linux/net/sched/cls_cgroup.c
<<
>>
Prefs
   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
  26static 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
  32static 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
  38static struct cgroup_subsys_state *cgrp_css_alloc(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        return &cs->css;
  46}
  47
  48static int cgrp_css_online(struct cgroup *cgrp)
  49{
  50        if (cgrp->parent)
  51                cgrp_cls_state(cgrp)->classid =
  52                        cgrp_cls_state(cgrp->parent)->classid;
  53        return 0;
  54}
  55
  56static void cgrp_css_free(struct cgroup *cgrp)
  57{
  58        kfree(cgrp_cls_state(cgrp));
  59}
  60
  61static int update_classid(const void *v, struct file *file, unsigned n)
  62{
  63        int err;
  64        struct socket *sock = sock_from_file(file, &err);
  65        if (sock)
  66                sock->sk->sk_classid = (u32)(unsigned long)v;
  67        return 0;
  68}
  69
  70static void cgrp_attach(struct cgroup *cgrp, struct cgroup_taskset *tset)
  71{
  72        struct task_struct *p;
  73        void *v;
  74
  75        cgroup_taskset_for_each(p, cgrp, tset) {
  76                task_lock(p);
  77                v = (void *)(unsigned long)task_cls_classid(p);
  78                iterate_fd(p->files, 0, update_classid, v);
  79                task_unlock(p);
  80        }
  81}
  82
  83static u64 read_classid(struct cgroup *cgrp, struct cftype *cft)
  84{
  85        return cgrp_cls_state(cgrp)->classid;
  86}
  87
  88static int write_classid(struct cgroup *cgrp, struct cftype *cft, u64 value)
  89{
  90        cgrp_cls_state(cgrp)->classid = (u32) value;
  91        return 0;
  92}
  93
  94static struct cftype ss_files[] = {
  95        {
  96                .name = "classid",
  97                .read_u64 = read_classid,
  98                .write_u64 = write_classid,
  99        },
 100        { }     /* terminate */
 101};
 102
 103struct cgroup_subsys net_cls_subsys = {
 104        .name           = "net_cls",
 105        .css_alloc      = cgrp_css_alloc,
 106        .css_online     = cgrp_css_online,
 107        .css_free       = cgrp_css_free,
 108        .attach         = cgrp_attach,
 109        .subsys_id      = net_cls_subsys_id,
 110        .base_cftypes   = ss_files,
 111        .module         = THIS_MODULE,
 112};
 113
 114struct cls_cgroup_head {
 115        u32                     handle;
 116        struct tcf_exts         exts;
 117        struct tcf_ematch_tree  ematches;
 118        struct tcf_proto        *tp;
 119        struct rcu_head         rcu;
 120};
 121
 122static int cls_cgroup_classify(struct sk_buff *skb, const struct tcf_proto *tp,
 123                               struct tcf_result *res)
 124{
 125        struct cls_cgroup_head *head = rcu_dereference_bh(tp->root);
 126        u32 classid;
 127
 128        classid = task_cls_state(current)->classid;
 129
 130        /*
 131         * Due to the nature of the classifier it is required to ignore all
 132         * packets originating from softirq context as accessing `current'
 133         * would lead to false results.
 134         *
 135         * This test assumes that all callers of dev_queue_xmit() explicitely
 136         * disable bh. Knowing this, it is possible to detect softirq based
 137         * calls by looking at the number of nested bh disable calls because
 138         * softirqs always disables bh.
 139         */
 140        if (in_serving_softirq()) {
 141                /* If there is an sk_classid we'll use that. */
 142                if (!skb->sk)
 143                        return -1;
 144                classid = skb->sk->sk_classid;
 145        }
 146
 147        if (!classid)
 148                return -1;
 149
 150        if (!tcf_em_tree_match(skb, &head->ematches, NULL))
 151                return -1;
 152
 153        res->classid = classid;
 154        res->class = 0;
 155        return tcf_exts_exec(skb, &head->exts, res);
 156}
 157
 158static unsigned long cls_cgroup_get(struct tcf_proto *tp, u32 handle)
 159{
 160        return 0UL;
 161}
 162
 163static int cls_cgroup_init(struct tcf_proto *tp)
 164{
 165        return 0;
 166}
 167
 168static const struct nla_policy cgroup_policy[TCA_CGROUP_MAX + 1] = {
 169        [TCA_CGROUP_EMATCHES]   = { .type = NLA_NESTED },
 170};
 171
 172static void cls_cgroup_destroy_rcu(struct rcu_head *root)
 173{
 174        struct cls_cgroup_head *head = container_of(root,
 175                                                    struct cls_cgroup_head,
 176                                                    rcu);
 177
 178        tcf_exts_destroy(&head->exts);
 179        tcf_em_tree_destroy(&head->ematches);
 180        kfree(head);
 181}
 182
 183static int cls_cgroup_change(struct net *net, struct sk_buff *in_skb,
 184                             struct tcf_proto *tp, unsigned long base,
 185                             u32 handle, struct nlattr **tca,
 186                             unsigned long *arg, bool ovr)
 187{
 188        struct nlattr *tb[TCA_CGROUP_MAX + 1];
 189        struct cls_cgroup_head *head = rtnl_dereference(tp->root);
 190        struct cls_cgroup_head *new;
 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 && !handle)
 199                return -EINVAL;
 200
 201        if (head && handle != head->handle)
 202                return -ENOENT;
 203
 204        new = kzalloc(sizeof(*head), GFP_KERNEL);
 205        if (!new)
 206                return -ENOBUFS;
 207
 208        tcf_exts_init(&new->exts, TCA_CGROUP_ACT, TCA_CGROUP_POLICE);
 209        new->handle = handle;
 210        new->tp = tp;
 211        err = nla_parse_nested(tb, TCA_CGROUP_MAX, tca[TCA_OPTIONS],
 212                               cgroup_policy);
 213        if (err < 0)
 214                goto errout;
 215
 216        tcf_exts_init(&e, TCA_CGROUP_ACT, TCA_CGROUP_POLICE);
 217        err = tcf_exts_validate(net, tp, tb, tca[TCA_RATE], &e, ovr);
 218        if (err < 0)
 219                goto errout;
 220
 221        err = tcf_em_tree_validate(tp, tb[TCA_CGROUP_EMATCHES], &t);
 222        if (err < 0) {
 223                tcf_exts_destroy(&e);
 224                goto errout;
 225        }
 226
 227        tcf_exts_change(tp, &new->exts, &e);
 228        tcf_em_tree_change(tp, &new->ematches, &t);
 229
 230        rcu_assign_pointer(tp->root, new);
 231        if (head)
 232                call_rcu(&head->rcu, cls_cgroup_destroy_rcu);
 233        return 0;
 234errout:
 235        kfree(new);
 236        return err;
 237}
 238
 239static bool cls_cgroup_destroy(struct tcf_proto *tp, bool force)
 240{
 241        struct cls_cgroup_head *head = rtnl_dereference(tp->root);
 242
 243        if (!force)
 244                return false;
 245
 246        /* Head can still be NULL due to cls_cgroup_init(). */
 247        if (head)
 248                call_rcu(&head->rcu, cls_cgroup_destroy_rcu);
 249        return true;
 250}
 251
 252static int cls_cgroup_delete(struct tcf_proto *tp, unsigned long arg)
 253{
 254        return -EOPNOTSUPP;
 255}
 256
 257static void cls_cgroup_walk(struct tcf_proto *tp, struct tcf_walker *arg)
 258{
 259        struct cls_cgroup_head *head = rtnl_dereference(tp->root);
 260
 261        if (arg->count < arg->skip)
 262                goto skip;
 263
 264        if (arg->fn(tp, (unsigned long) head, arg) < 0) {
 265                arg->stop = 1;
 266                return;
 267        }
 268skip:
 269        arg->count++;
 270}
 271
 272static int cls_cgroup_dump(struct net *net, struct tcf_proto *tp, unsigned long fh,
 273                           struct sk_buff *skb, struct tcmsg *t)
 274{
 275        struct cls_cgroup_head *head = rtnl_dereference(tp->root);
 276        struct nlattr *nest;
 277
 278        t->tcm_handle = head->handle;
 279
 280        nest = nla_nest_start(skb, TCA_OPTIONS);
 281        if (nest == NULL)
 282                goto nla_put_failure;
 283
 284        if (tcf_exts_dump(skb, &head->exts) < 0 ||
 285            tcf_em_tree_dump(skb, &head->ematches, TCA_CGROUP_EMATCHES) < 0)
 286                goto nla_put_failure;
 287
 288        nla_nest_end(skb, nest);
 289
 290        if (tcf_exts_dump_stats(skb, &head->exts) < 0)
 291                goto nla_put_failure;
 292
 293        return skb->len;
 294
 295nla_put_failure:
 296        nla_nest_cancel(skb, nest);
 297        return -1;
 298}
 299
 300static struct tcf_proto_ops cls_cgroup_ops __read_mostly = {
 301        .kind           =       "cgroup",
 302        .init           =       cls_cgroup_init,
 303        .change         =       cls_cgroup_change,
 304        .classify       =       cls_cgroup_classify,
 305        .destroy        =       cls_cgroup_destroy,
 306        .get            =       cls_cgroup_get,
 307        .delete         =       cls_cgroup_delete,
 308        .walk           =       cls_cgroup_walk,
 309        .dump           =       cls_cgroup_dump,
 310        .owner          =       THIS_MODULE,
 311};
 312
 313static int __init init_cgroup_cls(void)
 314{
 315        int ret;
 316
 317        ret = cgroup_load_subsys(&net_cls_subsys);
 318        if (ret)
 319                goto out;
 320
 321        ret = register_tcf_proto_ops(&cls_cgroup_ops);
 322        if (ret)
 323                cgroup_unload_subsys(&net_cls_subsys);
 324
 325out:
 326        return ret;
 327}
 328
 329static void __exit exit_cgroup_cls(void)
 330{
 331        unregister_tcf_proto_ops(&cls_cgroup_ops);
 332
 333        cgroup_unload_subsys(&net_cls_subsys);
 334}
 335
 336module_init(init_cgroup_cls);
 337module_exit(exit_cgroup_cls);
 338MODULE_LICENSE("GPL");
 339