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