net/sched/act_police.c

   1 /*
   2  * net/sched/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 static struct tcf_hashinfo police_hash_info;
  45
  46 /* old policer structure from before tc actions */
  47 struct tc_police_compat {
  48         u32                     index;
  49         int                     action;
  50         u32                     limit;
  51         u32                     burst;
  52         u32                     mtu;
  53         struct tc_ratespec      rate;
  54         struct tc_ratespec      peakrate;
  55 };
  56
  57 /* Each policer is serialized by its individual spinlock */
  58
  59 static int tcf_act_police_walker(struct sk_buff *skb, struct netlink_callback *cb,
  60                               int type, struct tc_action *a)
  61 {
  62         struct tcf_hashinfo *hinfo = a->ops->hinfo;
  63         struct hlist_head *head;
  64         struct tcf_common *p;
  65         int err = 0, index = -1, i = 0, s_i = 0, n_i = 0;
  66         struct nlattr *nest;
  67
  68         spin_lock_bh(&hinfo->lock);
  69
  70         s_i = cb->args[0];
  71
  72         for (i = 0; i < (POL_TAB_MASK + 1); i++) {
  73                 head = &hinfo->htab[tcf_hash(i, POL_TAB_MASK)];
  74
  75                 hlist_for_each_entry_rcu(p, head, tcfc_head) {
  76                         index++;
  77                         if (index < s_i)
  78                                 continue;
  79                         a->priv = p;
  80                         a->order = index;
  81                         nest = nla_nest_start(skb, a->order);
  82                         if (nest == NULL)
  83                                 goto nla_put_failure;
  84                         if (type == RTM_DELACTION)
  85                                 err = tcf_action_dump_1(skb, a, 0, 1);
  86                         else
  87                                 err = tcf_action_dump_1(skb, a, 0, 0);
  88                         if (err < 0) {
  89                                 index--;
  90                                 nla_nest_cancel(skb, nest);
  91                                 goto done;
  92                         }
  93                         nla_nest_end(skb, nest);
  94                         n_i++;
  95                 }
  96         }
  97 done:
  98         spin_unlock_bh(&hinfo->lock);
  99         if (n_i)
 100                 cb->args[0] += n_i;
 101         return n_i;
 102
 103 nla_put_failure:
 104         nla_nest_cancel(skb, nest);
 105         goto done;
 106 }
 107
 108 static const struct nla_policy police_policy[TCA_POLICE_MAX + 1] = {
 109         [TCA_POLICE_RATE]       = { .len = TC_RTAB_SIZE },
 110         [TCA_POLICE_PEAKRATE]   = { .len = TC_RTAB_SIZE },
 111         [TCA_POLICE_AVRATE]     = { .type = NLA_U32 },
 112         [TCA_POLICE_RESULT]     = { .type = NLA_U32 },
 113 };
 114
 115 static int tcf_act_police_locate(struct net *net, struct nlattr *nla,
 116                                  struct nlattr *est, struct tc_action *a,
 117                                  int ovr, int bind)
 118 {
 119         unsigned int h;
 120         int ret = 0, err;
 121         struct nlattr *tb[TCA_POLICE_MAX + 1];
 122         struct tc_police *parm;
 123         struct tcf_police *police;
 124         struct qdisc_rate_table *R_tab = NULL, *P_tab = NULL;
 125         struct tcf_hashinfo *hinfo = a->ops->hinfo;
 126         int size;
 127
 128         if (nla == NULL)
 129                 return -EINVAL;
 130
 131         err = nla_parse_nested(tb, TCA_POLICE_MAX, nla, police_policy);
 132         if (err < 0)
 133                 return err;
 134
 135         if (tb[TCA_POLICE_TBF] == NULL)
 136                 return -EINVAL;
 137         size = nla_len(tb[TCA_POLICE_TBF]);
 138         if (size != sizeof(*parm) && size != sizeof(struct tc_police_compat))
 139                 return -EINVAL;
 140         parm = nla_data(tb[TCA_POLICE_TBF]);
 141
 142         if (parm->index) {
 143                 if (tcf_hash_search(a, parm->index)) {
 144                         police = to_police(a->priv);
 145                         if (bind) {
 146                                 police->tcf_bindcnt += 1;
 147                                 police->tcf_refcnt += 1;
 148                                 return 0;
 149                         }
 150                         if (ovr)
 151                                 goto override;
 152                         /* not replacing */
 153                         return -EEXIST;
 154                 }
 155         }
 156
 157         police = kzalloc(sizeof(*police), GFP_KERNEL);
 158         if (police == NULL)
 159                 return -ENOMEM;
 160         ret = ACT_P_CREATED;
 161         police->tcf_refcnt = 1;
 162         spin_lock_init(&police->tcf_lock);
 163         if (bind)
 164                 police->tcf_bindcnt = 1;
 165 override:
 166         if (parm->rate.rate) {
 167                 err = -ENOMEM;
 168                 R_tab = qdisc_get_rtab(&parm->rate, tb[TCA_POLICE_RATE]);
 169                 if (R_tab == NULL)
 170                         goto failure;
 171
 172                 if (parm->peakrate.rate) {
 173                         P_tab = qdisc_get_rtab(&parm->peakrate,
 174                                                tb[TCA_POLICE_PEAKRATE]);
 175                         if (P_tab == NULL)
 176                                 goto failure;
 177                 }
 178         }
 179
 180         spin_lock_bh(&police->tcf_lock);
 181         if (est) {
 182                 err = gen_replace_estimator(&police->tcf_bstats,
 183                                             &police->tcf_rate_est,
 184                                             &police->tcf_lock, est);
 185                 if (err)
 186                         goto failure_unlock;
 187         } else if (tb[TCA_POLICE_AVRATE] &&
 188                    (ret == ACT_P_CREATED ||
 189                     !gen_estimator_active(&police->tcf_bstats,
 190                                           &police->tcf_rate_est))) {
 191                 err = -EINVAL;
 192                 goto failure_unlock;
 193         }
 194
 195         /* No failure allowed after this point */
 196         police->tcfp_mtu = parm->mtu;
 197         if (police->tcfp_mtu == 0) {
 198                 police->tcfp_mtu = ~0;
 199                 if (R_tab)
 200                         police->tcfp_mtu = 255 << R_tab->rate.cell_log;
 201         }
 202         if (R_tab) {
 203                 police->rate_present = true;
 204                 psched_ratecfg_precompute(&police->rate, &R_tab->rate, 0);
 205                 qdisc_put_rtab(R_tab);
 206         } else {
 207                 police->rate_present = false;
 208         }
 209         if (P_tab) {
 210                 police->peak_present = true;
 211                 psched_ratecfg_precompute(&police->peak, &P_tab->rate, 0);
 212                 qdisc_put_rtab(P_tab);
 213         } else {
 214                 police->peak_present = false;
 215         }
 216
 217         if (tb[TCA_POLICE_RESULT])
 218                 police->tcfp_result = nla_get_u32(tb[TCA_POLICE_RESULT]);
 219         police->tcfp_burst = PSCHED_TICKS2NS(parm->burst);
 220         police->tcfp_toks = police->tcfp_burst;
 221         if (police->peak_present) {
 222                 police->tcfp_mtu_ptoks = (s64) psched_l2t_ns(&police->peak,
 223                                                              police->tcfp_mtu);
 224                 police->tcfp_ptoks = police->tcfp_mtu_ptoks;
 225         }
 226         police->tcf_action = parm->action;
 227
 228         if (tb[TCA_POLICE_AVRATE])
 229                 police->tcfp_ewma_rate = nla_get_u32(tb[TCA_POLICE_AVRATE]);
 230
 231         spin_unlock_bh(&police->tcf_lock);
 232         if (ret != ACT_P_CREATED)
 233                 return ret;
 234
 235         police->tcfp_t_c = ktime_to_ns(ktime_get());
 236         police->tcf_index = parm->index ? parm->index :
 237                 tcf_hash_new_index(a->ops->hinfo);
 238         h = tcf_hash(police->tcf_index, POL_TAB_MASK);
 239         spin_lock_bh(&hinfo->lock);
 240         hlist_add_head(&police->tcf_head, &hinfo->htab[h]);
 241         spin_unlock_bh(&hinfo->lock);
 242
 243         a->priv = police;
 244         return ret;
 245
 246 failure_unlock:
 247         spin_unlock_bh(&police->tcf_lock);
 248 failure:
 249         qdisc_put_rtab(P_tab);
 250         qdisc_put_rtab(R_tab);
 251         if (ret == ACT_P_CREATED)
 252                 kfree(police);
 253         return err;
 254 }
 255
 256 static int tcf_act_police_cleanup(struct tc_action *a, int bind)
 257 {
 258         struct tcf_police *p = a->priv;
 259         if (p)
 260                 return tcf_hash_release(&p->common, bind, &police_hash_info);
 261         return 0;
 262 }
 263
 264 static int tcf_act_police(struct sk_buff *skb, const struct tc_action *a,
 265                           struct tcf_result *res)
 266 {
 267         struct tcf_police *police = a->priv;
 268         s64 now;
 269         s64 toks;
 270         s64 ptoks = 0;
 271
 272         spin_lock(&police->tcf_lock);
 273
 274         bstats_update(&police->tcf_bstats, skb);
 275
 276         if (police->tcfp_ewma_rate &&
 277             police->tcf_rate_est.bps >= police->tcfp_ewma_rate) {
 278                 police->tcf_qstats.overlimits++;
 279                 if (police->tcf_action == TC_ACT_SHOT)
 280                         police->tcf_qstats.drops++;
 281                 spin_unlock(&police->tcf_lock);
 282                 return police->tcf_action;
 283         }
 284
 285         if (qdisc_pkt_len(skb) <= police->tcfp_mtu) {
 286                 if (!police->rate_present) {
 287                         spin_unlock(&police->tcf_lock);
 288                         return police->tcfp_result;
 289                 }
 290
 291                 now = ktime_to_ns(ktime_get());
 292                 toks = min_t(s64, now - police->tcfp_t_c,
 293                              police->tcfp_burst);
 294                 if (police->peak_present) {
 295                         ptoks = toks + police->tcfp_ptoks;
 296                         if (ptoks > police->tcfp_mtu_ptoks)
 297                                 ptoks = police->tcfp_mtu_ptoks;
 298                         ptoks -= (s64) psched_l2t_ns(&police->peak,
 299                                                      qdisc_pkt_len(skb));
 300                 }
 301                 toks += police->tcfp_toks;
 302                 if (toks > police->tcfp_burst)
 303                         toks = police->tcfp_burst;
 304                 toks -= (s64) psched_l2t_ns(&police->rate, qdisc_pkt_len(skb));
 305                 if ((toks|ptoks) >= 0) {
 306                         police->tcfp_t_c = now;
 307                         police->tcfp_toks = toks;
 308                         police->tcfp_ptoks = ptoks;
 309                         spin_unlock(&police->tcf_lock);
 310                         return police->tcfp_result;
 311                 }
 312         }
 313
 314         police->tcf_qstats.overlimits++;
 315         if (police->tcf_action == TC_ACT_SHOT)
 316                 police->tcf_qstats.drops++;
 317         spin_unlock(&police->tcf_lock);
 318         return police->tcf_action;
 319 }
 320
 321 static int
 322 tcf_act_police_dump(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
 323 {
 324         unsigned char *b = skb_tail_pointer(skb);
 325         struct tcf_police *police = a->priv;
 326         struct tc_police opt = {
 327                 .index = police->tcf_index,
 328                 .action = police->tcf_action,
 329                 .mtu = police->tcfp_mtu,
 330                 .burst = PSCHED_NS2TICKS(police->tcfp_burst),
 331                 .refcnt = police->tcf_refcnt - ref,
 332                 .bindcnt = police->tcf_bindcnt - bind,
 333         };
 334
 335         if (police->rate_present)
 336                 psched_ratecfg_getrate(&opt.rate, &police->rate);
 337         if (police->peak_present)
 338                 psched_ratecfg_getrate(&opt.peakrate, &police->peak);
 339         if (nla_put(skb, TCA_POLICE_TBF, sizeof(opt), &opt))
 340                 goto nla_put_failure;
 341         if (police->tcfp_result &&
 342             nla_put_u32(skb, TCA_POLICE_RESULT, police->tcfp_result))
 343                 goto nla_put_failure;
 344         if (police->tcfp_ewma_rate &&
 345             nla_put_u32(skb, TCA_POLICE_AVRATE, police->tcfp_ewma_rate))
 346                 goto nla_put_failure;
 347         return skb->len;
 348
 349 nla_put_failure:
 350         nlmsg_trim(skb, b);
 351         return -1;
 352 }
 353
 354 MODULE_AUTHOR("Alexey Kuznetsov");
 355 MODULE_DESCRIPTION("Policing actions");
 356 MODULE_LICENSE("GPL");
 357
 358 static struct tc_action_ops act_police_ops = {
 359         .kind           =       "police",
 360         .hinfo          =       &police_hash_info,
 361         .type           =       TCA_ID_POLICE,
 362         .owner          =       THIS_MODULE,
 363         .act            =       tcf_act_police,
 364         .dump           =       tcf_act_police_dump,
 365         .cleanup        =       tcf_act_police_cleanup,
 366         .init           =       tcf_act_police_locate,
 367         .walk           =       tcf_act_police_walker
 368 };
 369
 370 static int __init
 371 police_init_module(void)
 372 {
 373         int err = tcf_hashinfo_init(&police_hash_info, POL_TAB_MASK);
 374         if (err)
 375                 return err;
 376         err = tcf_register_action(&act_police_ops);
 377         if (err)
 378                 tcf_hashinfo_destroy(&police_hash_info);
 379         return err;
 380 }
 381
 382 static void __exit
 383 police_cleanup_module(void)
 384 {
 385         tcf_hashinfo_destroy(&police_hash_info);
 386         tcf_unregister_action(&act_police_ops);
 387 }
 388
 389 module_init(police_init_module);
 390 module_exit(police_cleanup_module);