drivers/infiniband/core/ucma.c

   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         struct rdma_dev_addr *dev_addr;
 659         struct net_device *dev;
 660         u16 vid = 0;
 661
 662         resp->num_paths = route->num_paths;
 663         switch (route->num_paths) {
 664         case 0:
 665                 dev_addr = &route->addr.dev_addr;
 666                 dev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
 667                         if (dev) {
 668                                 vid = rdma_vlan_dev_vlan_id(dev);
 669                                 dev_put(dev);
 670                         }
 671
 672                 iboe_mac_vlan_to_ll((union ib_gid *) &resp->ib_route[0].dgid,
 673                                     dev_addr->dst_dev_addr, vid);
 674                 iboe_addr_get_sgid(dev_addr,
 675                                    (union ib_gid *) &resp->ib_route[0].sgid);
 676                 resp->ib_route[0].pkey = cpu_to_be16(0xffff);
 677                 break;
 678         case 2:
 679                 ib_copy_path_rec_to_user(&resp->ib_route[1],
 680                                          &route->path_rec[1]);
 681                 /* fall through */
 682         case 1:
 683                 ib_copy_path_rec_to_user(&resp->ib_route[0],
 684                                          &route->path_rec[0]);
 685                 break;
 686         default:
 687                 break;
 688         }
 689 }
 690
 691 static void ucma_copy_iw_route(struct rdma_ucm_query_route_resp *resp,
 692                                struct rdma_route *route)
 693 {
 694         struct rdma_dev_addr *dev_addr;
 695
 696         dev_addr = &route->addr.dev_addr;
 697         rdma_addr_get_dgid(dev_addr, (union ib_gid *) &resp->ib_route[0].dgid);
 698         rdma_addr_get_sgid(dev_addr, (union ib_gid *) &resp->ib_route[0].sgid);
 699 }
 700
 701 static ssize_t ucma_query_route(struct ucma_file *file,
 702                                 const char __user *inbuf,
 703                                 int in_len, int out_len)
 704 {
 705         struct rdma_ucm_query cmd;
 706         struct rdma_ucm_query_route_resp resp;
 707         struct ucma_context *ctx;
 708         struct sockaddr *addr;
 709         int ret = 0;
 710
 711         if (out_len < sizeof(resp))
 712                 return -ENOSPC;
 713
 714         if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
 715                 return -EFAULT;
 716
 717         ctx = ucma_get_ctx(file, cmd.id);
 718         if (IS_ERR(ctx))
 719                 return PTR_ERR(ctx);
 720
 721         memset(&resp, 0, sizeof resp);
 722         addr = (struct sockaddr *) &ctx->cm_id->route.addr.src_addr;
 723         memcpy(&resp.src_addr, addr, addr->sa_family == AF_INET ?
 724                                      sizeof(struct sockaddr_in) :
 725                                      sizeof(struct sockaddr_in6));
 726         addr = (struct sockaddr *) &ctx->cm_id->route.addr.dst_addr;
 727         memcpy(&resp.dst_addr, addr, addr->sa_family == AF_INET ?
 728                                      sizeof(struct sockaddr_in) :
 729                                      sizeof(struct sockaddr_in6));
 730         if (!ctx->cm_id->device)
 731                 goto out;
 732
 733         resp.node_guid = (__force __u64) ctx->cm_id->device->node_guid;
 734         resp.port_num = ctx->cm_id->port_num;
 735         switch (rdma_node_get_transport(ctx->cm_id->device->node_type)) {
 736         case RDMA_TRANSPORT_IB:
 737                 switch (rdma_port_get_link_layer(ctx->cm_id->device,
 738                         ctx->cm_id->port_num)) {
 739                 case IB_LINK_LAYER_INFINIBAND:
 740                         ucma_copy_ib_route(&resp, &ctx->cm_id->route);
 741                         break;
 742                 case IB_LINK_LAYER_ETHERNET:
 743                         ucma_copy_iboe_route(&resp, &ctx->cm_id->route);
 744                         break;
 745                 default:
 746                         break;
 747                 }
 748                 break;
 749         case RDMA_TRANSPORT_IWARP:
 750                 ucma_copy_iw_route(&resp, &ctx->cm_id->route);
 751                 break;
 752         default:
 753                 break;
 754         }
 755
 756 out:
 757         if (copy_to_user((void __user *)(unsigned long)cmd.response,
 758                          &resp, sizeof(resp)))
 759                 ret = -EFAULT;
 760
 761         ucma_put_ctx(ctx);
 762         return ret;
 763 }
 764
 765 static void ucma_query_device_addr(struct rdma_cm_id *cm_id,
 766                                    struct rdma_ucm_query_addr_resp *resp)
 767 {
 768         if (!cm_id->device)
 769                 return;
 770
 771         resp->node_guid = (__force __u64) cm_id->device->node_guid;
 772         resp->port_num = cm_id->port_num;
 773         resp->pkey = (__force __u16) cpu_to_be16(
 774                      ib_addr_get_pkey(&cm_id->route.addr.dev_addr));
 775 }
 776
 777 static ssize_t ucma_query_addr(struct ucma_context *ctx,
 778                                void __user *response, int out_len)
 779 {
 780         struct rdma_ucm_query_addr_resp resp;
 781         struct sockaddr *addr;
 782         int ret = 0;
 783
 784         if (out_len < sizeof(resp))
 785                 return -ENOSPC;
 786
 787         memset(&resp, 0, sizeof resp);
 788
 789         addr = (struct sockaddr *) &ctx->cm_id->route.addr.src_addr;
 790         resp.src_size = rdma_addr_size(addr);
 791         memcpy(&resp.src_addr, addr, resp.src_size);
 792
 793         addr = (struct sockaddr *) &ctx->cm_id->route.addr.dst_addr;
 794         resp.dst_size = rdma_addr_size(addr);
 795         memcpy(&resp.dst_addr, addr, resp.dst_size);
 796
 797         ucma_query_device_addr(ctx->cm_id, &resp);
 798
 799         if (copy_to_user(response, &resp, sizeof(resp)))
 800                 ret = -EFAULT;
 801
 802         return ret;
 803 }
 804
 805 static ssize_t ucma_query_path(struct ucma_context *ctx,
 806                                void __user *response, int out_len)
 807 {
 808         struct rdma_ucm_query_path_resp *resp;
 809         int i, ret = 0;
 810
 811         if (out_len < sizeof(*resp))
 812                 return -ENOSPC;
 813
 814         resp = kzalloc(out_len, GFP_KERNEL);
 815         if (!resp)
 816                 return -ENOMEM;
 817
 818         resp->num_paths = ctx->cm_id->route.num_paths;
 819         for (i = 0, out_len -= sizeof(*resp);
 820              i < resp->num_paths && out_len > sizeof(struct ib_path_rec_data);
 821              i++, out_len -= sizeof(struct ib_path_rec_data)) {
 822
 823                 resp->path_data[i].flags = IB_PATH_GMP | IB_PATH_PRIMARY |
 824                                            IB_PATH_BIDIRECTIONAL;
 825                 ib_sa_pack_path(&ctx->cm_id->route.path_rec[i],
 826                                 &resp->path_data[i].path_rec);
 827         }
 828
 829         if (copy_to_user(response, resp,
 830                          sizeof(*resp) + (i * sizeof(struct ib_path_rec_data))))
 831                 ret = -EFAULT;
 832
 833         kfree(resp);
 834         return ret;
 835 }
 836
 837 static ssize_t ucma_query_gid(struct ucma_context *ctx,
 838                               void __user *response, int out_len)
 839 {
 840         struct rdma_ucm_query_addr_resp resp;
 841         struct sockaddr_ib *addr;
 842         int ret = 0;
 843
 844         if (out_len < sizeof(resp))
 845                 return -ENOSPC;
 846
 847         memset(&resp, 0, sizeof resp);
 848
 849         ucma_query_device_addr(ctx->cm_id, &resp);
 850
 851         addr = (struct sockaddr_ib *) &resp.src_addr;
 852         resp.src_size = sizeof(*addr);
 853         if (ctx->cm_id->route.addr.src_addr.ss_family == AF_IB) {
 854                 memcpy(addr, &ctx->cm_id->route.addr.src_addr, resp.src_size);
 855         } else {
 856                 addr->sib_family = AF_IB;
 857                 addr->sib_pkey = (__force __be16) resp.pkey;
 858                 rdma_addr_get_sgid(&ctx->cm_id->route.addr.dev_addr,
 859                                    (union ib_gid *) &addr->sib_addr);
 860                 addr->sib_sid = rdma_get_service_id(ctx->cm_id, (struct sockaddr *)
 861                                                     &ctx->cm_id->route.addr.src_addr);
 862         }
 863
 864         addr = (struct sockaddr_ib *) &resp.dst_addr;
 865         resp.dst_size = sizeof(*addr);
 866         if (ctx->cm_id->route.addr.dst_addr.ss_family == AF_IB) {
 867                 memcpy(addr, &ctx->cm_id->route.addr.dst_addr, resp.dst_size);
 868         } else {
 869                 addr->sib_family = AF_IB;
 870                 addr->sib_pkey = (__force __be16) resp.pkey;
 871                 rdma_addr_get_dgid(&ctx->cm_id->route.addr.dev_addr,
 872                                    (union ib_gid *) &addr->sib_addr);
 873                 addr->sib_sid = rdma_get_service_id(ctx->cm_id, (struct sockaddr *)
 874                                                     &ctx->cm_id->route.addr.dst_addr);
 875         }
 876
 877         if (copy_to_user(response, &resp, sizeof(resp)))
 878                 ret = -EFAULT;
 879
 880         return ret;
 881 }
 882
 883 static ssize_t ucma_query(struct ucma_file *file,
 884                           const char __user *inbuf,
 885                           int in_len, int out_len)
 886 {
 887         struct rdma_ucm_query cmd;
 888         struct ucma_context *ctx;
 889         void __user *response;
 890         int ret;
 891
 892         if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
 893                 return -EFAULT;
 894
 895         response = (void __user *)(unsigned long) cmd.response;
 896         ctx = ucma_get_ctx(file, cmd.id);
 897         if (IS_ERR(ctx))
 898                 return PTR_ERR(ctx);
 899
 900         switch (cmd.option) {
 901         case RDMA_USER_CM_QUERY_ADDR:
 902                 ret = ucma_query_addr(ctx, response, out_len);
 903                 break;
 904         case RDMA_USER_CM_QUERY_PATH:
 905                 ret = ucma_query_path(ctx, response, out_len);
 906                 break;
 907         case RDMA_USER_CM_QUERY_GID:
 908                 ret = ucma_query_gid(ctx, response, out_len);
 909                 break;
 910         default:
 911                 ret = -ENOSYS;
 912                 break;
 913         }
 914
 915         ucma_put_ctx(ctx);
 916         return ret;
 917 }
 918
 919 static void ucma_copy_conn_param(struct rdma_cm_id *id,
 920                                  struct rdma_conn_param *dst,
 921                                  struct rdma_ucm_conn_param *src)
 922 {
 923         dst->private_data = src->private_data;
 924         dst->private_data_len = src->private_data_len;
 925         dst->responder_resources =src->responder_resources;
 926         dst->initiator_depth = src->initiator_depth;
 927         dst->flow_control = src->flow_control;
 928         dst->retry_count = src->retry_count;
 929         dst->rnr_retry_count = src->rnr_retry_count;
 930         dst->srq = src->srq;
 931         dst->qp_num = src->qp_num;
 932         dst->qkey = (id->route.addr.src_addr.ss_family == AF_IB) ? src->qkey : 0;
 933 }
 934
 935 static ssize_t ucma_connect(struct ucma_file *file, const char __user *inbuf,
 936                             int in_len, int out_len)
 937 {
 938         struct rdma_ucm_connect cmd;
 939         struct rdma_conn_param conn_param;
 940         struct ucma_context *ctx;
 941         int ret;
 942
 943         if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
 944                 return -EFAULT;
 945
 946         if (!cmd.conn_param.valid)
 947                 return -EINVAL;
 948
 949         ctx = ucma_get_ctx(file, cmd.id);
 950         if (IS_ERR(ctx))
 951                 return PTR_ERR(ctx);
 952
 953         ucma_copy_conn_param(ctx->cm_id, &conn_param, &cmd.conn_param);
 954         ret = rdma_connect(ctx->cm_id, &conn_param);
 955         ucma_put_ctx(ctx);
 956         return ret;
 957 }
 958
 959 static ssize_t ucma_listen(struct ucma_file *file, const char __user *inbuf,
 960                            int in_len, int out_len)
 961 {
 962         struct rdma_ucm_listen cmd;
 963         struct ucma_context *ctx;
 964         int ret;
 965
 966         if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
 967                 return -EFAULT;
 968
 969         ctx = ucma_get_ctx(file, cmd.id);
 970         if (IS_ERR(ctx))
 971                 return PTR_ERR(ctx);
 972
 973         ctx->backlog = cmd.backlog > 0 && cmd.backlog < max_backlog ?
 974                        cmd.backlog : max_backlog;
 975         ret = rdma_listen(ctx->cm_id, ctx->backlog);
 976         ucma_put_ctx(ctx);
 977         return ret;
 978 }
 979
 980 static ssize_t ucma_accept(struct ucma_file *file, const char __user *inbuf,
 981                            int in_len, int out_len)
 982 {
 983         struct rdma_ucm_accept cmd;
 984         struct rdma_conn_param conn_param;
 985         struct ucma_context *ctx;
 986         int ret;
 987
 988         if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
 989                 return -EFAULT;
 990
 991         ctx = ucma_get_ctx(file, cmd.id);
 992         if (IS_ERR(ctx))
 993                 return PTR_ERR(ctx);
 994
 995         if (cmd.conn_param.valid) {
 996                 ucma_copy_conn_param(ctx->cm_id, &conn_param, &cmd.conn_param);
 997                 mutex_lock(&file->mut);
 998                 ret = rdma_accept(ctx->cm_id, &conn_param);
 999                 if (!ret)
1000                         ctx->uid = cmd.uid;
1001                 mutex_unlock(&file->mut);
1002         } else
1003                 ret = rdma_accept(ctx->cm_id, NULL);
1004
1005         ucma_put_ctx(ctx);
1006         return ret;
1007 }
1008
1009 static ssize_t ucma_reject(struct ucma_file *file, const char __user *inbuf,
1010                            int in_len, int out_len)
1011 {
1012         struct rdma_ucm_reject cmd;
1013         struct ucma_context *ctx;
1014         int ret;
1015
1016         if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1017                 return -EFAULT;
1018
1019         ctx = ucma_get_ctx(file, cmd.id);
1020         if (IS_ERR(ctx))
1021                 return PTR_ERR(ctx);
1022
1023         ret = rdma_reject(ctx->cm_id, cmd.private_data, cmd.private_data_len);
1024         ucma_put_ctx(ctx);
1025         return ret;
1026 }
1027
1028 static ssize_t ucma_disconnect(struct ucma_file *file, const char __user *inbuf,
1029                                int in_len, int out_len)
1030 {
1031         struct rdma_ucm_disconnect cmd;
1032         struct ucma_context *ctx;
1033         int ret;
1034
1035         if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1036                 return -EFAULT;
1037
1038         ctx = ucma_get_ctx(file, cmd.id);
1039         if (IS_ERR(ctx))
1040                 return PTR_ERR(ctx);
1041
1042         ret = rdma_disconnect(ctx->cm_id);
1043         ucma_put_ctx(ctx);
1044         return ret;
1045 }
1046
1047 static ssize_t ucma_init_qp_attr(struct ucma_file *file,
1048                                  const char __user *inbuf,
1049                                  int in_len, int out_len)
1050 {
1051         struct rdma_ucm_init_qp_attr cmd;
1052         struct ib_uverbs_qp_attr resp;
1053         struct ucma_context *ctx;
1054         struct ib_qp_attr qp_attr;
1055         int ret;
1056
1057         if (out_len < sizeof(resp))
1058                 return -ENOSPC;
1059
1060         if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1061                 return -EFAULT;
1062
1063         ctx = ucma_get_ctx(file, cmd.id);
1064         if (IS_ERR(ctx))
1065                 return PTR_ERR(ctx);
1066
1067         resp.qp_attr_mask = 0;
1068         memset(&qp_attr, 0, sizeof qp_attr);
1069         qp_attr.qp_state = cmd.qp_state;
1070         ret = rdma_init_qp_attr(ctx->cm_id, &qp_attr, &resp.qp_attr_mask);
1071         if (ret)
1072                 goto out;
1073
1074         ib_copy_qp_attr_to_user(&resp, &qp_attr);
1075         if (copy_to_user((void __user *)(unsigned long)cmd.response,
1076                          &resp, sizeof(resp)))
1077                 ret = -EFAULT;
1078
1079 out:
1080         ucma_put_ctx(ctx);
1081         return ret;
1082 }
1083
1084 static int ucma_set_option_id(struct ucma_context *ctx, int optname,
1085                               void *optval, size_t optlen)
1086 {
1087         int ret = 0;
1088
1089         switch (optname) {
1090         case RDMA_OPTION_ID_TOS:
1091                 if (optlen != sizeof(u8)) {
1092                         ret = -EINVAL;
1093                         break;
1094                 }
1095                 rdma_set_service_type(ctx->cm_id, *((u8 *) optval));
1096                 break;
1097         case RDMA_OPTION_ID_REUSEADDR:
1098                 if (optlen != sizeof(int)) {
1099                         ret = -EINVAL;
1100                         break;
1101                 }
1102                 ret = rdma_set_reuseaddr(ctx->cm_id, *((int *) optval) ? 1 : 0);
1103                 break;
1104         case RDMA_OPTION_ID_AFONLY:
1105                 if (optlen != sizeof(int)) {
1106                         ret = -EINVAL;
1107                         break;
1108                 }
1109                 ret = rdma_set_afonly(ctx->cm_id, *((int *) optval) ? 1 : 0);
1110                 break;
1111         default:
1112                 ret = -ENOSYS;
1113         }
1114
1115         return ret;
1116 }
1117
1118 static int ucma_set_ib_path(struct ucma_context *ctx,
1119                             struct ib_path_rec_data *path_data, size_t optlen)
1120 {
1121         struct ib_sa_path_rec sa_path;
1122         struct rdma_cm_event event;
1123         int ret;
1124
1125         if (optlen % sizeof(*path_data))
1126                 return -EINVAL;
1127
1128         for (; optlen; optlen -= sizeof(*path_data), path_data++) {
1129                 if (path_data->flags == (IB_PATH_GMP | IB_PATH_PRIMARY |
1130                                          IB_PATH_BIDIRECTIONAL))
1131                         break;
1132         }
1133
1134         if (!optlen)
1135                 return -EINVAL;
1136
1137         ib_sa_unpack_path(path_data->path_rec, &sa_path);
1138         ret = rdma_set_ib_paths(ctx->cm_id, &sa_path, 1);
1139         if (ret)
1140                 return ret;
1141
1142         memset(&event, 0, sizeof event);
1143         event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
1144         return ucma_event_handler(ctx->cm_id, &event);
1145 }
1146
1147 static int ucma_set_option_ib(struct ucma_context *ctx, int optname,
1148                               void *optval, size_t optlen)
1149 {
1150         int ret;
1151
1152         switch (optname) {
1153         case RDMA_OPTION_IB_PATH:
1154                 ret = ucma_set_ib_path(ctx, optval, optlen);
1155                 break;
1156         default:
1157                 ret = -ENOSYS;
1158         }
1159
1160         return ret;
1161 }
1162
1163 static int ucma_set_option_level(struct ucma_context *ctx, int level,
1164                                  int optname, void *optval, size_t optlen)
1165 {
1166         int ret;
1167
1168         switch (level) {
1169         case RDMA_OPTION_ID:
1170                 ret = ucma_set_option_id(ctx, optname, optval, optlen);
1171                 break;
1172         case RDMA_OPTION_IB:
1173                 ret = ucma_set_option_ib(ctx, optname, optval, optlen);
1174                 break;
1175         default:
1176                 ret = -ENOSYS;
1177         }
1178
1179         return ret;
1180 }
1181
1182 static ssize_t ucma_set_option(struct ucma_file *file, const char __user *inbuf,
1183                                int in_len, int out_len)
1184 {
1185         struct rdma_ucm_set_option cmd;
1186         struct ucma_context *ctx;
1187         void *optval;
1188         int ret;
1189
1190         if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1191                 return -EFAULT;
1192
1193         ctx = ucma_get_ctx(file, cmd.id);
1194         if (IS_ERR(ctx))
1195                 return PTR_ERR(ctx);
1196
1197         optval = memdup_user((void __user *) (unsigned long) cmd.optval,
1198                              cmd.optlen);
1199         if (IS_ERR(optval)) {
1200                 ret = PTR_ERR(optval);
1201                 goto out;
1202         }
1203
1204         ret = ucma_set_option_level(ctx, cmd.level, cmd.optname, optval,
1205                                     cmd.optlen);
1206         kfree(optval);
1207
1208 out:
1209         ucma_put_ctx(ctx);
1210         return ret;
1211 }
1212
1213 static ssize_t ucma_notify(struct ucma_file *file, const char __user *inbuf,
1214                            int in_len, int out_len)
1215 {
1216         struct rdma_ucm_notify cmd;
1217         struct ucma_context *ctx;
1218         int ret;
1219
1220         if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1221                 return -EFAULT;
1222
1223         ctx = ucma_get_ctx(file, cmd.id);
1224         if (IS_ERR(ctx))
1225                 return PTR_ERR(ctx);
1226
1227         ret = rdma_notify(ctx->cm_id, (enum ib_event_type) cmd.event);
1228         ucma_put_ctx(ctx);
1229         return ret;
1230 }
1231
1232 static ssize_t ucma_process_join(struct ucma_file *file,
1233                                  struct rdma_ucm_join_mcast *cmd,  int out_len)
1234 {
1235         struct rdma_ucm_create_id_resp resp;
1236         struct ucma_context *ctx;
1237         struct ucma_multicast *mc;
1238         struct sockaddr *addr;
1239         int ret;
1240
1241         if (out_len < sizeof(resp))
1242                 return -ENOSPC;
1243
1244         addr = (struct sockaddr *) &cmd->addr;
1245         if (cmd->reserved || !cmd->addr_size || (cmd->addr_size != rdma_addr_size(addr)))
1246                 return -EINVAL;
1247
1248         ctx = ucma_get_ctx(file, cmd->id);
1249         if (IS_ERR(ctx))
1250                 return PTR_ERR(ctx);
1251
1252         mutex_lock(&file->mut);
1253         mc = ucma_alloc_multicast(ctx);
1254         if (!mc) {
1255                 ret = -ENOMEM;
1256                 goto err1;
1257         }
1258
1259         mc->uid = cmd->uid;
1260         memcpy(&mc->addr, addr, cmd->addr_size);
1261         ret = rdma_join_multicast(ctx->cm_id, (struct sockaddr *) &mc->addr, mc);
1262         if (ret)
1263                 goto err2;
1264
1265         resp.id = mc->id;
1266         if (copy_to_user((void __user *)(unsigned long) cmd->response,
1267                          &resp, sizeof(resp))) {
1268                 ret = -EFAULT;
1269                 goto err3;
1270         }
1271
1272         mutex_unlock(&file->mut);
1273         ucma_put_ctx(ctx);
1274         return 0;
1275
1276 err3:
1277         rdma_leave_multicast(ctx->cm_id, (struct sockaddr *) &mc->addr);
1278         ucma_cleanup_mc_events(mc);
1279 err2:
1280         mutex_lock(&mut);
1281         idr_remove(&multicast_idr, mc->id);
1282         mutex_unlock(&mut);
1283         list_del(&mc->list);
1284         kfree(mc);
1285 err1:
1286         mutex_unlock(&file->mut);
1287         ucma_put_ctx(ctx);
1288         return ret;
1289 }
1290
1291 static ssize_t ucma_join_ip_multicast(struct ucma_file *file,
1292                                       const char __user *inbuf,
1293                                       int in_len, int out_len)
1294 {
1295         struct rdma_ucm_join_ip_mcast cmd;
1296         struct rdma_ucm_join_mcast join_cmd;
1297
1298         if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1299                 return -EFAULT;
1300
1301         join_cmd.response = cmd.response;
1302         join_cmd.uid = cmd.uid;
1303         join_cmd.id = cmd.id;
1304         join_cmd.addr_size = rdma_addr_size((struct sockaddr *) &cmd.addr);
1305         join_cmd.reserved = 0;
1306         memcpy(&join_cmd.addr, &cmd.addr, join_cmd.addr_size);
1307
1308         return ucma_process_join(file, &join_cmd, out_len);
1309 }
1310
1311 static ssize_t ucma_join_multicast(struct ucma_file *file,
1312                                    const char __user *inbuf,
1313                                    int in_len, int out_len)
1314 {
1315         struct rdma_ucm_join_mcast cmd;
1316
1317         if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1318                 return -EFAULT;
1319
1320         return ucma_process_join(file, &cmd, out_len);
1321 }
1322
1323 static ssize_t ucma_leave_multicast(struct ucma_file *file,
1324                                     const char __user *inbuf,
1325                                     int in_len, int out_len)
1326 {
1327         struct rdma_ucm_destroy_id cmd;
1328         struct rdma_ucm_destroy_id_resp resp;
1329         struct ucma_multicast *mc;
1330         int ret = 0;
1331
1332         if (out_len < sizeof(resp))
1333                 return -ENOSPC;
1334
1335         if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1336                 return -EFAULT;
1337
1338         mutex_lock(&mut);
1339         mc = idr_find(&multicast_idr, cmd.id);
1340         if (!mc)
1341                 mc = ERR_PTR(-ENOENT);
1342         else if (mc->ctx->file != file)
1343                 mc = ERR_PTR(-EINVAL);
1344         else {
1345                 idr_remove(&multicast_idr, mc->id);
1346                 atomic_inc(&mc->ctx->ref);
1347         }
1348         mutex_unlock(&mut);
1349
1350         if (IS_ERR(mc)) {
1351                 ret = PTR_ERR(mc);
1352                 goto out;
1353         }
1354
1355         rdma_leave_multicast(mc->ctx->cm_id, (struct sockaddr *) &mc->addr);
1356         mutex_lock(&mc->ctx->file->mut);
1357         ucma_cleanup_mc_events(mc);
1358         list_del(&mc->list);
1359         mutex_unlock(&mc->ctx->file->mut);
1360
1361         ucma_put_ctx(mc->ctx);
1362         resp.events_reported = mc->events_reported;
1363         kfree(mc);
1364
1365         if (copy_to_user((void __user *)(unsigned long)cmd.response,
1366                          &resp, sizeof(resp)))
1367                 ret = -EFAULT;
1368 out:
1369         return ret;
1370 }
1371
1372 static void ucma_lock_files(struct ucma_file *file1, struct ucma_file *file2)
1373 {
1374         /* Acquire mutex's based on pointer comparison to prevent deadlock. */
1375         if (file1 < file2) {
1376                 mutex_lock(&file1->mut);
1377                 mutex_lock(&file2->mut);
1378         } else {
1379                 mutex_lock(&file2->mut);
1380                 mutex_lock(&file1->mut);
1381         }
1382 }
1383
1384 static void ucma_unlock_files(struct ucma_file *file1, struct ucma_file *file2)
1385 {
1386         if (file1 < file2) {
1387                 mutex_unlock(&file2->mut);
1388                 mutex_unlock(&file1->mut);
1389         } else {
1390                 mutex_unlock(&file1->mut);
1391                 mutex_unlock(&file2->mut);
1392         }
1393 }
1394
1395 static void ucma_move_events(struct ucma_context *ctx, struct ucma_file *file)
1396 {
1397         struct ucma_event *uevent, *tmp;
1398
1399         list_for_each_entry_safe(uevent, tmp, &ctx->file->event_list, list)
1400                 if (uevent->ctx == ctx)
1401                         list_move_tail(&uevent->list, &file->event_list);
1402 }
1403
1404 static ssize_t ucma_migrate_id(struct ucma_file *new_file,
1405                                const char __user *inbuf,
1406                                int in_len, int out_len)
1407 {
1408         struct rdma_ucm_migrate_id cmd;
1409         struct rdma_ucm_migrate_resp resp;
1410         struct ucma_context *ctx;
1411         struct fd f;
1412         struct ucma_file *cur_file;
1413         int ret = 0;
1414
1415         if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1416                 return -EFAULT;
1417
1418         /* Get current fd to protect against it being closed */
1419         f = fdget(cmd.fd);
1420         if (!f.file)
1421                 return -ENOENT;
1422
1423         /* Validate current fd and prevent destruction of id. */
1424         ctx = ucma_get_ctx(f.file->private_data, cmd.id);
1425         if (IS_ERR(ctx)) {
1426                 ret = PTR_ERR(ctx);
1427                 goto file_put;
1428         }
1429
1430         cur_file = ctx->file;
1431         if (cur_file == new_file) {
1432                 resp.events_reported = ctx->events_reported;
1433                 goto response;
1434         }
1435
1436         /*
1437          * Migrate events between fd's, maintaining order, and avoiding new
1438          * events being added before existing events.
1439          */
1440         ucma_lock_files(cur_file, new_file);
1441         mutex_lock(&mut);
1442
1443         list_move_tail(&ctx->list, &new_file->ctx_list);
1444         ucma_move_events(ctx, new_file);
1445         ctx->file = new_file;
1446         resp.events_reported = ctx->events_reported;
1447
1448         mutex_unlock(&mut);
1449         ucma_unlock_files(cur_file, new_file);
1450
1451 response:
1452         if (copy_to_user((void __user *)(unsigned long)cmd.response,
1453                          &resp, sizeof(resp)))
1454                 ret = -EFAULT;
1455
1456         ucma_put_ctx(ctx);
1457 file_put:
1458         fdput(f);
1459         return ret;
1460 }
1461
1462 static ssize_t (*ucma_cmd_table[])(struct ucma_file *file,
1463                                    const char __user *inbuf,
1464                                    int in_len, int out_len) = {
1465         [RDMA_USER_CM_CMD_CREATE_ID]     = ucma_create_id,
1466         [RDMA_USER_CM_CMD_DESTROY_ID]    = ucma_destroy_id,
1467         [RDMA_USER_CM_CMD_BIND_IP]       = ucma_bind_ip,
1468         [RDMA_USER_CM_CMD_RESOLVE_IP]    = ucma_resolve_ip,
1469         [RDMA_USER_CM_CMD_RESOLVE_ROUTE] = ucma_resolve_route,
1470         [RDMA_USER_CM_CMD_QUERY_ROUTE]   = ucma_query_route,
1471         [RDMA_USER_CM_CMD_CONNECT]       = ucma_connect,
1472         [RDMA_USER_CM_CMD_LISTEN]        = ucma_listen,
1473         [RDMA_USER_CM_CMD_ACCEPT]        = ucma_accept,
1474         [RDMA_USER_CM_CMD_REJECT]        = ucma_reject,
1475         [RDMA_USER_CM_CMD_DISCONNECT]    = ucma_disconnect,
1476         [RDMA_USER_CM_CMD_INIT_QP_ATTR]  = ucma_init_qp_attr,
1477         [RDMA_USER_CM_CMD_GET_EVENT]     = ucma_get_event,
1478         [RDMA_USER_CM_CMD_GET_OPTION]    = NULL,
1479         [RDMA_USER_CM_CMD_SET_OPTION]    = ucma_set_option,
1480         [RDMA_USER_CM_CMD_NOTIFY]        = ucma_notify,
1481         [RDMA_USER_CM_CMD_JOIN_IP_MCAST] = ucma_join_ip_multicast,
1482         [RDMA_USER_CM_CMD_LEAVE_MCAST]   = ucma_leave_multicast,
1483         [RDMA_USER_CM_CMD_MIGRATE_ID]    = ucma_migrate_id,
1484         [RDMA_USER_CM_CMD_QUERY]         = ucma_query,
1485         [RDMA_USER_CM_CMD_BIND]          = ucma_bind,
1486         [RDMA_USER_CM_CMD_RESOLVE_ADDR]  = ucma_resolve_addr,
1487         [RDMA_USER_CM_CMD_JOIN_MCAST]    = ucma_join_multicast
1488 };
1489
1490 static ssize_t ucma_write(struct file *filp, const char __user *buf,
1491                           size_t len, loff_t *pos)
1492 {
1493         struct ucma_file *file = filp->private_data;
1494         struct rdma_ucm_cmd_hdr hdr;
1495         ssize_t ret;
1496
1497         if (len < sizeof(hdr))
1498                 return -EINVAL;
1499
1500         if (copy_from_user(&hdr, buf, sizeof(hdr)))
1501                 return -EFAULT;
1502
1503         if (hdr.cmd >= ARRAY_SIZE(ucma_cmd_table))
1504                 return -EINVAL;
1505
1506         if (hdr.in + sizeof(hdr) > len)
1507                 return -EINVAL;
1508
1509         if (!ucma_cmd_table[hdr.cmd])
1510                 return -ENOSYS;
1511
1512         ret = ucma_cmd_table[hdr.cmd](file, buf + sizeof(hdr), hdr.in, hdr.out);
1513         if (!ret)
1514                 ret = len;
1515
1516         return ret;
1517 }
1518
1519 static unsigned int ucma_poll(struct file *filp, struct poll_table_struct *wait)
1520 {
1521         struct ucma_file *file = filp->private_data;
1522         unsigned int mask = 0;
1523
1524         poll_wait(filp, &file->poll_wait, wait);
1525
1526         if (!list_empty(&file->event_list))
1527                 mask = POLLIN | POLLRDNORM;
1528
1529         return mask;
1530 }
1531
1532 /*
1533  * ucma_open() does not need the BKL:
1534  *
1535  *  - no global state is referred to;
1536  *  - there is no ioctl method to race against;
1537  *  - no further module initialization is required for open to work
1538  *    after the device is registered.
1539  */
1540 static int ucma_open(struct inode *inode, struct file *filp)
1541 {
1542         struct ucma_file *file;
1543
1544         file = kmalloc(sizeof *file, GFP_KERNEL);
1545         if (!file)
1546                 return -ENOMEM;
1547
1548         INIT_LIST_HEAD(&file->event_list);
1549         INIT_LIST_HEAD(&file->ctx_list);
1550         init_waitqueue_head(&file->poll_wait);
1551         mutex_init(&file->mut);
1552
1553         filp->private_data = file;
1554         file->filp = filp;
1555
1556         return nonseekable_open(inode, filp);
1557 }
1558
1559 static int ucma_close(struct inode *inode, struct file *filp)
1560 {
1561         struct ucma_file *file = filp->private_data;
1562         struct ucma_context *ctx, *tmp;
1563
1564         mutex_lock(&file->mut);
1565         list_for_each_entry_safe(ctx, tmp, &file->ctx_list, list) {
1566                 mutex_unlock(&file->mut);
1567
1568                 mutex_lock(&mut);
1569                 idr_remove(&ctx_idr, ctx->id);
1570                 mutex_unlock(&mut);
1571
1572                 ucma_free_ctx(ctx);
1573                 mutex_lock(&file->mut);
1574         }
1575         mutex_unlock(&file->mut);
1576         kfree(file);
1577         return 0;
1578 }
1579
1580 static const struct file_operations ucma_fops = {
1581         .owner   = THIS_MODULE,
1582         .open    = ucma_open,
1583         .release = ucma_close,
1584         .write   = ucma_write,
1585         .poll    = ucma_poll,
1586         .llseek  = no_llseek,
1587 };
1588
1589 static struct miscdevice ucma_misc = {
1590         .minor          = MISC_DYNAMIC_MINOR,
1591         .name           = "rdma_cm",
1592         .nodename       = "infiniband/rdma_cm",
1593         .mode           = 0666,
1594         .fops           = &ucma_fops,
1595 };
1596
1597 static ssize_t show_abi_version(struct device *dev,
1598                                 struct device_attribute *attr,
1599                                 char *buf)
1600 {
1601         return sprintf(buf, "%d\n", RDMA_USER_CM_ABI_VERSION);
1602 }
1603 static DEVICE_ATTR(abi_version, S_IRUGO, show_abi_version, NULL);
1604
1605 static int __init ucma_init(void)
1606 {
1607         int ret;
1608
1609         ret = misc_register(&ucma_misc);
1610         if (ret)
1611                 return ret;
1612
1613         ret = device_create_file(ucma_misc.this_device, &dev_attr_abi_version);
1614         if (ret) {
1615                 printk(KERN_ERR "rdma_ucm: couldn't create abi_version attr\n");
1616                 goto err1;
1617         }
1618
1619         ucma_ctl_table_hdr = register_net_sysctl(&init_net, "net/rdma_ucm", ucma_ctl_table);
1620         if (!ucma_ctl_table_hdr) {
1621                 printk(KERN_ERR "rdma_ucm: couldn't register sysctl paths\n");
1622                 ret = -ENOMEM;
1623                 goto err2;
1624         }
1625         return 0;
1626 err2:
1627         device_remove_file(ucma_misc.this_device, &dev_attr_abi_version);
1628 err1:
1629         misc_deregister(&ucma_misc);
1630         return ret;
1631 }
1632
1633 static void __exit ucma_cleanup(void)
1634 {
1635         unregister_net_sysctl_table(ucma_ctl_table_hdr);
1636         device_remove_file(ucma_misc.this_device, &dev_attr_abi_version);
1637         misc_deregister(&ucma_misc);
1638         idr_destroy(&ctx_idr);
1639 }
1640
1641 module_init(ucma_init);
1642 module_exit(ucma_cleanup);