Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 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> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 34 | #include <linux/gfp.h> |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 35 | #include <net/sock.h> |
| 36 | #include <linux/in.h> |
| 37 | #include <linux/list.h> |
| 38 | |
| 39 | #include "rds.h" |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 40 | |
| 41 | /* When transmitting messages in rds_send_xmit, we need to emerge from |
| 42 | * time to time and briefly release the CPU. Otherwise the softlock watchdog |
| 43 | * will kick our shin. |
| 44 | * Also, it seems fairer to not let one busy connection stall all the |
| 45 | * others. |
| 46 | * |
| 47 | * send_batch_count is the number of times we'll loop in send_xmit. Setting |
| 48 | * it to 0 will restore the old behavior (where we looped until we had |
| 49 | * drained the queue). |
| 50 | */ |
| 51 | static int send_batch_count = 64; |
| 52 | module_param(send_batch_count, int, 0444); |
| 53 | MODULE_PARM_DESC(send_batch_count, " batch factor when working the send queue"); |
| 54 | |
| 55 | /* |
| 56 | * Reset the send state. Caller must hold c_send_lock when calling here. |
| 57 | */ |
| 58 | void rds_send_reset(struct rds_connection *conn) |
| 59 | { |
| 60 | struct rds_message *rm, *tmp; |
| 61 | unsigned long flags; |
| 62 | |
Chris Mason | 7e3f295 | 2010-05-11 15:11:11 -0700 | [diff] [blame] | 63 | spin_lock_irqsave(&conn->c_send_lock, flags); |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 64 | if (conn->c_xmit_rm) { |
Chris Mason | 7e3f295 | 2010-05-11 15:11:11 -0700 | [diff] [blame] | 65 | rm = conn->c_xmit_rm; |
| 66 | conn->c_xmit_rm = NULL; |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 67 | /* Tell the user the RDMA op is no longer mapped by the |
| 68 | * transport. This isn't entirely true (it's flushed out |
| 69 | * independently) but as the connection is down, there's |
| 70 | * no ongoing RDMA to/from that memory */ |
Chris Mason | 7e3f295 | 2010-05-11 15:11:11 -0700 | [diff] [blame] | 71 | rds_message_unmapped(rm); |
| 72 | spin_unlock_irqrestore(&conn->c_send_lock, flags); |
| 73 | |
| 74 | rds_message_put(rm); |
| 75 | } else { |
| 76 | spin_unlock_irqrestore(&conn->c_send_lock, flags); |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 77 | } |
Chris Mason | 7e3f295 | 2010-05-11 15:11:11 -0700 | [diff] [blame] | 78 | |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 79 | conn->c_xmit_sg = 0; |
| 80 | conn->c_xmit_hdr_off = 0; |
| 81 | conn->c_xmit_data_off = 0; |
Andy Grover | 15133f6 | 2010-01-12 14:33:38 -0800 | [diff] [blame] | 82 | conn->c_xmit_atomic_sent = 0; |
Andy Grover | 5b2366b | 2010-02-03 19:36:44 -0800 | [diff] [blame] | 83 | conn->c_xmit_rdma_sent = 0; |
| 84 | conn->c_xmit_data_sent = 0; |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 85 | |
| 86 | conn->c_map_queued = 0; |
| 87 | |
| 88 | conn->c_unacked_packets = rds_sysctl_max_unacked_packets; |
| 89 | conn->c_unacked_bytes = rds_sysctl_max_unacked_bytes; |
| 90 | |
| 91 | /* Mark messages as retransmissions, and move them to the send q */ |
| 92 | spin_lock_irqsave(&conn->c_lock, flags); |
| 93 | list_for_each_entry_safe(rm, tmp, &conn->c_retrans, m_conn_item) { |
| 94 | set_bit(RDS_MSG_ACK_REQUIRED, &rm->m_flags); |
| 95 | set_bit(RDS_MSG_RETRANSMITTED, &rm->m_flags); |
| 96 | } |
| 97 | list_splice_init(&conn->c_retrans, &conn->c_send_queue); |
| 98 | spin_unlock_irqrestore(&conn->c_lock, flags); |
| 99 | } |
| 100 | |
| 101 | /* |
| 102 | * We're making the concious trade-off here to only send one message |
| 103 | * down the connection at a time. |
| 104 | * Pro: |
| 105 | * - tx queueing is a simple fifo list |
| 106 | * - reassembly is optional and easily done by transports per conn |
| 107 | * - no per flow rx lookup at all, straight to the socket |
| 108 | * - less per-frag memory and wire overhead |
| 109 | * Con: |
| 110 | * - queued acks can be delayed behind large messages |
| 111 | * Depends: |
| 112 | * - small message latency is higher behind queued large messages |
| 113 | * - large message latency isn't starved by intervening small sends |
| 114 | */ |
| 115 | int rds_send_xmit(struct rds_connection *conn) |
| 116 | { |
| 117 | struct rds_message *rm; |
| 118 | unsigned long flags; |
| 119 | unsigned int tmp; |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 120 | struct scatterlist *sg; |
| 121 | int ret = 0; |
Chris Mason | 9e29db0 | 2010-04-15 16:38:14 -0400 | [diff] [blame] | 122 | int gen = 0; |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 123 | LIST_HEAD(to_be_dropped); |
| 124 | |
Andy Grover | fcc5450 | 2010-03-29 17:08:49 -0700 | [diff] [blame] | 125 | restart: |
Andy Grover | 049ee3f | 2010-03-23 17:39:07 -0700 | [diff] [blame] | 126 | if (!rds_conn_up(conn)) |
| 127 | goto out; |
| 128 | |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 129 | /* |
| 130 | * sendmsg calls here after having queued its message on the send |
| 131 | * queue. We only have one task feeding the connection at a time. If |
| 132 | * another thread is already feeding the queue then we back off. This |
| 133 | * avoids blocking the caller and trading per-connection data between |
| 134 | * caches per message. |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 135 | */ |
Andy Grover | 049ee3f | 2010-03-23 17:39:07 -0700 | [diff] [blame] | 136 | if (!spin_trylock_irqsave(&conn->c_send_lock, flags)) { |
| 137 | rds_stats_inc(s_send_lock_contention); |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 138 | ret = -ENOMEM; |
| 139 | goto out; |
| 140 | } |
Chris Mason | 7e3f295 | 2010-05-11 15:11:11 -0700 | [diff] [blame] | 141 | atomic_inc(&conn->c_senders); |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 142 | |
| 143 | if (conn->c_trans->xmit_prepare) |
| 144 | conn->c_trans->xmit_prepare(conn); |
| 145 | |
Chris Mason | 9e29db0 | 2010-04-15 16:38:14 -0400 | [diff] [blame] | 146 | gen = atomic_inc_return(&conn->c_send_generation); |
| 147 | |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 148 | /* |
| 149 | * spin trying to push headers and data down the connection until |
Andy Grover | 5b2366b | 2010-02-03 19:36:44 -0800 | [diff] [blame] | 150 | * the connection doesn't make forward progress. |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 151 | */ |
Andy Grover | fcc5450 | 2010-03-29 17:08:49 -0700 | [diff] [blame] | 152 | while (1) { |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 153 | |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 154 | rm = conn->c_xmit_rm; |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 155 | |
Andy Grover | 5b2366b | 2010-02-03 19:36:44 -0800 | [diff] [blame] | 156 | /* |
| 157 | * If between sending messages, we can send a pending congestion |
| 158 | * map update. |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 159 | */ |
Andy Grover | 8690bfa | 2010-01-12 11:56:44 -0800 | [diff] [blame] | 160 | if (!rm && test_and_clear_bit(0, &conn->c_map_queued)) { |
Andy Grover | 77dd550 | 2010-03-22 15:22:04 -0700 | [diff] [blame] | 161 | rm = rds_cong_update_alloc(conn); |
| 162 | if (IS_ERR(rm)) { |
| 163 | ret = PTR_ERR(rm); |
| 164 | break; |
Andy Grover | 5b2366b | 2010-02-03 19:36:44 -0800 | [diff] [blame] | 165 | } |
Andy Grover | 77dd550 | 2010-03-22 15:22:04 -0700 | [diff] [blame] | 166 | rm->data.op_active = 1; |
| 167 | |
| 168 | conn->c_xmit_rm = rm; |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 169 | } |
| 170 | |
| 171 | /* |
Andy Grover | 5b2366b | 2010-02-03 19:36:44 -0800 | [diff] [blame] | 172 | * If not already working on one, grab the next message. |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 173 | * |
| 174 | * c_xmit_rm holds a ref while we're sending this message down |
| 175 | * the connction. We can use this ref while holding the |
| 176 | * send_sem.. rds_send_reset() is serialized with it. |
| 177 | */ |
Andy Grover | 8690bfa | 2010-01-12 11:56:44 -0800 | [diff] [blame] | 178 | if (!rm) { |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 179 | unsigned int len; |
| 180 | |
Andy Grover | 2ad8099 | 2010-03-23 17:48:04 -0700 | [diff] [blame] | 181 | spin_lock(&conn->c_lock); |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 182 | |
| 183 | if (!list_empty(&conn->c_send_queue)) { |
| 184 | rm = list_entry(conn->c_send_queue.next, |
| 185 | struct rds_message, |
| 186 | m_conn_item); |
| 187 | rds_message_addref(rm); |
| 188 | |
| 189 | /* |
| 190 | * Move the message from the send queue to the retransmit |
| 191 | * list right away. |
| 192 | */ |
| 193 | list_move_tail(&rm->m_conn_item, &conn->c_retrans); |
| 194 | } |
| 195 | |
Andy Grover | 2ad8099 | 2010-03-23 17:48:04 -0700 | [diff] [blame] | 196 | spin_unlock(&conn->c_lock); |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 197 | |
Andy Grover | fcc5450 | 2010-03-29 17:08:49 -0700 | [diff] [blame] | 198 | if (!rm) |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 199 | break; |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 200 | |
| 201 | /* Unfortunately, the way Infiniband deals with |
| 202 | * RDMA to a bad MR key is by moving the entire |
| 203 | * queue pair to error state. We cold possibly |
| 204 | * recover from that, but right now we drop the |
| 205 | * connection. |
| 206 | * Therefore, we never retransmit messages with RDMA ops. |
| 207 | */ |
Andy Grover | f8b3aaf | 2010-03-01 14:11:53 -0800 | [diff] [blame] | 208 | if (rm->rdma.op_active && |
Joe Perches | f64f9e7 | 2009-11-29 16:55:45 -0800 | [diff] [blame] | 209 | test_bit(RDS_MSG_RETRANSMITTED, &rm->m_flags)) { |
Andy Grover | 2ad8099 | 2010-03-23 17:48:04 -0700 | [diff] [blame] | 210 | spin_lock(&conn->c_lock); |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 211 | if (test_and_clear_bit(RDS_MSG_ON_CONN, &rm->m_flags)) |
| 212 | list_move(&rm->m_conn_item, &to_be_dropped); |
Andy Grover | 2ad8099 | 2010-03-23 17:48:04 -0700 | [diff] [blame] | 213 | spin_unlock(&conn->c_lock); |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 214 | continue; |
| 215 | } |
| 216 | |
| 217 | /* Require an ACK every once in a while */ |
| 218 | len = ntohl(rm->m_inc.i_hdr.h_len); |
Joe Perches | f64f9e7 | 2009-11-29 16:55:45 -0800 | [diff] [blame] | 219 | if (conn->c_unacked_packets == 0 || |
| 220 | conn->c_unacked_bytes < len) { |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 221 | __set_bit(RDS_MSG_ACK_REQUIRED, &rm->m_flags); |
| 222 | |
| 223 | conn->c_unacked_packets = rds_sysctl_max_unacked_packets; |
| 224 | conn->c_unacked_bytes = rds_sysctl_max_unacked_bytes; |
| 225 | rds_stats_inc(s_send_ack_required); |
| 226 | } else { |
| 227 | conn->c_unacked_bytes -= len; |
| 228 | conn->c_unacked_packets--; |
| 229 | } |
| 230 | |
| 231 | conn->c_xmit_rm = rm; |
| 232 | } |
| 233 | |
Andy Grover | 2c3a5f9 | 2010-03-01 16:10:40 -0800 | [diff] [blame] | 234 | /* The transport either sends the whole rdma or none of it */ |
| 235 | if (rm->rdma.op_active && !conn->c_xmit_rdma_sent) { |
Andy Grover | ff3d7d3 | 2010-03-01 14:03:09 -0800 | [diff] [blame] | 236 | rm->m_final_op = &rm->rdma; |
Andy Grover | 2c3a5f9 | 2010-03-01 16:10:40 -0800 | [diff] [blame] | 237 | ret = conn->c_trans->xmit_rdma(conn, &rm->rdma); |
Chris Mason | 1cc2228 | 2010-05-11 16:15:35 -0700 | [diff] [blame^] | 238 | if (ret) |
Andy Grover | 2c3a5f9 | 2010-03-01 16:10:40 -0800 | [diff] [blame] | 239 | break; |
| 240 | conn->c_xmit_rdma_sent = 1; |
| 241 | |
| 242 | /* The transport owns the mapped memory for now. |
| 243 | * You can't unmap it while it's on the send queue */ |
| 244 | set_bit(RDS_MSG_MAPPED, &rm->m_flags); |
| 245 | } |
| 246 | |
Andy Grover | 15133f6 | 2010-01-12 14:33:38 -0800 | [diff] [blame] | 247 | if (rm->atomic.op_active && !conn->c_xmit_atomic_sent) { |
Andy Grover | ff3d7d3 | 2010-03-01 14:03:09 -0800 | [diff] [blame] | 248 | rm->m_final_op = &rm->atomic; |
| 249 | ret = conn->c_trans->xmit_atomic(conn, &rm->atomic); |
Chris Mason | 1cc2228 | 2010-05-11 16:15:35 -0700 | [diff] [blame^] | 250 | if (ret) |
Andy Grover | 15133f6 | 2010-01-12 14:33:38 -0800 | [diff] [blame] | 251 | break; |
| 252 | conn->c_xmit_atomic_sent = 1; |
Andy Grover | ff3d7d3 | 2010-03-01 14:03:09 -0800 | [diff] [blame] | 253 | |
Andy Grover | 15133f6 | 2010-01-12 14:33:38 -0800 | [diff] [blame] | 254 | /* The transport owns the mapped memory for now. |
| 255 | * You can't unmap it while it's on the send queue */ |
| 256 | set_bit(RDS_MSG_MAPPED, &rm->m_flags); |
| 257 | } |
| 258 | |
Andy Grover | 2c3a5f9 | 2010-03-01 16:10:40 -0800 | [diff] [blame] | 259 | /* |
| 260 | * A number of cases require an RDS header to be sent |
| 261 | * even if there is no data. |
| 262 | * We permit 0-byte sends; rds-ping depends on this. |
| 263 | * However, if there are exclusively attached silent ops, |
| 264 | * we skip the hdr/data send, to enable silent operation. |
| 265 | */ |
| 266 | if (rm->data.op_nents == 0) { |
| 267 | int ops_present; |
| 268 | int all_ops_are_silent = 1; |
Andy Grover | 241eef3 | 2010-01-19 21:25:26 -0800 | [diff] [blame] | 269 | |
Andy Grover | 2c3a5f9 | 2010-03-01 16:10:40 -0800 | [diff] [blame] | 270 | ops_present = (rm->atomic.op_active || rm->rdma.op_active); |
| 271 | if (rm->atomic.op_active && !rm->atomic.op_silent) |
| 272 | all_ops_are_silent = 0; |
| 273 | if (rm->rdma.op_active && !rm->rdma.op_silent) |
| 274 | all_ops_are_silent = 0; |
Andy Grover | 241eef3 | 2010-01-19 21:25:26 -0800 | [diff] [blame] | 275 | |
Andy Grover | 2c3a5f9 | 2010-03-01 16:10:40 -0800 | [diff] [blame] | 276 | if (ops_present && all_ops_are_silent |
| 277 | && !rm->m_rdma_cookie) |
| 278 | rm->data.op_active = 0; |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 279 | } |
| 280 | |
Andy Grover | 5b2366b | 2010-02-03 19:36:44 -0800 | [diff] [blame] | 281 | if (rm->data.op_active && !conn->c_xmit_data_sent) { |
Andy Grover | ff3d7d3 | 2010-03-01 14:03:09 -0800 | [diff] [blame] | 282 | rm->m_final_op = &rm->data; |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 283 | ret = conn->c_trans->xmit(conn, rm, |
| 284 | conn->c_xmit_hdr_off, |
| 285 | conn->c_xmit_sg, |
| 286 | conn->c_xmit_data_off); |
| 287 | if (ret <= 0) |
| 288 | break; |
| 289 | |
| 290 | if (conn->c_xmit_hdr_off < sizeof(struct rds_header)) { |
| 291 | tmp = min_t(int, ret, |
| 292 | sizeof(struct rds_header) - |
| 293 | conn->c_xmit_hdr_off); |
| 294 | conn->c_xmit_hdr_off += tmp; |
| 295 | ret -= tmp; |
| 296 | } |
| 297 | |
Andy Grover | 6c7cc6e | 2010-01-27 18:04:18 -0800 | [diff] [blame] | 298 | sg = &rm->data.op_sg[conn->c_xmit_sg]; |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 299 | while (ret) { |
| 300 | tmp = min_t(int, ret, sg->length - |
| 301 | conn->c_xmit_data_off); |
| 302 | conn->c_xmit_data_off += tmp; |
| 303 | ret -= tmp; |
| 304 | if (conn->c_xmit_data_off == sg->length) { |
| 305 | conn->c_xmit_data_off = 0; |
| 306 | sg++; |
| 307 | conn->c_xmit_sg++; |
| 308 | BUG_ON(ret != 0 && |
Andy Grover | 6c7cc6e | 2010-01-27 18:04:18 -0800 | [diff] [blame] | 309 | conn->c_xmit_sg == rm->data.op_nents); |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 310 | } |
| 311 | } |
Andy Grover | 5b2366b | 2010-02-03 19:36:44 -0800 | [diff] [blame] | 312 | |
| 313 | if (conn->c_xmit_hdr_off == sizeof(struct rds_header) && |
| 314 | (conn->c_xmit_sg == rm->data.op_nents)) |
| 315 | conn->c_xmit_data_sent = 1; |
| 316 | } |
| 317 | |
| 318 | /* |
| 319 | * A rm will only take multiple times through this loop |
| 320 | * if there is a data op. Thus, if the data is sent (or there was |
| 321 | * none), then we're done with the rm. |
| 322 | */ |
| 323 | if (!rm->data.op_active || conn->c_xmit_data_sent) { |
| 324 | conn->c_xmit_rm = NULL; |
| 325 | conn->c_xmit_sg = 0; |
| 326 | conn->c_xmit_hdr_off = 0; |
| 327 | conn->c_xmit_data_off = 0; |
| 328 | conn->c_xmit_rdma_sent = 0; |
| 329 | conn->c_xmit_atomic_sent = 0; |
| 330 | conn->c_xmit_data_sent = 0; |
| 331 | |
| 332 | rds_message_put(rm); |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 333 | } |
| 334 | } |
| 335 | |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 336 | if (conn->c_trans->xmit_complete) |
| 337 | conn->c_trans->xmit_complete(conn); |
| 338 | |
| 339 | /* |
| 340 | * We might be racing with another sender who queued a message but |
| 341 | * backed off on noticing that we held the c_send_lock. If we check |
| 342 | * for queued messages after dropping the sem then either we'll |
| 343 | * see the queued message or the queuer will get the sem. If we |
| 344 | * notice the queued message then we trigger an immediate retry. |
| 345 | * |
| 346 | * We need to be careful only to do this when we stopped processing |
| 347 | * the send queue because it was empty. It's the only way we |
| 348 | * stop processing the loop when the transport hasn't taken |
| 349 | * responsibility for forward progress. |
| 350 | */ |
Andy Grover | 049ee3f | 2010-03-23 17:39:07 -0700 | [diff] [blame] | 351 | spin_unlock_irqrestore(&conn->c_send_lock, flags); |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 352 | |
Andy Grover | 2ad8099 | 2010-03-23 17:48:04 -0700 | [diff] [blame] | 353 | /* Nuke any messages we decided not to retransmit. */ |
| 354 | if (!list_empty(&to_be_dropped)) { |
| 355 | /* irqs on here, so we can put(), unlike above */ |
| 356 | list_for_each_entry(rm, &to_be_dropped, m_conn_item) |
| 357 | rds_message_put(rm); |
| 358 | rds_send_remove_from_sock(&to_be_dropped, RDS_RDMA_DROPPED); |
| 359 | } |
| 360 | |
Chris Mason | 7e3f295 | 2010-05-11 15:11:11 -0700 | [diff] [blame] | 361 | atomic_dec(&conn->c_senders); |
| 362 | |
Andy Grover | fcc5450 | 2010-03-29 17:08:49 -0700 | [diff] [blame] | 363 | /* |
| 364 | * Other senders will see we have c_send_lock and exit. We |
| 365 | * need to recheck the send queue and race again for c_send_lock |
Andy Grover | ce47f52 | 2010-04-15 17:19:29 -0700 | [diff] [blame] | 366 | * to make sure messages don't just sit on the send queue, if |
| 367 | * somebody hasn't already beat us into the loop. |
Andy Grover | fcc5450 | 2010-03-29 17:08:49 -0700 | [diff] [blame] | 368 | * |
| 369 | * If the transport cannot continue (i.e ret != 0), then it must |
| 370 | * call us when more room is available, such as from the tx |
| 371 | * completion handler. |
| 372 | */ |
| 373 | if (ret == 0) { |
Chris Mason | 9e29db0 | 2010-04-15 16:38:14 -0400 | [diff] [blame] | 374 | smp_mb(); |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 375 | if (!list_empty(&conn->c_send_queue)) { |
Andy Grover | 049ee3f | 2010-03-23 17:39:07 -0700 | [diff] [blame] | 376 | rds_stats_inc(s_send_lock_queue_raced); |
Chris Mason | 9e29db0 | 2010-04-15 16:38:14 -0400 | [diff] [blame] | 377 | if (gen == atomic_read(&conn->c_send_generation)) { |
| 378 | goto restart; |
| 379 | } |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 380 | } |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 381 | } |
| 382 | out: |
| 383 | return ret; |
| 384 | } |
| 385 | |
| 386 | static void rds_send_sndbuf_remove(struct rds_sock *rs, struct rds_message *rm) |
| 387 | { |
| 388 | u32 len = be32_to_cpu(rm->m_inc.i_hdr.h_len); |
| 389 | |
| 390 | assert_spin_locked(&rs->rs_lock); |
| 391 | |
| 392 | BUG_ON(rs->rs_snd_bytes < len); |
| 393 | rs->rs_snd_bytes -= len; |
| 394 | |
| 395 | if (rs->rs_snd_bytes == 0) |
| 396 | rds_stats_inc(s_send_queue_empty); |
| 397 | } |
| 398 | |
| 399 | static inline int rds_send_is_acked(struct rds_message *rm, u64 ack, |
| 400 | is_acked_func is_acked) |
| 401 | { |
| 402 | if (is_acked) |
| 403 | return is_acked(rm, ack); |
| 404 | return be64_to_cpu(rm->m_inc.i_hdr.h_sequence) <= ack; |
| 405 | } |
| 406 | |
| 407 | /* |
| 408 | * Returns true if there are no messages on the send and retransmit queues |
| 409 | * which have a sequence number greater than or equal to the given sequence |
| 410 | * number. |
| 411 | */ |
| 412 | int rds_send_acked_before(struct rds_connection *conn, u64 seq) |
| 413 | { |
| 414 | struct rds_message *rm, *tmp; |
| 415 | int ret = 1; |
| 416 | |
| 417 | spin_lock(&conn->c_lock); |
| 418 | |
| 419 | list_for_each_entry_safe(rm, tmp, &conn->c_retrans, m_conn_item) { |
| 420 | if (be64_to_cpu(rm->m_inc.i_hdr.h_sequence) < seq) |
| 421 | ret = 0; |
| 422 | break; |
| 423 | } |
| 424 | |
| 425 | list_for_each_entry_safe(rm, tmp, &conn->c_send_queue, m_conn_item) { |
| 426 | if (be64_to_cpu(rm->m_inc.i_hdr.h_sequence) < seq) |
| 427 | ret = 0; |
| 428 | break; |
| 429 | } |
| 430 | |
| 431 | spin_unlock(&conn->c_lock); |
| 432 | |
| 433 | return ret; |
| 434 | } |
| 435 | |
| 436 | /* |
| 437 | * This is pretty similar to what happens below in the ACK |
| 438 | * handling code - except that we call here as soon as we get |
| 439 | * the IB send completion on the RDMA op and the accompanying |
| 440 | * message. |
| 441 | */ |
| 442 | void rds_rdma_send_complete(struct rds_message *rm, int status) |
| 443 | { |
| 444 | struct rds_sock *rs = NULL; |
Andy Grover | f8b3aaf | 2010-03-01 14:11:53 -0800 | [diff] [blame] | 445 | struct rm_rdma_op *ro; |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 446 | struct rds_notifier *notifier; |
Andy Grover | 9de0864 | 2010-03-29 16:50:54 -0700 | [diff] [blame] | 447 | unsigned long flags; |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 448 | |
Andy Grover | 9de0864 | 2010-03-29 16:50:54 -0700 | [diff] [blame] | 449 | spin_lock_irqsave(&rm->m_rs_lock, flags); |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 450 | |
Andy Grover | f8b3aaf | 2010-03-01 14:11:53 -0800 | [diff] [blame] | 451 | ro = &rm->rdma; |
Joe Perches | f64f9e7 | 2009-11-29 16:55:45 -0800 | [diff] [blame] | 452 | if (test_bit(RDS_MSG_ON_SOCK, &rm->m_flags) && |
Andy Grover | f8b3aaf | 2010-03-01 14:11:53 -0800 | [diff] [blame] | 453 | ro->op_active && ro->op_notify && ro->op_notifier) { |
| 454 | notifier = ro->op_notifier; |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 455 | rs = rm->m_rs; |
| 456 | sock_hold(rds_rs_to_sk(rs)); |
| 457 | |
| 458 | notifier->n_status = status; |
| 459 | spin_lock(&rs->rs_lock); |
| 460 | list_add_tail(¬ifier->n_list, &rs->rs_notify_queue); |
| 461 | spin_unlock(&rs->rs_lock); |
| 462 | |
Andy Grover | f8b3aaf | 2010-03-01 14:11:53 -0800 | [diff] [blame] | 463 | ro->op_notifier = NULL; |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 464 | } |
| 465 | |
Andy Grover | 9de0864 | 2010-03-29 16:50:54 -0700 | [diff] [blame] | 466 | spin_unlock_irqrestore(&rm->m_rs_lock, flags); |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 467 | |
| 468 | if (rs) { |
| 469 | rds_wake_sk_sleep(rs); |
| 470 | sock_put(rds_rs_to_sk(rs)); |
| 471 | } |
| 472 | } |
Andy Grover | 616b757 | 2009-08-21 12:28:32 +0000 | [diff] [blame] | 473 | EXPORT_SYMBOL_GPL(rds_rdma_send_complete); |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 474 | |
| 475 | /* |
Andy Grover | 15133f6 | 2010-01-12 14:33:38 -0800 | [diff] [blame] | 476 | * Just like above, except looks at atomic op |
| 477 | */ |
| 478 | void rds_atomic_send_complete(struct rds_message *rm, int status) |
| 479 | { |
| 480 | struct rds_sock *rs = NULL; |
| 481 | struct rm_atomic_op *ao; |
| 482 | struct rds_notifier *notifier; |
Andy Grover | cf4b738 | 2010-03-29 16:50:54 -0700 | [diff] [blame] | 483 | unsigned long flags; |
Andy Grover | 15133f6 | 2010-01-12 14:33:38 -0800 | [diff] [blame] | 484 | |
Andy Grover | cf4b738 | 2010-03-29 16:50:54 -0700 | [diff] [blame] | 485 | spin_lock_irqsave(&rm->m_rs_lock, flags); |
Andy Grover | 15133f6 | 2010-01-12 14:33:38 -0800 | [diff] [blame] | 486 | |
| 487 | ao = &rm->atomic; |
| 488 | if (test_bit(RDS_MSG_ON_SOCK, &rm->m_flags) |
| 489 | && ao->op_active && ao->op_notify && ao->op_notifier) { |
| 490 | notifier = ao->op_notifier; |
| 491 | rs = rm->m_rs; |
| 492 | sock_hold(rds_rs_to_sk(rs)); |
| 493 | |
| 494 | notifier->n_status = status; |
| 495 | spin_lock(&rs->rs_lock); |
| 496 | list_add_tail(¬ifier->n_list, &rs->rs_notify_queue); |
| 497 | spin_unlock(&rs->rs_lock); |
| 498 | |
| 499 | ao->op_notifier = NULL; |
| 500 | } |
| 501 | |
Andy Grover | cf4b738 | 2010-03-29 16:50:54 -0700 | [diff] [blame] | 502 | spin_unlock_irqrestore(&rm->m_rs_lock, flags); |
Andy Grover | 15133f6 | 2010-01-12 14:33:38 -0800 | [diff] [blame] | 503 | |
| 504 | if (rs) { |
| 505 | rds_wake_sk_sleep(rs); |
| 506 | sock_put(rds_rs_to_sk(rs)); |
| 507 | } |
| 508 | } |
| 509 | EXPORT_SYMBOL_GPL(rds_atomic_send_complete); |
| 510 | |
| 511 | /* |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 512 | * This is the same as rds_rdma_send_complete except we |
| 513 | * don't do any locking - we have all the ingredients (message, |
| 514 | * socket, socket lock) and can just move the notifier. |
| 515 | */ |
| 516 | static inline void |
Andy Grover | 940786e | 2010-02-19 18:04:58 -0800 | [diff] [blame] | 517 | __rds_send_complete(struct rds_sock *rs, struct rds_message *rm, int status) |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 518 | { |
Andy Grover | f8b3aaf | 2010-03-01 14:11:53 -0800 | [diff] [blame] | 519 | struct rm_rdma_op *ro; |
Andy Grover | 940786e | 2010-02-19 18:04:58 -0800 | [diff] [blame] | 520 | struct rm_atomic_op *ao; |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 521 | |
Andy Grover | f8b3aaf | 2010-03-01 14:11:53 -0800 | [diff] [blame] | 522 | ro = &rm->rdma; |
| 523 | if (ro->op_active && ro->op_notify && ro->op_notifier) { |
| 524 | ro->op_notifier->n_status = status; |
| 525 | list_add_tail(&ro->op_notifier->n_list, &rs->rs_notify_queue); |
| 526 | ro->op_notifier = NULL; |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 527 | } |
| 528 | |
Andy Grover | 940786e | 2010-02-19 18:04:58 -0800 | [diff] [blame] | 529 | ao = &rm->atomic; |
| 530 | if (ao->op_active && ao->op_notify && ao->op_notifier) { |
| 531 | ao->op_notifier->n_status = status; |
| 532 | list_add_tail(&ao->op_notifier->n_list, &rs->rs_notify_queue); |
| 533 | ao->op_notifier = NULL; |
| 534 | } |
| 535 | |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 536 | /* No need to wake the app - caller does this */ |
| 537 | } |
| 538 | |
| 539 | /* |
| 540 | * This is called from the IB send completion when we detect |
| 541 | * a RDMA operation that failed with remote access error. |
| 542 | * So speed is not an issue here. |
| 543 | */ |
| 544 | struct rds_message *rds_send_get_message(struct rds_connection *conn, |
Andy Grover | f8b3aaf | 2010-03-01 14:11:53 -0800 | [diff] [blame] | 545 | struct rm_rdma_op *op) |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 546 | { |
| 547 | struct rds_message *rm, *tmp, *found = NULL; |
| 548 | unsigned long flags; |
| 549 | |
| 550 | spin_lock_irqsave(&conn->c_lock, flags); |
| 551 | |
| 552 | list_for_each_entry_safe(rm, tmp, &conn->c_retrans, m_conn_item) { |
Andy Grover | f8b3aaf | 2010-03-01 14:11:53 -0800 | [diff] [blame] | 553 | if (&rm->rdma == op) { |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 554 | atomic_inc(&rm->m_refcount); |
| 555 | found = rm; |
| 556 | goto out; |
| 557 | } |
| 558 | } |
| 559 | |
| 560 | list_for_each_entry_safe(rm, tmp, &conn->c_send_queue, m_conn_item) { |
Andy Grover | f8b3aaf | 2010-03-01 14:11:53 -0800 | [diff] [blame] | 561 | if (&rm->rdma == op) { |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 562 | atomic_inc(&rm->m_refcount); |
| 563 | found = rm; |
| 564 | break; |
| 565 | } |
| 566 | } |
| 567 | |
| 568 | out: |
| 569 | spin_unlock_irqrestore(&conn->c_lock, flags); |
| 570 | |
| 571 | return found; |
| 572 | } |
Andy Grover | 616b757 | 2009-08-21 12:28:32 +0000 | [diff] [blame] | 573 | EXPORT_SYMBOL_GPL(rds_send_get_message); |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 574 | |
| 575 | /* |
| 576 | * This removes messages from the socket's list if they're on it. The list |
| 577 | * argument must be private to the caller, we must be able to modify it |
| 578 | * without locks. The messages must have a reference held for their |
| 579 | * position on the list. This function will drop that reference after |
| 580 | * removing the messages from the 'messages' list regardless of if it found |
| 581 | * the messages on the socket list or not. |
| 582 | */ |
| 583 | void rds_send_remove_from_sock(struct list_head *messages, int status) |
| 584 | { |
Andy Grover | 561c7df | 2010-03-11 13:50:06 +0000 | [diff] [blame] | 585 | unsigned long flags; |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 586 | struct rds_sock *rs = NULL; |
| 587 | struct rds_message *rm; |
| 588 | |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 589 | while (!list_empty(messages)) { |
Andy Grover | 561c7df | 2010-03-11 13:50:06 +0000 | [diff] [blame] | 590 | int was_on_sock = 0; |
| 591 | |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 592 | rm = list_entry(messages->next, struct rds_message, |
| 593 | m_conn_item); |
| 594 | list_del_init(&rm->m_conn_item); |
| 595 | |
| 596 | /* |
| 597 | * If we see this flag cleared then we're *sure* that someone |
| 598 | * else beat us to removing it from the sock. If we race |
| 599 | * with their flag update we'll get the lock and then really |
| 600 | * see that the flag has been cleared. |
| 601 | * |
| 602 | * The message spinlock makes sure nobody clears rm->m_rs |
| 603 | * while we're messing with it. It does not prevent the |
| 604 | * message from being removed from the socket, though. |
| 605 | */ |
Andy Grover | 561c7df | 2010-03-11 13:50:06 +0000 | [diff] [blame] | 606 | spin_lock_irqsave(&rm->m_rs_lock, flags); |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 607 | if (!test_bit(RDS_MSG_ON_SOCK, &rm->m_flags)) |
| 608 | goto unlock_and_drop; |
| 609 | |
| 610 | if (rs != rm->m_rs) { |
| 611 | if (rs) { |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 612 | rds_wake_sk_sleep(rs); |
| 613 | sock_put(rds_rs_to_sk(rs)); |
| 614 | } |
| 615 | rs = rm->m_rs; |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 616 | sock_hold(rds_rs_to_sk(rs)); |
| 617 | } |
Tina Yang | 048c15e | 2010-03-11 13:50:00 +0000 | [diff] [blame] | 618 | spin_lock(&rs->rs_lock); |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 619 | |
| 620 | if (test_and_clear_bit(RDS_MSG_ON_SOCK, &rm->m_flags)) { |
Andy Grover | f8b3aaf | 2010-03-01 14:11:53 -0800 | [diff] [blame] | 621 | struct rm_rdma_op *ro = &rm->rdma; |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 622 | struct rds_notifier *notifier; |
| 623 | |
| 624 | list_del_init(&rm->m_sock_item); |
| 625 | rds_send_sndbuf_remove(rs, rm); |
| 626 | |
Andy Grover | f8b3aaf | 2010-03-01 14:11:53 -0800 | [diff] [blame] | 627 | if (ro->op_active && ro->op_notifier && |
| 628 | (ro->op_notify || (ro->op_recverr && status))) { |
| 629 | notifier = ro->op_notifier; |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 630 | list_add_tail(¬ifier->n_list, |
| 631 | &rs->rs_notify_queue); |
| 632 | if (!notifier->n_status) |
| 633 | notifier->n_status = status; |
Andy Grover | f8b3aaf | 2010-03-01 14:11:53 -0800 | [diff] [blame] | 634 | rm->rdma.op_notifier = NULL; |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 635 | } |
Andy Grover | 561c7df | 2010-03-11 13:50:06 +0000 | [diff] [blame] | 636 | was_on_sock = 1; |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 637 | rm->m_rs = NULL; |
| 638 | } |
Tina Yang | 048c15e | 2010-03-11 13:50:00 +0000 | [diff] [blame] | 639 | spin_unlock(&rs->rs_lock); |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 640 | |
| 641 | unlock_and_drop: |
Andy Grover | 561c7df | 2010-03-11 13:50:06 +0000 | [diff] [blame] | 642 | spin_unlock_irqrestore(&rm->m_rs_lock, flags); |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 643 | rds_message_put(rm); |
Andy Grover | 561c7df | 2010-03-11 13:50:06 +0000 | [diff] [blame] | 644 | if (was_on_sock) |
| 645 | rds_message_put(rm); |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 646 | } |
| 647 | |
| 648 | if (rs) { |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 649 | rds_wake_sk_sleep(rs); |
| 650 | sock_put(rds_rs_to_sk(rs)); |
| 651 | } |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 652 | } |
| 653 | |
| 654 | /* |
| 655 | * Transports call here when they've determined that the receiver queued |
| 656 | * messages up to, and including, the given sequence number. Messages are |
| 657 | * moved to the retrans queue when rds_send_xmit picks them off the send |
| 658 | * queue. This means that in the TCP case, the message may not have been |
| 659 | * assigned the m_ack_seq yet - but that's fine as long as tcp_is_acked |
| 660 | * checks the RDS_MSG_HAS_ACK_SEQ bit. |
| 661 | * |
| 662 | * XXX It's not clear to me how this is safely serialized with socket |
| 663 | * destruction. Maybe it should bail if it sees SOCK_DEAD. |
| 664 | */ |
| 665 | void rds_send_drop_acked(struct rds_connection *conn, u64 ack, |
| 666 | is_acked_func is_acked) |
| 667 | { |
| 668 | struct rds_message *rm, *tmp; |
| 669 | unsigned long flags; |
| 670 | LIST_HEAD(list); |
| 671 | |
| 672 | spin_lock_irqsave(&conn->c_lock, flags); |
| 673 | |
| 674 | list_for_each_entry_safe(rm, tmp, &conn->c_retrans, m_conn_item) { |
| 675 | if (!rds_send_is_acked(rm, ack, is_acked)) |
| 676 | break; |
| 677 | |
| 678 | list_move(&rm->m_conn_item, &list); |
| 679 | clear_bit(RDS_MSG_ON_CONN, &rm->m_flags); |
| 680 | } |
| 681 | |
| 682 | /* order flag updates with spin locks */ |
| 683 | if (!list_empty(&list)) |
| 684 | smp_mb__after_clear_bit(); |
| 685 | |
| 686 | spin_unlock_irqrestore(&conn->c_lock, flags); |
| 687 | |
| 688 | /* now remove the messages from the sock list as needed */ |
| 689 | rds_send_remove_from_sock(&list, RDS_RDMA_SUCCESS); |
| 690 | } |
Andy Grover | 616b757 | 2009-08-21 12:28:32 +0000 | [diff] [blame] | 691 | EXPORT_SYMBOL_GPL(rds_send_drop_acked); |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 692 | |
| 693 | void rds_send_drop_to(struct rds_sock *rs, struct sockaddr_in *dest) |
| 694 | { |
| 695 | struct rds_message *rm, *tmp; |
| 696 | struct rds_connection *conn; |
Andy Grover | 7c82eaf | 2010-02-19 18:01:41 -0800 | [diff] [blame] | 697 | unsigned long flags; |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 698 | LIST_HEAD(list); |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 699 | |
| 700 | /* get all the messages we're dropping under the rs lock */ |
| 701 | spin_lock_irqsave(&rs->rs_lock, flags); |
| 702 | |
| 703 | list_for_each_entry_safe(rm, tmp, &rs->rs_send_queue, m_sock_item) { |
| 704 | if (dest && (dest->sin_addr.s_addr != rm->m_daddr || |
| 705 | dest->sin_port != rm->m_inc.i_hdr.h_dport)) |
| 706 | continue; |
| 707 | |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 708 | list_move(&rm->m_sock_item, &list); |
| 709 | rds_send_sndbuf_remove(rs, rm); |
| 710 | clear_bit(RDS_MSG_ON_SOCK, &rm->m_flags); |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 711 | } |
| 712 | |
| 713 | /* order flag updates with the rs lock */ |
Andy Grover | 7c82eaf | 2010-02-19 18:01:41 -0800 | [diff] [blame] | 714 | smp_mb__after_clear_bit(); |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 715 | |
| 716 | spin_unlock_irqrestore(&rs->rs_lock, flags); |
| 717 | |
Andy Grover | 7c82eaf | 2010-02-19 18:01:41 -0800 | [diff] [blame] | 718 | if (list_empty(&list)) |
| 719 | return; |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 720 | |
Andy Grover | 7c82eaf | 2010-02-19 18:01:41 -0800 | [diff] [blame] | 721 | /* Remove the messages from the conn */ |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 722 | list_for_each_entry(rm, &list, m_sock_item) { |
Andy Grover | 7c82eaf | 2010-02-19 18:01:41 -0800 | [diff] [blame] | 723 | |
| 724 | conn = rm->m_inc.i_conn; |
Andy Grover | 7c82eaf | 2010-02-19 18:01:41 -0800 | [diff] [blame] | 725 | |
Andy Grover | 9de0864 | 2010-03-29 16:50:54 -0700 | [diff] [blame] | 726 | spin_lock_irqsave(&conn->c_lock, flags); |
Andy Grover | 7c82eaf | 2010-02-19 18:01:41 -0800 | [diff] [blame] | 727 | /* |
| 728 | * Maybe someone else beat us to removing rm from the conn. |
| 729 | * If we race with their flag update we'll get the lock and |
| 730 | * then really see that the flag has been cleared. |
| 731 | */ |
| 732 | if (!test_and_clear_bit(RDS_MSG_ON_CONN, &rm->m_flags)) { |
| 733 | spin_unlock_irqrestore(&conn->c_lock, flags); |
| 734 | continue; |
| 735 | } |
Andy Grover | 9de0864 | 2010-03-29 16:50:54 -0700 | [diff] [blame] | 736 | list_del_init(&rm->m_conn_item); |
| 737 | spin_unlock_irqrestore(&conn->c_lock, flags); |
Andy Grover | 7c82eaf | 2010-02-19 18:01:41 -0800 | [diff] [blame] | 738 | |
| 739 | /* |
| 740 | * Couldn't grab m_rs_lock in top loop (lock ordering), |
| 741 | * but we can now. |
| 742 | */ |
Andy Grover | 9de0864 | 2010-03-29 16:50:54 -0700 | [diff] [blame] | 743 | spin_lock_irqsave(&rm->m_rs_lock, flags); |
Andy Grover | 7c82eaf | 2010-02-19 18:01:41 -0800 | [diff] [blame] | 744 | |
Tina Yang | 550a800 | 2010-03-11 13:50:03 +0000 | [diff] [blame] | 745 | spin_lock(&rs->rs_lock); |
Andy Grover | 940786e | 2010-02-19 18:04:58 -0800 | [diff] [blame] | 746 | __rds_send_complete(rs, rm, RDS_RDMA_CANCELED); |
Tina Yang | 550a800 | 2010-03-11 13:50:03 +0000 | [diff] [blame] | 747 | spin_unlock(&rs->rs_lock); |
Andy Grover | 7c82eaf | 2010-02-19 18:01:41 -0800 | [diff] [blame] | 748 | |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 749 | rm->m_rs = NULL; |
Andy Grover | 9de0864 | 2010-03-29 16:50:54 -0700 | [diff] [blame] | 750 | spin_unlock_irqrestore(&rm->m_rs_lock, flags); |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 751 | |
Andy Grover | 7c82eaf | 2010-02-19 18:01:41 -0800 | [diff] [blame] | 752 | rds_message_put(rm); |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 753 | } |
| 754 | |
Andy Grover | 7c82eaf | 2010-02-19 18:01:41 -0800 | [diff] [blame] | 755 | rds_wake_sk_sleep(rs); |
Tina Yang | 550a800 | 2010-03-11 13:50:03 +0000 | [diff] [blame] | 756 | |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 757 | while (!list_empty(&list)) { |
| 758 | rm = list_entry(list.next, struct rds_message, m_sock_item); |
| 759 | list_del_init(&rm->m_sock_item); |
| 760 | |
| 761 | rds_message_wait(rm); |
| 762 | rds_message_put(rm); |
| 763 | } |
| 764 | } |
| 765 | |
| 766 | /* |
| 767 | * we only want this to fire once so we use the callers 'queued'. It's |
| 768 | * possible that another thread can race with us and remove the |
| 769 | * message from the flow with RDS_CANCEL_SENT_TO. |
| 770 | */ |
| 771 | static int rds_send_queue_rm(struct rds_sock *rs, struct rds_connection *conn, |
| 772 | struct rds_message *rm, __be16 sport, |
| 773 | __be16 dport, int *queued) |
| 774 | { |
| 775 | unsigned long flags; |
| 776 | u32 len; |
| 777 | |
| 778 | if (*queued) |
| 779 | goto out; |
| 780 | |
| 781 | len = be32_to_cpu(rm->m_inc.i_hdr.h_len); |
| 782 | |
| 783 | /* this is the only place which holds both the socket's rs_lock |
| 784 | * and the connection's c_lock */ |
| 785 | spin_lock_irqsave(&rs->rs_lock, flags); |
| 786 | |
| 787 | /* |
| 788 | * If there is a little space in sndbuf, we don't queue anything, |
| 789 | * and userspace gets -EAGAIN. But poll() indicates there's send |
| 790 | * room. This can lead to bad behavior (spinning) if snd_bytes isn't |
| 791 | * freed up by incoming acks. So we check the *old* value of |
| 792 | * rs_snd_bytes here to allow the last msg to exceed the buffer, |
| 793 | * and poll() now knows no more data can be sent. |
| 794 | */ |
| 795 | if (rs->rs_snd_bytes < rds_sk_sndbuf(rs)) { |
| 796 | rs->rs_snd_bytes += len; |
| 797 | |
| 798 | /* let recv side know we are close to send space exhaustion. |
| 799 | * This is probably not the optimal way to do it, as this |
| 800 | * means we set the flag on *all* messages as soon as our |
| 801 | * throughput hits a certain threshold. |
| 802 | */ |
| 803 | if (rs->rs_snd_bytes >= rds_sk_sndbuf(rs) / 2) |
| 804 | __set_bit(RDS_MSG_ACK_REQUIRED, &rm->m_flags); |
| 805 | |
| 806 | list_add_tail(&rm->m_sock_item, &rs->rs_send_queue); |
| 807 | set_bit(RDS_MSG_ON_SOCK, &rm->m_flags); |
| 808 | rds_message_addref(rm); |
| 809 | rm->m_rs = rs; |
| 810 | |
| 811 | /* The code ordering is a little weird, but we're |
| 812 | trying to minimize the time we hold c_lock */ |
| 813 | rds_message_populate_header(&rm->m_inc.i_hdr, sport, dport, 0); |
| 814 | rm->m_inc.i_conn = conn; |
| 815 | rds_message_addref(rm); |
| 816 | |
| 817 | spin_lock(&conn->c_lock); |
| 818 | rm->m_inc.i_hdr.h_sequence = cpu_to_be64(conn->c_next_tx_seq++); |
| 819 | list_add_tail(&rm->m_conn_item, &conn->c_send_queue); |
| 820 | set_bit(RDS_MSG_ON_CONN, &rm->m_flags); |
| 821 | spin_unlock(&conn->c_lock); |
| 822 | |
| 823 | rdsdebug("queued msg %p len %d, rs %p bytes %d seq %llu\n", |
| 824 | rm, len, rs, rs->rs_snd_bytes, |
| 825 | (unsigned long long)be64_to_cpu(rm->m_inc.i_hdr.h_sequence)); |
| 826 | |
| 827 | *queued = 1; |
| 828 | } |
| 829 | |
| 830 | spin_unlock_irqrestore(&rs->rs_lock, flags); |
| 831 | out: |
| 832 | return *queued; |
| 833 | } |
| 834 | |
Andy Grover | fc44508 | 2010-01-12 12:56:06 -0800 | [diff] [blame] | 835 | /* |
| 836 | * rds_message is getting to be quite complicated, and we'd like to allocate |
| 837 | * it all in one go. This figures out how big it needs to be up front. |
| 838 | */ |
| 839 | static int rds_rm_size(struct msghdr *msg, int data_len) |
| 840 | { |
Andy Grover | ff87e97 | 2010-01-12 14:13:15 -0800 | [diff] [blame] | 841 | struct cmsghdr *cmsg; |
Andy Grover | fc44508 | 2010-01-12 12:56:06 -0800 | [diff] [blame] | 842 | int size = 0; |
Andy Grover | aa0a4ef | 2010-04-13 12:00:35 -0700 | [diff] [blame] | 843 | int cmsg_groups = 0; |
Andy Grover | ff87e97 | 2010-01-12 14:13:15 -0800 | [diff] [blame] | 844 | int retval; |
Andy Grover | fc44508 | 2010-01-12 12:56:06 -0800 | [diff] [blame] | 845 | |
Andy Grover | ff87e97 | 2010-01-12 14:13:15 -0800 | [diff] [blame] | 846 | for (cmsg = CMSG_FIRSTHDR(msg); cmsg; cmsg = CMSG_NXTHDR(msg, cmsg)) { |
| 847 | if (!CMSG_OK(msg, cmsg)) |
| 848 | return -EINVAL; |
| 849 | |
| 850 | if (cmsg->cmsg_level != SOL_RDS) |
| 851 | continue; |
| 852 | |
| 853 | switch (cmsg->cmsg_type) { |
| 854 | case RDS_CMSG_RDMA_ARGS: |
Andy Grover | aa0a4ef | 2010-04-13 12:00:35 -0700 | [diff] [blame] | 855 | cmsg_groups |= 1; |
Andy Grover | ff87e97 | 2010-01-12 14:13:15 -0800 | [diff] [blame] | 856 | retval = rds_rdma_extra_size(CMSG_DATA(cmsg)); |
| 857 | if (retval < 0) |
| 858 | return retval; |
| 859 | size += retval; |
Andy Grover | aa0a4ef | 2010-04-13 12:00:35 -0700 | [diff] [blame] | 860 | |
Andy Grover | ff87e97 | 2010-01-12 14:13:15 -0800 | [diff] [blame] | 861 | break; |
| 862 | |
| 863 | case RDS_CMSG_RDMA_DEST: |
| 864 | case RDS_CMSG_RDMA_MAP: |
Andy Grover | aa0a4ef | 2010-04-13 12:00:35 -0700 | [diff] [blame] | 865 | cmsg_groups |= 2; |
Andy Grover | ff87e97 | 2010-01-12 14:13:15 -0800 | [diff] [blame] | 866 | /* these are valid but do no add any size */ |
| 867 | break; |
| 868 | |
Andy Grover | 15133f6 | 2010-01-12 14:33:38 -0800 | [diff] [blame] | 869 | case RDS_CMSG_ATOMIC_CSWP: |
| 870 | case RDS_CMSG_ATOMIC_FADD: |
Andy Grover | aa0a4ef | 2010-04-13 12:00:35 -0700 | [diff] [blame] | 871 | cmsg_groups |= 1; |
Andy Grover | 15133f6 | 2010-01-12 14:33:38 -0800 | [diff] [blame] | 872 | size += sizeof(struct scatterlist); |
| 873 | break; |
| 874 | |
Andy Grover | ff87e97 | 2010-01-12 14:13:15 -0800 | [diff] [blame] | 875 | default: |
| 876 | return -EINVAL; |
| 877 | } |
| 878 | |
| 879 | } |
| 880 | |
| 881 | size += ceil(data_len, PAGE_SIZE) * sizeof(struct scatterlist); |
Andy Grover | fc44508 | 2010-01-12 12:56:06 -0800 | [diff] [blame] | 882 | |
Andy Grover | aa0a4ef | 2010-04-13 12:00:35 -0700 | [diff] [blame] | 883 | /* Ensure (DEST, MAP) are never used with (ARGS, ATOMIC) */ |
| 884 | if (cmsg_groups == 3) |
| 885 | return -EINVAL; |
| 886 | |
Andy Grover | fc44508 | 2010-01-12 12:56:06 -0800 | [diff] [blame] | 887 | return size; |
| 888 | } |
| 889 | |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 890 | static int rds_cmsg_send(struct rds_sock *rs, struct rds_message *rm, |
| 891 | struct msghdr *msg, int *allocated_mr) |
| 892 | { |
| 893 | struct cmsghdr *cmsg; |
| 894 | int ret = 0; |
| 895 | |
| 896 | for (cmsg = CMSG_FIRSTHDR(msg); cmsg; cmsg = CMSG_NXTHDR(msg, cmsg)) { |
| 897 | if (!CMSG_OK(msg, cmsg)) |
| 898 | return -EINVAL; |
| 899 | |
| 900 | if (cmsg->cmsg_level != SOL_RDS) |
| 901 | continue; |
| 902 | |
| 903 | /* As a side effect, RDMA_DEST and RDMA_MAP will set |
Andy Grover | 15133f6 | 2010-01-12 14:33:38 -0800 | [diff] [blame] | 904 | * rm->rdma.m_rdma_cookie and rm->rdma.m_rdma_mr. |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 905 | */ |
| 906 | switch (cmsg->cmsg_type) { |
| 907 | case RDS_CMSG_RDMA_ARGS: |
| 908 | ret = rds_cmsg_rdma_args(rs, rm, cmsg); |
| 909 | break; |
| 910 | |
| 911 | case RDS_CMSG_RDMA_DEST: |
| 912 | ret = rds_cmsg_rdma_dest(rs, rm, cmsg); |
| 913 | break; |
| 914 | |
| 915 | case RDS_CMSG_RDMA_MAP: |
| 916 | ret = rds_cmsg_rdma_map(rs, rm, cmsg); |
| 917 | if (!ret) |
| 918 | *allocated_mr = 1; |
| 919 | break; |
Andy Grover | 15133f6 | 2010-01-12 14:33:38 -0800 | [diff] [blame] | 920 | case RDS_CMSG_ATOMIC_CSWP: |
| 921 | case RDS_CMSG_ATOMIC_FADD: |
| 922 | ret = rds_cmsg_atomic(rs, rm, cmsg); |
| 923 | break; |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 924 | |
| 925 | default: |
| 926 | return -EINVAL; |
| 927 | } |
| 928 | |
| 929 | if (ret) |
| 930 | break; |
| 931 | } |
| 932 | |
| 933 | return ret; |
| 934 | } |
| 935 | |
| 936 | int rds_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg, |
| 937 | size_t payload_len) |
| 938 | { |
| 939 | struct sock *sk = sock->sk; |
| 940 | struct rds_sock *rs = rds_sk_to_rs(sk); |
| 941 | struct sockaddr_in *usin = (struct sockaddr_in *)msg->msg_name; |
| 942 | __be32 daddr; |
| 943 | __be16 dport; |
| 944 | struct rds_message *rm = NULL; |
| 945 | struct rds_connection *conn; |
| 946 | int ret = 0; |
| 947 | int queued = 0, allocated_mr = 0; |
| 948 | int nonblock = msg->msg_flags & MSG_DONTWAIT; |
Andy Grover | 1123fd7 | 2010-03-11 13:49:56 +0000 | [diff] [blame] | 949 | long timeo = sock_sndtimeo(sk, nonblock); |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 950 | |
| 951 | /* Mirror Linux UDP mirror of BSD error message compatibility */ |
| 952 | /* XXX: Perhaps MSG_MORE someday */ |
| 953 | if (msg->msg_flags & ~(MSG_DONTWAIT | MSG_CMSG_COMPAT)) { |
| 954 | printk(KERN_INFO "msg_flags 0x%08X\n", msg->msg_flags); |
| 955 | ret = -EOPNOTSUPP; |
| 956 | goto out; |
| 957 | } |
| 958 | |
| 959 | if (msg->msg_namelen) { |
| 960 | /* XXX fail non-unicast destination IPs? */ |
| 961 | if (msg->msg_namelen < sizeof(*usin) || usin->sin_family != AF_INET) { |
| 962 | ret = -EINVAL; |
| 963 | goto out; |
| 964 | } |
| 965 | daddr = usin->sin_addr.s_addr; |
| 966 | dport = usin->sin_port; |
| 967 | } else { |
| 968 | /* We only care about consistency with ->connect() */ |
| 969 | lock_sock(sk); |
| 970 | daddr = rs->rs_conn_addr; |
| 971 | dport = rs->rs_conn_port; |
| 972 | release_sock(sk); |
| 973 | } |
| 974 | |
| 975 | /* racing with another thread binding seems ok here */ |
| 976 | if (daddr == 0 || rs->rs_bound_addr == 0) { |
| 977 | ret = -ENOTCONN; /* XXX not a great errno */ |
| 978 | goto out; |
| 979 | } |
| 980 | |
Andy Grover | fc44508 | 2010-01-12 12:56:06 -0800 | [diff] [blame] | 981 | /* size of rm including all sgs */ |
| 982 | ret = rds_rm_size(msg, payload_len); |
| 983 | if (ret < 0) |
| 984 | goto out; |
| 985 | |
| 986 | rm = rds_message_alloc(ret, GFP_KERNEL); |
| 987 | if (!rm) { |
| 988 | ret = -ENOMEM; |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 989 | goto out; |
| 990 | } |
| 991 | |
Andy Grover | 372cd7d | 2010-02-03 19:40:32 -0800 | [diff] [blame] | 992 | /* Attach data to the rm */ |
| 993 | if (payload_len) { |
| 994 | rm->data.op_sg = rds_message_alloc_sgs(rm, ceil(payload_len, PAGE_SIZE)); |
| 995 | ret = rds_message_copy_from_user(rm, msg->msg_iov, payload_len); |
| 996 | if (ret) |
| 997 | goto out; |
| 998 | } |
| 999 | rm->data.op_active = 1; |
Andy Grover | fc44508 | 2010-01-12 12:56:06 -0800 | [diff] [blame] | 1000 | |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 1001 | rm->m_daddr = daddr; |
| 1002 | |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 1003 | /* rds_conn_create has a spinlock that runs with IRQ off. |
| 1004 | * Caching the conn in the socket helps a lot. */ |
| 1005 | if (rs->rs_conn && rs->rs_conn->c_faddr == daddr) |
| 1006 | conn = rs->rs_conn; |
| 1007 | else { |
| 1008 | conn = rds_conn_create_outgoing(rs->rs_bound_addr, daddr, |
| 1009 | rs->rs_transport, |
| 1010 | sock->sk->sk_allocation); |
| 1011 | if (IS_ERR(conn)) { |
| 1012 | ret = PTR_ERR(conn); |
| 1013 | goto out; |
| 1014 | } |
| 1015 | rs->rs_conn = conn; |
| 1016 | } |
| 1017 | |
Andy Grover | 49f6969 | 2009-04-09 14:09:41 +0000 | [diff] [blame] | 1018 | /* Parse any control messages the user may have included. */ |
| 1019 | ret = rds_cmsg_send(rs, rm, msg, &allocated_mr); |
| 1020 | if (ret) |
| 1021 | goto out; |
| 1022 | |
Andy Grover | 2c3a5f9 | 2010-03-01 16:10:40 -0800 | [diff] [blame] | 1023 | if (rm->rdma.op_active && !conn->c_trans->xmit_rdma) { |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 1024 | if (printk_ratelimit()) |
| 1025 | printk(KERN_NOTICE "rdma_op %p conn xmit_rdma %p\n", |
Andy Grover | f8b3aaf | 2010-03-01 14:11:53 -0800 | [diff] [blame] | 1026 | &rm->rdma, conn->c_trans->xmit_rdma); |
Andy Grover | 15133f6 | 2010-01-12 14:33:38 -0800 | [diff] [blame] | 1027 | ret = -EOPNOTSUPP; |
| 1028 | goto out; |
| 1029 | } |
| 1030 | |
| 1031 | if (rm->atomic.op_active && !conn->c_trans->xmit_atomic) { |
| 1032 | if (printk_ratelimit()) |
| 1033 | printk(KERN_NOTICE "atomic_op %p conn xmit_atomic %p\n", |
| 1034 | &rm->atomic, conn->c_trans->xmit_atomic); |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 1035 | ret = -EOPNOTSUPP; |
| 1036 | goto out; |
| 1037 | } |
| 1038 | |
| 1039 | /* If the connection is down, trigger a connect. We may |
| 1040 | * have scheduled a delayed reconnect however - in this case |
| 1041 | * we should not interfere. |
| 1042 | */ |
Joe Perches | f64f9e7 | 2009-11-29 16:55:45 -0800 | [diff] [blame] | 1043 | if (rds_conn_state(conn) == RDS_CONN_DOWN && |
| 1044 | !test_and_set_bit(RDS_RECONNECT_PENDING, &conn->c_flags)) |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 1045 | queue_delayed_work(rds_wq, &conn->c_conn_w, 0); |
| 1046 | |
| 1047 | ret = rds_cong_wait(conn->c_fcong, dport, nonblock, rs); |
Andy Grover | b98ba52 | 2010-03-11 13:50:04 +0000 | [diff] [blame] | 1048 | if (ret) { |
| 1049 | rs->rs_seen_congestion = 1; |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 1050 | goto out; |
Andy Grover | b98ba52 | 2010-03-11 13:50:04 +0000 | [diff] [blame] | 1051 | } |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 1052 | |
| 1053 | while (!rds_send_queue_rm(rs, conn, rm, rs->rs_bound_port, |
| 1054 | dport, &queued)) { |
| 1055 | rds_stats_inc(s_send_queue_full); |
| 1056 | /* XXX make sure this is reasonable */ |
| 1057 | if (payload_len > rds_sk_sndbuf(rs)) { |
| 1058 | ret = -EMSGSIZE; |
| 1059 | goto out; |
| 1060 | } |
| 1061 | if (nonblock) { |
| 1062 | ret = -EAGAIN; |
| 1063 | goto out; |
| 1064 | } |
| 1065 | |
Eric Dumazet | aa39514 | 2010-04-20 13:03:51 +0000 | [diff] [blame] | 1066 | timeo = wait_event_interruptible_timeout(*sk_sleep(sk), |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 1067 | rds_send_queue_rm(rs, conn, rm, |
| 1068 | rs->rs_bound_port, |
| 1069 | dport, |
| 1070 | &queued), |
| 1071 | timeo); |
| 1072 | rdsdebug("sendmsg woke queued %d timeo %ld\n", queued, timeo); |
| 1073 | if (timeo > 0 || timeo == MAX_SCHEDULE_TIMEOUT) |
| 1074 | continue; |
| 1075 | |
| 1076 | ret = timeo; |
| 1077 | if (ret == 0) |
| 1078 | ret = -ETIMEDOUT; |
| 1079 | goto out; |
| 1080 | } |
| 1081 | |
| 1082 | /* |
| 1083 | * By now we've committed to the send. We reuse rds_send_worker() |
| 1084 | * to retry sends in the rds thread if the transport asks us to. |
| 1085 | */ |
| 1086 | rds_stats_inc(s_send_queued); |
| 1087 | |
| 1088 | if (!test_bit(RDS_LL_SEND_FULL, &conn->c_flags)) |
Andy Grover | a7d3a28 | 2010-03-29 16:20:18 -0700 | [diff] [blame] | 1089 | rds_send_xmit(conn); |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 1090 | |
| 1091 | rds_message_put(rm); |
| 1092 | return payload_len; |
| 1093 | |
| 1094 | out: |
| 1095 | /* If the user included a RDMA_MAP cmsg, we allocated a MR on the fly. |
| 1096 | * If the sendmsg goes through, we keep the MR. If it fails with EAGAIN |
| 1097 | * or in any other way, we need to destroy the MR again */ |
| 1098 | if (allocated_mr) |
| 1099 | rds_rdma_unuse(rs, rds_rdma_cookie_key(rm->m_rdma_cookie), 1); |
| 1100 | |
| 1101 | if (rm) |
| 1102 | rds_message_put(rm); |
| 1103 | return ret; |
| 1104 | } |
| 1105 | |
| 1106 | /* |
| 1107 | * Reply to a ping packet. |
| 1108 | */ |
| 1109 | int |
| 1110 | rds_send_pong(struct rds_connection *conn, __be16 dport) |
| 1111 | { |
| 1112 | struct rds_message *rm; |
| 1113 | unsigned long flags; |
| 1114 | int ret = 0; |
| 1115 | |
| 1116 | rm = rds_message_alloc(0, GFP_ATOMIC); |
Andy Grover | 8690bfa | 2010-01-12 11:56:44 -0800 | [diff] [blame] | 1117 | if (!rm) { |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 1118 | ret = -ENOMEM; |
| 1119 | goto out; |
| 1120 | } |
| 1121 | |
| 1122 | rm->m_daddr = conn->c_faddr; |
Andy Grover | acfcd4d | 2010-03-31 18:56:25 -0700 | [diff] [blame] | 1123 | rm->data.op_active = 1; |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 1124 | |
| 1125 | /* If the connection is down, trigger a connect. We may |
| 1126 | * have scheduled a delayed reconnect however - in this case |
| 1127 | * we should not interfere. |
| 1128 | */ |
Joe Perches | f64f9e7 | 2009-11-29 16:55:45 -0800 | [diff] [blame] | 1129 | if (rds_conn_state(conn) == RDS_CONN_DOWN && |
| 1130 | !test_and_set_bit(RDS_RECONNECT_PENDING, &conn->c_flags)) |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 1131 | queue_delayed_work(rds_wq, &conn->c_conn_w, 0); |
| 1132 | |
| 1133 | ret = rds_cong_wait(conn->c_fcong, dport, 1, NULL); |
| 1134 | if (ret) |
| 1135 | goto out; |
| 1136 | |
| 1137 | spin_lock_irqsave(&conn->c_lock, flags); |
| 1138 | list_add_tail(&rm->m_conn_item, &conn->c_send_queue); |
| 1139 | set_bit(RDS_MSG_ON_CONN, &rm->m_flags); |
| 1140 | rds_message_addref(rm); |
| 1141 | rm->m_inc.i_conn = conn; |
| 1142 | |
| 1143 | rds_message_populate_header(&rm->m_inc.i_hdr, 0, dport, |
| 1144 | conn->c_next_tx_seq); |
| 1145 | conn->c_next_tx_seq++; |
| 1146 | spin_unlock_irqrestore(&conn->c_lock, flags); |
| 1147 | |
| 1148 | rds_stats_inc(s_send_queued); |
| 1149 | rds_stats_inc(s_send_pong); |
| 1150 | |
Andy Grover | acfcd4d | 2010-03-31 18:56:25 -0700 | [diff] [blame] | 1151 | if (!test_bit(RDS_LL_SEND_FULL, &conn->c_flags)) |
| 1152 | rds_send_xmit(conn); |
| 1153 | |
Andy Grover | 5c11559 | 2009-02-24 15:30:27 +0000 | [diff] [blame] | 1154 | rds_message_put(rm); |
| 1155 | return 0; |
| 1156 | |
| 1157 | out: |
| 1158 | if (rm) |
| 1159 | rds_message_put(rm); |
| 1160 | return ret; |
| 1161 | } |