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 static int sock_map_btf_id;
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_name = "bpf_stab",
809 .map_btf_id = &sock_map_btf_id,
810 .iter_seq_info = &sock_map_iter_seq_info,
813 struct bpf_shtab_elem {
817 struct hlist_node node;
821 struct bpf_shtab_bucket {
822 struct hlist_head head;
828 struct bpf_shtab_bucket *buckets;
831 struct sk_psock_progs progs;
835 static inline u32 sock_hash_bucket_hash(const void *key, u32 len)
837 return jhash(key, len, 0);
840 static struct bpf_shtab_bucket *sock_hash_select_bucket(struct bpf_shtab *htab,
843 return &htab->buckets[hash & (htab->buckets_num - 1)];
846 static struct bpf_shtab_elem *
847 sock_hash_lookup_elem_raw(struct hlist_head *head, u32 hash, void *key,
850 struct bpf_shtab_elem *elem;
852 hlist_for_each_entry_rcu(elem, head, node) {
853 if (elem->hash == hash &&
854 !memcmp(&elem->key, key, key_size))
861 static struct sock *__sock_hash_lookup_elem(struct bpf_map *map, void *key)
863 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
864 u32 key_size = map->key_size, hash;
865 struct bpf_shtab_bucket *bucket;
866 struct bpf_shtab_elem *elem;
868 WARN_ON_ONCE(!rcu_read_lock_held());
870 hash = sock_hash_bucket_hash(key, key_size);
871 bucket = sock_hash_select_bucket(htab, hash);
872 elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
874 return elem ? elem->sk : NULL;
877 static void sock_hash_free_elem(struct bpf_shtab *htab,
878 struct bpf_shtab_elem *elem)
880 atomic_dec(&htab->count);
881 kfree_rcu(elem, rcu);
884 static void sock_hash_delete_from_link(struct bpf_map *map, struct sock *sk,
887 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
888 struct bpf_shtab_elem *elem_probe, *elem = link_raw;
889 struct bpf_shtab_bucket *bucket;
891 WARN_ON_ONCE(!rcu_read_lock_held());
892 bucket = sock_hash_select_bucket(htab, elem->hash);
894 /* elem may be deleted in parallel from the map, but access here
895 * is okay since it's going away only after RCU grace period.
896 * However, we need to check whether it's still present.
898 raw_spin_lock_bh(&bucket->lock);
899 elem_probe = sock_hash_lookup_elem_raw(&bucket->head, elem->hash,
900 elem->key, map->key_size);
901 if (elem_probe && elem_probe == elem) {
902 hlist_del_rcu(&elem->node);
903 sock_map_unref(elem->sk, elem);
904 sock_hash_free_elem(htab, elem);
906 raw_spin_unlock_bh(&bucket->lock);
909 static int sock_hash_delete_elem(struct bpf_map *map, void *key)
911 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
912 u32 hash, key_size = map->key_size;
913 struct bpf_shtab_bucket *bucket;
914 struct bpf_shtab_elem *elem;
917 hash = sock_hash_bucket_hash(key, key_size);
918 bucket = sock_hash_select_bucket(htab, hash);
920 raw_spin_lock_bh(&bucket->lock);
921 elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
923 hlist_del_rcu(&elem->node);
924 sock_map_unref(elem->sk, elem);
925 sock_hash_free_elem(htab, elem);
928 raw_spin_unlock_bh(&bucket->lock);
932 static struct bpf_shtab_elem *sock_hash_alloc_elem(struct bpf_shtab *htab,
933 void *key, u32 key_size,
934 u32 hash, struct sock *sk,
935 struct bpf_shtab_elem *old)
937 struct bpf_shtab_elem *new;
939 if (atomic_inc_return(&htab->count) > htab->map.max_entries) {
941 atomic_dec(&htab->count);
942 return ERR_PTR(-E2BIG);
946 new = bpf_map_kmalloc_node(&htab->map, htab->elem_size,
947 GFP_ATOMIC | __GFP_NOWARN,
948 htab->map.numa_node);
950 atomic_dec(&htab->count);
951 return ERR_PTR(-ENOMEM);
953 memcpy(new->key, key, key_size);
959 static int sock_hash_update_common(struct bpf_map *map, void *key,
960 struct sock *sk, u64 flags)
962 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
963 u32 key_size = map->key_size, hash;
964 struct bpf_shtab_elem *elem, *elem_new;
965 struct bpf_shtab_bucket *bucket;
966 struct sk_psock_link *link;
967 struct sk_psock *psock;
970 WARN_ON_ONCE(!rcu_read_lock_held());
971 if (unlikely(flags > BPF_EXIST))
974 link = sk_psock_init_link();
978 ret = sock_map_link(map, sk);
982 psock = sk_psock(sk);
983 WARN_ON_ONCE(!psock);
985 hash = sock_hash_bucket_hash(key, key_size);
986 bucket = sock_hash_select_bucket(htab, hash);
988 raw_spin_lock_bh(&bucket->lock);
989 elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
990 if (elem && flags == BPF_NOEXIST) {
993 } else if (!elem && flags == BPF_EXIST) {
998 elem_new = sock_hash_alloc_elem(htab, key, key_size, hash, sk, elem);
999 if (IS_ERR(elem_new)) {
1000 ret = PTR_ERR(elem_new);
1004 sock_map_add_link(psock, link, map, elem_new);
1005 /* Add new element to the head of the list, so that
1006 * concurrent search will find it before old elem.
1008 hlist_add_head_rcu(&elem_new->node, &bucket->head);
1010 hlist_del_rcu(&elem->node);
1011 sock_map_unref(elem->sk, elem);
1012 sock_hash_free_elem(htab, elem);
1014 raw_spin_unlock_bh(&bucket->lock);
1017 raw_spin_unlock_bh(&bucket->lock);
1018 sk_psock_put(sk, psock);
1020 sk_psock_free_link(link);
1024 static int sock_hash_get_next_key(struct bpf_map *map, void *key,
1027 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1028 struct bpf_shtab_elem *elem, *elem_next;
1029 u32 hash, key_size = map->key_size;
1030 struct hlist_head *head;
1034 goto find_first_elem;
1035 hash = sock_hash_bucket_hash(key, key_size);
1036 head = &sock_hash_select_bucket(htab, hash)->head;
1037 elem = sock_hash_lookup_elem_raw(head, hash, key, key_size);
1039 goto find_first_elem;
1041 elem_next = hlist_entry_safe(rcu_dereference(hlist_next_rcu(&elem->node)),
1042 struct bpf_shtab_elem, node);
1044 memcpy(key_next, elem_next->key, key_size);
1048 i = hash & (htab->buckets_num - 1);
1051 for (; i < htab->buckets_num; i++) {
1052 head = &sock_hash_select_bucket(htab, i)->head;
1053 elem_next = hlist_entry_safe(rcu_dereference(hlist_first_rcu(head)),
1054 struct bpf_shtab_elem, node);
1056 memcpy(key_next, elem_next->key, key_size);
1064 static struct bpf_map *sock_hash_alloc(union bpf_attr *attr)
1066 struct bpf_shtab *htab;
1069 if (!capable(CAP_NET_ADMIN))
1070 return ERR_PTR(-EPERM);
1071 if (attr->max_entries == 0 ||
1072 attr->key_size == 0 ||
1073 (attr->value_size != sizeof(u32) &&
1074 attr->value_size != sizeof(u64)) ||
1075 attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
1076 return ERR_PTR(-EINVAL);
1077 if (attr->key_size > MAX_BPF_STACK)
1078 return ERR_PTR(-E2BIG);
1080 htab = kzalloc(sizeof(*htab), GFP_USER | __GFP_ACCOUNT);
1082 return ERR_PTR(-ENOMEM);
1084 bpf_map_init_from_attr(&htab->map, attr);
1086 htab->buckets_num = roundup_pow_of_two(htab->map.max_entries);
1087 htab->elem_size = sizeof(struct bpf_shtab_elem) +
1088 round_up(htab->map.key_size, 8);
1089 if (htab->buckets_num == 0 ||
1090 htab->buckets_num > U32_MAX / sizeof(struct bpf_shtab_bucket)) {
1095 htab->buckets = bpf_map_area_alloc(htab->buckets_num *
1096 sizeof(struct bpf_shtab_bucket),
1097 htab->map.numa_node);
1098 if (!htab->buckets) {
1103 for (i = 0; i < htab->buckets_num; i++) {
1104 INIT_HLIST_HEAD(&htab->buckets[i].head);
1105 raw_spin_lock_init(&htab->buckets[i].lock);
1111 return ERR_PTR(err);
1114 static void sock_hash_free(struct bpf_map *map)
1116 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1117 struct bpf_shtab_bucket *bucket;
1118 struct hlist_head unlink_list;
1119 struct bpf_shtab_elem *elem;
1120 struct hlist_node *node;
1123 /* After the sync no updates or deletes will be in-flight so it
1124 * is safe to walk map and remove entries without risking a race
1125 * in EEXIST update case.
1128 for (i = 0; i < htab->buckets_num; i++) {
1129 bucket = sock_hash_select_bucket(htab, i);
1131 /* We are racing with sock_hash_delete_from_link to
1132 * enter the spin-lock critical section. Every socket on
1133 * the list is still linked to sockhash. Since link
1134 * exists, psock exists and holds a ref to socket. That
1135 * lets us to grab a socket ref too.
1137 raw_spin_lock_bh(&bucket->lock);
1138 hlist_for_each_entry(elem, &bucket->head, node)
1139 sock_hold(elem->sk);
1140 hlist_move_list(&bucket->head, &unlink_list);
1141 raw_spin_unlock_bh(&bucket->lock);
1143 /* Process removed entries out of atomic context to
1144 * block for socket lock before deleting the psock's
1147 hlist_for_each_entry_safe(elem, node, &unlink_list, node) {
1148 hlist_del(&elem->node);
1149 lock_sock(elem->sk);
1151 sock_map_unref(elem->sk, elem);
1153 release_sock(elem->sk);
1155 sock_hash_free_elem(htab, elem);
1159 /* wait for psock readers accessing its map link */
1162 bpf_map_area_free(htab->buckets);
1166 static void *sock_hash_lookup_sys(struct bpf_map *map, void *key)
1170 if (map->value_size != sizeof(u64))
1171 return ERR_PTR(-ENOSPC);
1173 sk = __sock_hash_lookup_elem(map, key);
1175 return ERR_PTR(-ENOENT);
1177 __sock_gen_cookie(sk);
1178 return &sk->sk_cookie;
1181 static void *sock_hash_lookup(struct bpf_map *map, void *key)
1185 sk = __sock_hash_lookup_elem(map, key);
1188 if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
1193 static void sock_hash_release_progs(struct bpf_map *map)
1195 psock_progs_drop(&container_of(map, struct bpf_shtab, map)->progs);
1198 BPF_CALL_4(bpf_sock_hash_update, struct bpf_sock_ops_kern *, sops,
1199 struct bpf_map *, map, void *, key, u64, flags)
1201 WARN_ON_ONCE(!rcu_read_lock_held());
1203 if (likely(sock_map_sk_is_suitable(sops->sk) &&
1204 sock_map_op_okay(sops)))
1205 return sock_hash_update_common(map, key, sops->sk, flags);
1209 const struct bpf_func_proto bpf_sock_hash_update_proto = {
1210 .func = bpf_sock_hash_update,
1213 .ret_type = RET_INTEGER,
1214 .arg1_type = ARG_PTR_TO_CTX,
1215 .arg2_type = ARG_CONST_MAP_PTR,
1216 .arg3_type = ARG_PTR_TO_MAP_KEY,
1217 .arg4_type = ARG_ANYTHING,
1220 BPF_CALL_4(bpf_sk_redirect_hash, struct sk_buff *, skb,
1221 struct bpf_map *, map, void *, key, u64, flags)
1225 if (unlikely(flags & ~(BPF_F_INGRESS)))
1228 sk = __sock_hash_lookup_elem(map, key);
1229 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1232 skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
1236 const struct bpf_func_proto bpf_sk_redirect_hash_proto = {
1237 .func = bpf_sk_redirect_hash,
1239 .ret_type = RET_INTEGER,
1240 .arg1_type = ARG_PTR_TO_CTX,
1241 .arg2_type = ARG_CONST_MAP_PTR,
1242 .arg3_type = ARG_PTR_TO_MAP_KEY,
1243 .arg4_type = ARG_ANYTHING,
1246 BPF_CALL_4(bpf_msg_redirect_hash, struct sk_msg *, msg,
1247 struct bpf_map *, map, void *, key, u64, flags)
1251 if (unlikely(flags & ~(BPF_F_INGRESS)))
1254 sk = __sock_hash_lookup_elem(map, key);
1255 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1263 const struct bpf_func_proto bpf_msg_redirect_hash_proto = {
1264 .func = bpf_msg_redirect_hash,
1266 .ret_type = RET_INTEGER,
1267 .arg1_type = ARG_PTR_TO_CTX,
1268 .arg2_type = ARG_CONST_MAP_PTR,
1269 .arg3_type = ARG_PTR_TO_MAP_KEY,
1270 .arg4_type = ARG_ANYTHING,
1273 struct sock_hash_seq_info {
1274 struct bpf_map *map;
1275 struct bpf_shtab *htab;
1279 static void *sock_hash_seq_find_next(struct sock_hash_seq_info *info,
1280 struct bpf_shtab_elem *prev_elem)
1282 const struct bpf_shtab *htab = info->htab;
1283 struct bpf_shtab_bucket *bucket;
1284 struct bpf_shtab_elem *elem;
1285 struct hlist_node *node;
1287 /* try to find next elem in the same bucket */
1289 node = rcu_dereference(hlist_next_rcu(&prev_elem->node));
1290 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1294 /* no more elements, continue in the next bucket */
1298 for (; info->bucket_id < htab->buckets_num; info->bucket_id++) {
1299 bucket = &htab->buckets[info->bucket_id];
1300 node = rcu_dereference(hlist_first_rcu(&bucket->head));
1301 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1309 static void *sock_hash_seq_start(struct seq_file *seq, loff_t *pos)
1312 struct sock_hash_seq_info *info = seq->private;
1317 /* pairs with sock_hash_seq_stop */
1319 return sock_hash_seq_find_next(info, NULL);
1322 static void *sock_hash_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1325 struct sock_hash_seq_info *info = seq->private;
1328 return sock_hash_seq_find_next(info, v);
1331 static int sock_hash_seq_show(struct seq_file *seq, void *v)
1334 struct sock_hash_seq_info *info = seq->private;
1335 struct bpf_iter__sockmap ctx = {};
1336 struct bpf_shtab_elem *elem = v;
1337 struct bpf_iter_meta meta;
1338 struct bpf_prog *prog;
1341 prog = bpf_iter_get_info(&meta, !elem);
1346 ctx.map = info->map;
1348 ctx.key = elem->key;
1352 return bpf_iter_run_prog(prog, &ctx);
1355 static void sock_hash_seq_stop(struct seq_file *seq, void *v)
1359 (void)sock_hash_seq_show(seq, NULL);
1361 /* pairs with sock_hash_seq_start */
1365 static const struct seq_operations sock_hash_seq_ops = {
1366 .start = sock_hash_seq_start,
1367 .next = sock_hash_seq_next,
1368 .stop = sock_hash_seq_stop,
1369 .show = sock_hash_seq_show,
1372 static int sock_hash_init_seq_private(void *priv_data,
1373 struct bpf_iter_aux_info *aux)
1375 struct sock_hash_seq_info *info = priv_data;
1377 info->map = aux->map;
1378 info->htab = container_of(aux->map, struct bpf_shtab, map);
1382 static const struct bpf_iter_seq_info sock_hash_iter_seq_info = {
1383 .seq_ops = &sock_hash_seq_ops,
1384 .init_seq_private = sock_hash_init_seq_private,
1385 .seq_priv_size = sizeof(struct sock_hash_seq_info),
1388 static int sock_hash_map_btf_id;
1389 const struct bpf_map_ops sock_hash_ops = {
1390 .map_meta_equal = bpf_map_meta_equal,
1391 .map_alloc = sock_hash_alloc,
1392 .map_free = sock_hash_free,
1393 .map_get_next_key = sock_hash_get_next_key,
1394 .map_update_elem = sock_map_update_elem,
1395 .map_delete_elem = sock_hash_delete_elem,
1396 .map_lookup_elem = sock_hash_lookup,
1397 .map_lookup_elem_sys_only = sock_hash_lookup_sys,
1398 .map_release_uref = sock_hash_release_progs,
1399 .map_check_btf = map_check_no_btf,
1400 .map_btf_name = "bpf_shtab",
1401 .map_btf_id = &sock_hash_map_btf_id,
1402 .iter_seq_info = &sock_hash_iter_seq_info,
1405 static struct sk_psock_progs *sock_map_progs(struct bpf_map *map)
1407 switch (map->map_type) {
1408 case BPF_MAP_TYPE_SOCKMAP:
1409 return &container_of(map, struct bpf_stab, map)->progs;
1410 case BPF_MAP_TYPE_SOCKHASH:
1411 return &container_of(map, struct bpf_shtab, map)->progs;
1419 static int sock_map_prog_lookup(struct bpf_map *map, struct bpf_prog ***pprog,
1422 struct sk_psock_progs *progs = sock_map_progs(map);
1428 case BPF_SK_MSG_VERDICT:
1429 *pprog = &progs->msg_parser;
1431 #if IS_ENABLED(CONFIG_BPF_STREAM_PARSER)
1432 case BPF_SK_SKB_STREAM_PARSER:
1433 *pprog = &progs->stream_parser;
1436 case BPF_SK_SKB_STREAM_VERDICT:
1437 if (progs->skb_verdict)
1439 *pprog = &progs->stream_verdict;
1441 case BPF_SK_SKB_VERDICT:
1442 if (progs->stream_verdict)
1444 *pprog = &progs->skb_verdict;
1453 static int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog,
1454 struct bpf_prog *old, u32 which)
1456 struct bpf_prog **pprog;
1459 ret = sock_map_prog_lookup(map, &pprog, which);
1464 return psock_replace_prog(pprog, prog, old);
1466 psock_set_prog(pprog, prog);
1470 int sock_map_bpf_prog_query(const union bpf_attr *attr,
1471 union bpf_attr __user *uattr)
1473 __u32 __user *prog_ids = u64_to_user_ptr(attr->query.prog_ids);
1474 u32 prog_cnt = 0, flags = 0, ufd = attr->target_fd;
1475 struct bpf_prog **pprog;
1476 struct bpf_prog *prog;
1477 struct bpf_map *map;
1482 if (attr->query.query_flags)
1486 map = __bpf_map_get(f);
1488 return PTR_ERR(map);
1492 ret = sock_map_prog_lookup(map, &pprog, attr->query.attach_type);
1497 prog_cnt = !prog ? 0 : 1;
1499 if (!attr->query.prog_cnt || !prog_ids || !prog_cnt)
1502 /* we do not hold the refcnt, the bpf prog may be released
1503 * asynchronously and the id would be set to 0.
1505 id = data_race(prog->aux->id);
1512 if (copy_to_user(&uattr->query.attach_flags, &flags, sizeof(flags)) ||
1513 (id != 0 && copy_to_user(prog_ids, &id, sizeof(u32))) ||
1514 copy_to_user(&uattr->query.prog_cnt, &prog_cnt, sizeof(prog_cnt)))
1521 static void sock_map_unlink(struct sock *sk, struct sk_psock_link *link)
1523 switch (link->map->map_type) {
1524 case BPF_MAP_TYPE_SOCKMAP:
1525 return sock_map_delete_from_link(link->map, sk,
1527 case BPF_MAP_TYPE_SOCKHASH:
1528 return sock_hash_delete_from_link(link->map, sk,
1535 static void sock_map_remove_links(struct sock *sk, struct sk_psock *psock)
1537 struct sk_psock_link *link;
1539 while ((link = sk_psock_link_pop(psock))) {
1540 sock_map_unlink(sk, link);
1541 sk_psock_free_link(link);
1545 void sock_map_unhash(struct sock *sk)
1547 void (*saved_unhash)(struct sock *sk);
1548 struct sk_psock *psock;
1551 psock = sk_psock(sk);
1552 if (unlikely(!psock)) {
1554 if (sk->sk_prot->unhash)
1555 sk->sk_prot->unhash(sk);
1559 saved_unhash = psock->saved_unhash;
1560 sock_map_remove_links(sk, psock);
1564 EXPORT_SYMBOL_GPL(sock_map_unhash);
1566 void sock_map_close(struct sock *sk, long timeout)
1568 void (*saved_close)(struct sock *sk, long timeout);
1569 struct sk_psock *psock;
1573 psock = sk_psock_get(sk);
1574 if (unlikely(!psock)) {
1577 return sk->sk_prot->close(sk, timeout);
1580 saved_close = psock->saved_close;
1581 sock_map_remove_links(sk, psock);
1583 sk_psock_stop(psock, true);
1584 sk_psock_put(sk, psock);
1586 saved_close(sk, timeout);
1588 EXPORT_SYMBOL_GPL(sock_map_close);
1590 static int sock_map_iter_attach_target(struct bpf_prog *prog,
1591 union bpf_iter_link_info *linfo,
1592 struct bpf_iter_aux_info *aux)
1594 struct bpf_map *map;
1597 if (!linfo->map.map_fd)
1600 map = bpf_map_get_with_uref(linfo->map.map_fd);
1602 return PTR_ERR(map);
1604 if (map->map_type != BPF_MAP_TYPE_SOCKMAP &&
1605 map->map_type != BPF_MAP_TYPE_SOCKHASH)
1608 if (prog->aux->max_rdonly_access > map->key_size) {
1617 bpf_map_put_with_uref(map);
1621 static void sock_map_iter_detach_target(struct bpf_iter_aux_info *aux)
1623 bpf_map_put_with_uref(aux->map);
1626 static struct bpf_iter_reg sock_map_iter_reg = {
1627 .target = "sockmap",
1628 .attach_target = sock_map_iter_attach_target,
1629 .detach_target = sock_map_iter_detach_target,
1630 .show_fdinfo = bpf_iter_map_show_fdinfo,
1631 .fill_link_info = bpf_iter_map_fill_link_info,
1632 .ctx_arg_info_size = 2,
1634 { offsetof(struct bpf_iter__sockmap, key),
1635 PTR_TO_BUF | PTR_MAYBE_NULL | MEM_RDONLY },
1636 { offsetof(struct bpf_iter__sockmap, sk),
1637 PTR_TO_BTF_ID_OR_NULL },
1641 static int __init bpf_sockmap_iter_init(void)
1643 sock_map_iter_reg.ctx_arg_info[1].btf_id =
1644 btf_sock_ids[BTF_SOCK_TYPE_SOCK];
1645 return bpf_iter_reg_target(&sock_map_iter_reg);
1647 late_initcall(bpf_sockmap_iter_init);