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 82 static inline struct request_sock *inet_reqsk(const struct sock *sk) 83 { 84 return (struct request_sock *)sk; 85 } 86 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 */ 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 } 130 131 static inline struct request_sock * 132 reqsk_alloc_noprof(const struct request_sock_ops *ops, struct sock *sk_listener, 133 bool attach_listener) 134 { 135 struct request_sock *req; 136 137 req = kmem_cache_alloc_noprof(ops->slab, GFP_ATOMIC | __GFP_NOWARN); 138 if (!req) 139 return NULL; 140 req->rsk_listener = NULL; 141 if (attach_listener) { 142 if (unlikely(!refcount_inc_not_zero(&sk_listener->sk_refcnt))) { 143 kmem_cache_free(ops->slab, req); 144 return NULL; 145 } 146 req->rsk_listener = sk_listener; 147 } 148 req->rsk_ops = ops; 149 req_to_sk(req)->sk_prot = sk_listener->sk_prot; 150 sk_node_init(&req_to_sk(req)->sk_node); 151 sk_tx_queue_clear(req_to_sk(req)); 152 req->saved_syn = NULL; 153 req->syncookie = 0; 154 req->timeout = 0; 155 req->num_timeout = 0; 156 req->num_retrans = 0; 157 req->sk = NULL; 158 refcount_set(&req->rsk_refcnt, 0); 159 160 return req; 161 } 162 #define reqsk_alloc(...) alloc_hooks(reqsk_alloc_noprof(__VA_ARGS__)) 163 164 static inline void __reqsk_free(struct request_sock *req) 165 { 166 req->rsk_ops->destructor(req); 167 if (req->rsk_listener) 168 sock_put(req->rsk_listener); 169 kfree(req->saved_syn); 170 kmem_cache_free(req->rsk_ops->slab, req); 171 } 172 173 static inline void reqsk_free(struct request_sock *req) 174 { 175 WARN_ON_ONCE(refcount_read(&req->rsk_refcnt) != 0); 176 __reqsk_free(req); 177 } 178 179 static inline void reqsk_put(struct request_sock *req) 180 { 181 if (refcount_dec_and_test(&req->rsk_refcnt)) 182 reqsk_free(req); 183 } 184 185 /* 186 * For a TCP Fast Open listener - 187 * lock - protects the access to all the reqsk, which is co-owned by 188 * the listener and the child socket. 189 * qlen - pending TFO requests (still in TCP_SYN_RECV). 190 * max_qlen - max TFO reqs allowed before TFO is disabled. 191 * 192 * XXX (TFO) - ideally these fields can be made as part of "listen_sock" 193 * structure above. But there is some implementation difficulty due to 194 * listen_sock being part of request_sock_queue hence will be freed when 195 * a listener is stopped. But TFO related fields may continue to be 196 * accessed even after a listener is closed, until its sk_refcnt drops 197 * to 0 implying no more outstanding TFO reqs. One solution is to keep 198 * listen_opt around until sk_refcnt drops to 0. But there is some other 199 * complexity that needs to be resolved. E.g., a listener can be disabled 200 * temporarily through shutdown()->tcp_disconnect(), and re-enabled later. 201 */ 202 struct fastopen_queue { 203 struct request_sock *rskq_rst_head; /* Keep track of past TFO */ 204 struct request_sock *rskq_rst_tail; /* requests that caused RST. 205 * This is part of the defense 206 * against spoofing attack. 207 */ 208 spinlock_t lock; 209 int qlen; /* # of pending (TCP_SYN_RECV) reqs */ 210 int max_qlen; /* != 0 iff TFO is currently enabled */ 211 212 struct tcp_fastopen_context __rcu *ctx; /* cipher context for cookie */ 213 }; 214 215 /** struct request_sock_queue - queue of request_socks 216 * 217 * @rskq_accept_head - FIFO head of established children 218 * @rskq_accept_tail - FIFO tail of established children 219 * @rskq_defer_accept - User waits for some data after accept() 220 * 221 */ 222 struct request_sock_queue { 223 spinlock_t rskq_lock; 224 u8 rskq_defer_accept; 225 226 u32 synflood_warned; 227 atomic_t qlen; 228 atomic_t young; 229 230 struct request_sock *rskq_accept_head; 231 struct request_sock *rskq_accept_tail; 232 struct fastopen_queue fastopenq; /* Check max_qlen != 0 to determine 233 * if TFO is enabled. 234 */ 235 }; 236 237 void reqsk_queue_alloc(struct request_sock_queue *queue); 238 239 void reqsk_fastopen_remove(struct sock *sk, struct request_sock *req, 240 bool reset); 241 242 static inline bool reqsk_queue_empty(const struct request_sock_queue *queue) 243 { 244 return READ_ONCE(queue->rskq_accept_head) == NULL; 245 } 246 247 static inline struct request_sock *reqsk_queue_remove(struct request_sock_queue *queue, 248 struct sock *parent) 249 { 250 struct request_sock *req; 251 252 spin_lock_bh(&queue->rskq_lock); 253 req = queue->rskq_accept_head; 254 if (req) { 255 sk_acceptq_removed(parent); 256 WRITE_ONCE(queue->rskq_accept_head, req->dl_next); 257 if (queue->rskq_accept_head == NULL) 258 queue->rskq_accept_tail = NULL; 259 } 260 spin_unlock_bh(&queue->rskq_lock); 261 return req; 262 } 263 264 static inline void reqsk_queue_removed(struct request_sock_queue *queue, 265 const struct request_sock *req) 266 { 267 if (req->num_timeout == 0) 268 atomic_dec(&queue->young); 269 atomic_dec(&queue->qlen); 270 } 271 272 static inline void reqsk_queue_added(struct request_sock_queue *queue) 273 { 274 atomic_inc(&queue->young); 275 atomic_inc(&queue->qlen); 276 } 277 278 static inline int reqsk_queue_len(const struct request_sock_queue *queue) 279 { 280 return atomic_read(&queue->qlen); 281 } 282 283 static inline int reqsk_queue_len_young(const struct request_sock_queue *queue) 284 { 285 return atomic_read(&queue->young); 286 } 287 288 /* RFC 7323 2.3 Using the Window Scale Option 289 * The window field (SEG.WND) of every outgoing segment, with the 290 * exception of <SYN> segments, MUST be right-shifted by 291 * Rcv.Wind.Shift bits. 292 * 293 * This means the SEG.WND carried in SYNACK can not exceed 65535. 294 * We use this property to harden TCP stack while in NEW_SYN_RECV state. 295 */ 296 static inline u32 tcp_synack_window(const struct request_sock *req) 297 { 298 return min(req->rsk_rcv_wnd, 65535U); 299 } 300 #endif /* _REQUEST_SOCK_H */ 301