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/delay.h>
  13 #include <linux/etherdevice.h>
  14 #include <linux/if_bridge.h>
  15 #include <linux/rculist.h>
  16 #include <linux/netdevice.h>
  17 #include <linux/rtnetlink.h>
  18 #include <net/genetlink.h>
  19
  20 #include "compat.h"
  21 #include "openvswitch/brcompat-netlink.h"
  22 #include "brc_procfs.h"
  23 #include "datapath.h"
  24 #include "dp_dev.h"
  25
  26 static struct genl_family brc_genl_family;
  27 static struct genl_multicast_group brc_mc_group;
  28
  29 /* Time to wait for ovs-vswitchd to respond to a datapath action, in
  30  * jiffies. */
  31 #define BRC_TIMEOUT (HZ * 5)
  32
  33 /* Mutex to serialize ovs-brcompatd callbacks.  (Some callbacks naturally hold
  34  * br_ioctl_mutex, others hold rtnl_lock, but we can't take the former
  35  * ourselves and we don't want to hold the latter over a potentially long
  36  * period of time.) */
  37 static DEFINE_MUTEX(brc_serial);
  38
  39 /* Userspace communication. */
  40 static DEFINE_SPINLOCK(brc_lock);    /* Ensure atomic access to these vars. */
  41 static DECLARE_COMPLETION(brc_done); /* Userspace signaled operation done? */
  42 static struct sk_buff *brc_reply;    /* Reply from userspace. */
  43 static u32 brc_seq;                  /* Sequence number for current op. */
  44
  45 static struct sk_buff *brc_send_command(struct sk_buff *, struct nlattr **attrs);
  46 static int brc_send_simple_command(struct sk_buff *);
  47
  48 static struct sk_buff *
  49 brc_make_request(int op, const char *bridge, const char *port)
  50 {
  51         struct sk_buff *skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
  52         if (!skb)
  53                 goto error;
  54
  55         genlmsg_put(skb, 0, 0, &brc_genl_family, 0, op);
  56         if (bridge)
  57                 NLA_PUT_STRING(skb, BRC_GENL_A_DP_NAME, bridge);
  58         if (port)
  59                 NLA_PUT_STRING(skb, BRC_GENL_A_PORT_NAME, port);
  60         return skb;
  61
  62 nla_put_failure:
  63         kfree_skb(skb);
  64 error:
  65         return NULL;
  66 }
  67
  68 static int brc_send_simple_command(struct sk_buff *request)
  69 {
  70         struct nlattr *attrs[BRC_GENL_A_MAX + 1];
  71         struct sk_buff *reply;
  72         int error;
  73
  74         reply = brc_send_command(request, attrs);
  75         if (IS_ERR(reply))
  76                 return PTR_ERR(reply);
  77
  78         error = nla_get_u32(attrs[BRC_GENL_A_ERR_CODE]);
  79         kfree_skb(reply);
  80         return -error;
  81 }
  82
  83 static int brc_add_del_bridge(char __user *uname, int add)
  84 {
  85         struct sk_buff *request;
  86         char name[IFNAMSIZ];
  87
  88         if (copy_from_user(name, uname, IFNAMSIZ))
  89                 return -EFAULT;
  90
  91         name[IFNAMSIZ - 1] = 0;
  92         request = brc_make_request(add ? BRC_GENL_C_DP_ADD : BRC_GENL_C_DP_DEL,
  93                                    name, NULL);
  94         if (!request)
  95                 return -ENOMEM;
  96
  97         return brc_send_simple_command(request);
  98 }
  99
 100 static int brc_get_indices(int op, const char *br_name,
 101                            int __user *uindices, int n)
 102 {
 103         struct nlattr *attrs[BRC_GENL_A_MAX + 1];
 104         struct sk_buff *request, *reply;
 105         int *indices;
 106         int ret;
 107         int len;
 108
 109         if (n < 0)
 110                 return -EINVAL;
 111         if (n >= 2048)
 112                 return -ENOMEM;
 113
 114         request = brc_make_request(op, br_name, NULL);
 115         if (!request)
 116                 return -ENOMEM;
 117
 118         reply = brc_send_command(request, attrs);
 119         ret = PTR_ERR(reply);
 120         if (IS_ERR(reply))
 121                 goto exit;
 122
 123         ret = -nla_get_u32(attrs[BRC_GENL_A_ERR_CODE]);
 124         if (ret < 0)
 125                 goto exit_free_skb;
 126
 127         ret = -EINVAL;
 128         if (!attrs[BRC_GENL_A_IFINDEXES])
 129                 goto exit_free_skb;
 130
 131         len = nla_len(attrs[BRC_GENL_A_IFINDEXES]);
 132         indices = nla_data(attrs[BRC_GENL_A_IFINDEXES]);
 133         if (len % sizeof(int))
 134                 goto exit_free_skb;
 135
 136         n = min_t(int, n, len / sizeof(int));
 137         ret = copy_to_user(uindices, indices, n * sizeof(int)) ? -EFAULT : n;
 138
 139 exit_free_skb:
 140         kfree_skb(reply);
 141 exit:
 142         return ret;
 143 }
 144
 145 /* Called with br_ioctl_mutex. */
 146 static int brc_get_bridges(int __user *uindices, int n)
 147 {
 148         return brc_get_indices(BRC_GENL_C_GET_BRIDGES, NULL, uindices, n);
 149 }
 150
 151 /* Legacy deviceless bridge ioctl's.  Called with br_ioctl_mutex. */
 152 static int
 153 old_deviceless(void __user *uarg)
 154 {
 155         unsigned long args[3];
 156
 157         if (copy_from_user(args, uarg, sizeof(args)))
 158                 return -EFAULT;
 159
 160         switch (args[0]) {
 161         case BRCTL_GET_BRIDGES:
 162                 return brc_get_bridges((int __user *)args[1], args[2]);
 163
 164         case BRCTL_ADD_BRIDGE:
 165                 return brc_add_del_bridge((void __user *)args[1], 1);
 166         case BRCTL_DEL_BRIDGE:
 167                 return brc_add_del_bridge((void __user *)args[1], 0);
 168         }
 169
 170         return -EOPNOTSUPP;
 171 }
 172
 173 /* Called with the br_ioctl_mutex. */
 174 static int
 175 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
 176 brc_ioctl_deviceless_stub(unsigned int cmd, void __user *uarg)
 177 #else
 178 brc_ioctl_deviceless_stub(struct net *net, unsigned int cmd, void __user *uarg)
 179 #endif
 180 {
 181         switch (cmd) {
 182         case SIOCGIFBR:
 183         case SIOCSIFBR:
 184                 return old_deviceless(uarg);
 185
 186         case SIOCBRADDBR:
 187                 return brc_add_del_bridge(uarg, 1);
 188         case SIOCBRDELBR:
 189                 return brc_add_del_bridge(uarg, 0);
 190         }
 191
 192         return -EOPNOTSUPP;
 193 }
 194
 195 static int
 196 brc_add_del_port(struct net_device *dev, int port_ifindex, int add)
 197 {
 198         struct sk_buff *request;
 199         struct net_device *port;
 200         int err;
 201
 202         port = __dev_get_by_index(&init_net, port_ifindex);
 203         if (!port)
 204                 return -EINVAL;
 205
 206         /* Save name of dev and port because there's a race between the
 207          * rtnl_unlock() and the brc_send_simple_command(). */
 208         request = brc_make_request(add ? BRC_GENL_C_PORT_ADD : BRC_GENL_C_PORT_DEL,
 209                                    dev->name, port->name);
 210         if (!request)
 211                 return -ENOMEM;
 212
 213         rtnl_unlock();
 214         err = brc_send_simple_command(request);
 215         rtnl_lock();
 216
 217         return err;
 218 }
 219
 220 static int
 221 brc_get_bridge_info(struct net_device *dev, struct __bridge_info __user *ub)
 222 {
 223         struct __bridge_info b;
 224         u64 id = 0;
 225         int i;
 226
 227         memset(&b, 0, sizeof(struct __bridge_info));
 228
 229         for (i=0; i<ETH_ALEN; i++)
 230                 id |= (u64)dev->dev_addr[i] << (8*(ETH_ALEN-1 - i));
 231         b.bridge_id = cpu_to_be64(id);
 232         b.stp_enabled = 0;
 233
 234         if (copy_to_user(ub, &b, sizeof(struct __bridge_info)))
 235                 return -EFAULT;
 236
 237         return 0;
 238 }
 239
 240 static int
 241 brc_get_port_list(struct net_device *dev, int __user *uindices, int num)
 242 {
 243         int retval;
 244
 245         rtnl_unlock();
 246         retval = brc_get_indices(BRC_GENL_C_GET_PORTS, dev->name,
 247                                  uindices, num);
 248         rtnl_lock();
 249
 250         return retval;
 251 }
 252
 253 /*
 254  * Format up to a page worth of forwarding table entries
 255  * userbuf -- where to copy result
 256  * maxnum  -- maximum number of entries desired
 257  *            (limited to a page for sanity)
 258  * offset  -- number of records to skip
 259  */
 260 static int brc_get_fdb_entries(struct net_device *dev, void __user *userbuf,
 261                                unsigned long maxnum, unsigned long offset)
 262 {
 263         struct nlattr *attrs[BRC_GENL_A_MAX + 1];
 264         struct sk_buff *request, *reply;
 265         int retval;
 266         int len;
 267
 268         /* Clamp size to PAGE_SIZE, test maxnum to avoid overflow */
 269         if (maxnum > PAGE_SIZE/sizeof(struct __fdb_entry))
 270                 maxnum = PAGE_SIZE/sizeof(struct __fdb_entry);
 271
 272         request = brc_make_request(BRC_GENL_C_FDB_QUERY, dev->name, NULL);
 273         if (!request)
 274                 return -ENOMEM;
 275         NLA_PUT_U64(request, BRC_GENL_A_FDB_COUNT, maxnum);
 276         NLA_PUT_U64(request, BRC_GENL_A_FDB_SKIP, offset);
 277
 278         rtnl_unlock();
 279         reply = brc_send_command(request, attrs);
 280         retval = PTR_ERR(reply);
 281         if (IS_ERR(reply))
 282                 goto exit;
 283
 284         retval = -nla_get_u32(attrs[BRC_GENL_A_ERR_CODE]);
 285         if (retval < 0)
 286                 goto exit_free_skb;
 287
 288         retval = -EINVAL;
 289         if (!attrs[BRC_GENL_A_FDB_DATA])
 290                 goto exit_free_skb;
 291         len = nla_len(attrs[BRC_GENL_A_FDB_DATA]);
 292         if (len % sizeof(struct __fdb_entry) ||
 293             len / sizeof(struct __fdb_entry) > maxnum)
 294                 goto exit_free_skb;
 295
 296         retval = len / sizeof(struct __fdb_entry);
 297         if (copy_to_user(userbuf, nla_data(attrs[BRC_GENL_A_FDB_DATA]), len))
 298                 retval = -EFAULT;
 299
 300 exit_free_skb:
 301         kfree_skb(reply);
 302 exit:
 303         rtnl_lock();
 304         return retval;
 305
 306 nla_put_failure:
 307         kfree_skb(request);
 308         return -ENOMEM;
 309 }
 310
 311 /* Legacy ioctl's through SIOCDEVPRIVATE.  Called with rtnl_lock. */
 312 static int
 313 old_dev_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
 314 {
 315         unsigned long args[4];
 316
 317         if (copy_from_user(args, rq->ifr_data, sizeof(args)))
 318                 return -EFAULT;
 319
 320         switch (args[0]) {
 321         case BRCTL_ADD_IF:
 322                 return brc_add_del_port(dev, args[1], 1);
 323         case BRCTL_DEL_IF:
 324                 return brc_add_del_port(dev, args[1], 0);
 325
 326         case BRCTL_GET_BRIDGE_INFO:
 327                 return brc_get_bridge_info(dev, (struct __bridge_info __user *)args[1]);
 328
 329         case BRCTL_GET_PORT_LIST:
 330                 return brc_get_port_list(dev, (int __user *)args[1], args[2]);
 331
 332         case BRCTL_GET_FDB_ENTRIES:
 333                 return brc_get_fdb_entries(dev, (void __user *)args[1],
 334                                            args[2], args[3]);
 335         }
 336
 337         return -EOPNOTSUPP;
 338 }
 339
 340 /* Called with the rtnl_lock. */
 341 static int
 342 brc_dev_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
 343 {
 344         int err;
 345
 346         switch (cmd) {
 347                 case SIOCDEVPRIVATE:
 348                         err = old_dev_ioctl(dev, rq, cmd);
 349                         break;
 350
 351                 case SIOCBRADDIF:
 352                         return brc_add_del_port(dev, rq->ifr_ifindex, 1);
 353                 case SIOCBRDELIF:
 354                         return brc_add_del_port(dev, rq->ifr_ifindex, 0);
 355
 356                 default:
 357                         err = -EOPNOTSUPP;
 358                         break;
 359         }
 360
 361         return err;
 362 }
 363
 364
 365 static struct genl_family brc_genl_family = {
 366         .id = GENL_ID_GENERATE,
 367         .hdrsize = 0,
 368         .name = BRC_GENL_FAMILY_NAME,
 369         .version = 1,
 370         .maxattr = BRC_GENL_A_MAX,
 371 };
 372
 373 static int brc_genl_query(struct sk_buff *skb, struct genl_info *info)
 374 {
 375         int err = -EINVAL;
 376         struct sk_buff *ans_skb;
 377         void *data;
 378
 379         ans_skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
 380         if (!ans_skb)
 381                 return -ENOMEM;
 382
 383         data = genlmsg_put_reply(ans_skb, info, &brc_genl_family,
 384                                  0, BRC_GENL_C_QUERY_MC);
 385         if (data == NULL) {
 386                 err = -ENOMEM;
 387                 goto err;
 388         }
 389         NLA_PUT_U32(ans_skb, BRC_GENL_A_MC_GROUP, brc_mc_group.id);
 390
 391         genlmsg_end(ans_skb, data);
 392         return genlmsg_reply(ans_skb, info);
 393
 394 err:
 395 nla_put_failure:
 396         kfree_skb(ans_skb);
 397         return err;
 398 }
 399
 400 static struct genl_ops brc_genl_ops_query_dp = {
 401         .cmd = BRC_GENL_C_QUERY_MC,
 402         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privelege. */
 403         .policy = NULL,
 404         .doit = brc_genl_query,
 405         .dumpit = NULL
 406 };
 407
 408 /* Attribute policy: what each attribute may contain.  */
 409 static struct nla_policy brc_genl_policy[BRC_GENL_A_MAX + 1] = {
 410         [BRC_GENL_A_ERR_CODE] = { .type = NLA_U32 },
 411
 412         [BRC_GENL_A_PROC_DIR] = { .type = NLA_NUL_STRING },
 413         [BRC_GENL_A_PROC_NAME] = { .type = NLA_NUL_STRING },
 414         [BRC_GENL_A_PROC_DATA] = { .type = NLA_NUL_STRING },
 415
 416         [BRC_GENL_A_FDB_DATA] = { .type = NLA_UNSPEC },
 417 };
 418
 419 static int
 420 brc_genl_dp_result(struct sk_buff *skb, struct genl_info *info)
 421 {
 422         unsigned long int flags;
 423         int err;
 424
 425         if (!info->attrs[BRC_GENL_A_ERR_CODE])
 426                 return -EINVAL;
 427
 428         skb = skb_clone(skb, GFP_KERNEL);
 429         if (!skb)
 430                 return -ENOMEM;
 431
 432         spin_lock_irqsave(&brc_lock, flags);
 433         if (brc_seq == info->snd_seq) {
 434                 brc_seq++;
 435
 436                 if (brc_reply)
 437                         kfree_skb(brc_reply);
 438                 brc_reply = skb;
 439
 440                 complete(&brc_done);
 441                 err = 0;
 442         } else {
 443                 kfree_skb(skb);
 444                 err = -ESTALE;
 445         }
 446         spin_unlock_irqrestore(&brc_lock, flags);
 447
 448         return err;
 449 }
 450
 451 static struct genl_ops brc_genl_ops_dp_result = {
 452         .cmd = BRC_GENL_C_DP_RESULT,
 453         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privelege. */
 454         .policy = brc_genl_policy,
 455         .doit = brc_genl_dp_result,
 456         .dumpit = NULL
 457 };
 458
 459 static struct genl_ops brc_genl_ops_set_proc = {
 460         .cmd = BRC_GENL_C_SET_PROC,
 461         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privelege. */
 462         .policy = brc_genl_policy,
 463         .doit = brc_genl_set_proc,
 464         .dumpit = NULL
 465 };
 466
 467 static struct sk_buff *brc_send_command(struct sk_buff *request, struct nlattr **attrs)
 468 {
 469         unsigned long int flags;
 470         struct sk_buff *reply;
 471         int error;
 472
 473         mutex_lock(&brc_serial);
 474
 475         /* Increment sequence number first, so that we ignore any replies
 476          * to stale requests. */
 477         spin_lock_irqsave(&brc_lock, flags);
 478         nlmsg_hdr(request)->nlmsg_seq = ++brc_seq;
 479         INIT_COMPLETION(brc_done);
 480         spin_unlock_irqrestore(&brc_lock, flags);
 481
 482         nlmsg_end(request, nlmsg_hdr(request));
 483
 484         /* Send message. */
 485         error = genlmsg_multicast(request, 0, brc_mc_group.id, GFP_KERNEL);
 486         if (error < 0)
 487                 goto error;
 488
 489         /* Wait for reply. */
 490         error = -ETIMEDOUT;
 491         if (!wait_for_completion_timeout(&brc_done, BRC_TIMEOUT))
 492                 goto error;
 493
 494         /* Grab reply. */
 495         spin_lock_irqsave(&brc_lock, flags);
 496         reply = brc_reply;
 497         brc_reply = NULL;
 498         spin_unlock_irqrestore(&brc_lock, flags);
 499
 500         mutex_unlock(&brc_serial);
 501
 502         /* Re-parse message.  Can't fail, since it parsed correctly once
 503          * already. */
 504         error = nlmsg_parse(nlmsg_hdr(reply), GENL_HDRLEN,
 505                             attrs, BRC_GENL_A_MAX, brc_genl_policy);
 506         WARN_ON(error);
 507
 508         return reply;
 509
 510 error:
 511         mutex_unlock(&brc_serial);
 512         return ERR_PTR(error);
 513 }
 514
 515 static int
 516 __init brc_init(void)
 517 {
 518         int err;
 519
 520         printk("Open vSwitch Bridge Compatibility, built "__DATE__" "__TIME__"\n");
 521
 522         /* Set the bridge ioctl handler */
 523         brioctl_set(brc_ioctl_deviceless_stub);
 524
 525         /* Set the openvswitch_mod device ioctl handler */
 526         dp_ioctl_hook = brc_dev_ioctl;
 527
 528         /* Randomize the initial sequence number.  This is not a security
 529          * feature; it only helps avoid crossed wires between userspace and
 530          * the kernel when the module is unloaded and reloaded. */
 531         brc_seq = net_random();
 532
 533         /* Register generic netlink family to communicate changes to
 534          * userspace. */
 535         err = genl_register_family(&brc_genl_family);
 536         if (err)
 537                 goto error;
 538
 539         err = genl_register_ops(&brc_genl_family, &brc_genl_ops_query_dp);
 540         if (err != 0)
 541                 goto err_unregister;
 542
 543         err = genl_register_ops(&brc_genl_family, &brc_genl_ops_dp_result);
 544         if (err != 0)
 545                 goto err_unregister;
 546
 547         err = genl_register_ops(&brc_genl_family, &brc_genl_ops_set_proc);
 548         if (err != 0)
 549                 goto err_unregister;
 550
 551         strcpy(brc_mc_group.name, "brcompat");
 552         err = genl_register_mc_group(&brc_genl_family, &brc_mc_group);
 553         if (err < 0)
 554                 goto err_unregister;
 555
 556         return 0;
 557
 558 err_unregister:
 559         genl_unregister_family(&brc_genl_family);
 560 error:
 561         printk(KERN_EMERG "brcompat: failed to install!");
 562         return err;
 563 }
 564
 565 static void
 566 brc_cleanup(void)
 567 {
 568         /* Unregister ioctl hooks */
 569         dp_ioctl_hook = NULL;
 570         brioctl_set(NULL);
 571
 572         genl_unregister_family(&brc_genl_family);
 573         brc_procfs_exit();
 574 }
 575
 576 module_init(brc_init);
 577 module_exit(brc_cleanup);
 578
 579 MODULE_DESCRIPTION("Open vSwitch bridge compatibility");
 580 MODULE_AUTHOR("Nicira Networks");
 581 MODULE_LICENSE("GPL");