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