1 #include <linux/crypto.h>
3 #include <linux/init.h>
4 #include <linux/kernel.h>
5 #include <linux/list.h>
7 #include <linux/rcupdate.h>
8 #include <linux/rculist.h>
9 #include <net/inetpeer.h>
12 int sysctl_tcp_fastopen __read_mostly
= TFO_CLIENT_ENABLE
;
14 struct tcp_fastopen_context __rcu
*tcp_fastopen_ctx
;
16 static DEFINE_SPINLOCK(tcp_fastopen_ctx_lock
);
18 void tcp_fastopen_init_key_once(bool publish
)
20 static u8 key
[TCP_FASTOPEN_KEY_LENGTH
];
22 /* tcp_fastopen_reset_cipher publishes the new context
23 * atomically, so we allow this race happening here.
25 * All call sites of tcp_fastopen_cookie_gen also check
26 * for a valid cookie, so this is an acceptable risk.
28 if (net_get_random_once(key
, sizeof(key
)) && publish
)
29 tcp_fastopen_reset_cipher(key
, sizeof(key
));
32 static void tcp_fastopen_ctx_free(struct rcu_head
*head
)
34 struct tcp_fastopen_context
*ctx
=
35 container_of(head
, struct tcp_fastopen_context
, rcu
);
36 crypto_free_cipher(ctx
->tfm
);
40 int tcp_fastopen_reset_cipher(void *key
, unsigned int len
)
43 struct tcp_fastopen_context
*ctx
, *octx
;
45 ctx
= kmalloc(sizeof(*ctx
), GFP_KERNEL
);
48 ctx
->tfm
= crypto_alloc_cipher("aes", 0, 0);
50 if (IS_ERR(ctx
->tfm
)) {
51 err
= PTR_ERR(ctx
->tfm
);
53 pr_err("TCP: TFO aes cipher alloc error: %d\n", err
);
56 err
= crypto_cipher_setkey(ctx
->tfm
, key
, len
);
58 pr_err("TCP: TFO cipher key error: %d\n", err
);
59 crypto_free_cipher(ctx
->tfm
);
62 memcpy(ctx
->key
, key
, len
);
64 spin_lock(&tcp_fastopen_ctx_lock
);
66 octx
= rcu_dereference_protected(tcp_fastopen_ctx
,
67 lockdep_is_held(&tcp_fastopen_ctx_lock
));
68 rcu_assign_pointer(tcp_fastopen_ctx
, ctx
);
69 spin_unlock(&tcp_fastopen_ctx_lock
);
72 call_rcu(&octx
->rcu
, tcp_fastopen_ctx_free
);
76 static bool __tcp_fastopen_cookie_gen(const void *path
,
77 struct tcp_fastopen_cookie
*foc
)
79 struct tcp_fastopen_context
*ctx
;
83 ctx
= rcu_dereference(tcp_fastopen_ctx
);
85 crypto_cipher_encrypt_one(ctx
->tfm
, foc
->val
, path
);
86 foc
->len
= TCP_FASTOPEN_COOKIE_SIZE
;
93 /* Generate the fastopen cookie by doing aes128 encryption on both
94 * the source and destination addresses. Pad 0s for IPv4 or IPv4-mapped-IPv6
95 * addresses. For the longer IPv6 addresses use CBC-MAC.
97 * XXX (TFO) - refactor when TCP_FASTOPEN_COOKIE_SIZE != AES_BLOCK_SIZE.
99 static bool tcp_fastopen_cookie_gen(struct request_sock
*req
,
101 struct tcp_fastopen_cookie
*foc
)
103 if (req
->rsk_ops
->family
== AF_INET
) {
104 const struct iphdr
*iph
= ip_hdr(syn
);
106 __be32 path
[4] = { iph
->saddr
, iph
->daddr
, 0, 0 };
107 return __tcp_fastopen_cookie_gen(path
, foc
);
110 #if IS_ENABLED(CONFIG_IPV6)
111 if (req
->rsk_ops
->family
== AF_INET6
) {
112 const struct ipv6hdr
*ip6h
= ipv6_hdr(syn
);
113 struct tcp_fastopen_cookie tmp
;
115 if (__tcp_fastopen_cookie_gen(&ip6h
->saddr
, &tmp
)) {
116 struct in6_addr
*buf
= (struct in6_addr
*) tmp
.val
;
119 for (i
= 0; i
< 4; i
++)
120 buf
->s6_addr32
[i
] ^= ip6h
->daddr
.s6_addr32
[i
];
121 return __tcp_fastopen_cookie_gen(buf
, foc
);
128 static struct sock
*tcp_fastopen_create_child(struct sock
*sk
,
130 struct dst_entry
*dst
,
131 struct request_sock
*req
)
134 struct request_sock_queue
*queue
= &inet_csk(sk
)->icsk_accept_queue
;
139 req
->num_retrans
= 0;
140 req
->num_timeout
= 0;
143 child
= inet_csk(sk
)->icsk_af_ops
->syn_recv_sock(sk
, skb
, req
, NULL
,
148 spin_lock(&queue
->fastopenq
.lock
);
149 queue
->fastopenq
.qlen
++;
150 spin_unlock(&queue
->fastopenq
.lock
);
152 /* Initialize the child socket. Have to fix some values to take
153 * into account the child is a Fast Open socket and is created
154 * only out of the bits carried in the SYN packet.
158 tp
->fastopen_rsk
= req
;
159 tcp_rsk(req
)->tfo_listener
= true;
161 /* RFC1323: The window in SYN & SYN/ACK segments is never
162 * scaled. So correct it appropriately.
164 tp
->snd_wnd
= ntohs(tcp_hdr(skb
)->window
);
166 /* Activate the retrans timer so that SYNACK can be retransmitted.
167 * The request socket is not added to the ehash
168 * because it's been added to the accept queue directly.
170 inet_csk_reset_xmit_timer(child
, ICSK_TIME_RETRANS
,
171 TCP_TIMEOUT_INIT
, TCP_RTO_MAX
);
173 atomic_set(&req
->rsk_refcnt
, 2);
175 /* Now finish processing the fastopen child socket. */
176 inet_csk(child
)->icsk_af_ops
->rebuild_header(child
);
177 tcp_init_congestion_control(child
);
178 tcp_mtup_init(child
);
179 tcp_init_metrics(child
);
180 tcp_init_buffer_space(child
);
182 /* Queue the data carried in the SYN packet.
183 * We used to play tricky games with skb_get().
184 * With lockless listener, it is a dead end.
185 * Do not think about it.
187 * XXX (TFO) - we honor a zero-payload TFO request for now,
188 * (any reason not to?) but no need to queue the skb since
189 * there is no data. How about SYN+FIN?
191 end_seq
= TCP_SKB_CB(skb
)->end_seq
;
192 if (end_seq
!= TCP_SKB_CB(skb
)->seq
+ 1) {
193 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
197 __skb_pull(skb2
, tcp_hdrlen(skb
));
198 skb_set_owner_r(skb2
, child
);
199 __skb_queue_tail(&child
->sk_receive_queue
, skb2
);
200 tp
->syn_data_acked
= 1;
202 /* u64_stats_update_begin(&tp->syncp) not needed here,
203 * as we certainly are not changing upper 32bit value (0)
205 tp
->bytes_received
= end_seq
- TCP_SKB_CB(skb
)->seq
- 1;
207 end_seq
= TCP_SKB_CB(skb
)->seq
+ 1;
210 tcp_rsk(req
)->rcv_nxt
= tp
->rcv_nxt
= end_seq
;
211 /* tcp_conn_request() is sending the SYNACK,
212 * and queues the child into listener accept queue.
217 static bool tcp_fastopen_queue_check(struct sock
*sk
)
219 struct fastopen_queue
*fastopenq
;
221 /* Make sure the listener has enabled fastopen, and we don't
222 * exceed the max # of pending TFO requests allowed before trying
223 * to validating the cookie in order to avoid burning CPU cycles
226 * XXX (TFO) - The implication of checking the max_qlen before
227 * processing a cookie request is that clients can't differentiate
228 * between qlen overflow causing Fast Open to be disabled
229 * temporarily vs a server not supporting Fast Open at all.
231 fastopenq
= &inet_csk(sk
)->icsk_accept_queue
.fastopenq
;
232 if (fastopenq
->max_qlen
== 0)
235 if (fastopenq
->qlen
>= fastopenq
->max_qlen
) {
236 struct request_sock
*req1
;
237 spin_lock(&fastopenq
->lock
);
238 req1
= fastopenq
->rskq_rst_head
;
239 if (!req1
|| time_after(req1
->rsk_timer
.expires
, jiffies
)) {
240 spin_unlock(&fastopenq
->lock
);
241 NET_INC_STATS_BH(sock_net(sk
),
242 LINUX_MIB_TCPFASTOPENLISTENOVERFLOW
);
245 fastopenq
->rskq_rst_head
= req1
->dl_next
;
247 spin_unlock(&fastopenq
->lock
);
253 /* Returns true if we should perform Fast Open on the SYN. The cookie (foc)
254 * may be updated and return the client in the SYN-ACK later. E.g., Fast Open
255 * cookie request (foc->len == 0).
257 struct sock
*tcp_try_fastopen(struct sock
*sk
, struct sk_buff
*skb
,
258 struct request_sock
*req
,
259 struct tcp_fastopen_cookie
*foc
,
260 struct dst_entry
*dst
)
262 struct tcp_fastopen_cookie valid_foc
= { .len
= -1 };
263 bool syn_data
= TCP_SKB_CB(skb
)->end_seq
!= TCP_SKB_CB(skb
)->seq
+ 1;
266 if (foc
->len
== 0) /* Client requests a cookie */
267 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPFASTOPENCOOKIEREQD
);
269 if (!((sysctl_tcp_fastopen
& TFO_SERVER_ENABLE
) &&
270 (syn_data
|| foc
->len
>= 0) &&
271 tcp_fastopen_queue_check(sk
))) {
276 if (syn_data
&& (sysctl_tcp_fastopen
& TFO_SERVER_COOKIE_NOT_REQD
))
279 if (foc
->len
>= 0 && /* Client presents or requests a cookie */
280 tcp_fastopen_cookie_gen(req
, skb
, &valid_foc
) &&
281 foc
->len
== TCP_FASTOPEN_COOKIE_SIZE
&&
282 foc
->len
== valid_foc
.len
&&
283 !memcmp(foc
->val
, valid_foc
.val
, foc
->len
)) {
284 /* Cookie is valid. Create a (full) child socket to accept
285 * the data in SYN before returning a SYN-ACK to ack the
286 * data. If we fail to create the socket, fall back and
287 * ack the ISN only but includes the same cookie.
289 * Note: Data-less SYN with valid cookie is allowed to send
290 * data in SYN_RECV state.
293 child
= tcp_fastopen_create_child(sk
, skb
, dst
, req
);
296 NET_INC_STATS_BH(sock_net(sk
),
297 LINUX_MIB_TCPFASTOPENPASSIVE
);
300 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPFASTOPENPASSIVEFAIL
);
301 } else if (foc
->len
> 0) /* Client presents an invalid cookie */
302 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPFASTOPENPASSIVEFAIL
);
304 valid_foc
.exp
= foc
->exp
;