kernel/bpf/net_namespace.c

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include <linux/bpf.h>
   4 #include <linux/filter.h>
   5 #include <net/net_namespace.h>
   6
   7 /*
   8  * Functions to manage BPF programs attached to netns
   9  */
  10
  11 struct bpf_netns_link {
  12         struct bpf_link link;
  13         enum bpf_attach_type type;
  14         enum netns_bpf_attach_type netns_type;
  15
  16         /* We don't hold a ref to net in order to auto-detach the link
  17          * when netns is going away. Instead we rely on pernet
  18          * pre_exit callback to clear this pointer. Must be accessed
  19          * with netns_bpf_mutex held.
  20          */
  21         struct net *net;
  22         struct list_head node; /* node in list of links attached to net */
  23 };
  24
  25 /* Protects updates to netns_bpf */
  26 DEFINE_MUTEX(netns_bpf_mutex);
  27
  28 /* Must be called with netns_bpf_mutex held. */
  29 static void netns_bpf_run_array_detach(struct net *net,
  30                                        enum netns_bpf_attach_type type)
  31 {
  32         struct bpf_prog_array *run_array;
  33
  34         run_array = rcu_replace_pointer(net->bpf.run_array[type], NULL,
  35                                         lockdep_is_held(&netns_bpf_mutex));
  36         bpf_prog_array_free(run_array);
  37 }
  38
  39 static void bpf_netns_link_release(struct bpf_link *link)
  40 {
  41         struct bpf_netns_link *net_link =
  42                 container_of(link, struct bpf_netns_link, link);
  43         enum netns_bpf_attach_type type = net_link->netns_type;
  44         struct net *net;
  45
  46         mutex_lock(&netns_bpf_mutex);
  47
  48         /* We can race with cleanup_net, but if we see a non-NULL
  49          * struct net pointer, pre_exit has not run yet and wait for
  50          * netns_bpf_mutex.
  51          */
  52         net = net_link->net;
  53         if (!net)
  54                 goto out_unlock;
  55
  56         netns_bpf_run_array_detach(net, type);
  57         list_del(&net_link->node);
  58
  59 out_unlock:
  60         mutex_unlock(&netns_bpf_mutex);
  61 }
  62
  63 static void bpf_netns_link_dealloc(struct bpf_link *link)
  64 {
  65         struct bpf_netns_link *net_link =
  66                 container_of(link, struct bpf_netns_link, link);
  67
  68         kfree(net_link);
  69 }
  70
  71 static int bpf_netns_link_update_prog(struct bpf_link *link,
  72                                       struct bpf_prog *new_prog,
  73                                       struct bpf_prog *old_prog)
  74 {
  75         struct bpf_netns_link *net_link =
  76                 container_of(link, struct bpf_netns_link, link);
  77         enum netns_bpf_attach_type type = net_link->netns_type;
  78         struct bpf_prog_array *run_array;
  79         struct net *net;
  80         int ret = 0;
  81
  82         if (old_prog && old_prog != link->prog)
  83                 return -EPERM;
  84         if (new_prog->type != link->prog->type)
  85                 return -EINVAL;
  86
  87         mutex_lock(&netns_bpf_mutex);
  88
  89         net = net_link->net;
  90         if (!net || !check_net(net)) {
  91                 /* Link auto-detached or netns dying */
  92                 ret = -ENOLINK;
  93                 goto out_unlock;
  94         }
  95
  96         run_array = rcu_dereference_protected(net->bpf.run_array[type],
  97                                               lockdep_is_held(&netns_bpf_mutex));
  98         WRITE_ONCE(run_array->items[0].prog, new_prog);
  99
 100         old_prog = xchg(&link->prog, new_prog);
 101         bpf_prog_put(old_prog);
 102
 103 out_unlock:
 104         mutex_unlock(&netns_bpf_mutex);
 105         return ret;
 106 }
 107
 108 static int bpf_netns_link_fill_info(const struct bpf_link *link,
 109                                     struct bpf_link_info *info)
 110 {
 111         const struct bpf_netns_link *net_link =
 112                 container_of(link, struct bpf_netns_link, link);
 113         unsigned int inum = 0;
 114         struct net *net;
 115
 116         mutex_lock(&netns_bpf_mutex);
 117         net = net_link->net;
 118         if (net && check_net(net))
 119                 inum = net->ns.inum;
 120         mutex_unlock(&netns_bpf_mutex);
 121
 122         info->netns.netns_ino = inum;
 123         info->netns.attach_type = net_link->type;
 124         return 0;
 125 }
 126
 127 static void bpf_netns_link_show_fdinfo(const struct bpf_link *link,
 128                                        struct seq_file *seq)
 129 {
 130         struct bpf_link_info info = {};
 131
 132         bpf_netns_link_fill_info(link, &info);
 133         seq_printf(seq,
 134                    "netns_ino:\t%u\n"
 135                    "attach_type:\t%u\n",
 136                    info.netns.netns_ino,
 137                    info.netns.attach_type);
 138 }
 139
 140 static const struct bpf_link_ops bpf_netns_link_ops = {
 141         .release = bpf_netns_link_release,
 142         .dealloc = bpf_netns_link_dealloc,
 143         .update_prog = bpf_netns_link_update_prog,
 144         .fill_link_info = bpf_netns_link_fill_info,
 145         .show_fdinfo = bpf_netns_link_show_fdinfo,
 146 };
 147
 148 /* Must be called with netns_bpf_mutex held. */
 149 static int __netns_bpf_prog_query(const union bpf_attr *attr,
 150                                   union bpf_attr __user *uattr,
 151                                   struct net *net,
 152                                   enum netns_bpf_attach_type type)
 153 {
 154         __u32 __user *prog_ids = u64_to_user_ptr(attr->query.prog_ids);
 155         struct bpf_prog_array *run_array;
 156         u32 prog_cnt = 0, flags = 0;
 157
 158         run_array = rcu_dereference_protected(net->bpf.run_array[type],
 159                                               lockdep_is_held(&netns_bpf_mutex));
 160         if (run_array)
 161                 prog_cnt = bpf_prog_array_length(run_array);
 162
 163         if (copy_to_user(&uattr->query.attach_flags, &flags, sizeof(flags)))
 164                 return -EFAULT;
 165         if (copy_to_user(&uattr->query.prog_cnt, &prog_cnt, sizeof(prog_cnt)))
 166                 return -EFAULT;
 167         if (!attr->query.prog_cnt || !prog_ids || !prog_cnt)
 168                 return 0;
 169
 170         return bpf_prog_array_copy_to_user(run_array, prog_ids,
 171                                            attr->query.prog_cnt);
 172 }
 173
 174 int netns_bpf_prog_query(const union bpf_attr *attr,
 175                          union bpf_attr __user *uattr)
 176 {
 177         enum netns_bpf_attach_type type;
 178         struct net *net;
 179         int ret;
 180
 181         if (attr->query.query_flags)
 182                 return -EINVAL;
 183
 184         type = to_netns_bpf_attach_type(attr->query.attach_type);
 185         if (type < 0)
 186                 return -EINVAL;
 187
 188         net = get_net_ns_by_fd(attr->query.target_fd);
 189         if (IS_ERR(net))
 190                 return PTR_ERR(net);
 191
 192         mutex_lock(&netns_bpf_mutex);
 193         ret = __netns_bpf_prog_query(attr, uattr, net, type);
 194         mutex_unlock(&netns_bpf_mutex);
 195
 196         put_net(net);
 197         return ret;
 198 }
 199
 200 int netns_bpf_prog_attach(const union bpf_attr *attr, struct bpf_prog *prog)
 201 {
 202         struct bpf_prog_array *run_array;
 203         enum netns_bpf_attach_type type;
 204         struct bpf_prog *attached;
 205         struct net *net;
 206         int ret;
 207
 208         if (attr->target_fd || attr->attach_flags || attr->replace_bpf_fd)
 209                 return -EINVAL;
 210
 211         type = to_netns_bpf_attach_type(attr->attach_type);
 212         if (type < 0)
 213                 return -EINVAL;
 214
 215         net = current->nsproxy->net_ns;
 216         mutex_lock(&netns_bpf_mutex);
 217
 218         /* Attaching prog directly is not compatible with links */
 219         if (!list_empty(&net->bpf.links[type])) {
 220                 ret = -EEXIST;
 221                 goto out_unlock;
 222         }
 223
 224         switch (type) {
 225         case NETNS_BPF_FLOW_DISSECTOR:
 226                 ret = flow_dissector_bpf_prog_attach_check(net, prog);
 227                 break;
 228         default:
 229                 ret = -EINVAL;
 230                 break;
 231         }
 232         if (ret)
 233                 goto out_unlock;
 234
 235         attached = net->bpf.progs[type];
 236         if (attached == prog) {
 237                 /* The same program cannot be attached twice */
 238                 ret = -EINVAL;
 239                 goto out_unlock;
 240         }
 241
 242         run_array = rcu_dereference_protected(net->bpf.run_array[type],
 243                                               lockdep_is_held(&netns_bpf_mutex));
 244         if (run_array) {
 245                 WRITE_ONCE(run_array->items[0].prog, prog);
 246         } else {
 247                 run_array = bpf_prog_array_alloc(1, GFP_KERNEL);
 248                 if (!run_array) {
 249                         ret = -ENOMEM;
 250                         goto out_unlock;
 251                 }
 252                 run_array->items[0].prog = prog;
 253                 rcu_assign_pointer(net->bpf.run_array[type], run_array);
 254         }
 255
 256         net->bpf.progs[type] = prog;
 257         if (attached)
 258                 bpf_prog_put(attached);
 259
 260 out_unlock:
 261         mutex_unlock(&netns_bpf_mutex);
 262
 263         return ret;
 264 }
 265
 266 /* Must be called with netns_bpf_mutex held. */
 267 static int __netns_bpf_prog_detach(struct net *net,
 268                                    enum netns_bpf_attach_type type,
 269                                    struct bpf_prog *old)
 270 {
 271         struct bpf_prog *attached;
 272
 273         /* Progs attached via links cannot be detached */
 274         if (!list_empty(&net->bpf.links[type]))
 275                 return -EINVAL;
 276
 277         attached = net->bpf.progs[type];
 278         if (!attached || attached != old)
 279                 return -ENOENT;
 280         netns_bpf_run_array_detach(net, type);
 281         net->bpf.progs[type] = NULL;
 282         bpf_prog_put(attached);
 283         return 0;
 284 }
 285
 286 int netns_bpf_prog_detach(const union bpf_attr *attr, enum bpf_prog_type ptype)
 287 {
 288         enum netns_bpf_attach_type type;
 289         struct bpf_prog *prog;
 290         int ret;
 291
 292         if (attr->target_fd)
 293                 return -EINVAL;
 294
 295         type = to_netns_bpf_attach_type(attr->attach_type);
 296         if (type < 0)
 297                 return -EINVAL;
 298
 299         prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
 300         if (IS_ERR(prog))
 301                 return PTR_ERR(prog);
 302
 303         mutex_lock(&netns_bpf_mutex);
 304         ret = __netns_bpf_prog_detach(current->nsproxy->net_ns, type, prog);
 305         mutex_unlock(&netns_bpf_mutex);
 306
 307         bpf_prog_put(prog);
 308
 309         return ret;
 310 }
 311
 312 static int netns_bpf_link_attach(struct net *net, struct bpf_link *link,
 313                                  enum netns_bpf_attach_type type)
 314 {
 315         struct bpf_netns_link *net_link =
 316                 container_of(link, struct bpf_netns_link, link);
 317         struct bpf_prog_array *run_array;
 318         int err;
 319
 320         mutex_lock(&netns_bpf_mutex);
 321
 322         /* Allow attaching only one prog or link for now */
 323         if (!list_empty(&net->bpf.links[type])) {
 324                 err = -E2BIG;
 325                 goto out_unlock;
 326         }
 327         /* Links are not compatible with attaching prog directly */
 328         if (net->bpf.progs[type]) {
 329                 err = -EEXIST;
 330                 goto out_unlock;
 331         }
 332
 333         switch (type) {
 334         case NETNS_BPF_FLOW_DISSECTOR:
 335                 err = flow_dissector_bpf_prog_attach_check(net, link->prog);
 336                 break;
 337         default:
 338                 err = -EINVAL;
 339                 break;
 340         }
 341         if (err)
 342                 goto out_unlock;
 343
 344         run_array = bpf_prog_array_alloc(1, GFP_KERNEL);
 345         if (!run_array) {
 346                 err = -ENOMEM;
 347                 goto out_unlock;
 348         }
 349         run_array->items[0].prog = link->prog;
 350         rcu_assign_pointer(net->bpf.run_array[type], run_array);
 351
 352         list_add_tail(&net_link->node, &net->bpf.links[type]);
 353
 354 out_unlock:
 355         mutex_unlock(&netns_bpf_mutex);
 356         return err;
 357 }
 358
 359 int netns_bpf_link_create(const union bpf_attr *attr, struct bpf_prog *prog)
 360 {
 361         enum netns_bpf_attach_type netns_type;
 362         struct bpf_link_primer link_primer;
 363         struct bpf_netns_link *net_link;
 364         enum bpf_attach_type type;
 365         struct net *net;
 366         int err;
 367
 368         if (attr->link_create.flags)
 369                 return -EINVAL;
 370
 371         type = attr->link_create.attach_type;
 372         netns_type = to_netns_bpf_attach_type(type);
 373         if (netns_type < 0)
 374                 return -EINVAL;
 375
 376         net = get_net_ns_by_fd(attr->link_create.target_fd);
 377         if (IS_ERR(net))
 378                 return PTR_ERR(net);
 379
 380         net_link = kzalloc(sizeof(*net_link), GFP_USER);
 381         if (!net_link) {
 382                 err = -ENOMEM;
 383                 goto out_put_net;
 384         }
 385         bpf_link_init(&net_link->link, BPF_LINK_TYPE_NETNS,
 386                       &bpf_netns_link_ops, prog);
 387         net_link->net = net;
 388         net_link->type = type;
 389         net_link->netns_type = netns_type;
 390
 391         err = bpf_link_prime(&net_link->link, &link_primer);
 392         if (err) {
 393                 kfree(net_link);
 394                 goto out_put_net;
 395         }
 396
 397         err = netns_bpf_link_attach(net, &net_link->link, netns_type);
 398         if (err) {
 399                 bpf_link_cleanup(&link_primer);
 400                 goto out_put_net;
 401         }
 402
 403         put_net(net);
 404         return bpf_link_settle(&link_primer);
 405
 406 out_put_net:
 407         put_net(net);
 408         return err;
 409 }
 410
 411 static int __net_init netns_bpf_pernet_init(struct net *net)
 412 {
 413         int type;
 414
 415         for (type = 0; type < MAX_NETNS_BPF_ATTACH_TYPE; type++)
 416                 INIT_LIST_HEAD(&net->bpf.links[type]);
 417
 418         return 0;
 419 }
 420
 421 static void __net_exit netns_bpf_pernet_pre_exit(struct net *net)
 422 {
 423         enum netns_bpf_attach_type type;
 424         struct bpf_netns_link *net_link;
 425
 426         mutex_lock(&netns_bpf_mutex);
 427         for (type = 0; type < MAX_NETNS_BPF_ATTACH_TYPE; type++) {
 428                 netns_bpf_run_array_detach(net, type);
 429                 list_for_each_entry(net_link, &net->bpf.links[type], node)
 430                         net_link->net = NULL; /* auto-detach link */
 431                 if (net->bpf.progs[type])
 432                         bpf_prog_put(net->bpf.progs[type]);
 433         }
 434         mutex_unlock(&netns_bpf_mutex);
 435 }
 436
 437 static struct pernet_operations netns_bpf_pernet_ops __net_initdata = {
 438         .init = netns_bpf_pernet_init,
 439         .pre_exit = netns_bpf_pernet_pre_exit,
 440 };
 441
 442 static int __init netns_bpf_init(void)
 443 {
 444         return register_pernet_subsys(&netns_bpf_pernet_ops);
 445 }
 446
 447 subsys_initcall(netns_bpf_init);