Loading...
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 */
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; /* True if
65 * 1) tcpopts needs to be encoded in
66 * TS of SYN+ACK
67 * 2) ACK is validated by BPF kfunc.
68 */
69 u8 num_timeout:7; /* number of timeouts */
70 u32 ts_recent;
71 struct timer_list rsk_timer;
72 const struct request_sock_ops *rsk_ops;
73 struct sock *sk;
74 struct saved_syn *saved_syn;
75 u32 secid;
76 u32 peer_secid;
77 u32 timeout;
78};
79
80static inline struct request_sock *inet_reqsk(const struct sock *sk)
81{
82 return (struct request_sock *)sk;
83}
84
85static inline struct sock *req_to_sk(struct request_sock *req)
86{
87 return (struct sock *)req;
88}
89
90/**
91 * skb_steal_sock - steal a socket from an sk_buff
92 * @skb: sk_buff to steal the socket from
93 * @refcounted: is set to true if the socket is reference-counted
94 * @prefetched: is set to true if the socket was assigned from bpf
95 */
96static inline struct sock *skb_steal_sock(struct sk_buff *skb,
97 bool *refcounted, bool *prefetched)
98{
99 struct sock *sk = skb->sk;
100
101 if (!sk) {
102 *prefetched = false;
103 *refcounted = false;
104 return NULL;
105 }
106
107 *prefetched = skb_sk_is_prefetched(skb);
108 if (*prefetched) {
109#if IS_ENABLED(CONFIG_SYN_COOKIES)
110 if (sk->sk_state == TCP_NEW_SYN_RECV && inet_reqsk(sk)->syncookie) {
111 struct request_sock *req = inet_reqsk(sk);
112
113 *refcounted = false;
114 sk = req->rsk_listener;
115 req->rsk_listener = NULL;
116 return sk;
117 }
118#endif
119 *refcounted = sk_is_refcounted(sk);
120 } else {
121 *refcounted = true;
122 }
123
124 skb->destructor = NULL;
125 skb->sk = NULL;
126 return sk;
127}
128
129static inline struct request_sock *
130reqsk_alloc(const struct request_sock_ops *ops, struct sock *sk_listener,
131 bool attach_listener)
132{
133 struct request_sock *req;
134
135 req = kmem_cache_alloc(ops->slab, GFP_ATOMIC | __GFP_NOWARN);
136 if (!req)
137 return NULL;
138 req->rsk_listener = NULL;
139 if (attach_listener) {
140 if (unlikely(!refcount_inc_not_zero(&sk_listener->sk_refcnt))) {
141 kmem_cache_free(ops->slab, req);
142 return NULL;
143 }
144 req->rsk_listener = sk_listener;
145 }
146 req->rsk_ops = ops;
147 req_to_sk(req)->sk_prot = sk_listener->sk_prot;
148 sk_node_init(&req_to_sk(req)->sk_node);
149 sk_tx_queue_clear(req_to_sk(req));
150 req->saved_syn = NULL;
151 req->syncookie = 0;
152 req->timeout = 0;
153 req->num_timeout = 0;
154 req->num_retrans = 0;
155 req->sk = NULL;
156 refcount_set(&req->rsk_refcnt, 0);
157
158 return req;
159}
160
161static inline void __reqsk_free(struct request_sock *req)
162{
163 req->rsk_ops->destructor(req);
164 if (req->rsk_listener)
165 sock_put(req->rsk_listener);
166 kfree(req->saved_syn);
167 kmem_cache_free(req->rsk_ops->slab, req);
168}
169
170static inline void reqsk_free(struct request_sock *req)
171{
172 WARN_ON_ONCE(refcount_read(&req->rsk_refcnt) != 0);
173 __reqsk_free(req);
174}
175
176static inline void reqsk_put(struct request_sock *req)
177{
178 if (refcount_dec_and_test(&req->rsk_refcnt))
179 reqsk_free(req);
180}
181
182/*
183 * For a TCP Fast Open listener -
184 * lock - protects the access to all the reqsk, which is co-owned by
185 * the listener and the child socket.
186 * qlen - pending TFO requests (still in TCP_SYN_RECV).
187 * max_qlen - max TFO reqs allowed before TFO is disabled.
188 *
189 * XXX (TFO) - ideally these fields can be made as part of "listen_sock"
190 * structure above. But there is some implementation difficulty due to
191 * listen_sock being part of request_sock_queue hence will be freed when
192 * a listener is stopped. But TFO related fields may continue to be
193 * accessed even after a listener is closed, until its sk_refcnt drops
194 * to 0 implying no more outstanding TFO reqs. One solution is to keep
195 * listen_opt around until sk_refcnt drops to 0. But there is some other
196 * complexity that needs to be resolved. E.g., a listener can be disabled
197 * temporarily through shutdown()->tcp_disconnect(), and re-enabled later.
198 */
199struct fastopen_queue {
200 struct request_sock *rskq_rst_head; /* Keep track of past TFO */
201 struct request_sock *rskq_rst_tail; /* requests that caused RST.
202 * This is part of the defense
203 * against spoofing attack.
204 */
205 spinlock_t lock;
206 int qlen; /* # of pending (TCP_SYN_RECV) reqs */
207 int max_qlen; /* != 0 iff TFO is currently enabled */
208
209 struct tcp_fastopen_context __rcu *ctx; /* cipher context for cookie */
210};
211
212/** struct request_sock_queue - queue of request_socks
213 *
214 * @rskq_accept_head - FIFO head of established children
215 * @rskq_accept_tail - FIFO tail of established children
216 * @rskq_defer_accept - User waits for some data after accept()
217 *
218 */
219struct request_sock_queue {
220 spinlock_t rskq_lock;
221 u8 rskq_defer_accept;
222
223 u32 synflood_warned;
224 atomic_t qlen;
225 atomic_t young;
226
227 struct request_sock *rskq_accept_head;
228 struct request_sock *rskq_accept_tail;
229 struct fastopen_queue fastopenq; /* Check max_qlen != 0 to determine
230 * if TFO is enabled.
231 */
232};
233
234void reqsk_queue_alloc(struct request_sock_queue *queue);
235
236void reqsk_fastopen_remove(struct sock *sk, struct request_sock *req,
237 bool reset);
238
239static inline bool reqsk_queue_empty(const struct request_sock_queue *queue)
240{
241 return READ_ONCE(queue->rskq_accept_head) == NULL;
242}
243
244static inline struct request_sock *reqsk_queue_remove(struct request_sock_queue *queue,
245 struct sock *parent)
246{
247 struct request_sock *req;
248
249 spin_lock_bh(&queue->rskq_lock);
250 req = queue->rskq_accept_head;
251 if (req) {
252 sk_acceptq_removed(parent);
253 WRITE_ONCE(queue->rskq_accept_head, req->dl_next);
254 if (queue->rskq_accept_head == NULL)
255 queue->rskq_accept_tail = NULL;
256 }
257 spin_unlock_bh(&queue->rskq_lock);
258 return req;
259}
260
261static inline void reqsk_queue_removed(struct request_sock_queue *queue,
262 const struct request_sock *req)
263{
264 if (req->num_timeout == 0)
265 atomic_dec(&queue->young);
266 atomic_dec(&queue->qlen);
267}
268
269static inline void reqsk_queue_added(struct request_sock_queue *queue)
270{
271 atomic_inc(&queue->young);
272 atomic_inc(&queue->qlen);
273}
274
275static inline int reqsk_queue_len(const struct request_sock_queue *queue)
276{
277 return atomic_read(&queue->qlen);
278}
279
280static inline int reqsk_queue_len_young(const struct request_sock_queue *queue)
281{
282 return atomic_read(&queue->young);
283}
284
285#endif /* _REQUEST_SOCK_H */