1 /* 2 * Copyright (c) 2006 Oracle. 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 41 static void rds_tcp_cork(struct socket *sock, int val) 42 { 43 mm_segment_t oldfs; 44 45 oldfs = get_fs(); 46 set_fs(KERNEL_DS); 47 sock->ops->setsockopt(sock, SOL_TCP, TCP_CORK, (char __user *)&val, 48 sizeof(val)); 49 set_fs(oldfs); 50 } 51 52 void rds_tcp_xmit_path_prepare(struct rds_conn_path *cp) 53 { 54 struct rds_tcp_connection *tc = cp->cp_transport_data; 55 56 rds_tcp_cork(tc->t_sock, 1); 57 } 58 59 void rds_tcp_xmit_path_complete(struct rds_conn_path *cp) 60 { 61 struct rds_tcp_connection *tc = cp->cp_transport_data; 62 63 rds_tcp_cork(tc->t_sock, 0); 64 } 65 66 /* the core send_sem serializes this with other xmit and shutdown */ 67 static int rds_tcp_sendmsg(struct socket *sock, void *data, unsigned int len) 68 { 69 struct kvec vec = { 70 .iov_base = data, 71 .iov_len = len, 72 }; 73 struct msghdr msg = { 74 .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL, 75 }; 76 77 return kernel_sendmsg(sock, &msg, &vec, 1, vec.iov_len); 78 } 79 80 /* the core send_sem serializes this with other xmit and shutdown */ 81 int rds_tcp_xmit(struct rds_connection *conn, struct rds_message *rm, 82 unsigned int hdr_off, unsigned int sg, unsigned int off) 83 { 84 struct rds_tcp_connection *tc = conn->c_transport_data; 85 int done = 0; 86 int ret = 0; 87 int more; 88 89 if (hdr_off == 0) { 90 /* 91 * m_ack_seq is set to the sequence number of the last byte of 92 * header and data. see rds_tcp_is_acked(). 93 */ 94 tc->t_last_sent_nxt = rds_tcp_snd_nxt(tc); 95 rm->m_ack_seq = tc->t_last_sent_nxt + 96 sizeof(struct rds_header) + 97 be32_to_cpu(rm->m_inc.i_hdr.h_len) - 1; 98 smp_mb__before_atomic(); 99 set_bit(RDS_MSG_HAS_ACK_SEQ, &rm->m_flags); 100 tc->t_last_expected_una = rm->m_ack_seq + 1; 101 102 rdsdebug("rm %p tcp nxt %u ack_seq %llu\n", 103 rm, rds_tcp_snd_nxt(tc), 104 (unsigned long long)rm->m_ack_seq); 105 } 106 107 if (hdr_off < sizeof(struct rds_header)) { 108 /* see rds_tcp_write_space() */ 109 set_bit(SOCK_NOSPACE, &tc->t_sock->sk->sk_socket->flags); 110 111 ret = rds_tcp_sendmsg(tc->t_sock, 112 (void *)&rm->m_inc.i_hdr + hdr_off, 113 sizeof(rm->m_inc.i_hdr) - hdr_off); 114 if (ret < 0) 115 goto out; 116 done += ret; 117 if (hdr_off + done != sizeof(struct rds_header)) 118 goto out; 119 } 120 121 more = rm->data.op_nents > 1 ? (MSG_MORE | MSG_SENDPAGE_NOTLAST) : 0; 122 while (sg < rm->data.op_nents) { 123 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | more; 124 125 ret = tc->t_sock->ops->sendpage(tc->t_sock, 126 sg_page(&rm->data.op_sg[sg]), 127 rm->data.op_sg[sg].offset + off, 128 rm->data.op_sg[sg].length - off, 129 flags); 130 rdsdebug("tcp sendpage %p:%u:%u ret %d\n", (void *)sg_page(&rm->data.op_sg[sg]), 131 rm->data.op_sg[sg].offset + off, rm->data.op_sg[sg].length - off, 132 ret); 133 if (ret <= 0) 134 break; 135 136 off += ret; 137 done += ret; 138 if (off == rm->data.op_sg[sg].length) { 139 off = 0; 140 sg++; 141 } 142 if (sg == rm->data.op_nents - 1) 143 more = 0; 144 } 145 146 out: 147 if (ret <= 0) { 148 /* write_space will hit after EAGAIN, all else fatal */ 149 if (ret == -EAGAIN) { 150 rds_tcp_stats_inc(s_tcp_sndbuf_full); 151 ret = 0; 152 } else { 153 printk(KERN_WARNING "RDS/tcp: send to %pI4 " 154 "returned %d, disconnecting and reconnecting\n", 155 &conn->c_faddr, ret); 156 rds_conn_drop(conn); 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 */ 171 static 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 } 177 178 void rds_tcp_write_space(struct sock *sk) 179 { 180 void (*write_space)(struct sock *sk); 181 struct rds_connection *conn; 182 struct rds_tcp_connection *tc; 183 184 read_lock_bh(&sk->sk_callback_lock); 185 conn = sk->sk_user_data; 186 if (!conn) { 187 write_space = sk->sk_write_space; 188 goto out; 189 } 190 191 tc = conn->c_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_drop_acked(conn, rds_tcp_snd_una(tc), rds_tcp_is_acked); 199 200 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) 201 queue_delayed_work(rds_wq, &conn->c_send_w, 0); 202 203 out: 204 read_unlock_bh(&sk->sk_callback_lock); 205 206 /* 207 * write_space is only called when data leaves tcp's send queue if 208 * SOCK_NOSPACE is set. We set SOCK_NOSPACE every time we put 209 * data in tcp's send queue because we use write_space to parse the 210 * sequence numbers and notice that rds messages have been fully 211 * received. 212 * 213 * tcp's write_space clears SOCK_NOSPACE if the send queue has more 214 * than a certain amount of space. So we need to set it again *after* 215 * we call tcp's write_space or else we might only get called on the 216 * first of a series of incoming tcp acks. 217 */ 218 write_space(sk); 219 220 if (sk->sk_socket) 221 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 222 } 223