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
23 struct request_sock;
24 struct sk_buff;
25 struct dst_entry;
26 struct proto;
27
28 struct 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 void (*syn_ack_timeout)(const struct request_sock *req);
40 };
41
42 struct saved_syn {
43 u32 mac_hdrlen;
44 u32 network_hdrlen;
45 u32 tcp_hdrlen;
46 u8 data[];
47 };
48
49 /* struct request_sock - mini sock to represent a connection request
50 */
51 struct request_sock {
52 struct sock_common __req_common;
53 #define rsk_refcnt __req_common.skc_refcnt
54 #define rsk_hash __req_common.skc_hash
55 #define rsk_listener __req_common.skc_listener
56 #define rsk_window_clamp __req_common.skc_window_clamp
57 #define rsk_rcv_wnd __req_common.skc_rcv_wnd
58
59 struct request_sock *dl_next;
60 u16 mss;
61 u8 num_retrans; /* number of retransmits */
62 u8 syncookie:1; /* True if
63 * 1) tcpopts needs to be encoded in
64 * TS of SYN+ACK
65 * 2) ACK is validated by BPF kfunc.
66 */
67 u8 num_timeout:7; /* number of timeouts */
68 u32 ts_recent;
69 struct timer_list rsk_timer;
70 const struct request_sock_ops *rsk_ops;
71 struct sock *sk;
72 struct saved_syn *saved_syn;
73 u32 secid;
74 u32 peer_secid;
75 u32 timeout;
76 };
77
inet_reqsk(const struct sock * sk)78 static inline struct request_sock *inet_reqsk(const struct sock *sk)
79 {
80 return (struct request_sock *)sk;
81 }
82
req_to_sk(struct request_sock * req)83 static inline struct sock *req_to_sk(struct request_sock *req)
84 {
85 return (struct sock *)req;
86 }
87
88 /**
89 * skb_steal_sock - steal a socket from an sk_buff
90 * @skb: sk_buff to steal the socket from
91 * @refcounted: is set to true if the socket is reference-counted
92 * @prefetched: is set to true if the socket was assigned from bpf
93 */
skb_steal_sock(struct sk_buff * skb,bool * refcounted,bool * prefetched)94 static inline struct sock *skb_steal_sock(struct sk_buff *skb,
95 bool *refcounted, bool *prefetched)
96 {
97 struct sock *sk = skb->sk;
98
99 if (!sk) {
100 *prefetched = false;
101 *refcounted = false;
102 return NULL;
103 }
104
105 *prefetched = skb_sk_is_prefetched(skb);
106 if (*prefetched) {
107 #if IS_ENABLED(CONFIG_SYN_COOKIES)
108 if (sk->sk_state == TCP_NEW_SYN_RECV && inet_reqsk(sk)->syncookie) {
109 struct request_sock *req = inet_reqsk(sk);
110
111 *refcounted = false;
112 sk = req->rsk_listener;
113 req->rsk_listener = NULL;
114 return sk;
115 }
116 #endif
117 *refcounted = sk_is_refcounted(sk);
118 } else {
119 *refcounted = true;
120 }
121
122 skb->destructor = NULL;
123 skb->sk = NULL;
124 return sk;
125 }
126
__reqsk_free(struct request_sock * req)127 static inline void __reqsk_free(struct request_sock *req)
128 {
129 req->rsk_ops->destructor(req);
130 if (req->rsk_listener)
131 sock_put(req->rsk_listener);
132 kfree(req->saved_syn);
133 kmem_cache_free(req->rsk_ops->slab, req);
134 }
135
reqsk_free(struct request_sock * req)136 static inline void reqsk_free(struct request_sock *req)
137 {
138 DEBUG_NET_WARN_ON_ONCE(refcount_read(&req->rsk_refcnt) != 0);
139 __reqsk_free(req);
140 }
141
reqsk_put(struct request_sock * req)142 static inline void reqsk_put(struct request_sock *req)
143 {
144 if (refcount_dec_and_test(&req->rsk_refcnt))
145 __reqsk_free(req);
146 }
147
148 /*
149 * For a TCP Fast Open listener -
150 * lock - protects the access to all the reqsk, which is co-owned by
151 * the listener and the child socket.
152 * qlen - pending TFO requests (still in TCP_SYN_RECV).
153 * max_qlen - max TFO reqs allowed before TFO is disabled.
154 *
155 * XXX (TFO) - ideally these fields can be made as part of "listen_sock"
156 * structure above. But there is some implementation difficulty due to
157 * listen_sock being part of request_sock_queue hence will be freed when
158 * a listener is stopped. But TFO related fields may continue to be
159 * accessed even after a listener is closed, until its sk_refcnt drops
160 * to 0 implying no more outstanding TFO reqs. One solution is to keep
161 * listen_opt around until sk_refcnt drops to 0. But there is some other
162 * complexity that needs to be resolved. E.g., a listener can be disabled
163 * temporarily through shutdown()->tcp_disconnect(), and re-enabled later.
164 */
165 struct fastopen_queue {
166 struct request_sock *rskq_rst_head; /* Keep track of past TFO */
167 struct request_sock *rskq_rst_tail; /* requests that caused RST.
168 * This is part of the defense
169 * against spoofing attack.
170 */
171 spinlock_t lock;
172 int qlen; /* # of pending (TCP_SYN_RECV) reqs */
173 int max_qlen; /* != 0 iff TFO is currently enabled */
174
175 struct tcp_fastopen_context __rcu *ctx; /* cipher context for cookie */
176 };
177
178 /** struct request_sock_queue - queue of request_socks
179 *
180 * @rskq_accept_head - FIFO head of established children
181 * @rskq_accept_tail - FIFO tail of established children
182 * @rskq_defer_accept - User waits for some data after accept()
183 *
184 */
185 struct request_sock_queue {
186 spinlock_t rskq_lock;
187 u8 rskq_defer_accept;
188
189 u32 synflood_warned;
190 atomic_t qlen;
191 atomic_t young;
192
193 struct request_sock *rskq_accept_head;
194 struct request_sock *rskq_accept_tail;
195 struct fastopen_queue fastopenq; /* Check max_qlen != 0 to determine
196 * if TFO is enabled.
197 */
198 };
199
200 void reqsk_queue_alloc(struct request_sock_queue *queue);
201
202 void reqsk_fastopen_remove(struct sock *sk, struct request_sock *req,
203 bool reset);
204
reqsk_queue_empty(const struct request_sock_queue * queue)205 static inline bool reqsk_queue_empty(const struct request_sock_queue *queue)
206 {
207 return READ_ONCE(queue->rskq_accept_head) == NULL;
208 }
209
reqsk_queue_remove(struct request_sock_queue * queue,struct sock * parent)210 static inline struct request_sock *reqsk_queue_remove(struct request_sock_queue *queue,
211 struct sock *parent)
212 {
213 struct request_sock *req;
214
215 spin_lock_bh(&queue->rskq_lock);
216 req = queue->rskq_accept_head;
217 if (req) {
218 sk_acceptq_removed(parent);
219 WRITE_ONCE(queue->rskq_accept_head, req->dl_next);
220 if (queue->rskq_accept_head == NULL)
221 queue->rskq_accept_tail = NULL;
222 }
223 spin_unlock_bh(&queue->rskq_lock);
224 return req;
225 }
226
reqsk_queue_removed(struct request_sock_queue * queue,const struct request_sock * req)227 static inline void reqsk_queue_removed(struct request_sock_queue *queue,
228 const struct request_sock *req)
229 {
230 if (req->num_timeout == 0)
231 atomic_dec(&queue->young);
232 atomic_dec(&queue->qlen);
233 }
234
reqsk_queue_added(struct request_sock_queue * queue)235 static inline void reqsk_queue_added(struct request_sock_queue *queue)
236 {
237 atomic_inc(&queue->young);
238 atomic_inc(&queue->qlen);
239 }
240
reqsk_queue_len(const struct request_sock_queue * queue)241 static inline int reqsk_queue_len(const struct request_sock_queue *queue)
242 {
243 return atomic_read(&queue->qlen);
244 }
245
reqsk_queue_len_young(const struct request_sock_queue * queue)246 static inline int reqsk_queue_len_young(const struct request_sock_queue *queue)
247 {
248 return atomic_read(&queue->young);
249 }
250
251 /* RFC 7323 2.3 Using the Window Scale Option
252 * The window field (SEG.WND) of every outgoing segment, with the
253 * exception of <SYN> segments, MUST be right-shifted by
254 * Rcv.Wind.Shift bits.
255 *
256 * This means the SEG.WND carried in SYNACK can not exceed 65535.
257 * We use this property to harden TCP stack while in NEW_SYN_RECV state.
258 */
tcp_synack_window(const struct request_sock * req)259 static inline u32 tcp_synack_window(const struct request_sock *req)
260 {
261 return min(req->rsk_rcv_wnd, 65535U);
262 }
263 #endif /* _REQUEST_SOCK_H */
264