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/module.h> 34#include <linux/errno.h> 35#include <linux/kernel.h> 36#include <linux/gfp.h> 37#include <linux/in.h> 38#include <linux/poll.h> 39#include <net/sock.h> 40 41#include "rds.h" 42 43char *rds_str_array(char **array, size_t elements, size_t index) 44{ 45 if ((index < elements) && array[index]) 46 return array[index]; 47 else 48 return "unknown"; 49} 50EXPORT_SYMBOL(rds_str_array); 51 52/* this is just used for stats gathering :/ */ 53static DEFINE_SPINLOCK(rds_sock_lock); 54static unsigned long rds_sock_count; 55static LIST_HEAD(rds_sock_list); 56DECLARE_WAIT_QUEUE_HEAD(rds_poll_waitq); 57 58/* 59 * This is called as the final descriptor referencing this socket is closed. 60 * We have to unbind the socket so that another socket can be bound to the 61 * address it was using. 62 * 63 * We have to be careful about racing with the incoming path. sock_orphan() 64 * sets SOCK_DEAD and we use that as an indicator to the rx path that new 65 * messages shouldn't be queued. 66 */ 67static int rds_release(struct socket *sock) 68{ 69 struct sock *sk = sock->sk; 70 struct rds_sock *rs; 71 72 if (!sk) 73 goto out; 74 75 rs = rds_sk_to_rs(sk); 76 77 sock_orphan(sk); 78 /* Note - rds_clear_recv_queue grabs rs_recv_lock, so 79 * that ensures the recv path has completed messing 80 * with the socket. */ 81 rds_clear_recv_queue(rs); 82 rds_cong_remove_socket(rs); 83 84 /* 85 * the binding lookup hash uses rcu, we need to 86 * make sure we synchronize_rcu before we free our 87 * entry 88 */ 89 rds_remove_bound(rs); 90 synchronize_rcu(); 91 92 rds_send_drop_to(rs, NULL); 93 rds_rdma_drop_keys(rs); 94 rds_notify_queue_get(rs, NULL); 95 96 spin_lock_bh(&rds_sock_lock); 97 list_del_init(&rs->rs_item); 98 rds_sock_count--; 99 spin_unlock_bh(&rds_sock_lock); 100 101 rds_trans_put(rs->rs_transport); 102 103 sock->sk = NULL; 104 sock_put(sk); 105out: 106 return 0; 107} 108 109/* 110 * Careful not to race with rds_release -> sock_orphan which clears sk_sleep. 111 * _bh() isn't OK here, we're called from interrupt handlers. It's probably OK 112 * to wake the waitqueue after sk_sleep is clear as we hold a sock ref, but 113 * this seems more conservative. 114 * NB - normally, one would use sk_callback_lock for this, but we can 115 * get here from interrupts, whereas the network code grabs sk_callback_lock 116 * with _lock_bh only - so relying on sk_callback_lock introduces livelocks. 117 */ 118void rds_wake_sk_sleep(struct rds_sock *rs) 119{ 120 unsigned long flags; 121 122 read_lock_irqsave(&rs->rs_recv_lock, flags); 123 __rds_wake_sk_sleep(rds_rs_to_sk(rs)); 124 read_unlock_irqrestore(&rs->rs_recv_lock, flags); 125} 126 127static int rds_getname(struct socket *sock, struct sockaddr *uaddr, 128 int *uaddr_len, int peer) 129{ 130 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr; 131 struct rds_sock *rs = rds_sk_to_rs(sock->sk); 132 133 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 134 135 /* racey, don't care */ 136 if (peer) { 137 if (!rs->rs_conn_addr) 138 return -ENOTCONN; 139 140 sin->sin_port = rs->rs_conn_port; 141 sin->sin_addr.s_addr = rs->rs_conn_addr; 142 } else { 143 sin->sin_port = rs->rs_bound_port; 144 sin->sin_addr.s_addr = rs->rs_bound_addr; 145 } 146 147 sin->sin_family = AF_INET; 148 149 *uaddr_len = sizeof(*sin); 150 return 0; 151} 152 153/* 154 * RDS' poll is without a doubt the least intuitive part of the interface, 155 * as POLLIN and POLLOUT do not behave entirely as you would expect from 156 * a network protocol. 157 * 158 * POLLIN is asserted if 159 * - there is data on the receive queue. 160 * - to signal that a previously congested destination may have become 161 * uncongested 162 * - A notification has been queued to the socket (this can be a congestion 163 * update, or a RDMA completion). 164 * 165 * POLLOUT is asserted if there is room on the send queue. This does not mean 166 * however, that the next sendmsg() call will succeed. If the application tries 167 * to send to a congested destination, the system call may still fail (and 168 * return ENOBUFS). 169 */ 170static unsigned int rds_poll(struct file *file, struct socket *sock, 171 poll_table *wait) 172{ 173 struct sock *sk = sock->sk; 174 struct rds_sock *rs = rds_sk_to_rs(sk); 175 unsigned int mask = 0; 176 unsigned long flags; 177 178 poll_wait(file, sk_sleep(sk), wait); 179 180 if (rs->rs_seen_congestion) 181 poll_wait(file, &rds_poll_waitq, wait); 182 183 read_lock_irqsave(&rs->rs_recv_lock, flags); 184 if (!rs->rs_cong_monitor) { 185 /* When a congestion map was updated, we signal POLLIN for 186 * "historical" reasons. Applications can also poll for 187 * WRBAND instead. */ 188 if (rds_cong_updated_since(&rs->rs_cong_track)) 189 mask |= (POLLIN | POLLRDNORM | POLLWRBAND); 190 } else { 191 spin_lock(&rs->rs_lock); 192 if (rs->rs_cong_notify) 193 mask |= (POLLIN | POLLRDNORM); 194 spin_unlock(&rs->rs_lock); 195 } 196 if (!list_empty(&rs->rs_recv_queue) || 197 !list_empty(&rs->rs_notify_queue)) 198 mask |= (POLLIN | POLLRDNORM); 199 if (rs->rs_snd_bytes < rds_sk_sndbuf(rs)) 200 mask |= (POLLOUT | POLLWRNORM); 201 read_unlock_irqrestore(&rs->rs_recv_lock, flags); 202 203 /* clear state any time we wake a seen-congested socket */ 204 if (mask) 205 rs->rs_seen_congestion = 0; 206 207 return mask; 208} 209 210static int rds_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 211{ 212 return -ENOIOCTLCMD; 213} 214 215static int rds_cancel_sent_to(struct rds_sock *rs, char __user *optval, 216 int len) 217{ 218 struct sockaddr_in sin; 219 int ret = 0; 220 221 /* racing with another thread binding seems ok here */ 222 if (rs->rs_bound_addr == 0) { 223 ret = -ENOTCONN; /* XXX not a great errno */ 224 goto out; 225 } 226 227 if (len < sizeof(struct sockaddr_in)) { 228 ret = -EINVAL; 229 goto out; 230 } 231 232 if (copy_from_user(&sin, optval, sizeof(sin))) { 233 ret = -EFAULT; 234 goto out; 235 } 236 237 rds_send_drop_to(rs, &sin); 238out: 239 return ret; 240} 241 242static int rds_set_bool_option(unsigned char *optvar, char __user *optval, 243 int optlen) 244{ 245 int value; 246 247 if (optlen < sizeof(int)) 248 return -EINVAL; 249 if (get_user(value, (int __user *) optval)) 250 return -EFAULT; 251 *optvar = !!value; 252 return 0; 253} 254 255static int rds_cong_monitor(struct rds_sock *rs, char __user *optval, 256 int optlen) 257{ 258 int ret; 259 260 ret = rds_set_bool_option(&rs->rs_cong_monitor, optval, optlen); 261 if (ret == 0) { 262 if (rs->rs_cong_monitor) { 263 rds_cong_add_socket(rs); 264 } else { 265 rds_cong_remove_socket(rs); 266 rs->rs_cong_mask = 0; 267 rs->rs_cong_notify = 0; 268 } 269 } 270 return ret; 271} 272 273static int rds_setsockopt(struct socket *sock, int level, int optname, 274 char __user *optval, unsigned int optlen) 275{ 276 struct rds_sock *rs = rds_sk_to_rs(sock->sk); 277 int ret; 278 279 if (level != SOL_RDS) { 280 ret = -ENOPROTOOPT; 281 goto out; 282 } 283 284 switch (optname) { 285 case RDS_CANCEL_SENT_TO: 286 ret = rds_cancel_sent_to(rs, optval, optlen); 287 break; 288 case RDS_GET_MR: 289 ret = rds_get_mr(rs, optval, optlen); 290 break; 291 case RDS_GET_MR_FOR_DEST: 292 ret = rds_get_mr_for_dest(rs, optval, optlen); 293 break; 294 case RDS_FREE_MR: 295 ret = rds_free_mr(rs, optval, optlen); 296 break; 297 case RDS_RECVERR: 298 ret = rds_set_bool_option(&rs->rs_recverr, optval, optlen); 299 break; 300 case RDS_CONG_MONITOR: 301 ret = rds_cong_monitor(rs, optval, optlen); 302 break; 303 default: 304 ret = -ENOPROTOOPT; 305 } 306out: 307 return ret; 308} 309 310static int rds_getsockopt(struct socket *sock, int level, int optname, 311 char __user *optval, int __user *optlen) 312{ 313 struct rds_sock *rs = rds_sk_to_rs(sock->sk); 314 int ret = -ENOPROTOOPT, len; 315 316 if (level != SOL_RDS) 317 goto out; 318 319 if (get_user(len, optlen)) { 320 ret = -EFAULT; 321 goto out; 322 } 323 324 switch (optname) { 325 case RDS_INFO_FIRST ... RDS_INFO_LAST: 326 ret = rds_info_getsockopt(sock, optname, optval, 327 optlen); 328 break; 329 330 case RDS_RECVERR: 331 if (len < sizeof(int)) 332 ret = -EINVAL; 333 else 334 if (put_user(rs->rs_recverr, (int __user *) optval) || 335 put_user(sizeof(int), optlen)) 336 ret = -EFAULT; 337 else 338 ret = 0; 339 break; 340 default: 341 break; 342 } 343 344out: 345 return ret; 346 347} 348 349static int rds_connect(struct socket *sock, struct sockaddr *uaddr, 350 int addr_len, int flags) 351{ 352 struct sock *sk = sock->sk; 353 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr; 354 struct rds_sock *rs = rds_sk_to_rs(sk); 355 int ret = 0; 356 357 lock_sock(sk); 358 359 if (addr_len != sizeof(struct sockaddr_in)) { 360 ret = -EINVAL; 361 goto out; 362 } 363 364 if (sin->sin_family != AF_INET) { 365 ret = -EAFNOSUPPORT; 366 goto out; 367 } 368 369 if (sin->sin_addr.s_addr == htonl(INADDR_ANY)) { 370 ret = -EDESTADDRREQ; 371 goto out; 372 } 373 374 rs->rs_conn_addr = sin->sin_addr.s_addr; 375 rs->rs_conn_port = sin->sin_port; 376 377out: 378 release_sock(sk); 379 return ret; 380} 381 382static struct proto rds_proto = { 383 .name = "RDS", 384 .owner = THIS_MODULE, 385 .obj_size = sizeof(struct rds_sock), 386}; 387 388static const struct proto_ops rds_proto_ops = { 389 .family = AF_RDS, 390 .owner = THIS_MODULE, 391 .release = rds_release, 392 .bind = rds_bind, 393 .connect = rds_connect, 394 .socketpair = sock_no_socketpair, 395 .accept = sock_no_accept, 396 .getname = rds_getname, 397 .poll = rds_poll, 398 .ioctl = rds_ioctl, 399 .listen = sock_no_listen, 400 .shutdown = sock_no_shutdown, 401 .setsockopt = rds_setsockopt, 402 .getsockopt = rds_getsockopt, 403 .sendmsg = rds_sendmsg, 404 .recvmsg = rds_recvmsg, 405 .mmap = sock_no_mmap, 406 .sendpage = sock_no_sendpage, 407}; 408 409static int __rds_create(struct socket *sock, struct sock *sk, int protocol) 410{ 411 struct rds_sock *rs; 412 413 sock_init_data(sock, sk); 414 sock->ops = &rds_proto_ops; 415 sk->sk_protocol = protocol; 416 417 rs = rds_sk_to_rs(sk); 418 spin_lock_init(&rs->rs_lock); 419 rwlock_init(&rs->rs_recv_lock); 420 INIT_LIST_HEAD(&rs->rs_send_queue); 421 INIT_LIST_HEAD(&rs->rs_recv_queue); 422 INIT_LIST_HEAD(&rs->rs_notify_queue); 423 INIT_LIST_HEAD(&rs->rs_cong_list); 424 spin_lock_init(&rs->rs_rdma_lock); 425 rs->rs_rdma_keys = RB_ROOT; 426 427 spin_lock_bh(&rds_sock_lock); 428 list_add_tail(&rs->rs_item, &rds_sock_list); 429 rds_sock_count++; 430 spin_unlock_bh(&rds_sock_lock); 431 432 return 0; 433} 434 435static int rds_create(struct net *net, struct socket *sock, int protocol, 436 int kern) 437{ 438 struct sock *sk; 439 440 if (sock->type != SOCK_SEQPACKET || protocol) 441 return -ESOCKTNOSUPPORT; 442 443 sk = sk_alloc(net, AF_RDS, GFP_ATOMIC, &rds_proto); 444 if (!sk) 445 return -ENOMEM; 446 447 return __rds_create(sock, sk, protocol); 448} 449 450void rds_sock_addref(struct rds_sock *rs) 451{ 452 sock_hold(rds_rs_to_sk(rs)); 453} 454 455void rds_sock_put(struct rds_sock *rs) 456{ 457 sock_put(rds_rs_to_sk(rs)); 458} 459 460static const struct net_proto_family rds_family_ops = { 461 .family = AF_RDS, 462 .create = rds_create, 463 .owner = THIS_MODULE, 464}; 465 466static void rds_sock_inc_info(struct socket *sock, unsigned int len, 467 struct rds_info_iterator *iter, 468 struct rds_info_lengths *lens) 469{ 470 struct rds_sock *rs; 471 struct rds_incoming *inc; 472 unsigned int total = 0; 473 474 len /= sizeof(struct rds_info_message); 475 476 spin_lock_bh(&rds_sock_lock); 477 478 list_for_each_entry(rs, &rds_sock_list, rs_item) { 479 read_lock(&rs->rs_recv_lock); 480 481 /* XXX too lazy to maintain counts.. */ 482 list_for_each_entry(inc, &rs->rs_recv_queue, i_item) { 483 total++; 484 if (total <= len) 485 rds_inc_info_copy(inc, iter, inc->i_saddr, 486 rs->rs_bound_addr, 1); 487 } 488 489 read_unlock(&rs->rs_recv_lock); 490 } 491 492 spin_unlock_bh(&rds_sock_lock); 493 494 lens->nr = total; 495 lens->each = sizeof(struct rds_info_message); 496} 497 498static void rds_sock_info(struct socket *sock, unsigned int len, 499 struct rds_info_iterator *iter, 500 struct rds_info_lengths *lens) 501{ 502 struct rds_info_socket sinfo; 503 struct rds_sock *rs; 504 505 len /= sizeof(struct rds_info_socket); 506 507 spin_lock_bh(&rds_sock_lock); 508 509 if (len < rds_sock_count) 510 goto out; 511 512 list_for_each_entry(rs, &rds_sock_list, rs_item) { 513 sinfo.sndbuf = rds_sk_sndbuf(rs); 514 sinfo.rcvbuf = rds_sk_rcvbuf(rs); 515 sinfo.bound_addr = rs->rs_bound_addr; 516 sinfo.connected_addr = rs->rs_conn_addr; 517 sinfo.bound_port = rs->rs_bound_port; 518 sinfo.connected_port = rs->rs_conn_port; 519 sinfo.inum = sock_i_ino(rds_rs_to_sk(rs)); 520 521 rds_info_copy(iter, &sinfo, sizeof(sinfo)); 522 } 523 524out: 525 lens->nr = rds_sock_count; 526 lens->each = sizeof(struct rds_info_socket); 527 528 spin_unlock_bh(&rds_sock_lock); 529} 530 531static void rds_exit(void) 532{ 533 sock_unregister(rds_family_ops.family); 534 proto_unregister(&rds_proto); 535 rds_conn_exit(); 536 rds_cong_exit(); 537 rds_sysctl_exit(); 538 rds_threads_exit(); 539 rds_stats_exit(); 540 rds_page_exit(); 541 rds_info_deregister_func(RDS_INFO_SOCKETS, rds_sock_info); 542 rds_info_deregister_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info); 543} 544module_exit(rds_exit); 545 546static int rds_init(void) 547{ 548 int ret; 549 550 ret = rds_conn_init(); 551 if (ret) 552 goto out; 553 ret = rds_threads_init(); 554 if (ret) 555 goto out_conn; 556 ret = rds_sysctl_init(); 557 if (ret) 558 goto out_threads; 559 ret = rds_stats_init(); 560 if (ret) 561 goto out_sysctl; 562 ret = proto_register(&rds_proto, 1); 563 if (ret) 564 goto out_stats; 565 ret = sock_register(&rds_family_ops); 566 if (ret) 567 goto out_proto; 568 569 rds_info_register_func(RDS_INFO_SOCKETS, rds_sock_info); 570 rds_info_register_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info); 571 572 goto out; 573 574out_proto: 575 proto_unregister(&rds_proto); 576out_stats: 577 rds_stats_exit(); 578out_sysctl: 579 rds_sysctl_exit(); 580out_threads: 581 rds_threads_exit(); 582out_conn: 583 rds_conn_exit(); 584 rds_cong_exit(); 585 rds_page_exit(); 586out: 587 return ret; 588} 589module_init(rds_init); 590 591#define DRV_VERSION "4.0" 592#define DRV_RELDATE "Feb 12, 2009" 593 594MODULE_AUTHOR("Oracle Corporation <rds-devel@oss.oracle.com>"); 595MODULE_DESCRIPTION("RDS: Reliable Datagram Sockets" 596 " v" DRV_VERSION " (" DRV_RELDATE ")"); 597MODULE_VERSION(DRV_VERSION); 598MODULE_LICENSE("Dual BSD/GPL"); 599MODULE_ALIAS_NETPROTO(PF_RDS); 600