]>
Commit | Line | Data |
---|---|---|
7c657876 ACM |
1 | /* |
2 | * net/dccp/output.c | |
3 | * | |
4 | * An implementation of the DCCP protocol | |
5 | * Arnaldo Carvalho de Melo <acme@conectiva.com.br> | |
6 | * | |
7 | * This program is free software; you can redistribute it and/or | |
8 | * modify it under the terms of the GNU General Public License | |
9 | * as published by the Free Software Foundation; either version | |
10 | * 2 of the License, or (at your option) any later version. | |
11 | */ | |
12 | ||
13 | #include <linux/config.h> | |
14 | #include <linux/dccp.h> | |
15 | #include <linux/skbuff.h> | |
16 | ||
17 | #include <net/sock.h> | |
18 | ||
19 | #include "ccid.h" | |
20 | #include "dccp.h" | |
21 | ||
22 | static inline void dccp_event_ack_sent(struct sock *sk) | |
23 | { | |
24 | inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK); | |
25 | } | |
26 | ||
27 | /* | |
28 | * All SKB's seen here are completely headerless. It is our | |
29 | * job to build the DCCP header, and pass the packet down to | |
30 | * IP so it can do the same plus pass the packet off to the | |
31 | * device. | |
32 | */ | |
33 | int dccp_transmit_skb(struct sock *sk, struct sk_buff *skb) | |
34 | { | |
35 | if (likely(skb != NULL)) { | |
36 | const struct inet_sock *inet = inet_sk(sk); | |
37 | struct dccp_sock *dp = dccp_sk(sk); | |
38 | struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb); | |
39 | struct dccp_hdr *dh; | |
40 | /* XXX For now we're using only 48 bits sequence numbers */ | |
41 | const int dccp_header_size = sizeof(*dh) + | |
42 | sizeof(struct dccp_hdr_ext) + | |
43 | dccp_packet_hdr_len(dcb->dccpd_type); | |
44 | int err, set_ack = 1; | |
45 | u64 ackno = dp->dccps_gsr; | |
46 | ||
47 | /* | |
48 | * FIXME: study DCCP_PKT_SYNC[ACK] to see what is the right thing | |
49 | * to do here... | |
50 | */ | |
51 | dccp_inc_seqno(&dp->dccps_gss); | |
52 | ||
53 | dcb->dccpd_seq = dp->dccps_gss; | |
54 | dccp_insert_options(sk, skb); | |
55 | ||
56 | switch (dcb->dccpd_type) { | |
57 | case DCCP_PKT_DATA: | |
58 | set_ack = 0; | |
59 | break; | |
60 | case DCCP_PKT_SYNC: | |
61 | case DCCP_PKT_SYNCACK: | |
62 | ackno = dcb->dccpd_seq; | |
63 | break; | |
64 | } | |
65 | ||
66 | skb->h.raw = skb_push(skb, dccp_header_size); | |
67 | dh = dccp_hdr(skb); | |
68 | /* Data packets are not cloned as they are never retransmitted */ | |
69 | if (skb_cloned(skb)) | |
70 | skb_set_owner_w(skb, sk); | |
71 | ||
72 | /* Build DCCP header and checksum it. */ | |
73 | memset(dh, 0, dccp_header_size); | |
74 | dh->dccph_type = dcb->dccpd_type; | |
75 | dh->dccph_sport = inet->sport; | |
76 | dh->dccph_dport = inet->dport; | |
77 | dh->dccph_doff = (dccp_header_size + dcb->dccpd_opt_len) / 4; | |
78 | dh->dccph_ccval = dcb->dccpd_ccval; | |
79 | /* XXX For now we're using only 48 bits sequence numbers */ | |
80 | dh->dccph_x = 1; | |
81 | ||
82 | dp->dccps_awh = dp->dccps_gss; | |
83 | dccp_hdr_set_seq(dh, dp->dccps_gss); | |
84 | if (set_ack) | |
85 | dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), ackno); | |
86 | ||
87 | switch (dcb->dccpd_type) { | |
88 | case DCCP_PKT_REQUEST: | |
89 | dccp_hdr_request(skb)->dccph_req_service = dcb->dccpd_service; | |
90 | break; | |
91 | case DCCP_PKT_RESET: | |
92 | dccp_hdr_reset(skb)->dccph_reset_code = dcb->dccpd_reset_code; | |
93 | break; | |
94 | } | |
95 | ||
95b81ef7 YN |
96 | dh->dccph_checksum = dccp_v4_checksum(skb, inet->saddr, |
97 | inet->daddr); | |
7c657876 ACM |
98 | |
99 | if (dcb->dccpd_type == DCCP_PKT_ACK || | |
100 | dcb->dccpd_type == DCCP_PKT_DATAACK) | |
101 | dccp_event_ack_sent(sk); | |
102 | ||
103 | DCCP_INC_STATS(DCCP_MIB_OUTSEGS); | |
104 | ||
105 | err = ip_queue_xmit(skb, 0); | |
106 | if (err <= 0) | |
107 | return err; | |
108 | ||
109 | /* NET_XMIT_CN is special. It does not guarantee, | |
110 | * that this packet is lost. It tells that device | |
111 | * is about to start to drop packets or already | |
112 | * drops some packets of the same priority and | |
113 | * invokes us to send less aggressively. | |
114 | */ | |
115 | return err == NET_XMIT_CN ? 0 : err; | |
116 | } | |
117 | return -ENOBUFS; | |
118 | } | |
119 | ||
120 | unsigned int dccp_sync_mss(struct sock *sk, u32 pmtu) | |
121 | { | |
122 | struct dccp_sock *dp = dccp_sk(sk); | |
123 | int mss_now; | |
124 | ||
125 | /* | |
126 | * FIXME: we really should be using the af_specific thing to support IPv6. | |
127 | * mss_now = pmtu - tp->af_specific->net_header_len - sizeof(struct dccp_hdr) - sizeof(struct dccp_hdr_ext); | |
128 | */ | |
129 | mss_now = pmtu - sizeof(struct iphdr) - sizeof(struct dccp_hdr) - sizeof(struct dccp_hdr_ext); | |
130 | ||
131 | /* Now subtract optional transport overhead */ | |
132 | mss_now -= dp->dccps_ext_header_len; | |
133 | ||
134 | /* | |
135 | * FIXME: this should come from the CCID infrastructure, where, say, | |
136 | * TFRC will say it wants TIMESTAMPS, ELAPSED time, etc, for now lets | |
137 | * put a rough estimate for NDP + TIMESTAMP + TIMESTAMP_ECHO + ELAPSED | |
138 | * TIME + TFRC_OPT_LOSS_EVENT_RATE + TFRC_OPT_RECEIVE_RATE + padding to | |
139 | * make it a multiple of 4 | |
140 | */ | |
141 | ||
142 | mss_now -= ((5 + 6 + 10 + 6 + 6 + 6 + 3) / 4) * 4; | |
143 | ||
144 | /* And store cached results */ | |
145 | dp->dccps_pmtu_cookie = pmtu; | |
146 | dp->dccps_mss_cache = mss_now; | |
147 | ||
148 | return mss_now; | |
149 | } | |
150 | ||
151 | int dccp_retransmit_skb(struct sock *sk, struct sk_buff *skb) | |
152 | { | |
153 | if (inet_sk_rebuild_header(sk) != 0) | |
154 | return -EHOSTUNREACH; /* Routing failure or similar. */ | |
155 | ||
156 | return dccp_transmit_skb(sk, (skb_cloned(skb) ? | |
157 | pskb_copy(skb, GFP_ATOMIC): | |
158 | skb_clone(skb, GFP_ATOMIC))); | |
159 | } | |
160 | ||
161 | struct sk_buff *dccp_make_response(struct sock *sk, struct dst_entry *dst, | |
162 | struct request_sock *req) | |
163 | { | |
164 | struct dccp_hdr *dh; | |
165 | const int dccp_header_size = sizeof(struct dccp_hdr) + | |
166 | sizeof(struct dccp_hdr_ext) + | |
167 | sizeof(struct dccp_hdr_response); | |
168 | struct sk_buff *skb = sock_wmalloc(sk, MAX_HEADER + DCCP_MAX_OPT_LEN + | |
169 | dccp_header_size, 1, | |
170 | GFP_ATOMIC); | |
171 | if (skb == NULL) | |
172 | return NULL; | |
173 | ||
174 | /* Reserve space for headers. */ | |
175 | skb_reserve(skb, MAX_HEADER + DCCP_MAX_OPT_LEN + dccp_header_size); | |
176 | ||
177 | skb->dst = dst_clone(dst); | |
178 | skb->csum = 0; | |
179 | ||
180 | DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE; | |
181 | DCCP_SKB_CB(skb)->dccpd_seq = dccp_rsk(req)->dreq_iss; | |
182 | dccp_insert_options(sk, skb); | |
183 | ||
184 | skb->h.raw = skb_push(skb, dccp_header_size); | |
185 | ||
186 | dh = dccp_hdr(skb); | |
187 | memset(dh, 0, dccp_header_size); | |
188 | ||
189 | dh->dccph_sport = inet_sk(sk)->sport; | |
190 | dh->dccph_dport = inet_rsk(req)->rmt_port; | |
191 | dh->dccph_doff = (dccp_header_size + DCCP_SKB_CB(skb)->dccpd_opt_len) / 4; | |
192 | dh->dccph_type = DCCP_PKT_RESPONSE; | |
193 | dh->dccph_x = 1; | |
194 | dccp_hdr_set_seq(dh, dccp_rsk(req)->dreq_iss); | |
195 | dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dccp_rsk(req)->dreq_isr); | |
196 | ||
95b81ef7 YN |
197 | dh->dccph_checksum = dccp_v4_checksum(skb, inet_rsk(req)->loc_addr, |
198 | inet_rsk(req)->rmt_addr); | |
7c657876 ACM |
199 | |
200 | DCCP_INC_STATS(DCCP_MIB_OUTSEGS); | |
201 | return skb; | |
202 | } | |
203 | ||
204 | struct sk_buff *dccp_make_reset(struct sock *sk, struct dst_entry *dst, | |
205 | const enum dccp_reset_codes code) | |
206 | ||
207 | { | |
208 | struct dccp_hdr *dh; | |
209 | struct dccp_sock *dp = dccp_sk(sk); | |
210 | const int dccp_header_size = sizeof(struct dccp_hdr) + | |
211 | sizeof(struct dccp_hdr_ext) + | |
212 | sizeof(struct dccp_hdr_reset); | |
213 | struct sk_buff *skb = sock_wmalloc(sk, MAX_HEADER + DCCP_MAX_OPT_LEN + | |
214 | dccp_header_size, 1, | |
215 | GFP_ATOMIC); | |
216 | if (skb == NULL) | |
217 | return NULL; | |
218 | ||
219 | /* Reserve space for headers. */ | |
220 | skb_reserve(skb, MAX_HEADER + DCCP_MAX_OPT_LEN + dccp_header_size); | |
221 | ||
222 | skb->dst = dst_clone(dst); | |
223 | skb->csum = 0; | |
224 | ||
225 | dccp_inc_seqno(&dp->dccps_gss); | |
226 | ||
227 | DCCP_SKB_CB(skb)->dccpd_reset_code = code; | |
228 | DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESET; | |
229 | DCCP_SKB_CB(skb)->dccpd_seq = dp->dccps_gss; | |
230 | dccp_insert_options(sk, skb); | |
231 | ||
232 | skb->h.raw = skb_push(skb, dccp_header_size); | |
233 | ||
234 | dh = dccp_hdr(skb); | |
235 | memset(dh, 0, dccp_header_size); | |
236 | ||
237 | dh->dccph_sport = inet_sk(sk)->sport; | |
238 | dh->dccph_dport = inet_sk(sk)->dport; | |
239 | dh->dccph_doff = (dccp_header_size + DCCP_SKB_CB(skb)->dccpd_opt_len) / 4; | |
240 | dh->dccph_type = DCCP_PKT_RESET; | |
241 | dh->dccph_x = 1; | |
242 | dccp_hdr_set_seq(dh, dp->dccps_gss); | |
243 | dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dp->dccps_gsr); | |
244 | ||
245 | dccp_hdr_reset(skb)->dccph_reset_code = code; | |
246 | ||
95b81ef7 YN |
247 | dh->dccph_checksum = dccp_v4_checksum(skb, inet_sk(sk)->saddr, |
248 | inet_sk(sk)->daddr); | |
7c657876 ACM |
249 | |
250 | DCCP_INC_STATS(DCCP_MIB_OUTSEGS); | |
251 | return skb; | |
252 | } | |
253 | ||
254 | /* | |
255 | * Do all connect socket setups that can be done AF independent. | |
256 | */ | |
257 | static inline void dccp_connect_init(struct sock *sk) | |
258 | { | |
259 | struct dst_entry *dst = __sk_dst_get(sk); | |
260 | struct inet_connection_sock *icsk = inet_csk(sk); | |
261 | ||
262 | sk->sk_err = 0; | |
263 | sock_reset_flag(sk, SOCK_DONE); | |
264 | ||
265 | dccp_sync_mss(sk, dst_mtu(dst)); | |
266 | ||
267 | /* | |
268 | * FIXME: set dp->{dccps_swh,dccps_swl}, with | |
269 | * something like dccp_inc_seq | |
270 | */ | |
271 | ||
272 | icsk->icsk_retransmits = 0; | |
273 | } | |
274 | ||
275 | int dccp_connect(struct sock *sk) | |
276 | { | |
277 | struct sk_buff *skb; | |
278 | struct inet_connection_sock *icsk = inet_csk(sk); | |
279 | ||
280 | dccp_connect_init(sk); | |
281 | ||
282 | skb = alloc_skb(MAX_DCCP_HEADER + 15, sk->sk_allocation); | |
283 | if (unlikely(skb == NULL)) | |
284 | return -ENOBUFS; | |
285 | ||
286 | /* Reserve space for headers. */ | |
287 | skb_reserve(skb, MAX_DCCP_HEADER); | |
288 | ||
289 | DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST; | |
290 | /* FIXME: set service to something meaningful, coming | |
291 | * from userspace*/ | |
292 | DCCP_SKB_CB(skb)->dccpd_service = 0; | |
293 | skb->csum = 0; | |
294 | skb_set_owner_w(skb, sk); | |
295 | ||
296 | BUG_TRAP(sk->sk_send_head == NULL); | |
297 | sk->sk_send_head = skb; | |
298 | dccp_transmit_skb(sk, skb_clone(skb, GFP_KERNEL)); | |
299 | DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS); | |
300 | ||
301 | /* Timer for repeating the REQUEST until an answer. */ | |
302 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, icsk->icsk_rto, TCP_RTO_MAX); | |
303 | return 0; | |
304 | } | |
305 | ||
306 | void dccp_send_ack(struct sock *sk) | |
307 | { | |
308 | /* If we have been reset, we may not send again. */ | |
309 | if (sk->sk_state != DCCP_CLOSED) { | |
310 | struct sk_buff *skb = alloc_skb(MAX_DCCP_HEADER, GFP_ATOMIC); | |
311 | ||
312 | if (skb == NULL) { | |
313 | inet_csk_schedule_ack(sk); | |
314 | inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN; | |
315 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, TCP_DELACK_MAX, TCP_RTO_MAX); | |
316 | return; | |
317 | } | |
318 | ||
319 | /* Reserve space for headers */ | |
320 | skb_reserve(skb, MAX_DCCP_HEADER); | |
321 | skb->csum = 0; | |
322 | DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK; | |
323 | skb_set_owner_w(skb, sk); | |
324 | dccp_transmit_skb(sk, skb); | |
325 | } | |
326 | } | |
327 | ||
328 | EXPORT_SYMBOL_GPL(dccp_send_ack); | |
329 | ||
330 | void dccp_send_delayed_ack(struct sock *sk) | |
331 | { | |
332 | struct inet_connection_sock *icsk = inet_csk(sk); | |
333 | /* | |
334 | * FIXME: tune this timer. elapsed time fixes the skew, so no problem | |
335 | * with using 2s, and active senders also piggyback the ACK into a | |
336 | * DATAACK packet, so this is really for quiescent senders. | |
337 | */ | |
338 | unsigned long timeout = jiffies + 2 * HZ; | |
339 | ||
340 | /* Use new timeout only if there wasn't a older one earlier. */ | |
341 | if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) { | |
342 | /* If delack timer was blocked or is about to expire, | |
343 | * send ACK now. | |
344 | * | |
345 | * FIXME: check the "about to expire" part | |
346 | */ | |
347 | if (icsk->icsk_ack.blocked) { | |
348 | dccp_send_ack(sk); | |
349 | return; | |
350 | } | |
351 | ||
352 | if (!time_before(timeout, icsk->icsk_ack.timeout)) | |
353 | timeout = icsk->icsk_ack.timeout; | |
354 | } | |
355 | icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER; | |
356 | icsk->icsk_ack.timeout = timeout; | |
357 | sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout); | |
358 | } | |
359 | ||
360 | void dccp_send_sync(struct sock *sk, u64 seq) | |
361 | { | |
362 | /* | |
363 | * We are not putting this on the write queue, so | |
364 | * dccp_transmit_skb() will set the ownership to this | |
365 | * sock. | |
366 | */ | |
367 | struct sk_buff *skb = alloc_skb(MAX_DCCP_HEADER, GFP_ATOMIC); | |
368 | ||
369 | if (skb == NULL) | |
370 | /* FIXME: how to make sure the sync is sent? */ | |
371 | return; | |
372 | ||
373 | /* Reserve space for headers and prepare control bits. */ | |
374 | skb_reserve(skb, MAX_DCCP_HEADER); | |
375 | skb->csum = 0; | |
376 | DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_SYNC; | |
377 | DCCP_SKB_CB(skb)->dccpd_seq = seq; | |
378 | ||
379 | skb_set_owner_w(skb, sk); | |
380 | dccp_transmit_skb(sk, skb); | |
381 | } | |
382 | ||
383 | /* Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This cannot be | |
384 | * allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under any circumstances. | |
385 | */ | |
386 | void dccp_send_close(struct sock *sk) | |
387 | { | |
388 | struct dccp_sock *dp = dccp_sk(sk); | |
389 | struct sk_buff *skb; | |
390 | ||
391 | /* Socket is locked, keep trying until memory is available. */ | |
392 | for (;;) { | |
393 | skb = alloc_skb(sk->sk_prot->max_header, GFP_KERNEL); | |
394 | if (skb != NULL) | |
395 | break; | |
396 | yield(); | |
397 | } | |
398 | ||
399 | /* Reserve space for headers and prepare control bits. */ | |
400 | skb_reserve(skb, sk->sk_prot->max_header); | |
401 | skb->csum = 0; | |
402 | DCCP_SKB_CB(skb)->dccpd_type = dp->dccps_role == DCCP_ROLE_CLIENT ? DCCP_PKT_CLOSE : DCCP_PKT_CLOSEREQ; | |
403 | ||
404 | skb_set_owner_w(skb, sk); | |
405 | dccp_transmit_skb(sk, skb); | |
406 | ||
407 | ccid_hc_rx_exit(dp->dccps_hc_rx_ccid, sk); | |
408 | ccid_hc_tx_exit(dp->dccps_hc_tx_ccid, sk); | |
409 | } |