1 // SPDX-License-Identifier: GPL-2.0
4 #include <linux/filter.h>
5 #include <net/net_namespace.h>
8 * Functions to manage BPF programs attached to netns
11 struct bpf_netns_link {
13 enum bpf_attach_type type;
14 enum netns_bpf_attach_type netns_type;
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.
22 struct list_head node; /* node in list of links attached to net */
25 /* Protects updates to netns_bpf */
26 DEFINE_MUTEX(netns_bpf_mutex);
28 static void netns_bpf_attach_type_unneed(enum netns_bpf_attach_type type)
32 case NETNS_BPF_SK_LOOKUP:
33 static_branch_dec(&bpf_sk_lookup_enabled);
41 static void netns_bpf_attach_type_need(enum netns_bpf_attach_type type)
45 case NETNS_BPF_SK_LOOKUP:
46 static_branch_inc(&bpf_sk_lookup_enabled);
54 /* Must be called with netns_bpf_mutex held. */
55 static void netns_bpf_run_array_detach(struct net *net,
56 enum netns_bpf_attach_type type)
58 struct bpf_prog_array *run_array;
60 run_array = rcu_replace_pointer(net->bpf.run_array[type], NULL,
61 lockdep_is_held(&netns_bpf_mutex));
62 bpf_prog_array_free(run_array);
65 static int link_index(struct net *net, enum netns_bpf_attach_type type,
66 struct bpf_netns_link *link)
68 struct bpf_netns_link *pos;
71 list_for_each_entry(pos, &net->bpf.links[type], node) {
79 static int link_count(struct net *net, enum netns_bpf_attach_type type)
81 struct list_head *pos;
84 list_for_each(pos, &net->bpf.links[type])
89 static void fill_prog_array(struct net *net, enum netns_bpf_attach_type type,
90 struct bpf_prog_array *prog_array)
92 struct bpf_netns_link *pos;
95 list_for_each_entry(pos, &net->bpf.links[type], node) {
96 prog_array->items[i].prog = pos->link.prog;
101 static void bpf_netns_link_release(struct bpf_link *link)
103 struct bpf_netns_link *net_link =
104 container_of(link, struct bpf_netns_link, link);
105 enum netns_bpf_attach_type type = net_link->netns_type;
106 struct bpf_prog_array *old_array, *new_array;
110 mutex_lock(&netns_bpf_mutex);
112 /* We can race with cleanup_net, but if we see a non-NULL
113 * struct net pointer, pre_exit has not run yet and wait for
120 /* Mark attach point as unused */
121 netns_bpf_attach_type_unneed(type);
123 /* Remember link position in case of safe delete */
124 idx = link_index(net, type, net_link);
125 list_del(&net_link->node);
127 cnt = link_count(net, type);
129 netns_bpf_run_array_detach(net, type);
133 old_array = rcu_dereference_protected(net->bpf.run_array[type],
134 lockdep_is_held(&netns_bpf_mutex));
135 new_array = bpf_prog_array_alloc(cnt, GFP_KERNEL);
137 WARN_ON(bpf_prog_array_delete_safe_at(old_array, idx));
140 fill_prog_array(net, type, new_array);
141 rcu_assign_pointer(net->bpf.run_array[type], new_array);
142 bpf_prog_array_free(old_array);
145 net_link->net = NULL;
146 mutex_unlock(&netns_bpf_mutex);
149 static int bpf_netns_link_detach(struct bpf_link *link)
151 bpf_netns_link_release(link);
155 static void bpf_netns_link_dealloc(struct bpf_link *link)
157 struct bpf_netns_link *net_link =
158 container_of(link, struct bpf_netns_link, link);
163 static int bpf_netns_link_update_prog(struct bpf_link *link,
164 struct bpf_prog *new_prog,
165 struct bpf_prog *old_prog)
167 struct bpf_netns_link *net_link =
168 container_of(link, struct bpf_netns_link, link);
169 enum netns_bpf_attach_type type = net_link->netns_type;
170 struct bpf_prog_array *run_array;
174 if (old_prog && old_prog != link->prog)
176 if (new_prog->type != link->prog->type)
179 mutex_lock(&netns_bpf_mutex);
182 if (!net || !check_net(net)) {
183 /* Link auto-detached or netns dying */
188 run_array = rcu_dereference_protected(net->bpf.run_array[type],
189 lockdep_is_held(&netns_bpf_mutex));
190 idx = link_index(net, type, net_link);
191 ret = bpf_prog_array_update_at(run_array, idx, new_prog);
195 old_prog = xchg(&link->prog, new_prog);
196 bpf_prog_put(old_prog);
199 mutex_unlock(&netns_bpf_mutex);
203 static int bpf_netns_link_fill_info(const struct bpf_link *link,
204 struct bpf_link_info *info)
206 const struct bpf_netns_link *net_link =
207 container_of(link, struct bpf_netns_link, link);
208 unsigned int inum = 0;
211 mutex_lock(&netns_bpf_mutex);
213 if (net && check_net(net))
215 mutex_unlock(&netns_bpf_mutex);
217 info->netns.netns_ino = inum;
218 info->netns.attach_type = net_link->type;
222 static void bpf_netns_link_show_fdinfo(const struct bpf_link *link,
223 struct seq_file *seq)
225 struct bpf_link_info info = {};
227 bpf_netns_link_fill_info(link, &info);
230 "attach_type:\t%u\n",
231 info.netns.netns_ino,
232 info.netns.attach_type);
235 static const struct bpf_link_ops bpf_netns_link_ops = {
236 .release = bpf_netns_link_release,
237 .dealloc = bpf_netns_link_dealloc,
238 .detach = bpf_netns_link_detach,
239 .update_prog = bpf_netns_link_update_prog,
240 .fill_link_info = bpf_netns_link_fill_info,
241 .show_fdinfo = bpf_netns_link_show_fdinfo,
244 /* Must be called with netns_bpf_mutex held. */
245 static int __netns_bpf_prog_query(const union bpf_attr *attr,
246 union bpf_attr __user *uattr,
248 enum netns_bpf_attach_type type)
250 __u32 __user *prog_ids = u64_to_user_ptr(attr->query.prog_ids);
251 struct bpf_prog_array *run_array;
252 u32 prog_cnt = 0, flags = 0;
254 run_array = rcu_dereference_protected(net->bpf.run_array[type],
255 lockdep_is_held(&netns_bpf_mutex));
257 prog_cnt = bpf_prog_array_length(run_array);
259 if (copy_to_user(&uattr->query.attach_flags, &flags, sizeof(flags)))
261 if (copy_to_user(&uattr->query.prog_cnt, &prog_cnt, sizeof(prog_cnt)))
263 if (!attr->query.prog_cnt || !prog_ids || !prog_cnt)
266 return bpf_prog_array_copy_to_user(run_array, prog_ids,
267 attr->query.prog_cnt);
270 int netns_bpf_prog_query(const union bpf_attr *attr,
271 union bpf_attr __user *uattr)
273 enum netns_bpf_attach_type type;
277 if (attr->query.query_flags)
280 type = to_netns_bpf_attach_type(attr->query.attach_type);
284 net = get_net_ns_by_fd(attr->query.target_fd);
288 mutex_lock(&netns_bpf_mutex);
289 ret = __netns_bpf_prog_query(attr, uattr, net, type);
290 mutex_unlock(&netns_bpf_mutex);
296 int netns_bpf_prog_attach(const union bpf_attr *attr, struct bpf_prog *prog)
298 struct bpf_prog_array *run_array;
299 enum netns_bpf_attach_type type;
300 struct bpf_prog *attached;
304 if (attr->target_fd || attr->attach_flags || attr->replace_bpf_fd)
307 type = to_netns_bpf_attach_type(attr->attach_type);
311 net = current->nsproxy->net_ns;
312 mutex_lock(&netns_bpf_mutex);
314 /* Attaching prog directly is not compatible with links */
315 if (!list_empty(&net->bpf.links[type])) {
321 case NETNS_BPF_FLOW_DISSECTOR:
322 ret = flow_dissector_bpf_prog_attach_check(net, prog);
331 attached = net->bpf.progs[type];
332 if (attached == prog) {
333 /* The same program cannot be attached twice */
338 run_array = rcu_dereference_protected(net->bpf.run_array[type],
339 lockdep_is_held(&netns_bpf_mutex));
341 WRITE_ONCE(run_array->items[0].prog, prog);
343 run_array = bpf_prog_array_alloc(1, GFP_KERNEL);
348 run_array->items[0].prog = prog;
349 rcu_assign_pointer(net->bpf.run_array[type], run_array);
352 net->bpf.progs[type] = prog;
354 bpf_prog_put(attached);
357 mutex_unlock(&netns_bpf_mutex);
362 /* Must be called with netns_bpf_mutex held. */
363 static int __netns_bpf_prog_detach(struct net *net,
364 enum netns_bpf_attach_type type,
365 struct bpf_prog *old)
367 struct bpf_prog *attached;
369 /* Progs attached via links cannot be detached */
370 if (!list_empty(&net->bpf.links[type]))
373 attached = net->bpf.progs[type];
374 if (!attached || attached != old)
376 netns_bpf_run_array_detach(net, type);
377 net->bpf.progs[type] = NULL;
378 bpf_prog_put(attached);
382 int netns_bpf_prog_detach(const union bpf_attr *attr, enum bpf_prog_type ptype)
384 enum netns_bpf_attach_type type;
385 struct bpf_prog *prog;
391 type = to_netns_bpf_attach_type(attr->attach_type);
395 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
397 return PTR_ERR(prog);
399 mutex_lock(&netns_bpf_mutex);
400 ret = __netns_bpf_prog_detach(current->nsproxy->net_ns, type, prog);
401 mutex_unlock(&netns_bpf_mutex);
408 static int netns_bpf_max_progs(enum netns_bpf_attach_type type)
411 case NETNS_BPF_FLOW_DISSECTOR:
413 case NETNS_BPF_SK_LOOKUP:
420 static int netns_bpf_link_attach(struct net *net, struct bpf_link *link,
421 enum netns_bpf_attach_type type)
423 struct bpf_netns_link *net_link =
424 container_of(link, struct bpf_netns_link, link);
425 struct bpf_prog_array *run_array;
428 mutex_lock(&netns_bpf_mutex);
430 cnt = link_count(net, type);
431 if (cnt >= netns_bpf_max_progs(type)) {
435 /* Links are not compatible with attaching prog directly */
436 if (net->bpf.progs[type]) {
442 case NETNS_BPF_FLOW_DISSECTOR:
443 err = flow_dissector_bpf_prog_attach_check(net, link->prog);
445 case NETNS_BPF_SK_LOOKUP:
446 err = 0; /* nothing to check */
455 run_array = bpf_prog_array_alloc(cnt + 1, GFP_KERNEL);
461 list_add_tail(&net_link->node, &net->bpf.links[type]);
463 fill_prog_array(net, type, run_array);
464 run_array = rcu_replace_pointer(net->bpf.run_array[type], run_array,
465 lockdep_is_held(&netns_bpf_mutex));
466 bpf_prog_array_free(run_array);
468 /* Mark attach point as used */
469 netns_bpf_attach_type_need(type);
472 mutex_unlock(&netns_bpf_mutex);
476 int netns_bpf_link_create(const union bpf_attr *attr, struct bpf_prog *prog)
478 enum netns_bpf_attach_type netns_type;
479 struct bpf_link_primer link_primer;
480 struct bpf_netns_link *net_link;
481 enum bpf_attach_type type;
485 if (attr->link_create.flags)
488 type = attr->link_create.attach_type;
489 netns_type = to_netns_bpf_attach_type(type);
493 net = get_net_ns_by_fd(attr->link_create.target_fd);
497 net_link = kzalloc(sizeof(*net_link), GFP_USER);
502 bpf_link_init(&net_link->link, BPF_LINK_TYPE_NETNS,
503 &bpf_netns_link_ops, prog);
505 net_link->type = type;
506 net_link->netns_type = netns_type;
508 err = bpf_link_prime(&net_link->link, &link_primer);
514 err = netns_bpf_link_attach(net, &net_link->link, netns_type);
516 bpf_link_cleanup(&link_primer);
521 return bpf_link_settle(&link_primer);
528 static int __net_init netns_bpf_pernet_init(struct net *net)
532 for (type = 0; type < MAX_NETNS_BPF_ATTACH_TYPE; type++)
533 INIT_LIST_HEAD(&net->bpf.links[type]);
538 static void __net_exit netns_bpf_pernet_pre_exit(struct net *net)
540 enum netns_bpf_attach_type type;
541 struct bpf_netns_link *net_link;
543 mutex_lock(&netns_bpf_mutex);
544 for (type = 0; type < MAX_NETNS_BPF_ATTACH_TYPE; type++) {
545 netns_bpf_run_array_detach(net, type);
546 list_for_each_entry(net_link, &net->bpf.links[type], node) {
547 net_link->net = NULL; /* auto-detach link */
548 netns_bpf_attach_type_unneed(type);
550 if (net->bpf.progs[type])
551 bpf_prog_put(net->bpf.progs[type]);
553 mutex_unlock(&netns_bpf_mutex);
556 static struct pernet_operations netns_bpf_pernet_ops __net_initdata = {
557 .init = netns_bpf_pernet_init,
558 .pre_exit = netns_bpf_pernet_pre_exit,
561 static int __init netns_bpf_init(void)
563 return register_pernet_subsys(&netns_bpf_pernet_ops);
566 subsys_initcall(netns_bpf_init);