2 * Copyright (C) 2017-2018 Netronome Systems, Inc.
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
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.
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>
30 /* Protects offdevs, members of bpf_offload_netdev and offload members
32 * RTNL lock cannot be taken when holding this lock.
34 static DECLARE_RWSEM(bpf_devs_lock);
36 struct bpf_offload_dev {
37 const struct bpf_prog_offload_ops *ops;
38 struct list_head netdevs;
42 struct bpf_offload_netdev {
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;
51 static const struct rhashtable_params offdevs_params = {
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,
59 static struct rhashtable offdevs;
61 static int bpf_dev_offload_check(struct net_device *netdev)
65 if (!netdev->netdev_ops->ndo_bpf)
70 static struct bpf_offload_netdev *
71 bpf_offload_find_netdev(struct net_device *netdev)
73 lockdep_assert_held(&bpf_devs_lock);
75 return rhashtable_lookup_fast(&offdevs, &netdev, offdevs_params);
78 static int __bpf_offload_dev_netdev_register(struct bpf_offload_dev *offdev,
79 struct net_device *netdev)
81 struct bpf_offload_netdev *ondev;
84 ondev = kzalloc(sizeof(*ondev), GFP_KERNEL);
88 ondev->netdev = netdev;
89 ondev->offdev = offdev;
90 INIT_LIST_HEAD(&ondev->progs);
91 INIT_LIST_HEAD(&ondev->maps);
93 err = rhashtable_insert_fast(&offdevs, &ondev->l, offdevs_params);
95 netdev_warn(netdev, "failed to register for BPF offload\n");
100 list_add(&ondev->offdev_netdevs, &offdev->netdevs);
108 static void __bpf_prog_offload_destroy(struct bpf_prog *prog)
110 struct bpf_prog_offload *offload = prog->aux->offload;
112 if (offload->dev_state)
113 offload->offdev->ops->destroy(prog);
115 list_del_init(&offload->offloads);
117 prog->aux->offload = NULL;
120 static int bpf_map_offload_ndo(struct bpf_offloaded_map *offmap,
121 enum bpf_netdev_command cmd)
123 struct netdev_bpf data = {};
124 struct net_device *netdev;
129 data.offmap = offmap;
130 /* Caller must make sure netdev is valid */
131 netdev = offmap->netdev;
133 return netdev->netdev_ops->ndo_bpf(netdev, &data);
136 static void __bpf_map_offload_destroy(struct bpf_offloaded_map *offmap)
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;
145 static void __bpf_offload_dev_netdev_unregister(struct bpf_offload_dev *offdev,
146 struct net_device *netdev)
148 struct bpf_offload_netdev *ondev, *altdev = NULL;
149 struct bpf_offloaded_map *offmap, *mtmp;
150 struct bpf_prog_offload *offload, *ptmp;
154 ondev = rhashtable_lookup_fast(&offdevs, &netdev, offdevs_params);
158 WARN_ON(rhashtable_remove_fast(&offdevs, &ondev->l, offdevs_params));
160 /* Try to move the objects to another netdev of the device */
162 list_del(&ondev->offdev_netdevs);
163 altdev = list_first_entry_or_null(&offdev->netdevs,
164 struct bpf_offload_netdev,
169 list_for_each_entry(offload, &ondev->progs, offloads)
170 offload->netdev = altdev->netdev;
171 list_splice_init(&ondev->progs, &altdev->progs);
173 list_for_each_entry(offmap, &ondev->maps, offloads)
174 offmap->netdev = altdev->netdev;
175 list_splice_init(&ondev->maps, &altdev->maps);
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);
183 WARN_ON(!list_empty(&ondev->progs));
184 WARN_ON(!list_empty(&ondev->maps));
188 static int __bpf_prog_dev_bound_init(struct bpf_prog *prog, struct net_device *netdev)
190 struct bpf_offload_netdev *ondev;
191 struct bpf_prog_offload *offload;
194 offload = kzalloc(sizeof(*offload), GFP_USER);
198 offload->prog = prog;
199 offload->netdev = netdev;
201 ondev = bpf_offload_find_netdev(offload->netdev);
203 if (bpf_prog_is_offloaded(prog->aux)) {
208 /* When only binding to the device, explicitly
209 * create an entry in the hashtable.
211 err = __bpf_offload_dev_netdev_register(NULL, offload->netdev);
214 ondev = bpf_offload_find_netdev(offload->netdev);
216 offload->offdev = ondev->offdev;
217 prog->aux->offload = offload;
218 list_add_tail(&offload->offloads, &ondev->progs);
226 int bpf_prog_dev_bound_init(struct bpf_prog *prog, union bpf_attr *attr)
228 struct net_device *netdev;
231 if (attr->prog_type != BPF_PROG_TYPE_SCHED_CLS &&
232 attr->prog_type != BPF_PROG_TYPE_XDP)
235 if (attr->prog_flags & ~BPF_F_XDP_DEV_BOUND_ONLY)
238 if (attr->prog_type == BPF_PROG_TYPE_SCHED_CLS &&
239 attr->prog_flags & BPF_F_XDP_DEV_BOUND_ONLY)
242 netdev = dev_get_by_index(current->nsproxy->net_ns, attr->prog_ifindex);
246 err = bpf_dev_offload_check(netdev);
250 prog->aux->offload_requested = !(attr->prog_flags & BPF_F_XDP_DEV_BOUND_ONLY);
252 down_write(&bpf_devs_lock);
253 err = __bpf_prog_dev_bound_init(prog, netdev);
254 up_write(&bpf_devs_lock);
261 int bpf_prog_dev_bound_inherit(struct bpf_prog *new_prog, struct bpf_prog *old_prog)
265 if (!bpf_prog_is_dev_bound(old_prog->aux))
268 if (bpf_prog_is_offloaded(old_prog->aux))
271 new_prog->aux->dev_bound = old_prog->aux->dev_bound;
272 new_prog->aux->offload_requested = old_prog->aux->offload_requested;
274 down_write(&bpf_devs_lock);
275 if (!old_prog->aux->offload) {
280 err = __bpf_prog_dev_bound_init(new_prog, old_prog->aux->offload->netdev);
283 up_write(&bpf_devs_lock);
287 int bpf_prog_offload_verifier_prep(struct bpf_prog *prog)
289 struct bpf_prog_offload *offload;
292 down_read(&bpf_devs_lock);
293 offload = prog->aux->offload;
295 ret = offload->offdev->ops->prepare(prog);
296 offload->dev_state = !ret;
298 up_read(&bpf_devs_lock);
303 int bpf_prog_offload_verify_insn(struct bpf_verifier_env *env,
304 int insn_idx, int prev_insn_idx)
306 struct bpf_prog_offload *offload;
309 down_read(&bpf_devs_lock);
310 offload = env->prog->aux->offload;
312 ret = offload->offdev->ops->insn_hook(env, insn_idx,
314 up_read(&bpf_devs_lock);
319 int bpf_prog_offload_finalize(struct bpf_verifier_env *env)
321 struct bpf_prog_offload *offload;
324 down_read(&bpf_devs_lock);
325 offload = env->prog->aux->offload;
327 if (offload->offdev->ops->finalize)
328 ret = offload->offdev->ops->finalize(env);
332 up_read(&bpf_devs_lock);
338 bpf_prog_offload_replace_insn(struct bpf_verifier_env *env, u32 off,
339 struct bpf_insn *insn)
341 const struct bpf_prog_offload_ops *ops;
342 struct bpf_prog_offload *offload;
343 int ret = -EOPNOTSUPP;
345 down_read(&bpf_devs_lock);
346 offload = env->prog->aux->offload;
348 ops = offload->offdev->ops;
349 if (!offload->opt_failed && ops->replace_insn)
350 ret = ops->replace_insn(env, off, insn);
351 offload->opt_failed |= ret;
353 up_read(&bpf_devs_lock);
357 bpf_prog_offload_remove_insns(struct bpf_verifier_env *env, u32 off, u32 cnt)
359 struct bpf_prog_offload *offload;
360 int ret = -EOPNOTSUPP;
362 down_read(&bpf_devs_lock);
363 offload = env->prog->aux->offload;
365 if (!offload->opt_failed && offload->offdev->ops->remove_insns)
366 ret = offload->offdev->ops->remove_insns(env, off, cnt);
367 offload->opt_failed |= ret;
369 up_read(&bpf_devs_lock);
372 void bpf_prog_dev_bound_destroy(struct bpf_prog *prog)
374 struct bpf_offload_netdev *ondev;
375 struct net_device *netdev;
378 down_write(&bpf_devs_lock);
379 if (prog->aux->offload) {
380 list_del_init(&prog->aux->offload->offloads);
382 netdev = prog->aux->offload->netdev;
383 __bpf_prog_offload_destroy(prog);
385 ondev = bpf_offload_find_netdev(netdev);
386 if (!ondev->offdev && list_empty(&ondev->progs))
387 __bpf_offload_dev_netdev_unregister(NULL, netdev);
389 up_write(&bpf_devs_lock);
393 static int bpf_prog_offload_translate(struct bpf_prog *prog)
395 struct bpf_prog_offload *offload;
398 down_read(&bpf_devs_lock);
399 offload = prog->aux->offload;
401 ret = offload->offdev->ops->translate(prog);
402 up_read(&bpf_devs_lock);
407 static unsigned int bpf_prog_warn_on_exec(const void *ctx,
408 const struct bpf_insn *insn)
410 WARN(1, "attempt to execute device eBPF program on the host!");
414 int bpf_prog_offload_compile(struct bpf_prog *prog)
416 prog->bpf_func = bpf_prog_warn_on_exec;
418 return bpf_prog_offload_translate(prog);
421 struct ns_get_path_bpf_prog_args {
422 struct bpf_prog *prog;
423 struct bpf_prog_info *info;
426 static struct ns_common *bpf_prog_offload_info_fill_ns(void *private_data)
428 struct ns_get_path_bpf_prog_args *args = private_data;
429 struct bpf_prog_aux *aux = args->prog->aux;
430 struct ns_common *ns;
434 down_read(&bpf_devs_lock);
437 args->info->ifindex = aux->offload->netdev->ifindex;
438 net = dev_net(aux->offload->netdev);
442 args->info->ifindex = 0;
446 up_read(&bpf_devs_lock);
452 int bpf_prog_offload_info_fill(struct bpf_prog_info *info,
453 struct bpf_prog *prog)
455 struct ns_get_path_bpf_prog_args args = {
459 struct bpf_prog_aux *aux = prog->aux;
460 struct inode *ns_inode;
466 res = ns_get_path_cb(&ns_path, bpf_prog_offload_info_fill_ns, &args);
473 down_read(&bpf_devs_lock);
476 up_read(&bpf_devs_lock);
480 ulen = info->jited_prog_len;
481 info->jited_prog_len = aux->offload->jited_len;
482 if (info->jited_prog_len && ulen) {
483 uinsns = u64_to_user_ptr(info->jited_prog_insns);
484 ulen = min_t(u32, info->jited_prog_len, ulen);
485 if (copy_to_user(uinsns, aux->offload->jited_image, ulen)) {
486 up_read(&bpf_devs_lock);
491 up_read(&bpf_devs_lock);
493 ns_inode = ns_path.dentry->d_inode;
494 info->netns_dev = new_encode_dev(ns_inode->i_sb->s_dev);
495 info->netns_ino = ns_inode->i_ino;
501 const struct bpf_prog_ops bpf_offload_prog_ops = {
504 struct bpf_map *bpf_map_offload_map_alloc(union bpf_attr *attr)
506 struct net *net = current->nsproxy->net_ns;
507 struct bpf_offload_netdev *ondev;
508 struct bpf_offloaded_map *offmap;
511 if (!capable(CAP_SYS_ADMIN))
512 return ERR_PTR(-EPERM);
513 if (attr->map_type != BPF_MAP_TYPE_ARRAY &&
514 attr->map_type != BPF_MAP_TYPE_HASH)
515 return ERR_PTR(-EINVAL);
517 offmap = bpf_map_area_alloc(sizeof(*offmap), NUMA_NO_NODE);
519 return ERR_PTR(-ENOMEM);
521 bpf_map_init_from_attr(&offmap->map, attr);
524 down_write(&bpf_devs_lock);
525 offmap->netdev = __dev_get_by_index(net, attr->map_ifindex);
526 err = bpf_dev_offload_check(offmap->netdev);
530 ondev = bpf_offload_find_netdev(offmap->netdev);
536 err = bpf_map_offload_ndo(offmap, BPF_OFFLOAD_MAP_ALLOC);
540 list_add_tail(&offmap->offloads, &ondev->maps);
541 up_write(&bpf_devs_lock);
547 up_write(&bpf_devs_lock);
549 bpf_map_area_free(offmap);
553 void bpf_map_offload_map_free(struct bpf_map *map)
555 struct bpf_offloaded_map *offmap = map_to_offmap(map);
558 down_write(&bpf_devs_lock);
560 __bpf_map_offload_destroy(offmap);
561 up_write(&bpf_devs_lock);
564 bpf_map_area_free(offmap);
567 u64 bpf_map_offload_map_mem_usage(const struct bpf_map *map)
569 /* The memory dynamically allocated in netdev dev_ops is not counted */
570 return sizeof(struct bpf_offloaded_map);
573 int bpf_map_offload_lookup_elem(struct bpf_map *map, void *key, void *value)
575 struct bpf_offloaded_map *offmap = map_to_offmap(map);
578 down_read(&bpf_devs_lock);
580 ret = offmap->dev_ops->map_lookup_elem(offmap, key, value);
581 up_read(&bpf_devs_lock);
586 int bpf_map_offload_update_elem(struct bpf_map *map,
587 void *key, void *value, u64 flags)
589 struct bpf_offloaded_map *offmap = map_to_offmap(map);
592 if (unlikely(flags > BPF_EXIST))
595 down_read(&bpf_devs_lock);
597 ret = offmap->dev_ops->map_update_elem(offmap, key, value,
599 up_read(&bpf_devs_lock);
604 int bpf_map_offload_delete_elem(struct bpf_map *map, void *key)
606 struct bpf_offloaded_map *offmap = map_to_offmap(map);
609 down_read(&bpf_devs_lock);
611 ret = offmap->dev_ops->map_delete_elem(offmap, key);
612 up_read(&bpf_devs_lock);
617 int bpf_map_offload_get_next_key(struct bpf_map *map, void *key, void *next_key)
619 struct bpf_offloaded_map *offmap = map_to_offmap(map);
622 down_read(&bpf_devs_lock);
624 ret = offmap->dev_ops->map_get_next_key(offmap, key, next_key);
625 up_read(&bpf_devs_lock);
630 struct ns_get_path_bpf_map_args {
631 struct bpf_offloaded_map *offmap;
632 struct bpf_map_info *info;
635 static struct ns_common *bpf_map_offload_info_fill_ns(void *private_data)
637 struct ns_get_path_bpf_map_args *args = private_data;
638 struct ns_common *ns;
642 down_read(&bpf_devs_lock);
644 if (args->offmap->netdev) {
645 args->info->ifindex = args->offmap->netdev->ifindex;
646 net = dev_net(args->offmap->netdev);
650 args->info->ifindex = 0;
654 up_read(&bpf_devs_lock);
660 int bpf_map_offload_info_fill(struct bpf_map_info *info, struct bpf_map *map)
662 struct ns_get_path_bpf_map_args args = {
663 .offmap = map_to_offmap(map),
666 struct inode *ns_inode;
670 res = ns_get_path_cb(&ns_path, bpf_map_offload_info_fill_ns, &args);
677 ns_inode = ns_path.dentry->d_inode;
678 info->netns_dev = new_encode_dev(ns_inode->i_sb->s_dev);
679 info->netns_ino = ns_inode->i_ino;
685 static bool __bpf_offload_dev_match(struct bpf_prog *prog,
686 struct net_device *netdev)
688 struct bpf_offload_netdev *ondev1, *ondev2;
689 struct bpf_prog_offload *offload;
691 if (!bpf_prog_is_dev_bound(prog->aux))
694 offload = prog->aux->offload;
697 if (offload->netdev == netdev)
700 ondev1 = bpf_offload_find_netdev(offload->netdev);
701 ondev2 = bpf_offload_find_netdev(netdev);
703 return ondev1 && ondev2 && ondev1->offdev == ondev2->offdev;
706 bool bpf_offload_dev_match(struct bpf_prog *prog, struct net_device *netdev)
710 down_read(&bpf_devs_lock);
711 ret = __bpf_offload_dev_match(prog, netdev);
712 up_read(&bpf_devs_lock);
716 EXPORT_SYMBOL_GPL(bpf_offload_dev_match);
718 bool bpf_prog_dev_bound_match(const struct bpf_prog *lhs, const struct bpf_prog *rhs)
722 if (bpf_prog_is_offloaded(lhs->aux) != bpf_prog_is_offloaded(rhs->aux))
725 down_read(&bpf_devs_lock);
726 ret = lhs->aux->offload && rhs->aux->offload &&
727 lhs->aux->offload->netdev &&
728 lhs->aux->offload->netdev == rhs->aux->offload->netdev;
729 up_read(&bpf_devs_lock);
734 bool bpf_offload_prog_map_match(struct bpf_prog *prog, struct bpf_map *map)
736 struct bpf_offloaded_map *offmap;
739 if (!bpf_map_is_offloaded(map))
740 return bpf_map_offload_neutral(map);
741 offmap = map_to_offmap(map);
743 down_read(&bpf_devs_lock);
744 ret = __bpf_offload_dev_match(prog, offmap->netdev);
745 up_read(&bpf_devs_lock);
750 int bpf_offload_dev_netdev_register(struct bpf_offload_dev *offdev,
751 struct net_device *netdev)
755 down_write(&bpf_devs_lock);
756 err = __bpf_offload_dev_netdev_register(offdev, netdev);
757 up_write(&bpf_devs_lock);
760 EXPORT_SYMBOL_GPL(bpf_offload_dev_netdev_register);
762 void bpf_offload_dev_netdev_unregister(struct bpf_offload_dev *offdev,
763 struct net_device *netdev)
765 down_write(&bpf_devs_lock);
766 __bpf_offload_dev_netdev_unregister(offdev, netdev);
767 up_write(&bpf_devs_lock);
769 EXPORT_SYMBOL_GPL(bpf_offload_dev_netdev_unregister);
771 struct bpf_offload_dev *
772 bpf_offload_dev_create(const struct bpf_prog_offload_ops *ops, void *priv)
774 struct bpf_offload_dev *offdev;
776 offdev = kzalloc(sizeof(*offdev), GFP_KERNEL);
778 return ERR_PTR(-ENOMEM);
782 INIT_LIST_HEAD(&offdev->netdevs);
786 EXPORT_SYMBOL_GPL(bpf_offload_dev_create);
788 void bpf_offload_dev_destroy(struct bpf_offload_dev *offdev)
790 WARN_ON(!list_empty(&offdev->netdevs));
793 EXPORT_SYMBOL_GPL(bpf_offload_dev_destroy);
795 void *bpf_offload_dev_priv(struct bpf_offload_dev *offdev)
799 EXPORT_SYMBOL_GPL(bpf_offload_dev_priv);
801 void bpf_dev_bound_netdev_unregister(struct net_device *dev)
803 struct bpf_offload_netdev *ondev;
807 down_write(&bpf_devs_lock);
808 ondev = bpf_offload_find_netdev(dev);
809 if (ondev && !ondev->offdev)
810 __bpf_offload_dev_netdev_unregister(NULL, ondev->netdev);
811 up_write(&bpf_devs_lock);
814 int bpf_dev_bound_kfunc_check(struct bpf_verifier_log *log,
815 struct bpf_prog_aux *prog_aux)
817 if (!bpf_prog_is_dev_bound(prog_aux)) {
818 bpf_log(log, "metadata kfuncs require device-bound program\n");
822 if (bpf_prog_is_offloaded(prog_aux)) {
823 bpf_log(log, "metadata kfuncs can't be offloaded\n");
830 void *bpf_dev_bound_resolve_kfunc(struct bpf_prog *prog, u32 func_id)
832 const struct xdp_metadata_ops *ops;
835 /* We don't hold bpf_devs_lock while resolving several
836 * kfuncs and can race with the unregister_netdevice().
837 * We rely on bpf_dev_bound_match() check at attach
838 * to render this program unusable.
840 down_read(&bpf_devs_lock);
841 if (!prog->aux->offload)
844 ops = prog->aux->offload->netdev->xdp_metadata_ops;
848 if (func_id == bpf_xdp_metadata_kfunc_id(XDP_METADATA_KFUNC_RX_TIMESTAMP))
849 p = ops->xmo_rx_timestamp;
850 else if (func_id == bpf_xdp_metadata_kfunc_id(XDP_METADATA_KFUNC_RX_HASH))
851 p = ops->xmo_rx_hash;
853 up_read(&bpf_devs_lock);
858 static int __init bpf_offload_init(void)
860 return rhashtable_init(&offdevs, &offdevs_params);
863 core_initcall(bpf_offload_init);