xref: /openbmc/linux/net/dccp/proto.c (revision 227a0655)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  net/dccp/proto.c
4  *
5  *  An implementation of the DCCP protocol
6  *  Arnaldo Carvalho de Melo <acme@conectiva.com.br>
7  */
8 
9 #include <linux/dccp.h>
10 #include <linux/module.h>
11 #include <linux/types.h>
12 #include <linux/sched.h>
13 #include <linux/kernel.h>
14 #include <linux/skbuff.h>
15 #include <linux/netdevice.h>
16 #include <linux/in.h>
17 #include <linux/if_arp.h>
18 #include <linux/init.h>
19 #include <linux/random.h>
20 #include <linux/slab.h>
21 #include <net/checksum.h>
22 
23 #include <net/inet_sock.h>
24 #include <net/inet_common.h>
25 #include <net/sock.h>
26 #include <net/xfrm.h>
27 
28 #include <asm/ioctls.h>
29 #include <linux/spinlock.h>
30 #include <linux/timer.h>
31 #include <linux/delay.h>
32 #include <linux/poll.h>
33 
34 #include "ccid.h"
35 #include "dccp.h"
36 #include "feat.h"
37 
38 #define CREATE_TRACE_POINTS
39 #include "trace.h"
40 
41 DEFINE_SNMP_STAT(struct dccp_mib, dccp_statistics) __read_mostly;
42 
43 EXPORT_SYMBOL_GPL(dccp_statistics);
44 
45 DEFINE_PER_CPU(unsigned int, dccp_orphan_count);
46 EXPORT_PER_CPU_SYMBOL_GPL(dccp_orphan_count);
47 
48 struct inet_hashinfo dccp_hashinfo;
49 EXPORT_SYMBOL_GPL(dccp_hashinfo);
50 
51 /* the maximum queue length for tx in packets. 0 is no limit */
52 int sysctl_dccp_tx_qlen __read_mostly = 5;
53 
54 #ifdef CONFIG_IP_DCCP_DEBUG
55 static const char *dccp_state_name(const int state)
56 {
57 	static const char *const dccp_state_names[] = {
58 	[DCCP_OPEN]		= "OPEN",
59 	[DCCP_REQUESTING]	= "REQUESTING",
60 	[DCCP_PARTOPEN]		= "PARTOPEN",
61 	[DCCP_LISTEN]		= "LISTEN",
62 	[DCCP_RESPOND]		= "RESPOND",
63 	[DCCP_CLOSING]		= "CLOSING",
64 	[DCCP_ACTIVE_CLOSEREQ]	= "CLOSEREQ",
65 	[DCCP_PASSIVE_CLOSE]	= "PASSIVE_CLOSE",
66 	[DCCP_PASSIVE_CLOSEREQ]	= "PASSIVE_CLOSEREQ",
67 	[DCCP_TIME_WAIT]	= "TIME_WAIT",
68 	[DCCP_CLOSED]		= "CLOSED",
69 	};
70 
71 	if (state >= DCCP_MAX_STATES)
72 		return "INVALID STATE!";
73 	else
74 		return dccp_state_names[state];
75 }
76 #endif
77 
78 void dccp_set_state(struct sock *sk, const int state)
79 {
80 	const int oldstate = sk->sk_state;
81 
82 	dccp_pr_debug("%s(%p)  %s  -->  %s\n", dccp_role(sk), sk,
83 		      dccp_state_name(oldstate), dccp_state_name(state));
84 	WARN_ON(state == oldstate);
85 
86 	switch (state) {
87 	case DCCP_OPEN:
88 		if (oldstate != DCCP_OPEN)
89 			DCCP_INC_STATS(DCCP_MIB_CURRESTAB);
90 		/* Client retransmits all Confirm options until entering OPEN */
91 		if (oldstate == DCCP_PARTOPEN)
92 			dccp_feat_list_purge(&dccp_sk(sk)->dccps_featneg);
93 		break;
94 
95 	case DCCP_CLOSED:
96 		if (oldstate == DCCP_OPEN || oldstate == DCCP_ACTIVE_CLOSEREQ ||
97 		    oldstate == DCCP_CLOSING)
98 			DCCP_INC_STATS(DCCP_MIB_ESTABRESETS);
99 
100 		sk->sk_prot->unhash(sk);
101 		if (inet_csk(sk)->icsk_bind_hash != NULL &&
102 		    !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
103 			inet_put_port(sk);
104 		fallthrough;
105 	default:
106 		if (oldstate == DCCP_OPEN)
107 			DCCP_DEC_STATS(DCCP_MIB_CURRESTAB);
108 	}
109 
110 	/* Change state AFTER socket is unhashed to avoid closed
111 	 * socket sitting in hash tables.
112 	 */
113 	inet_sk_set_state(sk, state);
114 }
115 
116 EXPORT_SYMBOL_GPL(dccp_set_state);
117 
118 static void dccp_finish_passive_close(struct sock *sk)
119 {
120 	switch (sk->sk_state) {
121 	case DCCP_PASSIVE_CLOSE:
122 		/* Node (client or server) has received Close packet. */
123 		dccp_send_reset(sk, DCCP_RESET_CODE_CLOSED);
124 		dccp_set_state(sk, DCCP_CLOSED);
125 		break;
126 	case DCCP_PASSIVE_CLOSEREQ:
127 		/*
128 		 * Client received CloseReq. We set the `active' flag so that
129 		 * dccp_send_close() retransmits the Close as per RFC 4340, 8.3.
130 		 */
131 		dccp_send_close(sk, 1);
132 		dccp_set_state(sk, DCCP_CLOSING);
133 	}
134 }
135 
136 void dccp_done(struct sock *sk)
137 {
138 	dccp_set_state(sk, DCCP_CLOSED);
139 	dccp_clear_xmit_timers(sk);
140 
141 	sk->sk_shutdown = SHUTDOWN_MASK;
142 
143 	if (!sock_flag(sk, SOCK_DEAD))
144 		sk->sk_state_change(sk);
145 	else
146 		inet_csk_destroy_sock(sk);
147 }
148 
149 EXPORT_SYMBOL_GPL(dccp_done);
150 
151 const char *dccp_packet_name(const int type)
152 {
153 	static const char *const dccp_packet_names[] = {
154 		[DCCP_PKT_REQUEST]  = "REQUEST",
155 		[DCCP_PKT_RESPONSE] = "RESPONSE",
156 		[DCCP_PKT_DATA]	    = "DATA",
157 		[DCCP_PKT_ACK]	    = "ACK",
158 		[DCCP_PKT_DATAACK]  = "DATAACK",
159 		[DCCP_PKT_CLOSEREQ] = "CLOSEREQ",
160 		[DCCP_PKT_CLOSE]    = "CLOSE",
161 		[DCCP_PKT_RESET]    = "RESET",
162 		[DCCP_PKT_SYNC]	    = "SYNC",
163 		[DCCP_PKT_SYNCACK]  = "SYNCACK",
164 	};
165 
166 	if (type >= DCCP_NR_PKT_TYPES)
167 		return "INVALID";
168 	else
169 		return dccp_packet_names[type];
170 }
171 
172 EXPORT_SYMBOL_GPL(dccp_packet_name);
173 
174 static void dccp_sk_destruct(struct sock *sk)
175 {
176 	struct dccp_sock *dp = dccp_sk(sk);
177 
178 	ccid_hc_tx_delete(dp->dccps_hc_tx_ccid, sk);
179 	dp->dccps_hc_tx_ccid = NULL;
180 	inet_sock_destruct(sk);
181 }
182 
183 int dccp_init_sock(struct sock *sk, const __u8 ctl_sock_initialized)
184 {
185 	struct dccp_sock *dp = dccp_sk(sk);
186 	struct inet_connection_sock *icsk = inet_csk(sk);
187 
188 	icsk->icsk_rto		= DCCP_TIMEOUT_INIT;
189 	icsk->icsk_syn_retries	= sysctl_dccp_request_retries;
190 	sk->sk_state		= DCCP_CLOSED;
191 	sk->sk_write_space	= dccp_write_space;
192 	sk->sk_destruct		= dccp_sk_destruct;
193 	icsk->icsk_sync_mss	= dccp_sync_mss;
194 	dp->dccps_mss_cache	= 536;
195 	dp->dccps_rate_last	= jiffies;
196 	dp->dccps_role		= DCCP_ROLE_UNDEFINED;
197 	dp->dccps_service	= DCCP_SERVICE_CODE_IS_ABSENT;
198 	dp->dccps_tx_qlen	= sysctl_dccp_tx_qlen;
199 
200 	dccp_init_xmit_timers(sk);
201 
202 	INIT_LIST_HEAD(&dp->dccps_featneg);
203 	/* control socket doesn't need feat nego */
204 	if (likely(ctl_sock_initialized))
205 		return dccp_feat_init(sk);
206 	return 0;
207 }
208 
209 EXPORT_SYMBOL_GPL(dccp_init_sock);
210 
211 void dccp_destroy_sock(struct sock *sk)
212 {
213 	struct dccp_sock *dp = dccp_sk(sk);
214 
215 	__skb_queue_purge(&sk->sk_write_queue);
216 	if (sk->sk_send_head != NULL) {
217 		kfree_skb(sk->sk_send_head);
218 		sk->sk_send_head = NULL;
219 	}
220 
221 	/* Clean up a referenced DCCP bind bucket. */
222 	if (inet_csk(sk)->icsk_bind_hash != NULL)
223 		inet_put_port(sk);
224 
225 	kfree(dp->dccps_service_list);
226 	dp->dccps_service_list = NULL;
227 
228 	if (dp->dccps_hc_rx_ackvec != NULL) {
229 		dccp_ackvec_free(dp->dccps_hc_rx_ackvec);
230 		dp->dccps_hc_rx_ackvec = NULL;
231 	}
232 	ccid_hc_rx_delete(dp->dccps_hc_rx_ccid, sk);
233 	dp->dccps_hc_rx_ccid = NULL;
234 
235 	/* clean up feature negotiation state */
236 	dccp_feat_list_purge(&dp->dccps_featneg);
237 }
238 
239 EXPORT_SYMBOL_GPL(dccp_destroy_sock);
240 
241 static inline int dccp_need_reset(int state)
242 {
243 	return state != DCCP_CLOSED && state != DCCP_LISTEN &&
244 	       state != DCCP_REQUESTING;
245 }
246 
247 int dccp_disconnect(struct sock *sk, int flags)
248 {
249 	struct inet_connection_sock *icsk = inet_csk(sk);
250 	struct inet_sock *inet = inet_sk(sk);
251 	struct dccp_sock *dp = dccp_sk(sk);
252 	const int old_state = sk->sk_state;
253 
254 	if (old_state != DCCP_CLOSED)
255 		dccp_set_state(sk, DCCP_CLOSED);
256 
257 	/*
258 	 * This corresponds to the ABORT function of RFC793, sec. 3.8
259 	 * TCP uses a RST segment, DCCP a Reset packet with Code 2, "Aborted".
260 	 */
261 	if (old_state == DCCP_LISTEN) {
262 		inet_csk_listen_stop(sk);
263 	} else if (dccp_need_reset(old_state)) {
264 		dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED);
265 		sk->sk_err = ECONNRESET;
266 	} else if (old_state == DCCP_REQUESTING)
267 		sk->sk_err = ECONNRESET;
268 
269 	dccp_clear_xmit_timers(sk);
270 	ccid_hc_rx_delete(dp->dccps_hc_rx_ccid, sk);
271 	dp->dccps_hc_rx_ccid = NULL;
272 
273 	__skb_queue_purge(&sk->sk_receive_queue);
274 	__skb_queue_purge(&sk->sk_write_queue);
275 	if (sk->sk_send_head != NULL) {
276 		__kfree_skb(sk->sk_send_head);
277 		sk->sk_send_head = NULL;
278 	}
279 
280 	inet->inet_dport = 0;
281 
282 	if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
283 		inet_reset_saddr(sk);
284 
285 	sk->sk_shutdown = 0;
286 	sock_reset_flag(sk, SOCK_DONE);
287 
288 	icsk->icsk_backoff = 0;
289 	inet_csk_delack_init(sk);
290 	__sk_dst_reset(sk);
291 
292 	WARN_ON(inet->inet_num && !icsk->icsk_bind_hash);
293 
294 	sk_error_report(sk);
295 	return 0;
296 }
297 
298 EXPORT_SYMBOL_GPL(dccp_disconnect);
299 
300 /*
301  *	Wait for a DCCP event.
302  *
303  *	Note that we don't need to lock the socket, as the upper poll layers
304  *	take care of normal races (between the test and the event) and we don't
305  *	go look at any of the socket buffers directly.
306  */
307 __poll_t dccp_poll(struct file *file, struct socket *sock,
308 		       poll_table *wait)
309 {
310 	__poll_t mask;
311 	struct sock *sk = sock->sk;
312 
313 	sock_poll_wait(file, sock, wait);
314 	if (sk->sk_state == DCCP_LISTEN)
315 		return inet_csk_listen_poll(sk);
316 
317 	/* Socket is not locked. We are protected from async events
318 	   by poll logic and correct handling of state changes
319 	   made by another threads is impossible in any case.
320 	 */
321 
322 	mask = 0;
323 	if (sk->sk_err)
324 		mask = EPOLLERR;
325 
326 	if (sk->sk_shutdown == SHUTDOWN_MASK || sk->sk_state == DCCP_CLOSED)
327 		mask |= EPOLLHUP;
328 	if (sk->sk_shutdown & RCV_SHUTDOWN)
329 		mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
330 
331 	/* Connected? */
332 	if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_RESPOND)) {
333 		if (atomic_read(&sk->sk_rmem_alloc) > 0)
334 			mask |= EPOLLIN | EPOLLRDNORM;
335 
336 		if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
337 			if (sk_stream_is_writeable(sk)) {
338 				mask |= EPOLLOUT | EPOLLWRNORM;
339 			} else {  /* send SIGIO later */
340 				sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
341 				set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
342 
343 				/* Race breaker. If space is freed after
344 				 * wspace test but before the flags are set,
345 				 * IO signal will be lost.
346 				 */
347 				if (sk_stream_is_writeable(sk))
348 					mask |= EPOLLOUT | EPOLLWRNORM;
349 			}
350 		}
351 	}
352 	return mask;
353 }
354 
355 EXPORT_SYMBOL_GPL(dccp_poll);
356 
357 int dccp_ioctl(struct sock *sk, int cmd, unsigned long arg)
358 {
359 	int rc = -ENOTCONN;
360 
361 	lock_sock(sk);
362 
363 	if (sk->sk_state == DCCP_LISTEN)
364 		goto out;
365 
366 	switch (cmd) {
367 	case SIOCOUTQ: {
368 		int amount = sk_wmem_alloc_get(sk);
369 		/* Using sk_wmem_alloc here because sk_wmem_queued is not used by DCCP and
370 		 * always 0, comparably to UDP.
371 		 */
372 
373 		rc = put_user(amount, (int __user *)arg);
374 	}
375 		break;
376 	case SIOCINQ: {
377 		struct sk_buff *skb;
378 		unsigned long amount = 0;
379 
380 		skb = skb_peek(&sk->sk_receive_queue);
381 		if (skb != NULL) {
382 			/*
383 			 * We will only return the amount of this packet since
384 			 * that is all that will be read.
385 			 */
386 			amount = skb->len;
387 		}
388 		rc = put_user(amount, (int __user *)arg);
389 	}
390 		break;
391 	default:
392 		rc = -ENOIOCTLCMD;
393 		break;
394 	}
395 out:
396 	release_sock(sk);
397 	return rc;
398 }
399 
400 EXPORT_SYMBOL_GPL(dccp_ioctl);
401 
402 static int dccp_setsockopt_service(struct sock *sk, const __be32 service,
403 				   sockptr_t optval, unsigned int optlen)
404 {
405 	struct dccp_sock *dp = dccp_sk(sk);
406 	struct dccp_service_list *sl = NULL;
407 
408 	if (service == DCCP_SERVICE_INVALID_VALUE ||
409 	    optlen > DCCP_SERVICE_LIST_MAX_LEN * sizeof(u32))
410 		return -EINVAL;
411 
412 	if (optlen > sizeof(service)) {
413 		sl = kmalloc(optlen, GFP_KERNEL);
414 		if (sl == NULL)
415 			return -ENOMEM;
416 
417 		sl->dccpsl_nr = optlen / sizeof(u32) - 1;
418 		if (copy_from_sockptr_offset(sl->dccpsl_list, optval,
419 				sizeof(service), optlen - sizeof(service)) ||
420 		    dccp_list_has_service(sl, DCCP_SERVICE_INVALID_VALUE)) {
421 			kfree(sl);
422 			return -EFAULT;
423 		}
424 	}
425 
426 	lock_sock(sk);
427 	dp->dccps_service = service;
428 
429 	kfree(dp->dccps_service_list);
430 
431 	dp->dccps_service_list = sl;
432 	release_sock(sk);
433 	return 0;
434 }
435 
436 static int dccp_setsockopt_cscov(struct sock *sk, int cscov, bool rx)
437 {
438 	u8 *list, len;
439 	int i, rc;
440 
441 	if (cscov < 0 || cscov > 15)
442 		return -EINVAL;
443 	/*
444 	 * Populate a list of permissible values, in the range cscov...15. This
445 	 * is necessary since feature negotiation of single values only works if
446 	 * both sides incidentally choose the same value. Since the list starts
447 	 * lowest-value first, negotiation will pick the smallest shared value.
448 	 */
449 	if (cscov == 0)
450 		return 0;
451 	len = 16 - cscov;
452 
453 	list = kmalloc(len, GFP_KERNEL);
454 	if (list == NULL)
455 		return -ENOBUFS;
456 
457 	for (i = 0; i < len; i++)
458 		list[i] = cscov++;
459 
460 	rc = dccp_feat_register_sp(sk, DCCPF_MIN_CSUM_COVER, rx, list, len);
461 
462 	if (rc == 0) {
463 		if (rx)
464 			dccp_sk(sk)->dccps_pcrlen = cscov;
465 		else
466 			dccp_sk(sk)->dccps_pcslen = cscov;
467 	}
468 	kfree(list);
469 	return rc;
470 }
471 
472 static int dccp_setsockopt_ccid(struct sock *sk, int type,
473 				sockptr_t optval, unsigned int optlen)
474 {
475 	u8 *val;
476 	int rc = 0;
477 
478 	if (optlen < 1 || optlen > DCCP_FEAT_MAX_SP_VALS)
479 		return -EINVAL;
480 
481 	val = memdup_sockptr(optval, optlen);
482 	if (IS_ERR(val))
483 		return PTR_ERR(val);
484 
485 	lock_sock(sk);
486 	if (type == DCCP_SOCKOPT_TX_CCID || type == DCCP_SOCKOPT_CCID)
487 		rc = dccp_feat_register_sp(sk, DCCPF_CCID, 1, val, optlen);
488 
489 	if (!rc && (type == DCCP_SOCKOPT_RX_CCID || type == DCCP_SOCKOPT_CCID))
490 		rc = dccp_feat_register_sp(sk, DCCPF_CCID, 0, val, optlen);
491 	release_sock(sk);
492 
493 	kfree(val);
494 	return rc;
495 }
496 
497 static int do_dccp_setsockopt(struct sock *sk, int level, int optname,
498 		sockptr_t optval, unsigned int optlen)
499 {
500 	struct dccp_sock *dp = dccp_sk(sk);
501 	int val, err = 0;
502 
503 	switch (optname) {
504 	case DCCP_SOCKOPT_PACKET_SIZE:
505 		DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n");
506 		return 0;
507 	case DCCP_SOCKOPT_CHANGE_L:
508 	case DCCP_SOCKOPT_CHANGE_R:
509 		DCCP_WARN("sockopt(CHANGE_L/R) is deprecated: fix your app\n");
510 		return 0;
511 	case DCCP_SOCKOPT_CCID:
512 	case DCCP_SOCKOPT_RX_CCID:
513 	case DCCP_SOCKOPT_TX_CCID:
514 		return dccp_setsockopt_ccid(sk, optname, optval, optlen);
515 	}
516 
517 	if (optlen < (int)sizeof(int))
518 		return -EINVAL;
519 
520 	if (copy_from_sockptr(&val, optval, sizeof(int)))
521 		return -EFAULT;
522 
523 	if (optname == DCCP_SOCKOPT_SERVICE)
524 		return dccp_setsockopt_service(sk, val, optval, optlen);
525 
526 	lock_sock(sk);
527 	switch (optname) {
528 	case DCCP_SOCKOPT_SERVER_TIMEWAIT:
529 		if (dp->dccps_role != DCCP_ROLE_SERVER)
530 			err = -EOPNOTSUPP;
531 		else
532 			dp->dccps_server_timewait = (val != 0);
533 		break;
534 	case DCCP_SOCKOPT_SEND_CSCOV:
535 		err = dccp_setsockopt_cscov(sk, val, false);
536 		break;
537 	case DCCP_SOCKOPT_RECV_CSCOV:
538 		err = dccp_setsockopt_cscov(sk, val, true);
539 		break;
540 	case DCCP_SOCKOPT_QPOLICY_ID:
541 		if (sk->sk_state != DCCP_CLOSED)
542 			err = -EISCONN;
543 		else if (val < 0 || val >= DCCPQ_POLICY_MAX)
544 			err = -EINVAL;
545 		else
546 			dp->dccps_qpolicy = val;
547 		break;
548 	case DCCP_SOCKOPT_QPOLICY_TXQLEN:
549 		if (val < 0)
550 			err = -EINVAL;
551 		else
552 			dp->dccps_tx_qlen = val;
553 		break;
554 	default:
555 		err = -ENOPROTOOPT;
556 		break;
557 	}
558 	release_sock(sk);
559 
560 	return err;
561 }
562 
563 int dccp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
564 		    unsigned int optlen)
565 {
566 	if (level != SOL_DCCP)
567 		return inet_csk(sk)->icsk_af_ops->setsockopt(sk, level,
568 							     optname, optval,
569 							     optlen);
570 	return do_dccp_setsockopt(sk, level, optname, optval, optlen);
571 }
572 
573 EXPORT_SYMBOL_GPL(dccp_setsockopt);
574 
575 static int dccp_getsockopt_service(struct sock *sk, int len,
576 				   __be32 __user *optval,
577 				   int __user *optlen)
578 {
579 	const struct dccp_sock *dp = dccp_sk(sk);
580 	const struct dccp_service_list *sl;
581 	int err = -ENOENT, slen = 0, total_len = sizeof(u32);
582 
583 	lock_sock(sk);
584 	if ((sl = dp->dccps_service_list) != NULL) {
585 		slen = sl->dccpsl_nr * sizeof(u32);
586 		total_len += slen;
587 	}
588 
589 	err = -EINVAL;
590 	if (total_len > len)
591 		goto out;
592 
593 	err = 0;
594 	if (put_user(total_len, optlen) ||
595 	    put_user(dp->dccps_service, optval) ||
596 	    (sl != NULL && copy_to_user(optval + 1, sl->dccpsl_list, slen)))
597 		err = -EFAULT;
598 out:
599 	release_sock(sk);
600 	return err;
601 }
602 
603 static int do_dccp_getsockopt(struct sock *sk, int level, int optname,
604 		    char __user *optval, int __user *optlen)
605 {
606 	struct dccp_sock *dp;
607 	int val, len;
608 
609 	if (get_user(len, optlen))
610 		return -EFAULT;
611 
612 	if (len < (int)sizeof(int))
613 		return -EINVAL;
614 
615 	dp = dccp_sk(sk);
616 
617 	switch (optname) {
618 	case DCCP_SOCKOPT_PACKET_SIZE:
619 		DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n");
620 		return 0;
621 	case DCCP_SOCKOPT_SERVICE:
622 		return dccp_getsockopt_service(sk, len,
623 					       (__be32 __user *)optval, optlen);
624 	case DCCP_SOCKOPT_GET_CUR_MPS:
625 		val = dp->dccps_mss_cache;
626 		break;
627 	case DCCP_SOCKOPT_AVAILABLE_CCIDS:
628 		return ccid_getsockopt_builtin_ccids(sk, len, optval, optlen);
629 	case DCCP_SOCKOPT_TX_CCID:
630 		val = ccid_get_current_tx_ccid(dp);
631 		if (val < 0)
632 			return -ENOPROTOOPT;
633 		break;
634 	case DCCP_SOCKOPT_RX_CCID:
635 		val = ccid_get_current_rx_ccid(dp);
636 		if (val < 0)
637 			return -ENOPROTOOPT;
638 		break;
639 	case DCCP_SOCKOPT_SERVER_TIMEWAIT:
640 		val = dp->dccps_server_timewait;
641 		break;
642 	case DCCP_SOCKOPT_SEND_CSCOV:
643 		val = dp->dccps_pcslen;
644 		break;
645 	case DCCP_SOCKOPT_RECV_CSCOV:
646 		val = dp->dccps_pcrlen;
647 		break;
648 	case DCCP_SOCKOPT_QPOLICY_ID:
649 		val = dp->dccps_qpolicy;
650 		break;
651 	case DCCP_SOCKOPT_QPOLICY_TXQLEN:
652 		val = dp->dccps_tx_qlen;
653 		break;
654 	case 128 ... 191:
655 		return ccid_hc_rx_getsockopt(dp->dccps_hc_rx_ccid, sk, optname,
656 					     len, (u32 __user *)optval, optlen);
657 	case 192 ... 255:
658 		return ccid_hc_tx_getsockopt(dp->dccps_hc_tx_ccid, sk, optname,
659 					     len, (u32 __user *)optval, optlen);
660 	default:
661 		return -ENOPROTOOPT;
662 	}
663 
664 	len = sizeof(val);
665 	if (put_user(len, optlen) || copy_to_user(optval, &val, len))
666 		return -EFAULT;
667 
668 	return 0;
669 }
670 
671 int dccp_getsockopt(struct sock *sk, int level, int optname,
672 		    char __user *optval, int __user *optlen)
673 {
674 	if (level != SOL_DCCP)
675 		return inet_csk(sk)->icsk_af_ops->getsockopt(sk, level,
676 							     optname, optval,
677 							     optlen);
678 	return do_dccp_getsockopt(sk, level, optname, optval, optlen);
679 }
680 
681 EXPORT_SYMBOL_GPL(dccp_getsockopt);
682 
683 static int dccp_msghdr_parse(struct msghdr *msg, struct sk_buff *skb)
684 {
685 	struct cmsghdr *cmsg;
686 
687 	/*
688 	 * Assign an (opaque) qpolicy priority value to skb->priority.
689 	 *
690 	 * We are overloading this skb field for use with the qpolicy subystem.
691 	 * The skb->priority is normally used for the SO_PRIORITY option, which
692 	 * is initialised from sk_priority. Since the assignment of sk_priority
693 	 * to skb->priority happens later (on layer 3), we overload this field
694 	 * for use with queueing priorities as long as the skb is on layer 4.
695 	 * The default priority value (if nothing is set) is 0.
696 	 */
697 	skb->priority = 0;
698 
699 	for_each_cmsghdr(cmsg, msg) {
700 		if (!CMSG_OK(msg, cmsg))
701 			return -EINVAL;
702 
703 		if (cmsg->cmsg_level != SOL_DCCP)
704 			continue;
705 
706 		if (cmsg->cmsg_type <= DCCP_SCM_QPOLICY_MAX &&
707 		    !dccp_qpolicy_param_ok(skb->sk, cmsg->cmsg_type))
708 			return -EINVAL;
709 
710 		switch (cmsg->cmsg_type) {
711 		case DCCP_SCM_PRIORITY:
712 			if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u32)))
713 				return -EINVAL;
714 			skb->priority = *(__u32 *)CMSG_DATA(cmsg);
715 			break;
716 		default:
717 			return -EINVAL;
718 		}
719 	}
720 	return 0;
721 }
722 
723 int dccp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
724 {
725 	const struct dccp_sock *dp = dccp_sk(sk);
726 	const int flags = msg->msg_flags;
727 	const int noblock = flags & MSG_DONTWAIT;
728 	struct sk_buff *skb;
729 	int rc, size;
730 	long timeo;
731 
732 	trace_dccp_probe(sk, len);
733 
734 	if (len > dp->dccps_mss_cache)
735 		return -EMSGSIZE;
736 
737 	lock_sock(sk);
738 
739 	if (dccp_qpolicy_full(sk)) {
740 		rc = -EAGAIN;
741 		goto out_release;
742 	}
743 
744 	timeo = sock_sndtimeo(sk, noblock);
745 
746 	/*
747 	 * We have to use sk_stream_wait_connect here to set sk_write_pending,
748 	 * so that the trick in dccp_rcv_request_sent_state_process.
749 	 */
750 	/* Wait for a connection to finish. */
751 	if ((1 << sk->sk_state) & ~(DCCPF_OPEN | DCCPF_PARTOPEN))
752 		if ((rc = sk_stream_wait_connect(sk, &timeo)) != 0)
753 			goto out_release;
754 
755 	size = sk->sk_prot->max_header + len;
756 	release_sock(sk);
757 	skb = sock_alloc_send_skb(sk, size, noblock, &rc);
758 	lock_sock(sk);
759 	if (skb == NULL)
760 		goto out_release;
761 
762 	if (sk->sk_state == DCCP_CLOSED) {
763 		rc = -ENOTCONN;
764 		goto out_discard;
765 	}
766 
767 	skb_reserve(skb, sk->sk_prot->max_header);
768 	rc = memcpy_from_msg(skb_put(skb, len), msg, len);
769 	if (rc != 0)
770 		goto out_discard;
771 
772 	rc = dccp_msghdr_parse(msg, skb);
773 	if (rc != 0)
774 		goto out_discard;
775 
776 	dccp_qpolicy_push(sk, skb);
777 	/*
778 	 * The xmit_timer is set if the TX CCID is rate-based and will expire
779 	 * when congestion control permits to release further packets into the
780 	 * network. Window-based CCIDs do not use this timer.
781 	 */
782 	if (!timer_pending(&dp->dccps_xmit_timer))
783 		dccp_write_xmit(sk);
784 out_release:
785 	release_sock(sk);
786 	return rc ? : len;
787 out_discard:
788 	kfree_skb(skb);
789 	goto out_release;
790 }
791 
792 EXPORT_SYMBOL_GPL(dccp_sendmsg);
793 
794 int dccp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
795 		 int flags, int *addr_len)
796 {
797 	const struct dccp_hdr *dh;
798 	long timeo;
799 
800 	lock_sock(sk);
801 
802 	if (sk->sk_state == DCCP_LISTEN) {
803 		len = -ENOTCONN;
804 		goto out;
805 	}
806 
807 	timeo = sock_rcvtimeo(sk, nonblock);
808 
809 	do {
810 		struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
811 
812 		if (skb == NULL)
813 			goto verify_sock_status;
814 
815 		dh = dccp_hdr(skb);
816 
817 		switch (dh->dccph_type) {
818 		case DCCP_PKT_DATA:
819 		case DCCP_PKT_DATAACK:
820 			goto found_ok_skb;
821 
822 		case DCCP_PKT_CLOSE:
823 		case DCCP_PKT_CLOSEREQ:
824 			if (!(flags & MSG_PEEK))
825 				dccp_finish_passive_close(sk);
826 			fallthrough;
827 		case DCCP_PKT_RESET:
828 			dccp_pr_debug("found fin (%s) ok!\n",
829 				      dccp_packet_name(dh->dccph_type));
830 			len = 0;
831 			goto found_fin_ok;
832 		default:
833 			dccp_pr_debug("packet_type=%s\n",
834 				      dccp_packet_name(dh->dccph_type));
835 			sk_eat_skb(sk, skb);
836 		}
837 verify_sock_status:
838 		if (sock_flag(sk, SOCK_DONE)) {
839 			len = 0;
840 			break;
841 		}
842 
843 		if (sk->sk_err) {
844 			len = sock_error(sk);
845 			break;
846 		}
847 
848 		if (sk->sk_shutdown & RCV_SHUTDOWN) {
849 			len = 0;
850 			break;
851 		}
852 
853 		if (sk->sk_state == DCCP_CLOSED) {
854 			if (!sock_flag(sk, SOCK_DONE)) {
855 				/* This occurs when user tries to read
856 				 * from never connected socket.
857 				 */
858 				len = -ENOTCONN;
859 				break;
860 			}
861 			len = 0;
862 			break;
863 		}
864 
865 		if (!timeo) {
866 			len = -EAGAIN;
867 			break;
868 		}
869 
870 		if (signal_pending(current)) {
871 			len = sock_intr_errno(timeo);
872 			break;
873 		}
874 
875 		sk_wait_data(sk, &timeo, NULL);
876 		continue;
877 	found_ok_skb:
878 		if (len > skb->len)
879 			len = skb->len;
880 		else if (len < skb->len)
881 			msg->msg_flags |= MSG_TRUNC;
882 
883 		if (skb_copy_datagram_msg(skb, 0, msg, len)) {
884 			/* Exception. Bailout! */
885 			len = -EFAULT;
886 			break;
887 		}
888 		if (flags & MSG_TRUNC)
889 			len = skb->len;
890 	found_fin_ok:
891 		if (!(flags & MSG_PEEK))
892 			sk_eat_skb(sk, skb);
893 		break;
894 	} while (1);
895 out:
896 	release_sock(sk);
897 	return len;
898 }
899 
900 EXPORT_SYMBOL_GPL(dccp_recvmsg);
901 
902 int inet_dccp_listen(struct socket *sock, int backlog)
903 {
904 	struct sock *sk = sock->sk;
905 	unsigned char old_state;
906 	int err;
907 
908 	lock_sock(sk);
909 
910 	err = -EINVAL;
911 	if (sock->state != SS_UNCONNECTED || sock->type != SOCK_DCCP)
912 		goto out;
913 
914 	old_state = sk->sk_state;
915 	if (!((1 << old_state) & (DCCPF_CLOSED | DCCPF_LISTEN)))
916 		goto out;
917 
918 	WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
919 	/* Really, if the socket is already in listen state
920 	 * we can only allow the backlog to be adjusted.
921 	 */
922 	if (old_state != DCCP_LISTEN) {
923 		struct dccp_sock *dp = dccp_sk(sk);
924 
925 		dp->dccps_role = DCCP_ROLE_LISTEN;
926 
927 		/* do not start to listen if feature negotiation setup fails */
928 		if (dccp_feat_finalise_settings(dp)) {
929 			err = -EPROTO;
930 			goto out;
931 		}
932 
933 		err = inet_csk_listen_start(sk);
934 		if (err)
935 			goto out;
936 	}
937 	err = 0;
938 
939 out:
940 	release_sock(sk);
941 	return err;
942 }
943 
944 EXPORT_SYMBOL_GPL(inet_dccp_listen);
945 
946 static void dccp_terminate_connection(struct sock *sk)
947 {
948 	u8 next_state = DCCP_CLOSED;
949 
950 	switch (sk->sk_state) {
951 	case DCCP_PASSIVE_CLOSE:
952 	case DCCP_PASSIVE_CLOSEREQ:
953 		dccp_finish_passive_close(sk);
954 		break;
955 	case DCCP_PARTOPEN:
956 		dccp_pr_debug("Stop PARTOPEN timer (%p)\n", sk);
957 		inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
958 		fallthrough;
959 	case DCCP_OPEN:
960 		dccp_send_close(sk, 1);
961 
962 		if (dccp_sk(sk)->dccps_role == DCCP_ROLE_SERVER &&
963 		    !dccp_sk(sk)->dccps_server_timewait)
964 			next_state = DCCP_ACTIVE_CLOSEREQ;
965 		else
966 			next_state = DCCP_CLOSING;
967 		fallthrough;
968 	default:
969 		dccp_set_state(sk, next_state);
970 	}
971 }
972 
973 void dccp_close(struct sock *sk, long timeout)
974 {
975 	struct dccp_sock *dp = dccp_sk(sk);
976 	struct sk_buff *skb;
977 	u32 data_was_unread = 0;
978 	int state;
979 
980 	lock_sock(sk);
981 
982 	sk->sk_shutdown = SHUTDOWN_MASK;
983 
984 	if (sk->sk_state == DCCP_LISTEN) {
985 		dccp_set_state(sk, DCCP_CLOSED);
986 
987 		/* Special case. */
988 		inet_csk_listen_stop(sk);
989 
990 		goto adjudge_to_death;
991 	}
992 
993 	sk_stop_timer(sk, &dp->dccps_xmit_timer);
994 
995 	/*
996 	 * We need to flush the recv. buffs.  We do this only on the
997 	 * descriptor close, not protocol-sourced closes, because the
998 	  *reader process may not have drained the data yet!
999 	 */
1000 	while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
1001 		data_was_unread += skb->len;
1002 		__kfree_skb(skb);
1003 	}
1004 
1005 	/* If socket has been already reset kill it. */
1006 	if (sk->sk_state == DCCP_CLOSED)
1007 		goto adjudge_to_death;
1008 
1009 	if (data_was_unread) {
1010 		/* Unread data was tossed, send an appropriate Reset Code */
1011 		DCCP_WARN("ABORT with %u bytes unread\n", data_was_unread);
1012 		dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED);
1013 		dccp_set_state(sk, DCCP_CLOSED);
1014 	} else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
1015 		/* Check zero linger _after_ checking for unread data. */
1016 		sk->sk_prot->disconnect(sk, 0);
1017 	} else if (sk->sk_state != DCCP_CLOSED) {
1018 		/*
1019 		 * Normal connection termination. May need to wait if there are
1020 		 * still packets in the TX queue that are delayed by the CCID.
1021 		 */
1022 		dccp_flush_write_queue(sk, &timeout);
1023 		dccp_terminate_connection(sk);
1024 	}
1025 
1026 	/*
1027 	 * Flush write queue. This may be necessary in several cases:
1028 	 * - we have been closed by the peer but still have application data;
1029 	 * - abortive termination (unread data or zero linger time),
1030 	 * - normal termination but queue could not be flushed within time limit
1031 	 */
1032 	__skb_queue_purge(&sk->sk_write_queue);
1033 
1034 	sk_stream_wait_close(sk, timeout);
1035 
1036 adjudge_to_death:
1037 	state = sk->sk_state;
1038 	sock_hold(sk);
1039 	sock_orphan(sk);
1040 
1041 	/*
1042 	 * It is the last release_sock in its life. It will remove backlog.
1043 	 */
1044 	release_sock(sk);
1045 	/*
1046 	 * Now socket is owned by kernel and we acquire BH lock
1047 	 * to finish close. No need to check for user refs.
1048 	 */
1049 	local_bh_disable();
1050 	bh_lock_sock(sk);
1051 	WARN_ON(sock_owned_by_user(sk));
1052 
1053 	this_cpu_inc(dccp_orphan_count);
1054 
1055 	/* Have we already been destroyed by a softirq or backlog? */
1056 	if (state != DCCP_CLOSED && sk->sk_state == DCCP_CLOSED)
1057 		goto out;
1058 
1059 	if (sk->sk_state == DCCP_CLOSED)
1060 		inet_csk_destroy_sock(sk);
1061 
1062 	/* Otherwise, socket is reprieved until protocol close. */
1063 
1064 out:
1065 	bh_unlock_sock(sk);
1066 	local_bh_enable();
1067 	sock_put(sk);
1068 }
1069 
1070 EXPORT_SYMBOL_GPL(dccp_close);
1071 
1072 void dccp_shutdown(struct sock *sk, int how)
1073 {
1074 	dccp_pr_debug("called shutdown(%x)\n", how);
1075 }
1076 
1077 EXPORT_SYMBOL_GPL(dccp_shutdown);
1078 
1079 static inline int __init dccp_mib_init(void)
1080 {
1081 	dccp_statistics = alloc_percpu(struct dccp_mib);
1082 	if (!dccp_statistics)
1083 		return -ENOMEM;
1084 	return 0;
1085 }
1086 
1087 static inline void dccp_mib_exit(void)
1088 {
1089 	free_percpu(dccp_statistics);
1090 }
1091 
1092 static int thash_entries;
1093 module_param(thash_entries, int, 0444);
1094 MODULE_PARM_DESC(thash_entries, "Number of ehash buckets");
1095 
1096 #ifdef CONFIG_IP_DCCP_DEBUG
1097 bool dccp_debug;
1098 module_param(dccp_debug, bool, 0644);
1099 MODULE_PARM_DESC(dccp_debug, "Enable debug messages");
1100 
1101 EXPORT_SYMBOL_GPL(dccp_debug);
1102 #endif
1103 
1104 static int __init dccp_init(void)
1105 {
1106 	unsigned long goal;
1107 	unsigned long nr_pages = totalram_pages();
1108 	int ehash_order, bhash_order, i;
1109 	int rc;
1110 
1111 	BUILD_BUG_ON(sizeof(struct dccp_skb_cb) >
1112 		     sizeof_field(struct sk_buff, cb));
1113 	inet_hashinfo_init(&dccp_hashinfo);
1114 	rc = inet_hashinfo2_init_mod(&dccp_hashinfo);
1115 	if (rc)
1116 		goto out_fail;
1117 	rc = -ENOBUFS;
1118 	dccp_hashinfo.bind_bucket_cachep =
1119 		kmem_cache_create("dccp_bind_bucket",
1120 				  sizeof(struct inet_bind_bucket), 0,
1121 				  SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, NULL);
1122 	if (!dccp_hashinfo.bind_bucket_cachep)
1123 		goto out_free_hashinfo2;
1124 
1125 	/*
1126 	 * Size and allocate the main established and bind bucket
1127 	 * hash tables.
1128 	 *
1129 	 * The methodology is similar to that of the buffer cache.
1130 	 */
1131 	if (nr_pages >= (128 * 1024))
1132 		goal = nr_pages >> (21 - PAGE_SHIFT);
1133 	else
1134 		goal = nr_pages >> (23 - PAGE_SHIFT);
1135 
1136 	if (thash_entries)
1137 		goal = (thash_entries *
1138 			sizeof(struct inet_ehash_bucket)) >> PAGE_SHIFT;
1139 	for (ehash_order = 0; (1UL << ehash_order) < goal; ehash_order++)
1140 		;
1141 	do {
1142 		unsigned long hash_size = (1UL << ehash_order) * PAGE_SIZE /
1143 					sizeof(struct inet_ehash_bucket);
1144 
1145 		while (hash_size & (hash_size - 1))
1146 			hash_size--;
1147 		dccp_hashinfo.ehash_mask = hash_size - 1;
1148 		dccp_hashinfo.ehash = (struct inet_ehash_bucket *)
1149 			__get_free_pages(GFP_ATOMIC|__GFP_NOWARN, ehash_order);
1150 	} while (!dccp_hashinfo.ehash && --ehash_order > 0);
1151 
1152 	if (!dccp_hashinfo.ehash) {
1153 		DCCP_CRIT("Failed to allocate DCCP established hash table");
1154 		goto out_free_bind_bucket_cachep;
1155 	}
1156 
1157 	for (i = 0; i <= dccp_hashinfo.ehash_mask; i++)
1158 		INIT_HLIST_NULLS_HEAD(&dccp_hashinfo.ehash[i].chain, i);
1159 
1160 	if (inet_ehash_locks_alloc(&dccp_hashinfo))
1161 			goto out_free_dccp_ehash;
1162 
1163 	bhash_order = ehash_order;
1164 
1165 	do {
1166 		dccp_hashinfo.bhash_size = (1UL << bhash_order) * PAGE_SIZE /
1167 					sizeof(struct inet_bind_hashbucket);
1168 		if ((dccp_hashinfo.bhash_size > (64 * 1024)) &&
1169 		    bhash_order > 0)
1170 			continue;
1171 		dccp_hashinfo.bhash = (struct inet_bind_hashbucket *)
1172 			__get_free_pages(GFP_ATOMIC|__GFP_NOWARN, bhash_order);
1173 	} while (!dccp_hashinfo.bhash && --bhash_order >= 0);
1174 
1175 	if (!dccp_hashinfo.bhash) {
1176 		DCCP_CRIT("Failed to allocate DCCP bind hash table");
1177 		goto out_free_dccp_locks;
1178 	}
1179 
1180 	for (i = 0; i < dccp_hashinfo.bhash_size; i++) {
1181 		spin_lock_init(&dccp_hashinfo.bhash[i].lock);
1182 		INIT_HLIST_HEAD(&dccp_hashinfo.bhash[i].chain);
1183 	}
1184 
1185 	rc = dccp_mib_init();
1186 	if (rc)
1187 		goto out_free_dccp_bhash;
1188 
1189 	rc = dccp_ackvec_init();
1190 	if (rc)
1191 		goto out_free_dccp_mib;
1192 
1193 	rc = dccp_sysctl_init();
1194 	if (rc)
1195 		goto out_ackvec_exit;
1196 
1197 	rc = ccid_initialize_builtins();
1198 	if (rc)
1199 		goto out_sysctl_exit;
1200 
1201 	dccp_timestamping_init();
1202 
1203 	return 0;
1204 
1205 out_sysctl_exit:
1206 	dccp_sysctl_exit();
1207 out_ackvec_exit:
1208 	dccp_ackvec_exit();
1209 out_free_dccp_mib:
1210 	dccp_mib_exit();
1211 out_free_dccp_bhash:
1212 	free_pages((unsigned long)dccp_hashinfo.bhash, bhash_order);
1213 out_free_dccp_locks:
1214 	inet_ehash_locks_free(&dccp_hashinfo);
1215 out_free_dccp_ehash:
1216 	free_pages((unsigned long)dccp_hashinfo.ehash, ehash_order);
1217 out_free_bind_bucket_cachep:
1218 	kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1219 out_free_hashinfo2:
1220 	inet_hashinfo2_free_mod(&dccp_hashinfo);
1221 out_fail:
1222 	dccp_hashinfo.bhash = NULL;
1223 	dccp_hashinfo.ehash = NULL;
1224 	dccp_hashinfo.bind_bucket_cachep = NULL;
1225 	return rc;
1226 }
1227 
1228 static void __exit dccp_fini(void)
1229 {
1230 	ccid_cleanup_builtins();
1231 	dccp_mib_exit();
1232 	free_pages((unsigned long)dccp_hashinfo.bhash,
1233 		   get_order(dccp_hashinfo.bhash_size *
1234 			     sizeof(struct inet_bind_hashbucket)));
1235 	free_pages((unsigned long)dccp_hashinfo.ehash,
1236 		   get_order((dccp_hashinfo.ehash_mask + 1) *
1237 			     sizeof(struct inet_ehash_bucket)));
1238 	inet_ehash_locks_free(&dccp_hashinfo);
1239 	kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1240 	dccp_ackvec_exit();
1241 	dccp_sysctl_exit();
1242 	inet_hashinfo2_free_mod(&dccp_hashinfo);
1243 }
1244 
1245 module_init(dccp_init);
1246 module_exit(dccp_fini);
1247 
1248 MODULE_LICENSE("GPL");
1249 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@conectiva.com.br>");
1250 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");
1251