datapath/brcompat.c

   1 /*
   2  * Copyright (c) 2009 Nicira Networks.
   3  * Distributed under the terms of the GNU GPL version 2.
   4  *
   5  * Significant portions of this file may be copied from parts of the Linux
   6  * kernel, by Linus Torvalds and others.
   7  */
   8
   9 #include <linux/kernel.h>
  10 #include <asm/uaccess.h>
  11 #include <linux/completion.h>
  12 #include <linux/etherdevice.h>
  13 #include <linux/if_bridge.h>
  14 #include <linux/netdevice.h>
  15 #include <linux/rtnetlink.h>
  16 #include <net/genetlink.h>
  17
  18 #include "compat.h"
  19 #include "openvswitch/brcompat-netlink.h"
  20 #include "brc_procfs.h"
  21 #include "datapath.h"
  22
  23 static struct genl_family brc_genl_family;
  24 static struct genl_multicast_group brc_mc_group;
  25
  26 /* Time to wait for ovs-vswitchd to respond to a datapath action, in
  27  * jiffies. */
  28 #define BRC_TIMEOUT (HZ * 5)
  29
  30 /* Mutex to serialize ovs-brcompatd callbacks.  (Some callbacks naturally hold
  31  * br_ioctl_mutex, others hold rtnl_lock, but we can't take the former
  32  * ourselves and we don't want to hold the latter over a potentially long
  33  * period of time.) */
  34 static DEFINE_MUTEX(brc_serial);
  35
  36 /* Userspace communication. */
  37 static DEFINE_SPINLOCK(brc_lock);    /* Ensure atomic access to these vars. */
  38 static DECLARE_COMPLETION(brc_done); /* Userspace signaled operation done? */
  39 static struct sk_buff *brc_reply;    /* Reply from userspace. */
  40 static u32 brc_seq;                  /* Sequence number for current op. */
  41
  42 static struct sk_buff *brc_send_command(struct sk_buff *, struct nlattr **attrs);
  43 static int brc_send_simple_command(struct sk_buff *);
  44
  45 static struct sk_buff *brc_make_request(int op, const char *bridge,
  46                                         const char *port)
  47 {
  48         struct sk_buff *skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
  49         if (!skb)
  50                 goto error;
  51
  52         genlmsg_put(skb, 0, 0, &brc_genl_family, 0, op);
  53         if (bridge)
  54                 NLA_PUT_STRING(skb, BRC_GENL_A_DP_NAME, bridge);
  55         if (port)
  56                 NLA_PUT_STRING(skb, BRC_GENL_A_PORT_NAME, port);
  57         return skb;
  58
  59 nla_put_failure:
  60         kfree_skb(skb);
  61 error:
  62         return NULL;
  63 }
  64
  65 static int brc_send_simple_command(struct sk_buff *request)
  66 {
  67         struct nlattr *attrs[BRC_GENL_A_MAX + 1];
  68         struct sk_buff *reply;
  69         int error;
  70
  71         reply = brc_send_command(request, attrs);
  72         if (IS_ERR(reply))
  73                 return PTR_ERR(reply);
  74
  75         error = nla_get_u32(attrs[BRC_GENL_A_ERR_CODE]);
  76         kfree_skb(reply);
  77         return -error;
  78 }
  79
  80 static int brc_add_del_bridge(char __user *uname, int add)
  81 {
  82         struct sk_buff *request;
  83         char name[IFNAMSIZ];
  84
  85         if (copy_from_user(name, uname, IFNAMSIZ))
  86                 return -EFAULT;
  87
  88         name[IFNAMSIZ - 1] = 0;
  89         request = brc_make_request(add ? BRC_GENL_C_DP_ADD : BRC_GENL_C_DP_DEL,
  90                                    name, NULL);
  91         if (!request)
  92                 return -ENOMEM;
  93
  94         return brc_send_simple_command(request);
  95 }
  96
  97 static int brc_get_indices(int op, const char *br_name,
  98                            int __user *uindices, int n)
  99 {
 100         struct nlattr *attrs[BRC_GENL_A_MAX + 1];
 101         struct sk_buff *request, *reply;
 102         int *indices;
 103         int ret;
 104         int len;
 105
 106         if (n < 0)
 107                 return -EINVAL;
 108         if (n >= 2048)
 109                 return -ENOMEM;
 110
 111         request = brc_make_request(op, br_name, NULL);
 112         if (!request)
 113                 return -ENOMEM;
 114
 115         reply = brc_send_command(request, attrs);
 116         ret = PTR_ERR(reply);
 117         if (IS_ERR(reply))
 118                 goto exit;
 119
 120         ret = -nla_get_u32(attrs[BRC_GENL_A_ERR_CODE]);
 121         if (ret < 0)
 122                 goto exit_free_skb;
 123
 124         ret = -EINVAL;
 125         if (!attrs[BRC_GENL_A_IFINDEXES])
 126                 goto exit_free_skb;
 127
 128         len = nla_len(attrs[BRC_GENL_A_IFINDEXES]);
 129         indices = nla_data(attrs[BRC_GENL_A_IFINDEXES]);
 130         if (len % sizeof(int))
 131                 goto exit_free_skb;
 132
 133         n = min_t(int, n, len / sizeof(int));
 134         ret = copy_to_user(uindices, indices, n * sizeof(int)) ? -EFAULT : n;
 135
 136 exit_free_skb:
 137         kfree_skb(reply);
 138 exit:
 139         return ret;
 140 }
 141
 142 /* Called with br_ioctl_mutex. */
 143 static int brc_get_bridges(int __user *uindices, int n)
 144 {
 145         return brc_get_indices(BRC_GENL_C_GET_BRIDGES, NULL, uindices, n);
 146 }
 147
 148 /* Legacy deviceless bridge ioctl's.  Called with br_ioctl_mutex. */
 149 static int old_deviceless(void __user *uarg)
 150 {
 151         unsigned long args[3];
 152
 153         if (copy_from_user(args, uarg, sizeof(args)))
 154                 return -EFAULT;
 155
 156         switch (args[0]) {
 157         case BRCTL_GET_BRIDGES:
 158                 return brc_get_bridges((int __user *)args[1], args[2]);
 159
 160         case BRCTL_ADD_BRIDGE:
 161                 return brc_add_del_bridge((void __user *)args[1], 1);
 162         case BRCTL_DEL_BRIDGE:
 163                 return brc_add_del_bridge((void __user *)args[1], 0);
 164         }
 165
 166         return -EOPNOTSUPP;
 167 }
 168
 169 /* Called with the br_ioctl_mutex. */
 170 static int
 171 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
 172 brc_ioctl_deviceless_stub(unsigned int cmd, void __user *uarg)
 173 #else
 174 brc_ioctl_deviceless_stub(struct net *net, unsigned int cmd, void __user *uarg)
 175 #endif
 176 {
 177         switch (cmd) {
 178         case SIOCGIFBR:
 179         case SIOCSIFBR:
 180                 return old_deviceless(uarg);
 181
 182         case SIOCBRADDBR:
 183                 return brc_add_del_bridge(uarg, 1);
 184         case SIOCBRDELBR:
 185                 return brc_add_del_bridge(uarg, 0);
 186         }
 187
 188         return -EOPNOTSUPP;
 189 }
 190
 191 static int brc_add_del_port(struct net_device *dev, int port_ifindex, int add)
 192 {
 193         struct sk_buff *request;
 194         struct net_device *port;
 195         int err;
 196
 197         port = __dev_get_by_index(&init_net, port_ifindex);
 198         if (!port)
 199                 return -EINVAL;
 200
 201         /* Save name of dev and port because there's a race between the
 202          * rtnl_unlock() and the brc_send_simple_command(). */
 203         request = brc_make_request(add ? BRC_GENL_C_PORT_ADD : BRC_GENL_C_PORT_DEL,
 204                                    dev->name, port->name);
 205         if (!request)
 206                 return -ENOMEM;
 207
 208         rtnl_unlock();
 209         err = brc_send_simple_command(request);
 210         rtnl_lock();
 211
 212         return err;
 213 }
 214
 215 static int brc_get_bridge_info(struct net_device *dev,
 216                                struct __bridge_info __user *ub)
 217 {
 218         struct __bridge_info b;
 219         u64 id = 0;
 220         int i;
 221
 222         memset(&b, 0, sizeof(struct __bridge_info));
 223
 224         for (i=0; i<ETH_ALEN; i++)
 225                 id |= (u64)dev->dev_addr[i] << (8*(ETH_ALEN-1 - i));
 226         b.bridge_id = cpu_to_be64(id);
 227         b.stp_enabled = 0;
 228
 229         if (copy_to_user(ub, &b, sizeof(struct __bridge_info)))
 230                 return -EFAULT;
 231
 232         return 0;
 233 }
 234
 235 static int brc_get_port_list(struct net_device *dev, int __user *uindices,
 236                              int num)
 237 {
 238         int retval;
 239
 240         rtnl_unlock();
 241         retval = brc_get_indices(BRC_GENL_C_GET_PORTS, dev->name,
 242                                  uindices, num);
 243         rtnl_lock();
 244
 245         return retval;
 246 }
 247
 248 /*
 249  * Format up to a page worth of forwarding table entries
 250  * userbuf -- where to copy result
 251  * maxnum  -- maximum number of entries desired
 252  *            (limited to a page for sanity)
 253  * offset  -- number of records to skip
 254  */
 255 static int brc_get_fdb_entries(struct net_device *dev, void __user *userbuf,
 256                                unsigned long maxnum, unsigned long offset)
 257 {
 258         struct nlattr *attrs[BRC_GENL_A_MAX + 1];
 259         struct sk_buff *request, *reply;
 260         int retval;
 261         int len;
 262
 263         /* Clamp size to PAGE_SIZE, test maxnum to avoid overflow */
 264         if (maxnum > PAGE_SIZE/sizeof(struct __fdb_entry))
 265                 maxnum = PAGE_SIZE/sizeof(struct __fdb_entry);
 266
 267         request = brc_make_request(BRC_GENL_C_FDB_QUERY, dev->name, NULL);
 268         if (!request)
 269                 return -ENOMEM;
 270         NLA_PUT_U64(request, BRC_GENL_A_FDB_COUNT, maxnum);
 271         NLA_PUT_U64(request, BRC_GENL_A_FDB_SKIP, offset);
 272
 273         rtnl_unlock();
 274         reply = brc_send_command(request, attrs);
 275         retval = PTR_ERR(reply);
 276         if (IS_ERR(reply))
 277                 goto exit;
 278
 279         retval = -nla_get_u32(attrs[BRC_GENL_A_ERR_CODE]);
 280         if (retval < 0)
 281                 goto exit_free_skb;
 282
 283         retval = -EINVAL;
 284         if (!attrs[BRC_GENL_A_FDB_DATA])
 285                 goto exit_free_skb;
 286         len = nla_len(attrs[BRC_GENL_A_FDB_DATA]);
 287         if (len % sizeof(struct __fdb_entry) ||
 288             len / sizeof(struct __fdb_entry) > maxnum)
 289                 goto exit_free_skb;
 290
 291         retval = len / sizeof(struct __fdb_entry);
 292         if (copy_to_user(userbuf, nla_data(attrs[BRC_GENL_A_FDB_DATA]), len))
 293                 retval = -EFAULT;
 294
 295 exit_free_skb:
 296         kfree_skb(reply);
 297 exit:
 298         rtnl_lock();
 299         return retval;
 300
 301 nla_put_failure:
 302         kfree_skb(request);
 303         return -ENOMEM;
 304 }
 305
 306 /* Legacy ioctl's through SIOCDEVPRIVATE.  Called with rtnl_lock. */
 307 static int old_dev_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
 308 {
 309         unsigned long args[4];
 310
 311         if (copy_from_user(args, rq->ifr_data, sizeof(args)))
 312                 return -EFAULT;
 313
 314         switch (args[0]) {
 315         case BRCTL_ADD_IF:
 316                 return brc_add_del_port(dev, args[1], 1);
 317         case BRCTL_DEL_IF:
 318                 return brc_add_del_port(dev, args[1], 0);
 319
 320         case BRCTL_GET_BRIDGE_INFO:
 321                 return brc_get_bridge_info(dev, (struct __bridge_info __user *)args[1]);
 322
 323         case BRCTL_GET_PORT_LIST:
 324                 return brc_get_port_list(dev, (int __user *)args[1], args[2]);
 325
 326         case BRCTL_GET_FDB_ENTRIES:
 327                 return brc_get_fdb_entries(dev, (void __user *)args[1],
 328                                            args[2], args[3]);
 329         }
 330
 331         return -EOPNOTSUPP;
 332 }
 333
 334 /* Called with the rtnl_lock. */
 335 static int brc_dev_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
 336 {
 337         int err;
 338
 339         switch (cmd) {
 340                 case SIOCDEVPRIVATE:
 341                         err = old_dev_ioctl(dev, rq, cmd);
 342                         break;
 343
 344                 case SIOCBRADDIF:
 345                         return brc_add_del_port(dev, rq->ifr_ifindex, 1);
 346                 case SIOCBRDELIF:
 347                         return brc_add_del_port(dev, rq->ifr_ifindex, 0);
 348
 349                 default:
 350                         err = -EOPNOTSUPP;
 351                         break;
 352         }
 353
 354         return err;
 355 }
 356
 357
 358 static struct genl_family brc_genl_family = {
 359         .id = GENL_ID_GENERATE,
 360         .hdrsize = 0,
 361         .name = BRC_GENL_FAMILY_NAME,
 362         .version = 1,
 363         .maxattr = BRC_GENL_A_MAX,
 364 };
 365
 366 static int brc_genl_query(struct sk_buff *skb, struct genl_info *info)
 367 {
 368         int err = -EINVAL;
 369         struct sk_buff *ans_skb;
 370         void *data;
 371
 372         ans_skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
 373         if (!ans_skb)
 374                 return -ENOMEM;
 375
 376         data = genlmsg_put_reply(ans_skb, info, &brc_genl_family,
 377                                  0, BRC_GENL_C_QUERY_MC);
 378         if (data == NULL) {
 379                 err = -ENOMEM;
 380                 goto err;
 381         }
 382         NLA_PUT_U32(ans_skb, BRC_GENL_A_MC_GROUP, brc_mc_group.id);
 383
 384         genlmsg_end(ans_skb, data);
 385         return genlmsg_reply(ans_skb, info);
 386
 387 err:
 388 nla_put_failure:
 389         kfree_skb(ans_skb);
 390         return err;
 391 }
 392
 393 static struct genl_ops brc_genl_ops_query_dp = {
 394         .cmd = BRC_GENL_C_QUERY_MC,
 395         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privelege. */
 396         .policy = NULL,
 397         .doit = brc_genl_query,
 398         .dumpit = NULL
 399 };
 400
 401 /* Attribute policy: what each attribute may contain.  */
 402 static struct nla_policy brc_genl_policy[BRC_GENL_A_MAX + 1] = {
 403         [BRC_GENL_A_ERR_CODE] = { .type = NLA_U32 },
 404
 405         [BRC_GENL_A_PROC_DIR] = { .type = NLA_NUL_STRING },
 406         [BRC_GENL_A_PROC_NAME] = { .type = NLA_NUL_STRING },
 407         [BRC_GENL_A_PROC_DATA] = { .type = NLA_NUL_STRING },
 408
 409         [BRC_GENL_A_FDB_DATA] = { .type = NLA_UNSPEC },
 410 };
 411
 412 static int brc_genl_dp_result(struct sk_buff *skb, struct genl_info *info)
 413 {
 414         unsigned long int flags;
 415         int err;
 416
 417         if (!info->attrs[BRC_GENL_A_ERR_CODE])
 418                 return -EINVAL;
 419
 420         skb = skb_clone(skb, GFP_KERNEL);
 421         if (!skb)
 422                 return -ENOMEM;
 423
 424         spin_lock_irqsave(&brc_lock, flags);
 425         if (brc_seq == info->snd_seq) {
 426                 brc_seq++;
 427
 428                 kfree_skb(brc_reply);
 429                 brc_reply = skb;
 430
 431                 complete(&brc_done);
 432                 err = 0;
 433         } else {
 434                 kfree_skb(skb);
 435                 err = -ESTALE;
 436         }
 437         spin_unlock_irqrestore(&brc_lock, flags);
 438
 439         return err;
 440 }
 441
 442 static struct genl_ops brc_genl_ops_dp_result = {
 443         .cmd = BRC_GENL_C_DP_RESULT,
 444         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privelege. */
 445         .policy = brc_genl_policy,
 446         .doit = brc_genl_dp_result,
 447         .dumpit = NULL
 448 };
 449
 450 static struct genl_ops brc_genl_ops_set_proc = {
 451         .cmd = BRC_GENL_C_SET_PROC,
 452         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privelege. */
 453         .policy = brc_genl_policy,
 454         .doit = brc_genl_set_proc,
 455         .dumpit = NULL
 456 };
 457
 458 static struct sk_buff *brc_send_command(struct sk_buff *request,
 459                                         struct nlattr **attrs)
 460 {
 461         unsigned long int flags;
 462         struct sk_buff *reply;
 463         int error;
 464
 465         mutex_lock(&brc_serial);
 466
 467         /* Increment sequence number first, so that we ignore any replies
 468          * to stale requests. */
 469         spin_lock_irqsave(&brc_lock, flags);
 470         nlmsg_hdr(request)->nlmsg_seq = ++brc_seq;
 471         INIT_COMPLETION(brc_done);
 472         spin_unlock_irqrestore(&brc_lock, flags);
 473
 474         nlmsg_end(request, nlmsg_hdr(request));
 475
 476         /* Send message. */
 477         error = genlmsg_multicast(request, 0, brc_mc_group.id, GFP_KERNEL);
 478         if (error < 0)
 479                 goto error;
 480
 481         /* Wait for reply. */
 482         error = -ETIMEDOUT;
 483         if (!wait_for_completion_timeout(&brc_done, BRC_TIMEOUT)) {
 484                 printk(KERN_WARNING "brcompat: timed out waiting for userspace\n");
 485                 goto error;
 486     }
 487
 488         /* Grab reply. */
 489         spin_lock_irqsave(&brc_lock, flags);
 490         reply = brc_reply;
 491         brc_reply = NULL;
 492         spin_unlock_irqrestore(&brc_lock, flags);
 493
 494         mutex_unlock(&brc_serial);
 495
 496         /* Re-parse message.  Can't fail, since it parsed correctly once
 497          * already. */
 498         error = nlmsg_parse(nlmsg_hdr(reply), GENL_HDRLEN,
 499                             attrs, BRC_GENL_A_MAX, brc_genl_policy);
 500         WARN_ON(error);
 501
 502         return reply;
 503
 504 error:
 505         mutex_unlock(&brc_serial);
 506         return ERR_PTR(error);
 507 }
 508
 509 static int __init brc_init(void)
 510 {
 511         int err;
 512
 513         printk("Open vSwitch Bridge Compatibility, built "__DATE__" "__TIME__"\n");
 514
 515         /* Set the bridge ioctl handler */
 516         brioctl_set(brc_ioctl_deviceless_stub);
 517
 518         /* Set the openvswitch_mod device ioctl handler */
 519         dp_ioctl_hook = brc_dev_ioctl;
 520
 521         /* Randomize the initial sequence number.  This is not a security
 522          * feature; it only helps avoid crossed wires between userspace and
 523          * the kernel when the module is unloaded and reloaded. */
 524         brc_seq = net_random();
 525
 526         /* Register generic netlink family to communicate changes to
 527          * userspace. */
 528         err = genl_register_family(&brc_genl_family);
 529         if (err)
 530                 goto error;
 531
 532         err = genl_register_ops(&brc_genl_family, &brc_genl_ops_query_dp);
 533         if (err != 0)
 534                 goto err_unregister;
 535
 536         err = genl_register_ops(&brc_genl_family, &brc_genl_ops_dp_result);
 537         if (err != 0)
 538                 goto err_unregister;
 539
 540         err = genl_register_ops(&brc_genl_family, &brc_genl_ops_set_proc);
 541         if (err != 0)
 542                 goto err_unregister;
 543
 544         strcpy(brc_mc_group.name, "brcompat");
 545         err = genl_register_mc_group(&brc_genl_family, &brc_mc_group);
 546         if (err < 0)
 547                 goto err_unregister;
 548
 549         return 0;
 550
 551 err_unregister:
 552         genl_unregister_family(&brc_genl_family);
 553 error:
 554         printk(KERN_EMERG "brcompat: failed to install!");
 555         return err;
 556 }
 557
 558 static void brc_cleanup(void)
 559 {
 560         /* Unregister ioctl hooks */
 561         dp_ioctl_hook = NULL;
 562         brioctl_set(NULL);
 563
 564         genl_unregister_family(&brc_genl_family);
 565         brc_procfs_exit();
 566 }
 567
 568 module_init(brc_init);
 569 module_exit(brc_cleanup);
 570
 571 MODULE_DESCRIPTION("Open vSwitch bridge compatibility");
 572 MODULE_AUTHOR("Nicira Networks");
 573 MODULE_LICENSE("GPL");