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/major.h>
  26 #include <linux/sched.h>
  27 #include <linux/timer.h>
  28 #include <linux/string.h>
  29 #include <linux/sockios.h>
  30 #include <linux/net.h>
  31 #include <linux/fcntl.h>
  32 #include <linux/mm.h>
  33 #include <linux/slab.h>
  34 #include <linux/interrupt.h>
  35 #include <linux/capability.h>
  36 #include <linux/skbuff.h>
  37 #include <linux/init.h>
  38 #include <linux/security.h>
  39
  40 #include <asm/uaccess.h>
  41 #include <asm/system.h>
  42 #include <asm/string.h>
  43
  44 #include <linux/inet.h>
  45 #include <linux/netdevice.h>
  46 #include <net/ip.h>
  47 #include <net/protocol.h>
  48 #include <net/arp.h>
  49 #include <net/route.h>
  50 #include <net/udp.h>
  51 #include <net/sock.h>
  52 #include <net/pkt_sched.h>
  53
  54 DECLARE_MUTEX(rtnl_sem);
  55
  56 void rtnl_lock(void)
  57 {
  58         rtnl_shlock();
  59         rtnl_exlock();
  60 }
  61
  62 void rtnl_unlock(void)
  63 {
  64         rtnl_exunlock();
  65         rtnl_shunlock();
  66
  67         netdev_run_todo();
  68 }
  69
  70 int rtattr_parse(struct rtattr *tb[], int maxattr, struct rtattr *rta, int len)
  71 {
  72         memset(tb, 0, sizeof(struct rtattr*)*maxattr);
  73
  74         while (RTA_OK(rta, len)) {
  75                 unsigned flavor = rta->rta_type;
  76                 if (flavor && flavor <= maxattr)
  77                         tb[flavor-1] = rta;
  78                 rta = RTA_NEXT(rta, len);
  79         }
  80         return 0;
  81 }
  82
  83 struct sock *rtnl;
  84
  85 struct rtnetlink_link * rtnetlink_links[NPROTO];
  86
  87 static const int rtm_min[(RTM_MAX+1-RTM_BASE)/4] =
  88 {
  89         NLMSG_LENGTH(sizeof(struct ifinfomsg)),
  90         NLMSG_LENGTH(sizeof(struct ifaddrmsg)),
  91         NLMSG_LENGTH(sizeof(struct rtmsg)),
  92         NLMSG_LENGTH(sizeof(struct ndmsg)),
  93         NLMSG_LENGTH(sizeof(struct rtmsg)),
  94         NLMSG_LENGTH(sizeof(struct tcmsg)),
  95         NLMSG_LENGTH(sizeof(struct tcmsg)),
  96         NLMSG_LENGTH(sizeof(struct tcmsg)),
  97         NLMSG_LENGTH(sizeof(struct tcamsg))
  98 };
  99
 100 static const int rta_max[(RTM_MAX+1-RTM_BASE)/4] =
 101 {
 102         IFLA_MAX,
 103         IFA_MAX,
 104         RTA_MAX,
 105         NDA_MAX,
 106         RTA_MAX,
 107         TCA_MAX,
 108         TCA_MAX,
 109         TCA_MAX,
 110         TCAA_MAX
 111 };
 112
 113 void __rta_fill(struct sk_buff *skb, int attrtype, int attrlen, const void *data)
 114 {
 115         struct rtattr *rta;
 116         int size = RTA_LENGTH(attrlen);
 117
 118         rta = (struct rtattr*)skb_put(skb, RTA_ALIGN(size));
 119         rta->rta_type = attrtype;
 120         rta->rta_len = size;
 121         memcpy(RTA_DATA(rta), data, attrlen);
 122 }
 123
 124 int rtnetlink_send(struct sk_buff *skb, u32 pid, unsigned group, int echo)
 125 {
 126         int err = 0;
 127
 128         NETLINK_CB(skb).dst_groups = group;
 129         if (echo)
 130                 atomic_inc(&skb->users);
 131         netlink_broadcast(rtnl, skb, pid, group, GFP_KERNEL);
 132         if (echo)
 133                 err = netlink_unicast(rtnl, skb, pid, MSG_DONTWAIT);
 134         return err;
 135 }
 136
 137 int rtnetlink_put_metrics(struct sk_buff *skb, u32 *metrics)
 138 {
 139         struct rtattr *mx = (struct rtattr*)skb->tail;
 140         int i;
 141
 142         RTA_PUT(skb, RTA_METRICS, 0, NULL);
 143         for (i=0; i<RTAX_MAX; i++) {
 144                 if (metrics[i])
 145                         RTA_PUT(skb, i+1, sizeof(u32), metrics+i);
 146         }
 147         mx->rta_len = skb->tail - (u8*)mx;
 148         if (mx->rta_len == RTA_LENGTH(0))
 149                 skb_trim(skb, (u8*)mx - skb->data);
 150         return 0;
 151
 152 rtattr_failure:
 153         skb_trim(skb, (u8*)mx - skb->data);
 154         return -1;
 155 }
 156
 157
 158 static int rtnetlink_fill_ifinfo(struct sk_buff *skb, struct net_device *dev,
 159                                  int type, u32 pid, u32 seq, u32 change)
 160 {
 161         struct ifinfomsg *r;
 162         struct nlmsghdr  *nlh;
 163         unsigned char    *b = skb->tail;
 164
 165         nlh = NLMSG_PUT(skb, pid, seq, type, sizeof(*r));
 166         if (pid) nlh->nlmsg_flags |= NLM_F_MULTI;
 167         r = NLMSG_DATA(nlh);
 168         r->ifi_family = AF_UNSPEC;
 169         r->ifi_type = dev->type;
 170         r->ifi_index = dev->ifindex;
 171         r->ifi_flags = dev->flags;
 172         r->ifi_change = change;
 173
 174         if (!netif_running(dev) || !netif_carrier_ok(dev))
 175                 r->ifi_flags &= ~IFF_RUNNING;
 176         else
 177                 r->ifi_flags |= IFF_RUNNING;
 178
 179         RTA_PUT(skb, IFLA_IFNAME, strlen(dev->name)+1, dev->name);
 180         if (dev->addr_len) {
 181                 RTA_PUT(skb, IFLA_ADDRESS, dev->addr_len, dev->dev_addr);
 182                 RTA_PUT(skb, IFLA_BROADCAST, dev->addr_len, dev->broadcast);
 183         }
 184         if (1) {
 185                 unsigned mtu = dev->mtu;
 186                 RTA_PUT(skb, IFLA_MTU, sizeof(mtu), &mtu);
 187         }
 188         if (dev->ifindex != dev->iflink)
 189                 RTA_PUT(skb, IFLA_LINK, sizeof(int), &dev->iflink);
 190         if (dev->qdisc_sleeping)
 191                 RTA_PUT(skb, IFLA_QDISC,
 192                         strlen(dev->qdisc_sleeping->ops->id) + 1,
 193                         dev->qdisc_sleeping->ops->id);
 194         if (dev->master)
 195                 RTA_PUT(skb, IFLA_MASTER, sizeof(int), &dev->master->ifindex);
 196         if (dev->get_stats) {
 197                 unsigned long *stats = (unsigned long*)dev->get_stats(dev);
 198                 if (stats) {
 199                         struct rtattr  *a;
 200                         __u32          *s;
 201                         int             i;
 202                         int             n = sizeof(struct rtnl_link_stats)/4;
 203
 204                         a = __RTA_PUT(skb, IFLA_STATS, n*4);
 205                         s = RTA_DATA(a);
 206                         for (i=0; i<n; i++)
 207                                 s[i] = stats[i];
 208                 }
 209         }
 210         nlh->nlmsg_len = skb->tail - b;
 211         return skb->len;
 212
 213 nlmsg_failure:
 214 rtattr_failure:
 215         skb_trim(skb, b - skb->data);
 216         return -1;
 217 }
 218
 219 int rtnetlink_dump_ifinfo(struct sk_buff *skb, struct netlink_callback *cb)
 220 {
 221         int idx;
 222         int s_idx = cb->args[0];
 223         struct net_device *dev;
 224
 225         read_lock(&dev_base_lock);
 226         for (dev=dev_base, idx=0; dev; dev = dev->next, idx++) {
 227                 if (idx < s_idx)
 228                         continue;
 229                 if (!dev_in_nx_info(dev, current->nx_info))
 230                         continue;
 231                 if (rtnetlink_fill_ifinfo(skb, dev, RTM_NEWLINK, NETLINK_CB(cb->skb).pid, cb->nlh->nlmsg_seq, 0) <= 0)
 232                         break;
 233         }
 234         read_unlock(&dev_base_lock);
 235         cb->args[0] = idx;
 236
 237         return skb->len;
 238 }
 239
 240 static int do_setlink(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
 241 {
 242         struct ifinfomsg  *ifm = NLMSG_DATA(nlh);
 243         struct rtattr    **ida = arg;
 244         struct net_device *dev;
 245         int err;
 246
 247         dev = dev_get_by_index(ifm->ifi_index);
 248         if (!dev)
 249                 return -ENODEV;
 250
 251         err = -EINVAL;
 252
 253         if (ida[IFLA_ADDRESS - 1]) {
 254                 if (!dev->set_mac_address) {
 255                         err = -EOPNOTSUPP;
 256                         goto out;
 257                 }
 258                 if (!netif_device_present(dev)) {
 259                         err = -ENODEV;
 260                         goto out;
 261                 }
 262                 if (ida[IFLA_ADDRESS - 1]->rta_len != RTA_LENGTH(dev->addr_len))
 263                         goto out;
 264
 265                 err = dev->set_mac_address(dev, RTA_DATA(ida[IFLA_ADDRESS - 1]));
 266                 if (err)
 267                         goto out;
 268         }
 269
 270         if (ida[IFLA_BROADCAST - 1]) {
 271                 if (ida[IFLA_BROADCAST - 1]->rta_len != RTA_LENGTH(dev->addr_len))
 272                         goto out;
 273                 memcpy(dev->broadcast, RTA_DATA(ida[IFLA_BROADCAST - 1]),
 274                        dev->addr_len);
 275         }
 276
 277         err = 0;
 278
 279 out:
 280         if (!err)
 281                 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
 282
 283         dev_put(dev);
 284         return err;
 285 }
 286
 287 static int rtnetlink_dump_all(struct sk_buff *skb, struct netlink_callback *cb)
 288 {
 289         int idx;
 290         int s_idx = cb->family;
 291
 292         if (s_idx == 0)
 293                 s_idx = 1;
 294         for (idx=1; idx<NPROTO; idx++) {
 295                 int type = cb->nlh->nlmsg_type-RTM_BASE;
 296                 if (idx < s_idx || idx == PF_PACKET)
 297                         continue;
 298                 if (rtnetlink_links[idx] == NULL ||
 299                     rtnetlink_links[idx][type].dumpit == NULL)
 300                         continue;
 301                 if (idx > s_idx)
 302                         memset(&cb->args[0], 0, sizeof(cb->args));
 303                 if (rtnetlink_links[idx][type].dumpit(skb, cb))
 304                         break;
 305         }
 306         cb->family = idx;
 307
 308         return skb->len;
 309 }
 310
 311 void rtmsg_ifinfo(int type, struct net_device *dev, unsigned change)
 312 {
 313         struct sk_buff *skb;
 314         int size = NLMSG_GOODSIZE;
 315
 316         if (!dev_in_nx_info(dev, current->nx_info))
 317                 return;
 318         skb = alloc_skb(size, GFP_KERNEL);
 319         if (!skb)
 320                 return;
 321
 322         if (rtnetlink_fill_ifinfo(skb, dev, type, 0, 0, change) < 0) {
 323                 kfree_skb(skb);
 324                 return;
 325         }
 326         NETLINK_CB(skb).dst_groups = RTMGRP_LINK;
 327         netlink_broadcast(rtnl, skb, 0, RTMGRP_LINK, GFP_KERNEL);
 328 }
 329
 330 static int rtnetlink_done(struct netlink_callback *cb)
 331 {
 332         return 0;
 333 }
 334
 335 /* Process one rtnetlink message. */
 336
 337 static __inline__ int
 338 rtnetlink_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh, int *errp)
 339 {
 340         struct rtnetlink_link *link;
 341         struct rtnetlink_link *link_tab;
 342         struct rtattr   *rta[RTATTR_MAX];
 343
 344         int exclusive = 0;
 345         int sz_idx, kind;
 346         int min_len;
 347         int family;
 348         int type;
 349         int err;
 350
 351         /* Only requests are handled by kernel now */
 352         if (!(nlh->nlmsg_flags&NLM_F_REQUEST))
 353                 return 0;
 354
 355         type = nlh->nlmsg_type;
 356
 357         /* A control message: ignore them */
 358         if (type < RTM_BASE)
 359                 return 0;
 360
 361         /* Unknown message: reply with EINVAL */
 362         if (type > RTM_MAX)
 363                 goto err_inval;
 364
 365         type -= RTM_BASE;
 366
 367         /* All the messages must have at least 1 byte length */
 368         if (nlh->nlmsg_len < NLMSG_LENGTH(sizeof(struct rtgenmsg)))
 369                 return 0;
 370
 371         family = ((struct rtgenmsg*)NLMSG_DATA(nlh))->rtgen_family;
 372         if (family >= NPROTO) {
 373                 *errp = -EAFNOSUPPORT;
 374                 return -1;
 375         }
 376
 377         link_tab = rtnetlink_links[family];
 378         if (link_tab == NULL)
 379                 link_tab = rtnetlink_links[PF_UNSPEC];
 380         link = &link_tab[type];
 381
 382         sz_idx = type>>2;
 383         kind = type&3;
 384
 385         if (kind != 2 && security_netlink_recv(skb)) {
 386                 *errp = -EPERM;
 387                 return -1;
 388         }
 389
 390         if (kind == 2 && nlh->nlmsg_flags&NLM_F_DUMP) {
 391                 u32 rlen;
 392
 393                 if (link->dumpit == NULL)
 394                         link = &(rtnetlink_links[PF_UNSPEC][type]);
 395
 396                 if (link->dumpit == NULL)
 397                         goto err_inval;
 398
 399                 if ((*errp = netlink_dump_start(rtnl, skb, nlh,
 400                                                 link->dumpit,
 401                                                 rtnetlink_done)) != 0) {
 402                         return -1;
 403                 }
 404                 rlen = NLMSG_ALIGN(nlh->nlmsg_len);
 405                 if (rlen > skb->len)
 406                         rlen = skb->len;
 407                 skb_pull(skb, rlen);
 408                 return -1;
 409         }
 410
 411         if (kind != 2) {
 412                 if (rtnl_exlock_nowait()) {
 413                         *errp = 0;
 414                         return -1;
 415                 }
 416                 exclusive = 1;
 417         }
 418
 419         memset(&rta, 0, sizeof(rta));
 420
 421         min_len = rtm_min[sz_idx];
 422         if (nlh->nlmsg_len < min_len)
 423                 goto err_inval;
 424
 425         if (nlh->nlmsg_len > min_len) {
 426                 int attrlen = nlh->nlmsg_len - NLMSG_ALIGN(min_len);
 427                 struct rtattr *attr = (void*)nlh + NLMSG_ALIGN(min_len);
 428
 429                 while (RTA_OK(attr, attrlen)) {
 430                         unsigned flavor = attr->rta_type;
 431                         if (flavor) {
 432                                 if (flavor > rta_max[sz_idx])
 433                                         goto err_inval;
 434                                 rta[flavor-1] = attr;
 435                         }
 436                         attr = RTA_NEXT(attr, attrlen);
 437                 }
 438         }
 439
 440         if (link->doit == NULL)
 441                 link = &(rtnetlink_links[PF_UNSPEC][type]);
 442         if (link->doit == NULL)
 443                 goto err_inval;
 444         err = link->doit(skb, nlh, (void *)&rta);
 445
 446         if (exclusive)
 447                 rtnl_exunlock();
 448         *errp = err;
 449         return err;
 450
 451 err_inval:
 452         if (exclusive)
 453                 rtnl_exunlock();
 454         *errp = -EINVAL;
 455         return -1;
 456 }
 457
 458 /*
 459  * Process one packet of messages.
 460  * Malformed skbs with wrong lengths of messages are discarded silently.
 461  */
 462
 463 static inline int rtnetlink_rcv_skb(struct sk_buff *skb)
 464 {
 465         int err;
 466         struct nlmsghdr * nlh;
 467
 468         while (skb->len >= NLMSG_SPACE(0)) {
 469                 u32 rlen;
 470
 471                 nlh = (struct nlmsghdr *)skb->data;
 472                 if (nlh->nlmsg_len < sizeof(*nlh) || skb->len < nlh->nlmsg_len)
 473                         return 0;
 474                 rlen = NLMSG_ALIGN(nlh->nlmsg_len);
 475                 if (rlen > skb->len)
 476                         rlen = skb->len;
 477                 if (rtnetlink_rcv_msg(skb, nlh, &err)) {
 478                         /* Not error, but we must interrupt processing here:
 479                          *   Note, that in this case we do not pull message
 480                          *   from skb, it will be processed later.
 481                          */
 482                         if (err == 0)
 483                                 return -1;
 484                         netlink_ack(skb, nlh, err);
 485                 } else if (nlh->nlmsg_flags&NLM_F_ACK)
 486                         netlink_ack(skb, nlh, 0);
 487                 skb_pull(skb, rlen);
 488         }
 489
 490         return 0;
 491 }
 492
 493 /*
 494  *  rtnetlink input queue processing routine:
 495  *      - try to acquire shared lock. If it is failed, defer processing.
 496  *      - feed skbs to rtnetlink_rcv_skb, until it refuse a message,
 497  *        that will occur, when a dump started and/or acquisition of
 498  *        exclusive lock failed.
 499  */
 500
 501 static void rtnetlink_rcv(struct sock *sk, int len)
 502 {
 503         do {
 504                 struct sk_buff *skb;
 505
 506                 if (rtnl_shlock_nowait())
 507                         return;
 508
 509                 while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
 510                         if (rtnetlink_rcv_skb(skb)) {
 511                                 if (skb->len)
 512                                         skb_queue_head(&sk->sk_receive_queue,
 513                                                        skb);
 514                                 else
 515                                         kfree_skb(skb);
 516                                 break;
 517                         }
 518                         kfree_skb(skb);
 519                 }
 520
 521                 up(&rtnl_sem);
 522
 523                 netdev_run_todo();
 524         } while (rtnl && rtnl->sk_receive_queue.qlen);
 525 }
 526
 527 static struct rtnetlink_link link_rtnetlink_table[RTM_MAX-RTM_BASE+1] =
 528 {
 529         [RTM_GETLINK  - RTM_BASE] = { .dumpit = rtnetlink_dump_ifinfo },
 530         [RTM_SETLINK  - RTM_BASE] = { .doit   = do_setlink            },
 531         [RTM_GETADDR  - RTM_BASE] = { .dumpit = rtnetlink_dump_all    },
 532         [RTM_GETROUTE - RTM_BASE] = { .dumpit = rtnetlink_dump_all    },
 533         [RTM_NEWNEIGH - RTM_BASE] = { .doit   = neigh_add             },
 534         [RTM_DELNEIGH - RTM_BASE] = { .doit   = neigh_delete          },
 535         [RTM_GETNEIGH - RTM_BASE] = { .dumpit = neigh_dump_info       }
 536 };
 537
 538 static int rtnetlink_event(struct notifier_block *this, unsigned long event, void *ptr)
 539 {
 540         struct net_device *dev = ptr;
 541         switch (event) {
 542         case NETDEV_UNREGISTER:
 543                 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
 544                 break;
 545         case NETDEV_REGISTER:
 546                 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
 547                 break;
 548         case NETDEV_UP:
 549         case NETDEV_DOWN:
 550                 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
 551                 break;
 552         case NETDEV_CHANGE:
 553         case NETDEV_GOING_DOWN:
 554                 break;
 555         default:
 556                 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
 557                 break;
 558         }
 559         return NOTIFY_DONE;
 560 }
 561
 562 static struct notifier_block rtnetlink_dev_notifier = {
 563         .notifier_call  = rtnetlink_event,
 564 };
 565
 566 void __init rtnetlink_init(void)
 567 {
 568         rtnl = netlink_kernel_create(NETLINK_ROUTE, rtnetlink_rcv);
 569         if (rtnl == NULL)
 570                 panic("rtnetlink_init: cannot initialize rtnetlink\n");
 571         netlink_set_nonroot(NETLINK_ROUTE, NL_NONROOT_RECV);
 572         register_netdevice_notifier(&rtnetlink_dev_notifier);
 573         rtnetlink_links[PF_UNSPEC] = link_rtnetlink_table;
 574         rtnetlink_links[PF_PACKET] = link_rtnetlink_table;
 575 }
 576
 577 EXPORT_SYMBOL(__rta_fill);
 578 EXPORT_SYMBOL(rtattr_parse);
 579 EXPORT_SYMBOL(rtnetlink_dump_ifinfo);
 580 EXPORT_SYMBOL(rtnetlink_links);
 581 EXPORT_SYMBOL(rtnetlink_put_metrics);
 582 EXPORT_SYMBOL(rtnl);
 583 EXPORT_SYMBOL(rtnl_lock);
 584 EXPORT_SYMBOL(rtnl_sem);
 585 EXPORT_SYMBOL(rtnl_unlock);