SAFE public projects git trees. - safe/jmp/linux-2.6/blob - net/ipv4/devinet.c

   1 /*
   2  *      NET3    IP device support routines.
   3  *
   4  *      Version: $Id: devinet.c,v 1.44 2001/10/31 21:55:54 davem Exp $
   5  *
   6  *              This program is free software; you can redistribute it and/or
   7  *              modify it under the terms of the GNU General Public License
   8  *              as published by the Free Software Foundation; either version
   9  *              2 of the License, or (at your option) any later version.
  10  *
  11  *      Derived from the IP parts of dev.c 1.0.19
  12  *              Authors:        Ross Biro
  13  *                              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  14  *                              Mark Evans, <evansmp@uhura.aston.ac.uk>
  15  *
  16  *      Additional Authors:
  17  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
  18  *              Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
  19  *
  20  *      Changes:
  21  *              Alexey Kuznetsov:       pa_* fields are replaced with ifaddr
  22  *                                      lists.
  23  *              Cyrus Durgin:           updated for kmod
  24  *              Matthias Andree:        in devinet_ioctl, compare label and
  25  *                                      address (4.4BSD alias style support),
  26  *                                      fall back to comparing just the label
  27  *                                      if no match found.
  28  */
  29
  30
  31 #include <asm/uaccess.h>
  32 #include <asm/system.h>
  33 #include <linux/bitops.h>
  34 #include <linux/capability.h>
  35 #include <linux/module.h>
  36 #include <linux/types.h>
  37 #include <linux/kernel.h>
  38 #include <linux/string.h>
  39 #include <linux/mm.h>
  40 #include <linux/socket.h>
  41 #include <linux/sockios.h>
  42 #include <linux/in.h>
  43 #include <linux/errno.h>
  44 #include <linux/interrupt.h>
  45 #include <linux/if_addr.h>
  46 #include <linux/if_ether.h>
  47 #include <linux/inet.h>
  48 #include <linux/netdevice.h>
  49 #include <linux/etherdevice.h>
  50 #include <linux/skbuff.h>
  51 #include <linux/init.h>
  52 #include <linux/notifier.h>
  53 #include <linux/inetdevice.h>
  54 #include <linux/igmp.h>
  55 #ifdef CONFIG_SYSCTL
  56 #include <linux/sysctl.h>
  57 #endif
  58 #include <linux/kmod.h>
  59
  60 #include <net/arp.h>
  61 #include <net/ip.h>
  62 #include <net/route.h>
  63 #include <net/ip_fib.h>
  64 #include <net/rtnetlink.h>
  65
  66 struct ipv4_devconf ipv4_devconf = {
  67         .data = {
  68                 [NET_IPV4_CONF_ACCEPT_REDIRECTS - 1] = 1,
  69                 [NET_IPV4_CONF_SEND_REDIRECTS - 1] = 1,
  70                 [NET_IPV4_CONF_SECURE_REDIRECTS - 1] = 1,
  71                 [NET_IPV4_CONF_SHARED_MEDIA - 1] = 1,
  72         },
  73 };
  74
  75 static struct ipv4_devconf ipv4_devconf_dflt = {
  76         .data = {
  77                 [NET_IPV4_CONF_ACCEPT_REDIRECTS - 1] = 1,
  78                 [NET_IPV4_CONF_SEND_REDIRECTS - 1] = 1,
  79                 [NET_IPV4_CONF_SECURE_REDIRECTS - 1] = 1,
  80                 [NET_IPV4_CONF_SHARED_MEDIA - 1] = 1,
  81                 [NET_IPV4_CONF_ACCEPT_SOURCE_ROUTE - 1] = 1,
  82         },
  83 };
  84
  85 #define IPV4_DEVCONF_DFLT(attr) IPV4_DEVCONF(ipv4_devconf_dflt, attr)
  86
  87 static const struct nla_policy ifa_ipv4_policy[IFA_MAX+1] = {
  88         [IFA_LOCAL]             = { .type = NLA_U32 },
  89         [IFA_ADDRESS]           = { .type = NLA_U32 },
  90         [IFA_BROADCAST]         = { .type = NLA_U32 },
  91         [IFA_ANYCAST]           = { .type = NLA_U32 },
  92         [IFA_LABEL]             = { .type = NLA_STRING, .len = IFNAMSIZ - 1 },
  93 };
  94
  95 static void rtmsg_ifa(int event, struct in_ifaddr *, struct nlmsghdr *, u32);
  96
  97 static BLOCKING_NOTIFIER_HEAD(inetaddr_chain);
  98 static void inet_del_ifa(struct in_device *in_dev, struct in_ifaddr **ifap,
  99                          int destroy);
 100 #ifdef CONFIG_SYSCTL
 101 static void devinet_sysctl_register(struct in_device *in_dev,
 102                                     struct ipv4_devconf *p);
 103 static void devinet_sysctl_unregister(struct ipv4_devconf *p);
 104 #endif
 105
 106 /* Locks all the inet devices. */
 107
 108 static struct in_ifaddr *inet_alloc_ifa(void)
 109 {
 110         struct in_ifaddr *ifa = kzalloc(sizeof(*ifa), GFP_KERNEL);
 111
 112         if (ifa) {
 113                 INIT_RCU_HEAD(&ifa->rcu_head);
 114         }
 115
 116         return ifa;
 117 }
 118
 119 static void inet_rcu_free_ifa(struct rcu_head *head)
 120 {
 121         struct in_ifaddr *ifa = container_of(head, struct in_ifaddr, rcu_head);
 122         if (ifa->ifa_dev)
 123                 in_dev_put(ifa->ifa_dev);
 124         kfree(ifa);
 125 }
 126
 127 static inline void inet_free_ifa(struct in_ifaddr *ifa)
 128 {
 129         call_rcu(&ifa->rcu_head, inet_rcu_free_ifa);
 130 }
 131
 132 void in_dev_finish_destroy(struct in_device *idev)
 133 {
 134         struct net_device *dev = idev->dev;
 135
 136         BUG_TRAP(!idev->ifa_list);
 137         BUG_TRAP(!idev->mc_list);
 138 #ifdef NET_REFCNT_DEBUG
 139         printk(KERN_DEBUG "in_dev_finish_destroy: %p=%s\n",
 140                idev, dev ? dev->name : "NIL");
 141 #endif
 142         dev_put(dev);
 143         if (!idev->dead)
 144                 printk("Freeing alive in_device %p\n", idev);
 145         else {
 146                 kfree(idev);
 147         }
 148 }
 149
 150 static struct in_device *inetdev_init(struct net_device *dev)
 151 {
 152         struct in_device *in_dev;
 153
 154         ASSERT_RTNL();
 155
 156         in_dev = kzalloc(sizeof(*in_dev), GFP_KERNEL);
 157         if (!in_dev)
 158                 goto out;
 159         INIT_RCU_HEAD(&in_dev->rcu_head);
 160         memcpy(&in_dev->cnf, &ipv4_devconf_dflt, sizeof(in_dev->cnf));
 161         in_dev->cnf.sysctl = NULL;
 162         in_dev->dev = dev;
 163         if ((in_dev->arp_parms = neigh_parms_alloc(dev, &arp_tbl)) == NULL)
 164                 goto out_kfree;
 165         /* Reference in_dev->dev */
 166         dev_hold(dev);
 167 #ifdef CONFIG_SYSCTL
 168         neigh_sysctl_register(dev, in_dev->arp_parms, NET_IPV4,
 169                               NET_IPV4_NEIGH, "ipv4", NULL, NULL);
 170 #endif
 171
 172         /* Account for reference dev->ip_ptr (below) */
 173         in_dev_hold(in_dev);
 174
 175 #ifdef CONFIG_SYSCTL
 176         devinet_sysctl_register(in_dev, &in_dev->cnf);
 177 #endif
 178         ip_mc_init_dev(in_dev);
 179         if (dev->flags & IFF_UP)
 180                 ip_mc_up(in_dev);
 181
 182         /* we can receive as soon as ip_ptr is set -- do this last */
 183         rcu_assign_pointer(dev->ip_ptr, in_dev);
 184 out:
 185         return in_dev;
 186 out_kfree:
 187         kfree(in_dev);
 188         in_dev = NULL;
 189         goto out;
 190 }
 191
 192 static void in_dev_rcu_put(struct rcu_head *head)
 193 {
 194         struct in_device *idev = container_of(head, struct in_device, rcu_head);
 195         in_dev_put(idev);
 196 }
 197
 198 static void inetdev_destroy(struct in_device *in_dev)
 199 {
 200         struct in_ifaddr *ifa;
 201         struct net_device *dev;
 202
 203         ASSERT_RTNL();
 204
 205         dev = in_dev->dev;
 206
 207         in_dev->dead = 1;
 208
 209         ip_mc_destroy_dev(in_dev);
 210
 211         while ((ifa = in_dev->ifa_list) != NULL) {
 212                 inet_del_ifa(in_dev, &in_dev->ifa_list, 0);
 213                 inet_free_ifa(ifa);
 214         }
 215
 216 #ifdef CONFIG_SYSCTL
 217         devinet_sysctl_unregister(&in_dev->cnf);
 218 #endif
 219
 220         dev->ip_ptr = NULL;
 221
 222 #ifdef CONFIG_SYSCTL
 223         neigh_sysctl_unregister(in_dev->arp_parms);
 224 #endif
 225         neigh_parms_release(&arp_tbl, in_dev->arp_parms);
 226         arp_ifdown(dev);
 227
 228         call_rcu(&in_dev->rcu_head, in_dev_rcu_put);
 229 }
 230
 231 int inet_addr_onlink(struct in_device *in_dev, __be32 a, __be32 b)
 232 {
 233         rcu_read_lock();
 234         for_primary_ifa(in_dev) {
 235                 if (inet_ifa_match(a, ifa)) {
 236                         if (!b || inet_ifa_match(b, ifa)) {
 237                                 rcu_read_unlock();
 238                                 return 1;
 239                         }
 240                 }
 241         } endfor_ifa(in_dev);
 242         rcu_read_unlock();
 243         return 0;
 244 }
 245
 246 static void __inet_del_ifa(struct in_device *in_dev, struct in_ifaddr **ifap,
 247                          int destroy, struct nlmsghdr *nlh, u32 pid)
 248 {
 249         struct in_ifaddr *promote = NULL;
 250         struct in_ifaddr *ifa, *ifa1 = *ifap;
 251         struct in_ifaddr *last_prim = in_dev->ifa_list;
 252         struct in_ifaddr *prev_prom = NULL;
 253         int do_promote = IN_DEV_PROMOTE_SECONDARIES(in_dev);
 254
 255         ASSERT_RTNL();
 256
 257         /* 1. Deleting primary ifaddr forces deletion all secondaries
 258          * unless alias promotion is set
 259          **/
 260
 261         if (!(ifa1->ifa_flags & IFA_F_SECONDARY)) {
 262                 struct in_ifaddr **ifap1 = &ifa1->ifa_next;
 263
 264                 while ((ifa = *ifap1) != NULL) {
 265                         if (!(ifa->ifa_flags & IFA_F_SECONDARY) &&
 266                             ifa1->ifa_scope <= ifa->ifa_scope)
 267                                 last_prim = ifa;
 268
 269                         if (!(ifa->ifa_flags & IFA_F_SECONDARY) ||
 270                             ifa1->ifa_mask != ifa->ifa_mask ||
 271                             !inet_ifa_match(ifa1->ifa_address, ifa)) {
 272                                 ifap1 = &ifa->ifa_next;
 273                                 prev_prom = ifa;
 274                                 continue;
 275                         }
 276
 277                         if (!do_promote) {
 278                                 *ifap1 = ifa->ifa_next;
 279
 280                                 rtmsg_ifa(RTM_DELADDR, ifa, nlh, pid);
 281                                 blocking_notifier_call_chain(&inetaddr_chain,
 282                                                 NETDEV_DOWN, ifa);
 283                                 inet_free_ifa(ifa);
 284                         } else {
 285                                 promote = ifa;
 286                                 break;
 287                         }
 288                 }
 289         }
 290
 291         /* 2. Unlink it */
 292
 293         *ifap = ifa1->ifa_next;
 294
 295         /* 3. Announce address deletion */
 296
 297         /* Send message first, then call notifier.
 298            At first sight, FIB update triggered by notifier
 299            will refer to already deleted ifaddr, that could confuse
 300            netlink listeners. It is not true: look, gated sees
 301            that route deleted and if it still thinks that ifaddr
 302            is valid, it will try to restore deleted routes... Grr.
 303            So that, this order is correct.
 304          */
 305         rtmsg_ifa(RTM_DELADDR, ifa1, nlh, pid);
 306         blocking_notifier_call_chain(&inetaddr_chain, NETDEV_DOWN, ifa1);
 307
 308         if (promote) {
 309
 310                 if (prev_prom) {
 311                         prev_prom->ifa_next = promote->ifa_next;
 312                         promote->ifa_next = last_prim->ifa_next;
 313                         last_prim->ifa_next = promote;
 314                 }
 315
 316                 promote->ifa_flags &= ~IFA_F_SECONDARY;
 317                 rtmsg_ifa(RTM_NEWADDR, promote, nlh, pid);
 318                 blocking_notifier_call_chain(&inetaddr_chain,
 319                                 NETDEV_UP, promote);
 320                 for (ifa = promote->ifa_next; ifa; ifa = ifa->ifa_next) {
 321                         if (ifa1->ifa_mask != ifa->ifa_mask ||
 322                             !inet_ifa_match(ifa1->ifa_address, ifa))
 323                                         continue;
 324                         fib_add_ifaddr(ifa);
 325                 }
 326
 327         }
 328         if (destroy)
 329                 inet_free_ifa(ifa1);
 330 }
 331
 332 static void inet_del_ifa(struct in_device *in_dev, struct in_ifaddr **ifap,
 333                          int destroy)
 334 {
 335         __inet_del_ifa(in_dev, ifap, destroy, NULL, 0);
 336 }
 337
 338 static int __inet_insert_ifa(struct in_ifaddr *ifa, struct nlmsghdr *nlh,
 339                              u32 pid)
 340 {
 341         struct in_device *in_dev = ifa->ifa_dev;
 342         struct in_ifaddr *ifa1, **ifap, **last_primary;
 343
 344         ASSERT_RTNL();
 345
 346         if (!ifa->ifa_local) {
 347                 inet_free_ifa(ifa);
 348                 return 0;
 349         }
 350
 351         ifa->ifa_flags &= ~IFA_F_SECONDARY;
 352         last_primary = &in_dev->ifa_list;
 353
 354         for (ifap = &in_dev->ifa_list; (ifa1 = *ifap) != NULL;
 355              ifap = &ifa1->ifa_next) {
 356                 if (!(ifa1->ifa_flags & IFA_F_SECONDARY) &&
 357                     ifa->ifa_scope <= ifa1->ifa_scope)
 358                         last_primary = &ifa1->ifa_next;
 359                 if (ifa1->ifa_mask == ifa->ifa_mask &&
 360                     inet_ifa_match(ifa1->ifa_address, ifa)) {
 361                         if (ifa1->ifa_local == ifa->ifa_local) {
 362                                 inet_free_ifa(ifa);
 363                                 return -EEXIST;
 364                         }
 365                         if (ifa1->ifa_scope != ifa->ifa_scope) {
 366                                 inet_free_ifa(ifa);
 367                                 return -EINVAL;
 368                         }
 369                         ifa->ifa_flags |= IFA_F_SECONDARY;
 370                 }
 371         }
 372
 373         if (!(ifa->ifa_flags & IFA_F_SECONDARY)) {
 374                 net_srandom(ifa->ifa_local);
 375                 ifap = last_primary;
 376         }
 377
 378         ifa->ifa_next = *ifap;
 379         *ifap = ifa;
 380
 381         /* Send message first, then call notifier.
 382            Notifier will trigger FIB update, so that
 383            listeners of netlink will know about new ifaddr */
 384         rtmsg_ifa(RTM_NEWADDR, ifa, nlh, pid);
 385         blocking_notifier_call_chain(&inetaddr_chain, NETDEV_UP, ifa);
 386
 387         return 0;
 388 }
 389
 390 static int inet_insert_ifa(struct in_ifaddr *ifa)
 391 {
 392         return __inet_insert_ifa(ifa, NULL, 0);
 393 }
 394
 395 static int inet_set_ifa(struct net_device *dev, struct in_ifaddr *ifa)
 396 {
 397         struct in_device *in_dev = __in_dev_get_rtnl(dev);
 398
 399         ASSERT_RTNL();
 400
 401         if (!in_dev) {
 402                 inet_free_ifa(ifa);
 403                 return -ENOBUFS;
 404         }
 405         ipv4_devconf_setall(in_dev);
 406         if (ifa->ifa_dev != in_dev) {
 407                 BUG_TRAP(!ifa->ifa_dev);
 408                 in_dev_hold(in_dev);
 409                 ifa->ifa_dev = in_dev;
 410         }
 411         if (LOOPBACK(ifa->ifa_local))
 412                 ifa->ifa_scope = RT_SCOPE_HOST;
 413         return inet_insert_ifa(ifa);
 414 }
 415
 416 struct in_device *inetdev_by_index(int ifindex)
 417 {
 418         struct net_device *dev;
 419         struct in_device *in_dev = NULL;
 420         read_lock(&dev_base_lock);
 421         dev = __dev_get_by_index(&init_net, ifindex);
 422         if (dev)
 423                 in_dev = in_dev_get(dev);
 424         read_unlock(&dev_base_lock);
 425         return in_dev;
 426 }
 427
 428 /* Called only from RTNL semaphored context. No locks. */
 429
 430 struct in_ifaddr *inet_ifa_byprefix(struct in_device *in_dev, __be32 prefix,
 431                                     __be32 mask)
 432 {
 433         ASSERT_RTNL();
 434
 435         for_primary_ifa(in_dev) {
 436                 if (ifa->ifa_mask == mask && inet_ifa_match(prefix, ifa))
 437                         return ifa;
 438         } endfor_ifa(in_dev);
 439         return NULL;
 440 }
 441
 442 static int inet_rtm_deladdr(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
 443 {
 444         struct nlattr *tb[IFA_MAX+1];
 445         struct in_device *in_dev;
 446         struct ifaddrmsg *ifm;
 447         struct in_ifaddr *ifa, **ifap;
 448         int err = -EINVAL;
 449
 450         ASSERT_RTNL();
 451
 452         err = nlmsg_parse(nlh, sizeof(*ifm), tb, IFA_MAX, ifa_ipv4_policy);
 453         if (err < 0)
 454                 goto errout;
 455
 456         ifm = nlmsg_data(nlh);
 457         in_dev = inetdev_by_index(ifm->ifa_index);
 458         if (in_dev == NULL) {
 459                 err = -ENODEV;
 460                 goto errout;
 461         }
 462
 463         __in_dev_put(in_dev);
 464
 465         for (ifap = &in_dev->ifa_list; (ifa = *ifap) != NULL;
 466              ifap = &ifa->ifa_next) {
 467                 if (tb[IFA_LOCAL] &&
 468                     ifa->ifa_local != nla_get_be32(tb[IFA_LOCAL]))
 469                         continue;
 470
 471                 if (tb[IFA_LABEL] && nla_strcmp(tb[IFA_LABEL], ifa->ifa_label))
 472                         continue;
 473
 474                 if (tb[IFA_ADDRESS] &&
 475                     (ifm->ifa_prefixlen != ifa->ifa_prefixlen ||
 476                     !inet_ifa_match(nla_get_be32(tb[IFA_ADDRESS]), ifa)))
 477                         continue;
 478
 479                 __inet_del_ifa(in_dev, ifap, 1, nlh, NETLINK_CB(skb).pid);
 480                 return 0;
 481         }
 482
 483         err = -EADDRNOTAVAIL;
 484 errout:
 485         return err;
 486 }
 487
 488 static struct in_ifaddr *rtm_to_ifaddr(struct nlmsghdr *nlh)
 489 {
 490         struct nlattr *tb[IFA_MAX+1];
 491         struct in_ifaddr *ifa;
 492         struct ifaddrmsg *ifm;
 493         struct net_device *dev;
 494         struct in_device *in_dev;
 495         int err = -EINVAL;
 496
 497         err = nlmsg_parse(nlh, sizeof(*ifm), tb, IFA_MAX, ifa_ipv4_policy);
 498         if (err < 0)
 499                 goto errout;
 500
 501         ifm = nlmsg_data(nlh);
 502         if (ifm->ifa_prefixlen > 32 || tb[IFA_LOCAL] == NULL) {
 503                 err = -EINVAL;
 504                 goto errout;
 505         }
 506
 507         dev = __dev_get_by_index(&init_net, ifm->ifa_index);
 508         if (dev == NULL) {
 509                 err = -ENODEV;
 510                 goto errout;
 511         }
 512
 513         in_dev = __in_dev_get_rtnl(dev);
 514         if (in_dev == NULL) {
 515                 err = -ENOBUFS;
 516                 goto errout;
 517         }
 518
 519         ifa = inet_alloc_ifa();
 520         if (ifa == NULL) {
 521                 /*
 522                  * A potential indev allocation can be left alive, it stays
 523                  * assigned to its device and is destroy with it.
 524                  */
 525                 err = -ENOBUFS;
 526                 goto errout;
 527         }
 528
 529         ipv4_devconf_setall(in_dev);
 530         in_dev_hold(in_dev);
 531
 532         if (tb[IFA_ADDRESS] == NULL)
 533                 tb[IFA_ADDRESS] = tb[IFA_LOCAL];
 534
 535         ifa->ifa_prefixlen = ifm->ifa_prefixlen;
 536         ifa->ifa_mask = inet_make_mask(ifm->ifa_prefixlen);
 537         ifa->ifa_flags = ifm->ifa_flags;
 538         ifa->ifa_scope = ifm->ifa_scope;
 539         ifa->ifa_dev = in_dev;
 540
 541         ifa->ifa_local = nla_get_be32(tb[IFA_LOCAL]);
 542         ifa->ifa_address = nla_get_be32(tb[IFA_ADDRESS]);
 543
 544         if (tb[IFA_BROADCAST])
 545                 ifa->ifa_broadcast = nla_get_be32(tb[IFA_BROADCAST]);
 546
 547         if (tb[IFA_ANYCAST])
 548                 ifa->ifa_anycast = nla_get_be32(tb[IFA_ANYCAST]);
 549
 550         if (tb[IFA_LABEL])
 551                 nla_strlcpy(ifa->ifa_label, tb[IFA_LABEL], IFNAMSIZ);
 552         else
 553                 memcpy(ifa->ifa_label, dev->name, IFNAMSIZ);
 554
 555         return ifa;
 556
 557 errout:
 558         return ERR_PTR(err);
 559 }
 560
 561 static int inet_rtm_newaddr(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
 562 {
 563         struct in_ifaddr *ifa;
 564
 565         ASSERT_RTNL();
 566
 567         ifa = rtm_to_ifaddr(nlh);
 568         if (IS_ERR(ifa))
 569                 return PTR_ERR(ifa);
 570
 571         return __inet_insert_ifa(ifa, nlh, NETLINK_CB(skb).pid);
 572 }
 573
 574 /*
 575  *      Determine a default network mask, based on the IP address.
 576  */
 577
 578 static __inline__ int inet_abc_len(__be32 addr)
 579 {
 580         int rc = -1;    /* Something else, probably a multicast. */
 581
 582         if (ZERONET(addr))
 583                 rc = 0;
 584         else {
 585                 __u32 haddr = ntohl(addr);
 586
 587                 if (IN_CLASSA(haddr))
 588                         rc = 8;
 589                 else if (IN_CLASSB(haddr))
 590                         rc = 16;
 591                 else if (IN_CLASSC(haddr))
 592                         rc = 24;
 593         }
 594
 595         return rc;
 596 }
 597
 598
 599 int devinet_ioctl(unsigned int cmd, void __user *arg)
 600 {
 601         struct ifreq ifr;
 602         struct sockaddr_in sin_orig;
 603         struct sockaddr_in *sin = (struct sockaddr_in *)&ifr.ifr_addr;
 604         struct in_device *in_dev;
 605         struct in_ifaddr **ifap = NULL;
 606         struct in_ifaddr *ifa = NULL;
 607         struct net_device *dev;
 608         char *colon;
 609         int ret = -EFAULT;
 610         int tryaddrmatch = 0;
 611
 612         /*
 613          *      Fetch the caller's info block into kernel space
 614          */
 615
 616         if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
 617                 goto out;
 618         ifr.ifr_name[IFNAMSIZ - 1] = 0;
 619
 620         /* save original address for comparison */
 621         memcpy(&sin_orig, sin, sizeof(*sin));
 622
 623         colon = strchr(ifr.ifr_name, ':');
 624         if (colon)
 625                 *colon = 0;
 626
 627 #ifdef CONFIG_KMOD
 628         dev_load(&init_net, ifr.ifr_name);
 629 #endif
 630
 631         switch (cmd) {
 632         case SIOCGIFADDR:       /* Get interface address */
 633         case SIOCGIFBRDADDR:    /* Get the broadcast address */
 634         case SIOCGIFDSTADDR:    /* Get the destination address */
 635         case SIOCGIFNETMASK:    /* Get the netmask for the interface */
 636                 /* Note that these ioctls will not sleep,
 637                    so that we do not impose a lock.
 638                    One day we will be forced to put shlock here (I mean SMP)
 639                  */
 640                 tryaddrmatch = (sin_orig.sin_family == AF_INET);
 641                 memset(sin, 0, sizeof(*sin));
 642                 sin->sin_family = AF_INET;
 643                 break;
 644
 645         case SIOCSIFFLAGS:
 646                 ret = -EACCES;
 647                 if (!capable(CAP_NET_ADMIN))
 648                         goto out;
 649                 break;
 650         case SIOCSIFADDR:       /* Set interface address (and family) */
 651         case SIOCSIFBRDADDR:    /* Set the broadcast address */
 652         case SIOCSIFDSTADDR:    /* Set the destination address */
 653         case SIOCSIFNETMASK:    /* Set the netmask for the interface */
 654                 ret = -EACCES;
 655                 if (!capable(CAP_NET_ADMIN))
 656                         goto out;
 657                 ret = -EINVAL;
 658                 if (sin->sin_family != AF_INET)
 659                         goto out;
 660                 break;
 661         default:
 662                 ret = -EINVAL;
 663                 goto out;
 664         }
 665
 666         rtnl_lock();
 667
 668         ret = -ENODEV;
 669         if ((dev = __dev_get_by_name(&init_net, ifr.ifr_name)) == NULL)
 670                 goto done;
 671
 672         if (colon)
 673                 *colon = ':';
 674
 675         if ((in_dev = __in_dev_get_rtnl(dev)) != NULL) {
 676                 if (tryaddrmatch) {
 677                         /* Matthias Andree */
 678                         /* compare label and address (4.4BSD style) */
 679                         /* note: we only do this for a limited set of ioctls
 680                            and only if the original address family was AF_INET.
 681                            This is checked above. */
 682                         for (ifap = &in_dev->ifa_list; (ifa = *ifap) != NULL;
 683                              ifap = &ifa->ifa_next) {
 684                                 if (!strcmp(ifr.ifr_name, ifa->ifa_label) &&
 685                                     sin_orig.sin_addr.s_addr ==
 686                                                         ifa->ifa_address) {
 687                                         break; /* found */
 688                                 }
 689                         }
 690                 }
 691                 /* we didn't get a match, maybe the application is
 692                    4.3BSD-style and passed in junk so we fall back to
 693                    comparing just the label */
 694                 if (!ifa) {
 695                         for (ifap = &in_dev->ifa_list; (ifa = *ifap) != NULL;
 696                              ifap = &ifa->ifa_next)
 697                                 if (!strcmp(ifr.ifr_name, ifa->ifa_label))
 698                                         break;
 699                 }
 700         }
 701
 702         ret = -EADDRNOTAVAIL;
 703         if (!ifa && cmd != SIOCSIFADDR && cmd != SIOCSIFFLAGS)
 704                 goto done;
 705
 706         switch (cmd) {
 707         case SIOCGIFADDR:       /* Get interface address */
 708                 sin->sin_addr.s_addr = ifa->ifa_local;
 709                 goto rarok;
 710
 711         case SIOCGIFBRDADDR:    /* Get the broadcast address */
 712                 sin->sin_addr.s_addr = ifa->ifa_broadcast;
 713                 goto rarok;
 714
 715         case SIOCGIFDSTADDR:    /* Get the destination address */
 716                 sin->sin_addr.s_addr = ifa->ifa_address;
 717                 goto rarok;
 718
 719         case SIOCGIFNETMASK:    /* Get the netmask for the interface */
 720                 sin->sin_addr.s_addr = ifa->ifa_mask;
 721                 goto rarok;
 722
 723         case SIOCSIFFLAGS:
 724                 if (colon) {
 725                         ret = -EADDRNOTAVAIL;
 726                         if (!ifa)
 727                                 break;
 728                         ret = 0;
 729                         if (!(ifr.ifr_flags & IFF_UP))
 730                                 inet_del_ifa(in_dev, ifap, 1);
 731                         break;
 732                 }
 733                 ret = dev_change_flags(dev, ifr.ifr_flags);
 734                 break;
 735
 736         case SIOCSIFADDR:       /* Set interface address (and family) */
 737                 ret = -EINVAL;
 738                 if (inet_abc_len(sin->sin_addr.s_addr) < 0)
 739                         break;
 740
 741                 if (!ifa) {
 742                         ret = -ENOBUFS;
 743                         if ((ifa = inet_alloc_ifa()) == NULL)
 744                                 break;
 745                         if (colon)
 746                                 memcpy(ifa->ifa_label, ifr.ifr_name, IFNAMSIZ);
 747                         else
 748                                 memcpy(ifa->ifa_label, dev->name, IFNAMSIZ);
 749                 } else {
 750                         ret = 0;
 751                         if (ifa->ifa_local == sin->sin_addr.s_addr)
 752                                 break;
 753                         inet_del_ifa(in_dev, ifap, 0);
 754                         ifa->ifa_broadcast = 0;
 755                         ifa->ifa_anycast = 0;
 756                 }
 757
 758                 ifa->ifa_address = ifa->ifa_local = sin->sin_addr.s_addr;
 759
 760                 if (!(dev->flags & IFF_POINTOPOINT)) {
 761                         ifa->ifa_prefixlen = inet_abc_len(ifa->ifa_address);
 762                         ifa->ifa_mask = inet_make_mask(ifa->ifa_prefixlen);
 763                         if ((dev->flags & IFF_BROADCAST) &&
 764                             ifa->ifa_prefixlen < 31)
 765                                 ifa->ifa_broadcast = ifa->ifa_address |
 766                                                      ~ifa->ifa_mask;
 767                 } else {
 768                         ifa->ifa_prefixlen = 32;
 769                         ifa->ifa_mask = inet_make_mask(32);
 770                 }
 771                 ret = inet_set_ifa(dev, ifa);
 772                 break;
 773
 774         case SIOCSIFBRDADDR:    /* Set the broadcast address */
 775                 ret = 0;
 776                 if (ifa->ifa_broadcast != sin->sin_addr.s_addr) {
 777                         inet_del_ifa(in_dev, ifap, 0);
 778                         ifa->ifa_broadcast = sin->sin_addr.s_addr;
 779                         inet_insert_ifa(ifa);
 780                 }
 781                 break;
 782
 783         case SIOCSIFDSTADDR:    /* Set the destination address */
 784                 ret = 0;
 785                 if (ifa->ifa_address == sin->sin_addr.s_addr)
 786                         break;
 787                 ret = -EINVAL;
 788                 if (inet_abc_len(sin->sin_addr.s_addr) < 0)
 789                         break;
 790                 ret = 0;
 791                 inet_del_ifa(in_dev, ifap, 0);
 792                 ifa->ifa_address = sin->sin_addr.s_addr;
 793                 inet_insert_ifa(ifa);
 794                 break;
 795
 796         case SIOCSIFNETMASK:    /* Set the netmask for the interface */
 797
 798                 /*
 799                  *      The mask we set must be legal.
 800                  */
 801                 ret = -EINVAL;
 802                 if (bad_mask(sin->sin_addr.s_addr, 0))
 803                         break;
 804                 ret = 0;
 805                 if (ifa->ifa_mask != sin->sin_addr.s_addr) {
 806                         __be32 old_mask = ifa->ifa_mask;
 807                         inet_del_ifa(in_dev, ifap, 0);
 808                         ifa->ifa_mask = sin->sin_addr.s_addr;
 809                         ifa->ifa_prefixlen = inet_mask_len(ifa->ifa_mask);
 810
 811                         /* See if current broadcast address matches
 812                          * with current netmask, then recalculate
 813                          * the broadcast address. Otherwise it's a
 814                          * funny address, so don't touch it since
 815                          * the user seems to know what (s)he's doing...
 816                          */
 817                         if ((dev->flags & IFF_BROADCAST) &&
 818                             (ifa->ifa_prefixlen < 31) &&
 819                             (ifa->ifa_broadcast ==
 820                              (ifa->ifa_local|~old_mask))) {
 821                                 ifa->ifa_broadcast = (ifa->ifa_local |
 822                                                       ~sin->sin_addr.s_addr);
 823                         }
 824                         inet_insert_ifa(ifa);
 825                 }
 826                 break;
 827         }
 828 done:
 829         rtnl_unlock();
 830 out:
 831         return ret;
 832 rarok:
 833         rtnl_unlock();
 834         ret = copy_to_user(arg, &ifr, sizeof(struct ifreq)) ? -EFAULT : 0;
 835         goto out;
 836 }
 837
 838 static int inet_gifconf(struct net_device *dev, char __user *buf, int len)
 839 {
 840         struct in_device *in_dev = __in_dev_get_rtnl(dev);
 841         struct in_ifaddr *ifa;
 842         struct ifreq ifr;
 843         int done = 0;
 844
 845         if (!in_dev || (ifa = in_dev->ifa_list) == NULL)
 846                 goto out;
 847
 848         for (; ifa; ifa = ifa->ifa_next) {
 849                 if (!buf) {
 850                         done += sizeof(ifr);
 851                         continue;
 852                 }
 853                 if (len < (int) sizeof(ifr))
 854                         break;
 855                 memset(&ifr, 0, sizeof(struct ifreq));
 856                 if (ifa->ifa_label)
 857                         strcpy(ifr.ifr_name, ifa->ifa_label);
 858                 else
 859                         strcpy(ifr.ifr_name, dev->name);
 860
 861                 (*(struct sockaddr_in *)&ifr.ifr_addr).sin_family = AF_INET;
 862                 (*(struct sockaddr_in *)&ifr.ifr_addr).sin_addr.s_addr =
 863                                                                 ifa->ifa_local;
 864
 865                 if (copy_to_user(buf, &ifr, sizeof(struct ifreq))) {
 866                         done = -EFAULT;
 867                         break;
 868                 }
 869                 buf  += sizeof(struct ifreq);
 870                 len  -= sizeof(struct ifreq);
 871                 done += sizeof(struct ifreq);
 872         }
 873 out:
 874         return done;
 875 }
 876
 877 __be32 inet_select_addr(const struct net_device *dev, __be32 dst, int scope)
 878 {
 879         __be32 addr = 0;
 880         struct in_device *in_dev;
 881
 882         rcu_read_lock();
 883         in_dev = __in_dev_get_rcu(dev);
 884         if (!in_dev)
 885                 goto no_in_dev;
 886
 887         for_primary_ifa(in_dev) {
 888                 if (ifa->ifa_scope > scope)
 889                         continue;
 890                 if (!dst || inet_ifa_match(dst, ifa)) {
 891                         addr = ifa->ifa_local;
 892                         break;
 893                 }
 894                 if (!addr)
 895                         addr = ifa->ifa_local;
 896         } endfor_ifa(in_dev);
 897 no_in_dev:
 898         rcu_read_unlock();
 899
 900         if (addr)
 901                 goto out;
 902
 903         /* Not loopback addresses on loopback should be preferred
 904            in this case. It is importnat that lo is the first interface
 905            in dev_base list.
 906          */
 907         read_lock(&dev_base_lock);
 908         rcu_read_lock();
 909         for_each_netdev(&init_net, dev) {
 910                 if ((in_dev = __in_dev_get_rcu(dev)) == NULL)
 911                         continue;
 912
 913                 for_primary_ifa(in_dev) {
 914                         if (ifa->ifa_scope != RT_SCOPE_LINK &&
 915                             ifa->ifa_scope <= scope) {
 916                                 addr = ifa->ifa_local;
 917                                 goto out_unlock_both;
 918                         }
 919                 } endfor_ifa(in_dev);
 920         }
 921 out_unlock_both:
 922         read_unlock(&dev_base_lock);
 923         rcu_read_unlock();
 924 out:
 925         return addr;
 926 }
 927
 928 static __be32 confirm_addr_indev(struct in_device *in_dev, __be32 dst,
 929                               __be32 local, int scope)
 930 {
 931         int same = 0;
 932         __be32 addr = 0;
 933
 934         for_ifa(in_dev) {
 935                 if (!addr &&
 936                     (local == ifa->ifa_local || !local) &&
 937                     ifa->ifa_scope <= scope) {
 938                         addr = ifa->ifa_local;
 939                         if (same)
 940                                 break;
 941                 }
 942                 if (!same) {
 943                         same = (!local || inet_ifa_match(local, ifa)) &&
 944                                 (!dst || inet_ifa_match(dst, ifa));
 945                         if (same && addr) {
 946                                 if (local || !dst)
 947                                         break;
 948                                 /* Is the selected addr into dst subnet? */
 949                                 if (inet_ifa_match(addr, ifa))
 950                                         break;
 951                                 /* No, then can we use new local src? */
 952                                 if (ifa->ifa_scope <= scope) {
 953                                         addr = ifa->ifa_local;
 954                                         break;
 955                                 }
 956                                 /* search for large dst subnet for addr */
 957                                 same = 0;
 958                         }
 959                 }
 960         } endfor_ifa(in_dev);
 961
 962         return same? addr : 0;
 963 }
 964
 965 /*
 966  * Confirm that local IP address exists using wildcards:
 967  * - dev: only on this interface, 0=any interface
 968  * - dst: only in the same subnet as dst, 0=any dst
 969  * - local: address, 0=autoselect the local address
 970  * - scope: maximum allowed scope value for the local address
 971  */
 972 __be32 inet_confirm_addr(const struct net_device *dev, __be32 dst, __be32 local, int scope)
 973 {
 974         __be32 addr = 0;
 975         struct in_device *in_dev;
 976
 977         if (dev) {
 978                 rcu_read_lock();
 979                 if ((in_dev = __in_dev_get_rcu(dev)))
 980                         addr = confirm_addr_indev(in_dev, dst, local, scope);
 981                 rcu_read_unlock();
 982
 983                 return addr;
 984         }
 985
 986         read_lock(&dev_base_lock);
 987         rcu_read_lock();
 988         for_each_netdev(&init_net, dev) {
 989                 if ((in_dev = __in_dev_get_rcu(dev))) {
 990                         addr = confirm_addr_indev(in_dev, dst, local, scope);
 991                         if (addr)
 992                                 break;
 993                 }
 994         }
 995         rcu_read_unlock();
 996         read_unlock(&dev_base_lock);
 997
 998         return addr;
 999 }
1000
1001 /*
1002  *      Device notifier
1003  */
1004
1005 int register_inetaddr_notifier(struct notifier_block *nb)
1006 {
1007         return blocking_notifier_chain_register(&inetaddr_chain, nb);
1008 }
1009
1010 int unregister_inetaddr_notifier(struct notifier_block *nb)
1011 {
1012         return blocking_notifier_chain_unregister(&inetaddr_chain, nb);
1013 }
1014
1015 /* Rename ifa_labels for a device name change. Make some effort to preserve existing
1016  * alias numbering and to create unique labels if possible.
1017 */
1018 static void inetdev_changename(struct net_device *dev, struct in_device *in_dev)
1019 {
1020         struct in_ifaddr *ifa;
1021         int named = 0;
1022
1023         for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
1024                 char old[IFNAMSIZ], *dot;
1025
1026                 memcpy(old, ifa->ifa_label, IFNAMSIZ);
1027                 memcpy(ifa->ifa_label, dev->name, IFNAMSIZ);
1028                 if (named++ == 0)
1029                         continue;
1030                 dot = strchr(old, ':');
1031                 if (dot == NULL) {
1032                         sprintf(old, ":%d", named);
1033                         dot = old;
1034                 }
1035                 if (strlen(dot) + strlen(dev->name) < IFNAMSIZ) {
1036                         strcat(ifa->ifa_label, dot);
1037                 } else {
1038                         strcpy(ifa->ifa_label + (IFNAMSIZ - strlen(dot) - 1), dot);
1039                 }
1040         }
1041 }
1042
1043 /* Called only under RTNL semaphore */
1044
1045 static int inetdev_event(struct notifier_block *this, unsigned long event,
1046                          void *ptr)
1047 {
1048         struct net_device *dev = ptr;
1049         struct in_device *in_dev = __in_dev_get_rtnl(dev);
1050
1051         if (dev->nd_net != &init_net)
1052                 return NOTIFY_DONE;
1053
1054         ASSERT_RTNL();
1055
1056         if (!in_dev) {
1057                 if (event == NETDEV_REGISTER) {
1058                         in_dev = inetdev_init(dev);
1059                         if (!in_dev)
1060                                 return notifier_from_errno(-ENOMEM);
1061                         if (dev->flags & IFF_LOOPBACK) {
1062                                 IN_DEV_CONF_SET(in_dev, NOXFRM, 1);
1063                                 IN_DEV_CONF_SET(in_dev, NOPOLICY, 1);
1064                         }
1065                 }
1066                 goto out;
1067         }
1068
1069         switch (event) {
1070         case NETDEV_REGISTER:
1071                 printk(KERN_DEBUG "inetdev_event: bug\n");
1072                 dev->ip_ptr = NULL;
1073                 break;
1074         case NETDEV_UP:
1075                 if (dev->mtu < 68)
1076                         break;
1077                 if (dev->flags & IFF_LOOPBACK) {
1078                         struct in_ifaddr *ifa;
1079                         if ((ifa = inet_alloc_ifa()) != NULL) {
1080                                 ifa->ifa_local =
1081                                   ifa->ifa_address = htonl(INADDR_LOOPBACK);
1082                                 ifa->ifa_prefixlen = 8;
1083                                 ifa->ifa_mask = inet_make_mask(8);
1084                                 in_dev_hold(in_dev);
1085                                 ifa->ifa_dev = in_dev;
1086                                 ifa->ifa_scope = RT_SCOPE_HOST;
1087                                 memcpy(ifa->ifa_label, dev->name, IFNAMSIZ);
1088                                 inet_insert_ifa(ifa);
1089                         }
1090                 }
1091                 ip_mc_up(in_dev);
1092                 break;
1093         case NETDEV_DOWN:
1094                 ip_mc_down(in_dev);
1095                 break;
1096         case NETDEV_CHANGEMTU:
1097                 if (dev->mtu >= 68)
1098                         break;
1099                 /* MTU falled under 68, disable IP */
1100         case NETDEV_UNREGISTER:
1101                 inetdev_destroy(in_dev);
1102                 break;
1103         case NETDEV_CHANGENAME:
1104                 /* Do not notify about label change, this event is
1105                  * not interesting to applications using netlink.
1106                  */
1107                 inetdev_changename(dev, in_dev);
1108
1109 #ifdef CONFIG_SYSCTL
1110                 devinet_sysctl_unregister(&in_dev->cnf);
1111                 neigh_sysctl_unregister(in_dev->arp_parms);
1112                 neigh_sysctl_register(dev, in_dev->arp_parms, NET_IPV4,
1113                                       NET_IPV4_NEIGH, "ipv4", NULL, NULL);
1114                 devinet_sysctl_register(in_dev, &in_dev->cnf);
1115 #endif
1116                 break;
1117         }
1118 out:
1119         return NOTIFY_DONE;
1120 }
1121
1122 static struct notifier_block ip_netdev_notifier = {
1123         .notifier_call =inetdev_event,
1124 };
1125
1126 static inline size_t inet_nlmsg_size(void)
1127 {
1128         return NLMSG_ALIGN(sizeof(struct ifaddrmsg))
1129                + nla_total_size(4) /* IFA_ADDRESS */
1130                + nla_total_size(4) /* IFA_LOCAL */
1131                + nla_total_size(4) /* IFA_BROADCAST */
1132                + nla_total_size(4) /* IFA_ANYCAST */
1133                + nla_total_size(IFNAMSIZ); /* IFA_LABEL */
1134 }
1135
1136 static int inet_fill_ifaddr(struct sk_buff *skb, struct in_ifaddr *ifa,
1137                             u32 pid, u32 seq, int event, unsigned int flags)
1138 {
1139         struct ifaddrmsg *ifm;
1140         struct nlmsghdr  *nlh;
1141
1142         nlh = nlmsg_put(skb, pid, seq, event, sizeof(*ifm), flags);
1143         if (nlh == NULL)
1144                 return -EMSGSIZE;
1145
1146         ifm = nlmsg_data(nlh);
1147         ifm->ifa_family = AF_INET;
1148         ifm->ifa_prefixlen = ifa->ifa_prefixlen;
1149         ifm->ifa_flags = ifa->ifa_flags|IFA_F_PERMANENT;
1150         ifm->ifa_scope = ifa->ifa_scope;
1151         ifm->ifa_index = ifa->ifa_dev->dev->ifindex;
1152
1153         if (ifa->ifa_address)
1154                 NLA_PUT_BE32(skb, IFA_ADDRESS, ifa->ifa_address);
1155
1156         if (ifa->ifa_local)
1157                 NLA_PUT_BE32(skb, IFA_LOCAL, ifa->ifa_local);
1158
1159         if (ifa->ifa_broadcast)
1160                 NLA_PUT_BE32(skb, IFA_BROADCAST, ifa->ifa_broadcast);
1161
1162         if (ifa->ifa_anycast)
1163                 NLA_PUT_BE32(skb, IFA_ANYCAST, ifa->ifa_anycast);
1164
1165         if (ifa->ifa_label[0])
1166                 NLA_PUT_STRING(skb, IFA_LABEL, ifa->ifa_label);
1167
1168         return nlmsg_end(skb, nlh);
1169
1170 nla_put_failure:
1171         nlmsg_cancel(skb, nlh);
1172         return -EMSGSIZE;
1173 }
1174
1175 static int inet_dump_ifaddr(struct sk_buff *skb, struct netlink_callback *cb)
1176 {
1177         int idx, ip_idx;
1178         struct net_device *dev;
1179         struct in_device *in_dev;
1180         struct in_ifaddr *ifa;
1181         int s_ip_idx, s_idx = cb->args[0];
1182
1183         s_ip_idx = ip_idx = cb->args[1];
1184         idx = 0;
1185         for_each_netdev(&init_net, dev) {
1186                 if (idx < s_idx)
1187                         goto cont;
1188                 if (idx > s_idx)
1189                         s_ip_idx = 0;
1190                 if ((in_dev = __in_dev_get_rtnl(dev)) == NULL)
1191                         goto cont;
1192
1193                 for (ifa = in_dev->ifa_list, ip_idx = 0; ifa;
1194                      ifa = ifa->ifa_next, ip_idx++) {
1195                         if (ip_idx < s_ip_idx)
1196                                 continue;
1197                         if (inet_fill_ifaddr(skb, ifa, NETLINK_CB(cb->skb).pid,
1198                                              cb->nlh->nlmsg_seq,
1199                                              RTM_NEWADDR, NLM_F_MULTI) <= 0)
1200                                 goto done;
1201                 }
1202 cont:
1203                 idx++;
1204         }
1205
1206 done:
1207         cb->args[0] = idx;
1208         cb->args[1] = ip_idx;
1209
1210         return skb->len;
1211 }
1212
1213 static void rtmsg_ifa(int event, struct in_ifaddr* ifa, struct nlmsghdr *nlh,
1214                       u32 pid)
1215 {
1216         struct sk_buff *skb;
1217         u32 seq = nlh ? nlh->nlmsg_seq : 0;
1218         int err = -ENOBUFS;
1219
1220         skb = nlmsg_new(inet_nlmsg_size(), GFP_KERNEL);
1221         if (skb == NULL)
1222                 goto errout;
1223
1224         err = inet_fill_ifaddr(skb, ifa, pid, seq, event, 0);
1225         if (err < 0) {
1226                 /* -EMSGSIZE implies BUG in inet_nlmsg_size() */
1227                 WARN_ON(err == -EMSGSIZE);
1228                 kfree_skb(skb);
1229                 goto errout;
1230         }
1231         err = rtnl_notify(skb, pid, RTNLGRP_IPV4_IFADDR, nlh, GFP_KERNEL);
1232 errout:
1233         if (err < 0)
1234                 rtnl_set_sk_err(RTNLGRP_IPV4_IFADDR, err);
1235 }
1236
1237 #ifdef CONFIG_SYSCTL
1238
1239 static void devinet_copy_dflt_conf(int i)
1240 {
1241         struct net_device *dev;
1242
1243         read_lock(&dev_base_lock);
1244         for_each_netdev(&init_net, dev) {
1245                 struct in_device *in_dev;
1246                 rcu_read_lock();
1247                 in_dev = __in_dev_get_rcu(dev);
1248                 if (in_dev && !test_bit(i, in_dev->cnf.state))
1249                         in_dev->cnf.data[i] = ipv4_devconf_dflt.data[i];
1250                 rcu_read_unlock();
1251         }
1252         read_unlock(&dev_base_lock);
1253 }
1254
1255 static int devinet_conf_proc(ctl_table *ctl, int write,
1256                              struct file* filp, void __user *buffer,
1257                              size_t *lenp, loff_t *ppos)
1258 {
1259         int ret = proc_dointvec(ctl, write, filp, buffer, lenp, ppos);
1260
1261         if (write) {
1262                 struct ipv4_devconf *cnf = ctl->extra1;
1263                 int i = (int *)ctl->data - cnf->data;
1264
1265                 set_bit(i, cnf->state);
1266
1267                 if (cnf == &ipv4_devconf_dflt)
1268                         devinet_copy_dflt_conf(i);
1269         }
1270
1271         return ret;
1272 }
1273
1274 static int devinet_conf_sysctl(ctl_table *table, int __user *name, int nlen,
1275                                void __user *oldval, size_t __user *oldlenp,
1276                                void __user *newval, size_t newlen)
1277 {
1278         struct ipv4_devconf *cnf;
1279         int *valp = table->data;
1280         int new;
1281         int i;
1282
1283         if (!newval || !newlen)
1284                 return 0;
1285
1286         if (newlen != sizeof(int))
1287                 return -EINVAL;
1288
1289         if (get_user(new, (int __user *)newval))
1290                 return -EFAULT;
1291
1292         if (new == *valp)
1293                 return 0;
1294
1295         if (oldval && oldlenp) {
1296                 size_t len;
1297
1298                 if (get_user(len, oldlenp))
1299                         return -EFAULT;
1300
1301                 if (len) {
1302                         if (len > table->maxlen)
1303                                 len = table->maxlen;
1304                         if (copy_to_user(oldval, valp, len))
1305                                 return -EFAULT;
1306                         if (put_user(len, oldlenp))
1307                                 return -EFAULT;
1308                 }
1309         }
1310
1311         *valp = new;
1312
1313         cnf = table->extra1;
1314         i = (int *)table->data - cnf->data;
1315
1316         set_bit(i, cnf->state);
1317
1318         if (cnf == &ipv4_devconf_dflt)
1319                 devinet_copy_dflt_conf(i);
1320
1321         return 1;
1322 }
1323
1324 void inet_forward_change(void)
1325 {
1326         struct net_device *dev;
1327         int on = IPV4_DEVCONF_ALL(FORWARDING);
1328
1329         IPV4_DEVCONF_ALL(ACCEPT_REDIRECTS) = !on;
1330         IPV4_DEVCONF_DFLT(FORWARDING) = on;
1331
1332         read_lock(&dev_base_lock);
1333         for_each_netdev(&init_net, dev) {
1334                 struct in_device *in_dev;
1335                 rcu_read_lock();
1336                 in_dev = __in_dev_get_rcu(dev);
1337                 if (in_dev)
1338                         IN_DEV_CONF_SET(in_dev, FORWARDING, on);
1339                 rcu_read_unlock();
1340         }
1341         read_unlock(&dev_base_lock);
1342
1343         rt_cache_flush(0);
1344 }
1345
1346 static int devinet_sysctl_forward(ctl_table *ctl, int write,
1347                                   struct file* filp, void __user *buffer,
1348                                   size_t *lenp, loff_t *ppos)
1349 {
1350         int *valp = ctl->data;
1351         int val = *valp;
1352         int ret = proc_dointvec(ctl, write, filp, buffer, lenp, ppos);
1353
1354         if (write && *valp != val) {
1355                 if (valp == &IPV4_DEVCONF_ALL(FORWARDING))
1356                         inet_forward_change();
1357                 else if (valp != &IPV4_DEVCONF_DFLT(FORWARDING))
1358                         rt_cache_flush(0);
1359         }
1360
1361         return ret;
1362 }
1363
1364 int ipv4_doint_and_flush(ctl_table *ctl, int write,
1365                          struct file* filp, void __user *buffer,
1366                          size_t *lenp, loff_t *ppos)
1367 {
1368         int *valp = ctl->data;
1369         int val = *valp;
1370         int ret = proc_dointvec(ctl, write, filp, buffer, lenp, ppos);
1371
1372         if (write && *valp != val)
1373                 rt_cache_flush(0);
1374
1375         return ret;
1376 }
1377
1378 int ipv4_doint_and_flush_strategy(ctl_table *table, int __user *name, int nlen,
1379                                   void __user *oldval, size_t __user *oldlenp,
1380                                   void __user *newval, size_t newlen)
1381 {
1382         int ret = devinet_conf_sysctl(table, name, nlen, oldval, oldlenp,
1383                                       newval, newlen);
1384
1385         if (ret == 1)
1386                 rt_cache_flush(0);
1387
1388         return ret;
1389 }
1390
1391
1392 #define DEVINET_SYSCTL_ENTRY(attr, name, mval, proc, sysctl) \
1393         { \
1394                 .ctl_name       = NET_IPV4_CONF_ ## attr, \
1395                 .procname       = name, \
1396                 .data           = ipv4_devconf.data + \
1397                                   NET_IPV4_CONF_ ## attr - 1, \
1398                 .maxlen         = sizeof(int), \
1399                 .mode           = mval, \
1400                 .proc_handler   = proc, \
1401                 .strategy       = sysctl, \
1402                 .extra1         = &ipv4_devconf, \
1403         }
1404
1405 #define DEVINET_SYSCTL_RW_ENTRY(attr, name) \
1406         DEVINET_SYSCTL_ENTRY(attr, name, 0644, devinet_conf_proc, \
1407                              devinet_conf_sysctl)
1408
1409 #define DEVINET_SYSCTL_RO_ENTRY(attr, name) \
1410         DEVINET_SYSCTL_ENTRY(attr, name, 0444, devinet_conf_proc, \
1411                              devinet_conf_sysctl)
1412
1413 #define DEVINET_SYSCTL_COMPLEX_ENTRY(attr, name, proc, sysctl) \
1414         DEVINET_SYSCTL_ENTRY(attr, name, 0644, proc, sysctl)
1415
1416 #define DEVINET_SYSCTL_FLUSHING_ENTRY(attr, name) \
1417         DEVINET_SYSCTL_COMPLEX_ENTRY(attr, name, ipv4_doint_and_flush, \
1418                                      ipv4_doint_and_flush_strategy)
1419
1420 static struct devinet_sysctl_table {
1421         struct ctl_table_header *sysctl_header;
1422         ctl_table               devinet_vars[__NET_IPV4_CONF_MAX];
1423         ctl_table               devinet_dev[2];
1424         ctl_table               devinet_conf_dir[2];
1425         ctl_table               devinet_proto_dir[2];
1426         ctl_table               devinet_root_dir[2];
1427 } devinet_sysctl = {
1428         .devinet_vars = {
1429                 DEVINET_SYSCTL_COMPLEX_ENTRY(FORWARDING, "forwarding",
1430                                              devinet_sysctl_forward,
1431                                              devinet_conf_sysctl),
1432                 DEVINET_SYSCTL_RO_ENTRY(MC_FORWARDING, "mc_forwarding"),
1433
1434                 DEVINET_SYSCTL_RW_ENTRY(ACCEPT_REDIRECTS, "accept_redirects"),
1435                 DEVINET_SYSCTL_RW_ENTRY(SECURE_REDIRECTS, "secure_redirects"),
1436                 DEVINET_SYSCTL_RW_ENTRY(SHARED_MEDIA, "shared_media"),
1437                 DEVINET_SYSCTL_RW_ENTRY(RP_FILTER, "rp_filter"),
1438                 DEVINET_SYSCTL_RW_ENTRY(SEND_REDIRECTS, "send_redirects"),
1439                 DEVINET_SYSCTL_RW_ENTRY(ACCEPT_SOURCE_ROUTE,
1440                                         "accept_source_route"),
1441                 DEVINET_SYSCTL_RW_ENTRY(PROXY_ARP, "proxy_arp"),
1442                 DEVINET_SYSCTL_RW_ENTRY(MEDIUM_ID, "medium_id"),
1443                 DEVINET_SYSCTL_RW_ENTRY(BOOTP_RELAY, "bootp_relay"),
1444                 DEVINET_SYSCTL_RW_ENTRY(LOG_MARTIANS, "log_martians"),
1445                 DEVINET_SYSCTL_RW_ENTRY(TAG, "tag"),
1446                 DEVINET_SYSCTL_RW_ENTRY(ARPFILTER, "arp_filter"),
1447                 DEVINET_SYSCTL_RW_ENTRY(ARP_ANNOUNCE, "arp_announce"),
1448                 DEVINET_SYSCTL_RW_ENTRY(ARP_IGNORE, "arp_ignore"),
1449                 DEVINET_SYSCTL_RW_ENTRY(ARP_ACCEPT, "arp_accept"),
1450
1451                 DEVINET_SYSCTL_FLUSHING_ENTRY(NOXFRM, "disable_xfrm"),
1452                 DEVINET_SYSCTL_FLUSHING_ENTRY(NOPOLICY, "disable_policy"),
1453                 DEVINET_SYSCTL_FLUSHING_ENTRY(FORCE_IGMP_VERSION,
1454                                               "force_igmp_version"),
1455                 DEVINET_SYSCTL_FLUSHING_ENTRY(PROMOTE_SECONDARIES,
1456                                               "promote_secondaries"),
1457         },
1458         .devinet_dev = {
1459                 {
1460                         .ctl_name       = NET_PROTO_CONF_ALL,
1461                         .procname       = "all",
1462                         .mode           = 0555,
1463                         .child          = devinet_sysctl.devinet_vars,
1464                 },
1465         },
1466         .devinet_conf_dir = {
1467                 {
1468                         .ctl_name       = NET_IPV4_CONF,
1469                         .procname       = "conf",
1470                         .mode           = 0555,
1471                         .child          = devinet_sysctl.devinet_dev,
1472                 },
1473         },
1474         .devinet_proto_dir = {
1475                 {
1476                         .ctl_name       = NET_IPV4,
1477                         .procname       = "ipv4",
1478                         .mode           = 0555,
1479                         .child          = devinet_sysctl.devinet_conf_dir,
1480                 },
1481         },
1482         .devinet_root_dir = {
1483                 {
1484                         .ctl_name       = CTL_NET,
1485                         .procname       = "net",
1486                         .mode           = 0555,
1487                         .child          = devinet_sysctl.devinet_proto_dir,
1488                 },
1489         },
1490 };
1491
1492 static void devinet_sysctl_register(struct in_device *in_dev,
1493                                     struct ipv4_devconf *p)
1494 {
1495         int i;
1496         struct net_device *dev = in_dev ? in_dev->dev : NULL;
1497         struct devinet_sysctl_table *t = kmemdup(&devinet_sysctl, sizeof(*t),
1498                                                  GFP_KERNEL);
1499         char *dev_name = NULL;
1500
1501         if (!t)
1502                 return;
1503         for (i = 0; i < ARRAY_SIZE(t->devinet_vars) - 1; i++) {
1504                 t->devinet_vars[i].data += (char *)p - (char *)&ipv4_devconf;
1505                 t->devinet_vars[i].extra1 = p;
1506         }
1507
1508         if (dev) {
1509                 dev_name = dev->name;
1510                 t->devinet_dev[0].ctl_name = dev->ifindex;
1511         } else {
1512                 dev_name = "default";
1513                 t->devinet_dev[0].ctl_name = NET_PROTO_CONF_DEFAULT;
1514         }
1515
1516         /*
1517          * Make a copy of dev_name, because '.procname' is regarded as const
1518          * by sysctl and we wouldn't want anyone to change it under our feet
1519          * (see SIOCSIFNAME).
1520          */
1521         dev_name = kstrdup(dev_name, GFP_KERNEL);
1522         if (!dev_name)
1523             goto free;
1524
1525         t->devinet_dev[0].procname    = dev_name;
1526         t->devinet_dev[0].child       = t->devinet_vars;
1527         t->devinet_conf_dir[0].child  = t->devinet_dev;
1528         t->devinet_proto_dir[0].child = t->devinet_conf_dir;
1529         t->devinet_root_dir[0].child  = t->devinet_proto_dir;
1530
1531         t->sysctl_header = register_sysctl_table(t->devinet_root_dir);
1532         if (!t->sysctl_header)
1533             goto free_procname;
1534
1535         p->sysctl = t;
1536         return;
1537
1538         /* error path */
1539  free_procname:
1540         kfree(dev_name);
1541  free:
1542         kfree(t);
1543         return;
1544 }
1545
1546 static void devinet_sysctl_unregister(struct ipv4_devconf *p)
1547 {
1548         if (p->sysctl) {
1549                 struct devinet_sysctl_table *t = p->sysctl;
1550                 p->sysctl = NULL;
1551                 unregister_sysctl_table(t->sysctl_header);
1552                 kfree(t->devinet_dev[0].procname);
1553                 kfree(t);
1554         }
1555 }
1556 #endif
1557
1558 void __init devinet_init(void)
1559 {
1560         register_gifconf(PF_INET, inet_gifconf);
1561         register_netdevice_notifier(&ip_netdev_notifier);
1562
1563         rtnl_register(PF_INET, RTM_NEWADDR, inet_rtm_newaddr, NULL);
1564         rtnl_register(PF_INET, RTM_DELADDR, inet_rtm_deladdr, NULL);
1565         rtnl_register(PF_INET, RTM_GETADDR, NULL, inet_dump_ifaddr);
1566 #ifdef CONFIG_SYSCTL
1567         devinet_sysctl.sysctl_header =
1568                 register_sysctl_table(devinet_sysctl.devinet_root_dir);
1569         devinet_sysctl_register(NULL, &ipv4_devconf_dflt);
1570 #endif
1571 }
1572
1573 EXPORT_SYMBOL(in_dev_finish_destroy);
1574 EXPORT_SYMBOL(inet_select_addr);
1575 EXPORT_SYMBOL(inetdev_by_index);
1576 EXPORT_SYMBOL(register_inetaddr_notifier);
1577 EXPORT_SYMBOL(unregister_inetaddr_notifier);