1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright (c) 2017 - 2018 Covalent IO, Inc. http://covalent.io */
5 #include <linux/btf_ids.h>
6 #include <linux/filter.h>
7 #include <linux/errno.h>
8 #include <linux/file.h>
10 #include <linux/workqueue.h>
11 #include <linux/skmsg.h>
12 #include <linux/list.h>
13 #include <linux/jhash.h>
14 #include <linux/sock_diag.h>
20 struct sk_psock_progs progs;
24 #define SOCK_CREATE_FLAG_MASK \
25 (BPF_F_NUMA_NODE | BPF_F_RDONLY | BPF_F_WRONLY)
27 static int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog,
28 struct bpf_prog *old, u32 which);
29 static struct sk_psock_progs *sock_map_progs(struct bpf_map *map);
31 static struct bpf_map *sock_map_alloc(union bpf_attr *attr)
33 struct bpf_stab *stab;
35 if (!capable(CAP_NET_ADMIN))
36 return ERR_PTR(-EPERM);
37 if (attr->max_entries == 0 ||
38 attr->key_size != 4 ||
39 (attr->value_size != sizeof(u32) &&
40 attr->value_size != sizeof(u64)) ||
41 attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
42 return ERR_PTR(-EINVAL);
44 stab = bpf_map_area_alloc(sizeof(*stab), NUMA_NO_NODE);
46 return ERR_PTR(-ENOMEM);
48 bpf_map_init_from_attr(&stab->map, attr);
49 raw_spin_lock_init(&stab->lock);
51 stab->sks = bpf_map_area_alloc((u64) stab->map.max_entries *
52 sizeof(struct sock *),
55 bpf_map_area_free(stab);
56 return ERR_PTR(-ENOMEM);
62 int sock_map_get_from_fd(const union bpf_attr *attr, struct bpf_prog *prog)
64 u32 ufd = attr->target_fd;
69 if (attr->attach_flags || attr->replace_bpf_fd)
73 map = __bpf_map_get(f);
76 ret = sock_map_prog_update(map, prog, NULL, attr->attach_type);
81 int sock_map_prog_detach(const union bpf_attr *attr, enum bpf_prog_type ptype)
83 u32 ufd = attr->target_fd;
84 struct bpf_prog *prog;
89 if (attr->attach_flags || attr->replace_bpf_fd)
93 map = __bpf_map_get(f);
97 prog = bpf_prog_get(attr->attach_bpf_fd);
103 if (prog->type != ptype) {
108 ret = sock_map_prog_update(map, NULL, prog, attr->attach_type);
116 static void sock_map_sk_acquire(struct sock *sk)
117 __acquires(&sk->sk_lock.slock)
124 static void sock_map_sk_release(struct sock *sk)
125 __releases(&sk->sk_lock.slock)
132 static void sock_map_add_link(struct sk_psock *psock,
133 struct sk_psock_link *link,
134 struct bpf_map *map, void *link_raw)
136 link->link_raw = link_raw;
138 spin_lock_bh(&psock->link_lock);
139 list_add_tail(&link->list, &psock->link);
140 spin_unlock_bh(&psock->link_lock);
143 static void sock_map_del_link(struct sock *sk,
144 struct sk_psock *psock, void *link_raw)
146 bool strp_stop = false, verdict_stop = false;
147 struct sk_psock_link *link, *tmp;
149 spin_lock_bh(&psock->link_lock);
150 list_for_each_entry_safe(link, tmp, &psock->link, list) {
151 if (link->link_raw == link_raw) {
152 struct bpf_map *map = link->map;
153 struct bpf_stab *stab = container_of(map, struct bpf_stab,
155 if (psock->saved_data_ready && stab->progs.stream_parser)
157 if (psock->saved_data_ready && stab->progs.stream_verdict)
159 if (psock->saved_data_ready && stab->progs.skb_verdict)
161 list_del(&link->list);
162 sk_psock_free_link(link);
165 spin_unlock_bh(&psock->link_lock);
166 if (strp_stop || verdict_stop) {
167 write_lock_bh(&sk->sk_callback_lock);
169 sk_psock_stop_strp(sk, psock);
171 sk_psock_stop_verdict(sk, psock);
173 if (psock->psock_update_sk_prot)
174 psock->psock_update_sk_prot(sk, psock, false);
175 write_unlock_bh(&sk->sk_callback_lock);
179 static void sock_map_unref(struct sock *sk, void *link_raw)
181 struct sk_psock *psock = sk_psock(sk);
184 sock_map_del_link(sk, psock, link_raw);
185 sk_psock_put(sk, psock);
189 static int sock_map_init_proto(struct sock *sk, struct sk_psock *psock)
191 if (!sk->sk_prot->psock_update_sk_prot)
193 psock->psock_update_sk_prot = sk->sk_prot->psock_update_sk_prot;
194 return sk->sk_prot->psock_update_sk_prot(sk, psock, false);
197 static struct sk_psock *sock_map_psock_get_checked(struct sock *sk)
199 struct sk_psock *psock;
202 psock = sk_psock(sk);
204 if (sk->sk_prot->close != sock_map_close) {
205 psock = ERR_PTR(-EBUSY);
209 if (!refcount_inc_not_zero(&psock->refcnt))
210 psock = ERR_PTR(-EBUSY);
217 static int sock_map_link(struct bpf_map *map, struct sock *sk)
219 struct sk_psock_progs *progs = sock_map_progs(map);
220 struct bpf_prog *stream_verdict = NULL;
221 struct bpf_prog *stream_parser = NULL;
222 struct bpf_prog *skb_verdict = NULL;
223 struct bpf_prog *msg_parser = NULL;
224 struct sk_psock *psock;
227 stream_verdict = READ_ONCE(progs->stream_verdict);
228 if (stream_verdict) {
229 stream_verdict = bpf_prog_inc_not_zero(stream_verdict);
230 if (IS_ERR(stream_verdict))
231 return PTR_ERR(stream_verdict);
234 stream_parser = READ_ONCE(progs->stream_parser);
236 stream_parser = bpf_prog_inc_not_zero(stream_parser);
237 if (IS_ERR(stream_parser)) {
238 ret = PTR_ERR(stream_parser);
239 goto out_put_stream_verdict;
243 msg_parser = READ_ONCE(progs->msg_parser);
245 msg_parser = bpf_prog_inc_not_zero(msg_parser);
246 if (IS_ERR(msg_parser)) {
247 ret = PTR_ERR(msg_parser);
248 goto out_put_stream_parser;
252 skb_verdict = READ_ONCE(progs->skb_verdict);
254 skb_verdict = bpf_prog_inc_not_zero(skb_verdict);
255 if (IS_ERR(skb_verdict)) {
256 ret = PTR_ERR(skb_verdict);
257 goto out_put_msg_parser;
261 psock = sock_map_psock_get_checked(sk);
263 ret = PTR_ERR(psock);
268 if ((msg_parser && READ_ONCE(psock->progs.msg_parser)) ||
269 (stream_parser && READ_ONCE(psock->progs.stream_parser)) ||
270 (skb_verdict && READ_ONCE(psock->progs.skb_verdict)) ||
271 (skb_verdict && READ_ONCE(psock->progs.stream_verdict)) ||
272 (stream_verdict && READ_ONCE(psock->progs.skb_verdict)) ||
273 (stream_verdict && READ_ONCE(psock->progs.stream_verdict))) {
274 sk_psock_put(sk, psock);
279 psock = sk_psock_init(sk, map->numa_node);
281 ret = PTR_ERR(psock);
287 psock_set_prog(&psock->progs.msg_parser, msg_parser);
289 psock_set_prog(&psock->progs.stream_parser, stream_parser);
291 psock_set_prog(&psock->progs.stream_verdict, stream_verdict);
293 psock_set_prog(&psock->progs.skb_verdict, skb_verdict);
295 /* msg_* and stream_* programs references tracked in psock after this
296 * point. Reference dec and cleanup will occur through psock destructor
298 ret = sock_map_init_proto(sk, psock);
300 sk_psock_put(sk, psock);
304 write_lock_bh(&sk->sk_callback_lock);
305 if (stream_parser && stream_verdict && !psock->saved_data_ready) {
306 ret = sk_psock_init_strp(sk, psock);
308 write_unlock_bh(&sk->sk_callback_lock);
309 sk_psock_put(sk, psock);
312 sk_psock_start_strp(sk, psock);
313 } else if (!stream_parser && stream_verdict && !psock->saved_data_ready) {
314 sk_psock_start_verdict(sk,psock);
315 } else if (!stream_verdict && skb_verdict && !psock->saved_data_ready) {
316 sk_psock_start_verdict(sk, psock);
318 write_unlock_bh(&sk->sk_callback_lock);
322 bpf_prog_put(skb_verdict);
325 bpf_prog_put(msg_parser);
326 out_put_stream_parser:
328 bpf_prog_put(stream_parser);
329 out_put_stream_verdict:
331 bpf_prog_put(stream_verdict);
336 static void sock_map_free(struct bpf_map *map)
338 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
341 /* After the sync no updates or deletes will be in-flight so it
342 * is safe to walk map and remove entries without risking a race
343 * in EEXIST update case.
346 for (i = 0; i < stab->map.max_entries; i++) {
347 struct sock **psk = &stab->sks[i];
350 sk = xchg(psk, NULL);
355 sock_map_unref(sk, psk);
362 /* wait for psock readers accessing its map link */
365 bpf_map_area_free(stab->sks);
366 bpf_map_area_free(stab);
369 static void sock_map_release_progs(struct bpf_map *map)
371 psock_progs_drop(&container_of(map, struct bpf_stab, map)->progs);
374 static struct sock *__sock_map_lookup_elem(struct bpf_map *map, u32 key)
376 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
378 WARN_ON_ONCE(!rcu_read_lock_held());
380 if (unlikely(key >= map->max_entries))
382 return READ_ONCE(stab->sks[key]);
385 static void *sock_map_lookup(struct bpf_map *map, void *key)
389 sk = __sock_map_lookup_elem(map, *(u32 *)key);
392 if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
397 static void *sock_map_lookup_sys(struct bpf_map *map, void *key)
401 if (map->value_size != sizeof(u64))
402 return ERR_PTR(-ENOSPC);
404 sk = __sock_map_lookup_elem(map, *(u32 *)key);
406 return ERR_PTR(-ENOENT);
408 __sock_gen_cookie(sk);
409 return &sk->sk_cookie;
412 static int __sock_map_delete(struct bpf_stab *stab, struct sock *sk_test,
418 raw_spin_lock_bh(&stab->lock);
420 if (!sk_test || sk_test == sk)
421 sk = xchg(psk, NULL);
424 sock_map_unref(sk, psk);
428 raw_spin_unlock_bh(&stab->lock);
432 static void sock_map_delete_from_link(struct bpf_map *map, struct sock *sk,
435 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
437 __sock_map_delete(stab, sk, link_raw);
440 static int sock_map_delete_elem(struct bpf_map *map, void *key)
442 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
446 if (unlikely(i >= map->max_entries))
450 return __sock_map_delete(stab, NULL, psk);
453 static int sock_map_get_next_key(struct bpf_map *map, void *key, void *next)
455 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
456 u32 i = key ? *(u32 *)key : U32_MAX;
457 u32 *key_next = next;
459 if (i == stab->map.max_entries - 1)
461 if (i >= stab->map.max_entries)
468 static int sock_map_update_common(struct bpf_map *map, u32 idx,
469 struct sock *sk, u64 flags)
471 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
472 struct sk_psock_link *link;
473 struct sk_psock *psock;
477 WARN_ON_ONCE(!rcu_read_lock_held());
478 if (unlikely(flags > BPF_EXIST))
480 if (unlikely(idx >= map->max_entries))
483 link = sk_psock_init_link();
487 ret = sock_map_link(map, sk);
491 psock = sk_psock(sk);
492 WARN_ON_ONCE(!psock);
494 raw_spin_lock_bh(&stab->lock);
495 osk = stab->sks[idx];
496 if (osk && flags == BPF_NOEXIST) {
499 } else if (!osk && flags == BPF_EXIST) {
504 sock_map_add_link(psock, link, map, &stab->sks[idx]);
507 sock_map_unref(osk, &stab->sks[idx]);
508 raw_spin_unlock_bh(&stab->lock);
511 raw_spin_unlock_bh(&stab->lock);
513 sk_psock_put(sk, psock);
515 sk_psock_free_link(link);
519 static bool sock_map_op_okay(const struct bpf_sock_ops_kern *ops)
521 return ops->op == BPF_SOCK_OPS_PASSIVE_ESTABLISHED_CB ||
522 ops->op == BPF_SOCK_OPS_ACTIVE_ESTABLISHED_CB ||
523 ops->op == BPF_SOCK_OPS_TCP_LISTEN_CB;
526 static bool sock_map_redirect_allowed(const struct sock *sk)
529 return sk->sk_state != TCP_LISTEN;
531 return sk->sk_state == TCP_ESTABLISHED;
534 static bool sock_map_sk_is_suitable(const struct sock *sk)
536 return !!sk->sk_prot->psock_update_sk_prot;
539 static bool sock_map_sk_state_allowed(const struct sock *sk)
542 return (1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_LISTEN);
546 static int sock_hash_update_common(struct bpf_map *map, void *key,
547 struct sock *sk, u64 flags);
549 int sock_map_update_elem_sys(struct bpf_map *map, void *key, void *value,
557 if (map->value_size == sizeof(u64))
564 sock = sockfd_lookup(ufd, &ret);
572 if (!sock_map_sk_is_suitable(sk)) {
577 sock_map_sk_acquire(sk);
578 if (!sock_map_sk_state_allowed(sk))
580 else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
581 ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
583 ret = sock_hash_update_common(map, key, sk, flags);
584 sock_map_sk_release(sk);
590 static int sock_map_update_elem(struct bpf_map *map, void *key,
591 void *value, u64 flags)
593 struct sock *sk = (struct sock *)value;
596 if (unlikely(!sk || !sk_fullsock(sk)))
599 if (!sock_map_sk_is_suitable(sk))
604 if (!sock_map_sk_state_allowed(sk))
606 else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
607 ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
609 ret = sock_hash_update_common(map, key, sk, flags);
615 BPF_CALL_4(bpf_sock_map_update, struct bpf_sock_ops_kern *, sops,
616 struct bpf_map *, map, void *, key, u64, flags)
618 WARN_ON_ONCE(!rcu_read_lock_held());
620 if (likely(sock_map_sk_is_suitable(sops->sk) &&
621 sock_map_op_okay(sops)))
622 return sock_map_update_common(map, *(u32 *)key, sops->sk,
627 const struct bpf_func_proto bpf_sock_map_update_proto = {
628 .func = bpf_sock_map_update,
631 .ret_type = RET_INTEGER,
632 .arg1_type = ARG_PTR_TO_CTX,
633 .arg2_type = ARG_CONST_MAP_PTR,
634 .arg3_type = ARG_PTR_TO_MAP_KEY,
635 .arg4_type = ARG_ANYTHING,
638 BPF_CALL_4(bpf_sk_redirect_map, struct sk_buff *, skb,
639 struct bpf_map *, map, u32, key, u64, flags)
643 if (unlikely(flags & ~(BPF_F_INGRESS)))
646 sk = __sock_map_lookup_elem(map, key);
647 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
650 skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
654 const struct bpf_func_proto bpf_sk_redirect_map_proto = {
655 .func = bpf_sk_redirect_map,
657 .ret_type = RET_INTEGER,
658 .arg1_type = ARG_PTR_TO_CTX,
659 .arg2_type = ARG_CONST_MAP_PTR,
660 .arg3_type = ARG_ANYTHING,
661 .arg4_type = ARG_ANYTHING,
664 BPF_CALL_4(bpf_msg_redirect_map, struct sk_msg *, msg,
665 struct bpf_map *, map, u32, key, u64, flags)
669 if (unlikely(flags & ~(BPF_F_INGRESS)))
672 sk = __sock_map_lookup_elem(map, key);
673 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
681 const struct bpf_func_proto bpf_msg_redirect_map_proto = {
682 .func = bpf_msg_redirect_map,
684 .ret_type = RET_INTEGER,
685 .arg1_type = ARG_PTR_TO_CTX,
686 .arg2_type = ARG_CONST_MAP_PTR,
687 .arg3_type = ARG_ANYTHING,
688 .arg4_type = ARG_ANYTHING,
691 struct sock_map_seq_info {
697 struct bpf_iter__sockmap {
698 __bpf_md_ptr(struct bpf_iter_meta *, meta);
699 __bpf_md_ptr(struct bpf_map *, map);
700 __bpf_md_ptr(void *, key);
701 __bpf_md_ptr(struct sock *, sk);
704 DEFINE_BPF_ITER_FUNC(sockmap, struct bpf_iter_meta *meta,
705 struct bpf_map *map, void *key,
708 static void *sock_map_seq_lookup_elem(struct sock_map_seq_info *info)
710 if (unlikely(info->index >= info->map->max_entries))
713 info->sk = __sock_map_lookup_elem(info->map, info->index);
715 /* can't return sk directly, since that might be NULL */
719 static void *sock_map_seq_start(struct seq_file *seq, loff_t *pos)
722 struct sock_map_seq_info *info = seq->private;
727 /* pairs with sock_map_seq_stop */
729 return sock_map_seq_lookup_elem(info);
732 static void *sock_map_seq_next(struct seq_file *seq, void *v, loff_t *pos)
735 struct sock_map_seq_info *info = seq->private;
740 return sock_map_seq_lookup_elem(info);
743 static int sock_map_seq_show(struct seq_file *seq, void *v)
746 struct sock_map_seq_info *info = seq->private;
747 struct bpf_iter__sockmap ctx = {};
748 struct bpf_iter_meta meta;
749 struct bpf_prog *prog;
752 prog = bpf_iter_get_info(&meta, !v);
759 ctx.key = &info->index;
763 return bpf_iter_run_prog(prog, &ctx);
766 static void sock_map_seq_stop(struct seq_file *seq, void *v)
770 (void)sock_map_seq_show(seq, NULL);
772 /* pairs with sock_map_seq_start */
776 static const struct seq_operations sock_map_seq_ops = {
777 .start = sock_map_seq_start,
778 .next = sock_map_seq_next,
779 .stop = sock_map_seq_stop,
780 .show = sock_map_seq_show,
783 static int sock_map_init_seq_private(void *priv_data,
784 struct bpf_iter_aux_info *aux)
786 struct sock_map_seq_info *info = priv_data;
788 bpf_map_inc_with_uref(aux->map);
789 info->map = aux->map;
793 static void sock_map_fini_seq_private(void *priv_data)
795 struct sock_map_seq_info *info = priv_data;
797 bpf_map_put_with_uref(info->map);
800 static const struct bpf_iter_seq_info sock_map_iter_seq_info = {
801 .seq_ops = &sock_map_seq_ops,
802 .init_seq_private = sock_map_init_seq_private,
803 .fini_seq_private = sock_map_fini_seq_private,
804 .seq_priv_size = sizeof(struct sock_map_seq_info),
807 BTF_ID_LIST_SINGLE(sock_map_btf_ids, struct, bpf_stab)
808 const struct bpf_map_ops sock_map_ops = {
809 .map_meta_equal = bpf_map_meta_equal,
810 .map_alloc = sock_map_alloc,
811 .map_free = sock_map_free,
812 .map_get_next_key = sock_map_get_next_key,
813 .map_lookup_elem_sys_only = sock_map_lookup_sys,
814 .map_update_elem = sock_map_update_elem,
815 .map_delete_elem = sock_map_delete_elem,
816 .map_lookup_elem = sock_map_lookup,
817 .map_release_uref = sock_map_release_progs,
818 .map_check_btf = map_check_no_btf,
819 .map_btf_id = &sock_map_btf_ids[0],
820 .iter_seq_info = &sock_map_iter_seq_info,
823 struct bpf_shtab_elem {
827 struct hlist_node node;
831 struct bpf_shtab_bucket {
832 struct hlist_head head;
838 struct bpf_shtab_bucket *buckets;
841 struct sk_psock_progs progs;
845 static inline u32 sock_hash_bucket_hash(const void *key, u32 len)
847 return jhash(key, len, 0);
850 static struct bpf_shtab_bucket *sock_hash_select_bucket(struct bpf_shtab *htab,
853 return &htab->buckets[hash & (htab->buckets_num - 1)];
856 static struct bpf_shtab_elem *
857 sock_hash_lookup_elem_raw(struct hlist_head *head, u32 hash, void *key,
860 struct bpf_shtab_elem *elem;
862 hlist_for_each_entry_rcu(elem, head, node) {
863 if (elem->hash == hash &&
864 !memcmp(&elem->key, key, key_size))
871 static struct sock *__sock_hash_lookup_elem(struct bpf_map *map, void *key)
873 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
874 u32 key_size = map->key_size, hash;
875 struct bpf_shtab_bucket *bucket;
876 struct bpf_shtab_elem *elem;
878 WARN_ON_ONCE(!rcu_read_lock_held());
880 hash = sock_hash_bucket_hash(key, key_size);
881 bucket = sock_hash_select_bucket(htab, hash);
882 elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
884 return elem ? elem->sk : NULL;
887 static void sock_hash_free_elem(struct bpf_shtab *htab,
888 struct bpf_shtab_elem *elem)
890 atomic_dec(&htab->count);
891 kfree_rcu(elem, rcu);
894 static void sock_hash_delete_from_link(struct bpf_map *map, struct sock *sk,
897 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
898 struct bpf_shtab_elem *elem_probe, *elem = link_raw;
899 struct bpf_shtab_bucket *bucket;
901 WARN_ON_ONCE(!rcu_read_lock_held());
902 bucket = sock_hash_select_bucket(htab, elem->hash);
904 /* elem may be deleted in parallel from the map, but access here
905 * is okay since it's going away only after RCU grace period.
906 * However, we need to check whether it's still present.
908 raw_spin_lock_bh(&bucket->lock);
909 elem_probe = sock_hash_lookup_elem_raw(&bucket->head, elem->hash,
910 elem->key, map->key_size);
911 if (elem_probe && elem_probe == elem) {
912 hlist_del_rcu(&elem->node);
913 sock_map_unref(elem->sk, elem);
914 sock_hash_free_elem(htab, elem);
916 raw_spin_unlock_bh(&bucket->lock);
919 static int sock_hash_delete_elem(struct bpf_map *map, void *key)
921 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
922 u32 hash, key_size = map->key_size;
923 struct bpf_shtab_bucket *bucket;
924 struct bpf_shtab_elem *elem;
927 hash = sock_hash_bucket_hash(key, key_size);
928 bucket = sock_hash_select_bucket(htab, hash);
930 raw_spin_lock_bh(&bucket->lock);
931 elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
933 hlist_del_rcu(&elem->node);
934 sock_map_unref(elem->sk, elem);
935 sock_hash_free_elem(htab, elem);
938 raw_spin_unlock_bh(&bucket->lock);
942 static struct bpf_shtab_elem *sock_hash_alloc_elem(struct bpf_shtab *htab,
943 void *key, u32 key_size,
944 u32 hash, struct sock *sk,
945 struct bpf_shtab_elem *old)
947 struct bpf_shtab_elem *new;
949 if (atomic_inc_return(&htab->count) > htab->map.max_entries) {
951 atomic_dec(&htab->count);
952 return ERR_PTR(-E2BIG);
956 new = bpf_map_kmalloc_node(&htab->map, htab->elem_size,
957 GFP_ATOMIC | __GFP_NOWARN,
958 htab->map.numa_node);
960 atomic_dec(&htab->count);
961 return ERR_PTR(-ENOMEM);
963 memcpy(new->key, key, key_size);
969 static int sock_hash_update_common(struct bpf_map *map, void *key,
970 struct sock *sk, u64 flags)
972 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
973 u32 key_size = map->key_size, hash;
974 struct bpf_shtab_elem *elem, *elem_new;
975 struct bpf_shtab_bucket *bucket;
976 struct sk_psock_link *link;
977 struct sk_psock *psock;
980 WARN_ON_ONCE(!rcu_read_lock_held());
981 if (unlikely(flags > BPF_EXIST))
984 link = sk_psock_init_link();
988 ret = sock_map_link(map, sk);
992 psock = sk_psock(sk);
993 WARN_ON_ONCE(!psock);
995 hash = sock_hash_bucket_hash(key, key_size);
996 bucket = sock_hash_select_bucket(htab, hash);
998 raw_spin_lock_bh(&bucket->lock);
999 elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
1000 if (elem && flags == BPF_NOEXIST) {
1003 } else if (!elem && flags == BPF_EXIST) {
1008 elem_new = sock_hash_alloc_elem(htab, key, key_size, hash, sk, elem);
1009 if (IS_ERR(elem_new)) {
1010 ret = PTR_ERR(elem_new);
1014 sock_map_add_link(psock, link, map, elem_new);
1015 /* Add new element to the head of the list, so that
1016 * concurrent search will find it before old elem.
1018 hlist_add_head_rcu(&elem_new->node, &bucket->head);
1020 hlist_del_rcu(&elem->node);
1021 sock_map_unref(elem->sk, elem);
1022 sock_hash_free_elem(htab, elem);
1024 raw_spin_unlock_bh(&bucket->lock);
1027 raw_spin_unlock_bh(&bucket->lock);
1028 sk_psock_put(sk, psock);
1030 sk_psock_free_link(link);
1034 static int sock_hash_get_next_key(struct bpf_map *map, void *key,
1037 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1038 struct bpf_shtab_elem *elem, *elem_next;
1039 u32 hash, key_size = map->key_size;
1040 struct hlist_head *head;
1044 goto find_first_elem;
1045 hash = sock_hash_bucket_hash(key, key_size);
1046 head = &sock_hash_select_bucket(htab, hash)->head;
1047 elem = sock_hash_lookup_elem_raw(head, hash, key, key_size);
1049 goto find_first_elem;
1051 elem_next = hlist_entry_safe(rcu_dereference(hlist_next_rcu(&elem->node)),
1052 struct bpf_shtab_elem, node);
1054 memcpy(key_next, elem_next->key, key_size);
1058 i = hash & (htab->buckets_num - 1);
1061 for (; i < htab->buckets_num; i++) {
1062 head = &sock_hash_select_bucket(htab, i)->head;
1063 elem_next = hlist_entry_safe(rcu_dereference(hlist_first_rcu(head)),
1064 struct bpf_shtab_elem, node);
1066 memcpy(key_next, elem_next->key, key_size);
1074 static struct bpf_map *sock_hash_alloc(union bpf_attr *attr)
1076 struct bpf_shtab *htab;
1079 if (!capable(CAP_NET_ADMIN))
1080 return ERR_PTR(-EPERM);
1081 if (attr->max_entries == 0 ||
1082 attr->key_size == 0 ||
1083 (attr->value_size != sizeof(u32) &&
1084 attr->value_size != sizeof(u64)) ||
1085 attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
1086 return ERR_PTR(-EINVAL);
1087 if (attr->key_size > MAX_BPF_STACK)
1088 return ERR_PTR(-E2BIG);
1090 htab = bpf_map_area_alloc(sizeof(*htab), NUMA_NO_NODE);
1092 return ERR_PTR(-ENOMEM);
1094 bpf_map_init_from_attr(&htab->map, attr);
1096 htab->buckets_num = roundup_pow_of_two(htab->map.max_entries);
1097 htab->elem_size = sizeof(struct bpf_shtab_elem) +
1098 round_up(htab->map.key_size, 8);
1099 if (htab->buckets_num == 0 ||
1100 htab->buckets_num > U32_MAX / sizeof(struct bpf_shtab_bucket)) {
1105 htab->buckets = bpf_map_area_alloc(htab->buckets_num *
1106 sizeof(struct bpf_shtab_bucket),
1107 htab->map.numa_node);
1108 if (!htab->buckets) {
1113 for (i = 0; i < htab->buckets_num; i++) {
1114 INIT_HLIST_HEAD(&htab->buckets[i].head);
1115 raw_spin_lock_init(&htab->buckets[i].lock);
1120 bpf_map_area_free(htab);
1121 return ERR_PTR(err);
1124 static void sock_hash_free(struct bpf_map *map)
1126 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1127 struct bpf_shtab_bucket *bucket;
1128 struct hlist_head unlink_list;
1129 struct bpf_shtab_elem *elem;
1130 struct hlist_node *node;
1133 /* After the sync no updates or deletes will be in-flight so it
1134 * is safe to walk map and remove entries without risking a race
1135 * in EEXIST update case.
1138 for (i = 0; i < htab->buckets_num; i++) {
1139 bucket = sock_hash_select_bucket(htab, i);
1141 /* We are racing with sock_hash_delete_from_link to
1142 * enter the spin-lock critical section. Every socket on
1143 * the list is still linked to sockhash. Since link
1144 * exists, psock exists and holds a ref to socket. That
1145 * lets us to grab a socket ref too.
1147 raw_spin_lock_bh(&bucket->lock);
1148 hlist_for_each_entry(elem, &bucket->head, node)
1149 sock_hold(elem->sk);
1150 hlist_move_list(&bucket->head, &unlink_list);
1151 raw_spin_unlock_bh(&bucket->lock);
1153 /* Process removed entries out of atomic context to
1154 * block for socket lock before deleting the psock's
1157 hlist_for_each_entry_safe(elem, node, &unlink_list, node) {
1158 hlist_del(&elem->node);
1159 lock_sock(elem->sk);
1161 sock_map_unref(elem->sk, elem);
1163 release_sock(elem->sk);
1165 sock_hash_free_elem(htab, elem);
1169 /* wait for psock readers accessing its map link */
1172 bpf_map_area_free(htab->buckets);
1173 bpf_map_area_free(htab);
1176 static void *sock_hash_lookup_sys(struct bpf_map *map, void *key)
1180 if (map->value_size != sizeof(u64))
1181 return ERR_PTR(-ENOSPC);
1183 sk = __sock_hash_lookup_elem(map, key);
1185 return ERR_PTR(-ENOENT);
1187 __sock_gen_cookie(sk);
1188 return &sk->sk_cookie;
1191 static void *sock_hash_lookup(struct bpf_map *map, void *key)
1195 sk = __sock_hash_lookup_elem(map, key);
1198 if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
1203 static void sock_hash_release_progs(struct bpf_map *map)
1205 psock_progs_drop(&container_of(map, struct bpf_shtab, map)->progs);
1208 BPF_CALL_4(bpf_sock_hash_update, struct bpf_sock_ops_kern *, sops,
1209 struct bpf_map *, map, void *, key, u64, flags)
1211 WARN_ON_ONCE(!rcu_read_lock_held());
1213 if (likely(sock_map_sk_is_suitable(sops->sk) &&
1214 sock_map_op_okay(sops)))
1215 return sock_hash_update_common(map, key, sops->sk, flags);
1219 const struct bpf_func_proto bpf_sock_hash_update_proto = {
1220 .func = bpf_sock_hash_update,
1223 .ret_type = RET_INTEGER,
1224 .arg1_type = ARG_PTR_TO_CTX,
1225 .arg2_type = ARG_CONST_MAP_PTR,
1226 .arg3_type = ARG_PTR_TO_MAP_KEY,
1227 .arg4_type = ARG_ANYTHING,
1230 BPF_CALL_4(bpf_sk_redirect_hash, struct sk_buff *, skb,
1231 struct bpf_map *, map, void *, key, u64, flags)
1235 if (unlikely(flags & ~(BPF_F_INGRESS)))
1238 sk = __sock_hash_lookup_elem(map, key);
1239 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1242 skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
1246 const struct bpf_func_proto bpf_sk_redirect_hash_proto = {
1247 .func = bpf_sk_redirect_hash,
1249 .ret_type = RET_INTEGER,
1250 .arg1_type = ARG_PTR_TO_CTX,
1251 .arg2_type = ARG_CONST_MAP_PTR,
1252 .arg3_type = ARG_PTR_TO_MAP_KEY,
1253 .arg4_type = ARG_ANYTHING,
1256 BPF_CALL_4(bpf_msg_redirect_hash, struct sk_msg *, msg,
1257 struct bpf_map *, map, void *, key, u64, flags)
1261 if (unlikely(flags & ~(BPF_F_INGRESS)))
1264 sk = __sock_hash_lookup_elem(map, key);
1265 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1273 const struct bpf_func_proto bpf_msg_redirect_hash_proto = {
1274 .func = bpf_msg_redirect_hash,
1276 .ret_type = RET_INTEGER,
1277 .arg1_type = ARG_PTR_TO_CTX,
1278 .arg2_type = ARG_CONST_MAP_PTR,
1279 .arg3_type = ARG_PTR_TO_MAP_KEY,
1280 .arg4_type = ARG_ANYTHING,
1283 struct sock_hash_seq_info {
1284 struct bpf_map *map;
1285 struct bpf_shtab *htab;
1289 static void *sock_hash_seq_find_next(struct sock_hash_seq_info *info,
1290 struct bpf_shtab_elem *prev_elem)
1292 const struct bpf_shtab *htab = info->htab;
1293 struct bpf_shtab_bucket *bucket;
1294 struct bpf_shtab_elem *elem;
1295 struct hlist_node *node;
1297 /* try to find next elem in the same bucket */
1299 node = rcu_dereference(hlist_next_rcu(&prev_elem->node));
1300 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1304 /* no more elements, continue in the next bucket */
1308 for (; info->bucket_id < htab->buckets_num; info->bucket_id++) {
1309 bucket = &htab->buckets[info->bucket_id];
1310 node = rcu_dereference(hlist_first_rcu(&bucket->head));
1311 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1319 static void *sock_hash_seq_start(struct seq_file *seq, loff_t *pos)
1322 struct sock_hash_seq_info *info = seq->private;
1327 /* pairs with sock_hash_seq_stop */
1329 return sock_hash_seq_find_next(info, NULL);
1332 static void *sock_hash_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1335 struct sock_hash_seq_info *info = seq->private;
1338 return sock_hash_seq_find_next(info, v);
1341 static int sock_hash_seq_show(struct seq_file *seq, void *v)
1344 struct sock_hash_seq_info *info = seq->private;
1345 struct bpf_iter__sockmap ctx = {};
1346 struct bpf_shtab_elem *elem = v;
1347 struct bpf_iter_meta meta;
1348 struct bpf_prog *prog;
1351 prog = bpf_iter_get_info(&meta, !elem);
1356 ctx.map = info->map;
1358 ctx.key = elem->key;
1362 return bpf_iter_run_prog(prog, &ctx);
1365 static void sock_hash_seq_stop(struct seq_file *seq, void *v)
1369 (void)sock_hash_seq_show(seq, NULL);
1371 /* pairs with sock_hash_seq_start */
1375 static const struct seq_operations sock_hash_seq_ops = {
1376 .start = sock_hash_seq_start,
1377 .next = sock_hash_seq_next,
1378 .stop = sock_hash_seq_stop,
1379 .show = sock_hash_seq_show,
1382 static int sock_hash_init_seq_private(void *priv_data,
1383 struct bpf_iter_aux_info *aux)
1385 struct sock_hash_seq_info *info = priv_data;
1387 bpf_map_inc_with_uref(aux->map);
1388 info->map = aux->map;
1389 info->htab = container_of(aux->map, struct bpf_shtab, map);
1393 static void sock_hash_fini_seq_private(void *priv_data)
1395 struct sock_hash_seq_info *info = priv_data;
1397 bpf_map_put_with_uref(info->map);
1400 static const struct bpf_iter_seq_info sock_hash_iter_seq_info = {
1401 .seq_ops = &sock_hash_seq_ops,
1402 .init_seq_private = sock_hash_init_seq_private,
1403 .fini_seq_private = sock_hash_fini_seq_private,
1404 .seq_priv_size = sizeof(struct sock_hash_seq_info),
1407 BTF_ID_LIST_SINGLE(sock_hash_map_btf_ids, struct, bpf_shtab)
1408 const struct bpf_map_ops sock_hash_ops = {
1409 .map_meta_equal = bpf_map_meta_equal,
1410 .map_alloc = sock_hash_alloc,
1411 .map_free = sock_hash_free,
1412 .map_get_next_key = sock_hash_get_next_key,
1413 .map_update_elem = sock_map_update_elem,
1414 .map_delete_elem = sock_hash_delete_elem,
1415 .map_lookup_elem = sock_hash_lookup,
1416 .map_lookup_elem_sys_only = sock_hash_lookup_sys,
1417 .map_release_uref = sock_hash_release_progs,
1418 .map_check_btf = map_check_no_btf,
1419 .map_btf_id = &sock_hash_map_btf_ids[0],
1420 .iter_seq_info = &sock_hash_iter_seq_info,
1423 static struct sk_psock_progs *sock_map_progs(struct bpf_map *map)
1425 switch (map->map_type) {
1426 case BPF_MAP_TYPE_SOCKMAP:
1427 return &container_of(map, struct bpf_stab, map)->progs;
1428 case BPF_MAP_TYPE_SOCKHASH:
1429 return &container_of(map, struct bpf_shtab, map)->progs;
1437 static int sock_map_prog_lookup(struct bpf_map *map, struct bpf_prog ***pprog,
1440 struct sk_psock_progs *progs = sock_map_progs(map);
1446 case BPF_SK_MSG_VERDICT:
1447 *pprog = &progs->msg_parser;
1449 #if IS_ENABLED(CONFIG_BPF_STREAM_PARSER)
1450 case BPF_SK_SKB_STREAM_PARSER:
1451 *pprog = &progs->stream_parser;
1454 case BPF_SK_SKB_STREAM_VERDICT:
1455 if (progs->skb_verdict)
1457 *pprog = &progs->stream_verdict;
1459 case BPF_SK_SKB_VERDICT:
1460 if (progs->stream_verdict)
1462 *pprog = &progs->skb_verdict;
1471 static int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog,
1472 struct bpf_prog *old, u32 which)
1474 struct bpf_prog **pprog;
1477 ret = sock_map_prog_lookup(map, &pprog, which);
1482 return psock_replace_prog(pprog, prog, old);
1484 psock_set_prog(pprog, prog);
1488 int sock_map_bpf_prog_query(const union bpf_attr *attr,
1489 union bpf_attr __user *uattr)
1491 __u32 __user *prog_ids = u64_to_user_ptr(attr->query.prog_ids);
1492 u32 prog_cnt = 0, flags = 0, ufd = attr->target_fd;
1493 struct bpf_prog **pprog;
1494 struct bpf_prog *prog;
1495 struct bpf_map *map;
1500 if (attr->query.query_flags)
1504 map = __bpf_map_get(f);
1506 return PTR_ERR(map);
1510 ret = sock_map_prog_lookup(map, &pprog, attr->query.attach_type);
1515 prog_cnt = !prog ? 0 : 1;
1517 if (!attr->query.prog_cnt || !prog_ids || !prog_cnt)
1520 /* we do not hold the refcnt, the bpf prog may be released
1521 * asynchronously and the id would be set to 0.
1523 id = data_race(prog->aux->id);
1530 if (copy_to_user(&uattr->query.attach_flags, &flags, sizeof(flags)) ||
1531 (id != 0 && copy_to_user(prog_ids, &id, sizeof(u32))) ||
1532 copy_to_user(&uattr->query.prog_cnt, &prog_cnt, sizeof(prog_cnt)))
1539 static void sock_map_unlink(struct sock *sk, struct sk_psock_link *link)
1541 switch (link->map->map_type) {
1542 case BPF_MAP_TYPE_SOCKMAP:
1543 return sock_map_delete_from_link(link->map, sk,
1545 case BPF_MAP_TYPE_SOCKHASH:
1546 return sock_hash_delete_from_link(link->map, sk,
1553 static void sock_map_remove_links(struct sock *sk, struct sk_psock *psock)
1555 struct sk_psock_link *link;
1557 while ((link = sk_psock_link_pop(psock))) {
1558 sock_map_unlink(sk, link);
1559 sk_psock_free_link(link);
1563 void sock_map_unhash(struct sock *sk)
1565 void (*saved_unhash)(struct sock *sk);
1566 struct sk_psock *psock;
1569 psock = sk_psock(sk);
1570 if (unlikely(!psock)) {
1572 saved_unhash = READ_ONCE(sk->sk_prot)->unhash;
1574 saved_unhash = psock->saved_unhash;
1575 sock_map_remove_links(sk, psock);
1578 if (WARN_ON_ONCE(saved_unhash == sock_map_unhash))
1583 EXPORT_SYMBOL_GPL(sock_map_unhash);
1585 void sock_map_destroy(struct sock *sk)
1587 void (*saved_destroy)(struct sock *sk);
1588 struct sk_psock *psock;
1591 psock = sk_psock_get(sk);
1592 if (unlikely(!psock)) {
1594 saved_destroy = READ_ONCE(sk->sk_prot)->destroy;
1596 saved_destroy = psock->saved_destroy;
1597 sock_map_remove_links(sk, psock);
1599 sk_psock_stop(psock);
1600 sk_psock_put(sk, psock);
1602 if (WARN_ON_ONCE(saved_destroy == sock_map_destroy))
1607 EXPORT_SYMBOL_GPL(sock_map_destroy);
1609 void sock_map_close(struct sock *sk, long timeout)
1611 void (*saved_close)(struct sock *sk, long timeout);
1612 struct sk_psock *psock;
1616 psock = sk_psock_get(sk);
1617 if (unlikely(!psock)) {
1620 saved_close = READ_ONCE(sk->sk_prot)->close;
1622 saved_close = psock->saved_close;
1623 sock_map_remove_links(sk, psock);
1625 sk_psock_stop(psock);
1627 cancel_work_sync(&psock->work);
1628 sk_psock_put(sk, psock);
1630 /* Make sure we do not recurse. This is a bug.
1631 * Leak the socket instead of crashing on a stack overflow.
1633 if (WARN_ON_ONCE(saved_close == sock_map_close))
1635 saved_close(sk, timeout);
1637 EXPORT_SYMBOL_GPL(sock_map_close);
1639 static int sock_map_iter_attach_target(struct bpf_prog *prog,
1640 union bpf_iter_link_info *linfo,
1641 struct bpf_iter_aux_info *aux)
1643 struct bpf_map *map;
1646 if (!linfo->map.map_fd)
1649 map = bpf_map_get_with_uref(linfo->map.map_fd);
1651 return PTR_ERR(map);
1653 if (map->map_type != BPF_MAP_TYPE_SOCKMAP &&
1654 map->map_type != BPF_MAP_TYPE_SOCKHASH)
1657 if (prog->aux->max_rdonly_access > map->key_size) {
1666 bpf_map_put_with_uref(map);
1670 static void sock_map_iter_detach_target(struct bpf_iter_aux_info *aux)
1672 bpf_map_put_with_uref(aux->map);
1675 static struct bpf_iter_reg sock_map_iter_reg = {
1676 .target = "sockmap",
1677 .attach_target = sock_map_iter_attach_target,
1678 .detach_target = sock_map_iter_detach_target,
1679 .show_fdinfo = bpf_iter_map_show_fdinfo,
1680 .fill_link_info = bpf_iter_map_fill_link_info,
1681 .ctx_arg_info_size = 2,
1683 { offsetof(struct bpf_iter__sockmap, key),
1684 PTR_TO_BUF | PTR_MAYBE_NULL | MEM_RDONLY },
1685 { offsetof(struct bpf_iter__sockmap, sk),
1686 PTR_TO_BTF_ID_OR_NULL },
1690 static int __init bpf_sockmap_iter_init(void)
1692 sock_map_iter_reg.ctx_arg_info[1].btf_id =
1693 btf_sock_ids[BTF_SOCK_TYPE_SOCK];
1694 return bpf_iter_reg_target(&sock_map_iter_reg);
1696 late_initcall(bpf_sockmap_iter_init);