1/* 2 * Copyright (c) 2006, 2017 Oracle and/or its affiliates. All rights reserved. 3 * 4 * This software is available to you under a choice of one of two 5 * licenses. You may choose to be licensed under the terms of the GNU 6 * General Public License (GPL) Version 2, available from the file 7 * COPYING in the main directory of this source tree, or the 8 * OpenIB.org BSD license below: 9 * 10 * Redistribution and use in source and binary forms, with or 11 * without modification, are permitted provided that the following 12 * conditions are met: 13 * 14 * - Redistributions of source code must retain the above 15 * copyright notice, this list of conditions and the following 16 * disclaimer. 17 * 18 * - Redistributions in binary form must reproduce the above 19 * copyright notice, this list of conditions and the following 20 * disclaimer in the documentation and/or other materials 21 * provided with the distribution. 22 * 23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 30 * SOFTWARE. 31 * 32 */ 33#include <linux/kernel.h> 34#include <linux/in.h> 35#include <net/tcp.h> 36 37#include "rds_single_path.h" 38#include "rds.h" 39#include "tcp.h" 40 41void rds_tcp_xmit_path_prepare(struct rds_conn_path *cp) 42{ 43 struct rds_tcp_connection *tc = cp->cp_transport_data; 44 45 tcp_sock_set_cork(tc->t_sock->sk, true); 46} 47 48void rds_tcp_xmit_path_complete(struct rds_conn_path *cp) 49{ 50 struct rds_tcp_connection *tc = cp->cp_transport_data; 51 52 tcp_sock_set_cork(tc->t_sock->sk, false); 53} 54 55/* the core send_sem serializes this with other xmit and shutdown */ 56static int rds_tcp_sendmsg(struct socket *sock, void *data, unsigned int len) 57{ 58 struct kvec vec = { 59 .iov_base = data, 60 .iov_len = len, 61 }; 62 struct msghdr msg = { 63 .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL, 64 }; 65 66 return kernel_sendmsg(sock, &msg, &vec, 1, vec.iov_len); 67} 68 69/* the core send_sem serializes this with other xmit and shutdown */ 70int rds_tcp_xmit(struct rds_connection *conn, struct rds_message *rm, 71 unsigned int hdr_off, unsigned int sg, unsigned int off) 72{ 73 struct rds_conn_path *cp = rm->m_inc.i_conn_path; 74 struct rds_tcp_connection *tc = cp->cp_transport_data; 75 int done = 0; 76 int ret = 0; 77 int more; 78 79 if (hdr_off == 0) { 80 /* 81 * m_ack_seq is set to the sequence number of the last byte of 82 * header and data. see rds_tcp_is_acked(). 83 */ 84 tc->t_last_sent_nxt = rds_tcp_write_seq(tc); 85 rm->m_ack_seq = tc->t_last_sent_nxt + 86 sizeof(struct rds_header) + 87 be32_to_cpu(rm->m_inc.i_hdr.h_len) - 1; 88 smp_mb__before_atomic(); 89 set_bit(RDS_MSG_HAS_ACK_SEQ, &rm->m_flags); 90 tc->t_last_expected_una = rm->m_ack_seq + 1; 91 92 if (test_bit(RDS_MSG_RETRANSMITTED, &rm->m_flags)) 93 rm->m_inc.i_hdr.h_flags |= RDS_FLAG_RETRANSMITTED; 94 95 rdsdebug("rm %p tcp nxt %u ack_seq %llu\n", 96 rm, rds_tcp_write_seq(tc), 97 (unsigned long long)rm->m_ack_seq); 98 } 99 100 if (hdr_off < sizeof(struct rds_header)) { 101 /* see rds_tcp_write_space() */ 102 set_bit(SOCK_NOSPACE, &tc->t_sock->sk->sk_socket->flags); 103 104 ret = rds_tcp_sendmsg(tc->t_sock, 105 (void *)&rm->m_inc.i_hdr + hdr_off, 106 sizeof(rm->m_inc.i_hdr) - hdr_off); 107 if (ret < 0) 108 goto out; 109 done += ret; 110 if (hdr_off + done != sizeof(struct rds_header)) 111 goto out; 112 } 113 114 more = rm->data.op_nents > 1 ? (MSG_MORE | MSG_SENDPAGE_NOTLAST) : 0; 115 while (sg < rm->data.op_nents) { 116 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | more; 117 118 ret = tc->t_sock->ops->sendpage(tc->t_sock, 119 sg_page(&rm->data.op_sg[sg]), 120 rm->data.op_sg[sg].offset + off, 121 rm->data.op_sg[sg].length - off, 122 flags); 123 rdsdebug("tcp sendpage %p:%u:%u ret %d\n", (void *)sg_page(&rm->data.op_sg[sg]), 124 rm->data.op_sg[sg].offset + off, rm->data.op_sg[sg].length - off, 125 ret); 126 if (ret <= 0) 127 break; 128 129 off += ret; 130 done += ret; 131 if (off == rm->data.op_sg[sg].length) { 132 off = 0; 133 sg++; 134 } 135 if (sg == rm->data.op_nents - 1) 136 more = 0; 137 } 138 139out: 140 if (ret <= 0) { 141 /* write_space will hit after EAGAIN, all else fatal */ 142 if (ret == -EAGAIN) { 143 rds_tcp_stats_inc(s_tcp_sndbuf_full); 144 ret = 0; 145 } else { 146 /* No need to disconnect/reconnect if path_drop 147 * has already been triggered, because, e.g., of 148 * an incoming RST. 149 */ 150 if (rds_conn_path_up(cp)) { 151 pr_warn("RDS/tcp: send to %pI6c on cp [%d]" 152 "returned %d, " 153 "disconnecting and reconnecting\n", 154 &conn->c_faddr, cp->cp_index, ret); 155 rds_conn_path_drop(cp, false); 156 } 157 } 158 } 159 if (done == 0) 160 done = ret; 161 return done; 162} 163 164/* 165 * rm->m_ack_seq is set to the tcp sequence number that corresponds to the 166 * last byte of the message, including the header. This means that the 167 * entire message has been received if rm->m_ack_seq is "before" the next 168 * unacked byte of the TCP sequence space. We have to do very careful 169 * wrapping 32bit comparisons here. 170 */ 171static int rds_tcp_is_acked(struct rds_message *rm, uint64_t ack) 172{ 173 if (!test_bit(RDS_MSG_HAS_ACK_SEQ, &rm->m_flags)) 174 return 0; 175 return (__s32)((u32)rm->m_ack_seq - (u32)ack) < 0; 176} 177 178void rds_tcp_write_space(struct sock *sk) 179{ 180 void (*write_space)(struct sock *sk); 181 struct rds_conn_path *cp; 182 struct rds_tcp_connection *tc; 183 184 read_lock_bh(&sk->sk_callback_lock); 185 cp = sk->sk_user_data; 186 if (!cp) { 187 write_space = sk->sk_write_space; 188 goto out; 189 } 190 191 tc = cp->cp_transport_data; 192 rdsdebug("write_space for tc %p\n", tc); 193 write_space = tc->t_orig_write_space; 194 rds_tcp_stats_inc(s_tcp_write_space_calls); 195 196 rdsdebug("tcp una %u\n", rds_tcp_snd_una(tc)); 197 tc->t_last_seen_una = rds_tcp_snd_una(tc); 198 rds_send_path_drop_acked(cp, rds_tcp_snd_una(tc), rds_tcp_is_acked); 199 200 rcu_read_lock(); 201 if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf && 202 !rds_destroy_pending(cp->cp_conn)) 203 queue_delayed_work(rds_wq, &cp->cp_send_w, 0); 204 rcu_read_unlock(); 205 206out: 207 read_unlock_bh(&sk->sk_callback_lock); 208 209 /* 210 * write_space is only called when data leaves tcp's send queue if 211 * SOCK_NOSPACE is set. We set SOCK_NOSPACE every time we put 212 * data in tcp's send queue because we use write_space to parse the 213 * sequence numbers and notice that rds messages have been fully 214 * received. 215 * 216 * tcp's write_space clears SOCK_NOSPACE if the send queue has more 217 * than a certain amount of space. So we need to set it again *after* 218 * we call tcp's write_space or else we might only get called on the 219 * first of a series of incoming tcp acks. 220 */ 221 write_space(sk); 222 223 if (sk->sk_socket) 224 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 225} 226