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