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 *cgrp_cls_state(struct cgroup *cgrp)
  27 {
  28         return container_of(cgroup_subsys_state(cgrp, net_cls_subsys_id),
  29                             struct cgroup_cls_state, css);
  30 }
  31
  32 static inline struct cgroup_cls_state *task_cls_state(struct task_struct *p)
  33 {
  34         return container_of(task_subsys_state(p, net_cls_subsys_id),
  35                             struct cgroup_cls_state, css);
  36 }
  37
  38 static struct cgroup_subsys_state *cgrp_create(struct cgroup *cgrp)
  39 {
  40         struct cgroup_cls_state *cs;
  41
  42         cs = kzalloc(sizeof(*cs), GFP_KERNEL);
  43         if (!cs)
  44                 return ERR_PTR(-ENOMEM);
  45
  46         if (cgrp->parent)
  47                 cs->classid = cgrp_cls_state(cgrp->parent)->classid;
  48
  49         return &cs->css;
  50 }
  51
  52 static void cgrp_destroy(struct cgroup *cgrp)
  53 {
  54         kfree(cgrp_cls_state(cgrp));
  55 }
  56
  57 static int update_classid(const void *v, struct file *file, unsigned n)
  58 {
  59         int err;
  60         struct socket *sock = sock_from_file(file, &err);
  61         if (sock)
  62                 sock->sk->sk_classid = (u32)(unsigned long)v;
  63         return 0;
  64 }
  65
  66 static void cgrp_attach(struct cgroup *cgrp, struct cgroup_taskset *tset)
  67 {
  68         struct task_struct *p;
  69         void *v;
  70
  71         cgroup_taskset_for_each(p, cgrp, tset) {
  72                 task_lock(p);
  73                 v = (void *)(unsigned long)task_cls_classid(p);
  74                 iterate_fd(p->files, 0, update_classid, v);
  75                 task_unlock(p);
  76         }
  77 }
  78
  79 static u64 read_classid(struct cgroup *cgrp, struct cftype *cft)
  80 {
  81         return cgrp_cls_state(cgrp)->classid;
  82 }
  83
  84 static int write_classid(struct cgroup *cgrp, struct cftype *cft, u64 value)
  85 {
  86         cgrp_cls_state(cgrp)->classid = (u32) value;
  87         return 0;
  88 }
  89
  90 static struct cftype ss_files[] = {
  91         {
  92                 .name = "classid",
  93                 .read_u64 = read_classid,
  94                 .write_u64 = write_classid,
  95         },
  96         { }     /* terminate */
  97 };
  98
  99 struct cgroup_subsys net_cls_subsys = {
 100         .name           = "net_cls",
 101         .create         = cgrp_create,
 102         .destroy        = cgrp_destroy,
 103         .attach         = cgrp_attach,
 104         .subsys_id      = net_cls_subsys_id,
 105         .base_cftypes   = ss_files,
 106         .module         = THIS_MODULE,
 107
 108         /*
 109          * While net_cls cgroup has the rudimentary hierarchy support of
 110          * inheriting the parent's classid on cgroup creation, it doesn't
 111          * properly propagates config changes in ancestors to their
 112          * descendents.  A child should follow the parent's configuration
 113          * but be allowed to override it.  Fix it and remove the following.
 114          */
 115         .broken_hierarchy = true,
 116 };
 117
 118 struct cls_cgroup_head {
 119         u32                     handle;
 120         struct tcf_exts         exts;
 121         struct tcf_ematch_tree  ematches;
 122 };
 123
 124 static int cls_cgroup_classify(struct sk_buff *skb, const struct tcf_proto *tp,
 125                                struct tcf_result *res)
 126 {
 127         struct cls_cgroup_head *head = tp->root;
 128         u32 classid;
 129
 130         rcu_read_lock();
 131         classid = task_cls_state(current)->classid;
 132         rcu_read_unlock();
 133
 134         /*
 135          * Due to the nature of the classifier it is required to ignore all
 136          * packets originating from softirq context as accessing `current'
 137          * would lead to false results.
 138          *
 139          * This test assumes that all callers of dev_queue_xmit() explicitely
 140          * disable bh. Knowing this, it is possible to detect softirq based
 141          * calls by looking at the number of nested bh disable calls because
 142          * softirqs always disables bh.
 143          */
 144         if (in_serving_softirq()) {
 145                 /* If there is an sk_classid we'll use that. */
 146                 if (!skb->sk)
 147                         return -1;
 148                 classid = skb->sk->sk_classid;
 149         }
 150
 151         if (!classid)
 152                 return -1;
 153
 154         if (!tcf_em_tree_match(skb, &head->ematches, NULL))
 155                 return -1;
 156
 157         res->classid = classid;
 158         res->class = 0;
 159         return tcf_exts_exec(skb, &head->exts, res);
 160 }
 161
 162 static unsigned long cls_cgroup_get(struct tcf_proto *tp, u32 handle)
 163 {
 164         return 0UL;
 165 }
 166
 167 static void cls_cgroup_put(struct tcf_proto *tp, unsigned long f)
 168 {
 169 }
 170
 171 static int cls_cgroup_init(struct tcf_proto *tp)
 172 {
 173         return 0;
 174 }
 175
 176 static const struct tcf_ext_map cgroup_ext_map = {
 177         .action = TCA_CGROUP_ACT,
 178         .police = TCA_CGROUP_POLICE,
 179 };
 180
 181 static const struct nla_policy cgroup_policy[TCA_CGROUP_MAX + 1] = {
 182         [TCA_CGROUP_EMATCHES]   = { .type = NLA_NESTED },
 183 };
 184
 185 static int cls_cgroup_change(struct sk_buff *in_skb,
 186                              struct tcf_proto *tp, unsigned long base,
 187                              u32 handle, struct nlattr **tca,
 188                              unsigned long *arg)
 189 {
 190         struct nlattr *tb[TCA_CGROUP_MAX + 1];
 191         struct cls_cgroup_head *head = tp->root;
 192         struct tcf_ematch_tree t;
 193         struct tcf_exts e;
 194         int err;
 195
 196         if (!tca[TCA_OPTIONS])
 197                 return -EINVAL;
 198
 199         if (head == NULL) {
 200                 if (!handle)
 201                         return -EINVAL;
 202
 203                 head = kzalloc(sizeof(*head), GFP_KERNEL);
 204                 if (head == NULL)
 205                         return -ENOBUFS;
 206
 207                 head->handle = handle;
 208
 209                 tcf_tree_lock(tp);
 210                 tp->root = head;
 211                 tcf_tree_unlock(tp);
 212         }
 213
 214         if (handle != head->handle)
 215                 return -ENOENT;
 216
 217         err = nla_parse_nested(tb, TCA_CGROUP_MAX, tca[TCA_OPTIONS],
 218                                cgroup_policy);
 219         if (err < 0)
 220                 return err;
 221
 222         err = tcf_exts_validate(tp, tb, tca[TCA_RATE], &e, &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");