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);
129 /* If an incoming SYN or SYNACK frame contains a payload and/or FIN,
130 * queue this additional data / FIN.
132 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb)
134 struct tcp_sock *tp = tcp_sk(sk);
136 if (TCP_SKB_CB(skb)->end_seq == tp->rcv_nxt)
139 skb = skb_clone(skb, GFP_ATOMIC);
144 __skb_pull(skb, tcp_hdrlen(skb));
145 skb_set_owner_r(skb, sk);
147 TCP_SKB_CB(skb)->seq++;
148 TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_SYN;
150 tp->rcv_nxt = TCP_SKB_CB(skb)->end_seq;
151 __skb_queue_tail(&sk->sk_receive_queue, skb);
152 tp->syn_data_acked = 1;
154 /* u64_stats_update_begin(&tp->syncp) not needed here,
155 * as we certainly are not changing upper 32bit value (0)
157 tp->bytes_received = skb->len;
159 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
163 static struct sock *tcp_fastopen_create_child(struct sock *sk,
165 struct dst_entry *dst,
166 struct request_sock *req)
169 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
173 req->num_retrans = 0;
174 req->num_timeout = 0;
177 child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL,
182 spin_lock(&queue->fastopenq.lock);
183 queue->fastopenq.qlen++;
184 spin_unlock(&queue->fastopenq.lock);
186 /* Initialize the child socket. Have to fix some values to take
187 * into account the child is a Fast Open socket and is created
188 * only out of the bits carried in the SYN packet.
192 tp->fastopen_rsk = req;
193 tcp_rsk(req)->tfo_listener = true;
195 /* RFC1323: The window in SYN & SYN/ACK segments is never
196 * scaled. So correct it appropriately.
198 tp->snd_wnd = ntohs(tcp_hdr(skb)->window);
200 /* Activate the retrans timer so that SYNACK can be retransmitted.
201 * The request socket is not added to the ehash
202 * because it's been added to the accept queue directly.
204 inet_csk_reset_xmit_timer(child, ICSK_TIME_RETRANS,
205 TCP_TIMEOUT_INIT, TCP_RTO_MAX);
207 atomic_set(&req->rsk_refcnt, 2);
209 /* Now finish processing the fastopen child socket. */
210 inet_csk(child)->icsk_af_ops->rebuild_header(child);
211 tcp_init_congestion_control(child);
212 tcp_mtup_init(child);
213 tcp_init_metrics(child);
214 tcp_init_buffer_space(child);
216 tp->rcv_nxt = TCP_SKB_CB(skb)->seq + 1;
218 tcp_fastopen_add_skb(child, skb);
220 tcp_rsk(req)->rcv_nxt = tp->rcv_nxt;
221 /* tcp_conn_request() is sending the SYNACK,
222 * and queues the child into listener accept queue.
227 static bool tcp_fastopen_queue_check(struct sock *sk)
229 struct fastopen_queue *fastopenq;
231 /* Make sure the listener has enabled fastopen, and we don't
232 * exceed the max # of pending TFO requests allowed before trying
233 * to validating the cookie in order to avoid burning CPU cycles
236 * XXX (TFO) - The implication of checking the max_qlen before
237 * processing a cookie request is that clients can't differentiate
238 * between qlen overflow causing Fast Open to be disabled
239 * temporarily vs a server not supporting Fast Open at all.
241 fastopenq = &inet_csk(sk)->icsk_accept_queue.fastopenq;
242 if (fastopenq->max_qlen == 0)
245 if (fastopenq->qlen >= fastopenq->max_qlen) {
246 struct request_sock *req1;
247 spin_lock(&fastopenq->lock);
248 req1 = fastopenq->rskq_rst_head;
249 if (!req1 || time_after(req1->rsk_timer.expires, jiffies)) {
250 spin_unlock(&fastopenq->lock);
251 NET_INC_STATS_BH(sock_net(sk),
252 LINUX_MIB_TCPFASTOPENLISTENOVERFLOW);
255 fastopenq->rskq_rst_head = req1->dl_next;
257 spin_unlock(&fastopenq->lock);
263 /* Returns true if we should perform Fast Open on the SYN. The cookie (foc)
264 * may be updated and return the client in the SYN-ACK later. E.g., Fast Open
265 * cookie request (foc->len == 0).
267 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
268 struct request_sock *req,
269 struct tcp_fastopen_cookie *foc,
270 struct dst_entry *dst)
272 struct tcp_fastopen_cookie valid_foc = { .len = -1 };
273 bool syn_data = TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(skb)->seq + 1;
276 if (foc->len == 0) /* Client requests a cookie */
277 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPFASTOPENCOOKIEREQD);
279 if (!((sysctl_tcp_fastopen & TFO_SERVER_ENABLE) &&
280 (syn_data || foc->len >= 0) &&
281 tcp_fastopen_queue_check(sk))) {
286 if (syn_data && (sysctl_tcp_fastopen & TFO_SERVER_COOKIE_NOT_REQD))
289 if (foc->len >= 0 && /* Client presents or requests a cookie */
290 tcp_fastopen_cookie_gen(req, skb, &valid_foc) &&
291 foc->len == TCP_FASTOPEN_COOKIE_SIZE &&
292 foc->len == valid_foc.len &&
293 !memcmp(foc->val, valid_foc.val, foc->len)) {
294 /* Cookie is valid. Create a (full) child socket to accept
295 * the data in SYN before returning a SYN-ACK to ack the
296 * data. If we fail to create the socket, fall back and
297 * ack the ISN only but includes the same cookie.
299 * Note: Data-less SYN with valid cookie is allowed to send
300 * data in SYN_RECV state.
303 child = tcp_fastopen_create_child(sk, skb, dst, req);
306 NET_INC_STATS_BH(sock_net(sk),
307 LINUX_MIB_TCPFASTOPENPASSIVE);
310 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPFASTOPENPASSIVEFAIL);
311 } else if (foc->len > 0) /* Client presents an invalid cookie */
312 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPFASTOPENPASSIVEFAIL);
314 valid_foc.exp = foc->exp;