 2012-04-19 03:40:01 +0000 | [diff] [blame] | 5298 | |
| 5299 | /* Make sure socket is routed, for correct metrics. */ |
| 5300 | icsk->icsk_af_ops->rebuild_header(sk); |
| 5301 | |
| 5302 | tcp_init_metrics(sk); |
| 5303 | |
| 5304 | tcp_init_congestion_control(sk); |
| 5305 | |
| 5306 | /* Prevent spurious tcp_cwnd_restart() on first data |
| 5307 | * packet. |
| 5308 | */ |
| 5309 | tp->lsndtime = tcp_time_stamp; |
| 5310 | |
| 5311 | tcp_init_buffer_space(sk); |
| 5312 | |
| 5313 | if (sock_flag(sk, SOCK_KEEPOPEN)) |
| 5314 | inet_csk_reset_keepalive_timer(sk, keepalive_time_when(tp)); |
| 5315 | |
| 5316 | if (!tp->rx_opt.snd_wscale) |
| 5317 | __tcp_fast_path_on(tp, tp->snd_wnd); |
| 5318 | else |
| 5319 | tp->pred_flags = 0; |
| 5320 | |
| 5321 | if (!sock_flag(sk, SOCK_DEAD)) { |
| 5322 | sk->sk_state_change(sk); |
| 5323 | sk_wake_async(sk, SOCK_WAKE_IO, POLL_OUT); |
| 5324 | } |
| 5325 | } |
| 5326 | |
Yuchung Cheng | 8e4178c | 2012-07-19 06:43:08 +0000 | [diff] [blame] | 5327 | static bool tcp_rcv_fastopen_synack(struct sock *sk, struct sk_buff *synack, |
| 5328 | struct tcp_fastopen_cookie *cookie) |
| 5329 | { |
| 5330 | struct tcp_sock *tp = tcp_sk(sk); |
Yuchung Cheng | 67da22d | 2012-07-19 06:43:11 +0000 | [diff] [blame] | 5331 | 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] | 5332 | u16 mss = tp->rx_opt.mss_clamp; |
Yuchung Cheng | aab4874 | 2012-07-19 06:43:10 +0000 | [diff] [blame] | 5333 | bool syn_drop; |
Yuchung Cheng | 8e4178c | 2012-07-19 06:43:08 +0000 | [diff] [blame] | 5334 | |
| 5335 | if (mss == tp->rx_opt.user_mss) { |
| 5336 | struct tcp_options_received opt; |
Yuchung Cheng | 8e4178c | 2012-07-19 06:43:08 +0000 | [diff] [blame] | 5337 | |
| 5338 | /* Get original SYNACK MSS value if user MSS sets mss_clamp */ |
| 5339 | tcp_clear_options(&opt); |
| 5340 | opt.user_mss = opt.mss_clamp = 0; |
Christoph Paasch | 1a2c618 | 2013-03-17 08:23:34 +0000 | [diff] [blame] | 5341 | tcp_parse_options(synack, &opt, 0, NULL); |
Yuchung Cheng | 8e4178c | 2012-07-19 06:43:08 +0000 | [diff] [blame] | 5342 | mss = opt.mss_clamp; |
| 5343 | } |
| 5344 | |
Yuchung Cheng | 67da22d | 2012-07-19 06:43:11 +0000 | [diff] [blame] | 5345 | if (!tp->syn_fastopen) /* Ignore an unsolicited cookie */ |
| 5346 | cookie->len = -1; |
| 5347 | |
Yuchung Cheng | aab4874 | 2012-07-19 06:43:10 +0000 | [diff] [blame] | 5348 | /* The SYN-ACK neither has cookie nor acknowledges the data. Presumably |
| 5349 | * the remote receives only the retransmitted (regular) SYNs: either |
| 5350 | * the original SYN-data or the corresponding SYN-ACK is lost. |
| 5351 | */ |
Yuchung Cheng | 66555e9 | 2013-01-31 11:16:46 -0800 | [diff] [blame] | 5352 | syn_drop = (cookie->len <= 0 && data && tp->total_retrans); |
Yuchung Cheng | aab4874 | 2012-07-19 06:43:10 +0000 | [diff] [blame] | 5353 | |
| 5354 | tcp_fastopen_cache_set(sk, mss, cookie, syn_drop); |
Yuchung Cheng | 8e4178c | 2012-07-19 06:43:08 +0000 | [diff] [blame] | 5355 | |
| 5356 | if (data) { /* Retransmit unacked data in SYN */ |
Yuchung Cheng | 93b174a | 2012-12-06 08:45:32 +0000 | [diff] [blame] | 5357 | tcp_for_write_queue_from(data, sk) { |
| 5358 | if (data == tcp_send_head(sk) || |
| 5359 | __tcp_retransmit_skb(sk, data)) |
| 5360 | break; |
| 5361 | } |
Yuchung Cheng | 8e4178c | 2012-07-19 06:43:08 +0000 | [diff] [blame] | 5362 | tcp_rearm_rto(sk); |
Yuchung Cheng | f19c29e | 2014-03-03 12:31:36 -0800 | [diff] [blame] | 5363 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPFASTOPENACTIVEFAIL); |
Yuchung Cheng | 8e4178c | 2012-07-19 06:43:08 +0000 | [diff] [blame] | 5364 | return true; |
| 5365 | } |
Yuchung Cheng | 6f73601 | 2012-10-19 15:14:44 +0000 | [diff] [blame] | 5366 | tp->syn_data_acked = tp->syn_data; |
Yuchung Cheng | f19c29e | 2014-03-03 12:31:36 -0800 | [diff] [blame] | 5367 | if (tp->syn_data_acked) |
| 5368 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPFASTOPENACTIVE); |
Yuchung Cheng | 8e4178c | 2012-07-19 06:43:08 +0000 | [diff] [blame] | 5369 | return false; |
| 5370 | } |
| 5371 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5372 | 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] | 5373 | const struct tcphdr *th, unsigned int len) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5374 | { |
Arnaldo Carvalho de Melo | d83d846 | 2005-12-13 23:26:10 -0800 | [diff] [blame] | 5375 | struct inet_connection_sock *icsk = inet_csk(sk); |
William Allen Simpson | 4957faade | 2009-12-02 18:25:27 +0000 | [diff] [blame] | 5376 | struct tcp_sock *tp = tcp_sk(sk); |
Yuchung Cheng | 8e4178c | 2012-07-19 06:43:08 +0000 | [diff] [blame] | 5377 | struct tcp_fastopen_cookie foc = { .len = -1 }; |
William Allen Simpson | 4957faade | 2009-12-02 18:25:27 +0000 | [diff] [blame] | 5378 | int saved_clamp = tp->rx_opt.mss_clamp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5379 | |
Christoph Paasch | 1a2c618 | 2013-03-17 08:23:34 +0000 | [diff] [blame] | 5380 | tcp_parse_options(skb, &tp->rx_opt, 0, &foc); |
Andrew Vagin | e3e1202 | 2013-08-27 12:21:55 +0400 | [diff] [blame] | 5381 | if (tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr) |
Andrey Vagin | ee684b6 | 2013-02-11 05:50:19 +0000 | [diff] [blame] | 5382 | tp->rx_opt.rcv_tsecr -= tp->tsoffset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5383 | |
| 5384 | if (th->ack) { |
| 5385 | /* rfc793: |
| 5386 | * "If the state is SYN-SENT then |
| 5387 | * first check the ACK bit |
| 5388 | * If the ACK bit is set |
| 5389 | * If SEG.ACK =< ISS, or SEG.ACK > SND.NXT, send |
| 5390 | * a reset (unless the RST bit is set, if so drop |
| 5391 | * the segment and return)" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5392 | */ |
Yuchung Cheng | 8e4178c | 2012-07-19 06:43:08 +0000 | [diff] [blame] | 5393 | if (!after(TCP_SKB_CB(skb)->ack_seq, tp->snd_una) || |
| 5394 | after(TCP_SKB_CB(skb)->ack_seq, tp->snd_nxt)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5395 | goto reset_and_undo; |
| 5396 | |
| 5397 | if (tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr && |
| 5398 | !between(tp->rx_opt.rcv_tsecr, tp->retrans_stamp, |
| 5399 | tcp_time_stamp)) { |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 5400 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSACTIVEREJECTED); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5401 | goto reset_and_undo; |
| 5402 | } |
| 5403 | |
| 5404 | /* Now ACK is acceptable. |
| 5405 | * |
| 5406 | * "If the RST bit is set |
| 5407 | * If the ACK was acceptable then signal the user "error: |
| 5408 | * connection reset", drop the segment, enter CLOSED state, |
| 5409 | * delete TCB, and return." |
| 5410 | */ |
| 5411 | |
| 5412 | if (th->rst) { |
| 5413 | tcp_reset(sk); |
| 5414 | goto discard; |
| 5415 | } |
| 5416 | |
| 5417 | /* rfc793: |
| 5418 | * "fifth, if neither of the SYN or RST bits is set then |
| 5419 | * drop the segment and return." |
| 5420 | * |
| 5421 | * See note below! |
| 5422 | * --ANK(990513) |
| 5423 | */ |
| 5424 | if (!th->syn) |
| 5425 | goto discard_and_undo; |
| 5426 | |
| 5427 | /* rfc793: |
| 5428 | * "If the SYN bit is on ... |
| 5429 | * are acceptable then ... |
| 5430 | * (our SYN has been ACKed), change the connection |
| 5431 | * state to ESTABLISHED..." |
| 5432 | */ |
| 5433 | |
Florian Westphal | 735d383 | 2014-09-29 13:08:30 +0200 | [diff] [blame] | 5434 | tcp_ecn_rcv_synack(tp, th); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5435 | |
Razvan Ghitulete | 1b3a692 | 2012-08-14 16:30:20 +0300 | [diff] [blame] | 5436 | tcp_init_wl(tp, TCP_SKB_CB(skb)->seq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5437 | tcp_ack(sk, skb, FLAG_SLOWPATH); |
| 5438 | |
| 5439 | /* Ok.. it's good. Set up sequence numbers and |
| 5440 | * move to established. |
| 5441 | */ |
| 5442 | tp->rcv_nxt = TCP_SKB_CB(skb)->seq + 1; |
| 5443 | tp->rcv_wup = TCP_SKB_CB(skb)->seq + 1; |
| 5444 | |
| 5445 | /* RFC1323: The window in SYN & SYN/ACK segments is |
| 5446 | * never scaled. |
| 5447 | */ |
| 5448 | tp->snd_wnd = ntohs(th->window); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5449 | |
| 5450 | if (!tp->rx_opt.wscale_ok) { |
| 5451 | tp->rx_opt.snd_wscale = tp->rx_opt.rcv_wscale = 0; |
| 5452 | tp->window_clamp = min(tp->window_clamp, 65535U); |
| 5453 | } |
| 5454 | |
| 5455 | if (tp->rx_opt.saw_tstamp) { |
| 5456 | tp->rx_opt.tstamp_ok = 1; |
| 5457 | tp->tcp_header_len = |
| 5458 | sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED; |
| 5459 | tp->advmss -= TCPOLEN_TSTAMP_ALIGNED; |
| 5460 | tcp_store_ts_recent(tp); |
| 5461 | } else { |
| 5462 | tp->tcp_header_len = sizeof(struct tcphdr); |
| 5463 | } |
| 5464 | |
Ilpo Järvinen | e60402d | 2007-08-09 15:14:46 +0300 | [diff] [blame] | 5465 | if (tcp_is_sack(tp) && sysctl_tcp_fack) |
| 5466 | tcp_enable_fack(tp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5467 | |
John Heffner | 5d424d5 | 2006-03-20 17:53:41 -0800 | [diff] [blame] | 5468 | tcp_mtup_init(sk); |
Arnaldo Carvalho de Melo | d83d846 | 2005-12-13 23:26:10 -0800 | [diff] [blame] | 5469 | tcp_sync_mss(sk, icsk->icsk_pmtu_cookie); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5470 | tcp_initialize_rcv_mss(sk); |
| 5471 | |
| 5472 | /* Remember, tcp_poll() does not lock socket! |
| 5473 | * Change state from SYN-SENT only after copied_seq |
| 5474 | * is initialized. */ |
| 5475 | tp->copied_seq = tp->rcv_nxt; |
William Allen Simpson | 4957faade | 2009-12-02 18:25:27 +0000 | [diff] [blame] | 5476 | |
Ralf Baechle | e16aa20 | 2006-12-07 00:11:33 -0800 | [diff] [blame] | 5477 | smp_mb(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5478 | |
Pavel Emelyanov | 370816a | 2012-04-19 03:40:01 +0000 | [diff] [blame] | 5479 | tcp_finish_connect(sk, skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5480 | |
Yuchung Cheng | 67da22d | 2012-07-19 06:43:11 +0000 | [diff] [blame] | 5481 | if ((tp->syn_fastopen || tp->syn_data) && |
| 5482 | tcp_rcv_fastopen_synack(sk, skb, &foc)) |
Yuchung Cheng | 8e4178c | 2012-07-19 06:43:08 +0000 | [diff] [blame] | 5483 | return -1; |
| 5484 | |
Arnaldo Carvalho de Melo | 295f732 | 2005-08-09 20:11:56 -0700 | [diff] [blame] | 5485 | if (sk->sk_write_pending || |
| 5486 | icsk->icsk_accept_queue.rskq_defer_accept || |
| 5487 | icsk->icsk_ack.pingpong) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5488 | /* Save one ACK. Data will be ready after |
| 5489 | * several ticks, if write_pending is set. |
| 5490 | * |
| 5491 | * It may be deleted, but with this feature tcpdumps |
| 5492 | * look so _wonderfully_ clever, that I was not able |
| 5493 | * to stand against the temptation 8) --ANK |
| 5494 | */ |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 5495 | inet_csk_schedule_ack(sk); |
Arnaldo Carvalho de Melo | 295f732 | 2005-08-09 20:11:56 -0700 | [diff] [blame] | 5496 | icsk->icsk_ack.lrcvtime = tcp_time_stamp; |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 5497 | tcp_enter_quickack_mode(sk); |
Arnaldo Carvalho de Melo | 3f421ba | 2005-08-09 20:11:08 -0700 | [diff] [blame] | 5498 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, |
| 5499 | TCP_DELACK_MAX, TCP_RTO_MAX); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5500 | |
| 5501 | discard: |
| 5502 | __kfree_skb(skb); |
| 5503 | return 0; |
| 5504 | } else { |
| 5505 | tcp_send_ack(sk); |
| 5506 | } |
| 5507 | return -1; |
| 5508 | } |
| 5509 | |
| 5510 | /* No ACK in the segment */ |
| 5511 | |
| 5512 | if (th->rst) { |
| 5513 | /* rfc793: |
| 5514 | * "If the RST bit is set |
| 5515 | * |
| 5516 | * Otherwise (no ACK) drop the segment and return." |
| 5517 | */ |
| 5518 | |
| 5519 | goto discard_and_undo; |
| 5520 | } |
| 5521 | |
| 5522 | /* PAWS check. */ |
Ilpo Järvinen | 056834d | 2007-12-31 14:57:14 -0800 | [diff] [blame] | 5523 | if (tp->rx_opt.ts_recent_stamp && tp->rx_opt.saw_tstamp && |
Ilpo Järvinen | c887e6d | 2009-03-14 14:23:03 +0000 | [diff] [blame] | 5524 | tcp_paws_reject(&tp->rx_opt, 0)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5525 | goto discard_and_undo; |
| 5526 | |
| 5527 | if (th->syn) { |
| 5528 | /* We see SYN without ACK. It is attempt of |
| 5529 | * simultaneous connect with crossed SYNs. |
| 5530 | * Particularly, it can be connect to self. |
| 5531 | */ |
| 5532 | tcp_set_state(sk, TCP_SYN_RECV); |
| 5533 | |
| 5534 | if (tp->rx_opt.saw_tstamp) { |
| 5535 | tp->rx_opt.tstamp_ok = 1; |
| 5536 | tcp_store_ts_recent(tp); |
| 5537 | tp->tcp_header_len = |
| 5538 | sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED; |
| 5539 | } else { |
| 5540 | tp->tcp_header_len = sizeof(struct tcphdr); |
| 5541 | } |
| 5542 | |
| 5543 | tp->rcv_nxt = TCP_SKB_CB(skb)->seq + 1; |
| 5544 | tp->rcv_wup = TCP_SKB_CB(skb)->seq + 1; |
| 5545 | |
| 5546 | /* RFC1323: The window in SYN & SYN/ACK segments is |
| 5547 | * never scaled. |
| 5548 | */ |
| 5549 | tp->snd_wnd = ntohs(th->window); |
| 5550 | tp->snd_wl1 = TCP_SKB_CB(skb)->seq; |
| 5551 | tp->max_window = tp->snd_wnd; |
| 5552 | |
Florian Westphal | 735d383 | 2014-09-29 13:08:30 +0200 | [diff] [blame] | 5553 | tcp_ecn_rcv_syn(tp, th); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5554 | |
John Heffner | 5d424d5 | 2006-03-20 17:53:41 -0800 | [diff] [blame] | 5555 | tcp_mtup_init(sk); |
Arnaldo Carvalho de Melo | d83d846 | 2005-12-13 23:26:10 -0800 | [diff] [blame] | 5556 | tcp_sync_mss(sk, icsk->icsk_pmtu_cookie); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5557 | tcp_initialize_rcv_mss(sk); |
| 5558 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5559 | tcp_send_synack(sk); |
| 5560 | #if 0 |
| 5561 | /* Note, we could accept data and URG from this segment. |
Jerry Chu | 168a8f5 | 2012-08-31 12:29:13 +0000 | [diff] [blame] | 5562 | * There are no obstacles to make this (except that we must |
| 5563 | * either change tcp_recvmsg() to prevent it from returning data |
| 5564 | * before 3WHS completes per RFC793, or employ TCP Fast Open). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5565 | * |
| 5566 | * However, if we ignore data in ACKless segments sometimes, |
| 5567 | * we have no reasons to accept it sometimes. |
| 5568 | * Also, seems the code doing it in step6 of tcp_rcv_state_process |
| 5569 | * is not flawless. So, discard packet for sanity. |
| 5570 | * Uncomment this return to process the data. |
| 5571 | */ |
| 5572 | return -1; |
| 5573 | #else |
| 5574 | goto discard; |
| 5575 | #endif |
| 5576 | } |
| 5577 | /* "fifth, if neither of the SYN or RST bits is set then |
| 5578 | * drop the segment and return." |
| 5579 | */ |
| 5580 | |
| 5581 | discard_and_undo: |
| 5582 | tcp_clear_options(&tp->rx_opt); |
| 5583 | tp->rx_opt.mss_clamp = saved_clamp; |
| 5584 | goto discard; |
| 5585 | |
| 5586 | reset_and_undo: |
| 5587 | tcp_clear_options(&tp->rx_opt); |
| 5588 | tp->rx_opt.mss_clamp = saved_clamp; |
| 5589 | return 1; |
| 5590 | } |
| 5591 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5592 | /* |
| 5593 | * This function implements the receiving procedure of RFC 793 for |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5594 | * all states except ESTABLISHED and TIME_WAIT. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5595 | * It's called from both tcp_v4_rcv and tcp_v6_rcv and should be |
| 5596 | * address independent. |
| 5597 | */ |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5598 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5599 | int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb, |
Eric Dumazet | cf533ea | 2011-10-21 05:22:42 -0400 | [diff] [blame] | 5600 | const struct tcphdr *th, unsigned int len) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5601 | { |
| 5602 | struct tcp_sock *tp = tcp_sk(sk); |
Arnaldo Carvalho de Melo | 8292a17 | 2005-12-13 23:15:52 -0800 | [diff] [blame] | 5603 | struct inet_connection_sock *icsk = inet_csk(sk); |
Jerry Chu | 168a8f5 | 2012-08-31 12:29:13 +0000 | [diff] [blame] | 5604 | struct request_sock *req; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5605 | int queued = 0; |
Eric Dumazet | 1f6afc8 | 2013-05-24 15:03:54 +0000 | [diff] [blame] | 5606 | bool acceptable; |
Yuchung Cheng | bc15afa | 2013-10-24 08:44:25 -0700 | [diff] [blame] | 5607 | u32 synack_stamp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5608 | |
| 5609 | tp->rx_opt.saw_tstamp = 0; |
| 5610 | |
| 5611 | switch (sk->sk_state) { |
| 5612 | case TCP_CLOSE: |
| 5613 | goto discard; |
| 5614 | |
| 5615 | case TCP_LISTEN: |
Stephen Hemminger | 2de979b | 2007-03-08 20:45:19 -0800 | [diff] [blame] | 5616 | if (th->ack) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5617 | return 1; |
| 5618 | |
Stephen Hemminger | 2de979b | 2007-03-08 20:45:19 -0800 | [diff] [blame] | 5619 | if (th->rst) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5620 | goto discard; |
| 5621 | |
Stephen Hemminger | 2de979b | 2007-03-08 20:45:19 -0800 | [diff] [blame] | 5622 | if (th->syn) { |
Eric Dumazet | fdf5af0 | 2011-12-02 23:41:42 +0000 | [diff] [blame] | 5623 | if (th->fin) |
| 5624 | goto discard; |
Arnaldo Carvalho de Melo | 8292a17 | 2005-12-13 23:15:52 -0800 | [diff] [blame] | 5625 | if (icsk->icsk_af_ops->conn_request(sk, skb) < 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5626 | return 1; |
| 5627 | |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5628 | /* Now we have several options: In theory there is |
| 5629 | * nothing else in the frame. KA9Q has an option to |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5630 | * send data with the syn, BSD accepts data with the |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5631 | * syn up to the [to be] advertised window and |
| 5632 | * Solaris 2.1 gives you a protocol error. For now |
| 5633 | * we just ignore it, that fits the spec precisely |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5634 | * and avoids incompatibilities. It would be nice in |
| 5635 | * future to drop through and process the data. |
| 5636 | * |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5637 | * Now that TTCP is starting to be used we ought to |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5638 | * queue this data. |
| 5639 | * But, this leaves one open to an easy denial of |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5640 | * service attack, and SYN cookies can't defend |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5641 | * against this problem. So, we drop the data |
Masayuki Nakagawa | fb7e239 | 2007-01-23 20:15:06 -0800 | [diff] [blame] | 5642 | * in the interest of security over speed unless |
| 5643 | * it's still in use. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5644 | */ |
Masayuki Nakagawa | fb7e239 | 2007-01-23 20:15:06 -0800 | [diff] [blame] | 5645 | kfree_skb(skb); |
| 5646 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5647 | } |
| 5648 | goto discard; |
| 5649 | |
| 5650 | case TCP_SYN_SENT: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5651 | queued = tcp_rcv_synsent_state_process(sk, skb, th, len); |
| 5652 | if (queued >= 0) |
| 5653 | return queued; |
| 5654 | |
| 5655 | /* Do step6 onward by hand. */ |
| 5656 | tcp_urg(sk, skb, th); |
| 5657 | __kfree_skb(skb); |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 5658 | tcp_data_snd_check(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5659 | return 0; |
| 5660 | } |
| 5661 | |
Jerry Chu | 168a8f5 | 2012-08-31 12:29:13 +0000 | [diff] [blame] | 5662 | req = tp->fastopen_rsk; |
| 5663 | if (req != NULL) { |
Jerry Chu | 37561f6 | 2012-10-22 11:26:36 +0000 | [diff] [blame] | 5664 | WARN_ON_ONCE(sk->sk_state != TCP_SYN_RECV && |
Jerry Chu | 168a8f5 | 2012-08-31 12:29:13 +0000 | [diff] [blame] | 5665 | sk->sk_state != TCP_FIN_WAIT1); |
| 5666 | |
| 5667 | if (tcp_check_req(sk, skb, req, NULL, true) == NULL) |
| 5668 | goto discard; |
Neal Cardwell | e69bebd | 2012-09-22 04:18:57 +0000 | [diff] [blame] | 5669 | } |
Eric Dumazet | c3ae62af | 2012-12-26 12:44:34 +0000 | [diff] [blame] | 5670 | |
Calvin Owens | 0c228e8 | 2014-11-20 15:09:53 -0800 | [diff] [blame] | 5671 | if (!th->ack && !th->rst && !th->syn) |
Eric Dumazet | c3ae62af | 2012-12-26 12:44:34 +0000 | [diff] [blame] | 5672 | goto discard; |
| 5673 | |
Neal Cardwell | e69bebd | 2012-09-22 04:18:57 +0000 | [diff] [blame] | 5674 | if (!tcp_validate_incoming(sk, skb, th, 0)) |
Eric Dumazet | 0c24604 | 2012-07-17 01:41:30 +0000 | [diff] [blame] | 5675 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5676 | |
| 5677 | /* step 5: check the ACK field */ |
Eric Dumazet | 1f6afc8 | 2013-05-24 15:03:54 +0000 | [diff] [blame] | 5678 | acceptable = tcp_ack(sk, skb, FLAG_SLOWPATH | |
| 5679 | FLAG_UPDATE_TS_RECENT) > 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5680 | |
Eric Dumazet | 1f6afc8 | 2013-05-24 15:03:54 +0000 | [diff] [blame] | 5681 | switch (sk->sk_state) { |
| 5682 | case TCP_SYN_RECV: |
Joe Perches | 61eb9003 | 2013-05-24 18:36:13 +0000 | [diff] [blame] | 5683 | if (!acceptable) |
Eric Dumazet | 1f6afc8 | 2013-05-24 15:03:54 +0000 | [diff] [blame] | 5684 | return 1; |
Joe Perches | 61eb9003 | 2013-05-24 18:36:13 +0000 | [diff] [blame] | 5685 | |
| 5686 | /* Once we leave TCP_SYN_RECV, we no longer need req |
| 5687 | * so release it. |
| 5688 | */ |
| 5689 | if (req) { |
Yuchung Cheng | bc15afa | 2013-10-24 08:44:25 -0700 | [diff] [blame] | 5690 | synack_stamp = tcp_rsk(req)->snt_synack; |
Joe Perches | 61eb9003 | 2013-05-24 18:36:13 +0000 | [diff] [blame] | 5691 | tp->total_retrans = req->num_retrans; |
Joe Perches | 61eb9003 | 2013-05-24 18:36:13 +0000 | [diff] [blame] | 5692 | reqsk_fastopen_remove(sk, req, false); |
| 5693 | } else { |
Yuchung Cheng | bc15afa | 2013-10-24 08:44:25 -0700 | [diff] [blame] | 5694 | synack_stamp = tp->lsndtime; |
Joe Perches | 61eb9003 | 2013-05-24 18:36:13 +0000 | [diff] [blame] | 5695 | /* Make sure socket is routed, for correct metrics. */ |
| 5696 | icsk->icsk_af_ops->rebuild_header(sk); |
| 5697 | tcp_init_congestion_control(sk); |
| 5698 | |
| 5699 | tcp_mtup_init(sk); |
Joe Perches | 61eb9003 | 2013-05-24 18:36:13 +0000 | [diff] [blame] | 5700 | tp->copied_seq = tp->rcv_nxt; |
Eric Dumazet | b0983d3 | 2013-09-20 13:56:58 -0700 | [diff] [blame] | 5701 | tcp_init_buffer_space(sk); |
Eric Dumazet | 1f6afc8 | 2013-05-24 15:03:54 +0000 | [diff] [blame] | 5702 | } |
Joe Perches | 61eb9003 | 2013-05-24 18:36:13 +0000 | [diff] [blame] | 5703 | smp_mb(); |
| 5704 | tcp_set_state(sk, TCP_ESTABLISHED); |
| 5705 | sk->sk_state_change(sk); |
| 5706 | |
| 5707 | /* Note, that this wakeup is only for marginal crossed SYN case. |
| 5708 | * Passively open sockets are not waked up, because |
| 5709 | * sk->sk_sleep == NULL and sk->sk_socket == NULL. |
| 5710 | */ |
| 5711 | if (sk->sk_socket) |
| 5712 | sk_wake_async(sk, SOCK_WAKE_IO, POLL_OUT); |
| 5713 | |
| 5714 | tp->snd_una = TCP_SKB_CB(skb)->ack_seq; |
| 5715 | tp->snd_wnd = ntohs(th->window) << tp->rx_opt.snd_wscale; |
| 5716 | tcp_init_wl(tp, TCP_SKB_CB(skb)->seq); |
Yuchung Cheng | bc15afa | 2013-10-24 08:44:25 -0700 | [diff] [blame] | 5717 | tcp_synack_rtt_meas(sk, synack_stamp); |
Joe Perches | 61eb9003 | 2013-05-24 18:36:13 +0000 | [diff] [blame] | 5718 | |
| 5719 | if (tp->rx_opt.tstamp_ok) |
| 5720 | tp->advmss -= TCPOLEN_TSTAMP_ALIGNED; |
| 5721 | |
| 5722 | if (req) { |
| 5723 | /* Re-arm the timer because data may have been sent out. |
| 5724 | * This is similar to the regular data transmission case |
| 5725 | * when new data has just been ack'ed. |
| 5726 | * |
| 5727 | * (TFO) - we could try to be more aggressive and |
| 5728 | * retransmitting any data sooner based on when they |
| 5729 | * are sent out. |
| 5730 | */ |
| 5731 | tcp_rearm_rto(sk); |
| 5732 | } else |
| 5733 | tcp_init_metrics(sk); |
| 5734 | |
Neal Cardwell | 02cf4eb | 2013-10-21 15:40:19 -0400 | [diff] [blame] | 5735 | tcp_update_pacing_rate(sk); |
| 5736 | |
Joe Perches | 61eb9003 | 2013-05-24 18:36:13 +0000 | [diff] [blame] | 5737 | /* Prevent spurious tcp_cwnd_restart() on first data packet */ |
| 5738 | tp->lsndtime = tcp_time_stamp; |
| 5739 | |
| 5740 | tcp_initialize_rcv_mss(sk); |
| 5741 | tcp_fast_path_on(tp); |
Eric Dumazet | 1f6afc8 | 2013-05-24 15:03:54 +0000 | [diff] [blame] | 5742 | break; |
| 5743 | |
Joe Perches | c48b22d | 2013-05-24 18:06:58 +0000 | [diff] [blame] | 5744 | case TCP_FIN_WAIT1: { |
| 5745 | struct dst_entry *dst; |
| 5746 | int tmo; |
| 5747 | |
Eric Dumazet | 1f6afc8 | 2013-05-24 15:03:54 +0000 | [diff] [blame] | 5748 | /* If we enter the TCP_FIN_WAIT1 state and we are a |
| 5749 | * Fast Open socket and this is the first acceptable |
| 5750 | * ACK we have received, this would have acknowledged |
| 5751 | * our SYNACK so stop the SYNACK timer. |
| 5752 | */ |
| 5753 | if (req != NULL) { |
| 5754 | /* Return RST if ack_seq is invalid. |
| 5755 | * Note that RFC793 only says to generate a |
| 5756 | * DUPACK for it but for TCP Fast Open it seems |
| 5757 | * better to treat this case like TCP_SYN_RECV |
| 5758 | * above. |
| 5759 | */ |
| 5760 | if (!acceptable) |
| 5761 | return 1; |
| 5762 | /* We no longer need the request sock. */ |
| 5763 | reqsk_fastopen_remove(sk, req, false); |
| 5764 | tcp_rearm_rto(sk); |
| 5765 | } |
Joe Perches | c48b22d | 2013-05-24 18:06:58 +0000 | [diff] [blame] | 5766 | if (tp->snd_una != tp->write_seq) |
| 5767 | break; |
Eric Dumazet | 1f6afc8 | 2013-05-24 15:03:54 +0000 | [diff] [blame] | 5768 | |
Joe Perches | c48b22d | 2013-05-24 18:06:58 +0000 | [diff] [blame] | 5769 | tcp_set_state(sk, TCP_FIN_WAIT2); |
| 5770 | sk->sk_shutdown |= SEND_SHUTDOWN; |
Eric Dumazet | 1f6afc8 | 2013-05-24 15:03:54 +0000 | [diff] [blame] | 5771 | |
Joe Perches | c48b22d | 2013-05-24 18:06:58 +0000 | [diff] [blame] | 5772 | dst = __sk_dst_get(sk); |
| 5773 | if (dst) |
| 5774 | dst_confirm(dst); |
Eric Dumazet | 1f6afc8 | 2013-05-24 15:03:54 +0000 | [diff] [blame] | 5775 | |
Joe Perches | c48b22d | 2013-05-24 18:06:58 +0000 | [diff] [blame] | 5776 | if (!sock_flag(sk, SOCK_DEAD)) { |
| 5777 | /* Wake up lingering close() */ |
| 5778 | sk->sk_state_change(sk); |
| 5779 | break; |
| 5780 | } |
Eric Dumazet | 1f6afc8 | 2013-05-24 15:03:54 +0000 | [diff] [blame] | 5781 | |
Joe Perches | c48b22d | 2013-05-24 18:06:58 +0000 | [diff] [blame] | 5782 | if (tp->linger2 < 0 || |
| 5783 | (TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(skb)->seq && |
| 5784 | after(TCP_SKB_CB(skb)->end_seq - th->fin, tp->rcv_nxt))) { |
| 5785 | tcp_done(sk); |
| 5786 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPABORTONDATA); |
| 5787 | return 1; |
| 5788 | } |
Eric Dumazet | 1f6afc8 | 2013-05-24 15:03:54 +0000 | [diff] [blame] | 5789 | |
Joe Perches | c48b22d | 2013-05-24 18:06:58 +0000 | [diff] [blame] | 5790 | tmo = tcp_fin_time(sk); |
| 5791 | if (tmo > TCP_TIMEWAIT_LEN) { |
| 5792 | inet_csk_reset_keepalive_timer(sk, tmo - TCP_TIMEWAIT_LEN); |
| 5793 | } else if (th->fin || sock_owned_by_user(sk)) { |
| 5794 | /* Bad case. We could lose such FIN otherwise. |
| 5795 | * It is not a big problem, but it looks confusing |
| 5796 | * and not so rare event. We still can lose it now, |
| 5797 | * if it spins in bh_lock_sock(), but it is really |
| 5798 | * marginal case. |
| 5799 | */ |
| 5800 | inet_csk_reset_keepalive_timer(sk, tmo); |
| 5801 | } else { |
| 5802 | tcp_time_wait(sk, TCP_FIN_WAIT2, tmo); |
| 5803 | goto discard; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5804 | } |
Eric Dumazet | 1f6afc8 | 2013-05-24 15:03:54 +0000 | [diff] [blame] | 5805 | break; |
Joe Perches | c48b22d | 2013-05-24 18:06:58 +0000 | [diff] [blame] | 5806 | } |
Eric Dumazet | 1f6afc8 | 2013-05-24 15:03:54 +0000 | [diff] [blame] | 5807 | |
| 5808 | case TCP_CLOSING: |
| 5809 | if (tp->snd_una == tp->write_seq) { |
| 5810 | tcp_time_wait(sk, TCP_TIME_WAIT, 0); |
| 5811 | goto discard; |
| 5812 | } |
| 5813 | break; |
| 5814 | |
| 5815 | case TCP_LAST_ACK: |
| 5816 | if (tp->snd_una == tp->write_seq) { |
| 5817 | tcp_update_metrics(sk); |
| 5818 | tcp_done(sk); |
| 5819 | goto discard; |
| 5820 | } |
| 5821 | break; |
Eric Dumazet | c3ae62af | 2012-12-26 12:44:34 +0000 | [diff] [blame] | 5822 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5823 | |
| 5824 | /* step 6: check the URG bit */ |
| 5825 | tcp_urg(sk, skb, th); |
| 5826 | |
| 5827 | /* step 7: process the segment text */ |
| 5828 | switch (sk->sk_state) { |
| 5829 | case TCP_CLOSE_WAIT: |
| 5830 | case TCP_CLOSING: |
| 5831 | case TCP_LAST_ACK: |
| 5832 | if (!before(TCP_SKB_CB(skb)->seq, tp->rcv_nxt)) |
| 5833 | break; |
| 5834 | case TCP_FIN_WAIT1: |
| 5835 | case TCP_FIN_WAIT2: |
| 5836 | /* RFC 793 says to queue data in these states, |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5837 | * RFC 1122 says we MUST send a reset. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5838 | * BSD 4.4 also does reset. |
| 5839 | */ |
| 5840 | if (sk->sk_shutdown & RCV_SHUTDOWN) { |
| 5841 | if (TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(skb)->seq && |
| 5842 | after(TCP_SKB_CB(skb)->end_seq - th->fin, tp->rcv_nxt)) { |
Pavel Emelyanov | de0744a | 2008-07-16 20:31:16 -0700 | [diff] [blame] | 5843 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPABORTONDATA); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5844 | tcp_reset(sk); |
| 5845 | return 1; |
| 5846 | } |
| 5847 | } |
| 5848 | /* Fall through */ |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5849 | case TCP_ESTABLISHED: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5850 | tcp_data_queue(sk, skb); |
| 5851 | queued = 1; |
| 5852 | break; |
| 5853 | } |
| 5854 | |
| 5855 | /* tcp_data could move socket to TIME-WAIT */ |
| 5856 | if (sk->sk_state != TCP_CLOSE) { |
Ilpo Järvinen | 9e412ba | 2007-04-20 22:18:02 -0700 | [diff] [blame] | 5857 | tcp_data_snd_check(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5858 | tcp_ack_snd_check(sk); |
| 5859 | } |
| 5860 | |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 5861 | if (!queued) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5862 | discard: |
| 5863 | __kfree_skb(skb); |
| 5864 | } |
| 5865 | return 0; |
| 5866 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5867 | EXPORT_SYMBOL(tcp_rcv_state_process); |
Octavian Purdila | 1fb6f15 | 2014-06-25 17:10:02 +0300 | [diff] [blame] | 5868 | |
| 5869 | static inline void pr_drop_req(struct request_sock *req, __u16 port, int family) |
| 5870 | { |
| 5871 | struct inet_request_sock *ireq = inet_rsk(req); |
| 5872 | |
| 5873 | if (family == AF_INET) |
Joe Perches | ba7a46f | 2014-11-11 10:59:17 -0800 | [diff] [blame] | 5874 | net_dbg_ratelimited("drop open request from %pI4/%u\n", |
| 5875 | &ireq->ir_rmt_addr, port); |
Octavian Purdila | 4135ab8 | 2014-06-28 21:20:54 +0300 | [diff] [blame] | 5876 | #if IS_ENABLED(CONFIG_IPV6) |
| 5877 | else if (family == AF_INET6) |
Joe Perches | ba7a46f | 2014-11-11 10:59:17 -0800 | [diff] [blame] | 5878 | net_dbg_ratelimited("drop open request from %pI6/%u\n", |
| 5879 | &ireq->ir_v6_rmt_addr, port); |
Octavian Purdila | 4135ab8 | 2014-06-28 21:20:54 +0300 | [diff] [blame] | 5880 | #endif |
Octavian Purdila | 1fb6f15 | 2014-06-25 17:10:02 +0300 | [diff] [blame] | 5881 | } |
| 5882 | |
Florian Westphal | d82bd12 | 2014-09-29 13:08:29 +0200 | [diff] [blame] | 5883 | /* RFC3168 : 6.1.1 SYN packets must not have ECT/ECN bits set |
| 5884 | * |
| 5885 | * If we receive a SYN packet with these bits set, it means a |
| 5886 | * network is playing bad games with TOS bits. In order to |
| 5887 | * avoid possible false congestion notifications, we disable |
stephen hemminger | f4e715c | 2014-10-29 16:05:06 -0700 | [diff] [blame] | 5888 | * TCP ECN negotiation. |
Florian Westphal | d82bd12 | 2014-09-29 13:08:29 +0200 | [diff] [blame] | 5889 | * |
| 5890 | * Exception: tcp_ca wants ECN. This is required for DCTCP |
Florian Westphal | 843c2fd | 2015-01-30 20:45:20 +0100 | [diff] [blame] | 5891 | * congestion control: Linux DCTCP asserts ECT on all packets, |
| 5892 | * including SYN, which is most optimal solution; however, |
| 5893 | * others, such as FreeBSD do not. |
Florian Westphal | d82bd12 | 2014-09-29 13:08:29 +0200 | [diff] [blame] | 5894 | */ |
| 5895 | static void tcp_ecn_create_request(struct request_sock *req, |
| 5896 | const struct sk_buff *skb, |
Florian Westphal | f7b3bec | 2014-11-03 17:35:03 +0100 | [diff] [blame] | 5897 | const struct sock *listen_sk, |
| 5898 | const struct dst_entry *dst) |
Florian Westphal | d82bd12 | 2014-09-29 13:08:29 +0200 | [diff] [blame] | 5899 | { |
| 5900 | const struct tcphdr *th = tcp_hdr(skb); |
| 5901 | const struct net *net = sock_net(listen_sk); |
| 5902 | bool th_ecn = th->ece && th->cwr; |
Florian Westphal | 843c2fd | 2015-01-30 20:45:20 +0100 | [diff] [blame] | 5903 | bool ect, ecn_ok; |
Florian Westphal | d82bd12 | 2014-09-29 13:08:29 +0200 | [diff] [blame] | 5904 | |
| 5905 | if (!th_ecn) |
| 5906 | return; |
| 5907 | |
| 5908 | ect = !INET_ECN_is_not_ect(TCP_SKB_CB(skb)->ip_dsfield); |
Florian Westphal | f7b3bec | 2014-11-03 17:35:03 +0100 | [diff] [blame] | 5909 | ecn_ok = net->ipv4.sysctl_tcp_ecn || dst_feature(dst, RTAX_FEATURE_ECN); |
Florian Westphal | d82bd12 | 2014-09-29 13:08:29 +0200 | [diff] [blame] | 5910 | |
Florian Westphal | 843c2fd | 2015-01-30 20:45:20 +0100 | [diff] [blame] | 5911 | if ((!ect && ecn_ok) || tcp_ca_needs_ecn(listen_sk)) |
Florian Westphal | d82bd12 | 2014-09-29 13:08:29 +0200 | [diff] [blame] | 5912 | inet_rsk(req)->ecn_ok = 1; |
| 5913 | } |
| 5914 | |
Octavian Purdila | 1fb6f15 | 2014-06-25 17:10:02 +0300 | [diff] [blame] | 5915 | int tcp_conn_request(struct request_sock_ops *rsk_ops, |
| 5916 | const struct tcp_request_sock_ops *af_ops, |
| 5917 | struct sock *sk, struct sk_buff *skb) |
| 5918 | { |
| 5919 | struct tcp_options_received tmp_opt; |
| 5920 | struct request_sock *req; |
| 5921 | struct tcp_sock *tp = tcp_sk(sk); |
| 5922 | struct dst_entry *dst = NULL; |
Eric Dumazet | 04317da | 2014-09-05 15:33:32 -0700 | [diff] [blame] | 5923 | __u32 isn = TCP_SKB_CB(skb)->tcp_tw_isn; |
Octavian Purdila | 1fb6f15 | 2014-06-25 17:10:02 +0300 | [diff] [blame] | 5924 | bool want_cookie = false, fastopen; |
| 5925 | struct flowi fl; |
| 5926 | struct tcp_fastopen_cookie foc = { .len = -1 }; |
| 5927 | int err; |
| 5928 | |
| 5929 | |
| 5930 | /* TW buckets are converted to open requests without |
| 5931 | * limitations, they conserve resources and peer is |
| 5932 | * evidently real one. |
| 5933 | */ |
| 5934 | if ((sysctl_tcp_syncookies == 2 || |
| 5935 | inet_csk_reqsk_queue_is_full(sk)) && !isn) { |
| 5936 | want_cookie = tcp_syn_flood_action(sk, skb, rsk_ops->slab_name); |
| 5937 | if (!want_cookie) |
| 5938 | goto drop; |
| 5939 | } |
| 5940 | |
| 5941 | |
| 5942 | /* Accept backlog is full. If we have already queued enough |
| 5943 | * of warm entries in syn queue, drop request. It is better than |
| 5944 | * clogging syn queue with openreqs with exponentially increasing |
| 5945 | * timeout. |
| 5946 | */ |
| 5947 | if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1) { |
| 5948 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS); |
| 5949 | goto drop; |
| 5950 | } |
| 5951 | |
| 5952 | req = inet_reqsk_alloc(rsk_ops); |
| 5953 | if (!req) |
| 5954 | goto drop; |
| 5955 | |
| 5956 | tcp_rsk(req)->af_specific = af_ops; |
| 5957 | |
| 5958 | tcp_clear_options(&tmp_opt); |
| 5959 | tmp_opt.mss_clamp = af_ops->mss_clamp; |
| 5960 | tmp_opt.user_mss = tp->rx_opt.user_mss; |
| 5961 | tcp_parse_options(skb, &tmp_opt, 0, want_cookie ? NULL : &foc); |
| 5962 | |
| 5963 | if (want_cookie && !tmp_opt.saw_tstamp) |
| 5964 | tcp_clear_options(&tmp_opt); |
| 5965 | |
| 5966 | tmp_opt.tstamp_ok = tmp_opt.saw_tstamp; |
| 5967 | tcp_openreq_init(req, &tmp_opt, skb, sk); |
| 5968 | |
| 5969 | af_ops->init_req(req, sk, skb); |
| 5970 | |
| 5971 | if (security_inet_conn_request(sk, skb, req)) |
| 5972 | goto drop_and_free; |
| 5973 | |
Florian Westphal | f7b3bec | 2014-11-03 17:35:03 +0100 | [diff] [blame] | 5974 | if (!want_cookie && !isn) { |
Octavian Purdila | 1fb6f15 | 2014-06-25 17:10:02 +0300 | [diff] [blame] | 5975 | /* VJ's idea. We save last timestamp seen |
| 5976 | * from the destination in peer table, when entering |
| 5977 | * state TIME-WAIT, and check against it before |
| 5978 | * accepting new connection request. |
| 5979 | * |
| 5980 | * If "isn" is not zero, this request hit alive |
| 5981 | * timewait bucket, so that all the necessary checks |
| 5982 | * are made in the function processing timewait state. |
| 5983 | */ |
Hannes Frederic Sowa | a26552a | 2014-08-14 22:06:12 +0200 | [diff] [blame] | 5984 | if (tcp_death_row.sysctl_tw_recycle) { |
Octavian Purdila | 1fb6f15 | 2014-06-25 17:10:02 +0300 | [diff] [blame] | 5985 | bool strict; |
| 5986 | |
| 5987 | dst = af_ops->route_req(sk, &fl, req, &strict); |
Hannes Frederic Sowa | a26552a | 2014-08-14 22:06:12 +0200 | [diff] [blame] | 5988 | |
Octavian Purdila | 1fb6f15 | 2014-06-25 17:10:02 +0300 | [diff] [blame] | 5989 | if (dst && strict && |
Hannes Frederic Sowa | a26552a | 2014-08-14 22:06:12 +0200 | [diff] [blame] | 5990 | !tcp_peer_is_proven(req, dst, true, |
| 5991 | tmp_opt.saw_tstamp)) { |
Octavian Purdila | 1fb6f15 | 2014-06-25 17:10:02 +0300 | [diff] [blame] | 5992 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSPASSIVEREJECTED); |
| 5993 | goto drop_and_release; |
| 5994 | } |
| 5995 | } |
| 5996 | /* Kill the following clause, if you dislike this way. */ |
| 5997 | else if (!sysctl_tcp_syncookies && |
| 5998 | (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) < |
| 5999 | (sysctl_max_syn_backlog >> 2)) && |
Hannes Frederic Sowa | a26552a | 2014-08-14 22:06:12 +0200 | [diff] [blame] | 6000 | !tcp_peer_is_proven(req, dst, false, |
| 6001 | tmp_opt.saw_tstamp)) { |
Octavian Purdila | 1fb6f15 | 2014-06-25 17:10:02 +0300 | [diff] [blame] | 6002 | /* Without syncookies last quarter of |
| 6003 | * backlog is filled with destinations, |
| 6004 | * proven to be alive. |
| 6005 | * It means that we continue to communicate |
| 6006 | * to destinations, already remembered |
| 6007 | * to the moment of synflood. |
| 6008 | */ |
| 6009 | pr_drop_req(req, ntohs(tcp_hdr(skb)->source), |
| 6010 | rsk_ops->family); |
| 6011 | goto drop_and_release; |
| 6012 | } |
| 6013 | |
| 6014 | isn = af_ops->init_seq(skb); |
| 6015 | } |
| 6016 | if (!dst) { |
| 6017 | dst = af_ops->route_req(sk, &fl, req, NULL); |
| 6018 | if (!dst) |
| 6019 | goto drop_and_free; |
| 6020 | } |
| 6021 | |
Florian Westphal | f7b3bec | 2014-11-03 17:35:03 +0100 | [diff] [blame] | 6022 | tcp_ecn_create_request(req, skb, sk, dst); |
| 6023 | |
| 6024 | if (want_cookie) { |
| 6025 | isn = cookie_init_sequence(af_ops, sk, skb, &req->mss); |
| 6026 | req->cookie_ts = tmp_opt.tstamp_ok; |
| 6027 | if (!tmp_opt.tstamp_ok) |
| 6028 | inet_rsk(req)->ecn_ok = 0; |
| 6029 | } |
| 6030 | |
Octavian Purdila | 1fb6f15 | 2014-06-25 17:10:02 +0300 | [diff] [blame] | 6031 | tcp_rsk(req)->snt_isn = isn; |
| 6032 | tcp_openreq_init_rwin(req, sk, dst); |
| 6033 | fastopen = !want_cookie && |
| 6034 | tcp_try_fastopen(sk, skb, req, &foc, dst); |
| 6035 | err = af_ops->send_synack(sk, dst, &fl, req, |
| 6036 | skb_get_queue_mapping(skb), &foc); |
| 6037 | if (!fastopen) { |
| 6038 | if (err || want_cookie) |
| 6039 | goto drop_and_free; |
| 6040 | |
| 6041 | tcp_rsk(req)->listener = NULL; |
| 6042 | af_ops->queue_hash_add(sk, req, TCP_TIMEOUT_INIT); |
| 6043 | } |
| 6044 | |
| 6045 | return 0; |
| 6046 | |
| 6047 | drop_and_release: |
| 6048 | dst_release(dst); |
| 6049 | drop_and_free: |
| 6050 | reqsk_free(req); |
| 6051 | drop: |
| 6052 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS); |
| 6053 | return 0; |
| 6054 | } |
| 6055 | EXPORT_SYMBOL(tcp_conn_request); |