git.oblomov.eu Git - linux-2.6/blob - drivers/infiniband/core/addr.c

   1 /*
   2  * Copyright (c) 2005 Voltaire Inc.  All rights reserved.
   3  * Copyright (c) 2002-2005, Network Appliance, Inc. All rights reserved.
   4  * Copyright (c) 1999-2005, Mellanox Technologies, Inc. All rights reserved.
   5  * Copyright (c) 2005 Intel Corporation.  All rights reserved.
   6  *
   7  * This software is available to you under a choice of one of two
   8  * licenses.  You may choose to be licensed under the terms of the GNU
   9  * General Public License (GPL) Version 2, available from the file
  10  * COPYING in the main directory of this source tree, or the
  11  * OpenIB.org BSD license below:
  12  *
  13  *     Redistribution and use in source and binary forms, with or
  14  *     without modification, are permitted provided that the following
  15  *     conditions are met:
  16  *
  17  *      - Redistributions of source code must retain the above
  18  *        copyright notice, this list of conditions and the following
  19  *        disclaimer.
  20  *
  21  *      - Redistributions in binary form must reproduce the above
  22  *        copyright notice, this list of conditions and the following
  23  *        disclaimer in the documentation and/or other materials
  24  *        provided with the distribution.
  25  *
  26  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  27  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  28  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  29  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  30  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  31  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  32  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  33  * SOFTWARE.
  34  */
  35
  36 #include <linux/mutex.h>
  37 #include <linux/inetdevice.h>
  38 #include <linux/workqueue.h>
  39 #include <linux/if_arp.h>
  40 #include <net/arp.h>
  41 #include <net/neighbour.h>
  42 #include <net/route.h>
  43 #include <net/netevent.h>
  44 #include <rdma/ib_addr.h>
  45
  46 MODULE_AUTHOR("Sean Hefty");
  47 MODULE_DESCRIPTION("IB Address Translation");
  48 MODULE_LICENSE("Dual BSD/GPL");
  49
  50 struct addr_req {
  51         struct list_head list;
  52         struct sockaddr src_addr;
  53         struct sockaddr dst_addr;
  54         struct rdma_dev_addr *addr;
  55         struct rdma_addr_client *client;
  56         void *context;
  57         void (*callback)(int status, struct sockaddr *src_addr,
  58                          struct rdma_dev_addr *addr, void *context);
  59         unsigned long timeout;
  60         int status;
  61 };
  62
  63 static void process_req(struct work_struct *work);
  64
  65 static DEFINE_MUTEX(lock);
  66 static LIST_HEAD(req_list);
  67 static DECLARE_DELAYED_WORK(work, process_req);
  68 static struct workqueue_struct *addr_wq;
  69
  70 void rdma_addr_register_client(struct rdma_addr_client *client)
  71 {
  72         atomic_set(&client->refcount, 1);
  73         init_completion(&client->comp);
  74 }
  75 EXPORT_SYMBOL(rdma_addr_register_client);
  76
  77 static inline void put_client(struct rdma_addr_client *client)
  78 {
  79         if (atomic_dec_and_test(&client->refcount))
  80                 complete(&client->comp);
  81 }
  82
  83 void rdma_addr_unregister_client(struct rdma_addr_client *client)
  84 {
  85         put_client(client);
  86         wait_for_completion(&client->comp);
  87 }
  88 EXPORT_SYMBOL(rdma_addr_unregister_client);
  89
  90 int rdma_copy_addr(struct rdma_dev_addr *dev_addr, struct net_device *dev,
  91                      const unsigned char *dst_dev_addr)
  92 {
  93         switch (dev->type) {
  94         case ARPHRD_INFINIBAND:
  95                 dev_addr->dev_type = RDMA_NODE_IB_CA;
  96                 break;
  97         case ARPHRD_ETHER:
  98                 dev_addr->dev_type = RDMA_NODE_RNIC;
  99                 break;
 100         default:
 101                 return -EADDRNOTAVAIL;
 102         }
 103
 104         memcpy(dev_addr->src_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
 105         memcpy(dev_addr->broadcast, dev->broadcast, MAX_ADDR_LEN);
 106         if (dst_dev_addr)
 107                 memcpy(dev_addr->dst_dev_addr, dst_dev_addr, MAX_ADDR_LEN);
 108         return 0;
 109 }
 110 EXPORT_SYMBOL(rdma_copy_addr);
 111
 112 int rdma_translate_ip(struct sockaddr *addr, struct rdma_dev_addr *dev_addr)
 113 {
 114         struct net_device *dev;
 115         __be32 ip = ((struct sockaddr_in *) addr)->sin_addr.s_addr;
 116         int ret;
 117
 118         dev = ip_dev_find(&init_net, ip);
 119         if (!dev)
 120                 return -EADDRNOTAVAIL;
 121
 122         ret = rdma_copy_addr(dev_addr, dev, NULL);
 123         dev_put(dev);
 124         return ret;
 125 }
 126 EXPORT_SYMBOL(rdma_translate_ip);
 127
 128 static void set_timeout(unsigned long time)
 129 {
 130         unsigned long delay;
 131
 132         cancel_delayed_work(&work);
 133
 134         delay = time - jiffies;
 135         if ((long)delay <= 0)
 136                 delay = 1;
 137
 138         queue_delayed_work(addr_wq, &work, delay);
 139 }
 140
 141 static void queue_req(struct addr_req *req)
 142 {
 143         struct addr_req *temp_req;
 144
 145         mutex_lock(&lock);
 146         list_for_each_entry_reverse(temp_req, &req_list, list) {
 147                 if (time_after_eq(req->timeout, temp_req->timeout))
 148                         break;
 149         }
 150
 151         list_add(&req->list, &temp_req->list);
 152
 153         if (req_list.next == &req->list)
 154                 set_timeout(req->timeout);
 155         mutex_unlock(&lock);
 156 }
 157
 158 static void addr_send_arp(struct sockaddr_in *dst_in)
 159 {
 160         struct rtable *rt;
 161         struct flowi fl;
 162         __be32 dst_ip = dst_in->sin_addr.s_addr;
 163
 164         memset(&fl, 0, sizeof fl);
 165         fl.nl_u.ip4_u.daddr = dst_ip;
 166         if (ip_route_output_key(&init_net, &rt, &fl))
 167                 return;
 168
 169         neigh_event_send(rt->u.dst.neighbour, NULL);
 170         ip_rt_put(rt);
 171 }
 172
 173 static int addr_resolve_remote(struct sockaddr_in *src_in,
 174                                struct sockaddr_in *dst_in,
 175                                struct rdma_dev_addr *addr)
 176 {
 177         __be32 src_ip = src_in->sin_addr.s_addr;
 178         __be32 dst_ip = dst_in->sin_addr.s_addr;
 179         struct flowi fl;
 180         struct rtable *rt;
 181         struct neighbour *neigh;
 182         int ret;
 183
 184         memset(&fl, 0, sizeof fl);
 185         fl.nl_u.ip4_u.daddr = dst_ip;
 186         fl.nl_u.ip4_u.saddr = src_ip;
 187         ret = ip_route_output_key(&init_net, &rt, &fl);
 188         if (ret)
 189                 goto out;
 190
 191         /* If the device does ARP internally, return 'done' */
 192         if (rt->idev->dev->flags & IFF_NOARP) {
 193                 rdma_copy_addr(addr, rt->idev->dev, NULL);
 194                 goto put;
 195         }
 196
 197         neigh = neigh_lookup(&arp_tbl, &rt->rt_gateway, rt->idev->dev);
 198         if (!neigh) {
 199                 ret = -ENODATA;
 200                 goto put;
 201         }
 202
 203         if (!(neigh->nud_state & NUD_VALID)) {
 204                 ret = -ENODATA;
 205                 goto release;
 206         }
 207
 208         if (!src_ip) {
 209                 src_in->sin_family = dst_in->sin_family;
 210                 src_in->sin_addr.s_addr = rt->rt_src;
 211         }
 212
 213         ret = rdma_copy_addr(addr, neigh->dev, neigh->ha);
 214 release:
 215         neigh_release(neigh);
 216 put:
 217         ip_rt_put(rt);
 218 out:
 219         return ret;
 220 }
 221
 222 static void process_req(struct work_struct *work)
 223 {
 224         struct addr_req *req, *temp_req;
 225         struct sockaddr_in *src_in, *dst_in;
 226         struct list_head done_list;
 227
 228         INIT_LIST_HEAD(&done_list);
 229
 230         mutex_lock(&lock);
 231         list_for_each_entry_safe(req, temp_req, &req_list, list) {
 232                 if (req->status == -ENODATA) {
 233                         src_in = (struct sockaddr_in *) &req->src_addr;
 234                         dst_in = (struct sockaddr_in *) &req->dst_addr;
 235                         req->status = addr_resolve_remote(src_in, dst_in,
 236                                                           req->addr);
 237                         if (req->status && time_after_eq(jiffies, req->timeout))
 238                                 req->status = -ETIMEDOUT;
 239                         else if (req->status == -ENODATA)
 240                                 continue;
 241                 }
 242                 list_move_tail(&req->list, &done_list);
 243         }
 244
 245         if (!list_empty(&req_list)) {
 246                 req = list_entry(req_list.next, struct addr_req, list);
 247                 set_timeout(req->timeout);
 248         }
 249         mutex_unlock(&lock);
 250
 251         list_for_each_entry_safe(req, temp_req, &done_list, list) {
 252                 list_del(&req->list);
 253                 req->callback(req->status, &req->src_addr, req->addr,
 254                               req->context);
 255                 put_client(req->client);
 256                 kfree(req);
 257         }
 258 }
 259
 260 static int addr_resolve_local(struct sockaddr_in *src_in,
 261                               struct sockaddr_in *dst_in,
 262                               struct rdma_dev_addr *addr)
 263 {
 264         struct net_device *dev;
 265         __be32 src_ip = src_in->sin_addr.s_addr;
 266         __be32 dst_ip = dst_in->sin_addr.s_addr;
 267         int ret;
 268
 269         dev = ip_dev_find(&init_net, dst_ip);
 270         if (!dev)
 271                 return -EADDRNOTAVAIL;
 272
 273         if (ipv4_is_zeronet(src_ip)) {
 274                 src_in->sin_family = dst_in->sin_family;
 275                 src_in->sin_addr.s_addr = dst_ip;
 276                 ret = rdma_copy_addr(addr, dev, dev->dev_addr);
 277         } else if (ipv4_is_loopback(src_ip)) {
 278                 ret = rdma_translate_ip((struct sockaddr *)dst_in, addr);
 279                 if (!ret)
 280                         memcpy(addr->dst_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
 281         } else {
 282                 ret = rdma_translate_ip((struct sockaddr *)src_in, addr);
 283                 if (!ret)
 284                         memcpy(addr->dst_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
 285         }
 286
 287         dev_put(dev);
 288         return ret;
 289 }
 290
 291 int rdma_resolve_ip(struct rdma_addr_client *client,
 292                     struct sockaddr *src_addr, struct sockaddr *dst_addr,
 293                     struct rdma_dev_addr *addr, int timeout_ms,
 294                     void (*callback)(int status, struct sockaddr *src_addr,
 295                                      struct rdma_dev_addr *addr, void *context),
 296                     void *context)
 297 {
 298         struct sockaddr_in *src_in, *dst_in;
 299         struct addr_req *req;
 300         int ret = 0;
 301
 302         req = kzalloc(sizeof *req, GFP_KERNEL);
 303         if (!req)
 304                 return -ENOMEM;
 305
 306         if (src_addr)
 307                 memcpy(&req->src_addr, src_addr, ip_addr_size(src_addr));
 308         memcpy(&req->dst_addr, dst_addr, ip_addr_size(dst_addr));
 309         req->addr = addr;
 310         req->callback = callback;
 311         req->context = context;
 312         req->client = client;
 313         atomic_inc(&client->refcount);
 314
 315         src_in = (struct sockaddr_in *) &req->src_addr;
 316         dst_in = (struct sockaddr_in *) &req->dst_addr;
 317
 318         req->status = addr_resolve_local(src_in, dst_in, addr);
 319         if (req->status == -EADDRNOTAVAIL)
 320                 req->status = addr_resolve_remote(src_in, dst_in, addr);
 321
 322         switch (req->status) {
 323         case 0:
 324                 req->timeout = jiffies;
 325                 queue_req(req);
 326                 break;
 327         case -ENODATA:
 328                 req->timeout = msecs_to_jiffies(timeout_ms) + jiffies;
 329                 queue_req(req);
 330                 addr_send_arp(dst_in);
 331                 break;
 332         default:
 333                 ret = req->status;
 334                 atomic_dec(&client->refcount);
 335                 kfree(req);
 336                 break;
 337         }
 338         return ret;
 339 }
 340 EXPORT_SYMBOL(rdma_resolve_ip);
 341
 342 void rdma_addr_cancel(struct rdma_dev_addr *addr)
 343 {
 344         struct addr_req *req, *temp_req;
 345
 346         mutex_lock(&lock);
 347         list_for_each_entry_safe(req, temp_req, &req_list, list) {
 348                 if (req->addr == addr) {
 349                         req->status = -ECANCELED;
 350                         req->timeout = jiffies;
 351                         list_move(&req->list, &req_list);
 352                         set_timeout(req->timeout);
 353                         break;
 354                 }
 355         }
 356         mutex_unlock(&lock);
 357 }
 358 EXPORT_SYMBOL(rdma_addr_cancel);
 359
 360 static int netevent_callback(struct notifier_block *self, unsigned long event,
 361         void *ctx)
 362 {
 363         if (event == NETEVENT_NEIGH_UPDATE) {
 364                 struct neighbour *neigh = ctx;
 365
 366                 if (neigh->nud_state & NUD_VALID) {
 367                         set_timeout(jiffies);
 368                 }
 369         }
 370         return 0;
 371 }
 372
 373 static struct notifier_block nb = {
 374         .notifier_call = netevent_callback
 375 };
 376
 377 static int addr_init(void)
 378 {
 379         addr_wq = create_singlethread_workqueue("ib_addr");
 380         if (!addr_wq)
 381                 return -ENOMEM;
 382
 383         register_netevent_notifier(&nb);
 384         return 0;
 385 }
 386
 387 static void addr_cleanup(void)
 388 {
 389         unregister_netevent_notifier(&nb);
 390         destroy_workqueue(addr_wq);
 391 }
 392
 393 module_init(addr_init);
 394 module_exit(addr_cleanup);