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 struct bpf_map *sock_map_alloc(union bpf_attr *attr)
29 struct bpf_stab *stab;
31 if (!capable(CAP_NET_ADMIN))
32 return ERR_PTR(-EPERM);
33 if (attr->max_entries == 0 ||
34 attr->key_size != 4 ||
35 (attr->value_size != sizeof(u32) &&
36 attr->value_size != sizeof(u64)) ||
37 attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
38 return ERR_PTR(-EINVAL);
40 stab = kzalloc(sizeof(*stab), GFP_USER | __GFP_ACCOUNT);
42 return ERR_PTR(-ENOMEM);
44 bpf_map_init_from_attr(&stab->map, attr);
45 raw_spin_lock_init(&stab->lock);
47 stab->sks = bpf_map_area_alloc(stab->map.max_entries *
48 sizeof(struct sock *),
52 return ERR_PTR(-ENOMEM);
58 int sock_map_get_from_fd(const union bpf_attr *attr, struct bpf_prog *prog)
60 u32 ufd = attr->target_fd;
65 if (attr->attach_flags || attr->replace_bpf_fd)
69 map = __bpf_map_get(f);
72 ret = sock_map_prog_update(map, prog, NULL, attr->attach_type);
77 int sock_map_prog_detach(const union bpf_attr *attr, enum bpf_prog_type ptype)
79 u32 ufd = attr->target_fd;
80 struct bpf_prog *prog;
85 if (attr->attach_flags || attr->replace_bpf_fd)
89 map = __bpf_map_get(f);
93 prog = bpf_prog_get(attr->attach_bpf_fd);
99 if (prog->type != ptype) {
104 ret = sock_map_prog_update(map, NULL, prog, attr->attach_type);
112 static void sock_map_sk_acquire(struct sock *sk)
113 __acquires(&sk->sk_lock.slock)
120 static void sock_map_sk_release(struct sock *sk)
121 __releases(&sk->sk_lock.slock)
128 static void sock_map_add_link(struct sk_psock *psock,
129 struct sk_psock_link *link,
130 struct bpf_map *map, void *link_raw)
132 link->link_raw = link_raw;
134 spin_lock_bh(&psock->link_lock);
135 list_add_tail(&link->list, &psock->link);
136 spin_unlock_bh(&psock->link_lock);
139 static void sock_map_del_link(struct sock *sk,
140 struct sk_psock *psock, void *link_raw)
142 bool strp_stop = false, verdict_stop = false;
143 struct sk_psock_link *link, *tmp;
145 spin_lock_bh(&psock->link_lock);
146 list_for_each_entry_safe(link, tmp, &psock->link, list) {
147 if (link->link_raw == link_raw) {
148 struct bpf_map *map = link->map;
149 struct bpf_stab *stab = container_of(map, struct bpf_stab,
151 if (psock->saved_data_ready && stab->progs.stream_parser)
153 if (psock->saved_data_ready && stab->progs.stream_verdict)
155 list_del(&link->list);
156 sk_psock_free_link(link);
159 spin_unlock_bh(&psock->link_lock);
160 if (strp_stop || verdict_stop) {
161 write_lock_bh(&sk->sk_callback_lock);
163 sk_psock_stop_strp(sk, psock);
165 sk_psock_stop_verdict(sk, psock);
166 write_unlock_bh(&sk->sk_callback_lock);
170 static void sock_map_unref(struct sock *sk, void *link_raw)
172 struct sk_psock *psock = sk_psock(sk);
175 sock_map_del_link(sk, psock, link_raw);
176 sk_psock_put(sk, psock);
180 static int sock_map_init_proto(struct sock *sk, struct sk_psock *psock)
184 switch (sk->sk_type) {
186 prot = tcp_bpf_get_proto(sk, psock);
190 prot = udp_bpf_get_proto(sk, psock);
198 return PTR_ERR(prot);
200 sk_psock_update_proto(sk, psock, prot);
204 static struct sk_psock *sock_map_psock_get_checked(struct sock *sk)
206 struct sk_psock *psock;
209 psock = sk_psock(sk);
211 if (sk->sk_prot->close != sock_map_close) {
212 psock = ERR_PTR(-EBUSY);
216 if (!refcount_inc_not_zero(&psock->refcnt))
217 psock = ERR_PTR(-EBUSY);
224 static int sock_map_link(struct bpf_map *map, struct sk_psock_progs *progs,
227 struct bpf_prog *msg_parser, *stream_parser, *stream_verdict;
228 struct sk_psock *psock;
231 stream_verdict = READ_ONCE(progs->stream_verdict);
232 if (stream_verdict) {
233 stream_verdict = bpf_prog_inc_not_zero(stream_verdict);
234 if (IS_ERR(stream_verdict))
235 return PTR_ERR(stream_verdict);
238 stream_parser = READ_ONCE(progs->stream_parser);
240 stream_parser = bpf_prog_inc_not_zero(stream_parser);
241 if (IS_ERR(stream_parser)) {
242 ret = PTR_ERR(stream_parser);
243 goto out_put_stream_verdict;
247 msg_parser = READ_ONCE(progs->msg_parser);
249 msg_parser = bpf_prog_inc_not_zero(msg_parser);
250 if (IS_ERR(msg_parser)) {
251 ret = PTR_ERR(msg_parser);
252 goto out_put_stream_parser;
256 psock = sock_map_psock_get_checked(sk);
258 ret = PTR_ERR(psock);
263 if ((msg_parser && READ_ONCE(psock->progs.msg_parser)) ||
264 (stream_parser && READ_ONCE(psock->progs.stream_parser)) ||
265 (stream_verdict && READ_ONCE(psock->progs.stream_verdict))) {
266 sk_psock_put(sk, psock);
271 psock = sk_psock_init(sk, map->numa_node);
273 ret = PTR_ERR(psock);
279 psock_set_prog(&psock->progs.msg_parser, msg_parser);
281 ret = sock_map_init_proto(sk, psock);
285 write_lock_bh(&sk->sk_callback_lock);
286 if (stream_parser && stream_verdict && !psock->saved_data_ready) {
287 ret = sk_psock_init_strp(sk, psock);
289 goto out_unlock_drop;
290 psock_set_prog(&psock->progs.stream_verdict, stream_verdict);
291 psock_set_prog(&psock->progs.stream_parser, stream_parser);
292 sk_psock_start_strp(sk, psock);
293 } else if (!stream_parser && stream_verdict && !psock->saved_data_ready) {
294 psock_set_prog(&psock->progs.stream_verdict, stream_verdict);
295 sk_psock_start_verdict(sk,psock);
297 write_unlock_bh(&sk->sk_callback_lock);
300 write_unlock_bh(&sk->sk_callback_lock);
302 sk_psock_put(sk, psock);
305 bpf_prog_put(msg_parser);
306 out_put_stream_parser:
308 bpf_prog_put(stream_parser);
309 out_put_stream_verdict:
311 bpf_prog_put(stream_verdict);
315 static int sock_map_link_no_progs(struct bpf_map *map, struct sock *sk)
317 struct sk_psock *psock;
320 psock = sock_map_psock_get_checked(sk);
322 return PTR_ERR(psock);
325 psock = sk_psock_init(sk, map->numa_node);
327 return PTR_ERR(psock);
330 ret = sock_map_init_proto(sk, psock);
332 sk_psock_put(sk, psock);
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 bool sock_map_redirect_allowed(const struct sock *sk);
468 static int sock_map_update_common(struct bpf_map *map, u32 idx,
469 struct sock *sk, u64 flags)
471 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
472 struct sk_psock_link *link;
473 struct sk_psock *psock;
477 WARN_ON_ONCE(!rcu_read_lock_held());
478 if (unlikely(flags > BPF_EXIST))
480 if (unlikely(idx >= map->max_entries))
483 link = sk_psock_init_link();
487 /* Only sockets we can redirect into/from in BPF need to hold
488 * refs to parser/verdict progs and have their sk_data_ready
489 * and sk_write_space callbacks overridden.
491 if (sock_map_redirect_allowed(sk))
492 ret = sock_map_link(map, &stab->progs, sk);
494 ret = sock_map_link_no_progs(map, sk);
498 psock = sk_psock(sk);
499 WARN_ON_ONCE(!psock);
501 raw_spin_lock_bh(&stab->lock);
502 osk = stab->sks[idx];
503 if (osk && flags == BPF_NOEXIST) {
506 } else if (!osk && flags == BPF_EXIST) {
511 sock_map_add_link(psock, link, map, &stab->sks[idx]);
514 sock_map_unref(osk, &stab->sks[idx]);
515 raw_spin_unlock_bh(&stab->lock);
518 raw_spin_unlock_bh(&stab->lock);
520 sk_psock_put(sk, psock);
522 sk_psock_free_link(link);
526 static bool sock_map_op_okay(const struct bpf_sock_ops_kern *ops)
528 return ops->op == BPF_SOCK_OPS_PASSIVE_ESTABLISHED_CB ||
529 ops->op == BPF_SOCK_OPS_ACTIVE_ESTABLISHED_CB ||
530 ops->op == BPF_SOCK_OPS_TCP_LISTEN_CB;
533 static bool sk_is_tcp(const struct sock *sk)
535 return sk->sk_type == SOCK_STREAM &&
536 sk->sk_protocol == IPPROTO_TCP;
539 static bool sk_is_udp(const struct sock *sk)
541 return sk->sk_type == SOCK_DGRAM &&
542 sk->sk_protocol == IPPROTO_UDP;
545 static bool sock_map_redirect_allowed(const struct sock *sk)
547 return sk_is_tcp(sk) && sk->sk_state != TCP_LISTEN;
550 static bool sock_map_sk_is_suitable(const struct sock *sk)
552 return sk_is_tcp(sk) || sk_is_udp(sk);
555 static bool sock_map_sk_state_allowed(const struct sock *sk)
558 return (1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_LISTEN);
559 else if (sk_is_udp(sk))
560 return sk_hashed(sk);
565 static int sock_hash_update_common(struct bpf_map *map, void *key,
566 struct sock *sk, u64 flags);
568 int sock_map_update_elem_sys(struct bpf_map *map, void *key, void *value,
576 if (map->value_size == sizeof(u64))
583 sock = sockfd_lookup(ufd, &ret);
591 if (!sock_map_sk_is_suitable(sk)) {
596 sock_map_sk_acquire(sk);
597 if (!sock_map_sk_state_allowed(sk))
599 else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
600 ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
602 ret = sock_hash_update_common(map, key, sk, flags);
603 sock_map_sk_release(sk);
609 static int sock_map_update_elem(struct bpf_map *map, void *key,
610 void *value, u64 flags)
612 struct sock *sk = (struct sock *)value;
615 if (unlikely(!sk || !sk_fullsock(sk)))
618 if (!sock_map_sk_is_suitable(sk))
623 if (!sock_map_sk_state_allowed(sk))
625 else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
626 ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
628 ret = sock_hash_update_common(map, key, sk, flags);
634 BPF_CALL_4(bpf_sock_map_update, struct bpf_sock_ops_kern *, sops,
635 struct bpf_map *, map, void *, key, u64, flags)
637 WARN_ON_ONCE(!rcu_read_lock_held());
639 if (likely(sock_map_sk_is_suitable(sops->sk) &&
640 sock_map_op_okay(sops)))
641 return sock_map_update_common(map, *(u32 *)key, sops->sk,
646 const struct bpf_func_proto bpf_sock_map_update_proto = {
647 .func = bpf_sock_map_update,
650 .ret_type = RET_INTEGER,
651 .arg1_type = ARG_PTR_TO_CTX,
652 .arg2_type = ARG_CONST_MAP_PTR,
653 .arg3_type = ARG_PTR_TO_MAP_KEY,
654 .arg4_type = ARG_ANYTHING,
657 BPF_CALL_4(bpf_sk_redirect_map, struct sk_buff *, skb,
658 struct bpf_map *, map, u32, key, u64, flags)
662 if (unlikely(flags & ~(BPF_F_INGRESS)))
665 sk = __sock_map_lookup_elem(map, key);
666 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
669 skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
673 const struct bpf_func_proto bpf_sk_redirect_map_proto = {
674 .func = bpf_sk_redirect_map,
676 .ret_type = RET_INTEGER,
677 .arg1_type = ARG_PTR_TO_CTX,
678 .arg2_type = ARG_CONST_MAP_PTR,
679 .arg3_type = ARG_ANYTHING,
680 .arg4_type = ARG_ANYTHING,
683 BPF_CALL_4(bpf_msg_redirect_map, struct sk_msg *, msg,
684 struct bpf_map *, map, u32, key, u64, flags)
688 if (unlikely(flags & ~(BPF_F_INGRESS)))
691 sk = __sock_map_lookup_elem(map, key);
692 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
700 const struct bpf_func_proto bpf_msg_redirect_map_proto = {
701 .func = bpf_msg_redirect_map,
703 .ret_type = RET_INTEGER,
704 .arg1_type = ARG_PTR_TO_CTX,
705 .arg2_type = ARG_CONST_MAP_PTR,
706 .arg3_type = ARG_ANYTHING,
707 .arg4_type = ARG_ANYTHING,
710 struct sock_map_seq_info {
716 struct bpf_iter__sockmap {
717 __bpf_md_ptr(struct bpf_iter_meta *, meta);
718 __bpf_md_ptr(struct bpf_map *, map);
719 __bpf_md_ptr(void *, key);
720 __bpf_md_ptr(struct sock *, sk);
723 DEFINE_BPF_ITER_FUNC(sockmap, struct bpf_iter_meta *meta,
724 struct bpf_map *map, void *key,
727 static void *sock_map_seq_lookup_elem(struct sock_map_seq_info *info)
729 if (unlikely(info->index >= info->map->max_entries))
732 info->sk = __sock_map_lookup_elem(info->map, info->index);
734 /* can't return sk directly, since that might be NULL */
738 static void *sock_map_seq_start(struct seq_file *seq, loff_t *pos)
741 struct sock_map_seq_info *info = seq->private;
746 /* pairs with sock_map_seq_stop */
748 return sock_map_seq_lookup_elem(info);
751 static void *sock_map_seq_next(struct seq_file *seq, void *v, loff_t *pos)
754 struct sock_map_seq_info *info = seq->private;
759 return sock_map_seq_lookup_elem(info);
762 static int sock_map_seq_show(struct seq_file *seq, void *v)
765 struct sock_map_seq_info *info = seq->private;
766 struct bpf_iter__sockmap ctx = {};
767 struct bpf_iter_meta meta;
768 struct bpf_prog *prog;
771 prog = bpf_iter_get_info(&meta, !v);
778 ctx.key = &info->index;
782 return bpf_iter_run_prog(prog, &ctx);
785 static void sock_map_seq_stop(struct seq_file *seq, void *v)
789 (void)sock_map_seq_show(seq, NULL);
791 /* pairs with sock_map_seq_start */
795 static const struct seq_operations sock_map_seq_ops = {
796 .start = sock_map_seq_start,
797 .next = sock_map_seq_next,
798 .stop = sock_map_seq_stop,
799 .show = sock_map_seq_show,
802 static int sock_map_init_seq_private(void *priv_data,
803 struct bpf_iter_aux_info *aux)
805 struct sock_map_seq_info *info = priv_data;
807 info->map = aux->map;
811 static const struct bpf_iter_seq_info sock_map_iter_seq_info = {
812 .seq_ops = &sock_map_seq_ops,
813 .init_seq_private = sock_map_init_seq_private,
814 .seq_priv_size = sizeof(struct sock_map_seq_info),
817 static int sock_map_btf_id;
818 const struct bpf_map_ops sock_map_ops = {
819 .map_meta_equal = bpf_map_meta_equal,
820 .map_alloc = sock_map_alloc,
821 .map_free = sock_map_free,
822 .map_get_next_key = sock_map_get_next_key,
823 .map_lookup_elem_sys_only = sock_map_lookup_sys,
824 .map_update_elem = sock_map_update_elem,
825 .map_delete_elem = sock_map_delete_elem,
826 .map_lookup_elem = sock_map_lookup,
827 .map_release_uref = sock_map_release_progs,
828 .map_check_btf = map_check_no_btf,
829 .map_btf_name = "bpf_stab",
830 .map_btf_id = &sock_map_btf_id,
831 .iter_seq_info = &sock_map_iter_seq_info,
834 struct bpf_shtab_elem {
838 struct hlist_node node;
842 struct bpf_shtab_bucket {
843 struct hlist_head head;
849 struct bpf_shtab_bucket *buckets;
852 struct sk_psock_progs progs;
856 static inline u32 sock_hash_bucket_hash(const void *key, u32 len)
858 return jhash(key, len, 0);
861 static struct bpf_shtab_bucket *sock_hash_select_bucket(struct bpf_shtab *htab,
864 return &htab->buckets[hash & (htab->buckets_num - 1)];
867 static struct bpf_shtab_elem *
868 sock_hash_lookup_elem_raw(struct hlist_head *head, u32 hash, void *key,
871 struct bpf_shtab_elem *elem;
873 hlist_for_each_entry_rcu(elem, head, node) {
874 if (elem->hash == hash &&
875 !memcmp(&elem->key, key, key_size))
882 static struct sock *__sock_hash_lookup_elem(struct bpf_map *map, void *key)
884 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
885 u32 key_size = map->key_size, hash;
886 struct bpf_shtab_bucket *bucket;
887 struct bpf_shtab_elem *elem;
889 WARN_ON_ONCE(!rcu_read_lock_held());
891 hash = sock_hash_bucket_hash(key, key_size);
892 bucket = sock_hash_select_bucket(htab, hash);
893 elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
895 return elem ? elem->sk : NULL;
898 static void sock_hash_free_elem(struct bpf_shtab *htab,
899 struct bpf_shtab_elem *elem)
901 atomic_dec(&htab->count);
902 kfree_rcu(elem, rcu);
905 static void sock_hash_delete_from_link(struct bpf_map *map, struct sock *sk,
908 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
909 struct bpf_shtab_elem *elem_probe, *elem = link_raw;
910 struct bpf_shtab_bucket *bucket;
912 WARN_ON_ONCE(!rcu_read_lock_held());
913 bucket = sock_hash_select_bucket(htab, elem->hash);
915 /* elem may be deleted in parallel from the map, but access here
916 * is okay since it's going away only after RCU grace period.
917 * However, we need to check whether it's still present.
919 raw_spin_lock_bh(&bucket->lock);
920 elem_probe = sock_hash_lookup_elem_raw(&bucket->head, elem->hash,
921 elem->key, map->key_size);
922 if (elem_probe && elem_probe == elem) {
923 hlist_del_rcu(&elem->node);
924 sock_map_unref(elem->sk, elem);
925 sock_hash_free_elem(htab, elem);
927 raw_spin_unlock_bh(&bucket->lock);
930 static int sock_hash_delete_elem(struct bpf_map *map, void *key)
932 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
933 u32 hash, key_size = map->key_size;
934 struct bpf_shtab_bucket *bucket;
935 struct bpf_shtab_elem *elem;
938 hash = sock_hash_bucket_hash(key, key_size);
939 bucket = sock_hash_select_bucket(htab, hash);
941 raw_spin_lock_bh(&bucket->lock);
942 elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
944 hlist_del_rcu(&elem->node);
945 sock_map_unref(elem->sk, elem);
946 sock_hash_free_elem(htab, elem);
949 raw_spin_unlock_bh(&bucket->lock);
953 static struct bpf_shtab_elem *sock_hash_alloc_elem(struct bpf_shtab *htab,
954 void *key, u32 key_size,
955 u32 hash, struct sock *sk,
956 struct bpf_shtab_elem *old)
958 struct bpf_shtab_elem *new;
960 if (atomic_inc_return(&htab->count) > htab->map.max_entries) {
962 atomic_dec(&htab->count);
963 return ERR_PTR(-E2BIG);
967 new = bpf_map_kmalloc_node(&htab->map, htab->elem_size,
968 GFP_ATOMIC | __GFP_NOWARN,
969 htab->map.numa_node);
971 atomic_dec(&htab->count);
972 return ERR_PTR(-ENOMEM);
974 memcpy(new->key, key, key_size);
980 static int sock_hash_update_common(struct bpf_map *map, void *key,
981 struct sock *sk, u64 flags)
983 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
984 u32 key_size = map->key_size, hash;
985 struct bpf_shtab_elem *elem, *elem_new;
986 struct bpf_shtab_bucket *bucket;
987 struct sk_psock_link *link;
988 struct sk_psock *psock;
991 WARN_ON_ONCE(!rcu_read_lock_held());
992 if (unlikely(flags > BPF_EXIST))
995 link = sk_psock_init_link();
999 /* Only sockets we can redirect into/from in BPF need to hold
1000 * refs to parser/verdict progs and have their sk_data_ready
1001 * and sk_write_space callbacks overridden.
1003 if (sock_map_redirect_allowed(sk))
1004 ret = sock_map_link(map, &htab->progs, sk);
1006 ret = sock_map_link_no_progs(map, sk);
1010 psock = sk_psock(sk);
1011 WARN_ON_ONCE(!psock);
1013 hash = sock_hash_bucket_hash(key, key_size);
1014 bucket = sock_hash_select_bucket(htab, hash);
1016 raw_spin_lock_bh(&bucket->lock);
1017 elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
1018 if (elem && flags == BPF_NOEXIST) {
1021 } else if (!elem && flags == BPF_EXIST) {
1026 elem_new = sock_hash_alloc_elem(htab, key, key_size, hash, sk, elem);
1027 if (IS_ERR(elem_new)) {
1028 ret = PTR_ERR(elem_new);
1032 sock_map_add_link(psock, link, map, elem_new);
1033 /* Add new element to the head of the list, so that
1034 * concurrent search will find it before old elem.
1036 hlist_add_head_rcu(&elem_new->node, &bucket->head);
1038 hlist_del_rcu(&elem->node);
1039 sock_map_unref(elem->sk, elem);
1040 sock_hash_free_elem(htab, elem);
1042 raw_spin_unlock_bh(&bucket->lock);
1045 raw_spin_unlock_bh(&bucket->lock);
1046 sk_psock_put(sk, psock);
1048 sk_psock_free_link(link);
1052 static int sock_hash_get_next_key(struct bpf_map *map, void *key,
1055 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1056 struct bpf_shtab_elem *elem, *elem_next;
1057 u32 hash, key_size = map->key_size;
1058 struct hlist_head *head;
1062 goto find_first_elem;
1063 hash = sock_hash_bucket_hash(key, key_size);
1064 head = &sock_hash_select_bucket(htab, hash)->head;
1065 elem = sock_hash_lookup_elem_raw(head, hash, key, key_size);
1067 goto find_first_elem;
1069 elem_next = hlist_entry_safe(rcu_dereference(hlist_next_rcu(&elem->node)),
1070 struct bpf_shtab_elem, node);
1072 memcpy(key_next, elem_next->key, key_size);
1076 i = hash & (htab->buckets_num - 1);
1079 for (; i < htab->buckets_num; i++) {
1080 head = &sock_hash_select_bucket(htab, i)->head;
1081 elem_next = hlist_entry_safe(rcu_dereference(hlist_first_rcu(head)),
1082 struct bpf_shtab_elem, node);
1084 memcpy(key_next, elem_next->key, key_size);
1092 static struct bpf_map *sock_hash_alloc(union bpf_attr *attr)
1094 struct bpf_shtab *htab;
1097 if (!capable(CAP_NET_ADMIN))
1098 return ERR_PTR(-EPERM);
1099 if (attr->max_entries == 0 ||
1100 attr->key_size == 0 ||
1101 (attr->value_size != sizeof(u32) &&
1102 attr->value_size != sizeof(u64)) ||
1103 attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
1104 return ERR_PTR(-EINVAL);
1105 if (attr->key_size > MAX_BPF_STACK)
1106 return ERR_PTR(-E2BIG);
1108 htab = kzalloc(sizeof(*htab), GFP_USER | __GFP_ACCOUNT);
1110 return ERR_PTR(-ENOMEM);
1112 bpf_map_init_from_attr(&htab->map, attr);
1114 htab->buckets_num = roundup_pow_of_two(htab->map.max_entries);
1115 htab->elem_size = sizeof(struct bpf_shtab_elem) +
1116 round_up(htab->map.key_size, 8);
1117 if (htab->buckets_num == 0 ||
1118 htab->buckets_num > U32_MAX / sizeof(struct bpf_shtab_bucket)) {
1123 htab->buckets = bpf_map_area_alloc(htab->buckets_num *
1124 sizeof(struct bpf_shtab_bucket),
1125 htab->map.numa_node);
1126 if (!htab->buckets) {
1131 for (i = 0; i < htab->buckets_num; i++) {
1132 INIT_HLIST_HEAD(&htab->buckets[i].head);
1133 raw_spin_lock_init(&htab->buckets[i].lock);
1139 return ERR_PTR(err);
1142 static void sock_hash_free(struct bpf_map *map)
1144 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1145 struct bpf_shtab_bucket *bucket;
1146 struct hlist_head unlink_list;
1147 struct bpf_shtab_elem *elem;
1148 struct hlist_node *node;
1151 /* After the sync no updates or deletes will be in-flight so it
1152 * is safe to walk map and remove entries without risking a race
1153 * in EEXIST update case.
1156 for (i = 0; i < htab->buckets_num; i++) {
1157 bucket = sock_hash_select_bucket(htab, i);
1159 /* We are racing with sock_hash_delete_from_link to
1160 * enter the spin-lock critical section. Every socket on
1161 * the list is still linked to sockhash. Since link
1162 * exists, psock exists and holds a ref to socket. That
1163 * lets us to grab a socket ref too.
1165 raw_spin_lock_bh(&bucket->lock);
1166 hlist_for_each_entry(elem, &bucket->head, node)
1167 sock_hold(elem->sk);
1168 hlist_move_list(&bucket->head, &unlink_list);
1169 raw_spin_unlock_bh(&bucket->lock);
1171 /* Process removed entries out of atomic context to
1172 * block for socket lock before deleting the psock's
1175 hlist_for_each_entry_safe(elem, node, &unlink_list, node) {
1176 hlist_del(&elem->node);
1177 lock_sock(elem->sk);
1179 sock_map_unref(elem->sk, elem);
1181 release_sock(elem->sk);
1183 sock_hash_free_elem(htab, elem);
1187 /* wait for psock readers accessing its map link */
1190 bpf_map_area_free(htab->buckets);
1194 static void *sock_hash_lookup_sys(struct bpf_map *map, void *key)
1198 if (map->value_size != sizeof(u64))
1199 return ERR_PTR(-ENOSPC);
1201 sk = __sock_hash_lookup_elem(map, key);
1203 return ERR_PTR(-ENOENT);
1205 __sock_gen_cookie(sk);
1206 return &sk->sk_cookie;
1209 static void *sock_hash_lookup(struct bpf_map *map, void *key)
1213 sk = __sock_hash_lookup_elem(map, key);
1216 if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
1221 static void sock_hash_release_progs(struct bpf_map *map)
1223 psock_progs_drop(&container_of(map, struct bpf_shtab, map)->progs);
1226 BPF_CALL_4(bpf_sock_hash_update, struct bpf_sock_ops_kern *, sops,
1227 struct bpf_map *, map, void *, key, u64, flags)
1229 WARN_ON_ONCE(!rcu_read_lock_held());
1231 if (likely(sock_map_sk_is_suitable(sops->sk) &&
1232 sock_map_op_okay(sops)))
1233 return sock_hash_update_common(map, key, sops->sk, flags);
1237 const struct bpf_func_proto bpf_sock_hash_update_proto = {
1238 .func = bpf_sock_hash_update,
1241 .ret_type = RET_INTEGER,
1242 .arg1_type = ARG_PTR_TO_CTX,
1243 .arg2_type = ARG_CONST_MAP_PTR,
1244 .arg3_type = ARG_PTR_TO_MAP_KEY,
1245 .arg4_type = ARG_ANYTHING,
1248 BPF_CALL_4(bpf_sk_redirect_hash, struct sk_buff *, skb,
1249 struct bpf_map *, map, void *, key, u64, flags)
1253 if (unlikely(flags & ~(BPF_F_INGRESS)))
1256 sk = __sock_hash_lookup_elem(map, key);
1257 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1260 skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
1264 const struct bpf_func_proto bpf_sk_redirect_hash_proto = {
1265 .func = bpf_sk_redirect_hash,
1267 .ret_type = RET_INTEGER,
1268 .arg1_type = ARG_PTR_TO_CTX,
1269 .arg2_type = ARG_CONST_MAP_PTR,
1270 .arg3_type = ARG_PTR_TO_MAP_KEY,
1271 .arg4_type = ARG_ANYTHING,
1274 BPF_CALL_4(bpf_msg_redirect_hash, struct sk_msg *, msg,
1275 struct bpf_map *, map, void *, key, u64, flags)
1279 if (unlikely(flags & ~(BPF_F_INGRESS)))
1282 sk = __sock_hash_lookup_elem(map, key);
1283 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1291 const struct bpf_func_proto bpf_msg_redirect_hash_proto = {
1292 .func = bpf_msg_redirect_hash,
1294 .ret_type = RET_INTEGER,
1295 .arg1_type = ARG_PTR_TO_CTX,
1296 .arg2_type = ARG_CONST_MAP_PTR,
1297 .arg3_type = ARG_PTR_TO_MAP_KEY,
1298 .arg4_type = ARG_ANYTHING,
1301 struct sock_hash_seq_info {
1302 struct bpf_map *map;
1303 struct bpf_shtab *htab;
1307 static void *sock_hash_seq_find_next(struct sock_hash_seq_info *info,
1308 struct bpf_shtab_elem *prev_elem)
1310 const struct bpf_shtab *htab = info->htab;
1311 struct bpf_shtab_bucket *bucket;
1312 struct bpf_shtab_elem *elem;
1313 struct hlist_node *node;
1315 /* try to find next elem in the same bucket */
1317 node = rcu_dereference(hlist_next_rcu(&prev_elem->node));
1318 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1322 /* no more elements, continue in the next bucket */
1326 for (; info->bucket_id < htab->buckets_num; info->bucket_id++) {
1327 bucket = &htab->buckets[info->bucket_id];
1328 node = rcu_dereference(hlist_first_rcu(&bucket->head));
1329 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1337 static void *sock_hash_seq_start(struct seq_file *seq, loff_t *pos)
1340 struct sock_hash_seq_info *info = seq->private;
1345 /* pairs with sock_hash_seq_stop */
1347 return sock_hash_seq_find_next(info, NULL);
1350 static void *sock_hash_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1353 struct sock_hash_seq_info *info = seq->private;
1356 return sock_hash_seq_find_next(info, v);
1359 static int sock_hash_seq_show(struct seq_file *seq, void *v)
1362 struct sock_hash_seq_info *info = seq->private;
1363 struct bpf_iter__sockmap ctx = {};
1364 struct bpf_shtab_elem *elem = v;
1365 struct bpf_iter_meta meta;
1366 struct bpf_prog *prog;
1369 prog = bpf_iter_get_info(&meta, !elem);
1374 ctx.map = info->map;
1376 ctx.key = elem->key;
1380 return bpf_iter_run_prog(prog, &ctx);
1383 static void sock_hash_seq_stop(struct seq_file *seq, void *v)
1387 (void)sock_hash_seq_show(seq, NULL);
1389 /* pairs with sock_hash_seq_start */
1393 static const struct seq_operations sock_hash_seq_ops = {
1394 .start = sock_hash_seq_start,
1395 .next = sock_hash_seq_next,
1396 .stop = sock_hash_seq_stop,
1397 .show = sock_hash_seq_show,
1400 static int sock_hash_init_seq_private(void *priv_data,
1401 struct bpf_iter_aux_info *aux)
1403 struct sock_hash_seq_info *info = priv_data;
1405 info->map = aux->map;
1406 info->htab = container_of(aux->map, struct bpf_shtab, map);
1410 static const struct bpf_iter_seq_info sock_hash_iter_seq_info = {
1411 .seq_ops = &sock_hash_seq_ops,
1412 .init_seq_private = sock_hash_init_seq_private,
1413 .seq_priv_size = sizeof(struct sock_hash_seq_info),
1416 static int sock_hash_map_btf_id;
1417 const struct bpf_map_ops sock_hash_ops = {
1418 .map_meta_equal = bpf_map_meta_equal,
1419 .map_alloc = sock_hash_alloc,
1420 .map_free = sock_hash_free,
1421 .map_get_next_key = sock_hash_get_next_key,
1422 .map_update_elem = sock_map_update_elem,
1423 .map_delete_elem = sock_hash_delete_elem,
1424 .map_lookup_elem = sock_hash_lookup,
1425 .map_lookup_elem_sys_only = sock_hash_lookup_sys,
1426 .map_release_uref = sock_hash_release_progs,
1427 .map_check_btf = map_check_no_btf,
1428 .map_btf_name = "bpf_shtab",
1429 .map_btf_id = &sock_hash_map_btf_id,
1430 .iter_seq_info = &sock_hash_iter_seq_info,
1433 static struct sk_psock_progs *sock_map_progs(struct bpf_map *map)
1435 switch (map->map_type) {
1436 case BPF_MAP_TYPE_SOCKMAP:
1437 return &container_of(map, struct bpf_stab, map)->progs;
1438 case BPF_MAP_TYPE_SOCKHASH:
1439 return &container_of(map, struct bpf_shtab, map)->progs;
1447 int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog,
1448 struct bpf_prog *old, u32 which)
1450 struct sk_psock_progs *progs = sock_map_progs(map);
1451 struct bpf_prog **pprog;
1457 case BPF_SK_MSG_VERDICT:
1458 pprog = &progs->msg_parser;
1460 #if IS_ENABLED(CONFIG_BPF_STREAM_PARSER)
1461 case BPF_SK_SKB_STREAM_PARSER:
1462 pprog = &progs->stream_parser;
1465 case BPF_SK_SKB_STREAM_VERDICT:
1466 pprog = &progs->stream_verdict;
1473 return psock_replace_prog(pprog, prog, old);
1475 psock_set_prog(pprog, prog);
1479 static void sock_map_unlink(struct sock *sk, struct sk_psock_link *link)
1481 switch (link->map->map_type) {
1482 case BPF_MAP_TYPE_SOCKMAP:
1483 return sock_map_delete_from_link(link->map, sk,
1485 case BPF_MAP_TYPE_SOCKHASH:
1486 return sock_hash_delete_from_link(link->map, sk,
1493 static void sock_map_remove_links(struct sock *sk, struct sk_psock *psock)
1495 struct sk_psock_link *link;
1497 while ((link = sk_psock_link_pop(psock))) {
1498 sock_map_unlink(sk, link);
1499 sk_psock_free_link(link);
1503 void sock_map_unhash(struct sock *sk)
1505 void (*saved_unhash)(struct sock *sk);
1506 struct sk_psock *psock;
1509 psock = sk_psock(sk);
1510 if (unlikely(!psock)) {
1512 if (sk->sk_prot->unhash)
1513 sk->sk_prot->unhash(sk);
1517 saved_unhash = psock->saved_unhash;
1518 sock_map_remove_links(sk, psock);
1523 void sock_map_close(struct sock *sk, long timeout)
1525 void (*saved_close)(struct sock *sk, long timeout);
1526 struct sk_psock *psock;
1530 psock = sk_psock(sk);
1531 if (unlikely(!psock)) {
1534 return sk->sk_prot->close(sk, timeout);
1537 saved_close = psock->saved_close;
1538 sock_map_remove_links(sk, psock);
1541 saved_close(sk, timeout);
1544 static int sock_map_iter_attach_target(struct bpf_prog *prog,
1545 union bpf_iter_link_info *linfo,
1546 struct bpf_iter_aux_info *aux)
1548 struct bpf_map *map;
1551 if (!linfo->map.map_fd)
1554 map = bpf_map_get_with_uref(linfo->map.map_fd);
1556 return PTR_ERR(map);
1558 if (map->map_type != BPF_MAP_TYPE_SOCKMAP &&
1559 map->map_type != BPF_MAP_TYPE_SOCKHASH)
1562 if (prog->aux->max_rdonly_access > map->key_size) {
1571 bpf_map_put_with_uref(map);
1575 static void sock_map_iter_detach_target(struct bpf_iter_aux_info *aux)
1577 bpf_map_put_with_uref(aux->map);
1580 static struct bpf_iter_reg sock_map_iter_reg = {
1581 .target = "sockmap",
1582 .attach_target = sock_map_iter_attach_target,
1583 .detach_target = sock_map_iter_detach_target,
1584 .show_fdinfo = bpf_iter_map_show_fdinfo,
1585 .fill_link_info = bpf_iter_map_fill_link_info,
1586 .ctx_arg_info_size = 2,
1588 { offsetof(struct bpf_iter__sockmap, key),
1589 PTR_TO_RDONLY_BUF_OR_NULL },
1590 { offsetof(struct bpf_iter__sockmap, sk),
1591 PTR_TO_BTF_ID_OR_NULL },
1595 static int __init bpf_sockmap_iter_init(void)
1597 sock_map_iter_reg.ctx_arg_info[1].btf_id =
1598 btf_sock_ids[BTF_SOCK_TYPE_SOCK];
1599 return bpf_iter_reg_target(&sock_map_iter_reg);
1601 late_initcall(bpf_sockmap_iter_init);