Loading...
1/*
2 * Copyright (C) 2017 Netronome Systems, Inc.
3 *
4 * This software is licensed under the GNU General License Version 2,
5 * June 1991 as shown in the file COPYING in the top-level directory of this
6 * source tree.
7 *
8 * THE COPYRIGHT HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS"
9 * WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING,
10 * BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
11 * FOR A PARTICULAR PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE
12 * OF THE PROGRAM IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME
13 * THE COST OF ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
14 */
15
16#include <linux/bpf.h>
17#include <linux/bpf_verifier.h>
18#include <linux/bug.h>
19#include <linux/kdev_t.h>
20#include <linux/list.h>
21#include <linux/netdevice.h>
22#include <linux/printk.h>
23#include <linux/proc_ns.h>
24#include <linux/rtnetlink.h>
25#include <linux/rwsem.h>
26
27/* Protects bpf_prog_offload_devs, bpf_map_offload_devs and offload members
28 * of all progs.
29 * RTNL lock cannot be taken when holding this lock.
30 */
31static DECLARE_RWSEM(bpf_devs_lock);
32static LIST_HEAD(bpf_prog_offload_devs);
33static LIST_HEAD(bpf_map_offload_devs);
34
35static int bpf_dev_offload_check(struct net_device *netdev)
36{
37 if (!netdev)
38 return -EINVAL;
39 if (!netdev->netdev_ops->ndo_bpf)
40 return -EOPNOTSUPP;
41 return 0;
42}
43
44int bpf_prog_offload_init(struct bpf_prog *prog, union bpf_attr *attr)
45{
46 struct bpf_prog_offload *offload;
47 int err;
48
49 if (attr->prog_type != BPF_PROG_TYPE_SCHED_CLS &&
50 attr->prog_type != BPF_PROG_TYPE_XDP)
51 return -EINVAL;
52
53 if (attr->prog_flags)
54 return -EINVAL;
55
56 offload = kzalloc(sizeof(*offload), GFP_USER);
57 if (!offload)
58 return -ENOMEM;
59
60 offload->prog = prog;
61
62 offload->netdev = dev_get_by_index(current->nsproxy->net_ns,
63 attr->prog_ifindex);
64 err = bpf_dev_offload_check(offload->netdev);
65 if (err)
66 goto err_maybe_put;
67
68 down_write(&bpf_devs_lock);
69 if (offload->netdev->reg_state != NETREG_REGISTERED) {
70 err = -EINVAL;
71 goto err_unlock;
72 }
73 prog->aux->offload = offload;
74 list_add_tail(&offload->offloads, &bpf_prog_offload_devs);
75 dev_put(offload->netdev);
76 up_write(&bpf_devs_lock);
77
78 return 0;
79err_unlock:
80 up_write(&bpf_devs_lock);
81err_maybe_put:
82 if (offload->netdev)
83 dev_put(offload->netdev);
84 kfree(offload);
85 return err;
86}
87
88static int __bpf_offload_ndo(struct bpf_prog *prog, enum bpf_netdev_command cmd,
89 struct netdev_bpf *data)
90{
91 struct bpf_prog_offload *offload = prog->aux->offload;
92 struct net_device *netdev;
93
94 ASSERT_RTNL();
95
96 if (!offload)
97 return -ENODEV;
98 netdev = offload->netdev;
99
100 data->command = cmd;
101
102 return netdev->netdev_ops->ndo_bpf(netdev, data);
103}
104
105int bpf_prog_offload_verifier_prep(struct bpf_verifier_env *env)
106{
107 struct netdev_bpf data = {};
108 int err;
109
110 data.verifier.prog = env->prog;
111
112 rtnl_lock();
113 err = __bpf_offload_ndo(env->prog, BPF_OFFLOAD_VERIFIER_PREP, &data);
114 if (err)
115 goto exit_unlock;
116
117 env->prog->aux->offload->dev_ops = data.verifier.ops;
118 env->prog->aux->offload->dev_state = true;
119exit_unlock:
120 rtnl_unlock();
121 return err;
122}
123
124int bpf_prog_offload_verify_insn(struct bpf_verifier_env *env,
125 int insn_idx, int prev_insn_idx)
126{
127 struct bpf_prog_offload *offload;
128 int ret = -ENODEV;
129
130 down_read(&bpf_devs_lock);
131 offload = env->prog->aux->offload;
132 if (offload)
133 ret = offload->dev_ops->insn_hook(env, insn_idx, prev_insn_idx);
134 up_read(&bpf_devs_lock);
135
136 return ret;
137}
138
139static void __bpf_prog_offload_destroy(struct bpf_prog *prog)
140{
141 struct bpf_prog_offload *offload = prog->aux->offload;
142 struct netdev_bpf data = {};
143
144 data.offload.prog = prog;
145
146 if (offload->dev_state)
147 WARN_ON(__bpf_offload_ndo(prog, BPF_OFFLOAD_DESTROY, &data));
148
149 /* Make sure BPF_PROG_GET_NEXT_ID can't find this dead program */
150 bpf_prog_free_id(prog, true);
151
152 list_del_init(&offload->offloads);
153 kfree(offload);
154 prog->aux->offload = NULL;
155}
156
157void bpf_prog_offload_destroy(struct bpf_prog *prog)
158{
159 rtnl_lock();
160 down_write(&bpf_devs_lock);
161 if (prog->aux->offload)
162 __bpf_prog_offload_destroy(prog);
163 up_write(&bpf_devs_lock);
164 rtnl_unlock();
165}
166
167static int bpf_prog_offload_translate(struct bpf_prog *prog)
168{
169 struct netdev_bpf data = {};
170 int ret;
171
172 data.offload.prog = prog;
173
174 rtnl_lock();
175 ret = __bpf_offload_ndo(prog, BPF_OFFLOAD_TRANSLATE, &data);
176 rtnl_unlock();
177
178 return ret;
179}
180
181static unsigned int bpf_prog_warn_on_exec(const void *ctx,
182 const struct bpf_insn *insn)
183{
184 WARN(1, "attempt to execute device eBPF program on the host!");
185 return 0;
186}
187
188int bpf_prog_offload_compile(struct bpf_prog *prog)
189{
190 prog->bpf_func = bpf_prog_warn_on_exec;
191
192 return bpf_prog_offload_translate(prog);
193}
194
195struct ns_get_path_bpf_prog_args {
196 struct bpf_prog *prog;
197 struct bpf_prog_info *info;
198};
199
200static struct ns_common *bpf_prog_offload_info_fill_ns(void *private_data)
201{
202 struct ns_get_path_bpf_prog_args *args = private_data;
203 struct bpf_prog_aux *aux = args->prog->aux;
204 struct ns_common *ns;
205 struct net *net;
206
207 rtnl_lock();
208 down_read(&bpf_devs_lock);
209
210 if (aux->offload) {
211 args->info->ifindex = aux->offload->netdev->ifindex;
212 net = dev_net(aux->offload->netdev);
213 get_net(net);
214 ns = &net->ns;
215 } else {
216 args->info->ifindex = 0;
217 ns = NULL;
218 }
219
220 up_read(&bpf_devs_lock);
221 rtnl_unlock();
222
223 return ns;
224}
225
226int bpf_prog_offload_info_fill(struct bpf_prog_info *info,
227 struct bpf_prog *prog)
228{
229 struct ns_get_path_bpf_prog_args args = {
230 .prog = prog,
231 .info = info,
232 };
233 struct bpf_prog_aux *aux = prog->aux;
234 struct inode *ns_inode;
235 struct path ns_path;
236 char __user *uinsns;
237 void *res;
238 u32 ulen;
239
240 res = ns_get_path_cb(&ns_path, bpf_prog_offload_info_fill_ns, &args);
241 if (IS_ERR(res)) {
242 if (!info->ifindex)
243 return -ENODEV;
244 return PTR_ERR(res);
245 }
246
247 down_read(&bpf_devs_lock);
248
249 if (!aux->offload) {
250 up_read(&bpf_devs_lock);
251 return -ENODEV;
252 }
253
254 ulen = info->jited_prog_len;
255 info->jited_prog_len = aux->offload->jited_len;
256 if (info->jited_prog_len & ulen) {
257 uinsns = u64_to_user_ptr(info->jited_prog_insns);
258 ulen = min_t(u32, info->jited_prog_len, ulen);
259 if (copy_to_user(uinsns, aux->offload->jited_image, ulen)) {
260 up_read(&bpf_devs_lock);
261 return -EFAULT;
262 }
263 }
264
265 up_read(&bpf_devs_lock);
266
267 ns_inode = ns_path.dentry->d_inode;
268 info->netns_dev = new_encode_dev(ns_inode->i_sb->s_dev);
269 info->netns_ino = ns_inode->i_ino;
270 path_put(&ns_path);
271
272 return 0;
273}
274
275const struct bpf_prog_ops bpf_offload_prog_ops = {
276};
277
278static int bpf_map_offload_ndo(struct bpf_offloaded_map *offmap,
279 enum bpf_netdev_command cmd)
280{
281 struct netdev_bpf data = {};
282 struct net_device *netdev;
283
284 ASSERT_RTNL();
285
286 data.command = cmd;
287 data.offmap = offmap;
288 /* Caller must make sure netdev is valid */
289 netdev = offmap->netdev;
290
291 return netdev->netdev_ops->ndo_bpf(netdev, &data);
292}
293
294struct bpf_map *bpf_map_offload_map_alloc(union bpf_attr *attr)
295{
296 struct net *net = current->nsproxy->net_ns;
297 struct bpf_offloaded_map *offmap;
298 int err;
299
300 if (!capable(CAP_SYS_ADMIN))
301 return ERR_PTR(-EPERM);
302 if (attr->map_type != BPF_MAP_TYPE_ARRAY &&
303 attr->map_type != BPF_MAP_TYPE_HASH)
304 return ERR_PTR(-EINVAL);
305
306 offmap = kzalloc(sizeof(*offmap), GFP_USER);
307 if (!offmap)
308 return ERR_PTR(-ENOMEM);
309
310 bpf_map_init_from_attr(&offmap->map, attr);
311
312 rtnl_lock();
313 down_write(&bpf_devs_lock);
314 offmap->netdev = __dev_get_by_index(net, attr->map_ifindex);
315 err = bpf_dev_offload_check(offmap->netdev);
316 if (err)
317 goto err_unlock;
318
319 err = bpf_map_offload_ndo(offmap, BPF_OFFLOAD_MAP_ALLOC);
320 if (err)
321 goto err_unlock;
322
323 list_add_tail(&offmap->offloads, &bpf_map_offload_devs);
324 up_write(&bpf_devs_lock);
325 rtnl_unlock();
326
327 return &offmap->map;
328
329err_unlock:
330 up_write(&bpf_devs_lock);
331 rtnl_unlock();
332 kfree(offmap);
333 return ERR_PTR(err);
334}
335
336static void __bpf_map_offload_destroy(struct bpf_offloaded_map *offmap)
337{
338 WARN_ON(bpf_map_offload_ndo(offmap, BPF_OFFLOAD_MAP_FREE));
339 /* Make sure BPF_MAP_GET_NEXT_ID can't find this dead map */
340 bpf_map_free_id(&offmap->map, true);
341 list_del_init(&offmap->offloads);
342 offmap->netdev = NULL;
343}
344
345void bpf_map_offload_map_free(struct bpf_map *map)
346{
347 struct bpf_offloaded_map *offmap = map_to_offmap(map);
348
349 rtnl_lock();
350 down_write(&bpf_devs_lock);
351 if (offmap->netdev)
352 __bpf_map_offload_destroy(offmap);
353 up_write(&bpf_devs_lock);
354 rtnl_unlock();
355
356 kfree(offmap);
357}
358
359int bpf_map_offload_lookup_elem(struct bpf_map *map, void *key, void *value)
360{
361 struct bpf_offloaded_map *offmap = map_to_offmap(map);
362 int ret = -ENODEV;
363
364 down_read(&bpf_devs_lock);
365 if (offmap->netdev)
366 ret = offmap->dev_ops->map_lookup_elem(offmap, key, value);
367 up_read(&bpf_devs_lock);
368
369 return ret;
370}
371
372int bpf_map_offload_update_elem(struct bpf_map *map,
373 void *key, void *value, u64 flags)
374{
375 struct bpf_offloaded_map *offmap = map_to_offmap(map);
376 int ret = -ENODEV;
377
378 if (unlikely(flags > BPF_EXIST))
379 return -EINVAL;
380
381 down_read(&bpf_devs_lock);
382 if (offmap->netdev)
383 ret = offmap->dev_ops->map_update_elem(offmap, key, value,
384 flags);
385 up_read(&bpf_devs_lock);
386
387 return ret;
388}
389
390int bpf_map_offload_delete_elem(struct bpf_map *map, void *key)
391{
392 struct bpf_offloaded_map *offmap = map_to_offmap(map);
393 int ret = -ENODEV;
394
395 down_read(&bpf_devs_lock);
396 if (offmap->netdev)
397 ret = offmap->dev_ops->map_delete_elem(offmap, key);
398 up_read(&bpf_devs_lock);
399
400 return ret;
401}
402
403int bpf_map_offload_get_next_key(struct bpf_map *map, void *key, void *next_key)
404{
405 struct bpf_offloaded_map *offmap = map_to_offmap(map);
406 int ret = -ENODEV;
407
408 down_read(&bpf_devs_lock);
409 if (offmap->netdev)
410 ret = offmap->dev_ops->map_get_next_key(offmap, key, next_key);
411 up_read(&bpf_devs_lock);
412
413 return ret;
414}
415
416struct ns_get_path_bpf_map_args {
417 struct bpf_offloaded_map *offmap;
418 struct bpf_map_info *info;
419};
420
421static struct ns_common *bpf_map_offload_info_fill_ns(void *private_data)
422{
423 struct ns_get_path_bpf_map_args *args = private_data;
424 struct ns_common *ns;
425 struct net *net;
426
427 rtnl_lock();
428 down_read(&bpf_devs_lock);
429
430 if (args->offmap->netdev) {
431 args->info->ifindex = args->offmap->netdev->ifindex;
432 net = dev_net(args->offmap->netdev);
433 get_net(net);
434 ns = &net->ns;
435 } else {
436 args->info->ifindex = 0;
437 ns = NULL;
438 }
439
440 up_read(&bpf_devs_lock);
441 rtnl_unlock();
442
443 return ns;
444}
445
446int bpf_map_offload_info_fill(struct bpf_map_info *info, struct bpf_map *map)
447{
448 struct ns_get_path_bpf_map_args args = {
449 .offmap = map_to_offmap(map),
450 .info = info,
451 };
452 struct inode *ns_inode;
453 struct path ns_path;
454 void *res;
455
456 res = ns_get_path_cb(&ns_path, bpf_map_offload_info_fill_ns, &args);
457 if (IS_ERR(res)) {
458 if (!info->ifindex)
459 return -ENODEV;
460 return PTR_ERR(res);
461 }
462
463 ns_inode = ns_path.dentry->d_inode;
464 info->netns_dev = new_encode_dev(ns_inode->i_sb->s_dev);
465 info->netns_ino = ns_inode->i_ino;
466 path_put(&ns_path);
467
468 return 0;
469}
470
471bool bpf_offload_dev_match(struct bpf_prog *prog, struct bpf_map *map)
472{
473 struct bpf_offloaded_map *offmap;
474 struct bpf_prog_offload *offload;
475 bool ret;
476
477 if (!bpf_prog_is_dev_bound(prog->aux) || !bpf_map_is_dev_bound(map))
478 return false;
479
480 down_read(&bpf_devs_lock);
481 offload = prog->aux->offload;
482 offmap = map_to_offmap(map);
483
484 ret = offload && offload->netdev == offmap->netdev;
485 up_read(&bpf_devs_lock);
486
487 return ret;
488}
489
490static void bpf_offload_orphan_all_progs(struct net_device *netdev)
491{
492 struct bpf_prog_offload *offload, *tmp;
493
494 list_for_each_entry_safe(offload, tmp, &bpf_prog_offload_devs, offloads)
495 if (offload->netdev == netdev)
496 __bpf_prog_offload_destroy(offload->prog);
497}
498
499static void bpf_offload_orphan_all_maps(struct net_device *netdev)
500{
501 struct bpf_offloaded_map *offmap, *tmp;
502
503 list_for_each_entry_safe(offmap, tmp, &bpf_map_offload_devs, offloads)
504 if (offmap->netdev == netdev)
505 __bpf_map_offload_destroy(offmap);
506}
507
508static int bpf_offload_notification(struct notifier_block *notifier,
509 ulong event, void *ptr)
510{
511 struct net_device *netdev = netdev_notifier_info_to_dev(ptr);
512
513 ASSERT_RTNL();
514
515 switch (event) {
516 case NETDEV_UNREGISTER:
517 /* ignore namespace changes */
518 if (netdev->reg_state != NETREG_UNREGISTERING)
519 break;
520
521 down_write(&bpf_devs_lock);
522 bpf_offload_orphan_all_progs(netdev);
523 bpf_offload_orphan_all_maps(netdev);
524 up_write(&bpf_devs_lock);
525 break;
526 default:
527 break;
528 }
529 return NOTIFY_OK;
530}
531
532static struct notifier_block bpf_offload_notifier = {
533 .notifier_call = bpf_offload_notification,
534};
535
536static int __init bpf_offload_init(void)
537{
538 register_netdevice_notifier(&bpf_offload_notifier);
539 return 0;
540}
541
542subsys_initcall(bpf_offload_init);
1/*
2 * Copyright (C) 2017-2018 Netronome Systems, Inc.
3 *
4 * This software is licensed under the GNU General License Version 2,
5 * June 1991 as shown in the file COPYING in the top-level directory of this
6 * source tree.
7 *
8 * THE COPYRIGHT HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS"
9 * WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING,
10 * BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
11 * FOR A PARTICULAR PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE
12 * OF THE PROGRAM IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME
13 * THE COST OF ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
14 */
15
16#include <linux/bpf.h>
17#include <linux/bpf_verifier.h>
18#include <linux/bug.h>
19#include <linux/kdev_t.h>
20#include <linux/list.h>
21#include <linux/lockdep.h>
22#include <linux/netdevice.h>
23#include <linux/printk.h>
24#include <linux/proc_ns.h>
25#include <linux/rhashtable.h>
26#include <linux/rtnetlink.h>
27#include <linux/rwsem.h>
28#include <net/xdp.h>
29
30/* Protects offdevs, members of bpf_offload_netdev and offload members
31 * of all progs.
32 * RTNL lock cannot be taken when holding this lock.
33 */
34static DECLARE_RWSEM(bpf_devs_lock);
35
36struct bpf_offload_dev {
37 const struct bpf_prog_offload_ops *ops;
38 struct list_head netdevs;
39 void *priv;
40};
41
42struct bpf_offload_netdev {
43 struct rhash_head l;
44 struct net_device *netdev;
45 struct bpf_offload_dev *offdev; /* NULL when bound-only */
46 struct list_head progs;
47 struct list_head maps;
48 struct list_head offdev_netdevs;
49};
50
51static const struct rhashtable_params offdevs_params = {
52 .nelem_hint = 4,
53 .key_len = sizeof(struct net_device *),
54 .key_offset = offsetof(struct bpf_offload_netdev, netdev),
55 .head_offset = offsetof(struct bpf_offload_netdev, l),
56 .automatic_shrinking = true,
57};
58
59static struct rhashtable offdevs;
60
61static int bpf_dev_offload_check(struct net_device *netdev)
62{
63 if (!netdev)
64 return -EINVAL;
65 if (!netdev->netdev_ops->ndo_bpf)
66 return -EOPNOTSUPP;
67 return 0;
68}
69
70static struct bpf_offload_netdev *
71bpf_offload_find_netdev(struct net_device *netdev)
72{
73 lockdep_assert_held(&bpf_devs_lock);
74
75 return rhashtable_lookup_fast(&offdevs, &netdev, offdevs_params);
76}
77
78static int __bpf_offload_dev_netdev_register(struct bpf_offload_dev *offdev,
79 struct net_device *netdev)
80{
81 struct bpf_offload_netdev *ondev;
82 int err;
83
84 ondev = kzalloc(sizeof(*ondev), GFP_KERNEL);
85 if (!ondev)
86 return -ENOMEM;
87
88 ondev->netdev = netdev;
89 ondev->offdev = offdev;
90 INIT_LIST_HEAD(&ondev->progs);
91 INIT_LIST_HEAD(&ondev->maps);
92
93 err = rhashtable_insert_fast(&offdevs, &ondev->l, offdevs_params);
94 if (err) {
95 netdev_warn(netdev, "failed to register for BPF offload\n");
96 goto err_free;
97 }
98
99 if (offdev)
100 list_add(&ondev->offdev_netdevs, &offdev->netdevs);
101 return 0;
102
103err_free:
104 kfree(ondev);
105 return err;
106}
107
108static void __bpf_prog_offload_destroy(struct bpf_prog *prog)
109{
110 struct bpf_prog_offload *offload = prog->aux->offload;
111
112 if (offload->dev_state)
113 offload->offdev->ops->destroy(prog);
114
115 list_del_init(&offload->offloads);
116 kfree(offload);
117 prog->aux->offload = NULL;
118}
119
120static int bpf_map_offload_ndo(struct bpf_offloaded_map *offmap,
121 enum bpf_netdev_command cmd)
122{
123 struct netdev_bpf data = {};
124 struct net_device *netdev;
125
126 ASSERT_RTNL();
127
128 data.command = cmd;
129 data.offmap = offmap;
130 /* Caller must make sure netdev is valid */
131 netdev = offmap->netdev;
132
133 return netdev->netdev_ops->ndo_bpf(netdev, &data);
134}
135
136static void __bpf_map_offload_destroy(struct bpf_offloaded_map *offmap)
137{
138 WARN_ON(bpf_map_offload_ndo(offmap, BPF_OFFLOAD_MAP_FREE));
139 /* Make sure BPF_MAP_GET_NEXT_ID can't find this dead map */
140 bpf_map_free_id(&offmap->map);
141 list_del_init(&offmap->offloads);
142 offmap->netdev = NULL;
143}
144
145static void __bpf_offload_dev_netdev_unregister(struct bpf_offload_dev *offdev,
146 struct net_device *netdev)
147{
148 struct bpf_offload_netdev *ondev, *altdev = NULL;
149 struct bpf_offloaded_map *offmap, *mtmp;
150 struct bpf_prog_offload *offload, *ptmp;
151
152 ASSERT_RTNL();
153
154 ondev = rhashtable_lookup_fast(&offdevs, &netdev, offdevs_params);
155 if (WARN_ON(!ondev))
156 return;
157
158 WARN_ON(rhashtable_remove_fast(&offdevs, &ondev->l, offdevs_params));
159
160 /* Try to move the objects to another netdev of the device */
161 if (offdev) {
162 list_del(&ondev->offdev_netdevs);
163 altdev = list_first_entry_or_null(&offdev->netdevs,
164 struct bpf_offload_netdev,
165 offdev_netdevs);
166 }
167
168 if (altdev) {
169 list_for_each_entry(offload, &ondev->progs, offloads)
170 offload->netdev = altdev->netdev;
171 list_splice_init(&ondev->progs, &altdev->progs);
172
173 list_for_each_entry(offmap, &ondev->maps, offloads)
174 offmap->netdev = altdev->netdev;
175 list_splice_init(&ondev->maps, &altdev->maps);
176 } else {
177 list_for_each_entry_safe(offload, ptmp, &ondev->progs, offloads)
178 __bpf_prog_offload_destroy(offload->prog);
179 list_for_each_entry_safe(offmap, mtmp, &ondev->maps, offloads)
180 __bpf_map_offload_destroy(offmap);
181 }
182
183 WARN_ON(!list_empty(&ondev->progs));
184 WARN_ON(!list_empty(&ondev->maps));
185 kfree(ondev);
186}
187
188static int __bpf_prog_dev_bound_init(struct bpf_prog *prog, struct net_device *netdev)
189{
190 struct bpf_offload_netdev *ondev;
191 struct bpf_prog_offload *offload;
192 int err;
193
194 offload = kzalloc(sizeof(*offload), GFP_USER);
195 if (!offload)
196 return -ENOMEM;
197
198 offload->prog = prog;
199 offload->netdev = netdev;
200
201 ondev = bpf_offload_find_netdev(offload->netdev);
202 /* When program is offloaded require presence of "true"
203 * bpf_offload_netdev, avoid the one created for !ondev case below.
204 */
205 if (bpf_prog_is_offloaded(prog->aux) && (!ondev || !ondev->offdev)) {
206 err = -EINVAL;
207 goto err_free;
208 }
209 if (!ondev) {
210 /* When only binding to the device, explicitly
211 * create an entry in the hashtable.
212 */
213 err = __bpf_offload_dev_netdev_register(NULL, offload->netdev);
214 if (err)
215 goto err_free;
216 ondev = bpf_offload_find_netdev(offload->netdev);
217 }
218 offload->offdev = ondev->offdev;
219 prog->aux->offload = offload;
220 list_add_tail(&offload->offloads, &ondev->progs);
221
222 return 0;
223err_free:
224 kfree(offload);
225 return err;
226}
227
228int bpf_prog_dev_bound_init(struct bpf_prog *prog, union bpf_attr *attr)
229{
230 struct net_device *netdev;
231 int err;
232
233 if (attr->prog_type != BPF_PROG_TYPE_SCHED_CLS &&
234 attr->prog_type != BPF_PROG_TYPE_XDP)
235 return -EINVAL;
236
237 if (attr->prog_flags & ~(BPF_F_XDP_DEV_BOUND_ONLY | BPF_F_XDP_HAS_FRAGS))
238 return -EINVAL;
239
240 /* Frags are allowed only if program is dev-bound-only, but not
241 * if it is requesting bpf offload.
242 */
243 if (attr->prog_flags & BPF_F_XDP_HAS_FRAGS &&
244 !(attr->prog_flags & BPF_F_XDP_DEV_BOUND_ONLY))
245 return -EINVAL;
246
247 if (attr->prog_type == BPF_PROG_TYPE_SCHED_CLS &&
248 attr->prog_flags & BPF_F_XDP_DEV_BOUND_ONLY)
249 return -EINVAL;
250
251 netdev = dev_get_by_index(current->nsproxy->net_ns, attr->prog_ifindex);
252 if (!netdev)
253 return -EINVAL;
254
255 err = bpf_dev_offload_check(netdev);
256 if (err)
257 goto out;
258
259 prog->aux->offload_requested = !(attr->prog_flags & BPF_F_XDP_DEV_BOUND_ONLY);
260
261 down_write(&bpf_devs_lock);
262 err = __bpf_prog_dev_bound_init(prog, netdev);
263 up_write(&bpf_devs_lock);
264
265out:
266 dev_put(netdev);
267 return err;
268}
269
270int bpf_prog_dev_bound_inherit(struct bpf_prog *new_prog, struct bpf_prog *old_prog)
271{
272 int err;
273
274 if (!bpf_prog_is_dev_bound(old_prog->aux))
275 return 0;
276
277 if (bpf_prog_is_offloaded(old_prog->aux))
278 return -EINVAL;
279
280 new_prog->aux->dev_bound = old_prog->aux->dev_bound;
281 new_prog->aux->offload_requested = old_prog->aux->offload_requested;
282
283 down_write(&bpf_devs_lock);
284 if (!old_prog->aux->offload) {
285 err = -EINVAL;
286 goto out;
287 }
288
289 err = __bpf_prog_dev_bound_init(new_prog, old_prog->aux->offload->netdev);
290
291out:
292 up_write(&bpf_devs_lock);
293 return err;
294}
295
296int bpf_prog_offload_verifier_prep(struct bpf_prog *prog)
297{
298 struct bpf_prog_offload *offload;
299 int ret = -ENODEV;
300
301 down_read(&bpf_devs_lock);
302 offload = prog->aux->offload;
303 if (offload) {
304 ret = offload->offdev->ops->prepare(prog);
305 offload->dev_state = !ret;
306 }
307 up_read(&bpf_devs_lock);
308
309 return ret;
310}
311
312int bpf_prog_offload_verify_insn(struct bpf_verifier_env *env,
313 int insn_idx, int prev_insn_idx)
314{
315 struct bpf_prog_offload *offload;
316 int ret = -ENODEV;
317
318 down_read(&bpf_devs_lock);
319 offload = env->prog->aux->offload;
320 if (offload)
321 ret = offload->offdev->ops->insn_hook(env, insn_idx,
322 prev_insn_idx);
323 up_read(&bpf_devs_lock);
324
325 return ret;
326}
327
328int bpf_prog_offload_finalize(struct bpf_verifier_env *env)
329{
330 struct bpf_prog_offload *offload;
331 int ret = -ENODEV;
332
333 down_read(&bpf_devs_lock);
334 offload = env->prog->aux->offload;
335 if (offload) {
336 if (offload->offdev->ops->finalize)
337 ret = offload->offdev->ops->finalize(env);
338 else
339 ret = 0;
340 }
341 up_read(&bpf_devs_lock);
342
343 return ret;
344}
345
346void
347bpf_prog_offload_replace_insn(struct bpf_verifier_env *env, u32 off,
348 struct bpf_insn *insn)
349{
350 const struct bpf_prog_offload_ops *ops;
351 struct bpf_prog_offload *offload;
352 int ret = -EOPNOTSUPP;
353
354 down_read(&bpf_devs_lock);
355 offload = env->prog->aux->offload;
356 if (offload) {
357 ops = offload->offdev->ops;
358 if (!offload->opt_failed && ops->replace_insn)
359 ret = ops->replace_insn(env, off, insn);
360 offload->opt_failed |= ret;
361 }
362 up_read(&bpf_devs_lock);
363}
364
365void
366bpf_prog_offload_remove_insns(struct bpf_verifier_env *env, u32 off, u32 cnt)
367{
368 struct bpf_prog_offload *offload;
369 int ret = -EOPNOTSUPP;
370
371 down_read(&bpf_devs_lock);
372 offload = env->prog->aux->offload;
373 if (offload) {
374 if (!offload->opt_failed && offload->offdev->ops->remove_insns)
375 ret = offload->offdev->ops->remove_insns(env, off, cnt);
376 offload->opt_failed |= ret;
377 }
378 up_read(&bpf_devs_lock);
379}
380
381void bpf_prog_dev_bound_destroy(struct bpf_prog *prog)
382{
383 struct bpf_offload_netdev *ondev;
384 struct net_device *netdev;
385
386 rtnl_lock();
387 down_write(&bpf_devs_lock);
388 if (prog->aux->offload) {
389 list_del_init(&prog->aux->offload->offloads);
390
391 netdev = prog->aux->offload->netdev;
392 __bpf_prog_offload_destroy(prog);
393
394 ondev = bpf_offload_find_netdev(netdev);
395 if (!ondev->offdev && list_empty(&ondev->progs))
396 __bpf_offload_dev_netdev_unregister(NULL, netdev);
397 }
398 up_write(&bpf_devs_lock);
399 rtnl_unlock();
400}
401
402static int bpf_prog_offload_translate(struct bpf_prog *prog)
403{
404 struct bpf_prog_offload *offload;
405 int ret = -ENODEV;
406
407 down_read(&bpf_devs_lock);
408 offload = prog->aux->offload;
409 if (offload)
410 ret = offload->offdev->ops->translate(prog);
411 up_read(&bpf_devs_lock);
412
413 return ret;
414}
415
416static unsigned int bpf_prog_warn_on_exec(const void *ctx,
417 const struct bpf_insn *insn)
418{
419 WARN(1, "attempt to execute device eBPF program on the host!");
420 return 0;
421}
422
423int bpf_prog_offload_compile(struct bpf_prog *prog)
424{
425 prog->bpf_func = bpf_prog_warn_on_exec;
426
427 return bpf_prog_offload_translate(prog);
428}
429
430struct ns_get_path_bpf_prog_args {
431 struct bpf_prog *prog;
432 struct bpf_prog_info *info;
433};
434
435static struct ns_common *bpf_prog_offload_info_fill_ns(void *private_data)
436{
437 struct ns_get_path_bpf_prog_args *args = private_data;
438 struct bpf_prog_aux *aux = args->prog->aux;
439 struct ns_common *ns;
440 struct net *net;
441
442 rtnl_lock();
443 down_read(&bpf_devs_lock);
444
445 if (aux->offload) {
446 args->info->ifindex = aux->offload->netdev->ifindex;
447 net = dev_net(aux->offload->netdev);
448 get_net(net);
449 ns = &net->ns;
450 } else {
451 args->info->ifindex = 0;
452 ns = NULL;
453 }
454
455 up_read(&bpf_devs_lock);
456 rtnl_unlock();
457
458 return ns;
459}
460
461int bpf_prog_offload_info_fill(struct bpf_prog_info *info,
462 struct bpf_prog *prog)
463{
464 struct ns_get_path_bpf_prog_args args = {
465 .prog = prog,
466 .info = info,
467 };
468 struct bpf_prog_aux *aux = prog->aux;
469 struct inode *ns_inode;
470 struct path ns_path;
471 char __user *uinsns;
472 int res;
473 u32 ulen;
474
475 res = ns_get_path_cb(&ns_path, bpf_prog_offload_info_fill_ns, &args);
476 if (res) {
477 if (!info->ifindex)
478 return -ENODEV;
479 return res;
480 }
481
482 down_read(&bpf_devs_lock);
483
484 if (!aux->offload) {
485 up_read(&bpf_devs_lock);
486 return -ENODEV;
487 }
488
489 ulen = info->jited_prog_len;
490 info->jited_prog_len = aux->offload->jited_len;
491 if (info->jited_prog_len && ulen) {
492 uinsns = u64_to_user_ptr(info->jited_prog_insns);
493 ulen = min_t(u32, info->jited_prog_len, ulen);
494 if (copy_to_user(uinsns, aux->offload->jited_image, ulen)) {
495 up_read(&bpf_devs_lock);
496 return -EFAULT;
497 }
498 }
499
500 up_read(&bpf_devs_lock);
501
502 ns_inode = ns_path.dentry->d_inode;
503 info->netns_dev = new_encode_dev(ns_inode->i_sb->s_dev);
504 info->netns_ino = ns_inode->i_ino;
505 path_put(&ns_path);
506
507 return 0;
508}
509
510const struct bpf_prog_ops bpf_offload_prog_ops = {
511};
512
513struct bpf_map *bpf_map_offload_map_alloc(union bpf_attr *attr)
514{
515 struct net *net = current->nsproxy->net_ns;
516 struct bpf_offload_netdev *ondev;
517 struct bpf_offloaded_map *offmap;
518 int err;
519
520 if (!capable(CAP_SYS_ADMIN))
521 return ERR_PTR(-EPERM);
522 if (attr->map_type != BPF_MAP_TYPE_ARRAY &&
523 attr->map_type != BPF_MAP_TYPE_HASH)
524 return ERR_PTR(-EINVAL);
525
526 offmap = bpf_map_area_alloc(sizeof(*offmap), NUMA_NO_NODE);
527 if (!offmap)
528 return ERR_PTR(-ENOMEM);
529
530 bpf_map_init_from_attr(&offmap->map, attr);
531
532 rtnl_lock();
533 down_write(&bpf_devs_lock);
534 offmap->netdev = __dev_get_by_index(net, attr->map_ifindex);
535 err = bpf_dev_offload_check(offmap->netdev);
536 if (err)
537 goto err_unlock;
538
539 ondev = bpf_offload_find_netdev(offmap->netdev);
540 if (!ondev) {
541 err = -EINVAL;
542 goto err_unlock;
543 }
544
545 err = bpf_map_offload_ndo(offmap, BPF_OFFLOAD_MAP_ALLOC);
546 if (err)
547 goto err_unlock;
548
549 list_add_tail(&offmap->offloads, &ondev->maps);
550 up_write(&bpf_devs_lock);
551 rtnl_unlock();
552
553 return &offmap->map;
554
555err_unlock:
556 up_write(&bpf_devs_lock);
557 rtnl_unlock();
558 bpf_map_area_free(offmap);
559 return ERR_PTR(err);
560}
561
562void bpf_map_offload_map_free(struct bpf_map *map)
563{
564 struct bpf_offloaded_map *offmap = map_to_offmap(map);
565
566 rtnl_lock();
567 down_write(&bpf_devs_lock);
568 if (offmap->netdev)
569 __bpf_map_offload_destroy(offmap);
570 up_write(&bpf_devs_lock);
571 rtnl_unlock();
572
573 bpf_map_area_free(offmap);
574}
575
576u64 bpf_map_offload_map_mem_usage(const struct bpf_map *map)
577{
578 /* The memory dynamically allocated in netdev dev_ops is not counted */
579 return sizeof(struct bpf_offloaded_map);
580}
581
582int bpf_map_offload_lookup_elem(struct bpf_map *map, void *key, void *value)
583{
584 struct bpf_offloaded_map *offmap = map_to_offmap(map);
585 int ret = -ENODEV;
586
587 down_read(&bpf_devs_lock);
588 if (offmap->netdev)
589 ret = offmap->dev_ops->map_lookup_elem(offmap, key, value);
590 up_read(&bpf_devs_lock);
591
592 return ret;
593}
594
595int bpf_map_offload_update_elem(struct bpf_map *map,
596 void *key, void *value, u64 flags)
597{
598 struct bpf_offloaded_map *offmap = map_to_offmap(map);
599 int ret = -ENODEV;
600
601 if (unlikely(flags > BPF_EXIST))
602 return -EINVAL;
603
604 down_read(&bpf_devs_lock);
605 if (offmap->netdev)
606 ret = offmap->dev_ops->map_update_elem(offmap, key, value,
607 flags);
608 up_read(&bpf_devs_lock);
609
610 return ret;
611}
612
613int bpf_map_offload_delete_elem(struct bpf_map *map, void *key)
614{
615 struct bpf_offloaded_map *offmap = map_to_offmap(map);
616 int ret = -ENODEV;
617
618 down_read(&bpf_devs_lock);
619 if (offmap->netdev)
620 ret = offmap->dev_ops->map_delete_elem(offmap, key);
621 up_read(&bpf_devs_lock);
622
623 return ret;
624}
625
626int bpf_map_offload_get_next_key(struct bpf_map *map, void *key, void *next_key)
627{
628 struct bpf_offloaded_map *offmap = map_to_offmap(map);
629 int ret = -ENODEV;
630
631 down_read(&bpf_devs_lock);
632 if (offmap->netdev)
633 ret = offmap->dev_ops->map_get_next_key(offmap, key, next_key);
634 up_read(&bpf_devs_lock);
635
636 return ret;
637}
638
639struct ns_get_path_bpf_map_args {
640 struct bpf_offloaded_map *offmap;
641 struct bpf_map_info *info;
642};
643
644static struct ns_common *bpf_map_offload_info_fill_ns(void *private_data)
645{
646 struct ns_get_path_bpf_map_args *args = private_data;
647 struct ns_common *ns;
648 struct net *net;
649
650 rtnl_lock();
651 down_read(&bpf_devs_lock);
652
653 if (args->offmap->netdev) {
654 args->info->ifindex = args->offmap->netdev->ifindex;
655 net = dev_net(args->offmap->netdev);
656 get_net(net);
657 ns = &net->ns;
658 } else {
659 args->info->ifindex = 0;
660 ns = NULL;
661 }
662
663 up_read(&bpf_devs_lock);
664 rtnl_unlock();
665
666 return ns;
667}
668
669int bpf_map_offload_info_fill(struct bpf_map_info *info, struct bpf_map *map)
670{
671 struct ns_get_path_bpf_map_args args = {
672 .offmap = map_to_offmap(map),
673 .info = info,
674 };
675 struct inode *ns_inode;
676 struct path ns_path;
677 int res;
678
679 res = ns_get_path_cb(&ns_path, bpf_map_offload_info_fill_ns, &args);
680 if (res) {
681 if (!info->ifindex)
682 return -ENODEV;
683 return res;
684 }
685
686 ns_inode = ns_path.dentry->d_inode;
687 info->netns_dev = new_encode_dev(ns_inode->i_sb->s_dev);
688 info->netns_ino = ns_inode->i_ino;
689 path_put(&ns_path);
690
691 return 0;
692}
693
694static bool __bpf_offload_dev_match(struct bpf_prog *prog,
695 struct net_device *netdev)
696{
697 struct bpf_offload_netdev *ondev1, *ondev2;
698 struct bpf_prog_offload *offload;
699
700 if (!bpf_prog_is_dev_bound(prog->aux))
701 return false;
702
703 offload = prog->aux->offload;
704 if (!offload)
705 return false;
706 if (offload->netdev == netdev)
707 return true;
708
709 ondev1 = bpf_offload_find_netdev(offload->netdev);
710 ondev2 = bpf_offload_find_netdev(netdev);
711
712 return ondev1 && ondev2 && ondev1->offdev == ondev2->offdev;
713}
714
715bool bpf_offload_dev_match(struct bpf_prog *prog, struct net_device *netdev)
716{
717 bool ret;
718
719 down_read(&bpf_devs_lock);
720 ret = __bpf_offload_dev_match(prog, netdev);
721 up_read(&bpf_devs_lock);
722
723 return ret;
724}
725EXPORT_SYMBOL_GPL(bpf_offload_dev_match);
726
727bool bpf_prog_dev_bound_match(const struct bpf_prog *lhs, const struct bpf_prog *rhs)
728{
729 bool ret;
730
731 if (bpf_prog_is_offloaded(lhs->aux) != bpf_prog_is_offloaded(rhs->aux))
732 return false;
733
734 down_read(&bpf_devs_lock);
735 ret = lhs->aux->offload && rhs->aux->offload &&
736 lhs->aux->offload->netdev &&
737 lhs->aux->offload->netdev == rhs->aux->offload->netdev;
738 up_read(&bpf_devs_lock);
739
740 return ret;
741}
742
743bool bpf_offload_prog_map_match(struct bpf_prog *prog, struct bpf_map *map)
744{
745 struct bpf_offloaded_map *offmap;
746 bool ret;
747
748 if (!bpf_map_is_offloaded(map))
749 return bpf_map_offload_neutral(map);
750 offmap = map_to_offmap(map);
751
752 down_read(&bpf_devs_lock);
753 ret = __bpf_offload_dev_match(prog, offmap->netdev);
754 up_read(&bpf_devs_lock);
755
756 return ret;
757}
758
759int bpf_offload_dev_netdev_register(struct bpf_offload_dev *offdev,
760 struct net_device *netdev)
761{
762 int err;
763
764 down_write(&bpf_devs_lock);
765 err = __bpf_offload_dev_netdev_register(offdev, netdev);
766 up_write(&bpf_devs_lock);
767 return err;
768}
769EXPORT_SYMBOL_GPL(bpf_offload_dev_netdev_register);
770
771void bpf_offload_dev_netdev_unregister(struct bpf_offload_dev *offdev,
772 struct net_device *netdev)
773{
774 down_write(&bpf_devs_lock);
775 __bpf_offload_dev_netdev_unregister(offdev, netdev);
776 up_write(&bpf_devs_lock);
777}
778EXPORT_SYMBOL_GPL(bpf_offload_dev_netdev_unregister);
779
780struct bpf_offload_dev *
781bpf_offload_dev_create(const struct bpf_prog_offload_ops *ops, void *priv)
782{
783 struct bpf_offload_dev *offdev;
784
785 offdev = kzalloc(sizeof(*offdev), GFP_KERNEL);
786 if (!offdev)
787 return ERR_PTR(-ENOMEM);
788
789 offdev->ops = ops;
790 offdev->priv = priv;
791 INIT_LIST_HEAD(&offdev->netdevs);
792
793 return offdev;
794}
795EXPORT_SYMBOL_GPL(bpf_offload_dev_create);
796
797void bpf_offload_dev_destroy(struct bpf_offload_dev *offdev)
798{
799 WARN_ON(!list_empty(&offdev->netdevs));
800 kfree(offdev);
801}
802EXPORT_SYMBOL_GPL(bpf_offload_dev_destroy);
803
804void *bpf_offload_dev_priv(struct bpf_offload_dev *offdev)
805{
806 return offdev->priv;
807}
808EXPORT_SYMBOL_GPL(bpf_offload_dev_priv);
809
810void bpf_dev_bound_netdev_unregister(struct net_device *dev)
811{
812 struct bpf_offload_netdev *ondev;
813
814 ASSERT_RTNL();
815
816 down_write(&bpf_devs_lock);
817 ondev = bpf_offload_find_netdev(dev);
818 if (ondev && !ondev->offdev)
819 __bpf_offload_dev_netdev_unregister(NULL, ondev->netdev);
820 up_write(&bpf_devs_lock);
821}
822
823int bpf_dev_bound_kfunc_check(struct bpf_verifier_log *log,
824 struct bpf_prog_aux *prog_aux)
825{
826 if (!bpf_prog_is_dev_bound(prog_aux)) {
827 bpf_log(log, "metadata kfuncs require device-bound program\n");
828 return -EINVAL;
829 }
830
831 if (bpf_prog_is_offloaded(prog_aux)) {
832 bpf_log(log, "metadata kfuncs can't be offloaded\n");
833 return -EINVAL;
834 }
835
836 return 0;
837}
838
839void *bpf_dev_bound_resolve_kfunc(struct bpf_prog *prog, u32 func_id)
840{
841 const struct xdp_metadata_ops *ops;
842 void *p = NULL;
843
844 /* We don't hold bpf_devs_lock while resolving several
845 * kfuncs and can race with the unregister_netdevice().
846 * We rely on bpf_dev_bound_match() check at attach
847 * to render this program unusable.
848 */
849 down_read(&bpf_devs_lock);
850 if (!prog->aux->offload)
851 goto out;
852
853 ops = prog->aux->offload->netdev->xdp_metadata_ops;
854 if (!ops)
855 goto out;
856
857#define XDP_METADATA_KFUNC(name, _, __, xmo) \
858 if (func_id == bpf_xdp_metadata_kfunc_id(name)) p = ops->xmo;
859 XDP_METADATA_KFUNC_xxx
860#undef XDP_METADATA_KFUNC
861
862out:
863 up_read(&bpf_devs_lock);
864
865 return p;
866}
867
868static int __init bpf_offload_init(void)
869{
870 return rhashtable_init(&offdevs, &offdevs_params);
871}
872
873core_initcall(bpf_offload_init);