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 | #include <linux/module.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 22 | #include <linux/gfp.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 23 | #include <net/tcp.h> |
| 24 | |
Andreas Petlund | 36e31b0a | 2010-02-18 02:47:01 +0000 | [diff] [blame] | 25 | int sysctl_tcp_thin_linear_timeouts __read_mostly; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 26 | |
Ravi Joshi | 9252878 | 2013-12-18 16:14:15 -0800 | [diff] [blame] | 27 | static void set_tcp_default(void) |
| 28 | { |
| 29 | sysctl_tcp_delack_seg = TCP_DELACK_SEG; |
| 30 | } |
| 31 | |
| 32 | /*sysctl handler for tcp_ack realted master control */ |
| 33 | int tcp_proc_delayed_ack_control(struct ctl_table *table, int write, |
| 34 | void __user *buffer, size_t *length, |
| 35 | loff_t *ppos) |
| 36 | { |
| 37 | int ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
| 38 | |
| 39 | /* The ret value will be 0 if the input validation is successful |
| 40 | * and the values are written to sysctl table. If not, the stack |
| 41 | * will continue to work with currently configured values |
| 42 | */ |
| 43 | return ret; |
| 44 | } |
| 45 | |
| 46 | /*sysctl handler for tcp_ack realted master control */ |
| 47 | int tcp_use_userconfig_sysctl_handler(struct ctl_table *table, int write, |
| 48 | void __user *buffer, size_t *length, |
| 49 | loff_t *ppos) |
| 50 | { |
| 51 | int ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
| 52 | |
| 53 | if (write && ret == 0) { |
| 54 | if (!sysctl_tcp_use_userconfig) |
| 55 | set_tcp_default(); |
| 56 | } |
| 57 | return ret; |
| 58 | } |
| 59 | |
Richard Sailer | c380d37 | 2016-07-16 04:04:34 +0200 | [diff] [blame] | 60 | /** |
| 61 | * tcp_write_err() - close socket and save error info |
| 62 | * @sk: The socket the error has appeared on. |
| 63 | * |
| 64 | * Returns: Nothing (void) |
| 65 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 66 | static void tcp_write_err(struct sock *sk) |
| 67 | { |
| 68 | sk->sk_err = sk->sk_err_soft ? : ETIMEDOUT; |
| 69 | sk->sk_error_report(sk); |
| 70 | |
| 71 | tcp_done(sk); |
Eric Dumazet | 02a1d6e | 2016-04-27 16:44:39 -0700 | [diff] [blame] | 72 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONTIMEOUT); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 73 | } |
| 74 | |
Richard Sailer | c380d37 | 2016-07-16 04:04:34 +0200 | [diff] [blame] | 75 | /** |
| 76 | * tcp_out_of_resources() - Close socket if out of resources |
| 77 | * @sk: pointer to current socket |
| 78 | * @do_reset: send a last packet with reset flag |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 79 | * |
Richard Sailer | c380d37 | 2016-07-16 04:04:34 +0200 | [diff] [blame] | 80 | * Do not allow orphaned sockets to eat all our resources. |
| 81 | * This is direct violation of TCP specs, but it is required |
| 82 | * to prevent DoS attacks. It is called when a retransmission timeout |
| 83 | * or zero probe timeout occurs on orphaned socket. |
| 84 | * |
Dan Streetman | cf67be7 | 2018-01-18 16:14:26 -0500 | [diff] [blame] | 85 | * Also close if our net namespace is exiting; in that case there is no |
| 86 | * hope of ever communicating again since all netns interfaces are already |
| 87 | * down (or about to be down), and we need to release our dst references, |
| 88 | * which have been moved to the netns loopback interface, so the namespace |
| 89 | * can finish exiting. This condition is only possible if we are a kernel |
| 90 | * socket, as those do not hold references to the namespace. |
| 91 | * |
Richard Sailer | c380d37 | 2016-07-16 04:04:34 +0200 | [diff] [blame] | 92 | * Criteria is still not confirmed experimentally and may change. |
| 93 | * We kill the socket, if: |
| 94 | * 1. If number of orphaned sockets exceeds an administratively configured |
| 95 | * limit. |
| 96 | * 2. If we have strong memory pressure. |
Dan Streetman | cf67be7 | 2018-01-18 16:14:26 -0500 | [diff] [blame] | 97 | * 3. If our net namespace is exiting. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 98 | */ |
Yuchung Cheng | b248230 | 2014-09-29 13:20:38 -0700 | [diff] [blame] | 99 | static int tcp_out_of_resources(struct sock *sk, bool do_reset) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 100 | { |
| 101 | struct tcp_sock *tp = tcp_sk(sk); |
David S. Miller | ad1af0f | 2010-08-25 02:27:49 -0700 | [diff] [blame] | 102 | int shift = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 103 | |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 104 | /* If peer does not open window for long time, or did not transmit |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 105 | * anything for long time, penalize it. */ |
| 106 | if ((s32)(tcp_time_stamp - tp->lsndtime) > 2*TCP_RTO_MAX || !do_reset) |
David S. Miller | ad1af0f | 2010-08-25 02:27:49 -0700 | [diff] [blame] | 107 | shift++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 108 | |
| 109 | /* If some dubious ICMP arrived, penalize even more. */ |
| 110 | if (sk->sk_err_soft) |
David S. Miller | ad1af0f | 2010-08-25 02:27:49 -0700 | [diff] [blame] | 111 | shift++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 112 | |
Arun Sharma | efcdbf2 | 2012-01-30 14:16:06 -0800 | [diff] [blame] | 113 | if (tcp_check_oom(sk, shift)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 114 | /* Catch exceptional cases, when connection requires reset. |
| 115 | * 1. Last segment was sent recently. */ |
| 116 | if ((s32)(tcp_time_stamp - tp->lsndtime) <= TCP_TIMEWAIT_LEN || |
| 117 | /* 2. Window is closed. */ |
| 118 | (!tp->snd_wnd && !tp->packets_out)) |
Yuchung Cheng | b248230 | 2014-09-29 13:20:38 -0700 | [diff] [blame] | 119 | do_reset = true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 120 | if (do_reset) |
| 121 | tcp_send_active_reset(sk, GFP_ATOMIC); |
| 122 | tcp_done(sk); |
Eric Dumazet | 02a1d6e | 2016-04-27 16:44:39 -0700 | [diff] [blame] | 123 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONMEMORY); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 124 | return 1; |
| 125 | } |
Dan Streetman | cf67be7 | 2018-01-18 16:14:26 -0500 | [diff] [blame] | 126 | |
| 127 | if (!check_net(sock_net(sk))) { |
| 128 | /* Not possible to send reset; just close */ |
| 129 | tcp_done(sk); |
| 130 | return 1; |
| 131 | } |
| 132 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 133 | return 0; |
| 134 | } |
| 135 | |
Richard Sailer | c380d37 | 2016-07-16 04:04:34 +0200 | [diff] [blame] | 136 | /** |
| 137 | * tcp_orphan_retries() - Returns maximal number of retries on an orphaned socket |
| 138 | * @sk: Pointer to the current socket. |
| 139 | * @alive: bool, socket alive state |
| 140 | */ |
Richard Sailer | 7533ce3 | 2015-10-09 02:41:37 +0200 | [diff] [blame] | 141 | static int tcp_orphan_retries(struct sock *sk, bool alive) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 142 | { |
Nikolay Borisov | c402d9b | 2016-02-03 09:46:55 +0200 | [diff] [blame] | 143 | int retries = sock_net(sk)->ipv4.sysctl_tcp_orphan_retries; /* May be zero. */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 144 | |
| 145 | /* We know from an ICMP that something is wrong. */ |
| 146 | if (sk->sk_err_soft && !alive) |
| 147 | retries = 0; |
| 148 | |
| 149 | /* However, if socket sent something recently, select some safe |
| 150 | * number of retries. 8 corresponds to >100 seconds with minimal |
| 151 | * RTO of 200msec. */ |
| 152 | if (retries == 0 && alive) |
| 153 | retries = 8; |
| 154 | return retries; |
| 155 | } |
| 156 | |
Eric Dumazet | ce55dd3 | 2007-12-21 01:50:43 -0800 | [diff] [blame] | 157 | static void tcp_mtu_probing(struct inet_connection_sock *icsk, struct sock *sk) |
| 158 | { |
Fan Du | b0f9ca5 | 2015-02-10 09:53:16 +0800 | [diff] [blame] | 159 | struct net *net = sock_net(sk); |
| 160 | |
Eric Dumazet | ce55dd3 | 2007-12-21 01:50:43 -0800 | [diff] [blame] | 161 | /* Black hole detection */ |
Fan Du | b0f9ca5 | 2015-02-10 09:53:16 +0800 | [diff] [blame] | 162 | if (net->ipv4.sysctl_tcp_mtu_probing) { |
Eric Dumazet | ce55dd3 | 2007-12-21 01:50:43 -0800 | [diff] [blame] | 163 | if (!icsk->icsk_mtup.enabled) { |
| 164 | icsk->icsk_mtup.enabled = 1; |
Fan Du | 05cbc0d | 2015-03-06 11:18:24 +0800 | [diff] [blame] | 165 | icsk->icsk_mtup.probe_timestamp = tcp_time_stamp; |
Eric Dumazet | ce55dd3 | 2007-12-21 01:50:43 -0800 | [diff] [blame] | 166 | tcp_sync_mss(sk, icsk->icsk_pmtu_cookie); |
| 167 | } else { |
Fan Du | b0f9ca5 | 2015-02-10 09:53:16 +0800 | [diff] [blame] | 168 | struct net *net = sock_net(sk); |
Eric Dumazet | ce55dd3 | 2007-12-21 01:50:43 -0800 | [diff] [blame] | 169 | struct tcp_sock *tp = tcp_sk(sk); |
David S. Miller | 829942c | 2007-12-21 04:29:16 -0800 | [diff] [blame] | 170 | int mss; |
| 171 | |
Eric Dumazet | 8beb5c5 | 2007-12-21 05:58:29 -0800 | [diff] [blame] | 172 | mss = tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low) >> 1; |
Fan Du | b0f9ca5 | 2015-02-10 09:53:16 +0800 | [diff] [blame] | 173 | mss = min(net->ipv4.sysctl_tcp_base_mss, mss); |
Eric Dumazet | ce55dd3 | 2007-12-21 01:50:43 -0800 | [diff] [blame] | 174 | mss = max(mss, 68 - tp->tcp_header_len); |
Eric Dumazet | 7e90962 | 2019-06-15 17:47:27 -0700 | [diff] [blame] | 175 | mss = max(mss, net->ipv4.sysctl_tcp_min_snd_mss); |
Eric Dumazet | ce55dd3 | 2007-12-21 01:50:43 -0800 | [diff] [blame] | 176 | icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, mss); |
| 177 | tcp_sync_mss(sk, icsk->icsk_pmtu_cookie); |
| 178 | } |
| 179 | } |
| 180 | } |
| 181 | |
Richard Sailer | c380d37 | 2016-07-16 04:04:34 +0200 | [diff] [blame] | 182 | |
| 183 | /** |
| 184 | * retransmits_timed_out() - returns true if this connection has timed out |
| 185 | * @sk: The current socket |
| 186 | * @boundary: max number of retransmissions |
| 187 | * @timeout: A custom timeout value. |
| 188 | * If set to 0 the default timeout is calculated and used. |
| 189 | * Using TCP_RTO_MIN and the number of unsuccessful retransmits. |
| 190 | * @syn_set: true if the SYN Bit was set. |
| 191 | * |
| 192 | * The default "timeout" value this function can calculate and use |
| 193 | * is equivalent to the timeout of a TCP Connection |
| 194 | * after "boundary" unsuccessful, exponentially backed-off |
Damian Lukowski | 4d22f7d | 2010-09-28 13:08:32 -0700 | [diff] [blame] | 195 | * retransmissions with an initial RTO of TCP_RTO_MIN or TCP_TIMEOUT_INIT if |
| 196 | * syn_set flag is set. |
Richard Sailer | c380d37 | 2016-07-16 04:04:34 +0200 | [diff] [blame] | 197 | * |
Damian Lukowski | 2f7de57 | 2009-12-07 06:06:16 +0000 | [diff] [blame] | 198 | */ |
| 199 | static bool retransmits_timed_out(struct sock *sk, |
Jerry Chu | dca43c7 | 2010-08-27 19:13:28 +0000 | [diff] [blame] | 200 | unsigned int boundary, |
David S. Miller | 21a180c | 2010-10-04 11:56:38 -0700 | [diff] [blame] | 201 | unsigned int timeout, |
Damian Lukowski | 4d22f7d | 2010-09-28 13:08:32 -0700 | [diff] [blame] | 202 | bool syn_set) |
Damian Lukowski | 2f7de57 | 2009-12-07 06:06:16 +0000 | [diff] [blame] | 203 | { |
Jerry Chu | dca43c7 | 2010-08-27 19:13:28 +0000 | [diff] [blame] | 204 | unsigned int linear_backoff_thresh, start_ts; |
Damian Lukowski | 4d22f7d | 2010-09-28 13:08:32 -0700 | [diff] [blame] | 205 | unsigned int rto_base = syn_set ? TCP_TIMEOUT_INIT : TCP_RTO_MIN; |
Damian Lukowski | 2f7de57 | 2009-12-07 06:06:16 +0000 | [diff] [blame] | 206 | |
| 207 | if (!inet_csk(sk)->icsk_retransmits) |
| 208 | return false; |
| 209 | |
Eric Dumazet | 7faee5c | 2014-09-05 15:33:33 -0700 | [diff] [blame] | 210 | start_ts = tcp_sk(sk)->retrans_stamp; |
| 211 | if (unlikely(!start_ts)) |
| 212 | start_ts = tcp_skb_timestamp(tcp_write_queue_head(sk)); |
Damian Lukowski | 2f7de57 | 2009-12-07 06:06:16 +0000 | [diff] [blame] | 213 | |
Jerry Chu | dca43c7 | 2010-08-27 19:13:28 +0000 | [diff] [blame] | 214 | if (likely(timeout == 0)) { |
David S. Miller | 21a180c | 2010-10-04 11:56:38 -0700 | [diff] [blame] | 215 | linear_backoff_thresh = ilog2(TCP_RTO_MAX/rto_base); |
Damian Lukowski | 2f7de57 | 2009-12-07 06:06:16 +0000 | [diff] [blame] | 216 | |
Jerry Chu | dca43c7 | 2010-08-27 19:13:28 +0000 | [diff] [blame] | 217 | if (boundary <= linear_backoff_thresh) |
David S. Miller | 21a180c | 2010-10-04 11:56:38 -0700 | [diff] [blame] | 218 | timeout = ((2 << boundary) - 1) * rto_base; |
Jerry Chu | dca43c7 | 2010-08-27 19:13:28 +0000 | [diff] [blame] | 219 | else |
David S. Miller | 21a180c | 2010-10-04 11:56:38 -0700 | [diff] [blame] | 220 | timeout = ((2 << linear_backoff_thresh) - 1) * rto_base + |
Jerry Chu | dca43c7 | 2010-08-27 19:13:28 +0000 | [diff] [blame] | 221 | (boundary - linear_backoff_thresh) * TCP_RTO_MAX; |
| 222 | } |
Damian Lukowski | 2f7de57 | 2009-12-07 06:06:16 +0000 | [diff] [blame] | 223 | return (tcp_time_stamp - start_ts) >= timeout; |
| 224 | } |
| 225 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 226 | /* A write timeout has occurred. Process the after effects. */ |
| 227 | static int tcp_write_timeout(struct sock *sk) |
| 228 | { |
John Heffner | 5d424d5 | 2006-03-20 17:53:41 -0800 | [diff] [blame] | 229 | struct inet_connection_sock *icsk = inet_csk(sk); |
Yuchung Cheng | c968601 | 2013-10-29 10:09:05 -0700 | [diff] [blame] | 230 | struct tcp_sock *tp = tcp_sk(sk); |
Nikolay Borisov | 6fa2516 | 2016-02-03 09:46:49 +0200 | [diff] [blame] | 231 | struct net *net = sock_net(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 232 | int retry_until; |
Rusty Russell | 3db1cd5 | 2011-12-19 13:56:45 +0000 | [diff] [blame] | 233 | bool do_reset, syn_set = false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 234 | |
| 235 | if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { |
Yuchung Cheng | c968601 | 2013-10-29 10:09:05 -0700 | [diff] [blame] | 236 | if (icsk->icsk_retransmits) { |
Eric Dumazet | b6c6712 | 2010-04-08 23:03:29 +0000 | [diff] [blame] | 237 | dst_negative_advice(sk); |
Yuchung Cheng | c968601 | 2013-10-29 10:09:05 -0700 | [diff] [blame] | 238 | if (tp->syn_fastopen || tp->syn_data) |
Daniel Lee | 2646c83 | 2015-04-06 14:37:27 -0700 | [diff] [blame] | 239 | tcp_fastopen_cache_set(sk, 0, NULL, true, 0); |
Yuchung Cheng | dd52bc2 | 2015-11-18 18:17:31 -0800 | [diff] [blame] | 240 | if (tp->syn_data && icsk->icsk_retransmits == 1) |
Eric Dumazet | c10d931 | 2016-04-29 14:16:47 -0700 | [diff] [blame] | 241 | NET_INC_STATS(sock_net(sk), |
| 242 | LINUX_MIB_TCPFASTOPENACTIVEFAIL); |
Lawrence Brakmo | 3acf3ec | 2016-09-27 19:03:37 -0700 | [diff] [blame] | 243 | } else if (!tp->syn_data && !tp->syn_fastopen) { |
| 244 | sk_rethink_txhash(sk); |
Yuchung Cheng | c968601 | 2013-10-29 10:09:05 -0700 | [diff] [blame] | 245 | } |
Nikolay Borisov | 6fa2516 | 2016-02-03 09:46:49 +0200 | [diff] [blame] | 246 | retry_until = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_syn_retries; |
Rusty Russell | 3db1cd5 | 2011-12-19 13:56:45 +0000 | [diff] [blame] | 247 | syn_set = true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 248 | } else { |
Nikolay Borisov | ae5c3f4 | 2016-02-03 09:46:53 +0200 | [diff] [blame] | 249 | if (retransmits_timed_out(sk, net->ipv4.sysctl_tcp_retries1, 0, 0)) { |
Yuchung Cheng | 0e45f4d | 2015-11-18 18:17:30 -0800 | [diff] [blame] | 250 | /* Some middle-boxes may black-hole Fast Open _after_ |
| 251 | * the handshake. Therefore we conservatively disable |
| 252 | * Fast Open on this path on recurring timeouts with |
| 253 | * few or zero bytes acked after Fast Open. |
| 254 | */ |
| 255 | if (tp->syn_data_acked && |
| 256 | tp->bytes_acked <= tp->rx_opt.mss_clamp) { |
| 257 | tcp_fastopen_cache_set(sk, 0, NULL, true, 0); |
Nikolay Borisov | ae5c3f4 | 2016-02-03 09:46:53 +0200 | [diff] [blame] | 258 | if (icsk->icsk_retransmits == net->ipv4.sysctl_tcp_retries1) |
Eric Dumazet | c10d931 | 2016-04-29 14:16:47 -0700 | [diff] [blame] | 259 | NET_INC_STATS(sock_net(sk), |
| 260 | LINUX_MIB_TCPFASTOPENACTIVEFAIL); |
Yuchung Cheng | 0e45f4d | 2015-11-18 18:17:30 -0800 | [diff] [blame] | 261 | } |
John Heffner | 5d424d5 | 2006-03-20 17:53:41 -0800 | [diff] [blame] | 262 | /* Black hole detection */ |
Eric Dumazet | ce55dd3 | 2007-12-21 01:50:43 -0800 | [diff] [blame] | 263 | tcp_mtu_probing(icsk, sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 264 | |
Eric Dumazet | b6c6712 | 2010-04-08 23:03:29 +0000 | [diff] [blame] | 265 | dst_negative_advice(sk); |
Lawrence Brakmo | 3acf3ec | 2016-09-27 19:03:37 -0700 | [diff] [blame] | 266 | } else { |
| 267 | sk_rethink_txhash(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 268 | } |
| 269 | |
Nikolay Borisov | c6214a9 | 2016-02-03 09:46:54 +0200 | [diff] [blame] | 270 | retry_until = net->ipv4.sysctl_tcp_retries2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 271 | if (sock_flag(sk, SOCK_DEAD)) { |
Richard Sailer | 7533ce3 | 2015-10-09 02:41:37 +0200 | [diff] [blame] | 272 | const bool alive = icsk->icsk_rto < TCP_RTO_MAX; |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 273 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 274 | retry_until = tcp_orphan_retries(sk, alive); |
Damian Lukowski | 6fa12c8 | 2009-08-26 00:16:34 +0000 | [diff] [blame] | 275 | do_reset = alive || |
David S. Miller | 21a180c | 2010-10-04 11:56:38 -0700 | [diff] [blame] | 276 | !retransmits_timed_out(sk, retry_until, 0, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 277 | |
Damian Lukowski | 6fa12c8 | 2009-08-26 00:16:34 +0000 | [diff] [blame] | 278 | if (tcp_out_of_resources(sk, do_reset)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 279 | return 1; |
| 280 | } |
| 281 | } |
| 282 | |
Jerry Chu | dca43c7 | 2010-08-27 19:13:28 +0000 | [diff] [blame] | 283 | if (retransmits_timed_out(sk, retry_until, |
David S. Miller | 21a180c | 2010-10-04 11:56:38 -0700 | [diff] [blame] | 284 | syn_set ? 0 : icsk->icsk_user_timeout, syn_set)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 285 | /* Has it gone just too far? */ |
| 286 | tcp_write_err(sk); |
| 287 | return 1; |
| 288 | } |
| 289 | return 0; |
| 290 | } |
| 291 | |
Eric Dumazet | c10d931 | 2016-04-29 14:16:47 -0700 | [diff] [blame] | 292 | /* Called with BH disabled */ |
Eric Dumazet | 6f458df | 2012-07-20 05:45:50 +0000 | [diff] [blame] | 293 | void tcp_delack_timer_handler(struct sock *sk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 294 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 295 | struct tcp_sock *tp = tcp_sk(sk); |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 296 | struct inet_connection_sock *icsk = inet_csk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 297 | |
David S. Miller | 9993e7d | 2008-01-10 21:56:38 -0800 | [diff] [blame] | 298 | sk_mem_reclaim_partial(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 299 | |
Eric Dumazet | 07753bc | 2017-03-03 14:08:21 -0800 | [diff] [blame] | 300 | if (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)) || |
| 301 | !(icsk->icsk_ack.pending & ICSK_ACK_TIMER)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 302 | goto out; |
| 303 | |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 304 | if (time_after(icsk->icsk_ack.timeout, jiffies)) { |
| 305 | sk_reset_timer(sk, &icsk->icsk_delack_timer, icsk->icsk_ack.timeout); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 306 | goto out; |
| 307 | } |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 308 | icsk->icsk_ack.pending &= ~ICSK_ACK_TIMER; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 309 | |
David S. Miller | b03efcf | 2005-07-08 14:57:23 -0700 | [diff] [blame] | 310 | if (!skb_queue_empty(&tp->ucopy.prequeue)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 311 | struct sk_buff *skb; |
| 312 | |
Eric Dumazet | 02a1d6e | 2016-04-27 16:44:39 -0700 | [diff] [blame] | 313 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPSCHEDULERFAILED); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 314 | |
| 315 | while ((skb = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) |
Peter Zijlstra | c57943a | 2008-10-07 14:18:42 -0700 | [diff] [blame] | 316 | sk_backlog_rcv(sk, skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 317 | |
| 318 | tp->ucopy.memory = 0; |
| 319 | } |
| 320 | |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 321 | if (inet_csk_ack_scheduled(sk)) { |
| 322 | if (!icsk->icsk_ack.pingpong) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 323 | /* Delayed ACK missed: inflate ATO. */ |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 324 | icsk->icsk_ack.ato = min(icsk->icsk_ack.ato << 1, icsk->icsk_rto); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 325 | } else { |
| 326 | /* Delayed ACK missed: leave pingpong mode and |
| 327 | * deflate ATO. |
| 328 | */ |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 329 | icsk->icsk_ack.pingpong = 0; |
| 330 | icsk->icsk_ack.ato = TCP_ATO_MIN; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 331 | } |
| 332 | tcp_send_ack(sk); |
Eric Dumazet | 02a1d6e | 2016-04-27 16:44:39 -0700 | [diff] [blame] | 333 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_DELAYEDACKS); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 334 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 335 | |
| 336 | out: |
Eric Dumazet | b8da51e | 2015-05-15 12:39:27 -0700 | [diff] [blame] | 337 | if (tcp_under_memory_pressure(sk)) |
Hideo Aoki | 3ab224b | 2007-12-31 00:11:19 -0800 | [diff] [blame] | 338 | sk_mem_reclaim(sk); |
Eric Dumazet | 6f458df | 2012-07-20 05:45:50 +0000 | [diff] [blame] | 339 | } |
| 340 | |
Richard Sailer | c380d37 | 2016-07-16 04:04:34 +0200 | [diff] [blame] | 341 | |
| 342 | /** |
| 343 | * tcp_delack_timer() - The TCP delayed ACK timeout handler |
| 344 | * @data: Pointer to the current socket. (gets casted to struct sock *) |
| 345 | * |
| 346 | * This function gets (indirectly) called when the kernel timer for a TCP packet |
| 347 | * of this socket expires. Calls tcp_delack_timer_handler() to do the actual work. |
| 348 | * |
| 349 | * Returns: Nothing (void) |
| 350 | */ |
Eric Dumazet | 6f458df | 2012-07-20 05:45:50 +0000 | [diff] [blame] | 351 | static void tcp_delack_timer(unsigned long data) |
| 352 | { |
| 353 | struct sock *sk = (struct sock *)data; |
| 354 | |
| 355 | bh_lock_sock(sk); |
| 356 | if (!sock_owned_by_user(sk)) { |
| 357 | tcp_delack_timer_handler(sk); |
| 358 | } else { |
| 359 | inet_csk(sk)->icsk_ack.blocked = 1; |
Eric Dumazet | 02a1d6e | 2016-04-27 16:44:39 -0700 | [diff] [blame] | 360 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_DELAYEDACKLOCKED); |
Eric Dumazet | 6f458df | 2012-07-20 05:45:50 +0000 | [diff] [blame] | 361 | /* deleguate our work to tcp_release_cb() */ |
Eric Dumazet | 144d56e | 2012-08-20 00:22:46 +0000 | [diff] [blame] | 362 | if (!test_and_set_bit(TCP_DELACK_TIMER_DEFERRED, &tcp_sk(sk)->tsq_flags)) |
| 363 | sock_hold(sk); |
Eric Dumazet | 6f458df | 2012-07-20 05:45:50 +0000 | [diff] [blame] | 364 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 365 | bh_unlock_sock(sk); |
| 366 | sock_put(sk); |
| 367 | } |
| 368 | |
| 369 | static void tcp_probe_timer(struct sock *sk) |
| 370 | { |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 371 | struct inet_connection_sock *icsk = inet_csk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 372 | struct tcp_sock *tp = tcp_sk(sk); |
| 373 | int max_probes; |
Yuchung Cheng | b248230 | 2014-09-29 13:20:38 -0700 | [diff] [blame] | 374 | u32 start_ts; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 375 | |
David S. Miller | fe067e8 | 2007-03-07 12:12:44 -0800 | [diff] [blame] | 376 | if (tp->packets_out || !tcp_send_head(sk)) { |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 377 | icsk->icsk_probes_out = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 378 | return; |
| 379 | } |
| 380 | |
Yuchung Cheng | b248230 | 2014-09-29 13:20:38 -0700 | [diff] [blame] | 381 | /* RFC 1122 4.2.2.17 requires the sender to stay open indefinitely as |
| 382 | * long as the receiver continues to respond probes. We support this by |
| 383 | * default and reset icsk_probes_out with incoming ACKs. But if the |
| 384 | * socket is orphaned or the user specifies TCP_USER_TIMEOUT, we |
| 385 | * kill the socket when the retry count and the time exceeds the |
| 386 | * corresponding system limit. We also implement similar policy when |
| 387 | * we use RTO to probe window in tcp_retransmit_timer(). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 388 | */ |
Yuchung Cheng | b248230 | 2014-09-29 13:20:38 -0700 | [diff] [blame] | 389 | start_ts = tcp_skb_timestamp(tcp_send_head(sk)); |
| 390 | if (!start_ts) |
| 391 | skb_mstamp_get(&tcp_send_head(sk)->skb_mstamp); |
| 392 | else if (icsk->icsk_user_timeout && |
| 393 | (s32)(tcp_time_stamp - start_ts) > icsk->icsk_user_timeout) |
| 394 | goto abort; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 395 | |
Nikolay Borisov | c6214a9 | 2016-02-03 09:46:54 +0200 | [diff] [blame] | 396 | max_probes = sock_net(sk)->ipv4.sysctl_tcp_retries2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 397 | if (sock_flag(sk, SOCK_DEAD)) { |
Richard Sailer | 7533ce3 | 2015-10-09 02:41:37 +0200 | [diff] [blame] | 398 | const bool alive = inet_csk_rto_backoff(icsk, TCP_RTO_MAX) < TCP_RTO_MAX; |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 399 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 400 | max_probes = tcp_orphan_retries(sk, alive); |
Yuchung Cheng | b248230 | 2014-09-29 13:20:38 -0700 | [diff] [blame] | 401 | if (!alive && icsk->icsk_backoff >= max_probes) |
| 402 | goto abort; |
| 403 | if (tcp_out_of_resources(sk, true)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 404 | return; |
| 405 | } |
| 406 | |
Yuchung Cheng | 4ae7b6e | 2018-11-28 16:06:43 -0800 | [diff] [blame] | 407 | if (icsk->icsk_probes_out >= max_probes) { |
Yuchung Cheng | b248230 | 2014-09-29 13:20:38 -0700 | [diff] [blame] | 408 | abort: tcp_write_err(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 409 | } else { |
| 410 | /* Only send another probe if we didn't close things up. */ |
| 411 | tcp_send_probe0(sk); |
| 412 | } |
| 413 | } |
| 414 | |
| 415 | /* |
Jerry Chu | 8336886 | 2012-08-31 12:29:12 +0000 | [diff] [blame] | 416 | * Timer for Fast Open socket to retransmit SYNACK. Note that the |
| 417 | * sk here is the child socket, not the parent (listener) socket. |
| 418 | */ |
| 419 | static void tcp_fastopen_synack_timer(struct sock *sk) |
| 420 | { |
| 421 | struct inet_connection_sock *icsk = inet_csk(sk); |
| 422 | int max_retries = icsk->icsk_syn_retries ? : |
Nikolay Borisov | 7c083ec | 2016-02-03 09:46:50 +0200 | [diff] [blame] | 423 | sock_net(sk)->ipv4.sysctl_tcp_synack_retries + 1; /* add one more retry for fastopen */ |
Jerry Chu | 8336886 | 2012-08-31 12:29:12 +0000 | [diff] [blame] | 424 | struct request_sock *req; |
| 425 | |
| 426 | req = tcp_sk(sk)->fastopen_rsk; |
Eric Dumazet | 42cb80a | 2015-03-22 10:22:19 -0700 | [diff] [blame] | 427 | req->rsk_ops->syn_ack_timeout(req); |
Jerry Chu | 8336886 | 2012-08-31 12:29:12 +0000 | [diff] [blame] | 428 | |
Eric Dumazet | e6c022a | 2012-10-27 23:16:46 +0000 | [diff] [blame] | 429 | if (req->num_timeout >= max_retries) { |
Jerry Chu | 8336886 | 2012-08-31 12:29:12 +0000 | [diff] [blame] | 430 | tcp_write_err(sk); |
| 431 | return; |
| 432 | } |
| 433 | /* XXX (TFO) - Unlike regular SYN-ACK retransmit, we ignore error |
| 434 | * returned from rtx_syn_ack() to make it more persistent like |
| 435 | * regular retransmit because if the child socket has been accepted |
| 436 | * it's not good to give up too easily. |
| 437 | */ |
Eric Dumazet | e6c022a | 2012-10-27 23:16:46 +0000 | [diff] [blame] | 438 | inet_rtx_syn_ack(sk, req); |
| 439 | req->num_timeout++; |
Yuchung Cheng | 7e32b44 | 2016-09-21 16:16:15 -0700 | [diff] [blame] | 440 | icsk->icsk_retransmits++; |
Jerry Chu | 8336886 | 2012-08-31 12:29:12 +0000 | [diff] [blame] | 441 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, |
Eric Dumazet | e6c022a | 2012-10-27 23:16:46 +0000 | [diff] [blame] | 442 | TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX); |
Jerry Chu | 8336886 | 2012-08-31 12:29:12 +0000 | [diff] [blame] | 443 | } |
| 444 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 445 | |
Richard Sailer | c380d37 | 2016-07-16 04:04:34 +0200 | [diff] [blame] | 446 | /** |
| 447 | * tcp_retransmit_timer() - The TCP retransmit timeout handler |
| 448 | * @sk: Pointer to the current socket. |
| 449 | * |
| 450 | * This function gets called when the kernel timer for a TCP packet |
| 451 | * of this socket expires. |
| 452 | * |
| 453 | * It handles retransmission, timer adjustment and other necesarry measures. |
| 454 | * |
| 455 | * Returns: Nothing (void) |
| 456 | */ |
Damian Lukowski | f1ecd5d | 2009-08-26 00:16:31 +0000 | [diff] [blame] | 457 | void tcp_retransmit_timer(struct sock *sk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 458 | { |
| 459 | struct tcp_sock *tp = tcp_sk(sk); |
Nikolay Borisov | ae5c3f4 | 2016-02-03 09:46:53 +0200 | [diff] [blame] | 460 | struct net *net = sock_net(sk); |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 461 | struct inet_connection_sock *icsk = inet_csk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 462 | |
Jerry Chu | 8336886 | 2012-08-31 12:29:12 +0000 | [diff] [blame] | 463 | if (tp->fastopen_rsk) { |
Jerry Chu | 37561f6 | 2012-10-22 11:26:36 +0000 | [diff] [blame] | 464 | WARN_ON_ONCE(sk->sk_state != TCP_SYN_RECV && |
| 465 | sk->sk_state != TCP_FIN_WAIT1); |
Jerry Chu | 8336886 | 2012-08-31 12:29:12 +0000 | [diff] [blame] | 466 | tcp_fastopen_synack_timer(sk); |
| 467 | /* Before we receive ACK to our SYN-ACK don't retransmit |
| 468 | * anything else (e.g., data or FIN segments). |
| 469 | */ |
| 470 | return; |
| 471 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 472 | if (!tp->packets_out) |
| 473 | goto out; |
| 474 | |
Ilpo Järvinen | 547b792 | 2008-07-25 21:43:18 -0700 | [diff] [blame] | 475 | WARN_ON(tcp_write_queue_empty(sk)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 476 | |
Nandita Dukkipati | 9b717a8 | 2013-03-11 10:00:44 +0000 | [diff] [blame] | 477 | tp->tlp_high_seq = 0; |
| 478 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 479 | if (!tp->snd_wnd && !sock_flag(sk, SOCK_DEAD) && |
| 480 | !((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))) { |
| 481 | /* Receiver dastardly shrinks window. Our retransmits |
| 482 | * become zero probes, but we should not timeout this |
| 483 | * connection. If the socket is an orphan, time it out, |
| 484 | * we cannot allow such beasts to hang infinitely. |
| 485 | */ |
YOSHIFUJI Hideaki | 569508c | 2008-04-14 04:09:36 -0700 | [diff] [blame] | 486 | struct inet_sock *inet = inet_sk(sk); |
| 487 | if (sk->sk_family == AF_INET) { |
Joe Perches | ba7a46f | 2014-11-11 10:59:17 -0800 | [diff] [blame] | 488 | net_dbg_ratelimited("Peer %pI4:%u/%u unexpectedly shrunk window %u:%u (repaired)\n", |
| 489 | &inet->inet_daddr, |
| 490 | ntohs(inet->inet_dport), |
| 491 | inet->inet_num, |
| 492 | tp->snd_una, tp->snd_nxt); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 493 | } |
Eric Dumazet | dfd56b8 | 2011-12-10 09:48:31 +0000 | [diff] [blame] | 494 | #if IS_ENABLED(CONFIG_IPV6) |
YOSHIFUJI Hideaki | 569508c | 2008-04-14 04:09:36 -0700 | [diff] [blame] | 495 | else if (sk->sk_family == AF_INET6) { |
Joe Perches | ba7a46f | 2014-11-11 10:59:17 -0800 | [diff] [blame] | 496 | net_dbg_ratelimited("Peer %pI6:%u/%u unexpectedly shrunk window %u:%u (repaired)\n", |
| 497 | &sk->sk_v6_daddr, |
| 498 | ntohs(inet->inet_dport), |
| 499 | inet->inet_num, |
| 500 | tp->snd_una, tp->snd_nxt); |
YOSHIFUJI Hideaki | 569508c | 2008-04-14 04:09:36 -0700 | [diff] [blame] | 501 | } |
| 502 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 503 | if (tcp_time_stamp - tp->rcv_tstamp > TCP_RTO_MAX) { |
| 504 | tcp_write_err(sk); |
| 505 | goto out; |
| 506 | } |
Neal Cardwell | 5ae344c | 2014-08-04 19:12:29 -0400 | [diff] [blame] | 507 | tcp_enter_loss(sk); |
Eric Dumazet | 10d3be5 | 2016-04-21 10:55:23 -0700 | [diff] [blame] | 508 | tcp_retransmit_skb(sk, tcp_write_queue_head(sk), 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 509 | __sk_dst_reset(sk); |
| 510 | goto out_reset_timer; |
| 511 | } |
| 512 | |
Yuchung Cheng | 5471717 | 2018-11-28 16:06:45 -0800 | [diff] [blame] | 513 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPTIMEOUTS); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 514 | if (tcp_write_timeout(sk)) |
| 515 | goto out; |
| 516 | |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 517 | if (icsk->icsk_retransmits == 0) { |
Yuchung Cheng | 5471717 | 2018-11-28 16:06:45 -0800 | [diff] [blame] | 518 | int mib_idx = 0; |
Pavel Emelyanov | 40b215e | 2008-07-03 01:05:41 -0700 | [diff] [blame] | 519 | |
Ilpo Järvinen | c60ce4e | 2010-10-14 01:52:09 +0000 | [diff] [blame] | 520 | if (icsk->icsk_ca_state == TCP_CA_Recovery) { |
Ilpo Järvinen | bc079e9 | 2009-02-28 04:44:34 +0000 | [diff] [blame] | 521 | if (tcp_is_sack(tp)) |
| 522 | mib_idx = LINUX_MIB_TCPSACKRECOVERYFAIL; |
| 523 | else |
| 524 | mib_idx = LINUX_MIB_TCPRENORECOVERYFAIL; |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 525 | } else if (icsk->icsk_ca_state == TCP_CA_Loss) { |
Pavel Emelyanov | 40b215e | 2008-07-03 01:05:41 -0700 | [diff] [blame] | 526 | mib_idx = LINUX_MIB_TCPLOSSFAILURES; |
Ilpo Järvinen | c60ce4e | 2010-10-14 01:52:09 +0000 | [diff] [blame] | 527 | } else if ((icsk->icsk_ca_state == TCP_CA_Disorder) || |
| 528 | tp->sacked_out) { |
| 529 | if (tcp_is_sack(tp)) |
| 530 | mib_idx = LINUX_MIB_TCPSACKFAILURES; |
| 531 | else |
| 532 | mib_idx = LINUX_MIB_TCPRENOFAILURES; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 533 | } |
Yuchung Cheng | 5471717 | 2018-11-28 16:06:45 -0800 | [diff] [blame] | 534 | if (mib_idx) |
| 535 | __NET_INC_STATS(sock_net(sk), mib_idx); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 536 | } |
| 537 | |
Neal Cardwell | 5ae344c | 2014-08-04 19:12:29 -0400 | [diff] [blame] | 538 | tcp_enter_loss(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 539 | |
Yuchung Cheng | 5a95e5a | 2019-01-16 15:05:34 -0800 | [diff] [blame] | 540 | icsk->icsk_retransmits++; |
Eric Dumazet | 10d3be5 | 2016-04-21 10:55:23 -0700 | [diff] [blame] | 541 | if (tcp_retransmit_skb(sk, tcp_write_queue_head(sk), 1) > 0) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 542 | /* Retransmission failed because of local congestion, |
Yuchung Cheng | 5a95e5a | 2019-01-16 15:05:34 -0800 | [diff] [blame] | 543 | * Let senders fight for local resources conservatively. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 544 | */ |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 545 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, |
Yuchung Cheng | 5a95e5a | 2019-01-16 15:05:34 -0800 | [diff] [blame] | 546 | TCP_RESOURCE_PROBE_INTERVAL, |
Arnaldo Carvalho de Melo | 3f421ba | 2005-08-09 20:11:08 -0700 | [diff] [blame] | 547 | TCP_RTO_MAX); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 548 | goto out; |
| 549 | } |
| 550 | |
| 551 | /* Increase the timeout each time we retransmit. Note that |
| 552 | * we do not increase the rtt estimate. rto is initialized |
| 553 | * from rtt, but increases here. Jacobson (SIGCOMM 88) suggests |
| 554 | * that doubling rto each time is the least we can get away with. |
| 555 | * In KA9Q, Karn uses this for the first few times, and then |
| 556 | * goes to quadratic. netBSD doubles, but only goes up to *64, |
| 557 | * and clamps at 1 to 64 sec afterwards. Note that 120 sec is |
| 558 | * defined in the protocol as the maximum possible RTT. I guess |
| 559 | * we'll have to use something other than TCP to talk to the |
| 560 | * University of Mars. |
| 561 | * |
| 562 | * PAWS allows us longer timeouts and large windows, so once |
| 563 | * implemented ftp to mars will work nicely. We will have to fix |
| 564 | * the 120 second clamps though! |
| 565 | */ |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 566 | icsk->icsk_backoff++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 567 | |
| 568 | out_reset_timer: |
Andreas Petlund | 36e31b0a | 2010-02-18 02:47:01 +0000 | [diff] [blame] | 569 | /* If stream is thin, use linear timeouts. Since 'icsk_backoff' is |
| 570 | * used to reset timer, set to 0. Recalculate 'icsk_rto' as this |
| 571 | * might be increased if the stream oscillates between thin and thick, |
| 572 | * thus the old value might already be too high compared to the value |
| 573 | * set by 'tcp_set_rto' in tcp_input.c which resets the rto without |
| 574 | * backoff. Limit to TCP_THIN_LINEAR_RETRIES before initiating |
| 575 | * exponential backoff behaviour to avoid continue hammering |
| 576 | * linear-timeout retransmissions into a black hole |
| 577 | */ |
| 578 | if (sk->sk_state == TCP_ESTABLISHED && |
| 579 | (tp->thin_lto || sysctl_tcp_thin_linear_timeouts) && |
| 580 | tcp_stream_is_thin(tp) && |
| 581 | icsk->icsk_retransmits <= TCP_THIN_LINEAR_RETRIES) { |
| 582 | icsk->icsk_backoff = 0; |
| 583 | icsk->icsk_rto = min(__tcp_set_rto(tp), TCP_RTO_MAX); |
| 584 | } else { |
| 585 | /* Use normal (exponential) backoff */ |
| 586 | icsk->icsk_rto = min(icsk->icsk_rto << 1, TCP_RTO_MAX); |
| 587 | } |
Arnaldo Carvalho de Melo | 3f421ba | 2005-08-09 20:11:08 -0700 | [diff] [blame] | 588 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, icsk->icsk_rto, TCP_RTO_MAX); |
Nikolay Borisov | ae5c3f4 | 2016-02-03 09:46:53 +0200 | [diff] [blame] | 589 | if (retransmits_timed_out(sk, net->ipv4.sysctl_tcp_retries1 + 1, 0, 0)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 590 | __sk_dst_reset(sk); |
| 591 | |
| 592 | out:; |
| 593 | } |
| 594 | |
Richard Sailer | c380d37 | 2016-07-16 04:04:34 +0200 | [diff] [blame] | 595 | /* Called with bottom-half processing disabled. |
| 596 | Called by tcp_write_timer() */ |
Eric Dumazet | 6f458df | 2012-07-20 05:45:50 +0000 | [diff] [blame] | 597 | void tcp_write_timer_handler(struct sock *sk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 598 | { |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 599 | struct inet_connection_sock *icsk = inet_csk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 600 | int event; |
| 601 | |
Eric Dumazet | 07753bc | 2017-03-03 14:08:21 -0800 | [diff] [blame] | 602 | if (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)) || |
| 603 | !icsk->icsk_pending) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 604 | goto out; |
| 605 | |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 606 | if (time_after(icsk->icsk_timeout, jiffies)) { |
| 607 | sk_reset_timer(sk, &icsk->icsk_retransmit_timer, icsk->icsk_timeout); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 608 | goto out; |
| 609 | } |
| 610 | |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 611 | event = icsk->icsk_pending; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 612 | |
| 613 | switch (event) { |
Nandita Dukkipati | 6ba8a3b | 2013-03-11 10:00:43 +0000 | [diff] [blame] | 614 | case ICSK_TIME_EARLY_RETRANS: |
| 615 | tcp_resume_early_retransmit(sk); |
| 616 | break; |
| 617 | case ICSK_TIME_LOSS_PROBE: |
| 618 | tcp_send_loss_probe(sk); |
| 619 | break; |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 620 | case ICSK_TIME_RETRANS: |
Nandita Dukkipati | 6ba8a3b | 2013-03-11 10:00:43 +0000 | [diff] [blame] | 621 | icsk->icsk_pending = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 622 | tcp_retransmit_timer(sk); |
| 623 | break; |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 624 | case ICSK_TIME_PROBE0: |
Nandita Dukkipati | 6ba8a3b | 2013-03-11 10:00:43 +0000 | [diff] [blame] | 625 | icsk->icsk_pending = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 626 | tcp_probe_timer(sk); |
| 627 | break; |
| 628 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 629 | |
| 630 | out: |
Hideo Aoki | 3ab224b | 2007-12-31 00:11:19 -0800 | [diff] [blame] | 631 | sk_mem_reclaim(sk); |
Eric Dumazet | 6f458df | 2012-07-20 05:45:50 +0000 | [diff] [blame] | 632 | } |
| 633 | |
| 634 | static void tcp_write_timer(unsigned long data) |
| 635 | { |
| 636 | struct sock *sk = (struct sock *)data; |
| 637 | |
| 638 | bh_lock_sock(sk); |
| 639 | if (!sock_owned_by_user(sk)) { |
| 640 | tcp_write_timer_handler(sk); |
| 641 | } else { |
Richard Sailer | c380d37 | 2016-07-16 04:04:34 +0200 | [diff] [blame] | 642 | /* delegate our work to tcp_release_cb() */ |
Eric Dumazet | 144d56e | 2012-08-20 00:22:46 +0000 | [diff] [blame] | 643 | if (!test_and_set_bit(TCP_WRITE_TIMER_DEFERRED, &tcp_sk(sk)->tsq_flags)) |
| 644 | sock_hold(sk); |
Eric Dumazet | 6f458df | 2012-07-20 05:45:50 +0000 | [diff] [blame] | 645 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 646 | bh_unlock_sock(sk); |
| 647 | sock_put(sk); |
| 648 | } |
| 649 | |
Eric Dumazet | 42cb80a | 2015-03-22 10:22:19 -0700 | [diff] [blame] | 650 | void tcp_syn_ack_timeout(const struct request_sock *req) |
Octavian Purdila | 72659ec | 2010-01-17 19:09:39 -0800 | [diff] [blame] | 651 | { |
Eric Dumazet | 42cb80a | 2015-03-22 10:22:19 -0700 | [diff] [blame] | 652 | struct net *net = read_pnet(&inet_rsk(req)->ireq_net); |
| 653 | |
Eric Dumazet | 02a1d6e | 2016-04-27 16:44:39 -0700 | [diff] [blame] | 654 | __NET_INC_STATS(net, LINUX_MIB_TCPTIMEOUTS); |
Octavian Purdila | 72659ec | 2010-01-17 19:09:39 -0800 | [diff] [blame] | 655 | } |
| 656 | EXPORT_SYMBOL(tcp_syn_ack_timeout); |
| 657 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 658 | void tcp_set_keepalive(struct sock *sk, int val) |
| 659 | { |
| 660 | if ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)) |
| 661 | return; |
| 662 | |
| 663 | if (val && !sock_flag(sk, SOCK_KEEPOPEN)) |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 664 | inet_csk_reset_keepalive_timer(sk, keepalive_time_when(tcp_sk(sk))); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 665 | else if (!val) |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 666 | inet_csk_delete_keepalive_timer(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 667 | } |
| 668 | |
| 669 | |
| 670 | static void tcp_keepalive_timer (unsigned long data) |
| 671 | { |
| 672 | struct sock *sk = (struct sock *) data; |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 673 | struct inet_connection_sock *icsk = inet_csk(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 674 | struct tcp_sock *tp = tcp_sk(sk); |
Flavio Leitner | 6c37e5d | 2010-04-26 18:33:27 +0000 | [diff] [blame] | 675 | u32 elapsed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 676 | |
| 677 | /* Only process if socket is not in use. */ |
| 678 | bh_lock_sock(sk); |
| 679 | if (sock_owned_by_user(sk)) { |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 680 | /* Try again later. */ |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 681 | inet_csk_reset_keepalive_timer (sk, HZ/20); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 682 | goto out; |
| 683 | } |
| 684 | |
| 685 | if (sk->sk_state == TCP_LISTEN) { |
Eric Dumazet | fa76ce73 | 2015-03-19 19:04:20 -0700 | [diff] [blame] | 686 | pr_err("Hmm... keepalive on a LISTEN ???\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 687 | goto out; |
| 688 | } |
| 689 | |
| 690 | if (sk->sk_state == TCP_FIN_WAIT2 && sock_flag(sk, SOCK_DEAD)) { |
| 691 | if (tp->linger2 >= 0) { |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 692 | const int tmo = tcp_fin_time(sk) - TCP_TIMEWAIT_LEN; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 693 | |
| 694 | if (tmo > 0) { |
| 695 | tcp_time_wait(sk, TCP_FIN_WAIT2, tmo); |
| 696 | goto out; |
| 697 | } |
| 698 | } |
| 699 | tcp_send_active_reset(sk, GFP_ATOMIC); |
| 700 | goto death; |
| 701 | } |
| 702 | |
Eric Dumazet | 05046af | 2017-08-02 23:10:46 -0700 | [diff] [blame] | 703 | if (!sock_flag(sk, SOCK_KEEPOPEN) || |
| 704 | ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 705 | goto out; |
| 706 | |
| 707 | elapsed = keepalive_time_when(tp); |
| 708 | |
| 709 | /* It is alive without keepalive 8) */ |
David S. Miller | fe067e8 | 2007-03-07 12:12:44 -0800 | [diff] [blame] | 710 | if (tp->packets_out || tcp_send_head(sk)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 711 | goto resched; |
| 712 | |
Flavio Leitner | 6c37e5d | 2010-04-26 18:33:27 +0000 | [diff] [blame] | 713 | elapsed = keepalive_time_elapsed(tp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 714 | |
| 715 | if (elapsed >= keepalive_time_when(tp)) { |
Jerry Chu | dca43c7 | 2010-08-27 19:13:28 +0000 | [diff] [blame] | 716 | /* If the TCP_USER_TIMEOUT option is enabled, use that |
| 717 | * to determine when to timeout instead. |
| 718 | */ |
| 719 | if ((icsk->icsk_user_timeout != 0 && |
| 720 | elapsed >= icsk->icsk_user_timeout && |
| 721 | icsk->icsk_probes_out > 0) || |
| 722 | (icsk->icsk_user_timeout == 0 && |
| 723 | icsk->icsk_probes_out >= keepalive_probes(tp))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 724 | tcp_send_active_reset(sk, GFP_ATOMIC); |
| 725 | tcp_write_err(sk); |
| 726 | goto out; |
| 727 | } |
Eric Dumazet | e520af4 | 2015-05-06 14:26:25 -0700 | [diff] [blame] | 728 | if (tcp_write_wakeup(sk, LINUX_MIB_TCPKEEPALIVE) <= 0) { |
Arnaldo Carvalho de Melo | 6687e98 | 2005-08-10 04:03:31 -0300 | [diff] [blame] | 729 | icsk->icsk_probes_out++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 730 | elapsed = keepalive_intvl_when(tp); |
| 731 | } else { |
| 732 | /* If keepalive was lost due to local congestion, |
| 733 | * try harder. |
| 734 | */ |
| 735 | elapsed = TCP_RESOURCE_PROBE_INTERVAL; |
| 736 | } |
| 737 | } else { |
| 738 | /* It is tp->rcv_tstamp + keepalive_time_when(tp) */ |
| 739 | elapsed = keepalive_time_when(tp) - elapsed; |
| 740 | } |
| 741 | |
Hideo Aoki | 3ab224b | 2007-12-31 00:11:19 -0800 | [diff] [blame] | 742 | sk_mem_reclaim(sk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 743 | |
| 744 | resched: |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 745 | inet_csk_reset_keepalive_timer (sk, elapsed); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 746 | goto out; |
| 747 | |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 748 | death: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 749 | tcp_done(sk); |
| 750 | |
| 751 | out: |
| 752 | bh_unlock_sock(sk); |
| 753 | sock_put(sk); |
| 754 | } |
Eric Dumazet | 6f458df | 2012-07-20 05:45:50 +0000 | [diff] [blame] | 755 | |
| 756 | void tcp_init_xmit_timers(struct sock *sk) |
| 757 | { |
| 758 | inet_csk_init_xmit_timers(sk, &tcp_write_timer, &tcp_delack_timer, |
| 759 | &tcp_keepalive_timer); |
| 760 | } |