net/sched/act_police.c

   1 /*
   2  * net/sched/act_police.c       Input police filter
   3  *
   4  *              This program is free software; you can redistribute it and/or
   5  *              modify it under the terms of the GNU General Public License
   6  *              as published by the Free Software Foundation; either version
   7  *              2 of the License, or (at your option) any later version.
   8  *
   9  * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
  10  *              J Hadi Salim (action changes)
  11  */
  12
  13 #include <linux/module.h>
  14 #include <linux/types.h>
  15 #include <linux/kernel.h>
  16 #include <linux/string.h>
  17 #include <linux/errno.h>
  18 #include <linux/skbuff.h>
  19 #include <linux/rtnetlink.h>
  20 #include <linux/init.h>
  21 #include <linux/slab.h>
  22 #include <net/act_api.h>
  23 #include <net/netlink.h>
  24
  25 struct tcf_police {
  26         struct tcf_common       common;
  27         int                     tcfp_result;
  28         u32                     tcfp_ewma_rate;
  29         s64                     tcfp_burst;
  30         u32                     tcfp_mtu;
  31         s64                     tcfp_toks;
  32         s64                     tcfp_ptoks;
  33         s64                     tcfp_mtu_ptoks;
  34         s64                     tcfp_t_c;
  35         struct psched_ratecfg   rate;
  36         bool                    rate_present;
  37         struct psched_ratecfg   peak;
  38         bool                    peak_present;
  39 };
  40 #define to_police(pc)   \
  41         container_of(pc, struct tcf_police, common)
  42
  43 #define POL_TAB_MASK     15
  44
  45 /* old policer structure from before tc actions */
  46 struct tc_police_compat {
  47         u32                     index;
  48         int                     action;
  49         u32                     limit;
  50         u32                     burst;
  51         u32                     mtu;
  52         struct tc_ratespec      rate;
  53         struct tc_ratespec      peakrate;
  54 };
  55
  56 /* Each policer is serialized by its individual spinlock */
  57
  58 static int police_net_id;
  59
  60 static int tcf_act_police_walker(struct net *net, struct sk_buff *skb,
  61                                  struct netlink_callback *cb, int type,
  62                                  struct tc_action *a)
  63 {
  64         struct tc_action_net *tn = net_generic(net, police_net_id);
  65         struct tcf_hashinfo *hinfo = tn->hinfo;
  66         struct hlist_head *head;
  67         struct tcf_common *p;
  68         int err = 0, index = -1, i = 0, s_i = 0, n_i = 0;
  69         struct nlattr *nest;
  70
  71         spin_lock_bh(&hinfo->lock);
  72
  73         s_i = cb->args[0];
  74
  75         for (i = 0; i < (POL_TAB_MASK + 1); i++) {
  76                 head = &hinfo->htab[tcf_hash(i, POL_TAB_MASK)];
  77
  78                 hlist_for_each_entry_rcu(p, head, tcfc_head) {
  79                         index++;
  80                         if (index < s_i)
  81                                 continue;
  82                         a->priv = p;
  83                         a->order = index;
  84                         nest = nla_nest_start(skb, a->order);
  85                         if (nest == NULL)
  86                                 goto nla_put_failure;
  87                         if (type == RTM_DELACTION)
  88                                 err = tcf_action_dump_1(skb, a, 0, 1);
  89                         else
  90                                 err = tcf_action_dump_1(skb, a, 0, 0);
  91                         if (err < 0) {
  92                                 index--;
  93                                 nla_nest_cancel(skb, nest);
  94                                 goto done;
  95                         }
  96                         nla_nest_end(skb, nest);
  97                         n_i++;
  98                 }
  99         }
 100 done:
 101         spin_unlock_bh(&hinfo->lock);
 102         if (n_i)
 103                 cb->args[0] += n_i;
 104         return n_i;
 105
 106 nla_put_failure:
 107         nla_nest_cancel(skb, nest);
 108         goto done;
 109 }
 110
 111 static const struct nla_policy police_policy[TCA_POLICE_MAX + 1] = {
 112         [TCA_POLICE_RATE]       = { .len = TC_RTAB_SIZE },
 113         [TCA_POLICE_PEAKRATE]   = { .len = TC_RTAB_SIZE },
 114         [TCA_POLICE_AVRATE]     = { .type = NLA_U32 },
 115         [TCA_POLICE_RESULT]     = { .type = NLA_U32 },
 116 };
 117
 118 static int tcf_act_police_locate(struct net *net, struct nlattr *nla,
 119                                  struct nlattr *est, struct tc_action *a,
 120                                  int ovr, int bind)
 121 {
 122         unsigned int h;
 123         int ret = 0, err;
 124         struct nlattr *tb[TCA_POLICE_MAX + 1];
 125         struct tc_police *parm;
 126         struct tcf_police *police;
 127         struct qdisc_rate_table *R_tab = NULL, *P_tab = NULL;
 128         struct tc_action_net *tn = net_generic(net, police_net_id);
 129         struct tcf_hashinfo *hinfo = tn->hinfo;
 130         int size;
 131
 132         if (nla == NULL)
 133                 return -EINVAL;
 134
 135         err = nla_parse_nested(tb, TCA_POLICE_MAX, nla, police_policy);
 136         if (err < 0)
 137                 return err;
 138
 139         if (tb[TCA_POLICE_TBF] == NULL)
 140                 return -EINVAL;
 141         size = nla_len(tb[TCA_POLICE_TBF]);
 142         if (size != sizeof(*parm) && size != sizeof(struct tc_police_compat))
 143                 return -EINVAL;
 144         parm = nla_data(tb[TCA_POLICE_TBF]);
 145
 146         if (parm->index) {
 147                 if (tcf_hash_search(tn, a, parm->index)) {
 148                         police = to_police(a->priv);
 149                         if (bind) {
 150                                 police->tcf_bindcnt += 1;
 151                                 police->tcf_refcnt += 1;
 152                                 return 0;
 153                         }
 154                         if (ovr)
 155                                 goto override;
 156                         /* not replacing */
 157                         return -EEXIST;
 158                 }
 159         }
 160
 161         police = kzalloc(sizeof(*police), GFP_KERNEL);
 162         if (police == NULL)
 163                 return -ENOMEM;
 164         ret = ACT_P_CREATED;
 165         police->tcf_refcnt = 1;
 166         spin_lock_init(&police->tcf_lock);
 167         if (bind)
 168                 police->tcf_bindcnt = 1;
 169 override:
 170         if (parm->rate.rate) {
 171                 err = -ENOMEM;
 172                 R_tab = qdisc_get_rtab(&parm->rate, tb[TCA_POLICE_RATE]);
 173                 if (R_tab == NULL)
 174                         goto failure;
 175
 176                 if (parm->peakrate.rate) {
 177                         P_tab = qdisc_get_rtab(&parm->peakrate,
 178                                                tb[TCA_POLICE_PEAKRATE]);
 179                         if (P_tab == NULL)
 180                                 goto failure;
 181                 }
 182         }
 183
 184         spin_lock_bh(&police->tcf_lock);
 185         if (est) {
 186                 err = gen_replace_estimator(&police->tcf_bstats, NULL,
 187                                             &police->tcf_rate_est,
 188                                             &police->tcf_lock, est);
 189                 if (err)
 190                         goto failure_unlock;
 191         } else if (tb[TCA_POLICE_AVRATE] &&
 192                    (ret == ACT_P_CREATED ||
 193                     !gen_estimator_active(&police->tcf_bstats,
 194                                           &police->tcf_rate_est))) {
 195                 err = -EINVAL;
 196                 goto failure_unlock;
 197         }
 198
 199         /* No failure allowed after this point */
 200         police->tcfp_mtu = parm->mtu;
 201         if (police->tcfp_mtu == 0) {
 202                 police->tcfp_mtu = ~0;
 203                 if (R_tab)
 204                         police->tcfp_mtu = 255 << R_tab->rate.cell_log;
 205         }
 206         if (R_tab) {
 207                 police->rate_present = true;
 208                 psched_ratecfg_precompute(&police->rate, &R_tab->rate, 0);
 209                 qdisc_put_rtab(R_tab);
 210         } else {
 211                 police->rate_present = false;
 212         }
 213         if (P_tab) {
 214                 police->peak_present = true;
 215                 psched_ratecfg_precompute(&police->peak, &P_tab->rate, 0);
 216                 qdisc_put_rtab(P_tab);
 217         } else {
 218                 police->peak_present = false;
 219         }
 220
 221         if (tb[TCA_POLICE_RESULT])
 222                 police->tcfp_result = nla_get_u32(tb[TCA_POLICE_RESULT]);
 223         police->tcfp_burst = PSCHED_TICKS2NS(parm->burst);
 224         police->tcfp_toks = police->tcfp_burst;
 225         if (police->peak_present) {
 226                 police->tcfp_mtu_ptoks = (s64) psched_l2t_ns(&police->peak,
 227                                                              police->tcfp_mtu);
 228                 police->tcfp_ptoks = police->tcfp_mtu_ptoks;
 229         }
 230         police->tcf_action = parm->action;
 231
 232         if (tb[TCA_POLICE_AVRATE])
 233                 police->tcfp_ewma_rate = nla_get_u32(tb[TCA_POLICE_AVRATE]);
 234
 235         spin_unlock_bh(&police->tcf_lock);
 236         if (ret != ACT_P_CREATED)
 237                 return ret;
 238
 239         police->tcfp_t_c = ktime_get_ns();
 240         police->tcf_index = parm->index ? parm->index :
 241                 tcf_hash_new_index(tn);
 242         h = tcf_hash(police->tcf_index, POL_TAB_MASK);
 243         spin_lock_bh(&hinfo->lock);
 244         hlist_add_head(&police->tcf_head, &hinfo->htab[h]);
 245         spin_unlock_bh(&hinfo->lock);
 246
 247         a->priv = police;
 248         return ret;
 249
 250 failure_unlock:
 251         spin_unlock_bh(&police->tcf_lock);
 252 failure:
 253         qdisc_put_rtab(P_tab);
 254         qdisc_put_rtab(R_tab);
 255         if (ret == ACT_P_CREATED)
 256                 kfree(police);
 257         return err;
 258 }
 259
 260 static int tcf_act_police(struct sk_buff *skb, const struct tc_action *a,
 261                           struct tcf_result *res)
 262 {
 263         struct tcf_police *police = a->priv;
 264         s64 now;
 265         s64 toks;
 266         s64 ptoks = 0;
 267
 268         spin_lock(&police->tcf_lock);
 269
 270         bstats_update(&police->tcf_bstats, skb);
 271
 272         if (police->tcfp_ewma_rate &&
 273             police->tcf_rate_est.bps >= police->tcfp_ewma_rate) {
 274                 police->tcf_qstats.overlimits++;
 275                 if (police->tcf_action == TC_ACT_SHOT)
 276                         police->tcf_qstats.drops++;
 277                 spin_unlock(&police->tcf_lock);
 278                 return police->tcf_action;
 279         }
 280
 281         if (qdisc_pkt_len(skb) <= police->tcfp_mtu) {
 282                 if (!police->rate_present) {
 283                         spin_unlock(&police->tcf_lock);
 284                         return police->tcfp_result;
 285                 }
 286
 287                 now = ktime_get_ns();
 288                 toks = min_t(s64, now - police->tcfp_t_c,
 289                              police->tcfp_burst);
 290                 if (police->peak_present) {
 291                         ptoks = toks + police->tcfp_ptoks;
 292                         if (ptoks > police->tcfp_mtu_ptoks)
 293                                 ptoks = police->tcfp_mtu_ptoks;
 294                         ptoks -= (s64) psched_l2t_ns(&police->peak,
 295                                                      qdisc_pkt_len(skb));
 296                 }
 297                 toks += police->tcfp_toks;
 298                 if (toks > police->tcfp_burst)
 299                         toks = police->tcfp_burst;
 300                 toks -= (s64) psched_l2t_ns(&police->rate, qdisc_pkt_len(skb));
 301                 if ((toks|ptoks) >= 0) {
 302                         police->tcfp_t_c = now;
 303                         police->tcfp_toks = toks;
 304                         police->tcfp_ptoks = ptoks;
 305                         spin_unlock(&police->tcf_lock);
 306                         return police->tcfp_result;
 307                 }
 308         }
 309
 310         police->tcf_qstats.overlimits++;
 311         if (police->tcf_action == TC_ACT_SHOT)
 312                 police->tcf_qstats.drops++;
 313         spin_unlock(&police->tcf_lock);
 314         return police->tcf_action;
 315 }
 316
 317 static int
 318 tcf_act_police_dump(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
 319 {
 320         unsigned char *b = skb_tail_pointer(skb);
 321         struct tcf_police *police = a->priv;
 322         struct tc_police opt = {
 323                 .index = police->tcf_index,
 324                 .action = police->tcf_action,
 325                 .mtu = police->tcfp_mtu,
 326                 .burst = PSCHED_NS2TICKS(police->tcfp_burst),
 327                 .refcnt = police->tcf_refcnt - ref,
 328                 .bindcnt = police->tcf_bindcnt - bind,
 329         };
 330
 331         if (police->rate_present)
 332                 psched_ratecfg_getrate(&opt.rate, &police->rate);
 333         if (police->peak_present)
 334                 psched_ratecfg_getrate(&opt.peakrate, &police->peak);
 335         if (nla_put(skb, TCA_POLICE_TBF, sizeof(opt), &opt))
 336                 goto nla_put_failure;
 337         if (police->tcfp_result &&
 338             nla_put_u32(skb, TCA_POLICE_RESULT, police->tcfp_result))
 339                 goto nla_put_failure;
 340         if (police->tcfp_ewma_rate &&
 341             nla_put_u32(skb, TCA_POLICE_AVRATE, police->tcfp_ewma_rate))
 342                 goto nla_put_failure;
 343         return skb->len;
 344
 345 nla_put_failure:
 346         nlmsg_trim(skb, b);
 347         return -1;
 348 }
 349
 350 static int tcf_police_search(struct net *net, struct tc_action *a, u32 index)
 351 {
 352         struct tc_action_net *tn = net_generic(net, police_net_id);
 353
 354         return tcf_hash_search(tn, a, index);
 355 }
 356
 357 MODULE_AUTHOR("Alexey Kuznetsov");
 358 MODULE_DESCRIPTION("Policing actions");
 359 MODULE_LICENSE("GPL");
 360
 361 static struct tc_action_ops act_police_ops = {
 362         .kind           =       "police",
 363         .type           =       TCA_ID_POLICE,
 364         .owner          =       THIS_MODULE,
 365         .act            =       tcf_act_police,
 366         .dump           =       tcf_act_police_dump,
 367         .init           =       tcf_act_police_locate,
 368         .walk           =       tcf_act_police_walker,
 369         .lookup         =       tcf_police_search,
 370 };
 371
 372 static __net_init int police_init_net(struct net *net)
 373 {
 374         struct tc_action_net *tn = net_generic(net, police_net_id);
 375
 376         return tc_action_net_init(tn, &act_police_ops, POL_TAB_MASK);
 377 }
 378
 379 static void __net_exit police_exit_net(struct net *net)
 380 {
 381         struct tc_action_net *tn = net_generic(net, police_net_id);
 382
 383         tc_action_net_exit(tn);
 384 }
 385
 386 static struct pernet_operations police_net_ops = {
 387         .init = police_init_net,
 388         .exit = police_exit_net,
 389         .id   = &police_net_id,
 390         .size = sizeof(struct tc_action_net),
 391 };
 392
 393 static int __init
 394 police_init_module(void)
 395 {
 396         return tcf_register_action(&act_police_ops, &police_net_ops);
 397 }
 398
 399 static void __exit
 400 police_cleanup_module(void)
 401 {
 402         tcf_unregister_action(&act_police_ops, &police_net_ops);
 403 }
 404
 405 module_init(police_init_module);
 406 module_exit(police_cleanup_module);