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)
123 static void sock_map_sk_release(struct sock *sk)
124 __releases(&sk->sk_lock.slock)
130 static void sock_map_add_link(struct sk_psock *psock,
131 struct sk_psock_link *link,
132 struct bpf_map *map, void *link_raw)
134 link->link_raw = link_raw;
136 spin_lock_bh(&psock->link_lock);
137 list_add_tail(&link->list, &psock->link);
138 spin_unlock_bh(&psock->link_lock);
141 static void sock_map_del_link(struct sock *sk,
142 struct sk_psock *psock, void *link_raw)
144 bool strp_stop = false, verdict_stop = false;
145 struct sk_psock_link *link, *tmp;
147 spin_lock_bh(&psock->link_lock);
148 list_for_each_entry_safe(link, tmp, &psock->link, list) {
149 if (link->link_raw == link_raw) {
150 struct bpf_map *map = link->map;
151 struct sk_psock_progs *progs = sock_map_progs(map);
153 if (psock->saved_data_ready && progs->stream_parser)
155 if (psock->saved_data_ready && progs->stream_verdict)
157 if (psock->saved_data_ready && progs->skb_verdict)
159 list_del(&link->list);
160 sk_psock_free_link(link);
163 spin_unlock_bh(&psock->link_lock);
164 if (strp_stop || verdict_stop) {
165 write_lock_bh(&sk->sk_callback_lock);
167 sk_psock_stop_strp(sk, psock);
169 sk_psock_stop_verdict(sk, psock);
171 if (psock->psock_update_sk_prot)
172 psock->psock_update_sk_prot(sk, psock, false);
173 write_unlock_bh(&sk->sk_callback_lock);
177 static void sock_map_unref(struct sock *sk, void *link_raw)
179 struct sk_psock *psock = sk_psock(sk);
182 sock_map_del_link(sk, psock, link_raw);
183 sk_psock_put(sk, psock);
187 static int sock_map_init_proto(struct sock *sk, struct sk_psock *psock)
189 if (!sk->sk_prot->psock_update_sk_prot)
191 psock->psock_update_sk_prot = sk->sk_prot->psock_update_sk_prot;
192 return sk->sk_prot->psock_update_sk_prot(sk, psock, false);
195 static struct sk_psock *sock_map_psock_get_checked(struct sock *sk)
197 struct sk_psock *psock;
200 psock = sk_psock(sk);
202 if (sk->sk_prot->close != sock_map_close) {
203 psock = ERR_PTR(-EBUSY);
207 if (!refcount_inc_not_zero(&psock->refcnt))
208 psock = ERR_PTR(-EBUSY);
215 static int sock_map_link(struct bpf_map *map, struct sock *sk)
217 struct sk_psock_progs *progs = sock_map_progs(map);
218 struct bpf_prog *stream_verdict = NULL;
219 struct bpf_prog *stream_parser = NULL;
220 struct bpf_prog *skb_verdict = NULL;
221 struct bpf_prog *msg_parser = NULL;
222 struct sk_psock *psock;
225 stream_verdict = READ_ONCE(progs->stream_verdict);
226 if (stream_verdict) {
227 stream_verdict = bpf_prog_inc_not_zero(stream_verdict);
228 if (IS_ERR(stream_verdict))
229 return PTR_ERR(stream_verdict);
232 stream_parser = READ_ONCE(progs->stream_parser);
234 stream_parser = bpf_prog_inc_not_zero(stream_parser);
235 if (IS_ERR(stream_parser)) {
236 ret = PTR_ERR(stream_parser);
237 goto out_put_stream_verdict;
241 msg_parser = READ_ONCE(progs->msg_parser);
243 msg_parser = bpf_prog_inc_not_zero(msg_parser);
244 if (IS_ERR(msg_parser)) {
245 ret = PTR_ERR(msg_parser);
246 goto out_put_stream_parser;
250 skb_verdict = READ_ONCE(progs->skb_verdict);
252 skb_verdict = bpf_prog_inc_not_zero(skb_verdict);
253 if (IS_ERR(skb_verdict)) {
254 ret = PTR_ERR(skb_verdict);
255 goto out_put_msg_parser;
259 psock = sock_map_psock_get_checked(sk);
261 ret = PTR_ERR(psock);
266 if ((msg_parser && READ_ONCE(psock->progs.msg_parser)) ||
267 (stream_parser && READ_ONCE(psock->progs.stream_parser)) ||
268 (skb_verdict && READ_ONCE(psock->progs.skb_verdict)) ||
269 (skb_verdict && READ_ONCE(psock->progs.stream_verdict)) ||
270 (stream_verdict && READ_ONCE(psock->progs.skb_verdict)) ||
271 (stream_verdict && READ_ONCE(psock->progs.stream_verdict))) {
272 sk_psock_put(sk, psock);
277 psock = sk_psock_init(sk, map->numa_node);
279 ret = PTR_ERR(psock);
285 psock_set_prog(&psock->progs.msg_parser, msg_parser);
287 psock_set_prog(&psock->progs.stream_parser, stream_parser);
289 psock_set_prog(&psock->progs.stream_verdict, stream_verdict);
291 psock_set_prog(&psock->progs.skb_verdict, skb_verdict);
293 /* msg_* and stream_* programs references tracked in psock after this
294 * point. Reference dec and cleanup will occur through psock destructor
296 ret = sock_map_init_proto(sk, psock);
298 sk_psock_put(sk, psock);
302 write_lock_bh(&sk->sk_callback_lock);
303 if (stream_parser && stream_verdict && !psock->saved_data_ready) {
304 ret = sk_psock_init_strp(sk, psock);
306 write_unlock_bh(&sk->sk_callback_lock);
307 sk_psock_put(sk, psock);
310 sk_psock_start_strp(sk, psock);
311 } else if (!stream_parser && stream_verdict && !psock->saved_data_ready) {
312 sk_psock_start_verdict(sk,psock);
313 } else if (!stream_verdict && skb_verdict && !psock->saved_data_ready) {
314 sk_psock_start_verdict(sk, psock);
316 write_unlock_bh(&sk->sk_callback_lock);
320 bpf_prog_put(skb_verdict);
323 bpf_prog_put(msg_parser);
324 out_put_stream_parser:
326 bpf_prog_put(stream_parser);
327 out_put_stream_verdict:
329 bpf_prog_put(stream_verdict);
334 static void sock_map_free(struct bpf_map *map)
336 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
339 /* After the sync no updates or deletes will be in-flight so it
340 * is safe to walk map and remove entries without risking a race
341 * in EEXIST update case.
344 for (i = 0; i < stab->map.max_entries; i++) {
345 struct sock **psk = &stab->sks[i];
348 sk = xchg(psk, NULL);
353 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 sk_is_tcp(const struct sock *sk)
526 return sk->sk_type == SOCK_STREAM &&
527 sk->sk_protocol == IPPROTO_TCP;
530 static bool sock_map_redirect_allowed(const struct sock *sk)
533 return sk->sk_state != TCP_LISTEN;
535 return sk->sk_state == TCP_ESTABLISHED;
538 static bool sock_map_sk_is_suitable(const struct sock *sk)
540 return !!sk->sk_prot->psock_update_sk_prot;
543 static bool sock_map_sk_state_allowed(const struct sock *sk)
546 return (1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_LISTEN);
550 static int sock_hash_update_common(struct bpf_map *map, void *key,
551 struct sock *sk, u64 flags);
553 int sock_map_update_elem_sys(struct bpf_map *map, void *key, void *value,
561 if (map->value_size == sizeof(u64))
568 sock = sockfd_lookup(ufd, &ret);
576 if (!sock_map_sk_is_suitable(sk)) {
581 sock_map_sk_acquire(sk);
582 if (!sock_map_sk_state_allowed(sk))
584 else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
585 ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
587 ret = sock_hash_update_common(map, key, sk, flags);
588 sock_map_sk_release(sk);
594 static int sock_map_update_elem(struct bpf_map *map, void *key,
595 void *value, u64 flags)
597 struct sock *sk = (struct sock *)value;
600 if (unlikely(!sk || !sk_fullsock(sk)))
603 if (!sock_map_sk_is_suitable(sk))
608 if (!sock_map_sk_state_allowed(sk))
610 else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
611 ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
613 ret = sock_hash_update_common(map, key, sk, flags);
619 BPF_CALL_4(bpf_sock_map_update, struct bpf_sock_ops_kern *, sops,
620 struct bpf_map *, map, void *, key, u64, flags)
622 WARN_ON_ONCE(!rcu_read_lock_held());
624 if (likely(sock_map_sk_is_suitable(sops->sk) &&
625 sock_map_op_okay(sops)))
626 return sock_map_update_common(map, *(u32 *)key, sops->sk,
631 const struct bpf_func_proto bpf_sock_map_update_proto = {
632 .func = bpf_sock_map_update,
635 .ret_type = RET_INTEGER,
636 .arg1_type = ARG_PTR_TO_CTX,
637 .arg2_type = ARG_CONST_MAP_PTR,
638 .arg3_type = ARG_PTR_TO_MAP_KEY,
639 .arg4_type = ARG_ANYTHING,
642 BPF_CALL_4(bpf_sk_redirect_map, struct sk_buff *, skb,
643 struct bpf_map *, map, u32, key, u64, flags)
647 if (unlikely(flags & ~(BPF_F_INGRESS)))
650 sk = __sock_map_lookup_elem(map, key);
651 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
654 skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
658 const struct bpf_func_proto bpf_sk_redirect_map_proto = {
659 .func = bpf_sk_redirect_map,
661 .ret_type = RET_INTEGER,
662 .arg1_type = ARG_PTR_TO_CTX,
663 .arg2_type = ARG_CONST_MAP_PTR,
664 .arg3_type = ARG_ANYTHING,
665 .arg4_type = ARG_ANYTHING,
668 BPF_CALL_4(bpf_msg_redirect_map, struct sk_msg *, msg,
669 struct bpf_map *, map, u32, key, u64, flags)
673 if (unlikely(flags & ~(BPF_F_INGRESS)))
676 sk = __sock_map_lookup_elem(map, key);
677 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
679 if (!(flags & BPF_F_INGRESS) && !sk_is_tcp(sk))
687 const struct bpf_func_proto bpf_msg_redirect_map_proto = {
688 .func = bpf_msg_redirect_map,
690 .ret_type = RET_INTEGER,
691 .arg1_type = ARG_PTR_TO_CTX,
692 .arg2_type = ARG_CONST_MAP_PTR,
693 .arg3_type = ARG_ANYTHING,
694 .arg4_type = ARG_ANYTHING,
697 struct sock_map_seq_info {
703 struct bpf_iter__sockmap {
704 __bpf_md_ptr(struct bpf_iter_meta *, meta);
705 __bpf_md_ptr(struct bpf_map *, map);
706 __bpf_md_ptr(void *, key);
707 __bpf_md_ptr(struct sock *, sk);
710 DEFINE_BPF_ITER_FUNC(sockmap, struct bpf_iter_meta *meta,
711 struct bpf_map *map, void *key,
714 static void *sock_map_seq_lookup_elem(struct sock_map_seq_info *info)
716 if (unlikely(info->index >= info->map->max_entries))
719 info->sk = __sock_map_lookup_elem(info->map, info->index);
721 /* can't return sk directly, since that might be NULL */
725 static void *sock_map_seq_start(struct seq_file *seq, loff_t *pos)
728 struct sock_map_seq_info *info = seq->private;
733 /* pairs with sock_map_seq_stop */
735 return sock_map_seq_lookup_elem(info);
738 static void *sock_map_seq_next(struct seq_file *seq, void *v, loff_t *pos)
741 struct sock_map_seq_info *info = seq->private;
746 return sock_map_seq_lookup_elem(info);
749 static int sock_map_seq_show(struct seq_file *seq, void *v)
752 struct sock_map_seq_info *info = seq->private;
753 struct bpf_iter__sockmap ctx = {};
754 struct bpf_iter_meta meta;
755 struct bpf_prog *prog;
758 prog = bpf_iter_get_info(&meta, !v);
765 ctx.key = &info->index;
769 return bpf_iter_run_prog(prog, &ctx);
772 static void sock_map_seq_stop(struct seq_file *seq, void *v)
776 (void)sock_map_seq_show(seq, NULL);
778 /* pairs with sock_map_seq_start */
782 static const struct seq_operations sock_map_seq_ops = {
783 .start = sock_map_seq_start,
784 .next = sock_map_seq_next,
785 .stop = sock_map_seq_stop,
786 .show = sock_map_seq_show,
789 static int sock_map_init_seq_private(void *priv_data,
790 struct bpf_iter_aux_info *aux)
792 struct sock_map_seq_info *info = priv_data;
794 bpf_map_inc_with_uref(aux->map);
795 info->map = aux->map;
799 static void sock_map_fini_seq_private(void *priv_data)
801 struct sock_map_seq_info *info = priv_data;
803 bpf_map_put_with_uref(info->map);
806 static const struct bpf_iter_seq_info sock_map_iter_seq_info = {
807 .seq_ops = &sock_map_seq_ops,
808 .init_seq_private = sock_map_init_seq_private,
809 .fini_seq_private = sock_map_fini_seq_private,
810 .seq_priv_size = sizeof(struct sock_map_seq_info),
813 static int sock_map_btf_id;
814 const struct bpf_map_ops sock_map_ops = {
815 .map_meta_equal = bpf_map_meta_equal,
816 .map_alloc = sock_map_alloc,
817 .map_free = sock_map_free,
818 .map_get_next_key = sock_map_get_next_key,
819 .map_lookup_elem_sys_only = sock_map_lookup_sys,
820 .map_update_elem = sock_map_update_elem,
821 .map_delete_elem = sock_map_delete_elem,
822 .map_lookup_elem = sock_map_lookup,
823 .map_release_uref = sock_map_release_progs,
824 .map_check_btf = map_check_no_btf,
825 .map_btf_name = "bpf_stab",
826 .map_btf_id = &sock_map_btf_id,
827 .iter_seq_info = &sock_map_iter_seq_info,
830 struct bpf_shtab_elem {
834 struct hlist_node node;
838 struct bpf_shtab_bucket {
839 struct hlist_head head;
845 struct bpf_shtab_bucket *buckets;
848 struct sk_psock_progs progs;
852 static inline u32 sock_hash_bucket_hash(const void *key, u32 len)
854 return jhash(key, len, 0);
857 static struct bpf_shtab_bucket *sock_hash_select_bucket(struct bpf_shtab *htab,
860 return &htab->buckets[hash & (htab->buckets_num - 1)];
863 static struct bpf_shtab_elem *
864 sock_hash_lookup_elem_raw(struct hlist_head *head, u32 hash, void *key,
867 struct bpf_shtab_elem *elem;
869 hlist_for_each_entry_rcu(elem, head, node) {
870 if (elem->hash == hash &&
871 !memcmp(&elem->key, key, key_size))
878 static struct sock *__sock_hash_lookup_elem(struct bpf_map *map, void *key)
880 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
881 u32 key_size = map->key_size, hash;
882 struct bpf_shtab_bucket *bucket;
883 struct bpf_shtab_elem *elem;
885 WARN_ON_ONCE(!rcu_read_lock_held());
887 hash = sock_hash_bucket_hash(key, key_size);
888 bucket = sock_hash_select_bucket(htab, hash);
889 elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
891 return elem ? elem->sk : NULL;
894 static void sock_hash_free_elem(struct bpf_shtab *htab,
895 struct bpf_shtab_elem *elem)
897 atomic_dec(&htab->count);
898 kfree_rcu(elem, rcu);
901 static void sock_hash_delete_from_link(struct bpf_map *map, struct sock *sk,
904 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
905 struct bpf_shtab_elem *elem_probe, *elem = link_raw;
906 struct bpf_shtab_bucket *bucket;
908 WARN_ON_ONCE(!rcu_read_lock_held());
909 bucket = sock_hash_select_bucket(htab, elem->hash);
911 /* elem may be deleted in parallel from the map, but access here
912 * is okay since it's going away only after RCU grace period.
913 * However, we need to check whether it's still present.
915 raw_spin_lock_bh(&bucket->lock);
916 elem_probe = sock_hash_lookup_elem_raw(&bucket->head, elem->hash,
917 elem->key, map->key_size);
918 if (elem_probe && elem_probe == elem) {
919 hlist_del_rcu(&elem->node);
920 sock_map_unref(elem->sk, elem);
921 sock_hash_free_elem(htab, elem);
923 raw_spin_unlock_bh(&bucket->lock);
926 static int sock_hash_delete_elem(struct bpf_map *map, void *key)
928 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
929 u32 hash, key_size = map->key_size;
930 struct bpf_shtab_bucket *bucket;
931 struct bpf_shtab_elem *elem;
934 hash = sock_hash_bucket_hash(key, key_size);
935 bucket = sock_hash_select_bucket(htab, hash);
937 raw_spin_lock_bh(&bucket->lock);
938 elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
940 hlist_del_rcu(&elem->node);
941 sock_map_unref(elem->sk, elem);
942 sock_hash_free_elem(htab, elem);
945 raw_spin_unlock_bh(&bucket->lock);
949 static struct bpf_shtab_elem *sock_hash_alloc_elem(struct bpf_shtab *htab,
950 void *key, u32 key_size,
951 u32 hash, struct sock *sk,
952 struct bpf_shtab_elem *old)
954 struct bpf_shtab_elem *new;
956 if (atomic_inc_return(&htab->count) > htab->map.max_entries) {
958 atomic_dec(&htab->count);
959 return ERR_PTR(-E2BIG);
963 new = bpf_map_kmalloc_node(&htab->map, htab->elem_size,
964 GFP_ATOMIC | __GFP_NOWARN,
965 htab->map.numa_node);
967 atomic_dec(&htab->count);
968 return ERR_PTR(-ENOMEM);
970 memcpy(new->key, key, key_size);
976 static int sock_hash_update_common(struct bpf_map *map, void *key,
977 struct sock *sk, u64 flags)
979 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
980 u32 key_size = map->key_size, hash;
981 struct bpf_shtab_elem *elem, *elem_new;
982 struct bpf_shtab_bucket *bucket;
983 struct sk_psock_link *link;
984 struct sk_psock *psock;
987 WARN_ON_ONCE(!rcu_read_lock_held());
988 if (unlikely(flags > BPF_EXIST))
991 link = sk_psock_init_link();
995 ret = sock_map_link(map, sk);
999 psock = sk_psock(sk);
1000 WARN_ON_ONCE(!psock);
1002 hash = sock_hash_bucket_hash(key, key_size);
1003 bucket = sock_hash_select_bucket(htab, hash);
1005 raw_spin_lock_bh(&bucket->lock);
1006 elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
1007 if (elem && flags == BPF_NOEXIST) {
1010 } else if (!elem && flags == BPF_EXIST) {
1015 elem_new = sock_hash_alloc_elem(htab, key, key_size, hash, sk, elem);
1016 if (IS_ERR(elem_new)) {
1017 ret = PTR_ERR(elem_new);
1021 sock_map_add_link(psock, link, map, elem_new);
1022 /* Add new element to the head of the list, so that
1023 * concurrent search will find it before old elem.
1025 hlist_add_head_rcu(&elem_new->node, &bucket->head);
1027 hlist_del_rcu(&elem->node);
1028 sock_map_unref(elem->sk, elem);
1029 sock_hash_free_elem(htab, elem);
1031 raw_spin_unlock_bh(&bucket->lock);
1034 raw_spin_unlock_bh(&bucket->lock);
1035 sk_psock_put(sk, psock);
1037 sk_psock_free_link(link);
1041 static int sock_hash_get_next_key(struct bpf_map *map, void *key,
1044 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1045 struct bpf_shtab_elem *elem, *elem_next;
1046 u32 hash, key_size = map->key_size;
1047 struct hlist_head *head;
1051 goto find_first_elem;
1052 hash = sock_hash_bucket_hash(key, key_size);
1053 head = &sock_hash_select_bucket(htab, hash)->head;
1054 elem = sock_hash_lookup_elem_raw(head, hash, key, key_size);
1056 goto find_first_elem;
1058 elem_next = hlist_entry_safe(rcu_dereference(hlist_next_rcu(&elem->node)),
1059 struct bpf_shtab_elem, node);
1061 memcpy(key_next, elem_next->key, key_size);
1065 i = hash & (htab->buckets_num - 1);
1068 for (; i < htab->buckets_num; i++) {
1069 head = &sock_hash_select_bucket(htab, i)->head;
1070 elem_next = hlist_entry_safe(rcu_dereference(hlist_first_rcu(head)),
1071 struct bpf_shtab_elem, node);
1073 memcpy(key_next, elem_next->key, key_size);
1081 static struct bpf_map *sock_hash_alloc(union bpf_attr *attr)
1083 struct bpf_shtab *htab;
1086 if (!capable(CAP_NET_ADMIN))
1087 return ERR_PTR(-EPERM);
1088 if (attr->max_entries == 0 ||
1089 attr->key_size == 0 ||
1090 (attr->value_size != sizeof(u32) &&
1091 attr->value_size != sizeof(u64)) ||
1092 attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
1093 return ERR_PTR(-EINVAL);
1094 if (attr->key_size > MAX_BPF_STACK)
1095 return ERR_PTR(-E2BIG);
1097 htab = kzalloc(sizeof(*htab), GFP_USER | __GFP_ACCOUNT);
1099 return ERR_PTR(-ENOMEM);
1101 bpf_map_init_from_attr(&htab->map, attr);
1103 htab->buckets_num = roundup_pow_of_two(htab->map.max_entries);
1104 htab->elem_size = sizeof(struct bpf_shtab_elem) +
1105 round_up(htab->map.key_size, 8);
1106 if (htab->buckets_num == 0 ||
1107 htab->buckets_num > U32_MAX / sizeof(struct bpf_shtab_bucket)) {
1112 htab->buckets = bpf_map_area_alloc(htab->buckets_num *
1113 sizeof(struct bpf_shtab_bucket),
1114 htab->map.numa_node);
1115 if (!htab->buckets) {
1120 for (i = 0; i < htab->buckets_num; i++) {
1121 INIT_HLIST_HEAD(&htab->buckets[i].head);
1122 raw_spin_lock_init(&htab->buckets[i].lock);
1128 return ERR_PTR(err);
1131 static void sock_hash_free(struct bpf_map *map)
1133 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1134 struct bpf_shtab_bucket *bucket;
1135 struct hlist_head unlink_list;
1136 struct bpf_shtab_elem *elem;
1137 struct hlist_node *node;
1140 /* After the sync no updates or deletes will be in-flight so it
1141 * is safe to walk map and remove entries without risking a race
1142 * in EEXIST update case.
1145 for (i = 0; i < htab->buckets_num; i++) {
1146 bucket = sock_hash_select_bucket(htab, i);
1148 /* We are racing with sock_hash_delete_from_link to
1149 * enter the spin-lock critical section. Every socket on
1150 * the list is still linked to sockhash. Since link
1151 * exists, psock exists and holds a ref to socket. That
1152 * lets us to grab a socket ref too.
1154 raw_spin_lock_bh(&bucket->lock);
1155 hlist_for_each_entry(elem, &bucket->head, node)
1156 sock_hold(elem->sk);
1157 hlist_move_list(&bucket->head, &unlink_list);
1158 raw_spin_unlock_bh(&bucket->lock);
1160 /* Process removed entries out of atomic context to
1161 * block for socket lock before deleting the psock's
1164 hlist_for_each_entry_safe(elem, node, &unlink_list, node) {
1165 hlist_del(&elem->node);
1166 lock_sock(elem->sk);
1168 sock_map_unref(elem->sk, elem);
1170 release_sock(elem->sk);
1172 sock_hash_free_elem(htab, elem);
1176 /* wait for psock readers accessing its map link */
1179 bpf_map_area_free(htab->buckets);
1183 static void *sock_hash_lookup_sys(struct bpf_map *map, void *key)
1187 if (map->value_size != sizeof(u64))
1188 return ERR_PTR(-ENOSPC);
1190 sk = __sock_hash_lookup_elem(map, key);
1192 return ERR_PTR(-ENOENT);
1194 __sock_gen_cookie(sk);
1195 return &sk->sk_cookie;
1198 static void *sock_hash_lookup(struct bpf_map *map, void *key)
1202 sk = __sock_hash_lookup_elem(map, key);
1205 if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
1210 static void sock_hash_release_progs(struct bpf_map *map)
1212 psock_progs_drop(&container_of(map, struct bpf_shtab, map)->progs);
1215 BPF_CALL_4(bpf_sock_hash_update, struct bpf_sock_ops_kern *, sops,
1216 struct bpf_map *, map, void *, key, u64, flags)
1218 WARN_ON_ONCE(!rcu_read_lock_held());
1220 if (likely(sock_map_sk_is_suitable(sops->sk) &&
1221 sock_map_op_okay(sops)))
1222 return sock_hash_update_common(map, key, sops->sk, flags);
1226 const struct bpf_func_proto bpf_sock_hash_update_proto = {
1227 .func = bpf_sock_hash_update,
1230 .ret_type = RET_INTEGER,
1231 .arg1_type = ARG_PTR_TO_CTX,
1232 .arg2_type = ARG_CONST_MAP_PTR,
1233 .arg3_type = ARG_PTR_TO_MAP_KEY,
1234 .arg4_type = ARG_ANYTHING,
1237 BPF_CALL_4(bpf_sk_redirect_hash, struct sk_buff *, skb,
1238 struct bpf_map *, map, void *, key, u64, flags)
1242 if (unlikely(flags & ~(BPF_F_INGRESS)))
1245 sk = __sock_hash_lookup_elem(map, key);
1246 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1249 skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
1253 const struct bpf_func_proto bpf_sk_redirect_hash_proto = {
1254 .func = bpf_sk_redirect_hash,
1256 .ret_type = RET_INTEGER,
1257 .arg1_type = ARG_PTR_TO_CTX,
1258 .arg2_type = ARG_CONST_MAP_PTR,
1259 .arg3_type = ARG_PTR_TO_MAP_KEY,
1260 .arg4_type = ARG_ANYTHING,
1263 BPF_CALL_4(bpf_msg_redirect_hash, struct sk_msg *, msg,
1264 struct bpf_map *, map, void *, key, u64, flags)
1268 if (unlikely(flags & ~(BPF_F_INGRESS)))
1271 sk = __sock_hash_lookup_elem(map, key);
1272 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1274 if (!(flags & BPF_F_INGRESS) && !sk_is_tcp(sk))
1282 const struct bpf_func_proto bpf_msg_redirect_hash_proto = {
1283 .func = bpf_msg_redirect_hash,
1285 .ret_type = RET_INTEGER,
1286 .arg1_type = ARG_PTR_TO_CTX,
1287 .arg2_type = ARG_CONST_MAP_PTR,
1288 .arg3_type = ARG_PTR_TO_MAP_KEY,
1289 .arg4_type = ARG_ANYTHING,
1292 struct sock_hash_seq_info {
1293 struct bpf_map *map;
1294 struct bpf_shtab *htab;
1298 static void *sock_hash_seq_find_next(struct sock_hash_seq_info *info,
1299 struct bpf_shtab_elem *prev_elem)
1301 const struct bpf_shtab *htab = info->htab;
1302 struct bpf_shtab_bucket *bucket;
1303 struct bpf_shtab_elem *elem;
1304 struct hlist_node *node;
1306 /* try to find next elem in the same bucket */
1308 node = rcu_dereference(hlist_next_rcu(&prev_elem->node));
1309 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1313 /* no more elements, continue in the next bucket */
1317 for (; info->bucket_id < htab->buckets_num; info->bucket_id++) {
1318 bucket = &htab->buckets[info->bucket_id];
1319 node = rcu_dereference(hlist_first_rcu(&bucket->head));
1320 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1328 static void *sock_hash_seq_start(struct seq_file *seq, loff_t *pos)
1331 struct sock_hash_seq_info *info = seq->private;
1336 /* pairs with sock_hash_seq_stop */
1338 return sock_hash_seq_find_next(info, NULL);
1341 static void *sock_hash_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1344 struct sock_hash_seq_info *info = seq->private;
1347 return sock_hash_seq_find_next(info, v);
1350 static int sock_hash_seq_show(struct seq_file *seq, void *v)
1353 struct sock_hash_seq_info *info = seq->private;
1354 struct bpf_iter__sockmap ctx = {};
1355 struct bpf_shtab_elem *elem = v;
1356 struct bpf_iter_meta meta;
1357 struct bpf_prog *prog;
1360 prog = bpf_iter_get_info(&meta, !elem);
1365 ctx.map = info->map;
1367 ctx.key = elem->key;
1371 return bpf_iter_run_prog(prog, &ctx);
1374 static void sock_hash_seq_stop(struct seq_file *seq, void *v)
1378 (void)sock_hash_seq_show(seq, NULL);
1380 /* pairs with sock_hash_seq_start */
1384 static const struct seq_operations sock_hash_seq_ops = {
1385 .start = sock_hash_seq_start,
1386 .next = sock_hash_seq_next,
1387 .stop = sock_hash_seq_stop,
1388 .show = sock_hash_seq_show,
1391 static int sock_hash_init_seq_private(void *priv_data,
1392 struct bpf_iter_aux_info *aux)
1394 struct sock_hash_seq_info *info = priv_data;
1396 bpf_map_inc_with_uref(aux->map);
1397 info->map = aux->map;
1398 info->htab = container_of(aux->map, struct bpf_shtab, map);
1402 static void sock_hash_fini_seq_private(void *priv_data)
1404 struct sock_hash_seq_info *info = priv_data;
1406 bpf_map_put_with_uref(info->map);
1409 static const struct bpf_iter_seq_info sock_hash_iter_seq_info = {
1410 .seq_ops = &sock_hash_seq_ops,
1411 .init_seq_private = sock_hash_init_seq_private,
1412 .fini_seq_private = sock_hash_fini_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 static 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 if (progs->skb_verdict)
1468 pprog = &progs->stream_verdict;
1470 case BPF_SK_SKB_VERDICT:
1471 if (progs->stream_verdict)
1473 pprog = &progs->skb_verdict;
1480 return psock_replace_prog(pprog, prog, old);
1482 psock_set_prog(pprog, prog);
1486 static void sock_map_unlink(struct sock *sk, struct sk_psock_link *link)
1488 switch (link->map->map_type) {
1489 case BPF_MAP_TYPE_SOCKMAP:
1490 return sock_map_delete_from_link(link->map, sk,
1492 case BPF_MAP_TYPE_SOCKHASH:
1493 return sock_hash_delete_from_link(link->map, sk,
1500 static void sock_map_remove_links(struct sock *sk, struct sk_psock *psock)
1502 struct sk_psock_link *link;
1504 while ((link = sk_psock_link_pop(psock))) {
1505 sock_map_unlink(sk, link);
1506 sk_psock_free_link(link);
1510 void sock_map_unhash(struct sock *sk)
1512 void (*saved_unhash)(struct sock *sk);
1513 struct sk_psock *psock;
1516 psock = sk_psock(sk);
1517 if (unlikely(!psock)) {
1519 saved_unhash = READ_ONCE(sk->sk_prot)->unhash;
1521 saved_unhash = psock->saved_unhash;
1522 sock_map_remove_links(sk, psock);
1525 if (WARN_ON_ONCE(saved_unhash == sock_map_unhash))
1530 EXPORT_SYMBOL_GPL(sock_map_unhash);
1532 void sock_map_destroy(struct sock *sk)
1534 void (*saved_destroy)(struct sock *sk);
1535 struct sk_psock *psock;
1538 psock = sk_psock_get(sk);
1539 if (unlikely(!psock)) {
1541 saved_destroy = READ_ONCE(sk->sk_prot)->destroy;
1543 saved_destroy = psock->saved_destroy;
1544 sock_map_remove_links(sk, psock);
1546 sk_psock_stop(psock);
1547 sk_psock_put(sk, psock);
1549 if (WARN_ON_ONCE(saved_destroy == sock_map_destroy))
1554 EXPORT_SYMBOL_GPL(sock_map_destroy);
1556 void sock_map_close(struct sock *sk, long timeout)
1558 void (*saved_close)(struct sock *sk, long timeout);
1559 struct sk_psock *psock;
1563 psock = sk_psock_get(sk);
1564 if (unlikely(!psock)) {
1567 saved_close = READ_ONCE(sk->sk_prot)->close;
1569 saved_close = psock->saved_close;
1570 sock_map_remove_links(sk, psock);
1572 sk_psock_stop(psock);
1574 cancel_work_sync(&psock->work);
1575 sk_psock_put(sk, psock);
1577 /* Make sure we do not recurse. This is a bug.
1578 * Leak the socket instead of crashing on a stack overflow.
1580 if (WARN_ON_ONCE(saved_close == sock_map_close))
1582 saved_close(sk, timeout);
1584 EXPORT_SYMBOL_GPL(sock_map_close);
1586 static int sock_map_iter_attach_target(struct bpf_prog *prog,
1587 union bpf_iter_link_info *linfo,
1588 struct bpf_iter_aux_info *aux)
1590 struct bpf_map *map;
1593 if (!linfo->map.map_fd)
1596 map = bpf_map_get_with_uref(linfo->map.map_fd);
1598 return PTR_ERR(map);
1600 if (map->map_type != BPF_MAP_TYPE_SOCKMAP &&
1601 map->map_type != BPF_MAP_TYPE_SOCKHASH)
1604 if (prog->aux->max_rdonly_access > map->key_size) {
1613 bpf_map_put_with_uref(map);
1617 static void sock_map_iter_detach_target(struct bpf_iter_aux_info *aux)
1619 bpf_map_put_with_uref(aux->map);
1622 static struct bpf_iter_reg sock_map_iter_reg = {
1623 .target = "sockmap",
1624 .attach_target = sock_map_iter_attach_target,
1625 .detach_target = sock_map_iter_detach_target,
1626 .show_fdinfo = bpf_iter_map_show_fdinfo,
1627 .fill_link_info = bpf_iter_map_fill_link_info,
1628 .ctx_arg_info_size = 2,
1630 { offsetof(struct bpf_iter__sockmap, key),
1631 PTR_TO_BUF | PTR_MAYBE_NULL | MEM_RDONLY },
1632 { offsetof(struct bpf_iter__sockmap, sk),
1633 PTR_TO_BTF_ID_OR_NULL },
1637 static int __init bpf_sockmap_iter_init(void)
1639 sock_map_iter_reg.ctx_arg_info[1].btf_id =
1640 btf_sock_ids[BTF_SOCK_TYPE_SOCK];
1641 return bpf_iter_reg_target(&sock_map_iter_reg);
1643 late_initcall(bpf_sockmap_iter_init);