net/sched/cls_cgroup.c

   1 /*
   2  * net/sched/cls_cgroup.c       Control Group Classifier
   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:     Thomas Graf <tgraf@suug.ch>
  10  */
  11
  12 #include <linux/module.h>
  13 #include <linux/slab.h>
  14 #include <linux/types.h>
  15 #include <linux/string.h>
  16 #include <linux/errno.h>
  17 #include <linux/skbuff.h>
  18 #include <linux/cgroup.h>
  19 #include <linux/rcupdate.h>
  20 #include <linux/fdtable.h>
  21 #include <net/rtnetlink.h>
  22 #include <net/pkt_cls.h>
  23 #include <net/sock.h>
  24 #include <net/cls_cgroup.h>
  25
  26 static inline struct cgroup_cls_state *css_cls_state(struct cgroup_subsys_state *css)
  27 {
  28         return css ? container_of(css, struct cgroup_cls_state, css) : NULL;
  29 }
  30
  31 static inline struct cgroup_cls_state *task_cls_state(struct task_struct *p)
  32 {
  33         return css_cls_state(task_css(p, net_cls_subsys_id));
  34 }
  35
  36 static struct cgroup_subsys_state *
  37 cgrp_css_alloc(struct cgroup_subsys_state *parent_css)
  38 {
  39         struct cgroup_cls_state *cs;
  40
  41         cs = kzalloc(sizeof(*cs), GFP_KERNEL);
  42         if (!cs)
  43                 return ERR_PTR(-ENOMEM);
  44         return &cs->css;
  45 }
  46
  47 static int cgrp_css_online(struct cgroup_subsys_state *css)
  48 {
  49         struct cgroup_cls_state *cs = css_cls_state(css);
  50         struct cgroup_cls_state *parent = css_cls_state(css_parent(css));
  51
  52         if (parent)
  53                 cs->classid = parent->classid;
  54         return 0;
  55 }
  56
  57 static void cgrp_css_free(struct cgroup_subsys_state *css)
  58 {
  59         kfree(css_cls_state(css));
  60 }
  61
  62 static int update_classid(const void *v, struct file *file, unsigned n)
  63 {
  64         int err;
  65         struct socket *sock = sock_from_file(file, &err);
  66         if (sock)
  67                 sock->sk->sk_classid = (u32)(unsigned long)v;
  68         return 0;
  69 }
  70
  71 static void cgrp_attach(struct cgroup_subsys_state *css,
  72                         struct cgroup_taskset *tset)
  73 {
  74         struct task_struct *p;
  75         struct cgroup_cls_state *cs = css_cls_state(css);
  76         void *v = (void *)(unsigned long)cs->classid;
  77
  78         cgroup_taskset_for_each(p, css, tset) {
  79                 task_lock(p);
  80                 iterate_fd(p->files, 0, update_classid, v);
  81                 task_unlock(p);
  82         }
  83 }
  84
  85 static u64 read_classid(struct cgroup_subsys_state *css, struct cftype *cft)
  86 {
  87         return css_cls_state(css)->classid;
  88 }
  89
  90 static int write_classid(struct cgroup_subsys_state *css, struct cftype *cft,
  91                          u64 value)
  92 {
  93         css_cls_state(css)->classid = (u32) value;
  94         return 0;
  95 }
  96
  97 static struct cftype ss_files[] = {
  98         {
  99                 .name = "classid",
 100                 .read_u64 = read_classid,
 101                 .write_u64 = write_classid,
 102         },
 103         { }     /* terminate */
 104 };
 105
 106 struct cgroup_subsys net_cls_subsys = {
 107         .name           = "net_cls",
 108         .css_alloc      = cgrp_css_alloc,
 109         .css_online     = cgrp_css_online,
 110         .css_free       = cgrp_css_free,
 111         .attach         = cgrp_attach,
 112         .subsys_id      = net_cls_subsys_id,
 113         .base_cftypes   = ss_files,
 114         .module         = THIS_MODULE,
 115 };
 116
 117 struct cls_cgroup_head {
 118         u32                     handle;
 119         struct tcf_exts         exts;
 120         struct tcf_ematch_tree  ematches;
 121 };
 122
 123 static int cls_cgroup_classify(struct sk_buff *skb, const struct tcf_proto *tp,
 124                                struct tcf_result *res)
 125 {
 126         struct cls_cgroup_head *head = tp->root;
 127         u32 classid;
 128
 129         rcu_read_lock();
 130         classid = task_cls_state(current)->classid;
 131         rcu_read_unlock();
 132
 133         /*
 134          * Due to the nature of the classifier it is required to ignore all
 135          * packets originating from softirq context as accessing `current'
 136          * would lead to false results.
 137          *
 138          * This test assumes that all callers of dev_queue_xmit() explicitely
 139          * disable bh. Knowing this, it is possible to detect softirq based
 140          * calls by looking at the number of nested bh disable calls because
 141          * softirqs always disables bh.
 142          */
 143         if (in_serving_softirq()) {
 144                 /* If there is an sk_classid we'll use that. */
 145                 if (!skb->sk)
 146                         return -1;
 147                 classid = skb->sk->sk_classid;
 148         }
 149
 150         if (!classid)
 151                 return -1;
 152
 153         if (!tcf_em_tree_match(skb, &head->ematches, NULL))
 154                 return -1;
 155
 156         res->classid = classid;
 157         res->class = 0;
 158         return tcf_exts_exec(skb, &head->exts, res);
 159 }
 160
 161 static unsigned long cls_cgroup_get(struct tcf_proto *tp, u32 handle)
 162 {
 163         return 0UL;
 164 }
 165
 166 static void cls_cgroup_put(struct tcf_proto *tp, unsigned long f)
 167 {
 168 }
 169
 170 static int cls_cgroup_init(struct tcf_proto *tp)
 171 {
 172         return 0;
 173 }
 174
 175 static const struct tcf_ext_map cgroup_ext_map = {
 176         .action = TCA_CGROUP_ACT,
 177         .police = TCA_CGROUP_POLICE,
 178 };
 179
 180 static const struct nla_policy cgroup_policy[TCA_CGROUP_MAX + 1] = {
 181         [TCA_CGROUP_EMATCHES]   = { .type = NLA_NESTED },
 182 };
 183
 184 static int cls_cgroup_change(struct net *net, struct sk_buff *in_skb,
 185                              struct tcf_proto *tp, unsigned long base,
 186                              u32 handle, struct nlattr **tca,
 187                              unsigned long *arg)
 188 {
 189         struct nlattr *tb[TCA_CGROUP_MAX + 1];
 190         struct cls_cgroup_head *head = tp->root;
 191         struct tcf_ematch_tree t;
 192         struct tcf_exts e;
 193         int err;
 194
 195         if (!tca[TCA_OPTIONS])
 196                 return -EINVAL;
 197
 198         if (head == NULL) {
 199                 if (!handle)
 200                         return -EINVAL;
 201
 202                 head = kzalloc(sizeof(*head), GFP_KERNEL);
 203                 if (head == NULL)
 204                         return -ENOBUFS;
 205
 206                 head->handle = handle;
 207
 208                 tcf_tree_lock(tp);
 209                 tp->root = head;
 210                 tcf_tree_unlock(tp);
 211         }
 212
 213         if (handle != head->handle)
 214                 return -ENOENT;
 215
 216         err = nla_parse_nested(tb, TCA_CGROUP_MAX, tca[TCA_OPTIONS],
 217                                cgroup_policy);
 218         if (err < 0)
 219                 return err;
 220
 221         err = tcf_exts_validate(net, tp, tb, tca[TCA_RATE], &e,
 222                                 &cgroup_ext_map);
 223         if (err < 0)
 224                 return err;
 225
 226         err = tcf_em_tree_validate(tp, tb[TCA_CGROUP_EMATCHES], &t);
 227         if (err < 0)
 228                 return err;
 229
 230         tcf_exts_change(tp, &head->exts, &e);
 231         tcf_em_tree_change(tp, &head->ematches, &t);
 232
 233         return 0;
 234 }
 235
 236 static void cls_cgroup_destroy(struct tcf_proto *tp)
 237 {
 238         struct cls_cgroup_head *head = tp->root;
 239
 240         if (head) {
 241                 tcf_exts_destroy(tp, &head->exts);
 242                 tcf_em_tree_destroy(tp, &head->ematches);
 243                 kfree(head);
 244         }
 245 }
 246
 247 static int cls_cgroup_delete(struct tcf_proto *tp, unsigned long arg)
 248 {
 249         return -EOPNOTSUPP;
 250 }
 251
 252 static void cls_cgroup_walk(struct tcf_proto *tp, struct tcf_walker *arg)
 253 {
 254         struct cls_cgroup_head *head = tp->root;
 255
 256         if (arg->count < arg->skip)
 257                 goto skip;
 258
 259         if (arg->fn(tp, (unsigned long) head, arg) < 0) {
 260                 arg->stop = 1;
 261                 return;
 262         }
 263 skip:
 264         arg->count++;
 265 }
 266
 267 static int cls_cgroup_dump(struct tcf_proto *tp, unsigned long fh,
 268                            struct sk_buff *skb, struct tcmsg *t)
 269 {
 270         struct cls_cgroup_head *head = tp->root;
 271         unsigned char *b = skb_tail_pointer(skb);
 272         struct nlattr *nest;
 273
 274         t->tcm_handle = head->handle;
 275
 276         nest = nla_nest_start(skb, TCA_OPTIONS);
 277         if (nest == NULL)
 278                 goto nla_put_failure;
 279
 280         if (tcf_exts_dump(skb, &head->exts, &cgroup_ext_map) < 0 ||
 281             tcf_em_tree_dump(skb, &head->ematches, TCA_CGROUP_EMATCHES) < 0)
 282                 goto nla_put_failure;
 283
 284         nla_nest_end(skb, nest);
 285
 286         if (tcf_exts_dump_stats(skb, &head->exts, &cgroup_ext_map) < 0)
 287                 goto nla_put_failure;
 288
 289         return skb->len;
 290
 291 nla_put_failure:
 292         nlmsg_trim(skb, b);
 293         return -1;
 294 }
 295
 296 static struct tcf_proto_ops cls_cgroup_ops __read_mostly = {
 297         .kind           =       "cgroup",
 298         .init           =       cls_cgroup_init,
 299         .change         =       cls_cgroup_change,
 300         .classify       =       cls_cgroup_classify,
 301         .destroy        =       cls_cgroup_destroy,
 302         .get            =       cls_cgroup_get,
 303         .put            =       cls_cgroup_put,
 304         .delete         =       cls_cgroup_delete,
 305         .walk           =       cls_cgroup_walk,
 306         .dump           =       cls_cgroup_dump,
 307         .owner          =       THIS_MODULE,
 308 };
 309
 310 static int __init init_cgroup_cls(void)
 311 {
 312         int ret;
 313
 314         ret = cgroup_load_subsys(&net_cls_subsys);
 315         if (ret)
 316                 goto out;
 317
 318         ret = register_tcf_proto_ops(&cls_cgroup_ops);
 319         if (ret)
 320                 cgroup_unload_subsys(&net_cls_subsys);
 321
 322 out:
 323         return ret;
 324 }
 325
 326 static void __exit exit_cgroup_cls(void)
 327 {
 328         unregister_tcf_proto_ops(&cls_cgroup_ops);
 329
 330         cgroup_unload_subsys(&net_cls_subsys);
 331 }
 332
 333 module_init(init_cgroup_cls);
 334 module_exit(exit_cgroup_cls);
 335 MODULE_LICENSE("GPL");