Loading...
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 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
44int inet_rtx_syn_ack(const struct sock *parent, struct request_sock *req);
45
46struct 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 */
55struct 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
82static inline struct request_sock *inet_reqsk(const struct sock *sk)
83{
84 return (struct request_sock *)sk;
85}
86
87static 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 */
98static 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
131static 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
140static 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
146static 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 */
169struct 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 */
189struct 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
204void reqsk_queue_alloc(struct request_sock_queue *queue);
205
206void reqsk_fastopen_remove(struct sock *sk, struct request_sock *req,
207 bool reset);
208
209static inline bool reqsk_queue_empty(const struct request_sock_queue *queue)
210{
211 return READ_ONCE(queue->rskq_accept_head) == NULL;
212}
213
214static 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
231static 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
239static inline void reqsk_queue_added(struct request_sock_queue *queue)
240{
241 atomic_inc(&queue->young);
242 atomic_inc(&queue->qlen);
243}
244
245static inline int reqsk_queue_len(const struct request_sock_queue *queue)
246{
247 return atomic_read(&queue->qlen);
248}
249
250static 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 */
263static 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 */
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
22struct request_sock;
23struct sk_buff;
24struct dst_entry;
25struct proto;
26
27struct request_sock_ops {
28 int family;
29 unsigned int obj_size;
30 struct kmem_cache *slab;
31 char *slab_name;
32 int (*rtx_syn_ack)(const struct sock *sk,
33 struct request_sock *req);
34 void (*send_ack)(const struct sock *sk, struct sk_buff *skb,
35 struct request_sock *req);
36 void (*send_reset)(const struct sock *sk,
37 struct sk_buff *skb);
38 void (*destructor)(struct request_sock *req);
39 void (*syn_ack_timeout)(const struct request_sock *req);
40};
41
42int inet_rtx_syn_ack(const struct sock *parent, struct request_sock *req);
43
44struct saved_syn {
45 u32 mac_hdrlen;
46 u32 network_hdrlen;
47 u32 tcp_hdrlen;
48 u8 data[];
49};
50
51/* struct request_sock - mini sock to represent a connection request
52 */
53struct request_sock {
54 struct sock_common __req_common;
55#define rsk_refcnt __req_common.skc_refcnt
56#define rsk_hash __req_common.skc_hash
57#define rsk_listener __req_common.skc_listener
58#define rsk_window_clamp __req_common.skc_window_clamp
59#define rsk_rcv_wnd __req_common.skc_rcv_wnd
60
61 struct request_sock *dl_next;
62 u16 mss;
63 u8 num_retrans; /* number of retransmits */
64 u8 syncookie:1; /* syncookie: encode tcpopts in timestamp */
65 u8 num_timeout:7; /* number of timeouts */
66 u32 ts_recent;
67 struct timer_list rsk_timer;
68 const struct request_sock_ops *rsk_ops;
69 struct sock *sk;
70 struct saved_syn *saved_syn;
71 u32 secid;
72 u32 peer_secid;
73};
74
75static inline struct request_sock *inet_reqsk(const struct sock *sk)
76{
77 return (struct request_sock *)sk;
78}
79
80static inline struct sock *req_to_sk(struct request_sock *req)
81{
82 return (struct sock *)req;
83}
84
85static inline struct request_sock *
86reqsk_alloc(const struct request_sock_ops *ops, struct sock *sk_listener,
87 bool attach_listener)
88{
89 struct request_sock *req;
90
91 req = kmem_cache_alloc(ops->slab, GFP_ATOMIC | __GFP_NOWARN);
92 if (!req)
93 return NULL;
94 req->rsk_listener = NULL;
95 if (attach_listener) {
96 if (unlikely(!refcount_inc_not_zero(&sk_listener->sk_refcnt))) {
97 kmem_cache_free(ops->slab, req);
98 return NULL;
99 }
100 req->rsk_listener = sk_listener;
101 }
102 req->rsk_ops = ops;
103 req_to_sk(req)->sk_prot = sk_listener->sk_prot;
104 sk_node_init(&req_to_sk(req)->sk_node);
105 sk_tx_queue_clear(req_to_sk(req));
106 req->saved_syn = NULL;
107 req->num_timeout = 0;
108 req->num_retrans = 0;
109 req->sk = NULL;
110 refcount_set(&req->rsk_refcnt, 0);
111
112 return req;
113}
114
115static inline void __reqsk_free(struct request_sock *req)
116{
117 req->rsk_ops->destructor(req);
118 if (req->rsk_listener)
119 sock_put(req->rsk_listener);
120 kfree(req->saved_syn);
121 kmem_cache_free(req->rsk_ops->slab, req);
122}
123
124static inline void reqsk_free(struct request_sock *req)
125{
126 WARN_ON_ONCE(refcount_read(&req->rsk_refcnt) != 0);
127 __reqsk_free(req);
128}
129
130static inline void reqsk_put(struct request_sock *req)
131{
132 if (refcount_dec_and_test(&req->rsk_refcnt))
133 reqsk_free(req);
134}
135
136/*
137 * For a TCP Fast Open listener -
138 * lock - protects the access to all the reqsk, which is co-owned by
139 * the listener and the child socket.
140 * qlen - pending TFO requests (still in TCP_SYN_RECV).
141 * max_qlen - max TFO reqs allowed before TFO is disabled.
142 *
143 * XXX (TFO) - ideally these fields can be made as part of "listen_sock"
144 * structure above. But there is some implementation difficulty due to
145 * listen_sock being part of request_sock_queue hence will be freed when
146 * a listener is stopped. But TFO related fields may continue to be
147 * accessed even after a listener is closed, until its sk_refcnt drops
148 * to 0 implying no more outstanding TFO reqs. One solution is to keep
149 * listen_opt around until sk_refcnt drops to 0. But there is some other
150 * complexity that needs to be resolved. E.g., a listener can be disabled
151 * temporarily through shutdown()->tcp_disconnect(), and re-enabled later.
152 */
153struct fastopen_queue {
154 struct request_sock *rskq_rst_head; /* Keep track of past TFO */
155 struct request_sock *rskq_rst_tail; /* requests that caused RST.
156 * This is part of the defense
157 * against spoofing attack.
158 */
159 spinlock_t lock;
160 int qlen; /* # of pending (TCP_SYN_RECV) reqs */
161 int max_qlen; /* != 0 iff TFO is currently enabled */
162
163 struct tcp_fastopen_context __rcu *ctx; /* cipher context for cookie */
164};
165
166/** struct request_sock_queue - queue of request_socks
167 *
168 * @rskq_accept_head - FIFO head of established children
169 * @rskq_accept_tail - FIFO tail of established children
170 * @rskq_defer_accept - User waits for some data after accept()
171 *
172 */
173struct request_sock_queue {
174 spinlock_t rskq_lock;
175 u8 rskq_defer_accept;
176
177 u32 synflood_warned;
178 atomic_t qlen;
179 atomic_t young;
180
181 struct request_sock *rskq_accept_head;
182 struct request_sock *rskq_accept_tail;
183 struct fastopen_queue fastopenq; /* Check max_qlen != 0 to determine
184 * if TFO is enabled.
185 */
186};
187
188void reqsk_queue_alloc(struct request_sock_queue *queue);
189
190void reqsk_fastopen_remove(struct sock *sk, struct request_sock *req,
191 bool reset);
192
193static inline bool reqsk_queue_empty(const struct request_sock_queue *queue)
194{
195 return READ_ONCE(queue->rskq_accept_head) == NULL;
196}
197
198static inline struct request_sock *reqsk_queue_remove(struct request_sock_queue *queue,
199 struct sock *parent)
200{
201 struct request_sock *req;
202
203 spin_lock_bh(&queue->rskq_lock);
204 req = queue->rskq_accept_head;
205 if (req) {
206 sk_acceptq_removed(parent);
207 WRITE_ONCE(queue->rskq_accept_head, req->dl_next);
208 if (queue->rskq_accept_head == NULL)
209 queue->rskq_accept_tail = NULL;
210 }
211 spin_unlock_bh(&queue->rskq_lock);
212 return req;
213}
214
215static inline void reqsk_queue_removed(struct request_sock_queue *queue,
216 const struct request_sock *req)
217{
218 if (req->num_timeout == 0)
219 atomic_dec(&queue->young);
220 atomic_dec(&queue->qlen);
221}
222
223static inline void reqsk_queue_added(struct request_sock_queue *queue)
224{
225 atomic_inc(&queue->young);
226 atomic_inc(&queue->qlen);
227}
228
229static inline int reqsk_queue_len(const struct request_sock_queue *queue)
230{
231 return atomic_read(&queue->qlen);
232}
233
234static inline int reqsk_queue_len_young(const struct request_sock_queue *queue)
235{
236 return atomic_read(&queue->young);
237}
238
239#endif /* _REQUEST_SOCK_H */