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 int (*rtx_syn_ack)(const struct sock *sk,
34 struct request_sock *req);
35 void (*send_ack)(const struct sock *sk, struct sk_buff *skb,
36 struct request_sock *req);
37 void (*send_reset)(const struct sock *sk,
38 struct sk_buff *skb,
39 enum sk_rst_reason reason);
40 void (*destructor)(struct request_sock *req);
41 void (*syn_ack_timeout)(const struct request_sock *req);
42 };
43
44 int inet_rtx_syn_ack(const struct sock *parent, struct request_sock *req);
45
46 struct saved_syn {
47 u32 mac_hdrlen;
48 u32 network_hdrlen;
49 u32 tcp_hdrlen;
50 u8 data[];
51 };
52
53 /* struct request_sock - mini sock to represent a connection request
54 */
55 struct request_sock {
56 struct sock_common __req_common;
57 #define rsk_refcnt __req_common.skc_refcnt
58 #define rsk_hash __req_common.skc_hash
59 #define rsk_listener __req_common.skc_listener
60 #define rsk_window_clamp __req_common.skc_window_clamp
61 #define rsk_rcv_wnd __req_common.skc_rcv_wnd
62
63 struct request_sock *dl_next;
64 u16 mss;
65 u8 num_retrans; /* number of retransmits */
66 u8 syncookie:1; /* True if
67 * 1) tcpopts needs to be encoded in
68 * TS of SYN+ACK
69 * 2) ACK is validated by BPF kfunc.
70 */
71 u8 num_timeout:7; /* number of timeouts */
72 u32 ts_recent;
73 struct timer_list rsk_timer;
74 const struct request_sock_ops *rsk_ops;
75 struct sock *sk;
76 struct saved_syn *saved_syn;
77 u32 secid;
78 u32 peer_secid;
79 u32 timeout;
80 };
81
inet_reqsk(const struct sock * sk)82 static inline struct request_sock *inet_reqsk(const struct sock *sk)
83 {
84 return (struct request_sock *)sk;
85 }
86
req_to_sk(struct request_sock * req)87 static inline struct sock *req_to_sk(struct request_sock *req)
88 {
89 return (struct sock *)req;
90 }
91
92 /**
93 * skb_steal_sock - steal a socket from an sk_buff
94 * @skb: sk_buff to steal the socket from
95 * @refcounted: is set to true if the socket is reference-counted
96 * @prefetched: is set to true if the socket was assigned from bpf
97 */
skb_steal_sock(struct sk_buff * skb,bool * refcounted,bool * prefetched)98 static inline struct sock *skb_steal_sock(struct sk_buff *skb,
99 bool *refcounted, bool *prefetched)
100 {
101 struct sock *sk = skb->sk;
102
103 if (!sk) {
104 *prefetched = false;
105 *refcounted = false;
106 return NULL;
107 }
108
109 *prefetched = skb_sk_is_prefetched(skb);
110 if (*prefetched) {
111 #if IS_ENABLED(CONFIG_SYN_COOKIES)
112 if (sk->sk_state == TCP_NEW_SYN_RECV && inet_reqsk(sk)->syncookie) {
113 struct request_sock *req = inet_reqsk(sk);
114
115 *refcounted = false;
116 sk = req->rsk_listener;
117 req->rsk_listener = NULL;
118 return sk;
119 }
120 #endif
121 *refcounted = sk_is_refcounted(sk);
122 } else {
123 *refcounted = true;
124 }
125
126 skb->destructor = NULL;
127 skb->sk = NULL;
128 return sk;
129 }
130
__reqsk_free(struct request_sock * req)131 static inline void __reqsk_free(struct request_sock *req)
132 {
133 req->rsk_ops->destructor(req);
134 if (req->rsk_listener)
135 sock_put(req->rsk_listener);
136 kfree(req->saved_syn);
137 kmem_cache_free(req->rsk_ops->slab, req);
138 }
139
reqsk_free(struct request_sock * req)140 static inline void reqsk_free(struct request_sock *req)
141 {
142 DEBUG_NET_WARN_ON_ONCE(refcount_read(&req->rsk_refcnt) != 0);
143 __reqsk_free(req);
144 }
145
reqsk_put(struct request_sock * req)146 static inline void reqsk_put(struct request_sock *req)
147 {
148 if (refcount_dec_and_test(&req->rsk_refcnt))
149 __reqsk_free(req);
150 }
151
152 /*
153 * For a TCP Fast Open listener -
154 * lock - protects the access to all the reqsk, which is co-owned by
155 * the listener and the child socket.
156 * qlen - pending TFO requests (still in TCP_SYN_RECV).
157 * max_qlen - max TFO reqs allowed before TFO is disabled.
158 *
159 * XXX (TFO) - ideally these fields can be made as part of "listen_sock"
160 * structure above. But there is some implementation difficulty due to
161 * listen_sock being part of request_sock_queue hence will be freed when
162 * a listener is stopped. But TFO related fields may continue to be
163 * accessed even after a listener is closed, until its sk_refcnt drops
164 * to 0 implying no more outstanding TFO reqs. One solution is to keep
165 * listen_opt around until sk_refcnt drops to 0. But there is some other
166 * complexity that needs to be resolved. E.g., a listener can be disabled
167 * temporarily through shutdown()->tcp_disconnect(), and re-enabled later.
168 */
169 struct fastopen_queue {
170 struct request_sock *rskq_rst_head; /* Keep track of past TFO */
171 struct request_sock *rskq_rst_tail; /* requests that caused RST.
172 * This is part of the defense
173 * against spoofing attack.
174 */
175 spinlock_t lock;
176 int qlen; /* # of pending (TCP_SYN_RECV) reqs */
177 int max_qlen; /* != 0 iff TFO is currently enabled */
178
179 struct tcp_fastopen_context __rcu *ctx; /* cipher context for cookie */
180 };
181
182 /** struct request_sock_queue - queue of request_socks
183 *
184 * @rskq_accept_head - FIFO head of established children
185 * @rskq_accept_tail - FIFO tail of established children
186 * @rskq_defer_accept - User waits for some data after accept()
187 *
188 */
189 struct request_sock_queue {
190 spinlock_t rskq_lock;
191 u8 rskq_defer_accept;
192
193 u32 synflood_warned;
194 atomic_t qlen;
195 atomic_t young;
196
197 struct request_sock *rskq_accept_head;
198 struct request_sock *rskq_accept_tail;
199 struct fastopen_queue fastopenq; /* Check max_qlen != 0 to determine
200 * if TFO is enabled.
201 */
202 };
203
204 void reqsk_queue_alloc(struct request_sock_queue *queue);
205
206 void reqsk_fastopen_remove(struct sock *sk, struct request_sock *req,
207 bool reset);
208
reqsk_queue_empty(const struct request_sock_queue * queue)209 static inline bool reqsk_queue_empty(const struct request_sock_queue *queue)
210 {
211 return READ_ONCE(queue->rskq_accept_head) == NULL;
212 }
213
reqsk_queue_remove(struct request_sock_queue * queue,struct sock * parent)214 static inline struct request_sock *reqsk_queue_remove(struct request_sock_queue *queue,
215 struct sock *parent)
216 {
217 struct request_sock *req;
218
219 spin_lock_bh(&queue->rskq_lock);
220 req = queue->rskq_accept_head;
221 if (req) {
222 sk_acceptq_removed(parent);
223 WRITE_ONCE(queue->rskq_accept_head, req->dl_next);
224 if (queue->rskq_accept_head == NULL)
225 queue->rskq_accept_tail = NULL;
226 }
227 spin_unlock_bh(&queue->rskq_lock);
228 return req;
229 }
230
reqsk_queue_removed(struct request_sock_queue * queue,const struct request_sock * req)231 static inline void reqsk_queue_removed(struct request_sock_queue *queue,
232 const struct request_sock *req)
233 {
234 if (req->num_timeout == 0)
235 atomic_dec(&queue->young);
236 atomic_dec(&queue->qlen);
237 }
238
reqsk_queue_added(struct request_sock_queue * queue)239 static inline void reqsk_queue_added(struct request_sock_queue *queue)
240 {
241 atomic_inc(&queue->young);
242 atomic_inc(&queue->qlen);
243 }
244
reqsk_queue_len(const struct request_sock_queue * queue)245 static inline int reqsk_queue_len(const struct request_sock_queue *queue)
246 {
247 return atomic_read(&queue->qlen);
248 }
249
reqsk_queue_len_young(const struct request_sock_queue * queue)250 static inline int reqsk_queue_len_young(const struct request_sock_queue *queue)
251 {
252 return atomic_read(&queue->young);
253 }
254
255 /* RFC 7323 2.3 Using the Window Scale Option
256 * The window field (SEG.WND) of every outgoing segment, with the
257 * exception of <SYN> segments, MUST be right-shifted by
258 * Rcv.Wind.Shift bits.
259 *
260 * This means the SEG.WND carried in SYNACK can not exceed 65535.
261 * We use this property to harden TCP stack while in NEW_SYN_RECV state.
262 */
tcp_synack_window(const struct request_sock * req)263 static inline u32 tcp_synack_window(const struct request_sock *req)
264 {
265 return min(req->rsk_rcv_wnd, 65535U);
266 }
267 #endif /* _REQUEST_SOCK_H */
268