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;
33 if (!capable(CAP_NET_ADMIN))
34 return ERR_PTR(-EPERM);
35 if (attr->max_entries == 0 ||
36 attr->key_size != 4 ||
37 (attr->value_size != sizeof(u32) &&
38 attr->value_size != sizeof(u64)) ||
39 attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
40 return ERR_PTR(-EINVAL);
42 stab = kzalloc(sizeof(*stab), GFP_USER);
44 return ERR_PTR(-ENOMEM);
46 bpf_map_init_from_attr(&stab->map, attr);
47 raw_spin_lock_init(&stab->lock);
49 /* Make sure page count doesn't overflow. */
50 cost = (u64) stab->map.max_entries * sizeof(struct sock *);
51 err = bpf_map_charge_init(&stab->map.memory, cost);
55 stab->sks = bpf_map_area_alloc((u64) stab->map.max_entries *
56 sizeof(struct sock *),
61 bpf_map_charge_finish(&stab->map.memory);
67 int sock_map_get_from_fd(const union bpf_attr *attr, struct bpf_prog *prog)
69 u32 ufd = attr->target_fd;
74 if (attr->attach_flags || attr->replace_bpf_fd)
78 map = __bpf_map_get(f);
81 ret = sock_map_prog_update(map, prog, NULL, attr->attach_type);
86 int sock_map_prog_detach(const union bpf_attr *attr, enum bpf_prog_type ptype)
88 u32 ufd = attr->target_fd;
89 struct bpf_prog *prog;
94 if (attr->attach_flags || attr->replace_bpf_fd)
98 map = __bpf_map_get(f);
102 prog = bpf_prog_get(attr->attach_bpf_fd);
108 if (prog->type != ptype) {
113 ret = sock_map_prog_update(map, NULL, prog, attr->attach_type);
121 static void sock_map_sk_acquire(struct sock *sk)
122 __acquires(&sk->sk_lock.slock)
128 static void sock_map_sk_release(struct sock *sk)
129 __releases(&sk->sk_lock.slock)
135 static void sock_map_add_link(struct sk_psock *psock,
136 struct sk_psock_link *link,
137 struct bpf_map *map, void *link_raw)
139 link->link_raw = link_raw;
141 spin_lock_bh(&psock->link_lock);
142 list_add_tail(&link->list, &psock->link);
143 spin_unlock_bh(&psock->link_lock);
146 static void sock_map_del_link(struct sock *sk,
147 struct sk_psock *psock, void *link_raw)
149 bool strp_stop = false, verdict_stop = false;
150 struct sk_psock_link *link, *tmp;
152 spin_lock_bh(&psock->link_lock);
153 list_for_each_entry_safe(link, tmp, &psock->link, list) {
154 if (link->link_raw == link_raw) {
155 struct bpf_map *map = link->map;
156 struct bpf_stab *stab = container_of(map, struct bpf_stab,
158 if (psock->parser.enabled && stab->progs.skb_parser)
160 if (psock->parser.enabled && stab->progs.skb_verdict)
162 list_del(&link->list);
163 sk_psock_free_link(link);
166 spin_unlock_bh(&psock->link_lock);
167 if (strp_stop || verdict_stop) {
168 write_lock_bh(&sk->sk_callback_lock);
170 sk_psock_stop_strp(sk, psock);
172 sk_psock_stop_verdict(sk, psock);
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)
191 switch (sk->sk_type) {
193 prot = tcp_bpf_get_proto(sk, psock);
197 prot = udp_bpf_get_proto(sk, psock);
205 return PTR_ERR(prot);
207 sk_psock_update_proto(sk, psock, prot);
211 static struct sk_psock *sock_map_psock_get_checked(struct sock *sk)
213 struct sk_psock *psock;
216 psock = sk_psock(sk);
218 if (sk->sk_prot->close != sock_map_close) {
219 psock = ERR_PTR(-EBUSY);
223 if (!refcount_inc_not_zero(&psock->refcnt))
224 psock = ERR_PTR(-EBUSY);
231 static int sock_map_link(struct bpf_map *map, struct sk_psock_progs *progs,
234 struct bpf_prog *msg_parser, *skb_parser, *skb_verdict;
235 struct sk_psock *psock;
238 skb_verdict = READ_ONCE(progs->skb_verdict);
240 skb_verdict = bpf_prog_inc_not_zero(skb_verdict);
241 if (IS_ERR(skb_verdict))
242 return PTR_ERR(skb_verdict);
245 skb_parser = READ_ONCE(progs->skb_parser);
247 skb_parser = bpf_prog_inc_not_zero(skb_parser);
248 if (IS_ERR(skb_parser)) {
249 ret = PTR_ERR(skb_parser);
250 goto out_put_skb_verdict;
254 msg_parser = READ_ONCE(progs->msg_parser);
256 msg_parser = bpf_prog_inc_not_zero(msg_parser);
257 if (IS_ERR(msg_parser)) {
258 ret = PTR_ERR(msg_parser);
259 goto out_put_skb_parser;
263 psock = sock_map_psock_get_checked(sk);
265 ret = PTR_ERR(psock);
270 if ((msg_parser && READ_ONCE(psock->progs.msg_parser)) ||
271 (skb_parser && READ_ONCE(psock->progs.skb_parser)) ||
272 (skb_verdict && READ_ONCE(psock->progs.skb_verdict))) {
273 sk_psock_put(sk, psock);
278 psock = sk_psock_init(sk, map->numa_node);
280 ret = PTR_ERR(psock);
286 psock_set_prog(&psock->progs.msg_parser, msg_parser);
288 ret = sock_map_init_proto(sk, psock);
292 write_lock_bh(&sk->sk_callback_lock);
293 if (skb_parser && skb_verdict && !psock->parser.enabled) {
294 ret = sk_psock_init_strp(sk, psock);
296 goto out_unlock_drop;
297 psock_set_prog(&psock->progs.skb_verdict, skb_verdict);
298 psock_set_prog(&psock->progs.skb_parser, skb_parser);
299 sk_psock_start_strp(sk, psock);
300 } else if (!skb_parser && skb_verdict && !psock->parser.enabled) {
301 psock_set_prog(&psock->progs.skb_verdict, skb_verdict);
302 sk_psock_start_verdict(sk,psock);
304 write_unlock_bh(&sk->sk_callback_lock);
307 write_unlock_bh(&sk->sk_callback_lock);
309 sk_psock_put(sk, psock);
312 bpf_prog_put(msg_parser);
315 bpf_prog_put(skb_parser);
318 bpf_prog_put(skb_verdict);
322 static int sock_map_link_no_progs(struct bpf_map *map, struct sock *sk)
324 struct sk_psock *psock;
327 psock = sock_map_psock_get_checked(sk);
329 return PTR_ERR(psock);
332 psock = sk_psock_init(sk, map->numa_node);
334 return PTR_ERR(psock);
337 ret = sock_map_init_proto(sk, psock);
339 sk_psock_put(sk, psock);
343 static void sock_map_free(struct bpf_map *map)
345 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
348 /* After the sync no updates or deletes will be in-flight so it
349 * is safe to walk map and remove entries without risking a race
350 * in EEXIST update case.
353 for (i = 0; i < stab->map.max_entries; i++) {
354 struct sock **psk = &stab->sks[i];
357 sk = xchg(psk, NULL);
362 sock_map_unref(sk, psk);
369 /* wait for psock readers accessing its map link */
372 bpf_map_area_free(stab->sks);
376 static void sock_map_release_progs(struct bpf_map *map)
378 psock_progs_drop(&container_of(map, struct bpf_stab, map)->progs);
381 static struct sock *__sock_map_lookup_elem(struct bpf_map *map, u32 key)
383 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
385 WARN_ON_ONCE(!rcu_read_lock_held());
387 if (unlikely(key >= map->max_entries))
389 return READ_ONCE(stab->sks[key]);
392 static void *sock_map_lookup(struct bpf_map *map, void *key)
396 sk = __sock_map_lookup_elem(map, *(u32 *)key);
399 if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
404 static void *sock_map_lookup_sys(struct bpf_map *map, void *key)
408 if (map->value_size != sizeof(u64))
409 return ERR_PTR(-ENOSPC);
411 sk = __sock_map_lookup_elem(map, *(u32 *)key);
413 return ERR_PTR(-ENOENT);
415 __sock_gen_cookie(sk);
416 return &sk->sk_cookie;
419 static int __sock_map_delete(struct bpf_stab *stab, struct sock *sk_test,
426 return -EOPNOTSUPP; /* locks here are hardirq-unsafe */
428 raw_spin_lock_bh(&stab->lock);
430 if (!sk_test || sk_test == sk)
431 sk = xchg(psk, NULL);
434 sock_map_unref(sk, psk);
438 raw_spin_unlock_bh(&stab->lock);
442 static void sock_map_delete_from_link(struct bpf_map *map, struct sock *sk,
445 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
447 __sock_map_delete(stab, sk, link_raw);
450 static int sock_map_delete_elem(struct bpf_map *map, void *key)
452 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
456 if (unlikely(i >= map->max_entries))
460 return __sock_map_delete(stab, NULL, psk);
463 static int sock_map_get_next_key(struct bpf_map *map, void *key, void *next)
465 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
466 u32 i = key ? *(u32 *)key : U32_MAX;
467 u32 *key_next = next;
469 if (i == stab->map.max_entries - 1)
471 if (i >= stab->map.max_entries)
478 static bool sock_map_redirect_allowed(const struct sock *sk);
480 static int sock_map_update_common(struct bpf_map *map, u32 idx,
481 struct sock *sk, u64 flags)
483 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
484 struct sk_psock_link *link;
485 struct sk_psock *psock;
489 WARN_ON_ONCE(!rcu_read_lock_held());
490 if (unlikely(flags > BPF_EXIST))
492 if (unlikely(idx >= map->max_entries))
495 link = sk_psock_init_link();
499 /* Only sockets we can redirect into/from in BPF need to hold
500 * refs to parser/verdict progs and have their sk_data_ready
501 * and sk_write_space callbacks overridden.
503 if (sock_map_redirect_allowed(sk))
504 ret = sock_map_link(map, &stab->progs, sk);
506 ret = sock_map_link_no_progs(map, sk);
510 psock = sk_psock(sk);
511 WARN_ON_ONCE(!psock);
513 raw_spin_lock_bh(&stab->lock);
514 osk = stab->sks[idx];
515 if (osk && flags == BPF_NOEXIST) {
518 } else if (!osk && flags == BPF_EXIST) {
523 sock_map_add_link(psock, link, map, &stab->sks[idx]);
526 sock_map_unref(osk, &stab->sks[idx]);
527 raw_spin_unlock_bh(&stab->lock);
530 raw_spin_unlock_bh(&stab->lock);
532 sk_psock_put(sk, psock);
534 sk_psock_free_link(link);
538 static bool sock_map_op_okay(const struct bpf_sock_ops_kern *ops)
540 return ops->op == BPF_SOCK_OPS_PASSIVE_ESTABLISHED_CB ||
541 ops->op == BPF_SOCK_OPS_ACTIVE_ESTABLISHED_CB ||
542 ops->op == BPF_SOCK_OPS_TCP_LISTEN_CB;
545 static bool sk_is_tcp(const struct sock *sk)
547 return sk->sk_type == SOCK_STREAM &&
548 sk->sk_protocol == IPPROTO_TCP;
551 static bool sk_is_udp(const struct sock *sk)
553 return sk->sk_type == SOCK_DGRAM &&
554 sk->sk_protocol == IPPROTO_UDP;
557 static bool sock_map_redirect_allowed(const struct sock *sk)
559 return sk_is_tcp(sk) && sk->sk_state != TCP_LISTEN;
562 static bool sock_map_sk_is_suitable(const struct sock *sk)
564 return sk_is_tcp(sk) || sk_is_udp(sk);
567 static bool sock_map_sk_state_allowed(const struct sock *sk)
570 return (1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_LISTEN);
571 else if (sk_is_udp(sk))
572 return sk_hashed(sk);
577 static int sock_hash_update_common(struct bpf_map *map, void *key,
578 struct sock *sk, u64 flags);
580 int sock_map_update_elem_sys(struct bpf_map *map, void *key, void *value,
588 if (map->value_size == sizeof(u64))
595 sock = sockfd_lookup(ufd, &ret);
603 if (!sock_map_sk_is_suitable(sk)) {
608 sock_map_sk_acquire(sk);
609 if (!sock_map_sk_state_allowed(sk))
611 else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
612 ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
614 ret = sock_hash_update_common(map, key, sk, flags);
615 sock_map_sk_release(sk);
621 static int sock_map_update_elem(struct bpf_map *map, void *key,
622 void *value, u64 flags)
624 struct sock *sk = (struct sock *)value;
627 if (unlikely(!sk || !sk_fullsock(sk)))
630 if (!sock_map_sk_is_suitable(sk))
635 if (!sock_map_sk_state_allowed(sk))
637 else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
638 ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
640 ret = sock_hash_update_common(map, key, sk, flags);
646 BPF_CALL_4(bpf_sock_map_update, struct bpf_sock_ops_kern *, sops,
647 struct bpf_map *, map, void *, key, u64, flags)
649 WARN_ON_ONCE(!rcu_read_lock_held());
651 if (likely(sock_map_sk_is_suitable(sops->sk) &&
652 sock_map_op_okay(sops)))
653 return sock_map_update_common(map, *(u32 *)key, sops->sk,
658 const struct bpf_func_proto bpf_sock_map_update_proto = {
659 .func = bpf_sock_map_update,
662 .ret_type = RET_INTEGER,
663 .arg1_type = ARG_PTR_TO_CTX,
664 .arg2_type = ARG_CONST_MAP_PTR,
665 .arg3_type = ARG_PTR_TO_MAP_KEY,
666 .arg4_type = ARG_ANYTHING,
669 BPF_CALL_4(bpf_sk_redirect_map, struct sk_buff *, skb,
670 struct bpf_map *, map, u32, key, u64, flags)
672 struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
675 if (unlikely(flags & ~(BPF_F_INGRESS)))
678 sk = __sock_map_lookup_elem(map, key);
679 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
682 tcb->bpf.flags = flags;
683 tcb->bpf.sk_redir = sk;
687 const struct bpf_func_proto bpf_sk_redirect_map_proto = {
688 .func = bpf_sk_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 BPF_CALL_4(bpf_msg_redirect_map, struct sk_msg *, msg,
698 struct bpf_map *, map, u32, key, u64, flags)
702 if (unlikely(flags & ~(BPF_F_INGRESS)))
705 sk = __sock_map_lookup_elem(map, key);
706 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
714 const struct bpf_func_proto bpf_msg_redirect_map_proto = {
715 .func = bpf_msg_redirect_map,
717 .ret_type = RET_INTEGER,
718 .arg1_type = ARG_PTR_TO_CTX,
719 .arg2_type = ARG_CONST_MAP_PTR,
720 .arg3_type = ARG_ANYTHING,
721 .arg4_type = ARG_ANYTHING,
724 struct sock_map_seq_info {
730 struct bpf_iter__sockmap {
731 __bpf_md_ptr(struct bpf_iter_meta *, meta);
732 __bpf_md_ptr(struct bpf_map *, map);
733 __bpf_md_ptr(void *, key);
734 __bpf_md_ptr(struct sock *, sk);
737 DEFINE_BPF_ITER_FUNC(sockmap, struct bpf_iter_meta *meta,
738 struct bpf_map *map, void *key,
741 static void *sock_map_seq_lookup_elem(struct sock_map_seq_info *info)
743 if (unlikely(info->index >= info->map->max_entries))
746 info->sk = __sock_map_lookup_elem(info->map, info->index);
748 /* can't return sk directly, since that might be NULL */
752 static void *sock_map_seq_start(struct seq_file *seq, loff_t *pos)
755 struct sock_map_seq_info *info = seq->private;
760 /* pairs with sock_map_seq_stop */
762 return sock_map_seq_lookup_elem(info);
765 static void *sock_map_seq_next(struct seq_file *seq, void *v, loff_t *pos)
768 struct sock_map_seq_info *info = seq->private;
773 return sock_map_seq_lookup_elem(info);
776 static int sock_map_seq_show(struct seq_file *seq, void *v)
779 struct sock_map_seq_info *info = seq->private;
780 struct bpf_iter__sockmap ctx = {};
781 struct bpf_iter_meta meta;
782 struct bpf_prog *prog;
785 prog = bpf_iter_get_info(&meta, !v);
792 ctx.key = &info->index;
796 return bpf_iter_run_prog(prog, &ctx);
799 static void sock_map_seq_stop(struct seq_file *seq, void *v)
803 (void)sock_map_seq_show(seq, NULL);
805 /* pairs with sock_map_seq_start */
809 static const struct seq_operations sock_map_seq_ops = {
810 .start = sock_map_seq_start,
811 .next = sock_map_seq_next,
812 .stop = sock_map_seq_stop,
813 .show = sock_map_seq_show,
816 static int sock_map_init_seq_private(void *priv_data,
817 struct bpf_iter_aux_info *aux)
819 struct sock_map_seq_info *info = priv_data;
821 bpf_map_inc_with_uref(aux->map);
822 info->map = aux->map;
826 static void sock_map_fini_seq_private(void *priv_data)
828 struct sock_map_seq_info *info = priv_data;
830 bpf_map_put_with_uref(info->map);
833 static const struct bpf_iter_seq_info sock_map_iter_seq_info = {
834 .seq_ops = &sock_map_seq_ops,
835 .init_seq_private = sock_map_init_seq_private,
836 .fini_seq_private = sock_map_fini_seq_private,
837 .seq_priv_size = sizeof(struct sock_map_seq_info),
840 static int sock_map_btf_id;
841 const struct bpf_map_ops sock_map_ops = {
842 .map_meta_equal = bpf_map_meta_equal,
843 .map_alloc = sock_map_alloc,
844 .map_free = sock_map_free,
845 .map_get_next_key = sock_map_get_next_key,
846 .map_lookup_elem_sys_only = sock_map_lookup_sys,
847 .map_update_elem = sock_map_update_elem,
848 .map_delete_elem = sock_map_delete_elem,
849 .map_lookup_elem = sock_map_lookup,
850 .map_release_uref = sock_map_release_progs,
851 .map_check_btf = map_check_no_btf,
852 .map_btf_name = "bpf_stab",
853 .map_btf_id = &sock_map_btf_id,
854 .iter_seq_info = &sock_map_iter_seq_info,
857 struct bpf_shtab_elem {
861 struct hlist_node node;
865 struct bpf_shtab_bucket {
866 struct hlist_head head;
872 struct bpf_shtab_bucket *buckets;
875 struct sk_psock_progs progs;
879 static inline u32 sock_hash_bucket_hash(const void *key, u32 len)
881 return jhash(key, len, 0);
884 static struct bpf_shtab_bucket *sock_hash_select_bucket(struct bpf_shtab *htab,
887 return &htab->buckets[hash & (htab->buckets_num - 1)];
890 static struct bpf_shtab_elem *
891 sock_hash_lookup_elem_raw(struct hlist_head *head, u32 hash, void *key,
894 struct bpf_shtab_elem *elem;
896 hlist_for_each_entry_rcu(elem, head, node) {
897 if (elem->hash == hash &&
898 !memcmp(&elem->key, key, key_size))
905 static struct sock *__sock_hash_lookup_elem(struct bpf_map *map, void *key)
907 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
908 u32 key_size = map->key_size, hash;
909 struct bpf_shtab_bucket *bucket;
910 struct bpf_shtab_elem *elem;
912 WARN_ON_ONCE(!rcu_read_lock_held());
914 hash = sock_hash_bucket_hash(key, key_size);
915 bucket = sock_hash_select_bucket(htab, hash);
916 elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
918 return elem ? elem->sk : NULL;
921 static void sock_hash_free_elem(struct bpf_shtab *htab,
922 struct bpf_shtab_elem *elem)
924 atomic_dec(&htab->count);
925 kfree_rcu(elem, rcu);
928 static void sock_hash_delete_from_link(struct bpf_map *map, struct sock *sk,
931 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
932 struct bpf_shtab_elem *elem_probe, *elem = link_raw;
933 struct bpf_shtab_bucket *bucket;
935 WARN_ON_ONCE(!rcu_read_lock_held());
936 bucket = sock_hash_select_bucket(htab, elem->hash);
938 /* elem may be deleted in parallel from the map, but access here
939 * is okay since it's going away only after RCU grace period.
940 * However, we need to check whether it's still present.
942 raw_spin_lock_bh(&bucket->lock);
943 elem_probe = sock_hash_lookup_elem_raw(&bucket->head, elem->hash,
944 elem->key, map->key_size);
945 if (elem_probe && elem_probe == elem) {
946 hlist_del_rcu(&elem->node);
947 sock_map_unref(elem->sk, elem);
948 sock_hash_free_elem(htab, elem);
950 raw_spin_unlock_bh(&bucket->lock);
953 static int sock_hash_delete_elem(struct bpf_map *map, void *key)
955 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
956 u32 hash, key_size = map->key_size;
957 struct bpf_shtab_bucket *bucket;
958 struct bpf_shtab_elem *elem;
962 return -EOPNOTSUPP; /* locks here are hardirq-unsafe */
964 hash = sock_hash_bucket_hash(key, key_size);
965 bucket = sock_hash_select_bucket(htab, hash);
967 raw_spin_lock_bh(&bucket->lock);
968 elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
970 hlist_del_rcu(&elem->node);
971 sock_map_unref(elem->sk, elem);
972 sock_hash_free_elem(htab, elem);
975 raw_spin_unlock_bh(&bucket->lock);
979 static struct bpf_shtab_elem *sock_hash_alloc_elem(struct bpf_shtab *htab,
980 void *key, u32 key_size,
981 u32 hash, struct sock *sk,
982 struct bpf_shtab_elem *old)
984 struct bpf_shtab_elem *new;
986 if (atomic_inc_return(&htab->count) > htab->map.max_entries) {
988 atomic_dec(&htab->count);
989 return ERR_PTR(-E2BIG);
993 new = kmalloc_node(htab->elem_size, GFP_ATOMIC | __GFP_NOWARN,
994 htab->map.numa_node);
996 atomic_dec(&htab->count);
997 return ERR_PTR(-ENOMEM);
999 memcpy(new->key, key, key_size);
1005 static int sock_hash_update_common(struct bpf_map *map, void *key,
1006 struct sock *sk, u64 flags)
1008 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1009 u32 key_size = map->key_size, hash;
1010 struct bpf_shtab_elem *elem, *elem_new;
1011 struct bpf_shtab_bucket *bucket;
1012 struct sk_psock_link *link;
1013 struct sk_psock *psock;
1016 WARN_ON_ONCE(!rcu_read_lock_held());
1017 if (unlikely(flags > BPF_EXIST))
1020 link = sk_psock_init_link();
1024 /* Only sockets we can redirect into/from in BPF need to hold
1025 * refs to parser/verdict progs and have their sk_data_ready
1026 * and sk_write_space callbacks overridden.
1028 if (sock_map_redirect_allowed(sk))
1029 ret = sock_map_link(map, &htab->progs, sk);
1031 ret = sock_map_link_no_progs(map, sk);
1035 psock = sk_psock(sk);
1036 WARN_ON_ONCE(!psock);
1038 hash = sock_hash_bucket_hash(key, key_size);
1039 bucket = sock_hash_select_bucket(htab, hash);
1041 raw_spin_lock_bh(&bucket->lock);
1042 elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
1043 if (elem && flags == BPF_NOEXIST) {
1046 } else if (!elem && flags == BPF_EXIST) {
1051 elem_new = sock_hash_alloc_elem(htab, key, key_size, hash, sk, elem);
1052 if (IS_ERR(elem_new)) {
1053 ret = PTR_ERR(elem_new);
1057 sock_map_add_link(psock, link, map, elem_new);
1058 /* Add new element to the head of the list, so that
1059 * concurrent search will find it before old elem.
1061 hlist_add_head_rcu(&elem_new->node, &bucket->head);
1063 hlist_del_rcu(&elem->node);
1064 sock_map_unref(elem->sk, elem);
1065 sock_hash_free_elem(htab, elem);
1067 raw_spin_unlock_bh(&bucket->lock);
1070 raw_spin_unlock_bh(&bucket->lock);
1071 sk_psock_put(sk, psock);
1073 sk_psock_free_link(link);
1077 static int sock_hash_get_next_key(struct bpf_map *map, void *key,
1080 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1081 struct bpf_shtab_elem *elem, *elem_next;
1082 u32 hash, key_size = map->key_size;
1083 struct hlist_head *head;
1087 goto find_first_elem;
1088 hash = sock_hash_bucket_hash(key, key_size);
1089 head = &sock_hash_select_bucket(htab, hash)->head;
1090 elem = sock_hash_lookup_elem_raw(head, hash, key, key_size);
1092 goto find_first_elem;
1094 elem_next = hlist_entry_safe(rcu_dereference(hlist_next_rcu(&elem->node)),
1095 struct bpf_shtab_elem, node);
1097 memcpy(key_next, elem_next->key, key_size);
1101 i = hash & (htab->buckets_num - 1);
1104 for (; i < htab->buckets_num; i++) {
1105 head = &sock_hash_select_bucket(htab, i)->head;
1106 elem_next = hlist_entry_safe(rcu_dereference(hlist_first_rcu(head)),
1107 struct bpf_shtab_elem, node);
1109 memcpy(key_next, elem_next->key, key_size);
1117 static struct bpf_map *sock_hash_alloc(union bpf_attr *attr)
1119 struct bpf_shtab *htab;
1123 if (!capable(CAP_NET_ADMIN))
1124 return ERR_PTR(-EPERM);
1125 if (attr->max_entries == 0 ||
1126 attr->key_size == 0 ||
1127 (attr->value_size != sizeof(u32) &&
1128 attr->value_size != sizeof(u64)) ||
1129 attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
1130 return ERR_PTR(-EINVAL);
1131 if (attr->key_size > MAX_BPF_STACK)
1132 return ERR_PTR(-E2BIG);
1134 htab = kzalloc(sizeof(*htab), GFP_USER);
1136 return ERR_PTR(-ENOMEM);
1138 bpf_map_init_from_attr(&htab->map, attr);
1140 htab->buckets_num = roundup_pow_of_two(htab->map.max_entries);
1141 htab->elem_size = sizeof(struct bpf_shtab_elem) +
1142 round_up(htab->map.key_size, 8);
1143 if (htab->buckets_num == 0 ||
1144 htab->buckets_num > U32_MAX / sizeof(struct bpf_shtab_bucket)) {
1149 cost = (u64) htab->buckets_num * sizeof(struct bpf_shtab_bucket) +
1150 (u64) htab->elem_size * htab->map.max_entries;
1151 if (cost >= U32_MAX - PAGE_SIZE) {
1155 err = bpf_map_charge_init(&htab->map.memory, cost);
1159 htab->buckets = bpf_map_area_alloc(htab->buckets_num *
1160 sizeof(struct bpf_shtab_bucket),
1161 htab->map.numa_node);
1162 if (!htab->buckets) {
1163 bpf_map_charge_finish(&htab->map.memory);
1168 for (i = 0; i < htab->buckets_num; i++) {
1169 INIT_HLIST_HEAD(&htab->buckets[i].head);
1170 raw_spin_lock_init(&htab->buckets[i].lock);
1176 return ERR_PTR(err);
1179 static void sock_hash_free(struct bpf_map *map)
1181 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1182 struct bpf_shtab_bucket *bucket;
1183 struct hlist_head unlink_list;
1184 struct bpf_shtab_elem *elem;
1185 struct hlist_node *node;
1188 /* After the sync no updates or deletes will be in-flight so it
1189 * is safe to walk map and remove entries without risking a race
1190 * in EEXIST update case.
1193 for (i = 0; i < htab->buckets_num; i++) {
1194 bucket = sock_hash_select_bucket(htab, i);
1196 /* We are racing with sock_hash_delete_from_link to
1197 * enter the spin-lock critical section. Every socket on
1198 * the list is still linked to sockhash. Since link
1199 * exists, psock exists and holds a ref to socket. That
1200 * lets us to grab a socket ref too.
1202 raw_spin_lock_bh(&bucket->lock);
1203 hlist_for_each_entry(elem, &bucket->head, node)
1204 sock_hold(elem->sk);
1205 hlist_move_list(&bucket->head, &unlink_list);
1206 raw_spin_unlock_bh(&bucket->lock);
1208 /* Process removed entries out of atomic context to
1209 * block for socket lock before deleting the psock's
1212 hlist_for_each_entry_safe(elem, node, &unlink_list, node) {
1213 hlist_del(&elem->node);
1214 lock_sock(elem->sk);
1216 sock_map_unref(elem->sk, elem);
1218 release_sock(elem->sk);
1220 sock_hash_free_elem(htab, elem);
1224 /* wait for psock readers accessing its map link */
1227 bpf_map_area_free(htab->buckets);
1231 static void *sock_hash_lookup_sys(struct bpf_map *map, void *key)
1235 if (map->value_size != sizeof(u64))
1236 return ERR_PTR(-ENOSPC);
1238 sk = __sock_hash_lookup_elem(map, key);
1240 return ERR_PTR(-ENOENT);
1242 __sock_gen_cookie(sk);
1243 return &sk->sk_cookie;
1246 static void *sock_hash_lookup(struct bpf_map *map, void *key)
1250 sk = __sock_hash_lookup_elem(map, key);
1253 if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
1258 static void sock_hash_release_progs(struct bpf_map *map)
1260 psock_progs_drop(&container_of(map, struct bpf_shtab, map)->progs);
1263 BPF_CALL_4(bpf_sock_hash_update, struct bpf_sock_ops_kern *, sops,
1264 struct bpf_map *, map, void *, key, u64, flags)
1266 WARN_ON_ONCE(!rcu_read_lock_held());
1268 if (likely(sock_map_sk_is_suitable(sops->sk) &&
1269 sock_map_op_okay(sops)))
1270 return sock_hash_update_common(map, key, sops->sk, flags);
1274 const struct bpf_func_proto bpf_sock_hash_update_proto = {
1275 .func = bpf_sock_hash_update,
1278 .ret_type = RET_INTEGER,
1279 .arg1_type = ARG_PTR_TO_CTX,
1280 .arg2_type = ARG_CONST_MAP_PTR,
1281 .arg3_type = ARG_PTR_TO_MAP_KEY,
1282 .arg4_type = ARG_ANYTHING,
1285 BPF_CALL_4(bpf_sk_redirect_hash, struct sk_buff *, skb,
1286 struct bpf_map *, map, void *, key, u64, flags)
1288 struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
1291 if (unlikely(flags & ~(BPF_F_INGRESS)))
1294 sk = __sock_hash_lookup_elem(map, key);
1295 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1298 tcb->bpf.flags = flags;
1299 tcb->bpf.sk_redir = sk;
1303 const struct bpf_func_proto bpf_sk_redirect_hash_proto = {
1304 .func = bpf_sk_redirect_hash,
1306 .ret_type = RET_INTEGER,
1307 .arg1_type = ARG_PTR_TO_CTX,
1308 .arg2_type = ARG_CONST_MAP_PTR,
1309 .arg3_type = ARG_PTR_TO_MAP_KEY,
1310 .arg4_type = ARG_ANYTHING,
1313 BPF_CALL_4(bpf_msg_redirect_hash, struct sk_msg *, msg,
1314 struct bpf_map *, map, void *, key, u64, flags)
1318 if (unlikely(flags & ~(BPF_F_INGRESS)))
1321 sk = __sock_hash_lookup_elem(map, key);
1322 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1330 const struct bpf_func_proto bpf_msg_redirect_hash_proto = {
1331 .func = bpf_msg_redirect_hash,
1333 .ret_type = RET_INTEGER,
1334 .arg1_type = ARG_PTR_TO_CTX,
1335 .arg2_type = ARG_CONST_MAP_PTR,
1336 .arg3_type = ARG_PTR_TO_MAP_KEY,
1337 .arg4_type = ARG_ANYTHING,
1340 struct sock_hash_seq_info {
1341 struct bpf_map *map;
1342 struct bpf_shtab *htab;
1346 static void *sock_hash_seq_find_next(struct sock_hash_seq_info *info,
1347 struct bpf_shtab_elem *prev_elem)
1349 const struct bpf_shtab *htab = info->htab;
1350 struct bpf_shtab_bucket *bucket;
1351 struct bpf_shtab_elem *elem;
1352 struct hlist_node *node;
1354 /* try to find next elem in the same bucket */
1356 node = rcu_dereference(hlist_next_rcu(&prev_elem->node));
1357 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1361 /* no more elements, continue in the next bucket */
1365 for (; info->bucket_id < htab->buckets_num; info->bucket_id++) {
1366 bucket = &htab->buckets[info->bucket_id];
1367 node = rcu_dereference(hlist_first_rcu(&bucket->head));
1368 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1376 static void *sock_hash_seq_start(struct seq_file *seq, loff_t *pos)
1379 struct sock_hash_seq_info *info = seq->private;
1384 /* pairs with sock_hash_seq_stop */
1386 return sock_hash_seq_find_next(info, NULL);
1389 static void *sock_hash_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1392 struct sock_hash_seq_info *info = seq->private;
1395 return sock_hash_seq_find_next(info, v);
1398 static int sock_hash_seq_show(struct seq_file *seq, void *v)
1401 struct sock_hash_seq_info *info = seq->private;
1402 struct bpf_iter__sockmap ctx = {};
1403 struct bpf_shtab_elem *elem = v;
1404 struct bpf_iter_meta meta;
1405 struct bpf_prog *prog;
1408 prog = bpf_iter_get_info(&meta, !elem);
1413 ctx.map = info->map;
1415 ctx.key = elem->key;
1419 return bpf_iter_run_prog(prog, &ctx);
1422 static void sock_hash_seq_stop(struct seq_file *seq, void *v)
1426 (void)sock_hash_seq_show(seq, NULL);
1428 /* pairs with sock_hash_seq_start */
1432 static const struct seq_operations sock_hash_seq_ops = {
1433 .start = sock_hash_seq_start,
1434 .next = sock_hash_seq_next,
1435 .stop = sock_hash_seq_stop,
1436 .show = sock_hash_seq_show,
1439 static int sock_hash_init_seq_private(void *priv_data,
1440 struct bpf_iter_aux_info *aux)
1442 struct sock_hash_seq_info *info = priv_data;
1444 bpf_map_inc_with_uref(aux->map);
1445 info->map = aux->map;
1446 info->htab = container_of(aux->map, struct bpf_shtab, map);
1450 static void sock_hash_fini_seq_private(void *priv_data)
1452 struct sock_hash_seq_info *info = priv_data;
1454 bpf_map_put_with_uref(info->map);
1457 static const struct bpf_iter_seq_info sock_hash_iter_seq_info = {
1458 .seq_ops = &sock_hash_seq_ops,
1459 .init_seq_private = sock_hash_init_seq_private,
1460 .fini_seq_private = sock_hash_fini_seq_private,
1461 .seq_priv_size = sizeof(struct sock_hash_seq_info),
1464 static int sock_hash_map_btf_id;
1465 const struct bpf_map_ops sock_hash_ops = {
1466 .map_meta_equal = bpf_map_meta_equal,
1467 .map_alloc = sock_hash_alloc,
1468 .map_free = sock_hash_free,
1469 .map_get_next_key = sock_hash_get_next_key,
1470 .map_update_elem = sock_map_update_elem,
1471 .map_delete_elem = sock_hash_delete_elem,
1472 .map_lookup_elem = sock_hash_lookup,
1473 .map_lookup_elem_sys_only = sock_hash_lookup_sys,
1474 .map_release_uref = sock_hash_release_progs,
1475 .map_check_btf = map_check_no_btf,
1476 .map_btf_name = "bpf_shtab",
1477 .map_btf_id = &sock_hash_map_btf_id,
1478 .iter_seq_info = &sock_hash_iter_seq_info,
1481 static struct sk_psock_progs *sock_map_progs(struct bpf_map *map)
1483 switch (map->map_type) {
1484 case BPF_MAP_TYPE_SOCKMAP:
1485 return &container_of(map, struct bpf_stab, map)->progs;
1486 case BPF_MAP_TYPE_SOCKHASH:
1487 return &container_of(map, struct bpf_shtab, map)->progs;
1495 int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog,
1496 struct bpf_prog *old, u32 which)
1498 struct sk_psock_progs *progs = sock_map_progs(map);
1499 struct bpf_prog **pprog;
1505 case BPF_SK_MSG_VERDICT:
1506 pprog = &progs->msg_parser;
1508 case BPF_SK_SKB_STREAM_PARSER:
1509 pprog = &progs->skb_parser;
1511 case BPF_SK_SKB_STREAM_VERDICT:
1512 pprog = &progs->skb_verdict;
1519 return psock_replace_prog(pprog, prog, old);
1521 psock_set_prog(pprog, prog);
1525 static void sock_map_unlink(struct sock *sk, struct sk_psock_link *link)
1527 switch (link->map->map_type) {
1528 case BPF_MAP_TYPE_SOCKMAP:
1529 return sock_map_delete_from_link(link->map, sk,
1531 case BPF_MAP_TYPE_SOCKHASH:
1532 return sock_hash_delete_from_link(link->map, sk,
1539 static void sock_map_remove_links(struct sock *sk, struct sk_psock *psock)
1541 struct sk_psock_link *link;
1543 while ((link = sk_psock_link_pop(psock))) {
1544 sock_map_unlink(sk, link);
1545 sk_psock_free_link(link);
1549 void sock_map_unhash(struct sock *sk)
1551 void (*saved_unhash)(struct sock *sk);
1552 struct sk_psock *psock;
1555 psock = sk_psock(sk);
1556 if (unlikely(!psock)) {
1558 if (sk->sk_prot->unhash)
1559 sk->sk_prot->unhash(sk);
1563 saved_unhash = psock->saved_unhash;
1564 sock_map_remove_links(sk, psock);
1569 void sock_map_close(struct sock *sk, long timeout)
1571 void (*saved_close)(struct sock *sk, long timeout);
1572 struct sk_psock *psock;
1576 psock = sk_psock(sk);
1577 if (unlikely(!psock)) {
1580 return sk->sk_prot->close(sk, timeout);
1583 saved_close = psock->saved_close;
1584 sock_map_remove_links(sk, psock);
1587 saved_close(sk, timeout);
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_RDONLY_BUF_OR_NULL },
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);