2 #include <linux/init.h>
3 #include <linux/kernel.h>
4 #include <linux/list.h>
6 #include <linux/rcupdate.h>
7 #include <linux/rculist.h>
8 #include <net/inetpeer.h>
11 int sysctl_tcp_fastopen __read_mostly
= TFO_CLIENT_ENABLE
;
13 struct tcp_fastopen_context __rcu
*tcp_fastopen_ctx
;
15 static DEFINE_SPINLOCK(tcp_fastopen_ctx_lock
);
17 void tcp_fastopen_init_key_once(bool publish
)
19 static u8 key
[TCP_FASTOPEN_KEY_LENGTH
];
21 /* tcp_fastopen_reset_cipher publishes the new context
22 * atomically, so we allow this race happening here.
24 * All call sites of tcp_fastopen_cookie_gen also check
25 * for a valid cookie, so this is an acceptable risk.
27 if (net_get_random_once(key
, sizeof(key
)) && publish
)
28 tcp_fastopen_reset_cipher(key
, sizeof(key
));
31 static void tcp_fastopen_ctx_free(struct rcu_head
*head
)
33 struct tcp_fastopen_context
*ctx
=
34 container_of(head
, struct tcp_fastopen_context
, rcu
);
35 crypto_free_cipher(ctx
->tfm
);
39 int tcp_fastopen_reset_cipher(void *key
, unsigned int len
)
42 struct tcp_fastopen_context
*ctx
, *octx
;
44 ctx
= kmalloc(sizeof(*ctx
), GFP_KERNEL
);
47 ctx
->tfm
= crypto_alloc_cipher("aes", 0, 0);
49 if (IS_ERR(ctx
->tfm
)) {
50 err
= PTR_ERR(ctx
->tfm
);
52 pr_err("TCP: TFO aes cipher alloc error: %d\n", err
);
55 err
= crypto_cipher_setkey(ctx
->tfm
, key
, len
);
57 pr_err("TCP: TFO cipher key error: %d\n", err
);
58 crypto_free_cipher(ctx
->tfm
);
61 memcpy(ctx
->key
, key
, len
);
63 spin_lock(&tcp_fastopen_ctx_lock
);
65 octx
= rcu_dereference_protected(tcp_fastopen_ctx
,
66 lockdep_is_held(&tcp_fastopen_ctx_lock
));
67 rcu_assign_pointer(tcp_fastopen_ctx
, ctx
);
68 spin_unlock(&tcp_fastopen_ctx_lock
);
71 call_rcu(&octx
->rcu
, tcp_fastopen_ctx_free
);
75 static bool __tcp_fastopen_cookie_gen(const void *path
,
76 struct tcp_fastopen_cookie
*foc
)
78 struct tcp_fastopen_context
*ctx
;
82 ctx
= rcu_dereference(tcp_fastopen_ctx
);
84 crypto_cipher_encrypt_one(ctx
->tfm
, foc
->val
, path
);
85 foc
->len
= TCP_FASTOPEN_COOKIE_SIZE
;
92 /* Generate the fastopen cookie by doing aes128 encryption on both
93 * the source and destination addresses. Pad 0s for IPv4 or IPv4-mapped-IPv6
94 * addresses. For the longer IPv6 addresses use CBC-MAC.
96 * XXX (TFO) - refactor when TCP_FASTOPEN_COOKIE_SIZE != AES_BLOCK_SIZE.
98 static bool tcp_fastopen_cookie_gen(struct request_sock
*req
,
100 struct tcp_fastopen_cookie
*foc
)
102 if (req
->rsk_ops
->family
== AF_INET
) {
103 const struct iphdr
*iph
= ip_hdr(syn
);
105 __be32 path
[4] = { iph
->saddr
, iph
->daddr
, 0, 0 };
106 return __tcp_fastopen_cookie_gen(path
, foc
);
109 #if IS_ENABLED(CONFIG_IPV6)
110 if (req
->rsk_ops
->family
== AF_INET6
) {
111 const struct ipv6hdr
*ip6h
= ipv6_hdr(syn
);
112 struct tcp_fastopen_cookie tmp
;
114 if (__tcp_fastopen_cookie_gen(&ip6h
->saddr
, &tmp
)) {
115 struct in6_addr
*buf
= (struct in6_addr
*) tmp
.val
;
118 for (i
= 0; i
< 4; i
++)
119 buf
->s6_addr32
[i
] ^= ip6h
->daddr
.s6_addr32
[i
];
120 return __tcp_fastopen_cookie_gen(buf
, foc
);
128 /* If an incoming SYN or SYNACK frame contains a payload and/or FIN,
129 * queue this additional data / FIN.
131 void tcp_fastopen_add_skb(struct sock
*sk
, struct sk_buff
*skb
)
133 struct tcp_sock
*tp
= tcp_sk(sk
);
135 if (TCP_SKB_CB(skb
)->end_seq
== tp
->rcv_nxt
)
138 skb
= skb_clone(skb
, GFP_ATOMIC
);
143 __skb_pull(skb
, tcp_hdrlen(skb
));
144 skb_set_owner_r(skb
, sk
);
146 tp
->rcv_nxt
= TCP_SKB_CB(skb
)->end_seq
;
147 __skb_queue_tail(&sk
->sk_receive_queue
, skb
);
148 tp
->syn_data_acked
= 1;
150 /* u64_stats_update_begin(&tp->syncp) not needed here,
151 * as we certainly are not changing upper 32bit value (0)
153 tp
->bytes_received
= skb
->len
;
156 static struct sock
*tcp_fastopen_create_child(struct sock
*sk
,
158 struct dst_entry
*dst
,
159 struct request_sock
*req
)
162 struct request_sock_queue
*queue
= &inet_csk(sk
)->icsk_accept_queue
;
166 req
->num_retrans
= 0;
167 req
->num_timeout
= 0;
170 child
= inet_csk(sk
)->icsk_af_ops
->syn_recv_sock(sk
, skb
, req
, NULL
,
175 spin_lock(&queue
->fastopenq
.lock
);
176 queue
->fastopenq
.qlen
++;
177 spin_unlock(&queue
->fastopenq
.lock
);
179 /* Initialize the child socket. Have to fix some values to take
180 * into account the child is a Fast Open socket and is created
181 * only out of the bits carried in the SYN packet.
185 tp
->fastopen_rsk
= req
;
186 tcp_rsk(req
)->tfo_listener
= true;
188 /* RFC1323: The window in SYN & SYN/ACK segments is never
189 * scaled. So correct it appropriately.
191 tp
->snd_wnd
= ntohs(tcp_hdr(skb
)->window
);
193 /* Activate the retrans timer so that SYNACK can be retransmitted.
194 * The request socket is not added to the ehash
195 * because it's been added to the accept queue directly.
197 inet_csk_reset_xmit_timer(child
, ICSK_TIME_RETRANS
,
198 TCP_TIMEOUT_INIT
, TCP_RTO_MAX
);
200 atomic_set(&req
->rsk_refcnt
, 2);
202 /* Now finish processing the fastopen child socket. */
203 inet_csk(child
)->icsk_af_ops
->rebuild_header(child
);
204 tcp_init_congestion_control(child
);
205 tcp_mtup_init(child
);
206 tcp_init_metrics(child
);
207 tcp_init_buffer_space(child
);
209 tp
->rcv_nxt
= TCP_SKB_CB(skb
)->seq
+ 1;
211 tcp_fastopen_add_skb(child
, skb
);
213 tcp_rsk(req
)->rcv_nxt
= tp
->rcv_nxt
;
214 /* tcp_conn_request() is sending the SYNACK,
215 * and queues the child into listener accept queue.
220 static bool tcp_fastopen_queue_check(struct sock
*sk
)
222 struct fastopen_queue
*fastopenq
;
224 /* Make sure the listener has enabled fastopen, and we don't
225 * exceed the max # of pending TFO requests allowed before trying
226 * to validating the cookie in order to avoid burning CPU cycles
229 * XXX (TFO) - The implication of checking the max_qlen before
230 * processing a cookie request is that clients can't differentiate
231 * between qlen overflow causing Fast Open to be disabled
232 * temporarily vs a server not supporting Fast Open at all.
234 fastopenq
= &inet_csk(sk
)->icsk_accept_queue
.fastopenq
;
235 if (fastopenq
->max_qlen
== 0)
238 if (fastopenq
->qlen
>= fastopenq
->max_qlen
) {
239 struct request_sock
*req1
;
240 spin_lock(&fastopenq
->lock
);
241 req1
= fastopenq
->rskq_rst_head
;
242 if (!req1
|| time_after(req1
->rsk_timer
.expires
, jiffies
)) {
243 spin_unlock(&fastopenq
->lock
);
244 NET_INC_STATS_BH(sock_net(sk
),
245 LINUX_MIB_TCPFASTOPENLISTENOVERFLOW
);
248 fastopenq
->rskq_rst_head
= req1
->dl_next
;
250 spin_unlock(&fastopenq
->lock
);
256 /* Returns true if we should perform Fast Open on the SYN. The cookie (foc)
257 * may be updated and return the client in the SYN-ACK later. E.g., Fast Open
258 * cookie request (foc->len == 0).
260 struct sock
*tcp_try_fastopen(struct sock
*sk
, struct sk_buff
*skb
,
261 struct request_sock
*req
,
262 struct tcp_fastopen_cookie
*foc
,
263 struct dst_entry
*dst
)
265 struct tcp_fastopen_cookie valid_foc
= { .len
= -1 };
266 bool syn_data
= TCP_SKB_CB(skb
)->end_seq
!= TCP_SKB_CB(skb
)->seq
+ 1;
269 if (foc
->len
== 0) /* Client requests a cookie */
270 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPFASTOPENCOOKIEREQD
);
272 if (!((sysctl_tcp_fastopen
& TFO_SERVER_ENABLE
) &&
273 (syn_data
|| foc
->len
>= 0) &&
274 tcp_fastopen_queue_check(sk
))) {
279 if (syn_data
&& (sysctl_tcp_fastopen
& TFO_SERVER_COOKIE_NOT_REQD
))
282 if (foc
->len
>= 0 && /* Client presents or requests a cookie */
283 tcp_fastopen_cookie_gen(req
, skb
, &valid_foc
) &&
284 foc
->len
== TCP_FASTOPEN_COOKIE_SIZE
&&
285 foc
->len
== valid_foc
.len
&&
286 !memcmp(foc
->val
, valid_foc
.val
, foc
->len
)) {
287 /* Cookie is valid. Create a (full) child socket to accept
288 * the data in SYN before returning a SYN-ACK to ack the
289 * data. If we fail to create the socket, fall back and
290 * ack the ISN only but includes the same cookie.
292 * Note: Data-less SYN with valid cookie is allowed to send
293 * data in SYN_RECV state.
296 child
= tcp_fastopen_create_child(sk
, skb
, dst
, req
);
299 NET_INC_STATS_BH(sock_net(sk
),
300 LINUX_MIB_TCPFASTOPENPASSIVE
);
303 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPFASTOPENPASSIVEFAIL
);
304 } else if (foc
->len
> 0) /* Client presents an invalid cookie */
305 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPFASTOPENPASSIVEFAIL
);
307 valid_foc
.exp
= foc
->exp
;