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/*
2 * NET Generic infrastructure for Network protocols.
3 *
4 * Definitions for request_sock
5 *
6 * Authors: Arnaldo Carvalho de Melo <acme@conectiva.com.br>
7 *
8 * From code originally in include/net/tcp.h
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or (at your option) any later version.
14 */
15#ifndef _REQUEST_SOCK_H
16#define _REQUEST_SOCK_H
17
18#include <linux/slab.h>
19#include <linux/spinlock.h>
20#include <linux/types.h>
21#include <linux/bug.h>
22
23#include <net/sock.h>
24
25struct request_sock;
26struct sk_buff;
27struct dst_entry;
28struct proto;
29
30struct request_sock_ops {
31 int family;
32 int obj_size;
33 struct kmem_cache *slab;
34 char *slab_name;
35 int (*rtx_syn_ack)(struct sock *sk,
36 struct request_sock *req);
37 void (*send_ack)(struct sock *sk, struct sk_buff *skb,
38 struct request_sock *req);
39 void (*send_reset)(struct sock *sk,
40 struct sk_buff *skb);
41 void (*destructor)(struct request_sock *req);
42 void (*syn_ack_timeout)(struct sock *sk,
43 struct request_sock *req);
44};
45
46int inet_rtx_syn_ack(struct sock *parent, struct request_sock *req);
47
48/* struct request_sock - mini sock to represent a connection request
49 */
50struct request_sock {
51 struct sock_common __req_common;
52 struct request_sock *dl_next;
53 u16 mss;
54 u8 num_retrans; /* number of retransmits */
55 u8 cookie_ts:1; /* syncookie: encode tcpopts in timestamp */
56 u8 num_timeout:7; /* number of timeouts */
57 /* The following two fields can be easily recomputed I think -AK */
58 u32 window_clamp; /* window clamp at creation time */
59 u32 rcv_wnd; /* rcv_wnd offered first time */
60 u32 ts_recent;
61 unsigned long expires;
62 const struct request_sock_ops *rsk_ops;
63 struct sock *sk;
64 u32 secid;
65 u32 peer_secid;
66};
67
68static inline struct request_sock *reqsk_alloc(const struct request_sock_ops *ops)
69{
70 struct request_sock *req = kmem_cache_alloc(ops->slab, GFP_ATOMIC);
71
72 if (req != NULL)
73 req->rsk_ops = ops;
74
75 return req;
76}
77
78static inline void __reqsk_free(struct request_sock *req)
79{
80 kmem_cache_free(req->rsk_ops->slab, req);
81}
82
83static inline void reqsk_free(struct request_sock *req)
84{
85 req->rsk_ops->destructor(req);
86 __reqsk_free(req);
87}
88
89extern int sysctl_max_syn_backlog;
90
91/** struct listen_sock - listen state
92 *
93 * @max_qlen_log - log_2 of maximal queued SYNs/REQUESTs
94 */
95struct listen_sock {
96 u8 max_qlen_log;
97 u8 synflood_warned;
98 /* 2 bytes hole, try to use */
99 int qlen;
100 int qlen_young;
101 int clock_hand;
102 u32 hash_rnd;
103 u32 nr_table_entries;
104 struct request_sock *syn_table[0];
105};
106
107/*
108 * For a TCP Fast Open listener -
109 * lock - protects the access to all the reqsk, which is co-owned by
110 * the listener and the child socket.
111 * qlen - pending TFO requests (still in TCP_SYN_RECV).
112 * max_qlen - max TFO reqs allowed before TFO is disabled.
113 *
114 * XXX (TFO) - ideally these fields can be made as part of "listen_sock"
115 * structure above. But there is some implementation difficulty due to
116 * listen_sock being part of request_sock_queue hence will be freed when
117 * a listener is stopped. But TFO related fields may continue to be
118 * accessed even after a listener is closed, until its sk_refcnt drops
119 * to 0 implying no more outstanding TFO reqs. One solution is to keep
120 * listen_opt around until sk_refcnt drops to 0. But there is some other
121 * complexity that needs to be resolved. E.g., a listener can be disabled
122 * temporarily through shutdown()->tcp_disconnect(), and re-enabled later.
123 */
124struct fastopen_queue {
125 struct request_sock *rskq_rst_head; /* Keep track of past TFO */
126 struct request_sock *rskq_rst_tail; /* requests that caused RST.
127 * This is part of the defense
128 * against spoofing attack.
129 */
130 spinlock_t lock;
131 int qlen; /* # of pending (TCP_SYN_RECV) reqs */
132 int max_qlen; /* != 0 iff TFO is currently enabled */
133};
134
135/** struct request_sock_queue - queue of request_socks
136 *
137 * @rskq_accept_head - FIFO head of established children
138 * @rskq_accept_tail - FIFO tail of established children
139 * @rskq_defer_accept - User waits for some data after accept()
140 * @syn_wait_lock - serializer
141 *
142 * %syn_wait_lock is necessary only to avoid proc interface having to grab the main
143 * lock sock while browsing the listening hash (otherwise it's deadlock prone).
144 *
145 * This lock is acquired in read mode only from listening_get_next() seq_file
146 * op and it's acquired in write mode _only_ from code that is actively
147 * changing rskq_accept_head. All readers that are holding the master sock lock
148 * don't need to grab this lock in read mode too as rskq_accept_head. writes
149 * are always protected from the main sock lock.
150 */
151struct request_sock_queue {
152 struct request_sock *rskq_accept_head;
153 struct request_sock *rskq_accept_tail;
154 rwlock_t syn_wait_lock;
155 u8 rskq_defer_accept;
156 /* 3 bytes hole, try to pack */
157 struct listen_sock *listen_opt;
158 struct fastopen_queue *fastopenq; /* This is non-NULL iff TFO has been
159 * enabled on this listener. Check
160 * max_qlen != 0 in fastopen_queue
161 * to determine if TFO is enabled
162 * right at this moment.
163 */
164};
165
166int reqsk_queue_alloc(struct request_sock_queue *queue,
167 unsigned int nr_table_entries);
168
169void __reqsk_queue_destroy(struct request_sock_queue *queue);
170void reqsk_queue_destroy(struct request_sock_queue *queue);
171void reqsk_fastopen_remove(struct sock *sk, struct request_sock *req,
172 bool reset);
173
174static inline struct request_sock *
175 reqsk_queue_yank_acceptq(struct request_sock_queue *queue)
176{
177 struct request_sock *req = queue->rskq_accept_head;
178
179 queue->rskq_accept_head = NULL;
180 return req;
181}
182
183static inline int reqsk_queue_empty(struct request_sock_queue *queue)
184{
185 return queue->rskq_accept_head == NULL;
186}
187
188static inline void reqsk_queue_unlink(struct request_sock_queue *queue,
189 struct request_sock *req,
190 struct request_sock **prev_req)
191{
192 write_lock(&queue->syn_wait_lock);
193 *prev_req = req->dl_next;
194 write_unlock(&queue->syn_wait_lock);
195}
196
197static inline void reqsk_queue_add(struct request_sock_queue *queue,
198 struct request_sock *req,
199 struct sock *parent,
200 struct sock *child)
201{
202 req->sk = child;
203 sk_acceptq_added(parent);
204
205 if (queue->rskq_accept_head == NULL)
206 queue->rskq_accept_head = req;
207 else
208 queue->rskq_accept_tail->dl_next = req;
209
210 queue->rskq_accept_tail = req;
211 req->dl_next = NULL;
212}
213
214static inline struct request_sock *reqsk_queue_remove(struct request_sock_queue *queue)
215{
216 struct request_sock *req = queue->rskq_accept_head;
217
218 WARN_ON(req == NULL);
219
220 queue->rskq_accept_head = req->dl_next;
221 if (queue->rskq_accept_head == NULL)
222 queue->rskq_accept_tail = NULL;
223
224 return req;
225}
226
227static inline int reqsk_queue_removed(struct request_sock_queue *queue,
228 struct request_sock *req)
229{
230 struct listen_sock *lopt = queue->listen_opt;
231
232 if (req->num_timeout == 0)
233 --lopt->qlen_young;
234
235 return --lopt->qlen;
236}
237
238static inline int reqsk_queue_added(struct request_sock_queue *queue)
239{
240 struct listen_sock *lopt = queue->listen_opt;
241 const int prev_qlen = lopt->qlen;
242
243 lopt->qlen_young++;
244 lopt->qlen++;
245 return prev_qlen;
246}
247
248static inline int reqsk_queue_len(const struct request_sock_queue *queue)
249{
250 return queue->listen_opt != NULL ? queue->listen_opt->qlen : 0;
251}
252
253static inline int reqsk_queue_len_young(const struct request_sock_queue *queue)
254{
255 return queue->listen_opt->qlen_young;
256}
257
258static inline int reqsk_queue_is_full(const struct request_sock_queue *queue)
259{
260 return queue->listen_opt->qlen >> queue->listen_opt->max_qlen_log;
261}
262
263static inline void reqsk_queue_hash_req(struct request_sock_queue *queue,
264 u32 hash, struct request_sock *req,
265 unsigned long timeout)
266{
267 struct listen_sock *lopt = queue->listen_opt;
268
269 req->expires = jiffies + timeout;
270 req->num_retrans = 0;
271 req->num_timeout = 0;
272 req->sk = NULL;
273 req->dl_next = lopt->syn_table[hash];
274
275 write_lock(&queue->syn_wait_lock);
276 lopt->syn_table[hash] = req;
277 write_unlock(&queue->syn_wait_lock);
278}
279
280#endif /* _REQUEST_SOCK_H */