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 = kzalloc(sizeof(*stab), GFP_USER | __GFP_ACCOUNT);
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 *),
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);
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 info->map = aux->map;
790 static const struct bpf_iter_seq_info sock_map_iter_seq_info = {
791 .seq_ops = &sock_map_seq_ops,
792 .init_seq_private = sock_map_init_seq_private,
793 .seq_priv_size = sizeof(struct sock_map_seq_info),
796 BTF_ID_LIST_SINGLE(sock_map_btf_ids, struct, bpf_stab)
797 const struct bpf_map_ops sock_map_ops = {
798 .map_meta_equal = bpf_map_meta_equal,
799 .map_alloc = sock_map_alloc,
800 .map_free = sock_map_free,
801 .map_get_next_key = sock_map_get_next_key,
802 .map_lookup_elem_sys_only = sock_map_lookup_sys,
803 .map_update_elem = sock_map_update_elem,
804 .map_delete_elem = sock_map_delete_elem,
805 .map_lookup_elem = sock_map_lookup,
806 .map_release_uref = sock_map_release_progs,
807 .map_check_btf = map_check_no_btf,
808 .map_btf_id = &sock_map_btf_ids[0],
809 .iter_seq_info = &sock_map_iter_seq_info,
812 struct bpf_shtab_elem {
816 struct hlist_node node;
820 struct bpf_shtab_bucket {
821 struct hlist_head head;
827 struct bpf_shtab_bucket *buckets;
830 struct sk_psock_progs progs;
834 static inline u32 sock_hash_bucket_hash(const void *key, u32 len)
836 return jhash(key, len, 0);
839 static struct bpf_shtab_bucket *sock_hash_select_bucket(struct bpf_shtab *htab,
842 return &htab->buckets[hash & (htab->buckets_num - 1)];
845 static struct bpf_shtab_elem *
846 sock_hash_lookup_elem_raw(struct hlist_head *head, u32 hash, void *key,
849 struct bpf_shtab_elem *elem;
851 hlist_for_each_entry_rcu(elem, head, node) {
852 if (elem->hash == hash &&
853 !memcmp(&elem->key, key, key_size))
860 static struct sock *__sock_hash_lookup_elem(struct bpf_map *map, void *key)
862 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
863 u32 key_size = map->key_size, hash;
864 struct bpf_shtab_bucket *bucket;
865 struct bpf_shtab_elem *elem;
867 WARN_ON_ONCE(!rcu_read_lock_held());
869 hash = sock_hash_bucket_hash(key, key_size);
870 bucket = sock_hash_select_bucket(htab, hash);
871 elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
873 return elem ? elem->sk : NULL;
876 static void sock_hash_free_elem(struct bpf_shtab *htab,
877 struct bpf_shtab_elem *elem)
879 atomic_dec(&htab->count);
880 kfree_rcu(elem, rcu);
883 static void sock_hash_delete_from_link(struct bpf_map *map, struct sock *sk,
886 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
887 struct bpf_shtab_elem *elem_probe, *elem = link_raw;
888 struct bpf_shtab_bucket *bucket;
890 WARN_ON_ONCE(!rcu_read_lock_held());
891 bucket = sock_hash_select_bucket(htab, elem->hash);
893 /* elem may be deleted in parallel from the map, but access here
894 * is okay since it's going away only after RCU grace period.
895 * However, we need to check whether it's still present.
897 raw_spin_lock_bh(&bucket->lock);
898 elem_probe = sock_hash_lookup_elem_raw(&bucket->head, elem->hash,
899 elem->key, map->key_size);
900 if (elem_probe && elem_probe == elem) {
901 hlist_del_rcu(&elem->node);
902 sock_map_unref(elem->sk, elem);
903 sock_hash_free_elem(htab, elem);
905 raw_spin_unlock_bh(&bucket->lock);
908 static int sock_hash_delete_elem(struct bpf_map *map, void *key)
910 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
911 u32 hash, key_size = map->key_size;
912 struct bpf_shtab_bucket *bucket;
913 struct bpf_shtab_elem *elem;
916 hash = sock_hash_bucket_hash(key, key_size);
917 bucket = sock_hash_select_bucket(htab, hash);
919 raw_spin_lock_bh(&bucket->lock);
920 elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
922 hlist_del_rcu(&elem->node);
923 sock_map_unref(elem->sk, elem);
924 sock_hash_free_elem(htab, elem);
927 raw_spin_unlock_bh(&bucket->lock);
931 static struct bpf_shtab_elem *sock_hash_alloc_elem(struct bpf_shtab *htab,
932 void *key, u32 key_size,
933 u32 hash, struct sock *sk,
934 struct bpf_shtab_elem *old)
936 struct bpf_shtab_elem *new;
938 if (atomic_inc_return(&htab->count) > htab->map.max_entries) {
940 atomic_dec(&htab->count);
941 return ERR_PTR(-E2BIG);
945 new = bpf_map_kmalloc_node(&htab->map, htab->elem_size,
946 GFP_ATOMIC | __GFP_NOWARN,
947 htab->map.numa_node);
949 atomic_dec(&htab->count);
950 return ERR_PTR(-ENOMEM);
952 memcpy(new->key, key, key_size);
958 static int sock_hash_update_common(struct bpf_map *map, void *key,
959 struct sock *sk, u64 flags)
961 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
962 u32 key_size = map->key_size, hash;
963 struct bpf_shtab_elem *elem, *elem_new;
964 struct bpf_shtab_bucket *bucket;
965 struct sk_psock_link *link;
966 struct sk_psock *psock;
969 WARN_ON_ONCE(!rcu_read_lock_held());
970 if (unlikely(flags > BPF_EXIST))
973 link = sk_psock_init_link();
977 ret = sock_map_link(map, sk);
981 psock = sk_psock(sk);
982 WARN_ON_ONCE(!psock);
984 hash = sock_hash_bucket_hash(key, key_size);
985 bucket = sock_hash_select_bucket(htab, hash);
987 raw_spin_lock_bh(&bucket->lock);
988 elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
989 if (elem && flags == BPF_NOEXIST) {
992 } else if (!elem && flags == BPF_EXIST) {
997 elem_new = sock_hash_alloc_elem(htab, key, key_size, hash, sk, elem);
998 if (IS_ERR(elem_new)) {
999 ret = PTR_ERR(elem_new);
1003 sock_map_add_link(psock, link, map, elem_new);
1004 /* Add new element to the head of the list, so that
1005 * concurrent search will find it before old elem.
1007 hlist_add_head_rcu(&elem_new->node, &bucket->head);
1009 hlist_del_rcu(&elem->node);
1010 sock_map_unref(elem->sk, elem);
1011 sock_hash_free_elem(htab, elem);
1013 raw_spin_unlock_bh(&bucket->lock);
1016 raw_spin_unlock_bh(&bucket->lock);
1017 sk_psock_put(sk, psock);
1019 sk_psock_free_link(link);
1023 static int sock_hash_get_next_key(struct bpf_map *map, void *key,
1026 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1027 struct bpf_shtab_elem *elem, *elem_next;
1028 u32 hash, key_size = map->key_size;
1029 struct hlist_head *head;
1033 goto find_first_elem;
1034 hash = sock_hash_bucket_hash(key, key_size);
1035 head = &sock_hash_select_bucket(htab, hash)->head;
1036 elem = sock_hash_lookup_elem_raw(head, hash, key, key_size);
1038 goto find_first_elem;
1040 elem_next = hlist_entry_safe(rcu_dereference(hlist_next_rcu(&elem->node)),
1041 struct bpf_shtab_elem, node);
1043 memcpy(key_next, elem_next->key, key_size);
1047 i = hash & (htab->buckets_num - 1);
1050 for (; i < htab->buckets_num; i++) {
1051 head = &sock_hash_select_bucket(htab, i)->head;
1052 elem_next = hlist_entry_safe(rcu_dereference(hlist_first_rcu(head)),
1053 struct bpf_shtab_elem, node);
1055 memcpy(key_next, elem_next->key, key_size);
1063 static struct bpf_map *sock_hash_alloc(union bpf_attr *attr)
1065 struct bpf_shtab *htab;
1068 if (!capable(CAP_NET_ADMIN))
1069 return ERR_PTR(-EPERM);
1070 if (attr->max_entries == 0 ||
1071 attr->key_size == 0 ||
1072 (attr->value_size != sizeof(u32) &&
1073 attr->value_size != sizeof(u64)) ||
1074 attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
1075 return ERR_PTR(-EINVAL);
1076 if (attr->key_size > MAX_BPF_STACK)
1077 return ERR_PTR(-E2BIG);
1079 htab = kzalloc(sizeof(*htab), GFP_USER | __GFP_ACCOUNT);
1081 return ERR_PTR(-ENOMEM);
1083 bpf_map_init_from_attr(&htab->map, attr);
1085 htab->buckets_num = roundup_pow_of_two(htab->map.max_entries);
1086 htab->elem_size = sizeof(struct bpf_shtab_elem) +
1087 round_up(htab->map.key_size, 8);
1088 if (htab->buckets_num == 0 ||
1089 htab->buckets_num > U32_MAX / sizeof(struct bpf_shtab_bucket)) {
1094 htab->buckets = bpf_map_area_alloc(htab->buckets_num *
1095 sizeof(struct bpf_shtab_bucket),
1096 htab->map.numa_node);
1097 if (!htab->buckets) {
1102 for (i = 0; i < htab->buckets_num; i++) {
1103 INIT_HLIST_HEAD(&htab->buckets[i].head);
1104 raw_spin_lock_init(&htab->buckets[i].lock);
1110 return ERR_PTR(err);
1113 static void sock_hash_free(struct bpf_map *map)
1115 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1116 struct bpf_shtab_bucket *bucket;
1117 struct hlist_head unlink_list;
1118 struct bpf_shtab_elem *elem;
1119 struct hlist_node *node;
1122 /* After the sync no updates or deletes will be in-flight so it
1123 * is safe to walk map and remove entries without risking a race
1124 * in EEXIST update case.
1127 for (i = 0; i < htab->buckets_num; i++) {
1128 bucket = sock_hash_select_bucket(htab, i);
1130 /* We are racing with sock_hash_delete_from_link to
1131 * enter the spin-lock critical section. Every socket on
1132 * the list is still linked to sockhash. Since link
1133 * exists, psock exists and holds a ref to socket. That
1134 * lets us to grab a socket ref too.
1136 raw_spin_lock_bh(&bucket->lock);
1137 hlist_for_each_entry(elem, &bucket->head, node)
1138 sock_hold(elem->sk);
1139 hlist_move_list(&bucket->head, &unlink_list);
1140 raw_spin_unlock_bh(&bucket->lock);
1142 /* Process removed entries out of atomic context to
1143 * block for socket lock before deleting the psock's
1146 hlist_for_each_entry_safe(elem, node, &unlink_list, node) {
1147 hlist_del(&elem->node);
1148 lock_sock(elem->sk);
1150 sock_map_unref(elem->sk, elem);
1152 release_sock(elem->sk);
1154 sock_hash_free_elem(htab, elem);
1158 /* wait for psock readers accessing its map link */
1161 bpf_map_area_free(htab->buckets);
1165 static void *sock_hash_lookup_sys(struct bpf_map *map, void *key)
1169 if (map->value_size != sizeof(u64))
1170 return ERR_PTR(-ENOSPC);
1172 sk = __sock_hash_lookup_elem(map, key);
1174 return ERR_PTR(-ENOENT);
1176 __sock_gen_cookie(sk);
1177 return &sk->sk_cookie;
1180 static void *sock_hash_lookup(struct bpf_map *map, void *key)
1184 sk = __sock_hash_lookup_elem(map, key);
1187 if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
1192 static void sock_hash_release_progs(struct bpf_map *map)
1194 psock_progs_drop(&container_of(map, struct bpf_shtab, map)->progs);
1197 BPF_CALL_4(bpf_sock_hash_update, struct bpf_sock_ops_kern *, sops,
1198 struct bpf_map *, map, void *, key, u64, flags)
1200 WARN_ON_ONCE(!rcu_read_lock_held());
1202 if (likely(sock_map_sk_is_suitable(sops->sk) &&
1203 sock_map_op_okay(sops)))
1204 return sock_hash_update_common(map, key, sops->sk, flags);
1208 const struct bpf_func_proto bpf_sock_hash_update_proto = {
1209 .func = bpf_sock_hash_update,
1212 .ret_type = RET_INTEGER,
1213 .arg1_type = ARG_PTR_TO_CTX,
1214 .arg2_type = ARG_CONST_MAP_PTR,
1215 .arg3_type = ARG_PTR_TO_MAP_KEY,
1216 .arg4_type = ARG_ANYTHING,
1219 BPF_CALL_4(bpf_sk_redirect_hash, struct sk_buff *, skb,
1220 struct bpf_map *, map, void *, key, u64, flags)
1224 if (unlikely(flags & ~(BPF_F_INGRESS)))
1227 sk = __sock_hash_lookup_elem(map, key);
1228 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1231 skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
1235 const struct bpf_func_proto bpf_sk_redirect_hash_proto = {
1236 .func = bpf_sk_redirect_hash,
1238 .ret_type = RET_INTEGER,
1239 .arg1_type = ARG_PTR_TO_CTX,
1240 .arg2_type = ARG_CONST_MAP_PTR,
1241 .arg3_type = ARG_PTR_TO_MAP_KEY,
1242 .arg4_type = ARG_ANYTHING,
1245 BPF_CALL_4(bpf_msg_redirect_hash, struct sk_msg *, msg,
1246 struct bpf_map *, map, void *, key, u64, flags)
1250 if (unlikely(flags & ~(BPF_F_INGRESS)))
1253 sk = __sock_hash_lookup_elem(map, key);
1254 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1262 const struct bpf_func_proto bpf_msg_redirect_hash_proto = {
1263 .func = bpf_msg_redirect_hash,
1265 .ret_type = RET_INTEGER,
1266 .arg1_type = ARG_PTR_TO_CTX,
1267 .arg2_type = ARG_CONST_MAP_PTR,
1268 .arg3_type = ARG_PTR_TO_MAP_KEY,
1269 .arg4_type = ARG_ANYTHING,
1272 struct sock_hash_seq_info {
1273 struct bpf_map *map;
1274 struct bpf_shtab *htab;
1278 static void *sock_hash_seq_find_next(struct sock_hash_seq_info *info,
1279 struct bpf_shtab_elem *prev_elem)
1281 const struct bpf_shtab *htab = info->htab;
1282 struct bpf_shtab_bucket *bucket;
1283 struct bpf_shtab_elem *elem;
1284 struct hlist_node *node;
1286 /* try to find next elem in the same bucket */
1288 node = rcu_dereference(hlist_next_rcu(&prev_elem->node));
1289 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1293 /* no more elements, continue in the next bucket */
1297 for (; info->bucket_id < htab->buckets_num; info->bucket_id++) {
1298 bucket = &htab->buckets[info->bucket_id];
1299 node = rcu_dereference(hlist_first_rcu(&bucket->head));
1300 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1308 static void *sock_hash_seq_start(struct seq_file *seq, loff_t *pos)
1311 struct sock_hash_seq_info *info = seq->private;
1316 /* pairs with sock_hash_seq_stop */
1318 return sock_hash_seq_find_next(info, NULL);
1321 static void *sock_hash_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1324 struct sock_hash_seq_info *info = seq->private;
1327 return sock_hash_seq_find_next(info, v);
1330 static int sock_hash_seq_show(struct seq_file *seq, void *v)
1333 struct sock_hash_seq_info *info = seq->private;
1334 struct bpf_iter__sockmap ctx = {};
1335 struct bpf_shtab_elem *elem = v;
1336 struct bpf_iter_meta meta;
1337 struct bpf_prog *prog;
1340 prog = bpf_iter_get_info(&meta, !elem);
1345 ctx.map = info->map;
1347 ctx.key = elem->key;
1351 return bpf_iter_run_prog(prog, &ctx);
1354 static void sock_hash_seq_stop(struct seq_file *seq, void *v)
1358 (void)sock_hash_seq_show(seq, NULL);
1360 /* pairs with sock_hash_seq_start */
1364 static const struct seq_operations sock_hash_seq_ops = {
1365 .start = sock_hash_seq_start,
1366 .next = sock_hash_seq_next,
1367 .stop = sock_hash_seq_stop,
1368 .show = sock_hash_seq_show,
1371 static int sock_hash_init_seq_private(void *priv_data,
1372 struct bpf_iter_aux_info *aux)
1374 struct sock_hash_seq_info *info = priv_data;
1376 info->map = aux->map;
1377 info->htab = container_of(aux->map, struct bpf_shtab, map);
1381 static const struct bpf_iter_seq_info sock_hash_iter_seq_info = {
1382 .seq_ops = &sock_hash_seq_ops,
1383 .init_seq_private = sock_hash_init_seq_private,
1384 .seq_priv_size = sizeof(struct sock_hash_seq_info),
1387 BTF_ID_LIST_SINGLE(sock_hash_map_btf_ids, struct, bpf_shtab)
1388 const struct bpf_map_ops sock_hash_ops = {
1389 .map_meta_equal = bpf_map_meta_equal,
1390 .map_alloc = sock_hash_alloc,
1391 .map_free = sock_hash_free,
1392 .map_get_next_key = sock_hash_get_next_key,
1393 .map_update_elem = sock_map_update_elem,
1394 .map_delete_elem = sock_hash_delete_elem,
1395 .map_lookup_elem = sock_hash_lookup,
1396 .map_lookup_elem_sys_only = sock_hash_lookup_sys,
1397 .map_release_uref = sock_hash_release_progs,
1398 .map_check_btf = map_check_no_btf,
1399 .map_btf_id = &sock_hash_map_btf_ids[0],
1400 .iter_seq_info = &sock_hash_iter_seq_info,
1403 static struct sk_psock_progs *sock_map_progs(struct bpf_map *map)
1405 switch (map->map_type) {
1406 case BPF_MAP_TYPE_SOCKMAP:
1407 return &container_of(map, struct bpf_stab, map)->progs;
1408 case BPF_MAP_TYPE_SOCKHASH:
1409 return &container_of(map, struct bpf_shtab, map)->progs;
1417 static int sock_map_prog_lookup(struct bpf_map *map, struct bpf_prog ***pprog,
1420 struct sk_psock_progs *progs = sock_map_progs(map);
1426 case BPF_SK_MSG_VERDICT:
1427 *pprog = &progs->msg_parser;
1429 #if IS_ENABLED(CONFIG_BPF_STREAM_PARSER)
1430 case BPF_SK_SKB_STREAM_PARSER:
1431 *pprog = &progs->stream_parser;
1434 case BPF_SK_SKB_STREAM_VERDICT:
1435 if (progs->skb_verdict)
1437 *pprog = &progs->stream_verdict;
1439 case BPF_SK_SKB_VERDICT:
1440 if (progs->stream_verdict)
1442 *pprog = &progs->skb_verdict;
1451 static int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog,
1452 struct bpf_prog *old, u32 which)
1454 struct bpf_prog **pprog;
1457 ret = sock_map_prog_lookup(map, &pprog, which);
1462 return psock_replace_prog(pprog, prog, old);
1464 psock_set_prog(pprog, prog);
1468 int sock_map_bpf_prog_query(const union bpf_attr *attr,
1469 union bpf_attr __user *uattr)
1471 __u32 __user *prog_ids = u64_to_user_ptr(attr->query.prog_ids);
1472 u32 prog_cnt = 0, flags = 0, ufd = attr->target_fd;
1473 struct bpf_prog **pprog;
1474 struct bpf_prog *prog;
1475 struct bpf_map *map;
1480 if (attr->query.query_flags)
1484 map = __bpf_map_get(f);
1486 return PTR_ERR(map);
1490 ret = sock_map_prog_lookup(map, &pprog, attr->query.attach_type);
1495 prog_cnt = !prog ? 0 : 1;
1497 if (!attr->query.prog_cnt || !prog_ids || !prog_cnt)
1500 /* we do not hold the refcnt, the bpf prog may be released
1501 * asynchronously and the id would be set to 0.
1503 id = data_race(prog->aux->id);
1510 if (copy_to_user(&uattr->query.attach_flags, &flags, sizeof(flags)) ||
1511 (id != 0 && copy_to_user(prog_ids, &id, sizeof(u32))) ||
1512 copy_to_user(&uattr->query.prog_cnt, &prog_cnt, sizeof(prog_cnt)))
1519 static void sock_map_unlink(struct sock *sk, struct sk_psock_link *link)
1521 switch (link->map->map_type) {
1522 case BPF_MAP_TYPE_SOCKMAP:
1523 return sock_map_delete_from_link(link->map, sk,
1525 case BPF_MAP_TYPE_SOCKHASH:
1526 return sock_hash_delete_from_link(link->map, sk,
1533 static void sock_map_remove_links(struct sock *sk, struct sk_psock *psock)
1535 struct sk_psock_link *link;
1537 while ((link = sk_psock_link_pop(psock))) {
1538 sock_map_unlink(sk, link);
1539 sk_psock_free_link(link);
1543 void sock_map_unhash(struct sock *sk)
1545 void (*saved_unhash)(struct sock *sk);
1546 struct sk_psock *psock;
1549 psock = sk_psock(sk);
1550 if (unlikely(!psock)) {
1552 if (sk->sk_prot->unhash)
1553 sk->sk_prot->unhash(sk);
1557 saved_unhash = psock->saved_unhash;
1558 sock_map_remove_links(sk, psock);
1562 EXPORT_SYMBOL_GPL(sock_map_unhash);
1564 void sock_map_close(struct sock *sk, long timeout)
1566 void (*saved_close)(struct sock *sk, long timeout);
1567 struct sk_psock *psock;
1571 psock = sk_psock_get(sk);
1572 if (unlikely(!psock)) {
1575 return sk->sk_prot->close(sk, timeout);
1578 saved_close = psock->saved_close;
1579 sock_map_remove_links(sk, psock);
1581 sk_psock_stop(psock, true);
1582 sk_psock_put(sk, psock);
1584 saved_close(sk, timeout);
1586 EXPORT_SYMBOL_GPL(sock_map_close);
1588 static int sock_map_iter_attach_target(struct bpf_prog *prog,
1589 union bpf_iter_link_info *linfo,
1590 struct bpf_iter_aux_info *aux)
1592 struct bpf_map *map;
1595 if (!linfo->map.map_fd)
1598 map = bpf_map_get_with_uref(linfo->map.map_fd);
1600 return PTR_ERR(map);
1602 if (map->map_type != BPF_MAP_TYPE_SOCKMAP &&
1603 map->map_type != BPF_MAP_TYPE_SOCKHASH)
1606 if (prog->aux->max_rdonly_access > map->key_size) {
1615 bpf_map_put_with_uref(map);
1619 static void sock_map_iter_detach_target(struct bpf_iter_aux_info *aux)
1621 bpf_map_put_with_uref(aux->map);
1624 static struct bpf_iter_reg sock_map_iter_reg = {
1625 .target = "sockmap",
1626 .attach_target = sock_map_iter_attach_target,
1627 .detach_target = sock_map_iter_detach_target,
1628 .show_fdinfo = bpf_iter_map_show_fdinfo,
1629 .fill_link_info = bpf_iter_map_fill_link_info,
1630 .ctx_arg_info_size = 2,
1632 { offsetof(struct bpf_iter__sockmap, key),
1633 PTR_TO_BUF | PTR_MAYBE_NULL | MEM_RDONLY },
1634 { offsetof(struct bpf_iter__sockmap, sk),
1635 PTR_TO_BTF_ID_OR_NULL },
1639 static int __init bpf_sockmap_iter_init(void)
1641 sock_map_iter_reg.ctx_arg_info[1].btf_id =
1642 btf_sock_ids[BTF_SOCK_TYPE_SOCK];
1643 return bpf_iter_reg_target(&sock_map_iter_reg);
1645 late_initcall(bpf_sockmap_iter_init);