net/netfilter/nfnetlink_acct.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * (C) 2011 Pablo Neira Ayuso <pablo@netfilter.org>
   4  * (C) 2011 Intra2net AG <https://www.intra2net.com>
   5  */
   6 #include <linux/init.h>
   7 #include <linux/module.h>
   8 #include <linux/kernel.h>
   9 #include <linux/skbuff.h>
  10 #include <linux/atomic.h>
  11 #include <linux/refcount.h>
  12 #include <linux/netlink.h>
  13 #include <linux/rculist.h>
  14 #include <linux/slab.h>
  15 #include <linux/types.h>
  16 #include <linux/errno.h>
  17 #include <net/netlink.h>
  18 #include <net/sock.h>
  19 #include <net/netns/generic.h>
  20
  21 #include <linux/netfilter.h>
  22 #include <linux/netfilter/nfnetlink.h>
  23 #include <linux/netfilter/nfnetlink_acct.h>
  24
  25 MODULE_LICENSE("GPL");
  26 MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");
  27 MODULE_DESCRIPTION("nfacct: Extended Netfilter accounting infrastructure");
  28
  29 struct nf_acct {
  30         atomic64_t              pkts;
  31         atomic64_t              bytes;
  32         unsigned long           flags;
  33         struct list_head        head;
  34         refcount_t              refcnt;
  35         char                    name[NFACCT_NAME_MAX];
  36         struct rcu_head         rcu_head;
  37         char                    data[];
  38 };
  39
  40 struct nfacct_filter {
  41         u32 value;
  42         u32 mask;
  43 };
  44
  45 struct nfnl_acct_net {
  46         struct list_head        nfnl_acct_list;
  47 };
  48
  49 static unsigned int nfnl_acct_net_id __read_mostly;
  50
  51 static inline struct nfnl_acct_net *nfnl_acct_pernet(struct net *net)
  52 {
  53         return net_generic(net, nfnl_acct_net_id);
  54 }
  55
  56 #define NFACCT_F_QUOTA (NFACCT_F_QUOTA_PKTS | NFACCT_F_QUOTA_BYTES)
  57 #define NFACCT_OVERQUOTA_BIT    2       /* NFACCT_F_OVERQUOTA */
  58
  59 static int nfnl_acct_new(struct net *net, struct sock *nfnl,
  60                          struct sk_buff *skb, const struct nlmsghdr *nlh,
  61                          const struct nlattr * const tb[],
  62                          struct netlink_ext_ack *extack)
  63 {
  64         struct nfnl_acct_net *nfnl_acct_net = nfnl_acct_pernet(net);
  65         struct nf_acct *nfacct, *matching = NULL;
  66         char *acct_name;
  67         unsigned int size = 0;
  68         u32 flags = 0;
  69
  70         if (!tb[NFACCT_NAME])
  71                 return -EINVAL;
  72
  73         acct_name = nla_data(tb[NFACCT_NAME]);
  74         if (strlen(acct_name) == 0)
  75                 return -EINVAL;
  76
  77         list_for_each_entry(nfacct, &nfnl_acct_net->nfnl_acct_list, head) {
  78                 if (strncmp(nfacct->name, acct_name, NFACCT_NAME_MAX) != 0)
  79                         continue;
  80
  81                 if (nlh->nlmsg_flags & NLM_F_EXCL)
  82                         return -EEXIST;
  83
  84                 matching = nfacct;
  85                 break;
  86         }
  87
  88         if (matching) {
  89                 if (nlh->nlmsg_flags & NLM_F_REPLACE) {
  90                         /* reset counters if you request a replacement. */
  91                         atomic64_set(&matching->pkts, 0);
  92                         atomic64_set(&matching->bytes, 0);
  93                         smp_mb__before_atomic();
  94                         /* reset overquota flag if quota is enabled. */
  95                         if ((matching->flags & NFACCT_F_QUOTA))
  96                                 clear_bit(NFACCT_OVERQUOTA_BIT,
  97                                           &matching->flags);
  98                         return 0;
  99                 }
 100                 return -EBUSY;
 101         }
 102
 103         if (tb[NFACCT_FLAGS]) {
 104                 flags = ntohl(nla_get_be32(tb[NFACCT_FLAGS]));
 105                 if (flags & ~NFACCT_F_QUOTA)
 106                         return -EOPNOTSUPP;
 107                 if ((flags & NFACCT_F_QUOTA) == NFACCT_F_QUOTA)
 108                         return -EINVAL;
 109                 if (flags & NFACCT_F_OVERQUOTA)
 110                         return -EINVAL;
 111                 if ((flags & NFACCT_F_QUOTA) && !tb[NFACCT_QUOTA])
 112                         return -EINVAL;
 113
 114                 size += sizeof(u64);
 115         }
 116
 117         nfacct = kzalloc(sizeof(struct nf_acct) + size, GFP_KERNEL);
 118         if (nfacct == NULL)
 119                 return -ENOMEM;
 120
 121         if (flags & NFACCT_F_QUOTA) {
 122                 u64 *quota = (u64 *)nfacct->data;
 123
 124                 *quota = be64_to_cpu(nla_get_be64(tb[NFACCT_QUOTA]));
 125                 nfacct->flags = flags;
 126         }
 127
 128         nla_strscpy(nfacct->name, tb[NFACCT_NAME], NFACCT_NAME_MAX);
 129
 130         if (tb[NFACCT_BYTES]) {
 131                 atomic64_set(&nfacct->bytes,
 132                              be64_to_cpu(nla_get_be64(tb[NFACCT_BYTES])));
 133         }
 134         if (tb[NFACCT_PKTS]) {
 135                 atomic64_set(&nfacct->pkts,
 136                              be64_to_cpu(nla_get_be64(tb[NFACCT_PKTS])));
 137         }
 138         refcount_set(&nfacct->refcnt, 1);
 139         list_add_tail_rcu(&nfacct->head, &nfnl_acct_net->nfnl_acct_list);
 140         return 0;
 141 }
 142
 143 static int
 144 nfnl_acct_fill_info(struct sk_buff *skb, u32 portid, u32 seq, u32 type,
 145                    int event, struct nf_acct *acct)
 146 {
 147         struct nlmsghdr *nlh;
 148         unsigned int flags = portid ? NLM_F_MULTI : 0;
 149         u64 pkts, bytes;
 150         u32 old_flags;
 151
 152         event = nfnl_msg_type(NFNL_SUBSYS_ACCT, event);
 153         nlh = nfnl_msg_put(skb, portid, seq, event, flags, AF_UNSPEC,
 154                            NFNETLINK_V0, 0);
 155         if (!nlh)
 156                 goto nlmsg_failure;
 157
 158         if (nla_put_string(skb, NFACCT_NAME, acct->name))
 159                 goto nla_put_failure;
 160
 161         old_flags = acct->flags;
 162         if (type == NFNL_MSG_ACCT_GET_CTRZERO) {
 163                 pkts = atomic64_xchg(&acct->pkts, 0);
 164                 bytes = atomic64_xchg(&acct->bytes, 0);
 165                 smp_mb__before_atomic();
 166                 if (acct->flags & NFACCT_F_QUOTA)
 167                         clear_bit(NFACCT_OVERQUOTA_BIT, &acct->flags);
 168         } else {
 169                 pkts = atomic64_read(&acct->pkts);
 170                 bytes = atomic64_read(&acct->bytes);
 171         }
 172         if (nla_put_be64(skb, NFACCT_PKTS, cpu_to_be64(pkts),
 173                          NFACCT_PAD) ||
 174             nla_put_be64(skb, NFACCT_BYTES, cpu_to_be64(bytes),
 175                          NFACCT_PAD) ||
 176             nla_put_be32(skb, NFACCT_USE, htonl(refcount_read(&acct->refcnt))))
 177                 goto nla_put_failure;
 178         if (acct->flags & NFACCT_F_QUOTA) {
 179                 u64 *quota = (u64 *)acct->data;
 180
 181                 if (nla_put_be32(skb, NFACCT_FLAGS, htonl(old_flags)) ||
 182                     nla_put_be64(skb, NFACCT_QUOTA, cpu_to_be64(*quota),
 183                                  NFACCT_PAD))
 184                         goto nla_put_failure;
 185         }
 186         nlmsg_end(skb, nlh);
 187         return skb->len;
 188
 189 nlmsg_failure:
 190 nla_put_failure:
 191         nlmsg_cancel(skb, nlh);
 192         return -1;
 193 }
 194
 195 static int
 196 nfnl_acct_dump(struct sk_buff *skb, struct netlink_callback *cb)
 197 {
 198         struct net *net = sock_net(skb->sk);
 199         struct nfnl_acct_net *nfnl_acct_net = nfnl_acct_pernet(net);
 200         struct nf_acct *cur, *last;
 201         const struct nfacct_filter *filter = cb->data;
 202
 203         if (cb->args[2])
 204                 return 0;
 205
 206         last = (struct nf_acct *)cb->args[1];
 207         if (cb->args[1])
 208                 cb->args[1] = 0;
 209
 210         rcu_read_lock();
 211         list_for_each_entry_rcu(cur, &nfnl_acct_net->nfnl_acct_list, head) {
 212                 if (last) {
 213                         if (cur != last)
 214                                 continue;
 215
 216                         last = NULL;
 217                 }
 218
 219                 if (filter && (cur->flags & filter->mask) != filter->value)
 220                         continue;
 221
 222                 if (nfnl_acct_fill_info(skb, NETLINK_CB(cb->skb).portid,
 223                                        cb->nlh->nlmsg_seq,
 224                                        NFNL_MSG_TYPE(cb->nlh->nlmsg_type),
 225                                        NFNL_MSG_ACCT_NEW, cur) < 0) {
 226                         cb->args[1] = (unsigned long)cur;
 227                         break;
 228                 }
 229         }
 230         if (!cb->args[1])
 231                 cb->args[2] = 1;
 232         rcu_read_unlock();
 233         return skb->len;
 234 }
 235
 236 static int nfnl_acct_done(struct netlink_callback *cb)
 237 {
 238         kfree(cb->data);
 239         return 0;
 240 }
 241
 242 static const struct nla_policy filter_policy[NFACCT_FILTER_MAX + 1] = {
 243         [NFACCT_FILTER_MASK]    = { .type = NLA_U32 },
 244         [NFACCT_FILTER_VALUE]   = { .type = NLA_U32 },
 245 };
 246
 247 static int nfnl_acct_start(struct netlink_callback *cb)
 248 {
 249         const struct nlattr *const attr = cb->data;
 250         struct nlattr *tb[NFACCT_FILTER_MAX + 1];
 251         struct nfacct_filter *filter;
 252         int err;
 253
 254         if (!attr)
 255                 return 0;
 256
 257         err = nla_parse_nested_deprecated(tb, NFACCT_FILTER_MAX, attr,
 258                                           filter_policy, NULL);
 259         if (err < 0)
 260                 return err;
 261
 262         if (!tb[NFACCT_FILTER_MASK] || !tb[NFACCT_FILTER_VALUE])
 263                 return -EINVAL;
 264
 265         filter = kzalloc(sizeof(struct nfacct_filter), GFP_KERNEL);
 266         if (!filter)
 267                 return -ENOMEM;
 268
 269         filter->mask = ntohl(nla_get_be32(tb[NFACCT_FILTER_MASK]));
 270         filter->value = ntohl(nla_get_be32(tb[NFACCT_FILTER_VALUE]));
 271         cb->data = filter;
 272
 273         return 0;
 274 }
 275
 276 static int nfnl_acct_get(struct net *net, struct sock *nfnl,
 277                          struct sk_buff *skb, const struct nlmsghdr *nlh,
 278                          const struct nlattr * const tb[],
 279                          struct netlink_ext_ack *extack)
 280 {
 281         struct nfnl_acct_net *nfnl_acct_net = nfnl_acct_pernet(net);
 282         int ret = -ENOENT;
 283         struct nf_acct *cur;
 284         char *acct_name;
 285
 286         if (nlh->nlmsg_flags & NLM_F_DUMP) {
 287                 struct netlink_dump_control c = {
 288                         .dump = nfnl_acct_dump,
 289                         .start = nfnl_acct_start,
 290                         .done = nfnl_acct_done,
 291                         .data = (void *)tb[NFACCT_FILTER],
 292                 };
 293
 294                 return netlink_dump_start(nfnl, skb, nlh, &c);
 295         }
 296
 297         if (!tb[NFACCT_NAME])
 298                 return -EINVAL;
 299         acct_name = nla_data(tb[NFACCT_NAME]);
 300
 301         list_for_each_entry(cur, &nfnl_acct_net->nfnl_acct_list, head) {
 302                 struct sk_buff *skb2;
 303
 304                 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX)!= 0)
 305                         continue;
 306
 307                 skb2 = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
 308                 if (skb2 == NULL) {
 309                         ret = -ENOMEM;
 310                         break;
 311                 }
 312
 313                 ret = nfnl_acct_fill_info(skb2, NETLINK_CB(skb).portid,
 314                                          nlh->nlmsg_seq,
 315                                          NFNL_MSG_TYPE(nlh->nlmsg_type),
 316                                          NFNL_MSG_ACCT_NEW, cur);
 317                 if (ret <= 0) {
 318                         kfree_skb(skb2);
 319                         break;
 320                 }
 321                 ret = netlink_unicast(nfnl, skb2, NETLINK_CB(skb).portid,
 322                                         MSG_DONTWAIT);
 323                 if (ret > 0)
 324                         ret = 0;
 325
 326                 /* this avoids a loop in nfnetlink. */
 327                 return ret == -EAGAIN ? -ENOBUFS : ret;
 328         }
 329         return ret;
 330 }
 331
 332 /* try to delete object, fail if it is still in use. */
 333 static int nfnl_acct_try_del(struct nf_acct *cur)
 334 {
 335         int ret = 0;
 336
 337         /* We want to avoid races with nfnl_acct_put. So only when the current
 338          * refcnt is 1, we decrease it to 0.
 339          */
 340         if (refcount_dec_if_one(&cur->refcnt)) {
 341                 /* We are protected by nfnl mutex. */
 342                 list_del_rcu(&cur->head);
 343                 kfree_rcu(cur, rcu_head);
 344         } else {
 345                 ret = -EBUSY;
 346         }
 347         return ret;
 348 }
 349
 350 static int nfnl_acct_del(struct net *net, struct sock *nfnl,
 351                          struct sk_buff *skb, const struct nlmsghdr *nlh,
 352                          const struct nlattr * const tb[],
 353                          struct netlink_ext_ack *extack)
 354 {
 355         struct nfnl_acct_net *nfnl_acct_net = nfnl_acct_pernet(net);
 356         struct nf_acct *cur, *tmp;
 357         int ret = -ENOENT;
 358         char *acct_name;
 359
 360         if (!tb[NFACCT_NAME]) {
 361                 list_for_each_entry_safe(cur, tmp, &nfnl_acct_net->nfnl_acct_list, head)
 362                         nfnl_acct_try_del(cur);
 363
 364                 return 0;
 365         }
 366         acct_name = nla_data(tb[NFACCT_NAME]);
 367
 368         list_for_each_entry(cur, &nfnl_acct_net->nfnl_acct_list, head) {
 369                 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX) != 0)
 370                         continue;
 371
 372                 ret = nfnl_acct_try_del(cur);
 373                 if (ret < 0)
 374                         return ret;
 375
 376                 break;
 377         }
 378         return ret;
 379 }
 380
 381 static const struct nla_policy nfnl_acct_policy[NFACCT_MAX+1] = {
 382         [NFACCT_NAME] = { .type = NLA_NUL_STRING, .len = NFACCT_NAME_MAX-1 },
 383         [NFACCT_BYTES] = { .type = NLA_U64 },
 384         [NFACCT_PKTS] = { .type = NLA_U64 },
 385         [NFACCT_FLAGS] = { .type = NLA_U32 },
 386         [NFACCT_QUOTA] = { .type = NLA_U64 },
 387         [NFACCT_FILTER] = {.type = NLA_NESTED },
 388 };
 389
 390 static const struct nfnl_callback nfnl_acct_cb[NFNL_MSG_ACCT_MAX] = {
 391         [NFNL_MSG_ACCT_NEW]             = { .call = nfnl_acct_new,
 392                                             .attr_count = NFACCT_MAX,
 393                                             .policy = nfnl_acct_policy },
 394         [NFNL_MSG_ACCT_GET]             = { .call = nfnl_acct_get,
 395                                             .attr_count = NFACCT_MAX,
 396                                             .policy = nfnl_acct_policy },
 397         [NFNL_MSG_ACCT_GET_CTRZERO]     = { .call = nfnl_acct_get,
 398                                             .attr_count = NFACCT_MAX,
 399                                             .policy = nfnl_acct_policy },
 400         [NFNL_MSG_ACCT_DEL]             = { .call = nfnl_acct_del,
 401                                             .attr_count = NFACCT_MAX,
 402                                             .policy = nfnl_acct_policy },
 403 };
 404
 405 static const struct nfnetlink_subsystem nfnl_acct_subsys = {
 406         .name                           = "acct",
 407         .subsys_id                      = NFNL_SUBSYS_ACCT,
 408         .cb_count                       = NFNL_MSG_ACCT_MAX,
 409         .cb                             = nfnl_acct_cb,
 410 };
 411
 412 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ACCT);
 413
 414 struct nf_acct *nfnl_acct_find_get(struct net *net, const char *acct_name)
 415 {
 416         struct nfnl_acct_net *nfnl_acct_net = nfnl_acct_pernet(net);
 417         struct nf_acct *cur, *acct = NULL;
 418
 419         rcu_read_lock();
 420         list_for_each_entry_rcu(cur, &nfnl_acct_net->nfnl_acct_list, head) {
 421                 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX)!= 0)
 422                         continue;
 423
 424                 if (!try_module_get(THIS_MODULE))
 425                         goto err;
 426
 427                 if (!refcount_inc_not_zero(&cur->refcnt)) {
 428                         module_put(THIS_MODULE);
 429                         goto err;
 430                 }
 431
 432                 acct = cur;
 433                 break;
 434         }
 435 err:
 436         rcu_read_unlock();
 437         return acct;
 438 }
 439 EXPORT_SYMBOL_GPL(nfnl_acct_find_get);
 440
 441 void nfnl_acct_put(struct nf_acct *acct)
 442 {
 443         if (refcount_dec_and_test(&acct->refcnt))
 444                 kfree_rcu(acct, rcu_head);
 445
 446         module_put(THIS_MODULE);
 447 }
 448 EXPORT_SYMBOL_GPL(nfnl_acct_put);
 449
 450 void nfnl_acct_update(const struct sk_buff *skb, struct nf_acct *nfacct)
 451 {
 452         atomic64_inc(&nfacct->pkts);
 453         atomic64_add(skb->len, &nfacct->bytes);
 454 }
 455 EXPORT_SYMBOL_GPL(nfnl_acct_update);
 456
 457 static void nfnl_overquota_report(struct net *net, struct nf_acct *nfacct)
 458 {
 459         int ret;
 460         struct sk_buff *skb;
 461
 462         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
 463         if (skb == NULL)
 464                 return;
 465
 466         ret = nfnl_acct_fill_info(skb, 0, 0, NFNL_MSG_ACCT_OVERQUOTA, 0,
 467                                   nfacct);
 468         if (ret <= 0) {
 469                 kfree_skb(skb);
 470                 return;
 471         }
 472         netlink_broadcast(net->nfnl, skb, 0, NFNLGRP_ACCT_QUOTA,
 473                           GFP_ATOMIC);
 474 }
 475
 476 int nfnl_acct_overquota(struct net *net, struct nf_acct *nfacct)
 477 {
 478         u64 now;
 479         u64 *quota;
 480         int ret = NFACCT_UNDERQUOTA;
 481
 482         /* no place here if we don't have a quota */
 483         if (!(nfacct->flags & NFACCT_F_QUOTA))
 484                 return NFACCT_NO_QUOTA;
 485
 486         quota = (u64 *)nfacct->data;
 487         now = (nfacct->flags & NFACCT_F_QUOTA_PKTS) ?
 488                atomic64_read(&nfacct->pkts) : atomic64_read(&nfacct->bytes);
 489
 490         ret = now > *quota;
 491
 492         if (now >= *quota &&
 493             !test_and_set_bit(NFACCT_OVERQUOTA_BIT, &nfacct->flags)) {
 494                 nfnl_overquota_report(net, nfacct);
 495         }
 496
 497         return ret;
 498 }
 499 EXPORT_SYMBOL_GPL(nfnl_acct_overquota);
 500
 501 static int __net_init nfnl_acct_net_init(struct net *net)
 502 {
 503         INIT_LIST_HEAD(&nfnl_acct_pernet(net)->nfnl_acct_list);
 504
 505         return 0;
 506 }
 507
 508 static void __net_exit nfnl_acct_net_exit(struct net *net)
 509 {
 510         struct nfnl_acct_net *nfnl_acct_net = nfnl_acct_pernet(net);
 511         struct nf_acct *cur, *tmp;
 512
 513         list_for_each_entry_safe(cur, tmp, &nfnl_acct_net->nfnl_acct_list, head) {
 514                 list_del_rcu(&cur->head);
 515
 516                 if (refcount_dec_and_test(&cur->refcnt))
 517                         kfree_rcu(cur, rcu_head);
 518         }
 519 }
 520
 521 static struct pernet_operations nfnl_acct_ops = {
 522         .init   = nfnl_acct_net_init,
 523         .exit   = nfnl_acct_net_exit,
 524         .id     = &nfnl_acct_net_id,
 525         .size   = sizeof(struct nfnl_acct_net),
 526 };
 527
 528 static int __init nfnl_acct_init(void)
 529 {
 530         int ret;
 531
 532         ret = register_pernet_subsys(&nfnl_acct_ops);
 533         if (ret < 0) {
 534                 pr_err("nfnl_acct_init: failed to register pernet ops\n");
 535                 goto err_out;
 536         }
 537
 538         ret = nfnetlink_subsys_register(&nfnl_acct_subsys);
 539         if (ret < 0) {
 540                 pr_err("nfnl_acct_init: cannot register with nfnetlink.\n");
 541                 goto cleanup_pernet;
 542         }
 543         return 0;
 544
 545 cleanup_pernet:
 546         unregister_pernet_subsys(&nfnl_acct_ops);
 547 err_out:
 548         return ret;
 549 }
 550
 551 static void __exit nfnl_acct_exit(void)
 552 {
 553         nfnetlink_subsys_unregister(&nfnl_acct_subsys);
 554         unregister_pernet_subsys(&nfnl_acct_ops);
 555 }
 556
 557 module_init(nfnl_acct_init);
 558 module_exit(nfnl_acct_exit);