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 43 /* this is just used for stats gathering :/ */ 44 static DEFINE_SPINLOCK(rds_sock_lock); 45 static unsigned long rds_sock_count; 46 static LIST_HEAD(rds_sock_list); 47 DECLARE_WAIT_QUEUE_HEAD(rds_poll_waitq); 48 49 /* 50 * This is called as the final descriptor referencing this socket is closed. 51 * We have to unbind the socket so that another socket can be bound to the 52 * address it was using. 53 * 54 * We have to be careful about racing with the incoming path. sock_orphan() 55 * sets SOCK_DEAD and we use that as an indicator to the rx path that new 56 * messages shouldn't be queued. 57 */ 58 static int rds_release(struct socket *sock) 59 { 60 struct sock *sk = sock->sk; 61 struct rds_sock *rs; 62 63 if (!sk) 64 goto out; 65 66 rs = rds_sk_to_rs(sk); 67 68 sock_orphan(sk); 69 /* Note - rds_clear_recv_queue grabs rs_recv_lock, so 70 * that ensures the recv path has completed messing 71 * with the socket. */ 72 rds_clear_recv_queue(rs); 73 rds_cong_remove_socket(rs); 74 75 rds_remove_bound(rs); 76 77 rds_send_drop_to(rs, NULL); 78 rds_rdma_drop_keys(rs); 79 rds_notify_queue_get(rs, NULL); 80 81 spin_lock_bh(&rds_sock_lock); 82 list_del_init(&rs->rs_item); 83 rds_sock_count--; 84 spin_unlock_bh(&rds_sock_lock); 85 86 rds_trans_put(rs->rs_transport); 87 88 sock->sk = NULL; 89 sock_put(sk); 90 out: 91 return 0; 92 } 93 94 /* 95 * Careful not to race with rds_release -> sock_orphan which clears sk_sleep. 96 * _bh() isn't OK here, we're called from interrupt handlers. It's probably OK 97 * to wake the waitqueue after sk_sleep is clear as we hold a sock ref, but 98 * this seems more conservative. 99 * NB - normally, one would use sk_callback_lock for this, but we can 100 * get here from interrupts, whereas the network code grabs sk_callback_lock 101 * with _lock_bh only - so relying on sk_callback_lock introduces livelocks. 102 */ 103 void rds_wake_sk_sleep(struct rds_sock *rs) 104 { 105 unsigned long flags; 106 107 read_lock_irqsave(&rs->rs_recv_lock, flags); 108 __rds_wake_sk_sleep(rds_rs_to_sk(rs)); 109 read_unlock_irqrestore(&rs->rs_recv_lock, flags); 110 } 111 112 static int rds_getname(struct socket *sock, struct sockaddr *uaddr, 113 int *uaddr_len, int peer) 114 { 115 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr; 116 struct rds_sock *rs = rds_sk_to_rs(sock->sk); 117 118 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 119 120 /* racey, don't care */ 121 if (peer) { 122 if (!rs->rs_conn_addr) 123 return -ENOTCONN; 124 125 sin->sin_port = rs->rs_conn_port; 126 sin->sin_addr.s_addr = rs->rs_conn_addr; 127 } else { 128 sin->sin_port = rs->rs_bound_port; 129 sin->sin_addr.s_addr = rs->rs_bound_addr; 130 } 131 132 sin->sin_family = AF_INET; 133 134 *uaddr_len = sizeof(*sin); 135 return 0; 136 } 137 138 /* 139 * RDS' poll is without a doubt the least intuitive part of the interface, 140 * as POLLIN and POLLOUT do not behave entirely as you would expect from 141 * a network protocol. 142 * 143 * POLLIN is asserted if 144 * - there is data on the receive queue. 145 * - to signal that a previously congested destination may have become 146 * uncongested 147 * - A notification has been queued to the socket (this can be a congestion 148 * update, or a RDMA completion). 149 * 150 * POLLOUT is asserted if there is room on the send queue. This does not mean 151 * however, that the next sendmsg() call will succeed. If the application tries 152 * to send to a congested destination, the system call may still fail (and 153 * return ENOBUFS). 154 */ 155 static unsigned int rds_poll(struct file *file, struct socket *sock, 156 poll_table *wait) 157 { 158 struct sock *sk = sock->sk; 159 struct rds_sock *rs = rds_sk_to_rs(sk); 160 unsigned int mask = 0; 161 unsigned long flags; 162 163 poll_wait(file, sk_sleep(sk), wait); 164 165 if (rs->rs_seen_congestion) 166 poll_wait(file, &rds_poll_waitq, wait); 167 168 read_lock_irqsave(&rs->rs_recv_lock, flags); 169 if (!rs->rs_cong_monitor) { 170 /* When a congestion map was updated, we signal POLLIN for 171 * "historical" reasons. Applications can also poll for 172 * WRBAND instead. */ 173 if (rds_cong_updated_since(&rs->rs_cong_track)) 174 mask |= (POLLIN | POLLRDNORM | POLLWRBAND); 175 } else { 176 spin_lock(&rs->rs_lock); 177 if (rs->rs_cong_notify) 178 mask |= (POLLIN | POLLRDNORM); 179 spin_unlock(&rs->rs_lock); 180 } 181 if (!list_empty(&rs->rs_recv_queue) || 182 !list_empty(&rs->rs_notify_queue)) 183 mask |= (POLLIN | POLLRDNORM); 184 if (rs->rs_snd_bytes < rds_sk_sndbuf(rs)) 185 mask |= (POLLOUT | POLLWRNORM); 186 read_unlock_irqrestore(&rs->rs_recv_lock, flags); 187 188 /* clear state any time we wake a seen-congested socket */ 189 if (mask) 190 rs->rs_seen_congestion = 0; 191 192 return mask; 193 } 194 195 static int rds_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 196 { 197 return -ENOIOCTLCMD; 198 } 199 200 static int rds_cancel_sent_to(struct rds_sock *rs, char __user *optval, 201 int len) 202 { 203 struct sockaddr_in sin; 204 int ret = 0; 205 206 /* racing with another thread binding seems ok here */ 207 if (rs->rs_bound_addr == 0) { 208 ret = -ENOTCONN; /* XXX not a great errno */ 209 goto out; 210 } 211 212 if (len < sizeof(struct sockaddr_in)) { 213 ret = -EINVAL; 214 goto out; 215 } 216 217 if (copy_from_user(&sin, optval, sizeof(sin))) { 218 ret = -EFAULT; 219 goto out; 220 } 221 222 rds_send_drop_to(rs, &sin); 223 out: 224 return ret; 225 } 226 227 static int rds_set_bool_option(unsigned char *optvar, char __user *optval, 228 int optlen) 229 { 230 int value; 231 232 if (optlen < sizeof(int)) 233 return -EINVAL; 234 if (get_user(value, (int __user *) optval)) 235 return -EFAULT; 236 *optvar = !!value; 237 return 0; 238 } 239 240 static int rds_cong_monitor(struct rds_sock *rs, char __user *optval, 241 int optlen) 242 { 243 int ret; 244 245 ret = rds_set_bool_option(&rs->rs_cong_monitor, optval, optlen); 246 if (ret == 0) { 247 if (rs->rs_cong_monitor) { 248 rds_cong_add_socket(rs); 249 } else { 250 rds_cong_remove_socket(rs); 251 rs->rs_cong_mask = 0; 252 rs->rs_cong_notify = 0; 253 } 254 } 255 return ret; 256 } 257 258 static int rds_set_transport(struct rds_sock *rs, char __user *optval, 259 int optlen) 260 { 261 int t_type; 262 263 if (rs->rs_transport) 264 return -EOPNOTSUPP; /* previously attached to transport */ 265 266 if (optlen != sizeof(int)) 267 return -EINVAL; 268 269 if (copy_from_user(&t_type, (int __user *)optval, sizeof(t_type))) 270 return -EFAULT; 271 272 if (t_type < 0 || t_type >= RDS_TRANS_COUNT) 273 return -EINVAL; 274 275 rs->rs_transport = rds_trans_get(t_type); 276 277 return rs->rs_transport ? 0 : -ENOPROTOOPT; 278 } 279 280 static int rds_enable_recvtstamp(struct sock *sk, char __user *optval, 281 int optlen) 282 { 283 int val, valbool; 284 285 if (optlen != sizeof(int)) 286 return -EFAULT; 287 288 if (get_user(val, (int __user *)optval)) 289 return -EFAULT; 290 291 valbool = val ? 1 : 0; 292 293 if (valbool) 294 sock_set_flag(sk, SOCK_RCVTSTAMP); 295 else 296 sock_reset_flag(sk, SOCK_RCVTSTAMP); 297 298 return 0; 299 } 300 301 static int rds_recv_track_latency(struct rds_sock *rs, char __user *optval, 302 int optlen) 303 { 304 struct rds_rx_trace_so trace; 305 int i; 306 307 if (optlen != sizeof(struct rds_rx_trace_so)) 308 return -EFAULT; 309 310 if (copy_from_user(&trace, optval, sizeof(trace))) 311 return -EFAULT; 312 313 if (trace.rx_traces > RDS_MSG_RX_DGRAM_TRACE_MAX) 314 return -EFAULT; 315 316 rs->rs_rx_traces = trace.rx_traces; 317 for (i = 0; i < rs->rs_rx_traces; i++) { 318 if (trace.rx_trace_pos[i] > RDS_MSG_RX_DGRAM_TRACE_MAX) { 319 rs->rs_rx_traces = 0; 320 return -EFAULT; 321 } 322 rs->rs_rx_trace[i] = trace.rx_trace_pos[i]; 323 } 324 325 return 0; 326 } 327 328 static int rds_setsockopt(struct socket *sock, int level, int optname, 329 char __user *optval, unsigned int optlen) 330 { 331 struct rds_sock *rs = rds_sk_to_rs(sock->sk); 332 int ret; 333 334 if (level != SOL_RDS) { 335 ret = -ENOPROTOOPT; 336 goto out; 337 } 338 339 switch (optname) { 340 case RDS_CANCEL_SENT_TO: 341 ret = rds_cancel_sent_to(rs, optval, optlen); 342 break; 343 case RDS_GET_MR: 344 ret = rds_get_mr(rs, optval, optlen); 345 break; 346 case RDS_GET_MR_FOR_DEST: 347 ret = rds_get_mr_for_dest(rs, optval, optlen); 348 break; 349 case RDS_FREE_MR: 350 ret = rds_free_mr(rs, optval, optlen); 351 break; 352 case RDS_RECVERR: 353 ret = rds_set_bool_option(&rs->rs_recverr, optval, optlen); 354 break; 355 case RDS_CONG_MONITOR: 356 ret = rds_cong_monitor(rs, optval, optlen); 357 break; 358 case SO_RDS_TRANSPORT: 359 lock_sock(sock->sk); 360 ret = rds_set_transport(rs, optval, optlen); 361 release_sock(sock->sk); 362 break; 363 case SO_TIMESTAMP: 364 lock_sock(sock->sk); 365 ret = rds_enable_recvtstamp(sock->sk, optval, optlen); 366 release_sock(sock->sk); 367 break; 368 case SO_RDS_MSG_RXPATH_LATENCY: 369 ret = rds_recv_track_latency(rs, optval, optlen); 370 break; 371 default: 372 ret = -ENOPROTOOPT; 373 } 374 out: 375 return ret; 376 } 377 378 static int rds_getsockopt(struct socket *sock, int level, int optname, 379 char __user *optval, int __user *optlen) 380 { 381 struct rds_sock *rs = rds_sk_to_rs(sock->sk); 382 int ret = -ENOPROTOOPT, len; 383 int trans; 384 385 if (level != SOL_RDS) 386 goto out; 387 388 if (get_user(len, optlen)) { 389 ret = -EFAULT; 390 goto out; 391 } 392 393 switch (optname) { 394 case RDS_INFO_FIRST ... RDS_INFO_LAST: 395 ret = rds_info_getsockopt(sock, optname, optval, 396 optlen); 397 break; 398 399 case RDS_RECVERR: 400 if (len < sizeof(int)) 401 ret = -EINVAL; 402 else 403 if (put_user(rs->rs_recverr, (int __user *) optval) || 404 put_user(sizeof(int), optlen)) 405 ret = -EFAULT; 406 else 407 ret = 0; 408 break; 409 case SO_RDS_TRANSPORT: 410 if (len < sizeof(int)) { 411 ret = -EINVAL; 412 break; 413 } 414 trans = (rs->rs_transport ? rs->rs_transport->t_type : 415 RDS_TRANS_NONE); /* unbound */ 416 if (put_user(trans, (int __user *)optval) || 417 put_user(sizeof(int), optlen)) 418 ret = -EFAULT; 419 else 420 ret = 0; 421 break; 422 default: 423 break; 424 } 425 426 out: 427 return ret; 428 429 } 430 431 static int rds_connect(struct socket *sock, struct sockaddr *uaddr, 432 int addr_len, int flags) 433 { 434 struct sock *sk = sock->sk; 435 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr; 436 struct rds_sock *rs = rds_sk_to_rs(sk); 437 int ret = 0; 438 439 lock_sock(sk); 440 441 if (addr_len != sizeof(struct sockaddr_in)) { 442 ret = -EINVAL; 443 goto out; 444 } 445 446 if (sin->sin_family != AF_INET) { 447 ret = -EAFNOSUPPORT; 448 goto out; 449 } 450 451 if (sin->sin_addr.s_addr == htonl(INADDR_ANY)) { 452 ret = -EDESTADDRREQ; 453 goto out; 454 } 455 456 rs->rs_conn_addr = sin->sin_addr.s_addr; 457 rs->rs_conn_port = sin->sin_port; 458 459 out: 460 release_sock(sk); 461 return ret; 462 } 463 464 static struct proto rds_proto = { 465 .name = "RDS", 466 .owner = THIS_MODULE, 467 .obj_size = sizeof(struct rds_sock), 468 }; 469 470 static const struct proto_ops rds_proto_ops = { 471 .family = AF_RDS, 472 .owner = THIS_MODULE, 473 .release = rds_release, 474 .bind = rds_bind, 475 .connect = rds_connect, 476 .socketpair = sock_no_socketpair, 477 .accept = sock_no_accept, 478 .getname = rds_getname, 479 .poll = rds_poll, 480 .ioctl = rds_ioctl, 481 .listen = sock_no_listen, 482 .shutdown = sock_no_shutdown, 483 .setsockopt = rds_setsockopt, 484 .getsockopt = rds_getsockopt, 485 .sendmsg = rds_sendmsg, 486 .recvmsg = rds_recvmsg, 487 .mmap = sock_no_mmap, 488 .sendpage = sock_no_sendpage, 489 }; 490 491 static void rds_sock_destruct(struct sock *sk) 492 { 493 struct rds_sock *rs = rds_sk_to_rs(sk); 494 495 WARN_ON((&rs->rs_item != rs->rs_item.next || 496 &rs->rs_item != rs->rs_item.prev)); 497 } 498 499 static int __rds_create(struct socket *sock, struct sock *sk, int protocol) 500 { 501 struct rds_sock *rs; 502 503 sock_init_data(sock, sk); 504 sock->ops = &rds_proto_ops; 505 sk->sk_protocol = protocol; 506 sk->sk_destruct = rds_sock_destruct; 507 508 rs = rds_sk_to_rs(sk); 509 spin_lock_init(&rs->rs_lock); 510 rwlock_init(&rs->rs_recv_lock); 511 INIT_LIST_HEAD(&rs->rs_send_queue); 512 INIT_LIST_HEAD(&rs->rs_recv_queue); 513 INIT_LIST_HEAD(&rs->rs_notify_queue); 514 INIT_LIST_HEAD(&rs->rs_cong_list); 515 spin_lock_init(&rs->rs_rdma_lock); 516 rs->rs_rdma_keys = RB_ROOT; 517 rs->rs_rx_traces = 0; 518 519 spin_lock_bh(&rds_sock_lock); 520 list_add_tail(&rs->rs_item, &rds_sock_list); 521 rds_sock_count++; 522 spin_unlock_bh(&rds_sock_lock); 523 524 return 0; 525 } 526 527 static int rds_create(struct net *net, struct socket *sock, int protocol, 528 int kern) 529 { 530 struct sock *sk; 531 532 if (sock->type != SOCK_SEQPACKET || protocol) 533 return -ESOCKTNOSUPPORT; 534 535 sk = sk_alloc(net, AF_RDS, GFP_ATOMIC, &rds_proto, kern); 536 if (!sk) 537 return -ENOMEM; 538 539 return __rds_create(sock, sk, protocol); 540 } 541 542 void rds_sock_addref(struct rds_sock *rs) 543 { 544 sock_hold(rds_rs_to_sk(rs)); 545 } 546 547 void rds_sock_put(struct rds_sock *rs) 548 { 549 sock_put(rds_rs_to_sk(rs)); 550 } 551 552 static const struct net_proto_family rds_family_ops = { 553 .family = AF_RDS, 554 .create = rds_create, 555 .owner = THIS_MODULE, 556 }; 557 558 static void rds_sock_inc_info(struct socket *sock, unsigned int len, 559 struct rds_info_iterator *iter, 560 struct rds_info_lengths *lens) 561 { 562 struct rds_sock *rs; 563 struct rds_incoming *inc; 564 unsigned int total = 0; 565 566 len /= sizeof(struct rds_info_message); 567 568 spin_lock_bh(&rds_sock_lock); 569 570 list_for_each_entry(rs, &rds_sock_list, rs_item) { 571 read_lock(&rs->rs_recv_lock); 572 573 /* XXX too lazy to maintain counts.. */ 574 list_for_each_entry(inc, &rs->rs_recv_queue, i_item) { 575 total++; 576 if (total <= len) 577 rds_inc_info_copy(inc, iter, inc->i_saddr, 578 rs->rs_bound_addr, 1); 579 } 580 581 read_unlock(&rs->rs_recv_lock); 582 } 583 584 spin_unlock_bh(&rds_sock_lock); 585 586 lens->nr = total; 587 lens->each = sizeof(struct rds_info_message); 588 } 589 590 static void rds_sock_info(struct socket *sock, unsigned int len, 591 struct rds_info_iterator *iter, 592 struct rds_info_lengths *lens) 593 { 594 struct rds_info_socket sinfo; 595 struct rds_sock *rs; 596 597 len /= sizeof(struct rds_info_socket); 598 599 spin_lock_bh(&rds_sock_lock); 600 601 if (len < rds_sock_count) 602 goto out; 603 604 list_for_each_entry(rs, &rds_sock_list, rs_item) { 605 sinfo.sndbuf = rds_sk_sndbuf(rs); 606 sinfo.rcvbuf = rds_sk_rcvbuf(rs); 607 sinfo.bound_addr = rs->rs_bound_addr; 608 sinfo.connected_addr = rs->rs_conn_addr; 609 sinfo.bound_port = rs->rs_bound_port; 610 sinfo.connected_port = rs->rs_conn_port; 611 sinfo.inum = sock_i_ino(rds_rs_to_sk(rs)); 612 613 rds_info_copy(iter, &sinfo, sizeof(sinfo)); 614 } 615 616 out: 617 lens->nr = rds_sock_count; 618 lens->each = sizeof(struct rds_info_socket); 619 620 spin_unlock_bh(&rds_sock_lock); 621 } 622 623 static void rds_exit(void) 624 { 625 sock_unregister(rds_family_ops.family); 626 proto_unregister(&rds_proto); 627 rds_conn_exit(); 628 rds_cong_exit(); 629 rds_sysctl_exit(); 630 rds_threads_exit(); 631 rds_stats_exit(); 632 rds_page_exit(); 633 rds_bind_lock_destroy(); 634 rds_info_deregister_func(RDS_INFO_SOCKETS, rds_sock_info); 635 rds_info_deregister_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info); 636 } 637 module_exit(rds_exit); 638 639 u32 rds_gen_num; 640 641 static int rds_init(void) 642 { 643 int ret; 644 645 net_get_random_once(&rds_gen_num, sizeof(rds_gen_num)); 646 647 ret = rds_bind_lock_init(); 648 if (ret) 649 goto out; 650 651 ret = rds_conn_init(); 652 if (ret) 653 goto out_bind; 654 655 ret = rds_threads_init(); 656 if (ret) 657 goto out_conn; 658 ret = rds_sysctl_init(); 659 if (ret) 660 goto out_threads; 661 ret = rds_stats_init(); 662 if (ret) 663 goto out_sysctl; 664 ret = proto_register(&rds_proto, 1); 665 if (ret) 666 goto out_stats; 667 ret = sock_register(&rds_family_ops); 668 if (ret) 669 goto out_proto; 670 671 rds_info_register_func(RDS_INFO_SOCKETS, rds_sock_info); 672 rds_info_register_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info); 673 674 goto out; 675 676 out_proto: 677 proto_unregister(&rds_proto); 678 out_stats: 679 rds_stats_exit(); 680 out_sysctl: 681 rds_sysctl_exit(); 682 out_threads: 683 rds_threads_exit(); 684 out_conn: 685 rds_conn_exit(); 686 rds_cong_exit(); 687 rds_page_exit(); 688 out_bind: 689 rds_bind_lock_destroy(); 690 out: 691 return ret; 692 } 693 module_init(rds_init); 694 695 #define DRV_VERSION "4.0" 696 #define DRV_RELDATE "Feb 12, 2009" 697 698 MODULE_AUTHOR("Oracle Corporation <rds-devel@oss.oracle.com>"); 699 MODULE_DESCRIPTION("RDS: Reliable Datagram Sockets" 700 " v" DRV_VERSION " (" DRV_RELDATE ")"); 701 MODULE_VERSION(DRV_VERSION); 702 MODULE_LICENSE("Dual BSD/GPL"); 703 MODULE_ALIAS_NETPROTO(PF_RDS); 704