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);
354 sock_map_unref(sk, psk);
360 /* wait for psock readers accessing its map link */
363 bpf_map_area_free(stab->sks);
364 bpf_map_area_free(stab);
367 static void sock_map_release_progs(struct bpf_map *map)
369 psock_progs_drop(&container_of(map, struct bpf_stab, map)->progs);
372 static struct sock *__sock_map_lookup_elem(struct bpf_map *map, u32 key)
374 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
376 WARN_ON_ONCE(!rcu_read_lock_held());
378 if (unlikely(key >= map->max_entries))
380 return READ_ONCE(stab->sks[key]);
383 static void *sock_map_lookup(struct bpf_map *map, void *key)
387 sk = __sock_map_lookup_elem(map, *(u32 *)key);
390 if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
395 static void *sock_map_lookup_sys(struct bpf_map *map, void *key)
399 if (map->value_size != sizeof(u64))
400 return ERR_PTR(-ENOSPC);
402 sk = __sock_map_lookup_elem(map, *(u32 *)key);
404 return ERR_PTR(-ENOENT);
406 __sock_gen_cookie(sk);
407 return &sk->sk_cookie;
410 static int __sock_map_delete(struct bpf_stab *stab, struct sock *sk_test,
416 raw_spin_lock_bh(&stab->lock);
418 if (!sk_test || sk_test == sk)
419 sk = xchg(psk, NULL);
422 sock_map_unref(sk, psk);
426 raw_spin_unlock_bh(&stab->lock);
430 static void sock_map_delete_from_link(struct bpf_map *map, struct sock *sk,
433 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
435 __sock_map_delete(stab, sk, link_raw);
438 static int sock_map_delete_elem(struct bpf_map *map, void *key)
440 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
444 if (unlikely(i >= map->max_entries))
448 return __sock_map_delete(stab, NULL, psk);
451 static int sock_map_get_next_key(struct bpf_map *map, void *key, void *next)
453 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
454 u32 i = key ? *(u32 *)key : U32_MAX;
455 u32 *key_next = next;
457 if (i == stab->map.max_entries - 1)
459 if (i >= stab->map.max_entries)
466 static int sock_map_update_common(struct bpf_map *map, u32 idx,
467 struct sock *sk, u64 flags)
469 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
470 struct sk_psock_link *link;
471 struct sk_psock *psock;
475 WARN_ON_ONCE(!rcu_read_lock_held());
476 if (unlikely(flags > BPF_EXIST))
478 if (unlikely(idx >= map->max_entries))
481 link = sk_psock_init_link();
485 ret = sock_map_link(map, sk);
489 psock = sk_psock(sk);
490 WARN_ON_ONCE(!psock);
492 raw_spin_lock_bh(&stab->lock);
493 osk = stab->sks[idx];
494 if (osk && flags == BPF_NOEXIST) {
497 } else if (!osk && flags == BPF_EXIST) {
502 sock_map_add_link(psock, link, map, &stab->sks[idx]);
505 sock_map_unref(osk, &stab->sks[idx]);
506 raw_spin_unlock_bh(&stab->lock);
509 raw_spin_unlock_bh(&stab->lock);
511 sk_psock_put(sk, psock);
513 sk_psock_free_link(link);
517 static bool sock_map_op_okay(const struct bpf_sock_ops_kern *ops)
519 return ops->op == BPF_SOCK_OPS_PASSIVE_ESTABLISHED_CB ||
520 ops->op == BPF_SOCK_OPS_ACTIVE_ESTABLISHED_CB ||
521 ops->op == BPF_SOCK_OPS_TCP_LISTEN_CB;
524 static bool sock_map_redirect_allowed(const struct sock *sk)
527 return sk->sk_state != TCP_LISTEN;
529 return sk->sk_state == TCP_ESTABLISHED;
532 static bool sock_map_sk_is_suitable(const struct sock *sk)
534 return !!sk->sk_prot->psock_update_sk_prot;
537 static bool sock_map_sk_state_allowed(const struct sock *sk)
540 return (1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_LISTEN);
544 static int sock_hash_update_common(struct bpf_map *map, void *key,
545 struct sock *sk, u64 flags);
547 int sock_map_update_elem_sys(struct bpf_map *map, void *key, void *value,
555 if (map->value_size == sizeof(u64))
562 sock = sockfd_lookup(ufd, &ret);
570 if (!sock_map_sk_is_suitable(sk)) {
575 sock_map_sk_acquire(sk);
576 if (!sock_map_sk_state_allowed(sk))
578 else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
579 ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
581 ret = sock_hash_update_common(map, key, sk, flags);
582 sock_map_sk_release(sk);
588 static int sock_map_update_elem(struct bpf_map *map, void *key,
589 void *value, u64 flags)
591 struct sock *sk = (struct sock *)value;
594 if (unlikely(!sk || !sk_fullsock(sk)))
597 if (!sock_map_sk_is_suitable(sk))
602 if (!sock_map_sk_state_allowed(sk))
604 else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
605 ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
607 ret = sock_hash_update_common(map, key, sk, flags);
613 BPF_CALL_4(bpf_sock_map_update, struct bpf_sock_ops_kern *, sops,
614 struct bpf_map *, map, void *, key, u64, flags)
616 WARN_ON_ONCE(!rcu_read_lock_held());
618 if (likely(sock_map_sk_is_suitable(sops->sk) &&
619 sock_map_op_okay(sops)))
620 return sock_map_update_common(map, *(u32 *)key, sops->sk,
625 const struct bpf_func_proto bpf_sock_map_update_proto = {
626 .func = bpf_sock_map_update,
629 .ret_type = RET_INTEGER,
630 .arg1_type = ARG_PTR_TO_CTX,
631 .arg2_type = ARG_CONST_MAP_PTR,
632 .arg3_type = ARG_PTR_TO_MAP_KEY,
633 .arg4_type = ARG_ANYTHING,
636 BPF_CALL_4(bpf_sk_redirect_map, struct sk_buff *, skb,
637 struct bpf_map *, map, u32, key, u64, flags)
641 if (unlikely(flags & ~(BPF_F_INGRESS)))
644 sk = __sock_map_lookup_elem(map, key);
645 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
648 skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
652 const struct bpf_func_proto bpf_sk_redirect_map_proto = {
653 .func = bpf_sk_redirect_map,
655 .ret_type = RET_INTEGER,
656 .arg1_type = ARG_PTR_TO_CTX,
657 .arg2_type = ARG_CONST_MAP_PTR,
658 .arg3_type = ARG_ANYTHING,
659 .arg4_type = ARG_ANYTHING,
662 BPF_CALL_4(bpf_msg_redirect_map, struct sk_msg *, msg,
663 struct bpf_map *, map, u32, key, u64, flags)
667 if (unlikely(flags & ~(BPF_F_INGRESS)))
670 sk = __sock_map_lookup_elem(map, key);
671 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
679 const struct bpf_func_proto bpf_msg_redirect_map_proto = {
680 .func = bpf_msg_redirect_map,
682 .ret_type = RET_INTEGER,
683 .arg1_type = ARG_PTR_TO_CTX,
684 .arg2_type = ARG_CONST_MAP_PTR,
685 .arg3_type = ARG_ANYTHING,
686 .arg4_type = ARG_ANYTHING,
689 struct sock_map_seq_info {
695 struct bpf_iter__sockmap {
696 __bpf_md_ptr(struct bpf_iter_meta *, meta);
697 __bpf_md_ptr(struct bpf_map *, map);
698 __bpf_md_ptr(void *, key);
699 __bpf_md_ptr(struct sock *, sk);
702 DEFINE_BPF_ITER_FUNC(sockmap, struct bpf_iter_meta *meta,
703 struct bpf_map *map, void *key,
706 static void *sock_map_seq_lookup_elem(struct sock_map_seq_info *info)
708 if (unlikely(info->index >= info->map->max_entries))
711 info->sk = __sock_map_lookup_elem(info->map, info->index);
713 /* can't return sk directly, since that might be NULL */
717 static void *sock_map_seq_start(struct seq_file *seq, loff_t *pos)
720 struct sock_map_seq_info *info = seq->private;
725 /* pairs with sock_map_seq_stop */
727 return sock_map_seq_lookup_elem(info);
730 static void *sock_map_seq_next(struct seq_file *seq, void *v, loff_t *pos)
733 struct sock_map_seq_info *info = seq->private;
738 return sock_map_seq_lookup_elem(info);
741 static int sock_map_seq_show(struct seq_file *seq, void *v)
744 struct sock_map_seq_info *info = seq->private;
745 struct bpf_iter__sockmap ctx = {};
746 struct bpf_iter_meta meta;
747 struct bpf_prog *prog;
750 prog = bpf_iter_get_info(&meta, !v);
757 ctx.key = &info->index;
761 return bpf_iter_run_prog(prog, &ctx);
764 static void sock_map_seq_stop(struct seq_file *seq, void *v)
768 (void)sock_map_seq_show(seq, NULL);
770 /* pairs with sock_map_seq_start */
774 static const struct seq_operations sock_map_seq_ops = {
775 .start = sock_map_seq_start,
776 .next = sock_map_seq_next,
777 .stop = sock_map_seq_stop,
778 .show = sock_map_seq_show,
781 static int sock_map_init_seq_private(void *priv_data,
782 struct bpf_iter_aux_info *aux)
784 struct sock_map_seq_info *info = priv_data;
786 bpf_map_inc_with_uref(aux->map);
787 info->map = aux->map;
791 static void sock_map_fini_seq_private(void *priv_data)
793 struct sock_map_seq_info *info = priv_data;
795 bpf_map_put_with_uref(info->map);
798 static const struct bpf_iter_seq_info sock_map_iter_seq_info = {
799 .seq_ops = &sock_map_seq_ops,
800 .init_seq_private = sock_map_init_seq_private,
801 .fini_seq_private = sock_map_fini_seq_private,
802 .seq_priv_size = sizeof(struct sock_map_seq_info),
805 BTF_ID_LIST_SINGLE(sock_map_btf_ids, struct, bpf_stab)
806 const struct bpf_map_ops sock_map_ops = {
807 .map_meta_equal = bpf_map_meta_equal,
808 .map_alloc = sock_map_alloc,
809 .map_free = sock_map_free,
810 .map_get_next_key = sock_map_get_next_key,
811 .map_lookup_elem_sys_only = sock_map_lookup_sys,
812 .map_update_elem = sock_map_update_elem,
813 .map_delete_elem = sock_map_delete_elem,
814 .map_lookup_elem = sock_map_lookup,
815 .map_release_uref = sock_map_release_progs,
816 .map_check_btf = map_check_no_btf,
817 .map_btf_id = &sock_map_btf_ids[0],
818 .iter_seq_info = &sock_map_iter_seq_info,
821 struct bpf_shtab_elem {
825 struct hlist_node node;
829 struct bpf_shtab_bucket {
830 struct hlist_head head;
836 struct bpf_shtab_bucket *buckets;
839 struct sk_psock_progs progs;
843 static inline u32 sock_hash_bucket_hash(const void *key, u32 len)
845 return jhash(key, len, 0);
848 static struct bpf_shtab_bucket *sock_hash_select_bucket(struct bpf_shtab *htab,
851 return &htab->buckets[hash & (htab->buckets_num - 1)];
854 static struct bpf_shtab_elem *
855 sock_hash_lookup_elem_raw(struct hlist_head *head, u32 hash, void *key,
858 struct bpf_shtab_elem *elem;
860 hlist_for_each_entry_rcu(elem, head, node) {
861 if (elem->hash == hash &&
862 !memcmp(&elem->key, key, key_size))
869 static struct sock *__sock_hash_lookup_elem(struct bpf_map *map, void *key)
871 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
872 u32 key_size = map->key_size, hash;
873 struct bpf_shtab_bucket *bucket;
874 struct bpf_shtab_elem *elem;
876 WARN_ON_ONCE(!rcu_read_lock_held());
878 hash = sock_hash_bucket_hash(key, key_size);
879 bucket = sock_hash_select_bucket(htab, hash);
880 elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
882 return elem ? elem->sk : NULL;
885 static void sock_hash_free_elem(struct bpf_shtab *htab,
886 struct bpf_shtab_elem *elem)
888 atomic_dec(&htab->count);
889 kfree_rcu(elem, rcu);
892 static void sock_hash_delete_from_link(struct bpf_map *map, struct sock *sk,
895 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
896 struct bpf_shtab_elem *elem_probe, *elem = link_raw;
897 struct bpf_shtab_bucket *bucket;
899 WARN_ON_ONCE(!rcu_read_lock_held());
900 bucket = sock_hash_select_bucket(htab, elem->hash);
902 /* elem may be deleted in parallel from the map, but access here
903 * is okay since it's going away only after RCU grace period.
904 * However, we need to check whether it's still present.
906 raw_spin_lock_bh(&bucket->lock);
907 elem_probe = sock_hash_lookup_elem_raw(&bucket->head, elem->hash,
908 elem->key, map->key_size);
909 if (elem_probe && elem_probe == elem) {
910 hlist_del_rcu(&elem->node);
911 sock_map_unref(elem->sk, elem);
912 sock_hash_free_elem(htab, elem);
914 raw_spin_unlock_bh(&bucket->lock);
917 static int sock_hash_delete_elem(struct bpf_map *map, void *key)
919 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
920 u32 hash, key_size = map->key_size;
921 struct bpf_shtab_bucket *bucket;
922 struct bpf_shtab_elem *elem;
925 hash = sock_hash_bucket_hash(key, key_size);
926 bucket = sock_hash_select_bucket(htab, hash);
928 raw_spin_lock_bh(&bucket->lock);
929 elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
931 hlist_del_rcu(&elem->node);
932 sock_map_unref(elem->sk, elem);
933 sock_hash_free_elem(htab, elem);
936 raw_spin_unlock_bh(&bucket->lock);
940 static struct bpf_shtab_elem *sock_hash_alloc_elem(struct bpf_shtab *htab,
941 void *key, u32 key_size,
942 u32 hash, struct sock *sk,
943 struct bpf_shtab_elem *old)
945 struct bpf_shtab_elem *new;
947 if (atomic_inc_return(&htab->count) > htab->map.max_entries) {
949 atomic_dec(&htab->count);
950 return ERR_PTR(-E2BIG);
954 new = bpf_map_kmalloc_node(&htab->map, htab->elem_size,
955 GFP_ATOMIC | __GFP_NOWARN,
956 htab->map.numa_node);
958 atomic_dec(&htab->count);
959 return ERR_PTR(-ENOMEM);
961 memcpy(new->key, key, key_size);
967 static int sock_hash_update_common(struct bpf_map *map, void *key,
968 struct sock *sk, u64 flags)
970 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
971 u32 key_size = map->key_size, hash;
972 struct bpf_shtab_elem *elem, *elem_new;
973 struct bpf_shtab_bucket *bucket;
974 struct sk_psock_link *link;
975 struct sk_psock *psock;
978 WARN_ON_ONCE(!rcu_read_lock_held());
979 if (unlikely(flags > BPF_EXIST))
982 link = sk_psock_init_link();
986 ret = sock_map_link(map, sk);
990 psock = sk_psock(sk);
991 WARN_ON_ONCE(!psock);
993 hash = sock_hash_bucket_hash(key, key_size);
994 bucket = sock_hash_select_bucket(htab, hash);
996 raw_spin_lock_bh(&bucket->lock);
997 elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
998 if (elem && flags == BPF_NOEXIST) {
1001 } else if (!elem && flags == BPF_EXIST) {
1006 elem_new = sock_hash_alloc_elem(htab, key, key_size, hash, sk, elem);
1007 if (IS_ERR(elem_new)) {
1008 ret = PTR_ERR(elem_new);
1012 sock_map_add_link(psock, link, map, elem_new);
1013 /* Add new element to the head of the list, so that
1014 * concurrent search will find it before old elem.
1016 hlist_add_head_rcu(&elem_new->node, &bucket->head);
1018 hlist_del_rcu(&elem->node);
1019 sock_map_unref(elem->sk, elem);
1020 sock_hash_free_elem(htab, elem);
1022 raw_spin_unlock_bh(&bucket->lock);
1025 raw_spin_unlock_bh(&bucket->lock);
1026 sk_psock_put(sk, psock);
1028 sk_psock_free_link(link);
1032 static int sock_hash_get_next_key(struct bpf_map *map, void *key,
1035 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1036 struct bpf_shtab_elem *elem, *elem_next;
1037 u32 hash, key_size = map->key_size;
1038 struct hlist_head *head;
1042 goto find_first_elem;
1043 hash = sock_hash_bucket_hash(key, key_size);
1044 head = &sock_hash_select_bucket(htab, hash)->head;
1045 elem = sock_hash_lookup_elem_raw(head, hash, key, key_size);
1047 goto find_first_elem;
1049 elem_next = hlist_entry_safe(rcu_dereference(hlist_next_rcu(&elem->node)),
1050 struct bpf_shtab_elem, node);
1052 memcpy(key_next, elem_next->key, key_size);
1056 i = hash & (htab->buckets_num - 1);
1059 for (; i < htab->buckets_num; i++) {
1060 head = &sock_hash_select_bucket(htab, i)->head;
1061 elem_next = hlist_entry_safe(rcu_dereference(hlist_first_rcu(head)),
1062 struct bpf_shtab_elem, node);
1064 memcpy(key_next, elem_next->key, key_size);
1072 static struct bpf_map *sock_hash_alloc(union bpf_attr *attr)
1074 struct bpf_shtab *htab;
1077 if (!capable(CAP_NET_ADMIN))
1078 return ERR_PTR(-EPERM);
1079 if (attr->max_entries == 0 ||
1080 attr->key_size == 0 ||
1081 (attr->value_size != sizeof(u32) &&
1082 attr->value_size != sizeof(u64)) ||
1083 attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
1084 return ERR_PTR(-EINVAL);
1085 if (attr->key_size > MAX_BPF_STACK)
1086 return ERR_PTR(-E2BIG);
1088 htab = bpf_map_area_alloc(sizeof(*htab), NUMA_NO_NODE);
1090 return ERR_PTR(-ENOMEM);
1092 bpf_map_init_from_attr(&htab->map, attr);
1094 htab->buckets_num = roundup_pow_of_two(htab->map.max_entries);
1095 htab->elem_size = sizeof(struct bpf_shtab_elem) +
1096 round_up(htab->map.key_size, 8);
1097 if (htab->buckets_num == 0 ||
1098 htab->buckets_num > U32_MAX / sizeof(struct bpf_shtab_bucket)) {
1103 htab->buckets = bpf_map_area_alloc(htab->buckets_num *
1104 sizeof(struct bpf_shtab_bucket),
1105 htab->map.numa_node);
1106 if (!htab->buckets) {
1111 for (i = 0; i < htab->buckets_num; i++) {
1112 INIT_HLIST_HEAD(&htab->buckets[i].head);
1113 raw_spin_lock_init(&htab->buckets[i].lock);
1118 bpf_map_area_free(htab);
1119 return ERR_PTR(err);
1122 static void sock_hash_free(struct bpf_map *map)
1124 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1125 struct bpf_shtab_bucket *bucket;
1126 struct hlist_head unlink_list;
1127 struct bpf_shtab_elem *elem;
1128 struct hlist_node *node;
1131 /* After the sync no updates or deletes will be in-flight so it
1132 * is safe to walk map and remove entries without risking a race
1133 * in EEXIST update case.
1136 for (i = 0; i < htab->buckets_num; i++) {
1137 bucket = sock_hash_select_bucket(htab, i);
1139 /* We are racing with sock_hash_delete_from_link to
1140 * enter the spin-lock critical section. Every socket on
1141 * the list is still linked to sockhash. Since link
1142 * exists, psock exists and holds a ref to socket. That
1143 * lets us to grab a socket ref too.
1145 raw_spin_lock_bh(&bucket->lock);
1146 hlist_for_each_entry(elem, &bucket->head, node)
1147 sock_hold(elem->sk);
1148 hlist_move_list(&bucket->head, &unlink_list);
1149 raw_spin_unlock_bh(&bucket->lock);
1151 /* Process removed entries out of atomic context to
1152 * block for socket lock before deleting the psock's
1155 hlist_for_each_entry_safe(elem, node, &unlink_list, node) {
1156 hlist_del(&elem->node);
1157 lock_sock(elem->sk);
1159 sock_map_unref(elem->sk, elem);
1161 release_sock(elem->sk);
1163 sock_hash_free_elem(htab, elem);
1167 /* wait for psock readers accessing its map link */
1170 bpf_map_area_free(htab->buckets);
1171 bpf_map_area_free(htab);
1174 static void *sock_hash_lookup_sys(struct bpf_map *map, void *key)
1178 if (map->value_size != sizeof(u64))
1179 return ERR_PTR(-ENOSPC);
1181 sk = __sock_hash_lookup_elem(map, key);
1183 return ERR_PTR(-ENOENT);
1185 __sock_gen_cookie(sk);
1186 return &sk->sk_cookie;
1189 static void *sock_hash_lookup(struct bpf_map *map, void *key)
1193 sk = __sock_hash_lookup_elem(map, key);
1196 if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
1201 static void sock_hash_release_progs(struct bpf_map *map)
1203 psock_progs_drop(&container_of(map, struct bpf_shtab, map)->progs);
1206 BPF_CALL_4(bpf_sock_hash_update, struct bpf_sock_ops_kern *, sops,
1207 struct bpf_map *, map, void *, key, u64, flags)
1209 WARN_ON_ONCE(!rcu_read_lock_held());
1211 if (likely(sock_map_sk_is_suitable(sops->sk) &&
1212 sock_map_op_okay(sops)))
1213 return sock_hash_update_common(map, key, sops->sk, flags);
1217 const struct bpf_func_proto bpf_sock_hash_update_proto = {
1218 .func = bpf_sock_hash_update,
1221 .ret_type = RET_INTEGER,
1222 .arg1_type = ARG_PTR_TO_CTX,
1223 .arg2_type = ARG_CONST_MAP_PTR,
1224 .arg3_type = ARG_PTR_TO_MAP_KEY,
1225 .arg4_type = ARG_ANYTHING,
1228 BPF_CALL_4(bpf_sk_redirect_hash, struct sk_buff *, skb,
1229 struct bpf_map *, map, void *, key, u64, flags)
1233 if (unlikely(flags & ~(BPF_F_INGRESS)))
1236 sk = __sock_hash_lookup_elem(map, key);
1237 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1240 skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
1244 const struct bpf_func_proto bpf_sk_redirect_hash_proto = {
1245 .func = bpf_sk_redirect_hash,
1247 .ret_type = RET_INTEGER,
1248 .arg1_type = ARG_PTR_TO_CTX,
1249 .arg2_type = ARG_CONST_MAP_PTR,
1250 .arg3_type = ARG_PTR_TO_MAP_KEY,
1251 .arg4_type = ARG_ANYTHING,
1254 BPF_CALL_4(bpf_msg_redirect_hash, struct sk_msg *, msg,
1255 struct bpf_map *, map, void *, key, u64, flags)
1259 if (unlikely(flags & ~(BPF_F_INGRESS)))
1262 sk = __sock_hash_lookup_elem(map, key);
1263 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1271 const struct bpf_func_proto bpf_msg_redirect_hash_proto = {
1272 .func = bpf_msg_redirect_hash,
1274 .ret_type = RET_INTEGER,
1275 .arg1_type = ARG_PTR_TO_CTX,
1276 .arg2_type = ARG_CONST_MAP_PTR,
1277 .arg3_type = ARG_PTR_TO_MAP_KEY,
1278 .arg4_type = ARG_ANYTHING,
1281 struct sock_hash_seq_info {
1282 struct bpf_map *map;
1283 struct bpf_shtab *htab;
1287 static void *sock_hash_seq_find_next(struct sock_hash_seq_info *info,
1288 struct bpf_shtab_elem *prev_elem)
1290 const struct bpf_shtab *htab = info->htab;
1291 struct bpf_shtab_bucket *bucket;
1292 struct bpf_shtab_elem *elem;
1293 struct hlist_node *node;
1295 /* try to find next elem in the same bucket */
1297 node = rcu_dereference(hlist_next_rcu(&prev_elem->node));
1298 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1302 /* no more elements, continue in the next bucket */
1306 for (; info->bucket_id < htab->buckets_num; info->bucket_id++) {
1307 bucket = &htab->buckets[info->bucket_id];
1308 node = rcu_dereference(hlist_first_rcu(&bucket->head));
1309 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1317 static void *sock_hash_seq_start(struct seq_file *seq, loff_t *pos)
1320 struct sock_hash_seq_info *info = seq->private;
1325 /* pairs with sock_hash_seq_stop */
1327 return sock_hash_seq_find_next(info, NULL);
1330 static void *sock_hash_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1333 struct sock_hash_seq_info *info = seq->private;
1336 return sock_hash_seq_find_next(info, v);
1339 static int sock_hash_seq_show(struct seq_file *seq, void *v)
1342 struct sock_hash_seq_info *info = seq->private;
1343 struct bpf_iter__sockmap ctx = {};
1344 struct bpf_shtab_elem *elem = v;
1345 struct bpf_iter_meta meta;
1346 struct bpf_prog *prog;
1349 prog = bpf_iter_get_info(&meta, !elem);
1354 ctx.map = info->map;
1356 ctx.key = elem->key;
1360 return bpf_iter_run_prog(prog, &ctx);
1363 static void sock_hash_seq_stop(struct seq_file *seq, void *v)
1367 (void)sock_hash_seq_show(seq, NULL);
1369 /* pairs with sock_hash_seq_start */
1373 static const struct seq_operations sock_hash_seq_ops = {
1374 .start = sock_hash_seq_start,
1375 .next = sock_hash_seq_next,
1376 .stop = sock_hash_seq_stop,
1377 .show = sock_hash_seq_show,
1380 static int sock_hash_init_seq_private(void *priv_data,
1381 struct bpf_iter_aux_info *aux)
1383 struct sock_hash_seq_info *info = priv_data;
1385 bpf_map_inc_with_uref(aux->map);
1386 info->map = aux->map;
1387 info->htab = container_of(aux->map, struct bpf_shtab, map);
1391 static void sock_hash_fini_seq_private(void *priv_data)
1393 struct sock_hash_seq_info *info = priv_data;
1395 bpf_map_put_with_uref(info->map);
1398 static const struct bpf_iter_seq_info sock_hash_iter_seq_info = {
1399 .seq_ops = &sock_hash_seq_ops,
1400 .init_seq_private = sock_hash_init_seq_private,
1401 .fini_seq_private = sock_hash_fini_seq_private,
1402 .seq_priv_size = sizeof(struct sock_hash_seq_info),
1405 BTF_ID_LIST_SINGLE(sock_hash_map_btf_ids, struct, bpf_shtab)
1406 const struct bpf_map_ops sock_hash_ops = {
1407 .map_meta_equal = bpf_map_meta_equal,
1408 .map_alloc = sock_hash_alloc,
1409 .map_free = sock_hash_free,
1410 .map_get_next_key = sock_hash_get_next_key,
1411 .map_update_elem = sock_map_update_elem,
1412 .map_delete_elem = sock_hash_delete_elem,
1413 .map_lookup_elem = sock_hash_lookup,
1414 .map_lookup_elem_sys_only = sock_hash_lookup_sys,
1415 .map_release_uref = sock_hash_release_progs,
1416 .map_check_btf = map_check_no_btf,
1417 .map_btf_id = &sock_hash_map_btf_ids[0],
1418 .iter_seq_info = &sock_hash_iter_seq_info,
1421 static struct sk_psock_progs *sock_map_progs(struct bpf_map *map)
1423 switch (map->map_type) {
1424 case BPF_MAP_TYPE_SOCKMAP:
1425 return &container_of(map, struct bpf_stab, map)->progs;
1426 case BPF_MAP_TYPE_SOCKHASH:
1427 return &container_of(map, struct bpf_shtab, map)->progs;
1435 static int sock_map_prog_lookup(struct bpf_map *map, struct bpf_prog ***pprog,
1438 struct sk_psock_progs *progs = sock_map_progs(map);
1444 case BPF_SK_MSG_VERDICT:
1445 *pprog = &progs->msg_parser;
1447 #if IS_ENABLED(CONFIG_BPF_STREAM_PARSER)
1448 case BPF_SK_SKB_STREAM_PARSER:
1449 *pprog = &progs->stream_parser;
1452 case BPF_SK_SKB_STREAM_VERDICT:
1453 if (progs->skb_verdict)
1455 *pprog = &progs->stream_verdict;
1457 case BPF_SK_SKB_VERDICT:
1458 if (progs->stream_verdict)
1460 *pprog = &progs->skb_verdict;
1469 static int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog,
1470 struct bpf_prog *old, u32 which)
1472 struct bpf_prog **pprog;
1475 ret = sock_map_prog_lookup(map, &pprog, which);
1480 return psock_replace_prog(pprog, prog, old);
1482 psock_set_prog(pprog, prog);
1486 int sock_map_bpf_prog_query(const union bpf_attr *attr,
1487 union bpf_attr __user *uattr)
1489 __u32 __user *prog_ids = u64_to_user_ptr(attr->query.prog_ids);
1490 u32 prog_cnt = 0, flags = 0, ufd = attr->target_fd;
1491 struct bpf_prog **pprog;
1492 struct bpf_prog *prog;
1493 struct bpf_map *map;
1498 if (attr->query.query_flags)
1502 map = __bpf_map_get(f);
1504 return PTR_ERR(map);
1508 ret = sock_map_prog_lookup(map, &pprog, attr->query.attach_type);
1513 prog_cnt = !prog ? 0 : 1;
1515 if (!attr->query.prog_cnt || !prog_ids || !prog_cnt)
1518 /* we do not hold the refcnt, the bpf prog may be released
1519 * asynchronously and the id would be set to 0.
1521 id = data_race(prog->aux->id);
1528 if (copy_to_user(&uattr->query.attach_flags, &flags, sizeof(flags)) ||
1529 (id != 0 && copy_to_user(prog_ids, &id, sizeof(u32))) ||
1530 copy_to_user(&uattr->query.prog_cnt, &prog_cnt, sizeof(prog_cnt)))
1537 static void sock_map_unlink(struct sock *sk, struct sk_psock_link *link)
1539 switch (link->map->map_type) {
1540 case BPF_MAP_TYPE_SOCKMAP:
1541 return sock_map_delete_from_link(link->map, sk,
1543 case BPF_MAP_TYPE_SOCKHASH:
1544 return sock_hash_delete_from_link(link->map, sk,
1551 static void sock_map_remove_links(struct sock *sk, struct sk_psock *psock)
1553 struct sk_psock_link *link;
1555 while ((link = sk_psock_link_pop(psock))) {
1556 sock_map_unlink(sk, link);
1557 sk_psock_free_link(link);
1561 void sock_map_unhash(struct sock *sk)
1563 void (*saved_unhash)(struct sock *sk);
1564 struct sk_psock *psock;
1567 psock = sk_psock(sk);
1568 if (unlikely(!psock)) {
1570 if (sk->sk_prot->unhash)
1571 sk->sk_prot->unhash(sk);
1575 saved_unhash = psock->saved_unhash;
1576 sock_map_remove_links(sk, psock);
1580 EXPORT_SYMBOL_GPL(sock_map_unhash);
1582 void sock_map_destroy(struct sock *sk)
1584 void (*saved_destroy)(struct sock *sk);
1585 struct sk_psock *psock;
1588 psock = sk_psock_get(sk);
1589 if (unlikely(!psock)) {
1591 if (sk->sk_prot->destroy)
1592 sk->sk_prot->destroy(sk);
1596 saved_destroy = psock->saved_destroy;
1597 sock_map_remove_links(sk, psock);
1599 sk_psock_stop(psock);
1600 sk_psock_put(sk, psock);
1603 EXPORT_SYMBOL_GPL(sock_map_destroy);
1605 void sock_map_close(struct sock *sk, long timeout)
1607 void (*saved_close)(struct sock *sk, long timeout);
1608 struct sk_psock *psock;
1612 psock = sk_psock_get(sk);
1613 if (unlikely(!psock)) {
1616 return sk->sk_prot->close(sk, timeout);
1619 saved_close = psock->saved_close;
1620 sock_map_remove_links(sk, psock);
1622 sk_psock_stop(psock);
1624 cancel_work_sync(&psock->work);
1625 sk_psock_put(sk, psock);
1626 saved_close(sk, timeout);
1628 EXPORT_SYMBOL_GPL(sock_map_close);
1630 static int sock_map_iter_attach_target(struct bpf_prog *prog,
1631 union bpf_iter_link_info *linfo,
1632 struct bpf_iter_aux_info *aux)
1634 struct bpf_map *map;
1637 if (!linfo->map.map_fd)
1640 map = bpf_map_get_with_uref(linfo->map.map_fd);
1642 return PTR_ERR(map);
1644 if (map->map_type != BPF_MAP_TYPE_SOCKMAP &&
1645 map->map_type != BPF_MAP_TYPE_SOCKHASH)
1648 if (prog->aux->max_rdonly_access > map->key_size) {
1657 bpf_map_put_with_uref(map);
1661 static void sock_map_iter_detach_target(struct bpf_iter_aux_info *aux)
1663 bpf_map_put_with_uref(aux->map);
1666 static struct bpf_iter_reg sock_map_iter_reg = {
1667 .target = "sockmap",
1668 .attach_target = sock_map_iter_attach_target,
1669 .detach_target = sock_map_iter_detach_target,
1670 .show_fdinfo = bpf_iter_map_show_fdinfo,
1671 .fill_link_info = bpf_iter_map_fill_link_info,
1672 .ctx_arg_info_size = 2,
1674 { offsetof(struct bpf_iter__sockmap, key),
1675 PTR_TO_BUF | PTR_MAYBE_NULL | MEM_RDONLY },
1676 { offsetof(struct bpf_iter__sockmap, sk),
1677 PTR_TO_BTF_ID_OR_NULL },
1681 static int __init bpf_sockmap_iter_init(void)
1683 sock_map_iter_reg.ctx_arg_info[1].btf_id =
1684 btf_sock_ids[BTF_SOCK_TYPE_SOCK];
1685 return bpf_iter_reg_target(&sock_map_iter_reg);
1687 late_initcall(bpf_sockmap_iter_init);