1 /* 2 * Copyright (c) 2005-2006 Intel Corporation. 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/completion.h> 34 #include <linux/file.h> 35 #include <linux/mutex.h> 36 #include <linux/poll.h> 37 #include <linux/sched.h> 38 #include <linux/idr.h> 39 #include <linux/in.h> 40 #include <linux/in6.h> 41 #include <linux/miscdevice.h> 42 #include <linux/slab.h> 43 #include <linux/sysctl.h> 44 #include <linux/module.h> 45 46 #include <rdma/rdma_user_cm.h> 47 #include <rdma/ib_marshall.h> 48 #include <rdma/rdma_cm.h> 49 #include <rdma/rdma_cm_ib.h> 50 #include <rdma/ib_addr.h> 51 #include <rdma/ib.h> 52 53 MODULE_AUTHOR("Sean Hefty"); 54 MODULE_DESCRIPTION("RDMA Userspace Connection Manager Access"); 55 MODULE_LICENSE("Dual BSD/GPL"); 56 57 static unsigned int max_backlog = 1024; 58 59 static struct ctl_table_header *ucma_ctl_table_hdr; 60 static struct ctl_table ucma_ctl_table[] = { 61 { 62 .procname = "max_backlog", 63 .data = &max_backlog, 64 .maxlen = sizeof max_backlog, 65 .mode = 0644, 66 .proc_handler = proc_dointvec, 67 }, 68 { } 69 }; 70 71 struct ucma_file { 72 struct mutex mut; 73 struct file *filp; 74 struct list_head ctx_list; 75 struct list_head event_list; 76 wait_queue_head_t poll_wait; 77 }; 78 79 struct ucma_context { 80 int id; 81 struct completion comp; 82 atomic_t ref; 83 int events_reported; 84 int backlog; 85 86 struct ucma_file *file; 87 struct rdma_cm_id *cm_id; 88 u64 uid; 89 90 struct list_head list; 91 struct list_head mc_list; 92 }; 93 94 struct ucma_multicast { 95 struct ucma_context *ctx; 96 int id; 97 int events_reported; 98 99 u64 uid; 100 struct list_head list; 101 struct sockaddr_storage addr; 102 }; 103 104 struct ucma_event { 105 struct ucma_context *ctx; 106 struct ucma_multicast *mc; 107 struct list_head list; 108 struct rdma_cm_id *cm_id; 109 struct rdma_ucm_event_resp resp; 110 }; 111 112 static DEFINE_MUTEX(mut); 113 static DEFINE_IDR(ctx_idr); 114 static DEFINE_IDR(multicast_idr); 115 116 static inline struct ucma_context *_ucma_find_context(int id, 117 struct ucma_file *file) 118 { 119 struct ucma_context *ctx; 120 121 ctx = idr_find(&ctx_idr, id); 122 if (!ctx) 123 ctx = ERR_PTR(-ENOENT); 124 else if (ctx->file != file) 125 ctx = ERR_PTR(-EINVAL); 126 return ctx; 127 } 128 129 static struct ucma_context *ucma_get_ctx(struct ucma_file *file, int id) 130 { 131 struct ucma_context *ctx; 132 133 mutex_lock(&mut); 134 ctx = _ucma_find_context(id, file); 135 if (!IS_ERR(ctx)) 136 atomic_inc(&ctx->ref); 137 mutex_unlock(&mut); 138 return ctx; 139 } 140 141 static void ucma_put_ctx(struct ucma_context *ctx) 142 { 143 if (atomic_dec_and_test(&ctx->ref)) 144 complete(&ctx->comp); 145 } 146 147 static struct ucma_context *ucma_alloc_ctx(struct ucma_file *file) 148 { 149 struct ucma_context *ctx; 150 151 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); 152 if (!ctx) 153 return NULL; 154 155 atomic_set(&ctx->ref, 1); 156 init_completion(&ctx->comp); 157 INIT_LIST_HEAD(&ctx->mc_list); 158 ctx->file = file; 159 160 mutex_lock(&mut); 161 ctx->id = idr_alloc(&ctx_idr, ctx, 0, 0, GFP_KERNEL); 162 mutex_unlock(&mut); 163 if (ctx->id < 0) 164 goto error; 165 166 list_add_tail(&ctx->list, &file->ctx_list); 167 return ctx; 168 169 error: 170 kfree(ctx); 171 return NULL; 172 } 173 174 static struct ucma_multicast* ucma_alloc_multicast(struct ucma_context *ctx) 175 { 176 struct ucma_multicast *mc; 177 178 mc = kzalloc(sizeof(*mc), GFP_KERNEL); 179 if (!mc) 180 return NULL; 181 182 mutex_lock(&mut); 183 mc->id = idr_alloc(&multicast_idr, mc, 0, 0, GFP_KERNEL); 184 mutex_unlock(&mut); 185 if (mc->id < 0) 186 goto error; 187 188 mc->ctx = ctx; 189 list_add_tail(&mc->list, &ctx->mc_list); 190 return mc; 191 192 error: 193 kfree(mc); 194 return NULL; 195 } 196 197 static void ucma_copy_conn_event(struct rdma_ucm_conn_param *dst, 198 struct rdma_conn_param *src) 199 { 200 if (src->private_data_len) 201 memcpy(dst->private_data, src->private_data, 202 src->private_data_len); 203 dst->private_data_len = src->private_data_len; 204 dst->responder_resources =src->responder_resources; 205 dst->initiator_depth = src->initiator_depth; 206 dst->flow_control = src->flow_control; 207 dst->retry_count = src->retry_count; 208 dst->rnr_retry_count = src->rnr_retry_count; 209 dst->srq = src->srq; 210 dst->qp_num = src->qp_num; 211 } 212 213 static void ucma_copy_ud_event(struct rdma_ucm_ud_param *dst, 214 struct rdma_ud_param *src) 215 { 216 if (src->private_data_len) 217 memcpy(dst->private_data, src->private_data, 218 src->private_data_len); 219 dst->private_data_len = src->private_data_len; 220 ib_copy_ah_attr_to_user(&dst->ah_attr, &src->ah_attr); 221 dst->qp_num = src->qp_num; 222 dst->qkey = src->qkey; 223 } 224 225 static void ucma_set_event_context(struct ucma_context *ctx, 226 struct rdma_cm_event *event, 227 struct ucma_event *uevent) 228 { 229 uevent->ctx = ctx; 230 switch (event->event) { 231 case RDMA_CM_EVENT_MULTICAST_JOIN: 232 case RDMA_CM_EVENT_MULTICAST_ERROR: 233 uevent->mc = (struct ucma_multicast *) 234 event->param.ud.private_data; 235 uevent->resp.uid = uevent->mc->uid; 236 uevent->resp.id = uevent->mc->id; 237 break; 238 default: 239 uevent->resp.uid = ctx->uid; 240 uevent->resp.id = ctx->id; 241 break; 242 } 243 } 244 245 static int ucma_event_handler(struct rdma_cm_id *cm_id, 246 struct rdma_cm_event *event) 247 { 248 struct ucma_event *uevent; 249 struct ucma_context *ctx = cm_id->context; 250 int ret = 0; 251 252 uevent = kzalloc(sizeof(*uevent), GFP_KERNEL); 253 if (!uevent) 254 return event->event == RDMA_CM_EVENT_CONNECT_REQUEST; 255 256 mutex_lock(&ctx->file->mut); 257 uevent->cm_id = cm_id; 258 ucma_set_event_context(ctx, event, uevent); 259 uevent->resp.event = event->event; 260 uevent->resp.status = event->status; 261 if (cm_id->qp_type == IB_QPT_UD) 262 ucma_copy_ud_event(&uevent->resp.param.ud, &event->param.ud); 263 else 264 ucma_copy_conn_event(&uevent->resp.param.conn, 265 &event->param.conn); 266 267 if (event->event == RDMA_CM_EVENT_CONNECT_REQUEST) { 268 if (!ctx->backlog) { 269 ret = -ENOMEM; 270 kfree(uevent); 271 goto out; 272 } 273 ctx->backlog--; 274 } else if (!ctx->uid || ctx->cm_id != cm_id) { 275 /* 276 * We ignore events for new connections until userspace has set 277 * their context. This can only happen if an error occurs on a 278 * new connection before the user accepts it. This is okay, 279 * since the accept will just fail later. 280 */ 281 kfree(uevent); 282 goto out; 283 } 284 285 list_add_tail(&uevent->list, &ctx->file->event_list); 286 wake_up_interruptible(&ctx->file->poll_wait); 287 out: 288 mutex_unlock(&ctx->file->mut); 289 return ret; 290 } 291 292 static ssize_t ucma_get_event(struct ucma_file *file, const char __user *inbuf, 293 int in_len, int out_len) 294 { 295 struct ucma_context *ctx; 296 struct rdma_ucm_get_event cmd; 297 struct ucma_event *uevent; 298 int ret = 0; 299 300 if (out_len < sizeof uevent->resp) 301 return -ENOSPC; 302 303 if (copy_from_user(&cmd, inbuf, sizeof(cmd))) 304 return -EFAULT; 305 306 mutex_lock(&file->mut); 307 while (list_empty(&file->event_list)) { 308 mutex_unlock(&file->mut); 309 310 if (file->filp->f_flags & O_NONBLOCK) 311 return -EAGAIN; 312 313 if (wait_event_interruptible(file->poll_wait, 314 !list_empty(&file->event_list))) 315 return -ERESTARTSYS; 316 317 mutex_lock(&file->mut); 318 } 319 320 uevent = list_entry(file->event_list.next, struct ucma_event, list); 321 322 if (uevent->resp.event == RDMA_CM_EVENT_CONNECT_REQUEST) { 323 ctx = ucma_alloc_ctx(file); 324 if (!ctx) { 325 ret = -ENOMEM; 326 goto done; 327 } 328 uevent->ctx->backlog++; 329 ctx->cm_id = uevent->cm_id; 330 ctx->cm_id->context = ctx; 331 uevent->resp.id = ctx->id; 332 } 333 334 if (copy_to_user((void __user *)(unsigned long)cmd.response, 335 &uevent->resp, sizeof uevent->resp)) { 336 ret = -EFAULT; 337 goto done; 338 } 339 340 list_del(&uevent->list); 341 uevent->ctx->events_reported++; 342 if (uevent->mc) 343 uevent->mc->events_reported++; 344 kfree(uevent); 345 done: 346 mutex_unlock(&file->mut); 347 return ret; 348 } 349 350 static int ucma_get_qp_type(struct rdma_ucm_create_id *cmd, enum ib_qp_type *qp_type) 351 { 352 switch (cmd->ps) { 353 case RDMA_PS_TCP: 354 *qp_type = IB_QPT_RC; 355 return 0; 356 case RDMA_PS_UDP: 357 case RDMA_PS_IPOIB: 358 *qp_type = IB_QPT_UD; 359 return 0; 360 case RDMA_PS_IB: 361 *qp_type = cmd->qp_type; 362 return 0; 363 default: 364 return -EINVAL; 365 } 366 } 367 368 static ssize_t ucma_create_id(struct ucma_file *file, const char __user *inbuf, 369 int in_len, int out_len) 370 { 371 struct rdma_ucm_create_id cmd; 372 struct rdma_ucm_create_id_resp resp; 373 struct ucma_context *ctx; 374 enum ib_qp_type qp_type; 375 int ret; 376 377 if (out_len < sizeof(resp)) 378 return -ENOSPC; 379 380 if (copy_from_user(&cmd, inbuf, sizeof(cmd))) 381 return -EFAULT; 382 383 ret = ucma_get_qp_type(&cmd, &qp_type); 384 if (ret) 385 return ret; 386 387 mutex_lock(&file->mut); 388 ctx = ucma_alloc_ctx(file); 389 mutex_unlock(&file->mut); 390 if (!ctx) 391 return -ENOMEM; 392 393 ctx->uid = cmd.uid; 394 ctx->cm_id = rdma_create_id(ucma_event_handler, ctx, cmd.ps, qp_type); 395 if (IS_ERR(ctx->cm_id)) { 396 ret = PTR_ERR(ctx->cm_id); 397 goto err1; 398 } 399 400 resp.id = ctx->id; 401 if (copy_to_user((void __user *)(unsigned long)cmd.response, 402 &resp, sizeof(resp))) { 403 ret = -EFAULT; 404 goto err2; 405 } 406 return 0; 407 408 err2: 409 rdma_destroy_id(ctx->cm_id); 410 err1: 411 mutex_lock(&mut); 412 idr_remove(&ctx_idr, ctx->id); 413 mutex_unlock(&mut); 414 kfree(ctx); 415 return ret; 416 } 417 418 static void ucma_cleanup_multicast(struct ucma_context *ctx) 419 { 420 struct ucma_multicast *mc, *tmp; 421 422 mutex_lock(&mut); 423 list_for_each_entry_safe(mc, tmp, &ctx->mc_list, list) { 424 list_del(&mc->list); 425 idr_remove(&multicast_idr, mc->id); 426 kfree(mc); 427 } 428 mutex_unlock(&mut); 429 } 430 431 static void ucma_cleanup_mc_events(struct ucma_multicast *mc) 432 { 433 struct ucma_event *uevent, *tmp; 434 435 list_for_each_entry_safe(uevent, tmp, &mc->ctx->file->event_list, list) { 436 if (uevent->mc != mc) 437 continue; 438 439 list_del(&uevent->list); 440 kfree(uevent); 441 } 442 } 443 444 /* 445 * We cannot hold file->mut when calling rdma_destroy_id() or we can 446 * deadlock. We also acquire file->mut in ucma_event_handler(), and 447 * rdma_destroy_id() will wait until all callbacks have completed. 448 */ 449 static int ucma_free_ctx(struct ucma_context *ctx) 450 { 451 int events_reported; 452 struct ucma_event *uevent, *tmp; 453 LIST_HEAD(list); 454 455 /* No new events will be generated after destroying the id. */ 456 rdma_destroy_id(ctx->cm_id); 457 458 ucma_cleanup_multicast(ctx); 459 460 /* Cleanup events not yet reported to the user. */ 461 mutex_lock(&ctx->file->mut); 462 list_for_each_entry_safe(uevent, tmp, &ctx->file->event_list, list) { 463 if (uevent->ctx == ctx) 464 list_move_tail(&uevent->list, &list); 465 } 466 list_del(&ctx->list); 467 mutex_unlock(&ctx->file->mut); 468 469 list_for_each_entry_safe(uevent, tmp, &list, list) { 470 list_del(&uevent->list); 471 if (uevent->resp.event == RDMA_CM_EVENT_CONNECT_REQUEST) 472 rdma_destroy_id(uevent->cm_id); 473 kfree(uevent); 474 } 475 476 events_reported = ctx->events_reported; 477 kfree(ctx); 478 return events_reported; 479 } 480 481 static ssize_t ucma_destroy_id(struct ucma_file *file, const char __user *inbuf, 482 int in_len, int out_len) 483 { 484 struct rdma_ucm_destroy_id cmd; 485 struct rdma_ucm_destroy_id_resp resp; 486 struct ucma_context *ctx; 487 int ret = 0; 488 489 if (out_len < sizeof(resp)) 490 return -ENOSPC; 491 492 if (copy_from_user(&cmd, inbuf, sizeof(cmd))) 493 return -EFAULT; 494 495 mutex_lock(&mut); 496 ctx = _ucma_find_context(cmd.id, file); 497 if (!IS_ERR(ctx)) 498 idr_remove(&ctx_idr, ctx->id); 499 mutex_unlock(&mut); 500 501 if (IS_ERR(ctx)) 502 return PTR_ERR(ctx); 503 504 ucma_put_ctx(ctx); 505 wait_for_completion(&ctx->comp); 506 resp.events_reported = ucma_free_ctx(ctx); 507 508 if (copy_to_user((void __user *)(unsigned long)cmd.response, 509 &resp, sizeof(resp))) 510 ret = -EFAULT; 511 512 return ret; 513 } 514 515 static ssize_t ucma_bind_ip(struct ucma_file *file, const char __user *inbuf, 516 int in_len, int out_len) 517 { 518 struct rdma_ucm_bind_ip cmd; 519 struct ucma_context *ctx; 520 int ret; 521 522 if (copy_from_user(&cmd, inbuf, sizeof(cmd))) 523 return -EFAULT; 524 525 ctx = ucma_get_ctx(file, cmd.id); 526 if (IS_ERR(ctx)) 527 return PTR_ERR(ctx); 528 529 ret = rdma_bind_addr(ctx->cm_id, (struct sockaddr *) &cmd.addr); 530 ucma_put_ctx(ctx); 531 return ret; 532 } 533 534 static ssize_t ucma_bind(struct ucma_file *file, const char __user *inbuf, 535 int in_len, int out_len) 536 { 537 struct rdma_ucm_bind cmd; 538 struct sockaddr *addr; 539 struct ucma_context *ctx; 540 int ret; 541 542 if (copy_from_user(&cmd, inbuf, sizeof(cmd))) 543 return -EFAULT; 544 545 addr = (struct sockaddr *) &cmd.addr; 546 if (cmd.reserved || !cmd.addr_size || (cmd.addr_size != rdma_addr_size(addr))) 547 return -EINVAL; 548 549 ctx = ucma_get_ctx(file, cmd.id); 550 if (IS_ERR(ctx)) 551 return PTR_ERR(ctx); 552 553 ret = rdma_bind_addr(ctx->cm_id, addr); 554 ucma_put_ctx(ctx); 555 return ret; 556 } 557 558 static ssize_t ucma_resolve_ip(struct ucma_file *file, 559 const char __user *inbuf, 560 int in_len, int out_len) 561 { 562 struct rdma_ucm_resolve_ip cmd; 563 struct ucma_context *ctx; 564 int ret; 565 566 if (copy_from_user(&cmd, inbuf, sizeof(cmd))) 567 return -EFAULT; 568 569 ctx = ucma_get_ctx(file, cmd.id); 570 if (IS_ERR(ctx)) 571 return PTR_ERR(ctx); 572 573 ret = rdma_resolve_addr(ctx->cm_id, (struct sockaddr *) &cmd.src_addr, 574 (struct sockaddr *) &cmd.dst_addr, 575 cmd.timeout_ms); 576 ucma_put_ctx(ctx); 577 return ret; 578 } 579 580 static ssize_t ucma_resolve_addr(struct ucma_file *file, 581 const char __user *inbuf, 582 int in_len, int out_len) 583 { 584 struct rdma_ucm_resolve_addr cmd; 585 struct sockaddr *src, *dst; 586 struct ucma_context *ctx; 587 int ret; 588 589 if (copy_from_user(&cmd, inbuf, sizeof(cmd))) 590 return -EFAULT; 591 592 src = (struct sockaddr *) &cmd.src_addr; 593 dst = (struct sockaddr *) &cmd.dst_addr; 594 if (cmd.reserved || (cmd.src_size && (cmd.src_size != rdma_addr_size(src))) || 595 !cmd.dst_size || (cmd.dst_size != rdma_addr_size(dst))) 596 return -EINVAL; 597 598 ctx = ucma_get_ctx(file, cmd.id); 599 if (IS_ERR(ctx)) 600 return PTR_ERR(ctx); 601 602 ret = rdma_resolve_addr(ctx->cm_id, src, dst, cmd.timeout_ms); 603 ucma_put_ctx(ctx); 604 return ret; 605 } 606 607 static ssize_t ucma_resolve_route(struct ucma_file *file, 608 const char __user *inbuf, 609 int in_len, int out_len) 610 { 611 struct rdma_ucm_resolve_route cmd; 612 struct ucma_context *ctx; 613 int ret; 614 615 if (copy_from_user(&cmd, inbuf, sizeof(cmd))) 616 return -EFAULT; 617 618 ctx = ucma_get_ctx(file, cmd.id); 619 if (IS_ERR(ctx)) 620 return PTR_ERR(ctx); 621 622 ret = rdma_resolve_route(ctx->cm_id, cmd.timeout_ms); 623 ucma_put_ctx(ctx); 624 return ret; 625 } 626 627 static void ucma_copy_ib_route(struct rdma_ucm_query_route_resp *resp, 628 struct rdma_route *route) 629 { 630 struct rdma_dev_addr *dev_addr; 631 632 resp->num_paths = route->num_paths; 633 switch (route->num_paths) { 634 case 0: 635 dev_addr = &route->addr.dev_addr; 636 rdma_addr_get_dgid(dev_addr, 637 (union ib_gid *) &resp->ib_route[0].dgid); 638 rdma_addr_get_sgid(dev_addr, 639 (union ib_gid *) &resp->ib_route[0].sgid); 640 resp->ib_route[0].pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr)); 641 break; 642 case 2: 643 ib_copy_path_rec_to_user(&resp->ib_route[1], 644 &route->path_rec[1]); 645 /* fall through */ 646 case 1: 647 ib_copy_path_rec_to_user(&resp->ib_route[0], 648 &route->path_rec[0]); 649 break; 650 default: 651 break; 652 } 653 } 654 655 static void ucma_copy_iboe_route(struct rdma_ucm_query_route_resp *resp, 656 struct rdma_route *route) 657 { 658 659 resp->num_paths = route->num_paths; 660 switch (route->num_paths) { 661 case 0: 662 rdma_ip2gid((struct sockaddr *)&route->addr.dst_addr, 663 (union ib_gid *)&resp->ib_route[0].dgid); 664 rdma_ip2gid((struct sockaddr *)&route->addr.src_addr, 665 (union ib_gid *)&resp->ib_route[0].sgid); 666 resp->ib_route[0].pkey = cpu_to_be16(0xffff); 667 break; 668 case 2: 669 ib_copy_path_rec_to_user(&resp->ib_route[1], 670 &route->path_rec[1]); 671 /* fall through */ 672 case 1: 673 ib_copy_path_rec_to_user(&resp->ib_route[0], 674 &route->path_rec[0]); 675 break; 676 default: 677 break; 678 } 679 } 680 681 static void ucma_copy_iw_route(struct rdma_ucm_query_route_resp *resp, 682 struct rdma_route *route) 683 { 684 struct rdma_dev_addr *dev_addr; 685 686 dev_addr = &route->addr.dev_addr; 687 rdma_addr_get_dgid(dev_addr, (union ib_gid *) &resp->ib_route[0].dgid); 688 rdma_addr_get_sgid(dev_addr, (union ib_gid *) &resp->ib_route[0].sgid); 689 } 690 691 static ssize_t ucma_query_route(struct ucma_file *file, 692 const char __user *inbuf, 693 int in_len, int out_len) 694 { 695 struct rdma_ucm_query cmd; 696 struct rdma_ucm_query_route_resp resp; 697 struct ucma_context *ctx; 698 struct sockaddr *addr; 699 int ret = 0; 700 701 if (out_len < sizeof(resp)) 702 return -ENOSPC; 703 704 if (copy_from_user(&cmd, inbuf, sizeof(cmd))) 705 return -EFAULT; 706 707 ctx = ucma_get_ctx(file, cmd.id); 708 if (IS_ERR(ctx)) 709 return PTR_ERR(ctx); 710 711 memset(&resp, 0, sizeof resp); 712 addr = (struct sockaddr *) &ctx->cm_id->route.addr.src_addr; 713 memcpy(&resp.src_addr, addr, addr->sa_family == AF_INET ? 714 sizeof(struct sockaddr_in) : 715 sizeof(struct sockaddr_in6)); 716 addr = (struct sockaddr *) &ctx->cm_id->route.addr.dst_addr; 717 memcpy(&resp.dst_addr, addr, addr->sa_family == AF_INET ? 718 sizeof(struct sockaddr_in) : 719 sizeof(struct sockaddr_in6)); 720 if (!ctx->cm_id->device) 721 goto out; 722 723 resp.node_guid = (__force __u64) ctx->cm_id->device->node_guid; 724 resp.port_num = ctx->cm_id->port_num; 725 726 if (rdma_cap_ib_sa(ctx->cm_id->device, ctx->cm_id->port_num)) 727 ucma_copy_ib_route(&resp, &ctx->cm_id->route); 728 else if (rdma_protocol_roce(ctx->cm_id->device, ctx->cm_id->port_num)) 729 ucma_copy_iboe_route(&resp, &ctx->cm_id->route); 730 else if (rdma_protocol_iwarp(ctx->cm_id->device, ctx->cm_id->port_num)) 731 ucma_copy_iw_route(&resp, &ctx->cm_id->route); 732 733 out: 734 if (copy_to_user((void __user *)(unsigned long)cmd.response, 735 &resp, sizeof(resp))) 736 ret = -EFAULT; 737 738 ucma_put_ctx(ctx); 739 return ret; 740 } 741 742 static void ucma_query_device_addr(struct rdma_cm_id *cm_id, 743 struct rdma_ucm_query_addr_resp *resp) 744 { 745 if (!cm_id->device) 746 return; 747 748 resp->node_guid = (__force __u64) cm_id->device->node_guid; 749 resp->port_num = cm_id->port_num; 750 resp->pkey = (__force __u16) cpu_to_be16( 751 ib_addr_get_pkey(&cm_id->route.addr.dev_addr)); 752 } 753 754 static ssize_t ucma_query_addr(struct ucma_context *ctx, 755 void __user *response, int out_len) 756 { 757 struct rdma_ucm_query_addr_resp resp; 758 struct sockaddr *addr; 759 int ret = 0; 760 761 if (out_len < sizeof(resp)) 762 return -ENOSPC; 763 764 memset(&resp, 0, sizeof resp); 765 766 addr = (struct sockaddr *) &ctx->cm_id->route.addr.src_addr; 767 resp.src_size = rdma_addr_size(addr); 768 memcpy(&resp.src_addr, addr, resp.src_size); 769 770 addr = (struct sockaddr *) &ctx->cm_id->route.addr.dst_addr; 771 resp.dst_size = rdma_addr_size(addr); 772 memcpy(&resp.dst_addr, addr, resp.dst_size); 773 774 ucma_query_device_addr(ctx->cm_id, &resp); 775 776 if (copy_to_user(response, &resp, sizeof(resp))) 777 ret = -EFAULT; 778 779 return ret; 780 } 781 782 static ssize_t ucma_query_path(struct ucma_context *ctx, 783 void __user *response, int out_len) 784 { 785 struct rdma_ucm_query_path_resp *resp; 786 int i, ret = 0; 787 788 if (out_len < sizeof(*resp)) 789 return -ENOSPC; 790 791 resp = kzalloc(out_len, GFP_KERNEL); 792 if (!resp) 793 return -ENOMEM; 794 795 resp->num_paths = ctx->cm_id->route.num_paths; 796 for (i = 0, out_len -= sizeof(*resp); 797 i < resp->num_paths && out_len > sizeof(struct ib_path_rec_data); 798 i++, out_len -= sizeof(struct ib_path_rec_data)) { 799 800 resp->path_data[i].flags = IB_PATH_GMP | IB_PATH_PRIMARY | 801 IB_PATH_BIDIRECTIONAL; 802 ib_sa_pack_path(&ctx->cm_id->route.path_rec[i], 803 &resp->path_data[i].path_rec); 804 } 805 806 if (copy_to_user(response, resp, 807 sizeof(*resp) + (i * sizeof(struct ib_path_rec_data)))) 808 ret = -EFAULT; 809 810 kfree(resp); 811 return ret; 812 } 813 814 static ssize_t ucma_query_gid(struct ucma_context *ctx, 815 void __user *response, int out_len) 816 { 817 struct rdma_ucm_query_addr_resp resp; 818 struct sockaddr_ib *addr; 819 int ret = 0; 820 821 if (out_len < sizeof(resp)) 822 return -ENOSPC; 823 824 memset(&resp, 0, sizeof resp); 825 826 ucma_query_device_addr(ctx->cm_id, &resp); 827 828 addr = (struct sockaddr_ib *) &resp.src_addr; 829 resp.src_size = sizeof(*addr); 830 if (ctx->cm_id->route.addr.src_addr.ss_family == AF_IB) { 831 memcpy(addr, &ctx->cm_id->route.addr.src_addr, resp.src_size); 832 } else { 833 addr->sib_family = AF_IB; 834 addr->sib_pkey = (__force __be16) resp.pkey; 835 rdma_addr_get_sgid(&ctx->cm_id->route.addr.dev_addr, 836 (union ib_gid *) &addr->sib_addr); 837 addr->sib_sid = rdma_get_service_id(ctx->cm_id, (struct sockaddr *) 838 &ctx->cm_id->route.addr.src_addr); 839 } 840 841 addr = (struct sockaddr_ib *) &resp.dst_addr; 842 resp.dst_size = sizeof(*addr); 843 if (ctx->cm_id->route.addr.dst_addr.ss_family == AF_IB) { 844 memcpy(addr, &ctx->cm_id->route.addr.dst_addr, resp.dst_size); 845 } else { 846 addr->sib_family = AF_IB; 847 addr->sib_pkey = (__force __be16) resp.pkey; 848 rdma_addr_get_dgid(&ctx->cm_id->route.addr.dev_addr, 849 (union ib_gid *) &addr->sib_addr); 850 addr->sib_sid = rdma_get_service_id(ctx->cm_id, (struct sockaddr *) 851 &ctx->cm_id->route.addr.dst_addr); 852 } 853 854 if (copy_to_user(response, &resp, sizeof(resp))) 855 ret = -EFAULT; 856 857 return ret; 858 } 859 860 static ssize_t ucma_query(struct ucma_file *file, 861 const char __user *inbuf, 862 int in_len, int out_len) 863 { 864 struct rdma_ucm_query cmd; 865 struct ucma_context *ctx; 866 void __user *response; 867 int ret; 868 869 if (copy_from_user(&cmd, inbuf, sizeof(cmd))) 870 return -EFAULT; 871 872 response = (void __user *)(unsigned long) cmd.response; 873 ctx = ucma_get_ctx(file, cmd.id); 874 if (IS_ERR(ctx)) 875 return PTR_ERR(ctx); 876 877 switch (cmd.option) { 878 case RDMA_USER_CM_QUERY_ADDR: 879 ret = ucma_query_addr(ctx, response, out_len); 880 break; 881 case RDMA_USER_CM_QUERY_PATH: 882 ret = ucma_query_path(ctx, response, out_len); 883 break; 884 case RDMA_USER_CM_QUERY_GID: 885 ret = ucma_query_gid(ctx, response, out_len); 886 break; 887 default: 888 ret = -ENOSYS; 889 break; 890 } 891 892 ucma_put_ctx(ctx); 893 return ret; 894 } 895 896 static void ucma_copy_conn_param(struct rdma_cm_id *id, 897 struct rdma_conn_param *dst, 898 struct rdma_ucm_conn_param *src) 899 { 900 dst->private_data = src->private_data; 901 dst->private_data_len = src->private_data_len; 902 dst->responder_resources =src->responder_resources; 903 dst->initiator_depth = src->initiator_depth; 904 dst->flow_control = src->flow_control; 905 dst->retry_count = src->retry_count; 906 dst->rnr_retry_count = src->rnr_retry_count; 907 dst->srq = src->srq; 908 dst->qp_num = src->qp_num; 909 dst->qkey = (id->route.addr.src_addr.ss_family == AF_IB) ? src->qkey : 0; 910 } 911 912 static ssize_t ucma_connect(struct ucma_file *file, const char __user *inbuf, 913 int in_len, int out_len) 914 { 915 struct rdma_ucm_connect cmd; 916 struct rdma_conn_param conn_param; 917 struct ucma_context *ctx; 918 int ret; 919 920 if (copy_from_user(&cmd, inbuf, sizeof(cmd))) 921 return -EFAULT; 922 923 if (!cmd.conn_param.valid) 924 return -EINVAL; 925 926 ctx = ucma_get_ctx(file, cmd.id); 927 if (IS_ERR(ctx)) 928 return PTR_ERR(ctx); 929 930 ucma_copy_conn_param(ctx->cm_id, &conn_param, &cmd.conn_param); 931 ret = rdma_connect(ctx->cm_id, &conn_param); 932 ucma_put_ctx(ctx); 933 return ret; 934 } 935 936 static ssize_t ucma_listen(struct ucma_file *file, const char __user *inbuf, 937 int in_len, int out_len) 938 { 939 struct rdma_ucm_listen cmd; 940 struct ucma_context *ctx; 941 int ret; 942 943 if (copy_from_user(&cmd, inbuf, sizeof(cmd))) 944 return -EFAULT; 945 946 ctx = ucma_get_ctx(file, cmd.id); 947 if (IS_ERR(ctx)) 948 return PTR_ERR(ctx); 949 950 ctx->backlog = cmd.backlog > 0 && cmd.backlog < max_backlog ? 951 cmd.backlog : max_backlog; 952 ret = rdma_listen(ctx->cm_id, ctx->backlog); 953 ucma_put_ctx(ctx); 954 return ret; 955 } 956 957 static ssize_t ucma_accept(struct ucma_file *file, const char __user *inbuf, 958 int in_len, int out_len) 959 { 960 struct rdma_ucm_accept cmd; 961 struct rdma_conn_param conn_param; 962 struct ucma_context *ctx; 963 int ret; 964 965 if (copy_from_user(&cmd, inbuf, sizeof(cmd))) 966 return -EFAULT; 967 968 ctx = ucma_get_ctx(file, cmd.id); 969 if (IS_ERR(ctx)) 970 return PTR_ERR(ctx); 971 972 if (cmd.conn_param.valid) { 973 ucma_copy_conn_param(ctx->cm_id, &conn_param, &cmd.conn_param); 974 mutex_lock(&file->mut); 975 ret = rdma_accept(ctx->cm_id, &conn_param); 976 if (!ret) 977 ctx->uid = cmd.uid; 978 mutex_unlock(&file->mut); 979 } else 980 ret = rdma_accept(ctx->cm_id, NULL); 981 982 ucma_put_ctx(ctx); 983 return ret; 984 } 985 986 static ssize_t ucma_reject(struct ucma_file *file, const char __user *inbuf, 987 int in_len, int out_len) 988 { 989 struct rdma_ucm_reject cmd; 990 struct ucma_context *ctx; 991 int ret; 992 993 if (copy_from_user(&cmd, inbuf, sizeof(cmd))) 994 return -EFAULT; 995 996 ctx = ucma_get_ctx(file, cmd.id); 997 if (IS_ERR(ctx)) 998 return PTR_ERR(ctx); 999 1000 ret = rdma_reject(ctx->cm_id, cmd.private_data, cmd.private_data_len); 1001 ucma_put_ctx(ctx); 1002 return ret; 1003 } 1004 1005 static ssize_t ucma_disconnect(struct ucma_file *file, const char __user *inbuf, 1006 int in_len, int out_len) 1007 { 1008 struct rdma_ucm_disconnect cmd; 1009 struct ucma_context *ctx; 1010 int ret; 1011 1012 if (copy_from_user(&cmd, inbuf, sizeof(cmd))) 1013 return -EFAULT; 1014 1015 ctx = ucma_get_ctx(file, cmd.id); 1016 if (IS_ERR(ctx)) 1017 return PTR_ERR(ctx); 1018 1019 ret = rdma_disconnect(ctx->cm_id); 1020 ucma_put_ctx(ctx); 1021 return ret; 1022 } 1023 1024 static ssize_t ucma_init_qp_attr(struct ucma_file *file, 1025 const char __user *inbuf, 1026 int in_len, int out_len) 1027 { 1028 struct rdma_ucm_init_qp_attr cmd; 1029 struct ib_uverbs_qp_attr resp; 1030 struct ucma_context *ctx; 1031 struct ib_qp_attr qp_attr; 1032 int ret; 1033 1034 if (out_len < sizeof(resp)) 1035 return -ENOSPC; 1036 1037 if (copy_from_user(&cmd, inbuf, sizeof(cmd))) 1038 return -EFAULT; 1039 1040 ctx = ucma_get_ctx(file, cmd.id); 1041 if (IS_ERR(ctx)) 1042 return PTR_ERR(ctx); 1043 1044 resp.qp_attr_mask = 0; 1045 memset(&qp_attr, 0, sizeof qp_attr); 1046 qp_attr.qp_state = cmd.qp_state; 1047 ret = rdma_init_qp_attr(ctx->cm_id, &qp_attr, &resp.qp_attr_mask); 1048 if (ret) 1049 goto out; 1050 1051 ib_copy_qp_attr_to_user(&resp, &qp_attr); 1052 if (copy_to_user((void __user *)(unsigned long)cmd.response, 1053 &resp, sizeof(resp))) 1054 ret = -EFAULT; 1055 1056 out: 1057 ucma_put_ctx(ctx); 1058 return ret; 1059 } 1060 1061 static int ucma_set_option_id(struct ucma_context *ctx, int optname, 1062 void *optval, size_t optlen) 1063 { 1064 int ret = 0; 1065 1066 switch (optname) { 1067 case RDMA_OPTION_ID_TOS: 1068 if (optlen != sizeof(u8)) { 1069 ret = -EINVAL; 1070 break; 1071 } 1072 rdma_set_service_type(ctx->cm_id, *((u8 *) optval)); 1073 break; 1074 case RDMA_OPTION_ID_REUSEADDR: 1075 if (optlen != sizeof(int)) { 1076 ret = -EINVAL; 1077 break; 1078 } 1079 ret = rdma_set_reuseaddr(ctx->cm_id, *((int *) optval) ? 1 : 0); 1080 break; 1081 case RDMA_OPTION_ID_AFONLY: 1082 if (optlen != sizeof(int)) { 1083 ret = -EINVAL; 1084 break; 1085 } 1086 ret = rdma_set_afonly(ctx->cm_id, *((int *) optval) ? 1 : 0); 1087 break; 1088 default: 1089 ret = -ENOSYS; 1090 } 1091 1092 return ret; 1093 } 1094 1095 static int ucma_set_ib_path(struct ucma_context *ctx, 1096 struct ib_path_rec_data *path_data, size_t optlen) 1097 { 1098 struct ib_sa_path_rec sa_path; 1099 struct rdma_cm_event event; 1100 int ret; 1101 1102 if (optlen % sizeof(*path_data)) 1103 return -EINVAL; 1104 1105 for (; optlen; optlen -= sizeof(*path_data), path_data++) { 1106 if (path_data->flags == (IB_PATH_GMP | IB_PATH_PRIMARY | 1107 IB_PATH_BIDIRECTIONAL)) 1108 break; 1109 } 1110 1111 if (!optlen) 1112 return -EINVAL; 1113 1114 memset(&sa_path, 0, sizeof(sa_path)); 1115 sa_path.vlan_id = 0xffff; 1116 1117 ib_sa_unpack_path(path_data->path_rec, &sa_path); 1118 ret = rdma_set_ib_paths(ctx->cm_id, &sa_path, 1); 1119 if (ret) 1120 return ret; 1121 1122 memset(&event, 0, sizeof event); 1123 event.event = RDMA_CM_EVENT_ROUTE_RESOLVED; 1124 return ucma_event_handler(ctx->cm_id, &event); 1125 } 1126 1127 static int ucma_set_option_ib(struct ucma_context *ctx, int optname, 1128 void *optval, size_t optlen) 1129 { 1130 int ret; 1131 1132 switch (optname) { 1133 case RDMA_OPTION_IB_PATH: 1134 ret = ucma_set_ib_path(ctx, optval, optlen); 1135 break; 1136 default: 1137 ret = -ENOSYS; 1138 } 1139 1140 return ret; 1141 } 1142 1143 static int ucma_set_option_level(struct ucma_context *ctx, int level, 1144 int optname, void *optval, size_t optlen) 1145 { 1146 int ret; 1147 1148 switch (level) { 1149 case RDMA_OPTION_ID: 1150 ret = ucma_set_option_id(ctx, optname, optval, optlen); 1151 break; 1152 case RDMA_OPTION_IB: 1153 ret = ucma_set_option_ib(ctx, optname, optval, optlen); 1154 break; 1155 default: 1156 ret = -ENOSYS; 1157 } 1158 1159 return ret; 1160 } 1161 1162 static ssize_t ucma_set_option(struct ucma_file *file, const char __user *inbuf, 1163 int in_len, int out_len) 1164 { 1165 struct rdma_ucm_set_option cmd; 1166 struct ucma_context *ctx; 1167 void *optval; 1168 int ret; 1169 1170 if (copy_from_user(&cmd, inbuf, sizeof(cmd))) 1171 return -EFAULT; 1172 1173 ctx = ucma_get_ctx(file, cmd.id); 1174 if (IS_ERR(ctx)) 1175 return PTR_ERR(ctx); 1176 1177 optval = memdup_user((void __user *) (unsigned long) cmd.optval, 1178 cmd.optlen); 1179 if (IS_ERR(optval)) { 1180 ret = PTR_ERR(optval); 1181 goto out; 1182 } 1183 1184 ret = ucma_set_option_level(ctx, cmd.level, cmd.optname, optval, 1185 cmd.optlen); 1186 kfree(optval); 1187 1188 out: 1189 ucma_put_ctx(ctx); 1190 return ret; 1191 } 1192 1193 static ssize_t ucma_notify(struct ucma_file *file, const char __user *inbuf, 1194 int in_len, int out_len) 1195 { 1196 struct rdma_ucm_notify cmd; 1197 struct ucma_context *ctx; 1198 int ret; 1199 1200 if (copy_from_user(&cmd, inbuf, sizeof(cmd))) 1201 return -EFAULT; 1202 1203 ctx = ucma_get_ctx(file, cmd.id); 1204 if (IS_ERR(ctx)) 1205 return PTR_ERR(ctx); 1206 1207 ret = rdma_notify(ctx->cm_id, (enum ib_event_type) cmd.event); 1208 ucma_put_ctx(ctx); 1209 return ret; 1210 } 1211 1212 static ssize_t ucma_process_join(struct ucma_file *file, 1213 struct rdma_ucm_join_mcast *cmd, int out_len) 1214 { 1215 struct rdma_ucm_create_id_resp resp; 1216 struct ucma_context *ctx; 1217 struct ucma_multicast *mc; 1218 struct sockaddr *addr; 1219 int ret; 1220 1221 if (out_len < sizeof(resp)) 1222 return -ENOSPC; 1223 1224 addr = (struct sockaddr *) &cmd->addr; 1225 if (cmd->reserved || !cmd->addr_size || (cmd->addr_size != rdma_addr_size(addr))) 1226 return -EINVAL; 1227 1228 ctx = ucma_get_ctx(file, cmd->id); 1229 if (IS_ERR(ctx)) 1230 return PTR_ERR(ctx); 1231 1232 mutex_lock(&file->mut); 1233 mc = ucma_alloc_multicast(ctx); 1234 if (!mc) { 1235 ret = -ENOMEM; 1236 goto err1; 1237 } 1238 1239 mc->uid = cmd->uid; 1240 memcpy(&mc->addr, addr, cmd->addr_size); 1241 ret = rdma_join_multicast(ctx->cm_id, (struct sockaddr *) &mc->addr, mc); 1242 if (ret) 1243 goto err2; 1244 1245 resp.id = mc->id; 1246 if (copy_to_user((void __user *)(unsigned long) cmd->response, 1247 &resp, sizeof(resp))) { 1248 ret = -EFAULT; 1249 goto err3; 1250 } 1251 1252 mutex_unlock(&file->mut); 1253 ucma_put_ctx(ctx); 1254 return 0; 1255 1256 err3: 1257 rdma_leave_multicast(ctx->cm_id, (struct sockaddr *) &mc->addr); 1258 ucma_cleanup_mc_events(mc); 1259 err2: 1260 mutex_lock(&mut); 1261 idr_remove(&multicast_idr, mc->id); 1262 mutex_unlock(&mut); 1263 list_del(&mc->list); 1264 kfree(mc); 1265 err1: 1266 mutex_unlock(&file->mut); 1267 ucma_put_ctx(ctx); 1268 return ret; 1269 } 1270 1271 static ssize_t ucma_join_ip_multicast(struct ucma_file *file, 1272 const char __user *inbuf, 1273 int in_len, int out_len) 1274 { 1275 struct rdma_ucm_join_ip_mcast cmd; 1276 struct rdma_ucm_join_mcast join_cmd; 1277 1278 if (copy_from_user(&cmd, inbuf, sizeof(cmd))) 1279 return -EFAULT; 1280 1281 join_cmd.response = cmd.response; 1282 join_cmd.uid = cmd.uid; 1283 join_cmd.id = cmd.id; 1284 join_cmd.addr_size = rdma_addr_size((struct sockaddr *) &cmd.addr); 1285 join_cmd.reserved = 0; 1286 memcpy(&join_cmd.addr, &cmd.addr, join_cmd.addr_size); 1287 1288 return ucma_process_join(file, &join_cmd, out_len); 1289 } 1290 1291 static ssize_t ucma_join_multicast(struct ucma_file *file, 1292 const char __user *inbuf, 1293 int in_len, int out_len) 1294 { 1295 struct rdma_ucm_join_mcast cmd; 1296 1297 if (copy_from_user(&cmd, inbuf, sizeof(cmd))) 1298 return -EFAULT; 1299 1300 return ucma_process_join(file, &cmd, out_len); 1301 } 1302 1303 static ssize_t ucma_leave_multicast(struct ucma_file *file, 1304 const char __user *inbuf, 1305 int in_len, int out_len) 1306 { 1307 struct rdma_ucm_destroy_id cmd; 1308 struct rdma_ucm_destroy_id_resp resp; 1309 struct ucma_multicast *mc; 1310 int ret = 0; 1311 1312 if (out_len < sizeof(resp)) 1313 return -ENOSPC; 1314 1315 if (copy_from_user(&cmd, inbuf, sizeof(cmd))) 1316 return -EFAULT; 1317 1318 mutex_lock(&mut); 1319 mc = idr_find(&multicast_idr, cmd.id); 1320 if (!mc) 1321 mc = ERR_PTR(-ENOENT); 1322 else if (mc->ctx->file != file) 1323 mc = ERR_PTR(-EINVAL); 1324 else { 1325 idr_remove(&multicast_idr, mc->id); 1326 atomic_inc(&mc->ctx->ref); 1327 } 1328 mutex_unlock(&mut); 1329 1330 if (IS_ERR(mc)) { 1331 ret = PTR_ERR(mc); 1332 goto out; 1333 } 1334 1335 rdma_leave_multicast(mc->ctx->cm_id, (struct sockaddr *) &mc->addr); 1336 mutex_lock(&mc->ctx->file->mut); 1337 ucma_cleanup_mc_events(mc); 1338 list_del(&mc->list); 1339 mutex_unlock(&mc->ctx->file->mut); 1340 1341 ucma_put_ctx(mc->ctx); 1342 resp.events_reported = mc->events_reported; 1343 kfree(mc); 1344 1345 if (copy_to_user((void __user *)(unsigned long)cmd.response, 1346 &resp, sizeof(resp))) 1347 ret = -EFAULT; 1348 out: 1349 return ret; 1350 } 1351 1352 static void ucma_lock_files(struct ucma_file *file1, struct ucma_file *file2) 1353 { 1354 /* Acquire mutex's based on pointer comparison to prevent deadlock. */ 1355 if (file1 < file2) { 1356 mutex_lock(&file1->mut); 1357 mutex_lock(&file2->mut); 1358 } else { 1359 mutex_lock(&file2->mut); 1360 mutex_lock(&file1->mut); 1361 } 1362 } 1363 1364 static void ucma_unlock_files(struct ucma_file *file1, struct ucma_file *file2) 1365 { 1366 if (file1 < file2) { 1367 mutex_unlock(&file2->mut); 1368 mutex_unlock(&file1->mut); 1369 } else { 1370 mutex_unlock(&file1->mut); 1371 mutex_unlock(&file2->mut); 1372 } 1373 } 1374 1375 static void ucma_move_events(struct ucma_context *ctx, struct ucma_file *file) 1376 { 1377 struct ucma_event *uevent, *tmp; 1378 1379 list_for_each_entry_safe(uevent, tmp, &ctx->file->event_list, list) 1380 if (uevent->ctx == ctx) 1381 list_move_tail(&uevent->list, &file->event_list); 1382 } 1383 1384 static ssize_t ucma_migrate_id(struct ucma_file *new_file, 1385 const char __user *inbuf, 1386 int in_len, int out_len) 1387 { 1388 struct rdma_ucm_migrate_id cmd; 1389 struct rdma_ucm_migrate_resp resp; 1390 struct ucma_context *ctx; 1391 struct fd f; 1392 struct ucma_file *cur_file; 1393 int ret = 0; 1394 1395 if (copy_from_user(&cmd, inbuf, sizeof(cmd))) 1396 return -EFAULT; 1397 1398 /* Get current fd to protect against it being closed */ 1399 f = fdget(cmd.fd); 1400 if (!f.file) 1401 return -ENOENT; 1402 1403 /* Validate current fd and prevent destruction of id. */ 1404 ctx = ucma_get_ctx(f.file->private_data, cmd.id); 1405 if (IS_ERR(ctx)) { 1406 ret = PTR_ERR(ctx); 1407 goto file_put; 1408 } 1409 1410 cur_file = ctx->file; 1411 if (cur_file == new_file) { 1412 resp.events_reported = ctx->events_reported; 1413 goto response; 1414 } 1415 1416 /* 1417 * Migrate events between fd's, maintaining order, and avoiding new 1418 * events being added before existing events. 1419 */ 1420 ucma_lock_files(cur_file, new_file); 1421 mutex_lock(&mut); 1422 1423 list_move_tail(&ctx->list, &new_file->ctx_list); 1424 ucma_move_events(ctx, new_file); 1425 ctx->file = new_file; 1426 resp.events_reported = ctx->events_reported; 1427 1428 mutex_unlock(&mut); 1429 ucma_unlock_files(cur_file, new_file); 1430 1431 response: 1432 if (copy_to_user((void __user *)(unsigned long)cmd.response, 1433 &resp, sizeof(resp))) 1434 ret = -EFAULT; 1435 1436 ucma_put_ctx(ctx); 1437 file_put: 1438 fdput(f); 1439 return ret; 1440 } 1441 1442 static ssize_t (*ucma_cmd_table[])(struct ucma_file *file, 1443 const char __user *inbuf, 1444 int in_len, int out_len) = { 1445 [RDMA_USER_CM_CMD_CREATE_ID] = ucma_create_id, 1446 [RDMA_USER_CM_CMD_DESTROY_ID] = ucma_destroy_id, 1447 [RDMA_USER_CM_CMD_BIND_IP] = ucma_bind_ip, 1448 [RDMA_USER_CM_CMD_RESOLVE_IP] = ucma_resolve_ip, 1449 [RDMA_USER_CM_CMD_RESOLVE_ROUTE] = ucma_resolve_route, 1450 [RDMA_USER_CM_CMD_QUERY_ROUTE] = ucma_query_route, 1451 [RDMA_USER_CM_CMD_CONNECT] = ucma_connect, 1452 [RDMA_USER_CM_CMD_LISTEN] = ucma_listen, 1453 [RDMA_USER_CM_CMD_ACCEPT] = ucma_accept, 1454 [RDMA_USER_CM_CMD_REJECT] = ucma_reject, 1455 [RDMA_USER_CM_CMD_DISCONNECT] = ucma_disconnect, 1456 [RDMA_USER_CM_CMD_INIT_QP_ATTR] = ucma_init_qp_attr, 1457 [RDMA_USER_CM_CMD_GET_EVENT] = ucma_get_event, 1458 [RDMA_USER_CM_CMD_GET_OPTION] = NULL, 1459 [RDMA_USER_CM_CMD_SET_OPTION] = ucma_set_option, 1460 [RDMA_USER_CM_CMD_NOTIFY] = ucma_notify, 1461 [RDMA_USER_CM_CMD_JOIN_IP_MCAST] = ucma_join_ip_multicast, 1462 [RDMA_USER_CM_CMD_LEAVE_MCAST] = ucma_leave_multicast, 1463 [RDMA_USER_CM_CMD_MIGRATE_ID] = ucma_migrate_id, 1464 [RDMA_USER_CM_CMD_QUERY] = ucma_query, 1465 [RDMA_USER_CM_CMD_BIND] = ucma_bind, 1466 [RDMA_USER_CM_CMD_RESOLVE_ADDR] = ucma_resolve_addr, 1467 [RDMA_USER_CM_CMD_JOIN_MCAST] = ucma_join_multicast 1468 }; 1469 1470 static ssize_t ucma_write(struct file *filp, const char __user *buf, 1471 size_t len, loff_t *pos) 1472 { 1473 struct ucma_file *file = filp->private_data; 1474 struct rdma_ucm_cmd_hdr hdr; 1475 ssize_t ret; 1476 1477 if (len < sizeof(hdr)) 1478 return -EINVAL; 1479 1480 if (copy_from_user(&hdr, buf, sizeof(hdr))) 1481 return -EFAULT; 1482 1483 if (hdr.cmd >= ARRAY_SIZE(ucma_cmd_table)) 1484 return -EINVAL; 1485 1486 if (hdr.in + sizeof(hdr) > len) 1487 return -EINVAL; 1488 1489 if (!ucma_cmd_table[hdr.cmd]) 1490 return -ENOSYS; 1491 1492 ret = ucma_cmd_table[hdr.cmd](file, buf + sizeof(hdr), hdr.in, hdr.out); 1493 if (!ret) 1494 ret = len; 1495 1496 return ret; 1497 } 1498 1499 static unsigned int ucma_poll(struct file *filp, struct poll_table_struct *wait) 1500 { 1501 struct ucma_file *file = filp->private_data; 1502 unsigned int mask = 0; 1503 1504 poll_wait(filp, &file->poll_wait, wait); 1505 1506 if (!list_empty(&file->event_list)) 1507 mask = POLLIN | POLLRDNORM; 1508 1509 return mask; 1510 } 1511 1512 /* 1513 * ucma_open() does not need the BKL: 1514 * 1515 * - no global state is referred to; 1516 * - there is no ioctl method to race against; 1517 * - no further module initialization is required for open to work 1518 * after the device is registered. 1519 */ 1520 static int ucma_open(struct inode *inode, struct file *filp) 1521 { 1522 struct ucma_file *file; 1523 1524 file = kmalloc(sizeof *file, GFP_KERNEL); 1525 if (!file) 1526 return -ENOMEM; 1527 1528 INIT_LIST_HEAD(&file->event_list); 1529 INIT_LIST_HEAD(&file->ctx_list); 1530 init_waitqueue_head(&file->poll_wait); 1531 mutex_init(&file->mut); 1532 1533 filp->private_data = file; 1534 file->filp = filp; 1535 1536 return nonseekable_open(inode, filp); 1537 } 1538 1539 static int ucma_close(struct inode *inode, struct file *filp) 1540 { 1541 struct ucma_file *file = filp->private_data; 1542 struct ucma_context *ctx, *tmp; 1543 1544 mutex_lock(&file->mut); 1545 list_for_each_entry_safe(ctx, tmp, &file->ctx_list, list) { 1546 mutex_unlock(&file->mut); 1547 1548 mutex_lock(&mut); 1549 idr_remove(&ctx_idr, ctx->id); 1550 mutex_unlock(&mut); 1551 1552 ucma_free_ctx(ctx); 1553 mutex_lock(&file->mut); 1554 } 1555 mutex_unlock(&file->mut); 1556 kfree(file); 1557 return 0; 1558 } 1559 1560 static const struct file_operations ucma_fops = { 1561 .owner = THIS_MODULE, 1562 .open = ucma_open, 1563 .release = ucma_close, 1564 .write = ucma_write, 1565 .poll = ucma_poll, 1566 .llseek = no_llseek, 1567 }; 1568 1569 static struct miscdevice ucma_misc = { 1570 .minor = MISC_DYNAMIC_MINOR, 1571 .name = "rdma_cm", 1572 .nodename = "infiniband/rdma_cm", 1573 .mode = 0666, 1574 .fops = &ucma_fops, 1575 }; 1576 1577 static ssize_t show_abi_version(struct device *dev, 1578 struct device_attribute *attr, 1579 char *buf) 1580 { 1581 return sprintf(buf, "%d\n", RDMA_USER_CM_ABI_VERSION); 1582 } 1583 static DEVICE_ATTR(abi_version, S_IRUGO, show_abi_version, NULL); 1584 1585 static int __init ucma_init(void) 1586 { 1587 int ret; 1588 1589 ret = misc_register(&ucma_misc); 1590 if (ret) 1591 return ret; 1592 1593 ret = device_create_file(ucma_misc.this_device, &dev_attr_abi_version); 1594 if (ret) { 1595 printk(KERN_ERR "rdma_ucm: couldn't create abi_version attr\n"); 1596 goto err1; 1597 } 1598 1599 ucma_ctl_table_hdr = register_net_sysctl(&init_net, "net/rdma_ucm", ucma_ctl_table); 1600 if (!ucma_ctl_table_hdr) { 1601 printk(KERN_ERR "rdma_ucm: couldn't register sysctl paths\n"); 1602 ret = -ENOMEM; 1603 goto err2; 1604 } 1605 return 0; 1606 err2: 1607 device_remove_file(ucma_misc.this_device, &dev_attr_abi_version); 1608 err1: 1609 misc_deregister(&ucma_misc); 1610 return ret; 1611 } 1612 1613 static void __exit ucma_cleanup(void) 1614 { 1615 unregister_net_sysctl_table(ucma_ctl_table_hdr); 1616 device_remove_file(ucma_misc.this_device, &dev_attr_abi_version); 1617 misc_deregister(&ucma_misc); 1618 idr_destroy(&ctx_idr); 1619 } 1620 1621 module_init(ucma_init); 1622 module_exit(ucma_cleanup); 1623