request_sock.h

  1/* SPDX-License-Identifier: GPL-2.0-or-later */
  2/*
  3 * NET		Generic infrastructure for Network protocols.
  4 *
  5 *		Definitions for request_sock
  6 *
  7 * Authors:	Arnaldo Carvalho de Melo <acme@conectiva.com.br>
  8 *
  9 * 		From code originally in include/net/tcp.h
 10 */
 11#ifndef _REQUEST_SOCK_H
 12#define _REQUEST_SOCK_H
 13
 14#include <linux/slab.h>
 15#include <linux/spinlock.h>
 16#include <linux/types.h>
 17#include <linux/bug.h>
 18#include <linux/refcount.h>
 19
 20#include <net/sock.h>
 21#include <net/rstreason.h>
 22
 23struct request_sock;
 24struct sk_buff;
 25struct dst_entry;
 26struct proto;
 27
 28struct request_sock_ops {
 29	int		family;
 30	unsigned int	obj_size;
 31	struct kmem_cache	*slab;
 32	char		*slab_name;
 33	void		(*send_ack)(const struct sock *sk, struct sk_buff *skb,
 34				    struct request_sock *req);
 35	void		(*send_reset)(const struct sock *sk,
 36				      struct sk_buff *skb,
 37				      enum sk_rst_reason reason);
 38	void		(*destructor)(struct request_sock *req);
 39};
 40
 41struct saved_syn {
 42	u32 mac_hdrlen;
 43	u32 network_hdrlen;
 44	u32 tcp_hdrlen;
 45	u8 data[];
 46};
 47
 48/* struct request_sock - mini sock to represent a connection request
 49 */
 50struct request_sock {
 51	struct sock_common		__req_common;
 52#define rsk_refcnt			__req_common.skc_refcnt
 53#define rsk_hash			__req_common.skc_hash
 54#define rsk_listener			__req_common.skc_listener
 55#define rsk_window_clamp		__req_common.skc_window_clamp
 56#define rsk_rcv_wnd			__req_common.skc_rcv_wnd
 57
 58	struct request_sock		*dl_next;
 59	u16				mss;
 60	u8				num_retrans; /* number of retransmits */
 61	u8				syncookie:1; /* True if
 62						      * 1) tcpopts needs to be encoded in
 63						      *    TS of SYN+ACK
 64						      * 2) ACK is validated by BPF kfunc.
 65						      */
 66	u8				num_timeout:7; /* number of timeouts */
 67	u32				ts_recent;
 68	struct timer_list		rsk_timer;
 69	const struct request_sock_ops	*rsk_ops;
 70	struct sock			*sk;
 71	struct saved_syn		*saved_syn;
 72	u32				secid;
 73	u32				peer_secid;
 74	u32				timeout;
 75};
 76
 77static inline struct request_sock *inet_reqsk(const struct sock *sk)
 78{
 79	return (struct request_sock *)sk;
 80}
 81
 82static inline struct sock *req_to_sk(struct request_sock *req)
 83{
 84	return (struct sock *)req;
 85}
 86
 87/**
 88 * skb_steal_sock - steal a socket from an sk_buff
 89 * @skb: sk_buff to steal the socket from
 90 * @refcounted: is set to true if the socket is reference-counted
 91 * @prefetched: is set to true if the socket was assigned from bpf
 92 */
 93static inline struct sock *skb_steal_sock(struct sk_buff *skb,
 94					  bool *refcounted, bool *prefetched)
 95{
 96	struct sock *sk = skb->sk;
 97
 98	if (!sk) {
 99		*prefetched = false;
100		*refcounted = false;
101		return NULL;
102	}
103
104	*prefetched = skb_sk_is_prefetched(skb);
105	if (*prefetched) {
106#if IS_ENABLED(CONFIG_SYN_COOKIES)
107		if (sk->sk_state == TCP_NEW_SYN_RECV && inet_reqsk(sk)->syncookie) {
108			struct request_sock *req = inet_reqsk(sk);
109
110			*refcounted = false;
111			sk = req->rsk_listener;
112			req->rsk_listener = NULL;
113			return sk;
114		}
115#endif
116		*refcounted = sk_is_refcounted(sk);
117	} else {
118		*refcounted = true;
119	}
120
121	skb->destructor = NULL;
122	skb->sk = NULL;
123	return sk;
124}
125
126static inline void __reqsk_free(struct request_sock *req)
127{
128	req->rsk_ops->destructor(req);
129	if (req->rsk_listener)
130		sock_put(req->rsk_listener);
131	kfree(req->saved_syn);
132	kmem_cache_free(req->rsk_ops->slab, req);
133}
134
135static inline void reqsk_free(struct request_sock *req)
136{
137	DEBUG_NET_WARN_ON_ONCE(refcount_read(&req->rsk_refcnt) != 0);
138	__reqsk_free(req);
139}
140
141static inline void reqsk_put(struct request_sock *req)
142{
143	if (refcount_dec_and_test(&req->rsk_refcnt))
144		__reqsk_free(req);
145}
146
147/*
148 * For a TCP Fast Open listener -
149 *	lock - protects the access to all the reqsk, which is co-owned by
150 *		the listener and the child socket.
151 *	qlen - pending TFO requests (still in TCP_SYN_RECV).
152 *	max_qlen - max TFO reqs allowed before TFO is disabled.
153 *
154 *	XXX (TFO) - ideally these fields can be made as part of "listen_sock"
155 *	structure above. But there is some implementation difficulty due to
156 *	listen_sock being part of request_sock_queue hence will be freed when
157 *	a listener is stopped. But TFO related fields may continue to be
158 *	accessed even after a listener is closed, until its sk_refcnt drops
159 *	to 0 implying no more outstanding TFO reqs. One solution is to keep
160 *	listen_opt around until	sk_refcnt drops to 0. But there is some other
161 *	complexity that needs to be resolved. E.g., a listener can be disabled
162 *	temporarily through shutdown()->tcp_disconnect(), and re-enabled later.
163 */
164struct fastopen_queue {
165	struct request_sock	*rskq_rst_head; /* Keep track of past TFO */
166	struct request_sock	*rskq_rst_tail; /* requests that caused RST.
167						 * This is part of the defense
168						 * against spoofing attack.
169						 */
170	spinlock_t	lock;
171	int		qlen;		/* # of pending (TCP_SYN_RECV) reqs */
172	int		max_qlen;	/* != 0 iff TFO is currently enabled */
173
174	struct tcp_fastopen_context __rcu *ctx; /* cipher context for cookie */
175};
176
177/** struct request_sock_queue - queue of request_socks
178 *
179 * @rskq_accept_head - FIFO head of established children
180 * @rskq_accept_tail - FIFO tail of established children
181 * @rskq_defer_accept - User waits for some data after accept()
182 *
183 */
184struct request_sock_queue {
185	spinlock_t		rskq_lock;
186	u8			rskq_defer_accept;
187	u8			synflood_warned;
188
189	atomic_t		qlen;
190	atomic_t		young;
191
192	struct request_sock	*rskq_accept_head;
193	struct request_sock	*rskq_accept_tail;
194	struct fastopen_queue	fastopenq;  /* Check max_qlen != 0 to determine
195					     * if TFO is enabled.
196					     */
197};
198
199void reqsk_queue_alloc(struct request_sock_queue *queue);
200
201void reqsk_fastopen_remove(struct sock *sk, struct request_sock *req,
202			   bool reset);
203
204static inline bool reqsk_queue_empty(const struct request_sock_queue *queue)
205{
206	return READ_ONCE(queue->rskq_accept_head) == NULL;
207}
208
209static inline struct request_sock *reqsk_queue_remove(struct request_sock_queue *queue,
210						      struct sock *parent)
211{
212	struct request_sock *req;
213
214	spin_lock_bh(&queue->rskq_lock);
215	req = queue->rskq_accept_head;
216	if (req) {
217		sk_acceptq_removed(parent);
218		WRITE_ONCE(queue->rskq_accept_head, req->dl_next);
219		if (queue->rskq_accept_head == NULL)
220			queue->rskq_accept_tail = NULL;
221	}
222	spin_unlock_bh(&queue->rskq_lock);
223	return req;
224}
225
226static inline void reqsk_queue_removed(struct request_sock_queue *queue,
227				       const struct request_sock *req)
228{
229	if (req->num_timeout == 0)
230		atomic_dec(&queue->young);
231	atomic_dec(&queue->qlen);
232}
233
234static inline void reqsk_queue_added(struct request_sock_queue *queue)
235{
236	atomic_inc(&queue->young);
237	atomic_inc(&queue->qlen);
238}
239
240static inline int reqsk_queue_len(const struct request_sock_queue *queue)
241{
242	return atomic_read(&queue->qlen);
243}
244
245static inline int reqsk_queue_len_young(const struct request_sock_queue *queue)
246{
247	return atomic_read(&queue->young);
248}
249
250/* RFC 7323 2.3 Using the Window Scale Option
251 *  The window field (SEG.WND) of every outgoing segment, with the
252 *  exception of <SYN> segments, MUST be right-shifted by
253 *  Rcv.Wind.Shift bits.
254 *
255 * This means the SEG.WND carried in SYNACK can not exceed 65535.
256 * We use this property to harden TCP stack while in NEW_SYN_RECV state.
257 */
258static inline u32 tcp_synack_window(const struct request_sock *req)
259{
260	return min(req->rsk_rcv_wnd, 65535U);
261}
262#endif /* _REQUEST_SOCK_H */