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->sk_sleep, wait); 162 163 poll_wait(file, &rds_poll_waitq, wait); 164 165 read_lock_irqsave(&rs->rs_recv_lock, flags); 166 if (!rs->rs_cong_monitor) { 167 /* When a congestion map was updated, we signal POLLIN for 168 * "historical" reasons. Applications can also poll for 169 * WRBAND instead. */ 170 if (rds_cong_updated_since(&rs->rs_cong_track)) 171 mask |= (POLLIN | POLLRDNORM | POLLWRBAND); 172 } else { 173 spin_lock(&rs->rs_lock); 174 if (rs->rs_cong_notify) 175 mask |= (POLLIN | POLLRDNORM); 176 spin_unlock(&rs->rs_lock); 177 } 178 if (!list_empty(&rs->rs_recv_queue) || 179 !list_empty(&rs->rs_notify_queue)) 180 mask |= (POLLIN | POLLRDNORM); 181 if (rs->rs_snd_bytes < rds_sk_sndbuf(rs)) 182 mask |= (POLLOUT | POLLWRNORM); 183 read_unlock_irqrestore(&rs->rs_recv_lock, flags); 184 185 return mask; 186 } 187 188 static int rds_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 189 { 190 return -ENOIOCTLCMD; 191 } 192 193 static int rds_cancel_sent_to(struct rds_sock *rs, char __user *optval, 194 int len) 195 { 196 struct sockaddr_in sin; 197 int ret = 0; 198 199 /* racing with another thread binding seems ok here */ 200 if (rs->rs_bound_addr == 0) { 201 ret = -ENOTCONN; /* XXX not a great errno */ 202 goto out; 203 } 204 205 if (len < sizeof(struct sockaddr_in)) { 206 ret = -EINVAL; 207 goto out; 208 } 209 210 if (copy_from_user(&sin, optval, sizeof(sin))) { 211 ret = -EFAULT; 212 goto out; 213 } 214 215 rds_send_drop_to(rs, &sin); 216 out: 217 return ret; 218 } 219 220 static int rds_set_bool_option(unsigned char *optvar, char __user *optval, 221 int optlen) 222 { 223 int value; 224 225 if (optlen < sizeof(int)) 226 return -EINVAL; 227 if (get_user(value, (int __user *) optval)) 228 return -EFAULT; 229 *optvar = !!value; 230 return 0; 231 } 232 233 static int rds_cong_monitor(struct rds_sock *rs, char __user *optval, 234 int optlen) 235 { 236 int ret; 237 238 ret = rds_set_bool_option(&rs->rs_cong_monitor, optval, optlen); 239 if (ret == 0) { 240 if (rs->rs_cong_monitor) { 241 rds_cong_add_socket(rs); 242 } else { 243 rds_cong_remove_socket(rs); 244 rs->rs_cong_mask = 0; 245 rs->rs_cong_notify = 0; 246 } 247 } 248 return ret; 249 } 250 251 static int rds_setsockopt(struct socket *sock, int level, int optname, 252 char __user *optval, unsigned int optlen) 253 { 254 struct rds_sock *rs = rds_sk_to_rs(sock->sk); 255 int ret; 256 257 if (level != SOL_RDS) { 258 ret = -ENOPROTOOPT; 259 goto out; 260 } 261 262 switch (optname) { 263 case RDS_CANCEL_SENT_TO: 264 ret = rds_cancel_sent_to(rs, optval, optlen); 265 break; 266 case RDS_GET_MR: 267 ret = rds_get_mr(rs, optval, optlen); 268 break; 269 case RDS_GET_MR_FOR_DEST: 270 ret = rds_get_mr_for_dest(rs, optval, optlen); 271 break; 272 case RDS_FREE_MR: 273 ret = rds_free_mr(rs, optval, optlen); 274 break; 275 case RDS_RECVERR: 276 ret = rds_set_bool_option(&rs->rs_recverr, optval, optlen); 277 break; 278 case RDS_CONG_MONITOR: 279 ret = rds_cong_monitor(rs, optval, optlen); 280 break; 281 default: 282 ret = -ENOPROTOOPT; 283 } 284 out: 285 return ret; 286 } 287 288 static int rds_getsockopt(struct socket *sock, int level, int optname, 289 char __user *optval, int __user *optlen) 290 { 291 struct rds_sock *rs = rds_sk_to_rs(sock->sk); 292 int ret = -ENOPROTOOPT, len; 293 294 if (level != SOL_RDS) 295 goto out; 296 297 if (get_user(len, optlen)) { 298 ret = -EFAULT; 299 goto out; 300 } 301 302 switch (optname) { 303 case RDS_INFO_FIRST ... RDS_INFO_LAST: 304 ret = rds_info_getsockopt(sock, optname, optval, 305 optlen); 306 break; 307 308 case RDS_RECVERR: 309 if (len < sizeof(int)) 310 ret = -EINVAL; 311 else 312 if (put_user(rs->rs_recverr, (int __user *) optval) || 313 put_user(sizeof(int), optlen)) 314 ret = -EFAULT; 315 else 316 ret = 0; 317 break; 318 default: 319 break; 320 } 321 322 out: 323 return ret; 324 325 } 326 327 static int rds_connect(struct socket *sock, struct sockaddr *uaddr, 328 int addr_len, int flags) 329 { 330 struct sock *sk = sock->sk; 331 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr; 332 struct rds_sock *rs = rds_sk_to_rs(sk); 333 int ret = 0; 334 335 lock_sock(sk); 336 337 if (addr_len != sizeof(struct sockaddr_in)) { 338 ret = -EINVAL; 339 goto out; 340 } 341 342 if (sin->sin_family != AF_INET) { 343 ret = -EAFNOSUPPORT; 344 goto out; 345 } 346 347 if (sin->sin_addr.s_addr == htonl(INADDR_ANY)) { 348 ret = -EDESTADDRREQ; 349 goto out; 350 } 351 352 rs->rs_conn_addr = sin->sin_addr.s_addr; 353 rs->rs_conn_port = sin->sin_port; 354 355 out: 356 release_sock(sk); 357 return ret; 358 } 359 360 static struct proto rds_proto = { 361 .name = "RDS", 362 .owner = THIS_MODULE, 363 .obj_size = sizeof(struct rds_sock), 364 }; 365 366 static const struct proto_ops rds_proto_ops = { 367 .family = AF_RDS, 368 .owner = THIS_MODULE, 369 .release = rds_release, 370 .bind = rds_bind, 371 .connect = rds_connect, 372 .socketpair = sock_no_socketpair, 373 .accept = sock_no_accept, 374 .getname = rds_getname, 375 .poll = rds_poll, 376 .ioctl = rds_ioctl, 377 .listen = sock_no_listen, 378 .shutdown = sock_no_shutdown, 379 .setsockopt = rds_setsockopt, 380 .getsockopt = rds_getsockopt, 381 .sendmsg = rds_sendmsg, 382 .recvmsg = rds_recvmsg, 383 .mmap = sock_no_mmap, 384 .sendpage = sock_no_sendpage, 385 }; 386 387 static int __rds_create(struct socket *sock, struct sock *sk, int protocol) 388 { 389 unsigned long flags; 390 struct rds_sock *rs; 391 392 sock_init_data(sock, sk); 393 sock->ops = &rds_proto_ops; 394 sk->sk_protocol = protocol; 395 396 rs = rds_sk_to_rs(sk); 397 spin_lock_init(&rs->rs_lock); 398 rwlock_init(&rs->rs_recv_lock); 399 INIT_LIST_HEAD(&rs->rs_send_queue); 400 INIT_LIST_HEAD(&rs->rs_recv_queue); 401 INIT_LIST_HEAD(&rs->rs_notify_queue); 402 INIT_LIST_HEAD(&rs->rs_cong_list); 403 spin_lock_init(&rs->rs_rdma_lock); 404 rs->rs_rdma_keys = RB_ROOT; 405 406 spin_lock_irqsave(&rds_sock_lock, flags); 407 list_add_tail(&rs->rs_item, &rds_sock_list); 408 rds_sock_count++; 409 spin_unlock_irqrestore(&rds_sock_lock, flags); 410 411 return 0; 412 } 413 414 static int rds_create(struct net *net, struct socket *sock, int protocol, 415 int kern) 416 { 417 struct sock *sk; 418 419 if (sock->type != SOCK_SEQPACKET || protocol) 420 return -ESOCKTNOSUPPORT; 421 422 sk = sk_alloc(net, AF_RDS, GFP_ATOMIC, &rds_proto); 423 if (!sk) 424 return -ENOMEM; 425 426 return __rds_create(sock, sk, protocol); 427 } 428 429 void rds_sock_addref(struct rds_sock *rs) 430 { 431 sock_hold(rds_rs_to_sk(rs)); 432 } 433 434 void rds_sock_put(struct rds_sock *rs) 435 { 436 sock_put(rds_rs_to_sk(rs)); 437 } 438 439 static const struct net_proto_family rds_family_ops = { 440 .family = AF_RDS, 441 .create = rds_create, 442 .owner = THIS_MODULE, 443 }; 444 445 static void rds_sock_inc_info(struct socket *sock, unsigned int len, 446 struct rds_info_iterator *iter, 447 struct rds_info_lengths *lens) 448 { 449 struct rds_sock *rs; 450 struct sock *sk; 451 struct rds_incoming *inc; 452 unsigned long flags; 453 unsigned int total = 0; 454 455 len /= sizeof(struct rds_info_message); 456 457 spin_lock_irqsave(&rds_sock_lock, flags); 458 459 list_for_each_entry(rs, &rds_sock_list, rs_item) { 460 sk = rds_rs_to_sk(rs); 461 read_lock(&rs->rs_recv_lock); 462 463 /* XXX too lazy to maintain counts.. */ 464 list_for_each_entry(inc, &rs->rs_recv_queue, i_item) { 465 total++; 466 if (total <= len) 467 rds_inc_info_copy(inc, iter, inc->i_saddr, 468 rs->rs_bound_addr, 1); 469 } 470 471 read_unlock(&rs->rs_recv_lock); 472 } 473 474 spin_unlock_irqrestore(&rds_sock_lock, flags); 475 476 lens->nr = total; 477 lens->each = sizeof(struct rds_info_message); 478 } 479 480 static void rds_sock_info(struct socket *sock, unsigned int len, 481 struct rds_info_iterator *iter, 482 struct rds_info_lengths *lens) 483 { 484 struct rds_info_socket sinfo; 485 struct rds_sock *rs; 486 unsigned long flags; 487 488 len /= sizeof(struct rds_info_socket); 489 490 spin_lock_irqsave(&rds_sock_lock, flags); 491 492 if (len < rds_sock_count) 493 goto out; 494 495 list_for_each_entry(rs, &rds_sock_list, rs_item) { 496 sinfo.sndbuf = rds_sk_sndbuf(rs); 497 sinfo.rcvbuf = rds_sk_rcvbuf(rs); 498 sinfo.bound_addr = rs->rs_bound_addr; 499 sinfo.connected_addr = rs->rs_conn_addr; 500 sinfo.bound_port = rs->rs_bound_port; 501 sinfo.connected_port = rs->rs_conn_port; 502 sinfo.inum = sock_i_ino(rds_rs_to_sk(rs)); 503 504 rds_info_copy(iter, &sinfo, sizeof(sinfo)); 505 } 506 507 out: 508 lens->nr = rds_sock_count; 509 lens->each = sizeof(struct rds_info_socket); 510 511 spin_unlock_irqrestore(&rds_sock_lock, flags); 512 } 513 514 static void __exit rds_exit(void) 515 { 516 sock_unregister(rds_family_ops.family); 517 proto_unregister(&rds_proto); 518 rds_conn_exit(); 519 rds_cong_exit(); 520 rds_sysctl_exit(); 521 rds_threads_exit(); 522 rds_stats_exit(); 523 rds_page_exit(); 524 rds_info_deregister_func(RDS_INFO_SOCKETS, rds_sock_info); 525 rds_info_deregister_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info); 526 } 527 module_exit(rds_exit); 528 529 static int __init rds_init(void) 530 { 531 int ret; 532 533 ret = rds_conn_init(); 534 if (ret) 535 goto out; 536 ret = rds_threads_init(); 537 if (ret) 538 goto out_conn; 539 ret = rds_sysctl_init(); 540 if (ret) 541 goto out_threads; 542 ret = rds_stats_init(); 543 if (ret) 544 goto out_sysctl; 545 ret = proto_register(&rds_proto, 1); 546 if (ret) 547 goto out_stats; 548 ret = sock_register(&rds_family_ops); 549 if (ret) 550 goto out_proto; 551 552 rds_info_register_func(RDS_INFO_SOCKETS, rds_sock_info); 553 rds_info_register_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info); 554 555 goto out; 556 557 out_proto: 558 proto_unregister(&rds_proto); 559 out_stats: 560 rds_stats_exit(); 561 out_sysctl: 562 rds_sysctl_exit(); 563 out_threads: 564 rds_threads_exit(); 565 out_conn: 566 rds_conn_exit(); 567 rds_cong_exit(); 568 rds_page_exit(); 569 out: 570 return ret; 571 } 572 module_init(rds_init); 573 574 #define DRV_VERSION "4.0" 575 #define DRV_RELDATE "Feb 12, 2009" 576 577 MODULE_AUTHOR("Oracle Corporation <rds-devel@oss.oracle.com>"); 578 MODULE_DESCRIPTION("RDS: Reliable Datagram Sockets" 579 " v" DRV_VERSION " (" DRV_RELDATE ")"); 580 MODULE_VERSION(DRV_VERSION); 581 MODULE_LICENSE("Dual BSD/GPL"); 582 MODULE_ALIAS_NETPROTO(PF_RDS); 583