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);
202 /* When program is offloaded require presence of "true"
203 * bpf_offload_netdev, avoid the one created for !ondev case below.
205 if (bpf_prog_is_offloaded(prog->aux) && (!ondev || !ondev->offdev)) {
210 /* When only binding to the device, explicitly
211 * create an entry in the hashtable.
213 err = __bpf_offload_dev_netdev_register(NULL, offload->netdev);
216 ondev = bpf_offload_find_netdev(offload->netdev);
218 offload->offdev = ondev->offdev;
219 prog->aux->offload = offload;
220 list_add_tail(&offload->offloads, &ondev->progs);
228 int bpf_prog_dev_bound_init(struct bpf_prog *prog, union bpf_attr *attr)
230 struct net_device *netdev;
233 if (attr->prog_type != BPF_PROG_TYPE_SCHED_CLS &&
234 attr->prog_type != BPF_PROG_TYPE_XDP)
237 if (attr->prog_flags & ~BPF_F_XDP_DEV_BOUND_ONLY)
240 if (attr->prog_type == BPF_PROG_TYPE_SCHED_CLS &&
241 attr->prog_flags & BPF_F_XDP_DEV_BOUND_ONLY)
244 netdev = dev_get_by_index(current->nsproxy->net_ns, attr->prog_ifindex);
248 err = bpf_dev_offload_check(netdev);
252 prog->aux->offload_requested = !(attr->prog_flags & BPF_F_XDP_DEV_BOUND_ONLY);
254 down_write(&bpf_devs_lock);
255 err = __bpf_prog_dev_bound_init(prog, netdev);
256 up_write(&bpf_devs_lock);
263 int bpf_prog_dev_bound_inherit(struct bpf_prog *new_prog, struct bpf_prog *old_prog)
267 if (!bpf_prog_is_dev_bound(old_prog->aux))
270 if (bpf_prog_is_offloaded(old_prog->aux))
273 new_prog->aux->dev_bound = old_prog->aux->dev_bound;
274 new_prog->aux->offload_requested = old_prog->aux->offload_requested;
276 down_write(&bpf_devs_lock);
277 if (!old_prog->aux->offload) {
282 err = __bpf_prog_dev_bound_init(new_prog, old_prog->aux->offload->netdev);
285 up_write(&bpf_devs_lock);
289 int bpf_prog_offload_verifier_prep(struct bpf_prog *prog)
291 struct bpf_prog_offload *offload;
294 down_read(&bpf_devs_lock);
295 offload = prog->aux->offload;
297 ret = offload->offdev->ops->prepare(prog);
298 offload->dev_state = !ret;
300 up_read(&bpf_devs_lock);
305 int bpf_prog_offload_verify_insn(struct bpf_verifier_env *env,
306 int insn_idx, int prev_insn_idx)
308 struct bpf_prog_offload *offload;
311 down_read(&bpf_devs_lock);
312 offload = env->prog->aux->offload;
314 ret = offload->offdev->ops->insn_hook(env, insn_idx,
316 up_read(&bpf_devs_lock);
321 int bpf_prog_offload_finalize(struct bpf_verifier_env *env)
323 struct bpf_prog_offload *offload;
326 down_read(&bpf_devs_lock);
327 offload = env->prog->aux->offload;
329 if (offload->offdev->ops->finalize)
330 ret = offload->offdev->ops->finalize(env);
334 up_read(&bpf_devs_lock);
340 bpf_prog_offload_replace_insn(struct bpf_verifier_env *env, u32 off,
341 struct bpf_insn *insn)
343 const struct bpf_prog_offload_ops *ops;
344 struct bpf_prog_offload *offload;
345 int ret = -EOPNOTSUPP;
347 down_read(&bpf_devs_lock);
348 offload = env->prog->aux->offload;
350 ops = offload->offdev->ops;
351 if (!offload->opt_failed && ops->replace_insn)
352 ret = ops->replace_insn(env, off, insn);
353 offload->opt_failed |= ret;
355 up_read(&bpf_devs_lock);
359 bpf_prog_offload_remove_insns(struct bpf_verifier_env *env, u32 off, u32 cnt)
361 struct bpf_prog_offload *offload;
362 int ret = -EOPNOTSUPP;
364 down_read(&bpf_devs_lock);
365 offload = env->prog->aux->offload;
367 if (!offload->opt_failed && offload->offdev->ops->remove_insns)
368 ret = offload->offdev->ops->remove_insns(env, off, cnt);
369 offload->opt_failed |= ret;
371 up_read(&bpf_devs_lock);
374 void bpf_prog_dev_bound_destroy(struct bpf_prog *prog)
376 struct bpf_offload_netdev *ondev;
377 struct net_device *netdev;
380 down_write(&bpf_devs_lock);
381 if (prog->aux->offload) {
382 list_del_init(&prog->aux->offload->offloads);
384 netdev = prog->aux->offload->netdev;
385 __bpf_prog_offload_destroy(prog);
387 ondev = bpf_offload_find_netdev(netdev);
388 if (!ondev->offdev && list_empty(&ondev->progs))
389 __bpf_offload_dev_netdev_unregister(NULL, netdev);
391 up_write(&bpf_devs_lock);
395 static int bpf_prog_offload_translate(struct bpf_prog *prog)
397 struct bpf_prog_offload *offload;
400 down_read(&bpf_devs_lock);
401 offload = prog->aux->offload;
403 ret = offload->offdev->ops->translate(prog);
404 up_read(&bpf_devs_lock);
409 static unsigned int bpf_prog_warn_on_exec(const void *ctx,
410 const struct bpf_insn *insn)
412 WARN(1, "attempt to execute device eBPF program on the host!");
416 int bpf_prog_offload_compile(struct bpf_prog *prog)
418 prog->bpf_func = bpf_prog_warn_on_exec;
420 return bpf_prog_offload_translate(prog);
423 struct ns_get_path_bpf_prog_args {
424 struct bpf_prog *prog;
425 struct bpf_prog_info *info;
428 static struct ns_common *bpf_prog_offload_info_fill_ns(void *private_data)
430 struct ns_get_path_bpf_prog_args *args = private_data;
431 struct bpf_prog_aux *aux = args->prog->aux;
432 struct ns_common *ns;
436 down_read(&bpf_devs_lock);
439 args->info->ifindex = aux->offload->netdev->ifindex;
440 net = dev_net(aux->offload->netdev);
444 args->info->ifindex = 0;
448 up_read(&bpf_devs_lock);
454 int bpf_prog_offload_info_fill(struct bpf_prog_info *info,
455 struct bpf_prog *prog)
457 struct ns_get_path_bpf_prog_args args = {
461 struct bpf_prog_aux *aux = prog->aux;
462 struct inode *ns_inode;
468 res = ns_get_path_cb(&ns_path, bpf_prog_offload_info_fill_ns, &args);
475 down_read(&bpf_devs_lock);
478 up_read(&bpf_devs_lock);
482 ulen = info->jited_prog_len;
483 info->jited_prog_len = aux->offload->jited_len;
484 if (info->jited_prog_len && ulen) {
485 uinsns = u64_to_user_ptr(info->jited_prog_insns);
486 ulen = min_t(u32, info->jited_prog_len, ulen);
487 if (copy_to_user(uinsns, aux->offload->jited_image, ulen)) {
488 up_read(&bpf_devs_lock);
493 up_read(&bpf_devs_lock);
495 ns_inode = ns_path.dentry->d_inode;
496 info->netns_dev = new_encode_dev(ns_inode->i_sb->s_dev);
497 info->netns_ino = ns_inode->i_ino;
503 const struct bpf_prog_ops bpf_offload_prog_ops = {
506 struct bpf_map *bpf_map_offload_map_alloc(union bpf_attr *attr)
508 struct net *net = current->nsproxy->net_ns;
509 struct bpf_offload_netdev *ondev;
510 struct bpf_offloaded_map *offmap;
513 if (!capable(CAP_SYS_ADMIN))
514 return ERR_PTR(-EPERM);
515 if (attr->map_type != BPF_MAP_TYPE_ARRAY &&
516 attr->map_type != BPF_MAP_TYPE_HASH)
517 return ERR_PTR(-EINVAL);
519 offmap = bpf_map_area_alloc(sizeof(*offmap), NUMA_NO_NODE);
521 return ERR_PTR(-ENOMEM);
523 bpf_map_init_from_attr(&offmap->map, attr);
526 down_write(&bpf_devs_lock);
527 offmap->netdev = __dev_get_by_index(net, attr->map_ifindex);
528 err = bpf_dev_offload_check(offmap->netdev);
532 ondev = bpf_offload_find_netdev(offmap->netdev);
538 err = bpf_map_offload_ndo(offmap, BPF_OFFLOAD_MAP_ALLOC);
542 list_add_tail(&offmap->offloads, &ondev->maps);
543 up_write(&bpf_devs_lock);
549 up_write(&bpf_devs_lock);
551 bpf_map_area_free(offmap);
555 void bpf_map_offload_map_free(struct bpf_map *map)
557 struct bpf_offloaded_map *offmap = map_to_offmap(map);
560 down_write(&bpf_devs_lock);
562 __bpf_map_offload_destroy(offmap);
563 up_write(&bpf_devs_lock);
566 bpf_map_area_free(offmap);
569 u64 bpf_map_offload_map_mem_usage(const struct bpf_map *map)
571 /* The memory dynamically allocated in netdev dev_ops is not counted */
572 return sizeof(struct bpf_offloaded_map);
575 int bpf_map_offload_lookup_elem(struct bpf_map *map, void *key, void *value)
577 struct bpf_offloaded_map *offmap = map_to_offmap(map);
580 down_read(&bpf_devs_lock);
582 ret = offmap->dev_ops->map_lookup_elem(offmap, key, value);
583 up_read(&bpf_devs_lock);
588 int bpf_map_offload_update_elem(struct bpf_map *map,
589 void *key, void *value, u64 flags)
591 struct bpf_offloaded_map *offmap = map_to_offmap(map);
594 if (unlikely(flags > BPF_EXIST))
597 down_read(&bpf_devs_lock);
599 ret = offmap->dev_ops->map_update_elem(offmap, key, value,
601 up_read(&bpf_devs_lock);
606 int bpf_map_offload_delete_elem(struct bpf_map *map, void *key)
608 struct bpf_offloaded_map *offmap = map_to_offmap(map);
611 down_read(&bpf_devs_lock);
613 ret = offmap->dev_ops->map_delete_elem(offmap, key);
614 up_read(&bpf_devs_lock);
619 int bpf_map_offload_get_next_key(struct bpf_map *map, void *key, void *next_key)
621 struct bpf_offloaded_map *offmap = map_to_offmap(map);
624 down_read(&bpf_devs_lock);
626 ret = offmap->dev_ops->map_get_next_key(offmap, key, next_key);
627 up_read(&bpf_devs_lock);
632 struct ns_get_path_bpf_map_args {
633 struct bpf_offloaded_map *offmap;
634 struct bpf_map_info *info;
637 static struct ns_common *bpf_map_offload_info_fill_ns(void *private_data)
639 struct ns_get_path_bpf_map_args *args = private_data;
640 struct ns_common *ns;
644 down_read(&bpf_devs_lock);
646 if (args->offmap->netdev) {
647 args->info->ifindex = args->offmap->netdev->ifindex;
648 net = dev_net(args->offmap->netdev);
652 args->info->ifindex = 0;
656 up_read(&bpf_devs_lock);
662 int bpf_map_offload_info_fill(struct bpf_map_info *info, struct bpf_map *map)
664 struct ns_get_path_bpf_map_args args = {
665 .offmap = map_to_offmap(map),
668 struct inode *ns_inode;
672 res = ns_get_path_cb(&ns_path, bpf_map_offload_info_fill_ns, &args);
679 ns_inode = ns_path.dentry->d_inode;
680 info->netns_dev = new_encode_dev(ns_inode->i_sb->s_dev);
681 info->netns_ino = ns_inode->i_ino;
687 static bool __bpf_offload_dev_match(struct bpf_prog *prog,
688 struct net_device *netdev)
690 struct bpf_offload_netdev *ondev1, *ondev2;
691 struct bpf_prog_offload *offload;
693 if (!bpf_prog_is_dev_bound(prog->aux))
696 offload = prog->aux->offload;
699 if (offload->netdev == netdev)
702 ondev1 = bpf_offload_find_netdev(offload->netdev);
703 ondev2 = bpf_offload_find_netdev(netdev);
705 return ondev1 && ondev2 && ondev1->offdev == ondev2->offdev;
708 bool bpf_offload_dev_match(struct bpf_prog *prog, struct net_device *netdev)
712 down_read(&bpf_devs_lock);
713 ret = __bpf_offload_dev_match(prog, netdev);
714 up_read(&bpf_devs_lock);
718 EXPORT_SYMBOL_GPL(bpf_offload_dev_match);
720 bool bpf_prog_dev_bound_match(const struct bpf_prog *lhs, const struct bpf_prog *rhs)
724 if (bpf_prog_is_offloaded(lhs->aux) != bpf_prog_is_offloaded(rhs->aux))
727 down_read(&bpf_devs_lock);
728 ret = lhs->aux->offload && rhs->aux->offload &&
729 lhs->aux->offload->netdev &&
730 lhs->aux->offload->netdev == rhs->aux->offload->netdev;
731 up_read(&bpf_devs_lock);
736 bool bpf_offload_prog_map_match(struct bpf_prog *prog, struct bpf_map *map)
738 struct bpf_offloaded_map *offmap;
741 if (!bpf_map_is_offloaded(map))
742 return bpf_map_offload_neutral(map);
743 offmap = map_to_offmap(map);
745 down_read(&bpf_devs_lock);
746 ret = __bpf_offload_dev_match(prog, offmap->netdev);
747 up_read(&bpf_devs_lock);
752 int bpf_offload_dev_netdev_register(struct bpf_offload_dev *offdev,
753 struct net_device *netdev)
757 down_write(&bpf_devs_lock);
758 err = __bpf_offload_dev_netdev_register(offdev, netdev);
759 up_write(&bpf_devs_lock);
762 EXPORT_SYMBOL_GPL(bpf_offload_dev_netdev_register);
764 void bpf_offload_dev_netdev_unregister(struct bpf_offload_dev *offdev,
765 struct net_device *netdev)
767 down_write(&bpf_devs_lock);
768 __bpf_offload_dev_netdev_unregister(offdev, netdev);
769 up_write(&bpf_devs_lock);
771 EXPORT_SYMBOL_GPL(bpf_offload_dev_netdev_unregister);
773 struct bpf_offload_dev *
774 bpf_offload_dev_create(const struct bpf_prog_offload_ops *ops, void *priv)
776 struct bpf_offload_dev *offdev;
778 offdev = kzalloc(sizeof(*offdev), GFP_KERNEL);
780 return ERR_PTR(-ENOMEM);
784 INIT_LIST_HEAD(&offdev->netdevs);
788 EXPORT_SYMBOL_GPL(bpf_offload_dev_create);
790 void bpf_offload_dev_destroy(struct bpf_offload_dev *offdev)
792 WARN_ON(!list_empty(&offdev->netdevs));
795 EXPORT_SYMBOL_GPL(bpf_offload_dev_destroy);
797 void *bpf_offload_dev_priv(struct bpf_offload_dev *offdev)
801 EXPORT_SYMBOL_GPL(bpf_offload_dev_priv);
803 void bpf_dev_bound_netdev_unregister(struct net_device *dev)
805 struct bpf_offload_netdev *ondev;
809 down_write(&bpf_devs_lock);
810 ondev = bpf_offload_find_netdev(dev);
811 if (ondev && !ondev->offdev)
812 __bpf_offload_dev_netdev_unregister(NULL, ondev->netdev);
813 up_write(&bpf_devs_lock);
816 int bpf_dev_bound_kfunc_check(struct bpf_verifier_log *log,
817 struct bpf_prog_aux *prog_aux)
819 if (!bpf_prog_is_dev_bound(prog_aux)) {
820 bpf_log(log, "metadata kfuncs require device-bound program\n");
824 if (bpf_prog_is_offloaded(prog_aux)) {
825 bpf_log(log, "metadata kfuncs can't be offloaded\n");
832 void *bpf_dev_bound_resolve_kfunc(struct bpf_prog *prog, u32 func_id)
834 const struct xdp_metadata_ops *ops;
837 /* We don't hold bpf_devs_lock while resolving several
838 * kfuncs and can race with the unregister_netdevice().
839 * We rely on bpf_dev_bound_match() check at attach
840 * to render this program unusable.
842 down_read(&bpf_devs_lock);
843 if (!prog->aux->offload)
846 ops = prog->aux->offload->netdev->xdp_metadata_ops;
850 if (func_id == bpf_xdp_metadata_kfunc_id(XDP_METADATA_KFUNC_RX_TIMESTAMP))
851 p = ops->xmo_rx_timestamp;
852 else if (func_id == bpf_xdp_metadata_kfunc_id(XDP_METADATA_KFUNC_RX_HASH))
853 p = ops->xmo_rx_hash;
855 up_read(&bpf_devs_lock);
860 static int __init bpf_offload_init(void)
862 return rhashtable_init(&offdevs, &offdevs_params);
865 core_initcall(bpf_offload_init);