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>
29 /* Protects offdevs, members of bpf_offload_netdev and offload members
31 * RTNL lock cannot be taken when holding this lock.
33 static DECLARE_RWSEM(bpf_devs_lock
);
35 struct bpf_offload_dev
{
36 const struct bpf_prog_offload_ops
*ops
;
37 struct list_head netdevs
;
41 struct bpf_offload_netdev
{
43 struct net_device
*netdev
;
44 struct bpf_offload_dev
*offdev
;
45 struct list_head progs
;
46 struct list_head maps
;
47 struct list_head offdev_netdevs
;
50 static const struct rhashtable_params offdevs_params
= {
52 .key_len
= sizeof(struct net_device
*),
53 .key_offset
= offsetof(struct bpf_offload_netdev
, netdev
),
54 .head_offset
= offsetof(struct bpf_offload_netdev
, l
),
55 .automatic_shrinking
= true,
58 static struct rhashtable offdevs
;
59 static bool offdevs_inited
;
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
);
77 return rhashtable_lookup_fast(&offdevs
, &netdev
, offdevs_params
);
80 int bpf_prog_offload_init(struct bpf_prog
*prog
, union bpf_attr
*attr
)
82 struct bpf_offload_netdev
*ondev
;
83 struct bpf_prog_offload
*offload
;
86 if (attr
->prog_type
!= BPF_PROG_TYPE_SCHED_CLS
&&
87 attr
->prog_type
!= BPF_PROG_TYPE_XDP
)
93 offload
= kzalloc(sizeof(*offload
), GFP_USER
);
99 offload
->netdev
= dev_get_by_index(current
->nsproxy
->net_ns
,
101 err
= bpf_dev_offload_check(offload
->netdev
);
105 down_write(&bpf_devs_lock
);
106 ondev
= bpf_offload_find_netdev(offload
->netdev
);
111 offload
->offdev
= ondev
->offdev
;
112 prog
->aux
->offload
= offload
;
113 list_add_tail(&offload
->offloads
, &ondev
->progs
);
114 dev_put(offload
->netdev
);
115 up_write(&bpf_devs_lock
);
119 up_write(&bpf_devs_lock
);
122 dev_put(offload
->netdev
);
127 int bpf_prog_offload_verifier_prep(struct bpf_prog
*prog
)
129 struct bpf_prog_offload
*offload
;
132 down_read(&bpf_devs_lock
);
133 offload
= prog
->aux
->offload
;
135 ret
= offload
->offdev
->ops
->prepare(prog
);
136 offload
->dev_state
= !ret
;
138 up_read(&bpf_devs_lock
);
143 int bpf_prog_offload_verify_insn(struct bpf_verifier_env
*env
,
144 int insn_idx
, int prev_insn_idx
)
146 struct bpf_prog_offload
*offload
;
149 down_read(&bpf_devs_lock
);
150 offload
= env
->prog
->aux
->offload
;
152 ret
= offload
->offdev
->ops
->insn_hook(env
, insn_idx
,
154 up_read(&bpf_devs_lock
);
159 int bpf_prog_offload_finalize(struct bpf_verifier_env
*env
)
161 struct bpf_prog_offload
*offload
;
164 down_read(&bpf_devs_lock
);
165 offload
= env
->prog
->aux
->offload
;
167 if (offload
->offdev
->ops
->finalize
)
168 ret
= offload
->offdev
->ops
->finalize(env
);
172 up_read(&bpf_devs_lock
);
178 bpf_prog_offload_replace_insn(struct bpf_verifier_env
*env
, u32 off
,
179 struct bpf_insn
*insn
)
181 const struct bpf_prog_offload_ops
*ops
;
182 struct bpf_prog_offload
*offload
;
183 int ret
= -EOPNOTSUPP
;
185 down_read(&bpf_devs_lock
);
186 offload
= env
->prog
->aux
->offload
;
188 ops
= offload
->offdev
->ops
;
189 if (!offload
->opt_failed
&& ops
->replace_insn
)
190 ret
= ops
->replace_insn(env
, off
, insn
);
191 offload
->opt_failed
|= ret
;
193 up_read(&bpf_devs_lock
);
197 bpf_prog_offload_remove_insns(struct bpf_verifier_env
*env
, u32 off
, u32 cnt
)
199 struct bpf_prog_offload
*offload
;
200 int ret
= -EOPNOTSUPP
;
202 down_read(&bpf_devs_lock
);
203 offload
= env
->prog
->aux
->offload
;
205 if (!offload
->opt_failed
&& offload
->offdev
->ops
->remove_insns
)
206 ret
= offload
->offdev
->ops
->remove_insns(env
, off
, cnt
);
207 offload
->opt_failed
|= ret
;
209 up_read(&bpf_devs_lock
);
212 static void __bpf_prog_offload_destroy(struct bpf_prog
*prog
)
214 struct bpf_prog_offload
*offload
= prog
->aux
->offload
;
216 if (offload
->dev_state
)
217 offload
->offdev
->ops
->destroy(prog
);
219 /* Make sure BPF_PROG_GET_NEXT_ID can't find this dead program */
220 bpf_prog_free_id(prog
, true);
222 list_del_init(&offload
->offloads
);
224 prog
->aux
->offload
= NULL
;
227 void bpf_prog_offload_destroy(struct bpf_prog
*prog
)
229 down_write(&bpf_devs_lock
);
230 if (prog
->aux
->offload
)
231 __bpf_prog_offload_destroy(prog
);
232 up_write(&bpf_devs_lock
);
235 static int bpf_prog_offload_translate(struct bpf_prog
*prog
)
237 struct bpf_prog_offload
*offload
;
240 down_read(&bpf_devs_lock
);
241 offload
= prog
->aux
->offload
;
243 ret
= offload
->offdev
->ops
->translate(prog
);
244 up_read(&bpf_devs_lock
);
249 static unsigned int bpf_prog_warn_on_exec(const void *ctx
,
250 const struct bpf_insn
*insn
)
252 WARN(1, "attempt to execute device eBPF program on the host!");
256 int bpf_prog_offload_compile(struct bpf_prog
*prog
)
258 prog
->bpf_func
= bpf_prog_warn_on_exec
;
260 return bpf_prog_offload_translate(prog
);
263 struct ns_get_path_bpf_prog_args
{
264 struct bpf_prog
*prog
;
265 struct bpf_prog_info
*info
;
268 static struct ns_common
*bpf_prog_offload_info_fill_ns(void *private_data
)
270 struct ns_get_path_bpf_prog_args
*args
= private_data
;
271 struct bpf_prog_aux
*aux
= args
->prog
->aux
;
272 struct ns_common
*ns
;
276 down_read(&bpf_devs_lock
);
279 args
->info
->ifindex
= aux
->offload
->netdev
->ifindex
;
280 net
= dev_net(aux
->offload
->netdev
);
284 args
->info
->ifindex
= 0;
288 up_read(&bpf_devs_lock
);
294 int bpf_prog_offload_info_fill(struct bpf_prog_info
*info
,
295 struct bpf_prog
*prog
)
297 struct ns_get_path_bpf_prog_args args
= {
301 struct bpf_prog_aux
*aux
= prog
->aux
;
302 struct inode
*ns_inode
;
308 res
= ns_get_path_cb(&ns_path
, bpf_prog_offload_info_fill_ns
, &args
);
315 down_read(&bpf_devs_lock
);
318 up_read(&bpf_devs_lock
);
322 ulen
= info
->jited_prog_len
;
323 info
->jited_prog_len
= aux
->offload
->jited_len
;
324 if (info
->jited_prog_len
& ulen
) {
325 uinsns
= u64_to_user_ptr(info
->jited_prog_insns
);
326 ulen
= min_t(u32
, info
->jited_prog_len
, ulen
);
327 if (copy_to_user(uinsns
, aux
->offload
->jited_image
, ulen
)) {
328 up_read(&bpf_devs_lock
);
333 up_read(&bpf_devs_lock
);
335 ns_inode
= ns_path
.dentry
->d_inode
;
336 info
->netns_dev
= new_encode_dev(ns_inode
->i_sb
->s_dev
);
337 info
->netns_ino
= ns_inode
->i_ino
;
343 const struct bpf_prog_ops bpf_offload_prog_ops
= {
346 static int bpf_map_offload_ndo(struct bpf_offloaded_map
*offmap
,
347 enum bpf_netdev_command cmd
)
349 struct netdev_bpf data
= {};
350 struct net_device
*netdev
;
355 data
.offmap
= offmap
;
356 /* Caller must make sure netdev is valid */
357 netdev
= offmap
->netdev
;
359 return netdev
->netdev_ops
->ndo_bpf(netdev
, &data
);
362 struct bpf_map
*bpf_map_offload_map_alloc(union bpf_attr
*attr
)
364 struct net
*net
= current
->nsproxy
->net_ns
;
365 struct bpf_offload_netdev
*ondev
;
366 struct bpf_offloaded_map
*offmap
;
369 if (!capable(CAP_SYS_ADMIN
))
370 return ERR_PTR(-EPERM
);
371 if (attr
->map_type
!= BPF_MAP_TYPE_ARRAY
&&
372 attr
->map_type
!= BPF_MAP_TYPE_HASH
)
373 return ERR_PTR(-EINVAL
);
375 offmap
= kzalloc(sizeof(*offmap
), GFP_USER
);
377 return ERR_PTR(-ENOMEM
);
379 bpf_map_init_from_attr(&offmap
->map
, attr
);
382 down_write(&bpf_devs_lock
);
383 offmap
->netdev
= __dev_get_by_index(net
, attr
->map_ifindex
);
384 err
= bpf_dev_offload_check(offmap
->netdev
);
388 ondev
= bpf_offload_find_netdev(offmap
->netdev
);
394 err
= bpf_map_offload_ndo(offmap
, BPF_OFFLOAD_MAP_ALLOC
);
398 list_add_tail(&offmap
->offloads
, &ondev
->maps
);
399 up_write(&bpf_devs_lock
);
405 up_write(&bpf_devs_lock
);
411 static void __bpf_map_offload_destroy(struct bpf_offloaded_map
*offmap
)
413 WARN_ON(bpf_map_offload_ndo(offmap
, BPF_OFFLOAD_MAP_FREE
));
414 /* Make sure BPF_MAP_GET_NEXT_ID can't find this dead map */
415 bpf_map_free_id(&offmap
->map
, true);
416 list_del_init(&offmap
->offloads
);
417 offmap
->netdev
= NULL
;
420 void bpf_map_offload_map_free(struct bpf_map
*map
)
422 struct bpf_offloaded_map
*offmap
= map_to_offmap(map
);
425 down_write(&bpf_devs_lock
);
427 __bpf_map_offload_destroy(offmap
);
428 up_write(&bpf_devs_lock
);
434 int bpf_map_offload_lookup_elem(struct bpf_map
*map
, void *key
, void *value
)
436 struct bpf_offloaded_map
*offmap
= map_to_offmap(map
);
439 down_read(&bpf_devs_lock
);
441 ret
= offmap
->dev_ops
->map_lookup_elem(offmap
, key
, value
);
442 up_read(&bpf_devs_lock
);
447 int bpf_map_offload_update_elem(struct bpf_map
*map
,
448 void *key
, void *value
, u64 flags
)
450 struct bpf_offloaded_map
*offmap
= map_to_offmap(map
);
453 if (unlikely(flags
> BPF_EXIST
))
456 down_read(&bpf_devs_lock
);
458 ret
= offmap
->dev_ops
->map_update_elem(offmap
, key
, value
,
460 up_read(&bpf_devs_lock
);
465 int bpf_map_offload_delete_elem(struct bpf_map
*map
, void *key
)
467 struct bpf_offloaded_map
*offmap
= map_to_offmap(map
);
470 down_read(&bpf_devs_lock
);
472 ret
= offmap
->dev_ops
->map_delete_elem(offmap
, key
);
473 up_read(&bpf_devs_lock
);
478 int bpf_map_offload_get_next_key(struct bpf_map
*map
, void *key
, void *next_key
)
480 struct bpf_offloaded_map
*offmap
= map_to_offmap(map
);
483 down_read(&bpf_devs_lock
);
485 ret
= offmap
->dev_ops
->map_get_next_key(offmap
, key
, next_key
);
486 up_read(&bpf_devs_lock
);
491 struct ns_get_path_bpf_map_args
{
492 struct bpf_offloaded_map
*offmap
;
493 struct bpf_map_info
*info
;
496 static struct ns_common
*bpf_map_offload_info_fill_ns(void *private_data
)
498 struct ns_get_path_bpf_map_args
*args
= private_data
;
499 struct ns_common
*ns
;
503 down_read(&bpf_devs_lock
);
505 if (args
->offmap
->netdev
) {
506 args
->info
->ifindex
= args
->offmap
->netdev
->ifindex
;
507 net
= dev_net(args
->offmap
->netdev
);
511 args
->info
->ifindex
= 0;
515 up_read(&bpf_devs_lock
);
521 int bpf_map_offload_info_fill(struct bpf_map_info
*info
, struct bpf_map
*map
)
523 struct ns_get_path_bpf_map_args args
= {
524 .offmap
= map_to_offmap(map
),
527 struct inode
*ns_inode
;
531 res
= ns_get_path_cb(&ns_path
, bpf_map_offload_info_fill_ns
, &args
);
538 ns_inode
= ns_path
.dentry
->d_inode
;
539 info
->netns_dev
= new_encode_dev(ns_inode
->i_sb
->s_dev
);
540 info
->netns_ino
= ns_inode
->i_ino
;
546 static bool __bpf_offload_dev_match(struct bpf_prog
*prog
,
547 struct net_device
*netdev
)
549 struct bpf_offload_netdev
*ondev1
, *ondev2
;
550 struct bpf_prog_offload
*offload
;
552 if (!bpf_prog_is_dev_bound(prog
->aux
))
555 offload
= prog
->aux
->offload
;
558 if (offload
->netdev
== netdev
)
561 ondev1
= bpf_offload_find_netdev(offload
->netdev
);
562 ondev2
= bpf_offload_find_netdev(netdev
);
564 return ondev1
&& ondev2
&& ondev1
->offdev
== ondev2
->offdev
;
567 bool bpf_offload_dev_match(struct bpf_prog
*prog
, struct net_device
*netdev
)
571 down_read(&bpf_devs_lock
);
572 ret
= __bpf_offload_dev_match(prog
, netdev
);
573 up_read(&bpf_devs_lock
);
577 EXPORT_SYMBOL_GPL(bpf_offload_dev_match
);
579 bool bpf_offload_prog_map_match(struct bpf_prog
*prog
, struct bpf_map
*map
)
581 struct bpf_offloaded_map
*offmap
;
584 if (!bpf_map_is_dev_bound(map
))
585 return bpf_map_offload_neutral(map
);
586 offmap
= map_to_offmap(map
);
588 down_read(&bpf_devs_lock
);
589 ret
= __bpf_offload_dev_match(prog
, offmap
->netdev
);
590 up_read(&bpf_devs_lock
);
595 int bpf_offload_dev_netdev_register(struct bpf_offload_dev
*offdev
,
596 struct net_device
*netdev
)
598 struct bpf_offload_netdev
*ondev
;
601 ondev
= kzalloc(sizeof(*ondev
), GFP_KERNEL
);
605 ondev
->netdev
= netdev
;
606 ondev
->offdev
= offdev
;
607 INIT_LIST_HEAD(&ondev
->progs
);
608 INIT_LIST_HEAD(&ondev
->maps
);
610 down_write(&bpf_devs_lock
);
611 err
= rhashtable_insert_fast(&offdevs
, &ondev
->l
, offdevs_params
);
613 netdev_warn(netdev
, "failed to register for BPF offload\n");
614 goto err_unlock_free
;
617 list_add(&ondev
->offdev_netdevs
, &offdev
->netdevs
);
618 up_write(&bpf_devs_lock
);
622 up_write(&bpf_devs_lock
);
626 EXPORT_SYMBOL_GPL(bpf_offload_dev_netdev_register
);
628 void bpf_offload_dev_netdev_unregister(struct bpf_offload_dev
*offdev
,
629 struct net_device
*netdev
)
631 struct bpf_offload_netdev
*ondev
, *altdev
;
632 struct bpf_offloaded_map
*offmap
, *mtmp
;
633 struct bpf_prog_offload
*offload
, *ptmp
;
637 down_write(&bpf_devs_lock
);
638 ondev
= rhashtable_lookup_fast(&offdevs
, &netdev
, offdevs_params
);
642 WARN_ON(rhashtable_remove_fast(&offdevs
, &ondev
->l
, offdevs_params
));
643 list_del(&ondev
->offdev_netdevs
);
645 /* Try to move the objects to another netdev of the device */
646 altdev
= list_first_entry_or_null(&offdev
->netdevs
,
647 struct bpf_offload_netdev
,
650 list_for_each_entry(offload
, &ondev
->progs
, offloads
)
651 offload
->netdev
= altdev
->netdev
;
652 list_splice_init(&ondev
->progs
, &altdev
->progs
);
654 list_for_each_entry(offmap
, &ondev
->maps
, offloads
)
655 offmap
->netdev
= altdev
->netdev
;
656 list_splice_init(&ondev
->maps
, &altdev
->maps
);
658 list_for_each_entry_safe(offload
, ptmp
, &ondev
->progs
, offloads
)
659 __bpf_prog_offload_destroy(offload
->prog
);
660 list_for_each_entry_safe(offmap
, mtmp
, &ondev
->maps
, offloads
)
661 __bpf_map_offload_destroy(offmap
);
664 WARN_ON(!list_empty(&ondev
->progs
));
665 WARN_ON(!list_empty(&ondev
->maps
));
668 up_write(&bpf_devs_lock
);
670 EXPORT_SYMBOL_GPL(bpf_offload_dev_netdev_unregister
);
672 struct bpf_offload_dev
*
673 bpf_offload_dev_create(const struct bpf_prog_offload_ops
*ops
, void *priv
)
675 struct bpf_offload_dev
*offdev
;
678 down_write(&bpf_devs_lock
);
679 if (!offdevs_inited
) {
680 err
= rhashtable_init(&offdevs
, &offdevs_params
);
683 offdevs_inited
= true;
685 up_write(&bpf_devs_lock
);
687 offdev
= kzalloc(sizeof(*offdev
), GFP_KERNEL
);
689 return ERR_PTR(-ENOMEM
);
693 INIT_LIST_HEAD(&offdev
->netdevs
);
697 EXPORT_SYMBOL_GPL(bpf_offload_dev_create
);
699 void bpf_offload_dev_destroy(struct bpf_offload_dev
*offdev
)
701 WARN_ON(!list_empty(&offdev
->netdevs
));
704 EXPORT_SYMBOL_GPL(bpf_offload_dev_destroy
);
706 void *bpf_offload_dev_priv(struct bpf_offload_dev
*offdev
)
710 EXPORT_SYMBOL_GPL(bpf_offload_dev_priv
);