SAFE public projects git trees. - safe/jmp/linux-2.6/blob - net/core/rtnetlink.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              Routing netlink socket interface: protocol independent part.
   7  *
   8  * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
   9  *
  10  *              This program is free software; you can redistribute it and/or
  11  *              modify it under the terms of the GNU General Public License
  12  *              as published by the Free Software Foundation; either version
  13  *              2 of the License, or (at your option) any later version.
  14  *
  15  *      Fixes:
  16  *      Vitaly E. Lavrov                RTA_OK arithmetics was wrong.
  17  */
  18
  19 #include <linux/config.h>
  20 #include <linux/errno.h>
  21 #include <linux/module.h>
  22 #include <linux/types.h>
  23 #include <linux/socket.h>
  24 #include <linux/kernel.h>
  25 #include <linux/sched.h>
  26 #include <linux/timer.h>
  27 #include <linux/string.h>
  28 #include <linux/sockios.h>
  29 #include <linux/net.h>
  30 #include <linux/fcntl.h>
  31 #include <linux/mm.h>
  32 #include <linux/slab.h>
  33 #include <linux/interrupt.h>
  34 #include <linux/capability.h>
  35 #include <linux/skbuff.h>
  36 #include <linux/init.h>
  37 #include <linux/security.h>
  38
  39 #include <asm/uaccess.h>
  40 #include <asm/system.h>
  41 #include <asm/string.h>
  42
  43 #include <linux/inet.h>
  44 #include <linux/netdevice.h>
  45 #include <net/ip.h>
  46 #include <net/protocol.h>
  47 #include <net/arp.h>
  48 #include <net/route.h>
  49 #include <net/udp.h>
  50 #include <net/sock.h>
  51 #include <net/pkt_sched.h>
  52
  53 DECLARE_MUTEX(rtnl_sem);
  54
  55 void rtnl_lock(void)
  56 {
  57         rtnl_shlock();
  58 }
  59
  60 int rtnl_lock_interruptible(void)
  61 {
  62         return down_interruptible(&rtnl_sem);
  63 }
  64
  65 void rtnl_unlock(void)
  66 {
  67         rtnl_shunlock();
  68
  69         netdev_run_todo();
  70 }
  71
  72 int rtattr_parse(struct rtattr *tb[], int maxattr, struct rtattr *rta, int len)
  73 {
  74         memset(tb, 0, sizeof(struct rtattr*)*maxattr);
  75
  76         while (RTA_OK(rta, len)) {
  77                 unsigned flavor = rta->rta_type;
  78                 if (flavor && flavor <= maxattr)
  79                         tb[flavor-1] = rta;
  80                 rta = RTA_NEXT(rta, len);
  81         }
  82         return 0;
  83 }
  84
  85 struct sock *rtnl;
  86
  87 struct rtnetlink_link * rtnetlink_links[NPROTO];
  88
  89 static const int rtm_min[RTM_NR_FAMILIES] =
  90 {
  91         [RTM_FAM(RTM_NEWLINK)]      = NLMSG_LENGTH(sizeof(struct ifinfomsg)),
  92         [RTM_FAM(RTM_NEWADDR)]      = NLMSG_LENGTH(sizeof(struct ifaddrmsg)),
  93         [RTM_FAM(RTM_NEWROUTE)]     = NLMSG_LENGTH(sizeof(struct rtmsg)),
  94         [RTM_FAM(RTM_NEWNEIGH)]     = NLMSG_LENGTH(sizeof(struct ndmsg)),
  95         [RTM_FAM(RTM_NEWRULE)]      = NLMSG_LENGTH(sizeof(struct rtmsg)),
  96         [RTM_FAM(RTM_NEWQDISC)]     = NLMSG_LENGTH(sizeof(struct tcmsg)),
  97         [RTM_FAM(RTM_NEWTCLASS)]    = NLMSG_LENGTH(sizeof(struct tcmsg)),
  98         [RTM_FAM(RTM_NEWTFILTER)]   = NLMSG_LENGTH(sizeof(struct tcmsg)),
  99         [RTM_FAM(RTM_NEWACTION)]    = NLMSG_LENGTH(sizeof(struct tcamsg)),
 100         [RTM_FAM(RTM_NEWPREFIX)]    = NLMSG_LENGTH(sizeof(struct rtgenmsg)),
 101         [RTM_FAM(RTM_GETMULTICAST)] = NLMSG_LENGTH(sizeof(struct rtgenmsg)),
 102         [RTM_FAM(RTM_GETANYCAST)]   = NLMSG_LENGTH(sizeof(struct rtgenmsg)),
 103         [RTM_FAM(RTM_NEWNEIGHTBL)]  = NLMSG_LENGTH(sizeof(struct ndtmsg)),
 104 };
 105
 106 static const int rta_max[RTM_NR_FAMILIES] =
 107 {
 108         [RTM_FAM(RTM_NEWLINK)]      = IFLA_MAX,
 109         [RTM_FAM(RTM_NEWADDR)]      = IFA_MAX,
 110         [RTM_FAM(RTM_NEWROUTE)]     = RTA_MAX,
 111         [RTM_FAM(RTM_NEWNEIGH)]     = NDA_MAX,
 112         [RTM_FAM(RTM_NEWRULE)]      = RTA_MAX,
 113         [RTM_FAM(RTM_NEWQDISC)]     = TCA_MAX,
 114         [RTM_FAM(RTM_NEWTCLASS)]    = TCA_MAX,
 115         [RTM_FAM(RTM_NEWTFILTER)]   = TCA_MAX,
 116         [RTM_FAM(RTM_NEWACTION)]    = TCAA_MAX,
 117         [RTM_FAM(RTM_NEWNEIGHTBL)]  = NDTA_MAX,
 118 };
 119
 120 void __rta_fill(struct sk_buff *skb, int attrtype, int attrlen, const void *data)
 121 {
 122         struct rtattr *rta;
 123         int size = RTA_LENGTH(attrlen);
 124
 125         rta = (struct rtattr*)skb_put(skb, RTA_ALIGN(size));
 126         rta->rta_type = attrtype;
 127         rta->rta_len = size;
 128         memcpy(RTA_DATA(rta), data, attrlen);
 129 }
 130
 131 size_t rtattr_strlcpy(char *dest, const struct rtattr *rta, size_t size)
 132 {
 133         size_t ret = RTA_PAYLOAD(rta);
 134         char *src = RTA_DATA(rta);
 135
 136         if (ret > 0 && src[ret - 1] == '\0')
 137                 ret--;
 138         if (size > 0) {
 139                 size_t len = (ret >= size) ? size - 1 : ret;
 140                 memset(dest, 0, size);
 141                 memcpy(dest, src, len);
 142         }
 143         return ret;
 144 }
 145
 146 int rtnetlink_send(struct sk_buff *skb, u32 pid, unsigned group, int echo)
 147 {
 148         int err = 0;
 149
 150         NETLINK_CB(skb).dst_groups = group;
 151         if (echo)
 152                 atomic_inc(&skb->users);
 153         netlink_broadcast(rtnl, skb, pid, group, GFP_KERNEL);
 154         if (echo)
 155                 err = netlink_unicast(rtnl, skb, pid, MSG_DONTWAIT);
 156         return err;
 157 }
 158
 159 int rtnetlink_put_metrics(struct sk_buff *skb, u32 *metrics)
 160 {
 161         struct rtattr *mx = (struct rtattr*)skb->tail;
 162         int i;
 163
 164         RTA_PUT(skb, RTA_METRICS, 0, NULL);
 165         for (i=0; i<RTAX_MAX; i++) {
 166                 if (metrics[i])
 167                         RTA_PUT(skb, i+1, sizeof(u32), metrics+i);
 168         }
 169         mx->rta_len = skb->tail - (u8*)mx;
 170         if (mx->rta_len == RTA_LENGTH(0))
 171                 skb_trim(skb, (u8*)mx - skb->data);
 172         return 0;
 173
 174 rtattr_failure:
 175         skb_trim(skb, (u8*)mx - skb->data);
 176         return -1;
 177 }
 178
 179
 180 static int rtnetlink_fill_ifinfo(struct sk_buff *skb, struct net_device *dev,
 181                                  int type, u32 pid, u32 seq, u32 change,
 182                                  unsigned int flags)
 183 {
 184         struct ifinfomsg *r;
 185         struct nlmsghdr  *nlh;
 186         unsigned char    *b = skb->tail;
 187
 188         nlh = NLMSG_NEW(skb, pid, seq, type, sizeof(*r), flags);
 189         r = NLMSG_DATA(nlh);
 190         r->ifi_family = AF_UNSPEC;
 191         r->ifi_type = dev->type;
 192         r->ifi_index = dev->ifindex;
 193         r->ifi_flags = dev_get_flags(dev);
 194         r->ifi_change = change;
 195
 196         RTA_PUT(skb, IFLA_IFNAME, strlen(dev->name)+1, dev->name);
 197
 198         if (1) {
 199                 u32 txqlen = dev->tx_queue_len;
 200                 RTA_PUT(skb, IFLA_TXQLEN, sizeof(txqlen), &txqlen);
 201         }
 202
 203         if (1) {
 204                 u32 weight = dev->weight;
 205                 RTA_PUT(skb, IFLA_WEIGHT, sizeof(weight), &weight);
 206         }
 207
 208         if (1) {
 209                 struct rtnl_link_ifmap map = {
 210                         .mem_start   = dev->mem_start,
 211                         .mem_end     = dev->mem_end,
 212                         .base_addr   = dev->base_addr,
 213                         .irq         = dev->irq,
 214                         .dma         = dev->dma,
 215                         .port        = dev->if_port,
 216                 };
 217                 RTA_PUT(skb, IFLA_MAP, sizeof(map), &map);
 218         }
 219
 220         if (dev->addr_len) {
 221                 RTA_PUT(skb, IFLA_ADDRESS, dev->addr_len, dev->dev_addr);
 222                 RTA_PUT(skb, IFLA_BROADCAST, dev->addr_len, dev->broadcast);
 223         }
 224
 225         if (1) {
 226                 u32 mtu = dev->mtu;
 227                 RTA_PUT(skb, IFLA_MTU, sizeof(mtu), &mtu);
 228         }
 229
 230         if (dev->ifindex != dev->iflink) {
 231                 u32 iflink = dev->iflink;
 232                 RTA_PUT(skb, IFLA_LINK, sizeof(iflink), &iflink);
 233         }
 234
 235         if (dev->qdisc_sleeping)
 236                 RTA_PUT(skb, IFLA_QDISC,
 237                         strlen(dev->qdisc_sleeping->ops->id) + 1,
 238                         dev->qdisc_sleeping->ops->id);
 239
 240         if (dev->master) {
 241                 u32 master = dev->master->ifindex;
 242                 RTA_PUT(skb, IFLA_MASTER, sizeof(master), &master);
 243         }
 244
 245         if (dev->get_stats) {
 246                 unsigned long *stats = (unsigned long*)dev->get_stats(dev);
 247                 if (stats) {
 248                         struct rtattr  *a;
 249                         __u32          *s;
 250                         int             i;
 251                         int             n = sizeof(struct rtnl_link_stats)/4;
 252
 253                         a = __RTA_PUT(skb, IFLA_STATS, n*4);
 254                         s = RTA_DATA(a);
 255                         for (i=0; i<n; i++)
 256                                 s[i] = stats[i];
 257                 }
 258         }
 259         nlh->nlmsg_len = skb->tail - b;
 260         return skb->len;
 261
 262 nlmsg_failure:
 263 rtattr_failure:
 264         skb_trim(skb, b - skb->data);
 265         return -1;
 266 }
 267
 268 static int rtnetlink_dump_ifinfo(struct sk_buff *skb, struct netlink_callback *cb)
 269 {
 270         int idx;
 271         int s_idx = cb->args[0];
 272         struct net_device *dev;
 273
 274         read_lock(&dev_base_lock);
 275         for (dev=dev_base, idx=0; dev; dev = dev->next, idx++) {
 276                 if (idx < s_idx)
 277                         continue;
 278                 if (rtnetlink_fill_ifinfo(skb, dev, RTM_NEWLINK,
 279                                           NETLINK_CB(cb->skb).pid,
 280                                           cb->nlh->nlmsg_seq, 0,
 281                                           NLM_F_MULTI) <= 0)
 282                         break;
 283         }
 284         read_unlock(&dev_base_lock);
 285         cb->args[0] = idx;
 286
 287         return skb->len;
 288 }
 289
 290 static int do_setlink(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
 291 {
 292         struct ifinfomsg  *ifm = NLMSG_DATA(nlh);
 293         struct rtattr    **ida = arg;
 294         struct net_device *dev;
 295         int err, send_addr_notify = 0;
 296
 297         if (ifm->ifi_index >= 0)
 298                 dev = dev_get_by_index(ifm->ifi_index);
 299         else if (ida[IFLA_IFNAME - 1]) {
 300                 char ifname[IFNAMSIZ];
 301
 302                 if (rtattr_strlcpy(ifname, ida[IFLA_IFNAME - 1],
 303                                    IFNAMSIZ) >= IFNAMSIZ)
 304                         return -EINVAL;
 305                 dev = dev_get_by_name(ifname);
 306         } else
 307                 return -EINVAL;
 308
 309         if (!dev)
 310                 return -ENODEV;
 311
 312         err = -EINVAL;
 313
 314         if (ifm->ifi_flags)
 315                 dev_change_flags(dev, ifm->ifi_flags);
 316
 317         if (ida[IFLA_MAP - 1]) {
 318                 struct rtnl_link_ifmap *u_map;
 319                 struct ifmap k_map;
 320
 321                 if (!dev->set_config) {
 322                         err = -EOPNOTSUPP;
 323                         goto out;
 324                 }
 325
 326                 if (!netif_device_present(dev)) {
 327                         err = -ENODEV;
 328                         goto out;
 329                 }
 330
 331                 if (ida[IFLA_MAP - 1]->rta_len != RTA_LENGTH(sizeof(*u_map)))
 332                         goto out;
 333
 334                 u_map = RTA_DATA(ida[IFLA_MAP - 1]);
 335
 336                 k_map.mem_start = (unsigned long) u_map->mem_start;
 337                 k_map.mem_end = (unsigned long) u_map->mem_end;
 338                 k_map.base_addr = (unsigned short) u_map->base_addr;
 339                 k_map.irq = (unsigned char) u_map->irq;
 340                 k_map.dma = (unsigned char) u_map->dma;
 341                 k_map.port = (unsigned char) u_map->port;
 342
 343                 err = dev->set_config(dev, &k_map);
 344
 345                 if (err)
 346                         goto out;
 347         }
 348
 349         if (ida[IFLA_ADDRESS - 1]) {
 350                 if (!dev->set_mac_address) {
 351                         err = -EOPNOTSUPP;
 352                         goto out;
 353                 }
 354                 if (!netif_device_present(dev)) {
 355                         err = -ENODEV;
 356                         goto out;
 357                 }
 358                 if (ida[IFLA_ADDRESS - 1]->rta_len != RTA_LENGTH(dev->addr_len))
 359                         goto out;
 360
 361                 err = dev->set_mac_address(dev, RTA_DATA(ida[IFLA_ADDRESS - 1]));
 362                 if (err)
 363                         goto out;
 364                 send_addr_notify = 1;
 365         }
 366
 367         if (ida[IFLA_BROADCAST - 1]) {
 368                 if (ida[IFLA_BROADCAST - 1]->rta_len != RTA_LENGTH(dev->addr_len))
 369                         goto out;
 370                 memcpy(dev->broadcast, RTA_DATA(ida[IFLA_BROADCAST - 1]),
 371                        dev->addr_len);
 372                 send_addr_notify = 1;
 373         }
 374
 375         if (ida[IFLA_MTU - 1]) {
 376                 if (ida[IFLA_MTU - 1]->rta_len != RTA_LENGTH(sizeof(u32)))
 377                         goto out;
 378                 err = dev_set_mtu(dev, *((u32 *) RTA_DATA(ida[IFLA_MTU - 1])));
 379
 380                 if (err)
 381                         goto out;
 382
 383         }
 384
 385         if (ida[IFLA_TXQLEN - 1]) {
 386                 if (ida[IFLA_TXQLEN - 1]->rta_len != RTA_LENGTH(sizeof(u32)))
 387                         goto out;
 388
 389                 dev->tx_queue_len = *((u32 *) RTA_DATA(ida[IFLA_TXQLEN - 1]));
 390         }
 391
 392         if (ida[IFLA_WEIGHT - 1]) {
 393                 if (ida[IFLA_WEIGHT - 1]->rta_len != RTA_LENGTH(sizeof(u32)))
 394                         goto out;
 395
 396                 dev->weight = *((u32 *) RTA_DATA(ida[IFLA_WEIGHT - 1]));
 397         }
 398
 399         if (ifm->ifi_index >= 0 && ida[IFLA_IFNAME - 1]) {
 400                 char ifname[IFNAMSIZ];
 401
 402                 if (rtattr_strlcpy(ifname, ida[IFLA_IFNAME - 1],
 403                                    IFNAMSIZ) >= IFNAMSIZ)
 404                         goto out;
 405                 err = dev_change_name(dev, ifname);
 406                 if (err)
 407                         goto out;
 408         }
 409
 410         err = 0;
 411
 412 out:
 413         if (send_addr_notify)
 414                 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
 415
 416         dev_put(dev);
 417         return err;
 418 }
 419
 420 static int rtnetlink_dump_all(struct sk_buff *skb, struct netlink_callback *cb)
 421 {
 422         int idx;
 423         int s_idx = cb->family;
 424
 425         if (s_idx == 0)
 426                 s_idx = 1;
 427         for (idx=1; idx<NPROTO; idx++) {
 428                 int type = cb->nlh->nlmsg_type-RTM_BASE;
 429                 if (idx < s_idx || idx == PF_PACKET)
 430                         continue;
 431                 if (rtnetlink_links[idx] == NULL ||
 432                     rtnetlink_links[idx][type].dumpit == NULL)
 433                         continue;
 434                 if (idx > s_idx)
 435                         memset(&cb->args[0], 0, sizeof(cb->args));
 436                 if (rtnetlink_links[idx][type].dumpit(skb, cb))
 437                         break;
 438         }
 439         cb->family = idx;
 440
 441         return skb->len;
 442 }
 443
 444 void rtmsg_ifinfo(int type, struct net_device *dev, unsigned change)
 445 {
 446         struct sk_buff *skb;
 447         int size = NLMSG_SPACE(sizeof(struct ifinfomsg) +
 448                                sizeof(struct rtnl_link_ifmap) +
 449                                sizeof(struct rtnl_link_stats) + 128);
 450
 451         skb = alloc_skb(size, GFP_KERNEL);
 452         if (!skb)
 453                 return;
 454
 455         if (rtnetlink_fill_ifinfo(skb, dev, type, current->pid, 0, change, 0) < 0) {
 456                 kfree_skb(skb);
 457                 return;
 458         }
 459         NETLINK_CB(skb).dst_groups = RTMGRP_LINK;
 460         netlink_broadcast(rtnl, skb, 0, RTMGRP_LINK, GFP_KERNEL);
 461 }
 462
 463 static int rtnetlink_done(struct netlink_callback *cb)
 464 {
 465         return 0;
 466 }
 467
 468 /* Protected by RTNL sempahore.  */
 469 static struct rtattr **rta_buf;
 470 static int rtattr_max;
 471
 472 /* Process one rtnetlink message. */
 473
 474 static __inline__ int
 475 rtnetlink_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh, int *errp)
 476 {
 477         struct rtnetlink_link *link;
 478         struct rtnetlink_link *link_tab;
 479         int sz_idx, kind;
 480         int min_len;
 481         int family;
 482         int type;
 483         int err;
 484
 485         /* Only requests are handled by kernel now */
 486         if (!(nlh->nlmsg_flags&NLM_F_REQUEST))
 487                 return 0;
 488
 489         type = nlh->nlmsg_type;
 490
 491         /* A control message: ignore them */
 492         if (type < RTM_BASE)
 493                 return 0;
 494
 495         /* Unknown message: reply with EINVAL */
 496         if (type > RTM_MAX)
 497                 goto err_inval;
 498
 499         type -= RTM_BASE;
 500
 501         /* All the messages must have at least 1 byte length */
 502         if (nlh->nlmsg_len < NLMSG_LENGTH(sizeof(struct rtgenmsg)))
 503                 return 0;
 504
 505         family = ((struct rtgenmsg*)NLMSG_DATA(nlh))->rtgen_family;
 506         if (family >= NPROTO) {
 507                 *errp = -EAFNOSUPPORT;
 508                 return -1;
 509         }
 510
 511         link_tab = rtnetlink_links[family];
 512         if (link_tab == NULL)
 513                 link_tab = rtnetlink_links[PF_UNSPEC];
 514         link = &link_tab[type];
 515
 516         sz_idx = type>>2;
 517         kind = type&3;
 518
 519         if (kind != 2 && security_netlink_recv(skb)) {
 520                 *errp = -EPERM;
 521                 return -1;
 522         }
 523
 524         if (kind == 2 && nlh->nlmsg_flags&NLM_F_DUMP) {
 525                 u32 rlen;
 526
 527                 if (link->dumpit == NULL)
 528                         link = &(rtnetlink_links[PF_UNSPEC][type]);
 529
 530                 if (link->dumpit == NULL)
 531                         goto err_inval;
 532
 533                 if ((*errp = netlink_dump_start(rtnl, skb, nlh,
 534                                                 link->dumpit,
 535                                                 rtnetlink_done)) != 0) {
 536                         return -1;
 537                 }
 538                 rlen = NLMSG_ALIGN(nlh->nlmsg_len);
 539                 if (rlen > skb->len)
 540                         rlen = skb->len;
 541                 skb_pull(skb, rlen);
 542                 return -1;
 543         }
 544
 545         memset(rta_buf, 0, (rtattr_max * sizeof(struct rtattr *)));
 546
 547         min_len = rtm_min[sz_idx];
 548         if (nlh->nlmsg_len < min_len)
 549                 goto err_inval;
 550
 551         if (nlh->nlmsg_len > min_len) {
 552                 int attrlen = nlh->nlmsg_len - NLMSG_ALIGN(min_len);
 553                 struct rtattr *attr = (void*)nlh + NLMSG_ALIGN(min_len);
 554
 555                 while (RTA_OK(attr, attrlen)) {
 556                         unsigned flavor = attr->rta_type;
 557                         if (flavor) {
 558                                 if (flavor > rta_max[sz_idx])
 559                                         goto err_inval;
 560                                 rta_buf[flavor-1] = attr;
 561                         }
 562                         attr = RTA_NEXT(attr, attrlen);
 563                 }
 564         }
 565
 566         if (link->doit == NULL)
 567                 link = &(rtnetlink_links[PF_UNSPEC][type]);
 568         if (link->doit == NULL)
 569                 goto err_inval;
 570         err = link->doit(skb, nlh, (void *)&rta_buf[0]);
 571
 572         *errp = err;
 573         return err;
 574
 575 err_inval:
 576         *errp = -EINVAL;
 577         return -1;
 578 }
 579
 580 /*
 581  * Process one packet of messages.
 582  * Malformed skbs with wrong lengths of messages are discarded silently.
 583  */
 584
 585 static inline int rtnetlink_rcv_skb(struct sk_buff *skb)
 586 {
 587         int err;
 588         struct nlmsghdr * nlh;
 589
 590         while (skb->len >= NLMSG_SPACE(0)) {
 591                 u32 rlen;
 592
 593                 nlh = (struct nlmsghdr *)skb->data;
 594                 if (nlh->nlmsg_len < sizeof(*nlh) || skb->len < nlh->nlmsg_len)
 595                         return 0;
 596                 rlen = NLMSG_ALIGN(nlh->nlmsg_len);
 597                 if (rlen > skb->len)
 598                         rlen = skb->len;
 599                 if (rtnetlink_rcv_msg(skb, nlh, &err)) {
 600                         /* Not error, but we must interrupt processing here:
 601                          *   Note, that in this case we do not pull message
 602                          *   from skb, it will be processed later.
 603                          */
 604                         if (err == 0)
 605                                 return -1;
 606                         netlink_ack(skb, nlh, err);
 607                 } else if (nlh->nlmsg_flags&NLM_F_ACK)
 608                         netlink_ack(skb, nlh, 0);
 609                 skb_pull(skb, rlen);
 610         }
 611
 612         return 0;
 613 }
 614
 615 /*
 616  *  rtnetlink input queue processing routine:
 617  *      - process as much as there was in the queue upon entry.
 618  *      - feed skbs to rtnetlink_rcv_skb, until it refuse a message,
 619  *        that will occur, when a dump started.
 620  */
 621
 622 static void rtnetlink_rcv(struct sock *sk, int len)
 623 {
 624         unsigned int qlen = skb_queue_len(&sk->sk_receive_queue);
 625
 626         do {
 627                 struct sk_buff *skb;
 628
 629                 rtnl_lock();
 630
 631                 if (qlen > skb_queue_len(&sk->sk_receive_queue))
 632                         qlen = skb_queue_len(&sk->sk_receive_queue);
 633
 634                 for (; qlen; qlen--) {
 635                         skb = skb_dequeue(&sk->sk_receive_queue);
 636                         if (rtnetlink_rcv_skb(skb)) {
 637                                 if (skb->len)
 638                                         skb_queue_head(&sk->sk_receive_queue,
 639                                                        skb);
 640                                 else {
 641                                         kfree_skb(skb);
 642                                         qlen--;
 643                                 }
 644                                 break;
 645                         }
 646                         kfree_skb(skb);
 647                 }
 648
 649                 up(&rtnl_sem);
 650
 651                 netdev_run_todo();
 652         } while (qlen);
 653 }
 654
 655 static struct rtnetlink_link link_rtnetlink_table[RTM_NR_MSGTYPES] =
 656 {
 657         [RTM_GETLINK     - RTM_BASE] = { .dumpit = rtnetlink_dump_ifinfo },
 658         [RTM_SETLINK     - RTM_BASE] = { .doit   = do_setlink            },
 659         [RTM_GETADDR     - RTM_BASE] = { .dumpit = rtnetlink_dump_all    },
 660         [RTM_GETROUTE    - RTM_BASE] = { .dumpit = rtnetlink_dump_all    },
 661         [RTM_NEWNEIGH    - RTM_BASE] = { .doit   = neigh_add             },
 662         [RTM_DELNEIGH    - RTM_BASE] = { .doit   = neigh_delete          },
 663         [RTM_GETNEIGH    - RTM_BASE] = { .dumpit = neigh_dump_info       },
 664         [RTM_GETRULE     - RTM_BASE] = { .dumpit = rtnetlink_dump_all    },
 665         [RTM_GETNEIGHTBL - RTM_BASE] = { .dumpit = neightbl_dump_info    },
 666         [RTM_SETNEIGHTBL - RTM_BASE] = { .doit   = neightbl_set          },
 667 };
 668
 669 static int rtnetlink_event(struct notifier_block *this, unsigned long event, void *ptr)
 670 {
 671         struct net_device *dev = ptr;
 672         switch (event) {
 673         case NETDEV_UNREGISTER:
 674                 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
 675                 break;
 676         case NETDEV_REGISTER:
 677                 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
 678                 break;
 679         case NETDEV_UP:
 680         case NETDEV_DOWN:
 681                 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
 682                 break;
 683         case NETDEV_CHANGE:
 684         case NETDEV_GOING_DOWN:
 685                 break;
 686         default:
 687                 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
 688                 break;
 689         }
 690         return NOTIFY_DONE;
 691 }
 692
 693 static struct notifier_block rtnetlink_dev_notifier = {
 694         .notifier_call  = rtnetlink_event,
 695 };
 696
 697 void __init rtnetlink_init(void)
 698 {
 699         int i;
 700
 701         rtattr_max = 0;
 702         for (i = 0; i < ARRAY_SIZE(rta_max); i++)
 703                 if (rta_max[i] > rtattr_max)
 704                         rtattr_max = rta_max[i];
 705         rta_buf = kmalloc(rtattr_max * sizeof(struct rtattr *), GFP_KERNEL);
 706         if (!rta_buf)
 707                 panic("rtnetlink_init: cannot allocate rta_buf\n");
 708
 709         rtnl = netlink_kernel_create(NETLINK_ROUTE, rtnetlink_rcv);
 710         if (rtnl == NULL)
 711                 panic("rtnetlink_init: cannot initialize rtnetlink\n");
 712         netlink_set_nonroot(NETLINK_ROUTE, NL_NONROOT_RECV);
 713         register_netdevice_notifier(&rtnetlink_dev_notifier);
 714         rtnetlink_links[PF_UNSPEC] = link_rtnetlink_table;
 715         rtnetlink_links[PF_PACKET] = link_rtnetlink_table;
 716 }
 717
 718 EXPORT_SYMBOL(__rta_fill);
 719 EXPORT_SYMBOL(rtattr_strlcpy);
 720 EXPORT_SYMBOL(rtattr_parse);
 721 EXPORT_SYMBOL(rtnetlink_links);
 722 EXPORT_SYMBOL(rtnetlink_put_metrics);
 723 EXPORT_SYMBOL(rtnl);
 724 EXPORT_SYMBOL(rtnl_lock);
 725 EXPORT_SYMBOL(rtnl_lock_interruptible);
 726 EXPORT_SYMBOL(rtnl_sem);
 727 EXPORT_SYMBOL(rtnl_unlock);