xref: /openbmc/linux/net/dccp/proto.c (revision 48cc39c3)
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 struct percpu_counter dccp_orphan_count;
46 EXPORT_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_listen_start(struct sock *sk, int backlog)
242 {
243 	struct dccp_sock *dp = dccp_sk(sk);
244 
245 	dp->dccps_role = DCCP_ROLE_LISTEN;
246 	/* do not start to listen if feature negotiation setup fails */
247 	if (dccp_feat_finalise_settings(dp))
248 		return -EPROTO;
249 	return inet_csk_listen_start(sk, backlog);
250 }
251 
252 static inline int dccp_need_reset(int state)
253 {
254 	return state != DCCP_CLOSED && state != DCCP_LISTEN &&
255 	       state != DCCP_REQUESTING;
256 }
257 
258 int dccp_disconnect(struct sock *sk, int flags)
259 {
260 	struct inet_connection_sock *icsk = inet_csk(sk);
261 	struct inet_sock *inet = inet_sk(sk);
262 	struct dccp_sock *dp = dccp_sk(sk);
263 	const int old_state = sk->sk_state;
264 
265 	if (old_state != DCCP_CLOSED)
266 		dccp_set_state(sk, DCCP_CLOSED);
267 
268 	/*
269 	 * This corresponds to the ABORT function of RFC793, sec. 3.8
270 	 * TCP uses a RST segment, DCCP a Reset packet with Code 2, "Aborted".
271 	 */
272 	if (old_state == DCCP_LISTEN) {
273 		inet_csk_listen_stop(sk);
274 	} else if (dccp_need_reset(old_state)) {
275 		dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED);
276 		sk->sk_err = ECONNRESET;
277 	} else if (old_state == DCCP_REQUESTING)
278 		sk->sk_err = ECONNRESET;
279 
280 	dccp_clear_xmit_timers(sk);
281 	ccid_hc_rx_delete(dp->dccps_hc_rx_ccid, sk);
282 	dp->dccps_hc_rx_ccid = NULL;
283 
284 	__skb_queue_purge(&sk->sk_receive_queue);
285 	__skb_queue_purge(&sk->sk_write_queue);
286 	if (sk->sk_send_head != NULL) {
287 		__kfree_skb(sk->sk_send_head);
288 		sk->sk_send_head = NULL;
289 	}
290 
291 	inet->inet_dport = 0;
292 
293 	if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
294 		inet_reset_saddr(sk);
295 
296 	sk->sk_shutdown = 0;
297 	sock_reset_flag(sk, SOCK_DONE);
298 
299 	icsk->icsk_backoff = 0;
300 	inet_csk_delack_init(sk);
301 	__sk_dst_reset(sk);
302 
303 	WARN_ON(inet->inet_num && !icsk->icsk_bind_hash);
304 
305 	sk_error_report(sk);
306 	return 0;
307 }
308 
309 EXPORT_SYMBOL_GPL(dccp_disconnect);
310 
311 /*
312  *	Wait for a DCCP event.
313  *
314  *	Note that we don't need to lock the socket, as the upper poll layers
315  *	take care of normal races (between the test and the event) and we don't
316  *	go look at any of the socket buffers directly.
317  */
318 __poll_t dccp_poll(struct file *file, struct socket *sock,
319 		       poll_table *wait)
320 {
321 	__poll_t mask;
322 	struct sock *sk = sock->sk;
323 
324 	sock_poll_wait(file, sock, wait);
325 	if (sk->sk_state == DCCP_LISTEN)
326 		return inet_csk_listen_poll(sk);
327 
328 	/* Socket is not locked. We are protected from async events
329 	   by poll logic and correct handling of state changes
330 	   made by another threads is impossible in any case.
331 	 */
332 
333 	mask = 0;
334 	if (sk->sk_err)
335 		mask = EPOLLERR;
336 
337 	if (sk->sk_shutdown == SHUTDOWN_MASK || sk->sk_state == DCCP_CLOSED)
338 		mask |= EPOLLHUP;
339 	if (sk->sk_shutdown & RCV_SHUTDOWN)
340 		mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
341 
342 	/* Connected? */
343 	if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_RESPOND)) {
344 		if (atomic_read(&sk->sk_rmem_alloc) > 0)
345 			mask |= EPOLLIN | EPOLLRDNORM;
346 
347 		if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
348 			if (sk_stream_is_writeable(sk)) {
349 				mask |= EPOLLOUT | EPOLLWRNORM;
350 			} else {  /* send SIGIO later */
351 				sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
352 				set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
353 
354 				/* Race breaker. If space is freed after
355 				 * wspace test but before the flags are set,
356 				 * IO signal will be lost.
357 				 */
358 				if (sk_stream_is_writeable(sk))
359 					mask |= EPOLLOUT | EPOLLWRNORM;
360 			}
361 		}
362 	}
363 	return mask;
364 }
365 
366 EXPORT_SYMBOL_GPL(dccp_poll);
367 
368 int dccp_ioctl(struct sock *sk, int cmd, unsigned long arg)
369 {
370 	int rc = -ENOTCONN;
371 
372 	lock_sock(sk);
373 
374 	if (sk->sk_state == DCCP_LISTEN)
375 		goto out;
376 
377 	switch (cmd) {
378 	case SIOCOUTQ: {
379 		int amount = sk_wmem_alloc_get(sk);
380 		/* Using sk_wmem_alloc here because sk_wmem_queued is not used by DCCP and
381 		 * always 0, comparably to UDP.
382 		 */
383 
384 		rc = put_user(amount, (int __user *)arg);
385 	}
386 		break;
387 	case SIOCINQ: {
388 		struct sk_buff *skb;
389 		unsigned long amount = 0;
390 
391 		skb = skb_peek(&sk->sk_receive_queue);
392 		if (skb != NULL) {
393 			/*
394 			 * We will only return the amount of this packet since
395 			 * that is all that will be read.
396 			 */
397 			amount = skb->len;
398 		}
399 		rc = put_user(amount, (int __user *)arg);
400 	}
401 		break;
402 	default:
403 		rc = -ENOIOCTLCMD;
404 		break;
405 	}
406 out:
407 	release_sock(sk);
408 	return rc;
409 }
410 
411 EXPORT_SYMBOL_GPL(dccp_ioctl);
412 
413 static int dccp_setsockopt_service(struct sock *sk, const __be32 service,
414 				   sockptr_t optval, unsigned int optlen)
415 {
416 	struct dccp_sock *dp = dccp_sk(sk);
417 	struct dccp_service_list *sl = NULL;
418 
419 	if (service == DCCP_SERVICE_INVALID_VALUE ||
420 	    optlen > DCCP_SERVICE_LIST_MAX_LEN * sizeof(u32))
421 		return -EINVAL;
422 
423 	if (optlen > sizeof(service)) {
424 		sl = kmalloc(optlen, GFP_KERNEL);
425 		if (sl == NULL)
426 			return -ENOMEM;
427 
428 		sl->dccpsl_nr = optlen / sizeof(u32) - 1;
429 		if (copy_from_sockptr_offset(sl->dccpsl_list, optval,
430 				sizeof(service), optlen - sizeof(service)) ||
431 		    dccp_list_has_service(sl, DCCP_SERVICE_INVALID_VALUE)) {
432 			kfree(sl);
433 			return -EFAULT;
434 		}
435 	}
436 
437 	lock_sock(sk);
438 	dp->dccps_service = service;
439 
440 	kfree(dp->dccps_service_list);
441 
442 	dp->dccps_service_list = sl;
443 	release_sock(sk);
444 	return 0;
445 }
446 
447 static int dccp_setsockopt_cscov(struct sock *sk, int cscov, bool rx)
448 {
449 	u8 *list, len;
450 	int i, rc;
451 
452 	if (cscov < 0 || cscov > 15)
453 		return -EINVAL;
454 	/*
455 	 * Populate a list of permissible values, in the range cscov...15. This
456 	 * is necessary since feature negotiation of single values only works if
457 	 * both sides incidentally choose the same value. Since the list starts
458 	 * lowest-value first, negotiation will pick the smallest shared value.
459 	 */
460 	if (cscov == 0)
461 		return 0;
462 	len = 16 - cscov;
463 
464 	list = kmalloc(len, GFP_KERNEL);
465 	if (list == NULL)
466 		return -ENOBUFS;
467 
468 	for (i = 0; i < len; i++)
469 		list[i] = cscov++;
470 
471 	rc = dccp_feat_register_sp(sk, DCCPF_MIN_CSUM_COVER, rx, list, len);
472 
473 	if (rc == 0) {
474 		if (rx)
475 			dccp_sk(sk)->dccps_pcrlen = cscov;
476 		else
477 			dccp_sk(sk)->dccps_pcslen = cscov;
478 	}
479 	kfree(list);
480 	return rc;
481 }
482 
483 static int dccp_setsockopt_ccid(struct sock *sk, int type,
484 				sockptr_t optval, unsigned int optlen)
485 {
486 	u8 *val;
487 	int rc = 0;
488 
489 	if (optlen < 1 || optlen > DCCP_FEAT_MAX_SP_VALS)
490 		return -EINVAL;
491 
492 	val = memdup_sockptr(optval, optlen);
493 	if (IS_ERR(val))
494 		return PTR_ERR(val);
495 
496 	lock_sock(sk);
497 	if (type == DCCP_SOCKOPT_TX_CCID || type == DCCP_SOCKOPT_CCID)
498 		rc = dccp_feat_register_sp(sk, DCCPF_CCID, 1, val, optlen);
499 
500 	if (!rc && (type == DCCP_SOCKOPT_RX_CCID || type == DCCP_SOCKOPT_CCID))
501 		rc = dccp_feat_register_sp(sk, DCCPF_CCID, 0, val, optlen);
502 	release_sock(sk);
503 
504 	kfree(val);
505 	return rc;
506 }
507 
508 static int do_dccp_setsockopt(struct sock *sk, int level, int optname,
509 		sockptr_t optval, unsigned int optlen)
510 {
511 	struct dccp_sock *dp = dccp_sk(sk);
512 	int val, err = 0;
513 
514 	switch (optname) {
515 	case DCCP_SOCKOPT_PACKET_SIZE:
516 		DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n");
517 		return 0;
518 	case DCCP_SOCKOPT_CHANGE_L:
519 	case DCCP_SOCKOPT_CHANGE_R:
520 		DCCP_WARN("sockopt(CHANGE_L/R) is deprecated: fix your app\n");
521 		return 0;
522 	case DCCP_SOCKOPT_CCID:
523 	case DCCP_SOCKOPT_RX_CCID:
524 	case DCCP_SOCKOPT_TX_CCID:
525 		return dccp_setsockopt_ccid(sk, optname, optval, optlen);
526 	}
527 
528 	if (optlen < (int)sizeof(int))
529 		return -EINVAL;
530 
531 	if (copy_from_sockptr(&val, optval, sizeof(int)))
532 		return -EFAULT;
533 
534 	if (optname == DCCP_SOCKOPT_SERVICE)
535 		return dccp_setsockopt_service(sk, val, optval, optlen);
536 
537 	lock_sock(sk);
538 	switch (optname) {
539 	case DCCP_SOCKOPT_SERVER_TIMEWAIT:
540 		if (dp->dccps_role != DCCP_ROLE_SERVER)
541 			err = -EOPNOTSUPP;
542 		else
543 			dp->dccps_server_timewait = (val != 0);
544 		break;
545 	case DCCP_SOCKOPT_SEND_CSCOV:
546 		err = dccp_setsockopt_cscov(sk, val, false);
547 		break;
548 	case DCCP_SOCKOPT_RECV_CSCOV:
549 		err = dccp_setsockopt_cscov(sk, val, true);
550 		break;
551 	case DCCP_SOCKOPT_QPOLICY_ID:
552 		if (sk->sk_state != DCCP_CLOSED)
553 			err = -EISCONN;
554 		else if (val < 0 || val >= DCCPQ_POLICY_MAX)
555 			err = -EINVAL;
556 		else
557 			dp->dccps_qpolicy = val;
558 		break;
559 	case DCCP_SOCKOPT_QPOLICY_TXQLEN:
560 		if (val < 0)
561 			err = -EINVAL;
562 		else
563 			dp->dccps_tx_qlen = val;
564 		break;
565 	default:
566 		err = -ENOPROTOOPT;
567 		break;
568 	}
569 	release_sock(sk);
570 
571 	return err;
572 }
573 
574 int dccp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
575 		    unsigned int optlen)
576 {
577 	if (level != SOL_DCCP)
578 		return inet_csk(sk)->icsk_af_ops->setsockopt(sk, level,
579 							     optname, optval,
580 							     optlen);
581 	return do_dccp_setsockopt(sk, level, optname, optval, optlen);
582 }
583 
584 EXPORT_SYMBOL_GPL(dccp_setsockopt);
585 
586 static int dccp_getsockopt_service(struct sock *sk, int len,
587 				   __be32 __user *optval,
588 				   int __user *optlen)
589 {
590 	const struct dccp_sock *dp = dccp_sk(sk);
591 	const struct dccp_service_list *sl;
592 	int err = -ENOENT, slen = 0, total_len = sizeof(u32);
593 
594 	lock_sock(sk);
595 	if ((sl = dp->dccps_service_list) != NULL) {
596 		slen = sl->dccpsl_nr * sizeof(u32);
597 		total_len += slen;
598 	}
599 
600 	err = -EINVAL;
601 	if (total_len > len)
602 		goto out;
603 
604 	err = 0;
605 	if (put_user(total_len, optlen) ||
606 	    put_user(dp->dccps_service, optval) ||
607 	    (sl != NULL && copy_to_user(optval + 1, sl->dccpsl_list, slen)))
608 		err = -EFAULT;
609 out:
610 	release_sock(sk);
611 	return err;
612 }
613 
614 static int do_dccp_getsockopt(struct sock *sk, int level, int optname,
615 		    char __user *optval, int __user *optlen)
616 {
617 	struct dccp_sock *dp;
618 	int val, len;
619 
620 	if (get_user(len, optlen))
621 		return -EFAULT;
622 
623 	if (len < (int)sizeof(int))
624 		return -EINVAL;
625 
626 	dp = dccp_sk(sk);
627 
628 	switch (optname) {
629 	case DCCP_SOCKOPT_PACKET_SIZE:
630 		DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n");
631 		return 0;
632 	case DCCP_SOCKOPT_SERVICE:
633 		return dccp_getsockopt_service(sk, len,
634 					       (__be32 __user *)optval, optlen);
635 	case DCCP_SOCKOPT_GET_CUR_MPS:
636 		val = dp->dccps_mss_cache;
637 		break;
638 	case DCCP_SOCKOPT_AVAILABLE_CCIDS:
639 		return ccid_getsockopt_builtin_ccids(sk, len, optval, optlen);
640 	case DCCP_SOCKOPT_TX_CCID:
641 		val = ccid_get_current_tx_ccid(dp);
642 		if (val < 0)
643 			return -ENOPROTOOPT;
644 		break;
645 	case DCCP_SOCKOPT_RX_CCID:
646 		val = ccid_get_current_rx_ccid(dp);
647 		if (val < 0)
648 			return -ENOPROTOOPT;
649 		break;
650 	case DCCP_SOCKOPT_SERVER_TIMEWAIT:
651 		val = dp->dccps_server_timewait;
652 		break;
653 	case DCCP_SOCKOPT_SEND_CSCOV:
654 		val = dp->dccps_pcslen;
655 		break;
656 	case DCCP_SOCKOPT_RECV_CSCOV:
657 		val = dp->dccps_pcrlen;
658 		break;
659 	case DCCP_SOCKOPT_QPOLICY_ID:
660 		val = dp->dccps_qpolicy;
661 		break;
662 	case DCCP_SOCKOPT_QPOLICY_TXQLEN:
663 		val = dp->dccps_tx_qlen;
664 		break;
665 	case 128 ... 191:
666 		return ccid_hc_rx_getsockopt(dp->dccps_hc_rx_ccid, sk, optname,
667 					     len, (u32 __user *)optval, optlen);
668 	case 192 ... 255:
669 		return ccid_hc_tx_getsockopt(dp->dccps_hc_tx_ccid, sk, optname,
670 					     len, (u32 __user *)optval, optlen);
671 	default:
672 		return -ENOPROTOOPT;
673 	}
674 
675 	len = sizeof(val);
676 	if (put_user(len, optlen) || copy_to_user(optval, &val, len))
677 		return -EFAULT;
678 
679 	return 0;
680 }
681 
682 int dccp_getsockopt(struct sock *sk, int level, int optname,
683 		    char __user *optval, int __user *optlen)
684 {
685 	if (level != SOL_DCCP)
686 		return inet_csk(sk)->icsk_af_ops->getsockopt(sk, level,
687 							     optname, optval,
688 							     optlen);
689 	return do_dccp_getsockopt(sk, level, optname, optval, optlen);
690 }
691 
692 EXPORT_SYMBOL_GPL(dccp_getsockopt);
693 
694 static int dccp_msghdr_parse(struct msghdr *msg, struct sk_buff *skb)
695 {
696 	struct cmsghdr *cmsg;
697 
698 	/*
699 	 * Assign an (opaque) qpolicy priority value to skb->priority.
700 	 *
701 	 * We are overloading this skb field for use with the qpolicy subystem.
702 	 * The skb->priority is normally used for the SO_PRIORITY option, which
703 	 * is initialised from sk_priority. Since the assignment of sk_priority
704 	 * to skb->priority happens later (on layer 3), we overload this field
705 	 * for use with queueing priorities as long as the skb is on layer 4.
706 	 * The default priority value (if nothing is set) is 0.
707 	 */
708 	skb->priority = 0;
709 
710 	for_each_cmsghdr(cmsg, msg) {
711 		if (!CMSG_OK(msg, cmsg))
712 			return -EINVAL;
713 
714 		if (cmsg->cmsg_level != SOL_DCCP)
715 			continue;
716 
717 		if (cmsg->cmsg_type <= DCCP_SCM_QPOLICY_MAX &&
718 		    !dccp_qpolicy_param_ok(skb->sk, cmsg->cmsg_type))
719 			return -EINVAL;
720 
721 		switch (cmsg->cmsg_type) {
722 		case DCCP_SCM_PRIORITY:
723 			if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u32)))
724 				return -EINVAL;
725 			skb->priority = *(__u32 *)CMSG_DATA(cmsg);
726 			break;
727 		default:
728 			return -EINVAL;
729 		}
730 	}
731 	return 0;
732 }
733 
734 int dccp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
735 {
736 	const struct dccp_sock *dp = dccp_sk(sk);
737 	const int flags = msg->msg_flags;
738 	const int noblock = flags & MSG_DONTWAIT;
739 	struct sk_buff *skb;
740 	int rc, size;
741 	long timeo;
742 
743 	trace_dccp_probe(sk, len);
744 
745 	if (len > dp->dccps_mss_cache)
746 		return -EMSGSIZE;
747 
748 	lock_sock(sk);
749 
750 	if (dccp_qpolicy_full(sk)) {
751 		rc = -EAGAIN;
752 		goto out_release;
753 	}
754 
755 	timeo = sock_sndtimeo(sk, noblock);
756 
757 	/*
758 	 * We have to use sk_stream_wait_connect here to set sk_write_pending,
759 	 * so that the trick in dccp_rcv_request_sent_state_process.
760 	 */
761 	/* Wait for a connection to finish. */
762 	if ((1 << sk->sk_state) & ~(DCCPF_OPEN | DCCPF_PARTOPEN))
763 		if ((rc = sk_stream_wait_connect(sk, &timeo)) != 0)
764 			goto out_release;
765 
766 	size = sk->sk_prot->max_header + len;
767 	release_sock(sk);
768 	skb = sock_alloc_send_skb(sk, size, noblock, &rc);
769 	lock_sock(sk);
770 	if (skb == NULL)
771 		goto out_release;
772 
773 	if (sk->sk_state == DCCP_CLOSED) {
774 		rc = -ENOTCONN;
775 		goto out_discard;
776 	}
777 
778 	skb_reserve(skb, sk->sk_prot->max_header);
779 	rc = memcpy_from_msg(skb_put(skb, len), msg, len);
780 	if (rc != 0)
781 		goto out_discard;
782 
783 	rc = dccp_msghdr_parse(msg, skb);
784 	if (rc != 0)
785 		goto out_discard;
786 
787 	dccp_qpolicy_push(sk, skb);
788 	/*
789 	 * The xmit_timer is set if the TX CCID is rate-based and will expire
790 	 * when congestion control permits to release further packets into the
791 	 * network. Window-based CCIDs do not use this timer.
792 	 */
793 	if (!timer_pending(&dp->dccps_xmit_timer))
794 		dccp_write_xmit(sk);
795 out_release:
796 	release_sock(sk);
797 	return rc ? : len;
798 out_discard:
799 	kfree_skb(skb);
800 	goto out_release;
801 }
802 
803 EXPORT_SYMBOL_GPL(dccp_sendmsg);
804 
805 int dccp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
806 		 int flags, int *addr_len)
807 {
808 	const struct dccp_hdr *dh;
809 	long timeo;
810 
811 	lock_sock(sk);
812 
813 	if (sk->sk_state == DCCP_LISTEN) {
814 		len = -ENOTCONN;
815 		goto out;
816 	}
817 
818 	timeo = sock_rcvtimeo(sk, nonblock);
819 
820 	do {
821 		struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
822 
823 		if (skb == NULL)
824 			goto verify_sock_status;
825 
826 		dh = dccp_hdr(skb);
827 
828 		switch (dh->dccph_type) {
829 		case DCCP_PKT_DATA:
830 		case DCCP_PKT_DATAACK:
831 			goto found_ok_skb;
832 
833 		case DCCP_PKT_CLOSE:
834 		case DCCP_PKT_CLOSEREQ:
835 			if (!(flags & MSG_PEEK))
836 				dccp_finish_passive_close(sk);
837 			fallthrough;
838 		case DCCP_PKT_RESET:
839 			dccp_pr_debug("found fin (%s) ok!\n",
840 				      dccp_packet_name(dh->dccph_type));
841 			len = 0;
842 			goto found_fin_ok;
843 		default:
844 			dccp_pr_debug("packet_type=%s\n",
845 				      dccp_packet_name(dh->dccph_type));
846 			sk_eat_skb(sk, skb);
847 		}
848 verify_sock_status:
849 		if (sock_flag(sk, SOCK_DONE)) {
850 			len = 0;
851 			break;
852 		}
853 
854 		if (sk->sk_err) {
855 			len = sock_error(sk);
856 			break;
857 		}
858 
859 		if (sk->sk_shutdown & RCV_SHUTDOWN) {
860 			len = 0;
861 			break;
862 		}
863 
864 		if (sk->sk_state == DCCP_CLOSED) {
865 			if (!sock_flag(sk, SOCK_DONE)) {
866 				/* This occurs when user tries to read
867 				 * from never connected socket.
868 				 */
869 				len = -ENOTCONN;
870 				break;
871 			}
872 			len = 0;
873 			break;
874 		}
875 
876 		if (!timeo) {
877 			len = -EAGAIN;
878 			break;
879 		}
880 
881 		if (signal_pending(current)) {
882 			len = sock_intr_errno(timeo);
883 			break;
884 		}
885 
886 		sk_wait_data(sk, &timeo, NULL);
887 		continue;
888 	found_ok_skb:
889 		if (len > skb->len)
890 			len = skb->len;
891 		else if (len < skb->len)
892 			msg->msg_flags |= MSG_TRUNC;
893 
894 		if (skb_copy_datagram_msg(skb, 0, msg, len)) {
895 			/* Exception. Bailout! */
896 			len = -EFAULT;
897 			break;
898 		}
899 		if (flags & MSG_TRUNC)
900 			len = skb->len;
901 	found_fin_ok:
902 		if (!(flags & MSG_PEEK))
903 			sk_eat_skb(sk, skb);
904 		break;
905 	} while (1);
906 out:
907 	release_sock(sk);
908 	return len;
909 }
910 
911 EXPORT_SYMBOL_GPL(dccp_recvmsg);
912 
913 int inet_dccp_listen(struct socket *sock, int backlog)
914 {
915 	struct sock *sk = sock->sk;
916 	unsigned char old_state;
917 	int err;
918 
919 	lock_sock(sk);
920 
921 	err = -EINVAL;
922 	if (sock->state != SS_UNCONNECTED || sock->type != SOCK_DCCP)
923 		goto out;
924 
925 	old_state = sk->sk_state;
926 	if (!((1 << old_state) & (DCCPF_CLOSED | DCCPF_LISTEN)))
927 		goto out;
928 
929 	WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
930 	/* Really, if the socket is already in listen state
931 	 * we can only allow the backlog to be adjusted.
932 	 */
933 	if (old_state != DCCP_LISTEN) {
934 		/*
935 		 * FIXME: here it probably should be sk->sk_prot->listen_start
936 		 * see tcp_listen_start
937 		 */
938 		err = dccp_listen_start(sk, backlog);
939 		if (err)
940 			goto out;
941 	}
942 	err = 0;
943 
944 out:
945 	release_sock(sk);
946 	return err;
947 }
948 
949 EXPORT_SYMBOL_GPL(inet_dccp_listen);
950 
951 static void dccp_terminate_connection(struct sock *sk)
952 {
953 	u8 next_state = DCCP_CLOSED;
954 
955 	switch (sk->sk_state) {
956 	case DCCP_PASSIVE_CLOSE:
957 	case DCCP_PASSIVE_CLOSEREQ:
958 		dccp_finish_passive_close(sk);
959 		break;
960 	case DCCP_PARTOPEN:
961 		dccp_pr_debug("Stop PARTOPEN timer (%p)\n", sk);
962 		inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
963 		fallthrough;
964 	case DCCP_OPEN:
965 		dccp_send_close(sk, 1);
966 
967 		if (dccp_sk(sk)->dccps_role == DCCP_ROLE_SERVER &&
968 		    !dccp_sk(sk)->dccps_server_timewait)
969 			next_state = DCCP_ACTIVE_CLOSEREQ;
970 		else
971 			next_state = DCCP_CLOSING;
972 		fallthrough;
973 	default:
974 		dccp_set_state(sk, next_state);
975 	}
976 }
977 
978 void dccp_close(struct sock *sk, long timeout)
979 {
980 	struct dccp_sock *dp = dccp_sk(sk);
981 	struct sk_buff *skb;
982 	u32 data_was_unread = 0;
983 	int state;
984 
985 	lock_sock(sk);
986 
987 	sk->sk_shutdown = SHUTDOWN_MASK;
988 
989 	if (sk->sk_state == DCCP_LISTEN) {
990 		dccp_set_state(sk, DCCP_CLOSED);
991 
992 		/* Special case. */
993 		inet_csk_listen_stop(sk);
994 
995 		goto adjudge_to_death;
996 	}
997 
998 	sk_stop_timer(sk, &dp->dccps_xmit_timer);
999 
1000 	/*
1001 	 * We need to flush the recv. buffs.  We do this only on the
1002 	 * descriptor close, not protocol-sourced closes, because the
1003 	  *reader process may not have drained the data yet!
1004 	 */
1005 	while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
1006 		data_was_unread += skb->len;
1007 		__kfree_skb(skb);
1008 	}
1009 
1010 	/* If socket has been already reset kill it. */
1011 	if (sk->sk_state == DCCP_CLOSED)
1012 		goto adjudge_to_death;
1013 
1014 	if (data_was_unread) {
1015 		/* Unread data was tossed, send an appropriate Reset Code */
1016 		DCCP_WARN("ABORT with %u bytes unread\n", data_was_unread);
1017 		dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED);
1018 		dccp_set_state(sk, DCCP_CLOSED);
1019 	} else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
1020 		/* Check zero linger _after_ checking for unread data. */
1021 		sk->sk_prot->disconnect(sk, 0);
1022 	} else if (sk->sk_state != DCCP_CLOSED) {
1023 		/*
1024 		 * Normal connection termination. May need to wait if there are
1025 		 * still packets in the TX queue that are delayed by the CCID.
1026 		 */
1027 		dccp_flush_write_queue(sk, &timeout);
1028 		dccp_terminate_connection(sk);
1029 	}
1030 
1031 	/*
1032 	 * Flush write queue. This may be necessary in several cases:
1033 	 * - we have been closed by the peer but still have application data;
1034 	 * - abortive termination (unread data or zero linger time),
1035 	 * - normal termination but queue could not be flushed within time limit
1036 	 */
1037 	__skb_queue_purge(&sk->sk_write_queue);
1038 
1039 	sk_stream_wait_close(sk, timeout);
1040 
1041 adjudge_to_death:
1042 	state = sk->sk_state;
1043 	sock_hold(sk);
1044 	sock_orphan(sk);
1045 
1046 	/*
1047 	 * It is the last release_sock in its life. It will remove backlog.
1048 	 */
1049 	release_sock(sk);
1050 	/*
1051 	 * Now socket is owned by kernel and we acquire BH lock
1052 	 * to finish close. No need to check for user refs.
1053 	 */
1054 	local_bh_disable();
1055 	bh_lock_sock(sk);
1056 	WARN_ON(sock_owned_by_user(sk));
1057 
1058 	percpu_counter_inc(sk->sk_prot->orphan_count);
1059 
1060 	/* Have we already been destroyed by a softirq or backlog? */
1061 	if (state != DCCP_CLOSED && sk->sk_state == DCCP_CLOSED)
1062 		goto out;
1063 
1064 	if (sk->sk_state == DCCP_CLOSED)
1065 		inet_csk_destroy_sock(sk);
1066 
1067 	/* Otherwise, socket is reprieved until protocol close. */
1068 
1069 out:
1070 	bh_unlock_sock(sk);
1071 	local_bh_enable();
1072 	sock_put(sk);
1073 }
1074 
1075 EXPORT_SYMBOL_GPL(dccp_close);
1076 
1077 void dccp_shutdown(struct sock *sk, int how)
1078 {
1079 	dccp_pr_debug("called shutdown(%x)\n", how);
1080 }
1081 
1082 EXPORT_SYMBOL_GPL(dccp_shutdown);
1083 
1084 static inline int __init dccp_mib_init(void)
1085 {
1086 	dccp_statistics = alloc_percpu(struct dccp_mib);
1087 	if (!dccp_statistics)
1088 		return -ENOMEM;
1089 	return 0;
1090 }
1091 
1092 static inline void dccp_mib_exit(void)
1093 {
1094 	free_percpu(dccp_statistics);
1095 }
1096 
1097 static int thash_entries;
1098 module_param(thash_entries, int, 0444);
1099 MODULE_PARM_DESC(thash_entries, "Number of ehash buckets");
1100 
1101 #ifdef CONFIG_IP_DCCP_DEBUG
1102 bool dccp_debug;
1103 module_param(dccp_debug, bool, 0644);
1104 MODULE_PARM_DESC(dccp_debug, "Enable debug messages");
1105 
1106 EXPORT_SYMBOL_GPL(dccp_debug);
1107 #endif
1108 
1109 static int __init dccp_init(void)
1110 {
1111 	unsigned long goal;
1112 	unsigned long nr_pages = totalram_pages();
1113 	int ehash_order, bhash_order, i;
1114 	int rc;
1115 
1116 	BUILD_BUG_ON(sizeof(struct dccp_skb_cb) >
1117 		     sizeof_field(struct sk_buff, cb));
1118 	rc = percpu_counter_init(&dccp_orphan_count, 0, GFP_KERNEL);
1119 	if (rc)
1120 		goto out_fail;
1121 	inet_hashinfo_init(&dccp_hashinfo);
1122 	rc = inet_hashinfo2_init_mod(&dccp_hashinfo);
1123 	if (rc)
1124 		goto out_free_percpu;
1125 	rc = -ENOBUFS;
1126 	dccp_hashinfo.bind_bucket_cachep =
1127 		kmem_cache_create("dccp_bind_bucket",
1128 				  sizeof(struct inet_bind_bucket), 0,
1129 				  SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, NULL);
1130 	if (!dccp_hashinfo.bind_bucket_cachep)
1131 		goto out_free_hashinfo2;
1132 
1133 	/*
1134 	 * Size and allocate the main established and bind bucket
1135 	 * hash tables.
1136 	 *
1137 	 * The methodology is similar to that of the buffer cache.
1138 	 */
1139 	if (nr_pages >= (128 * 1024))
1140 		goal = nr_pages >> (21 - PAGE_SHIFT);
1141 	else
1142 		goal = nr_pages >> (23 - PAGE_SHIFT);
1143 
1144 	if (thash_entries)
1145 		goal = (thash_entries *
1146 			sizeof(struct inet_ehash_bucket)) >> PAGE_SHIFT;
1147 	for (ehash_order = 0; (1UL << ehash_order) < goal; ehash_order++)
1148 		;
1149 	do {
1150 		unsigned long hash_size = (1UL << ehash_order) * PAGE_SIZE /
1151 					sizeof(struct inet_ehash_bucket);
1152 
1153 		while (hash_size & (hash_size - 1))
1154 			hash_size--;
1155 		dccp_hashinfo.ehash_mask = hash_size - 1;
1156 		dccp_hashinfo.ehash = (struct inet_ehash_bucket *)
1157 			__get_free_pages(GFP_ATOMIC|__GFP_NOWARN, ehash_order);
1158 	} while (!dccp_hashinfo.ehash && --ehash_order > 0);
1159 
1160 	if (!dccp_hashinfo.ehash) {
1161 		DCCP_CRIT("Failed to allocate DCCP established hash table");
1162 		goto out_free_bind_bucket_cachep;
1163 	}
1164 
1165 	for (i = 0; i <= dccp_hashinfo.ehash_mask; i++)
1166 		INIT_HLIST_NULLS_HEAD(&dccp_hashinfo.ehash[i].chain, i);
1167 
1168 	if (inet_ehash_locks_alloc(&dccp_hashinfo))
1169 			goto out_free_dccp_ehash;
1170 
1171 	bhash_order = ehash_order;
1172 
1173 	do {
1174 		dccp_hashinfo.bhash_size = (1UL << bhash_order) * PAGE_SIZE /
1175 					sizeof(struct inet_bind_hashbucket);
1176 		if ((dccp_hashinfo.bhash_size > (64 * 1024)) &&
1177 		    bhash_order > 0)
1178 			continue;
1179 		dccp_hashinfo.bhash = (struct inet_bind_hashbucket *)
1180 			__get_free_pages(GFP_ATOMIC|__GFP_NOWARN, bhash_order);
1181 	} while (!dccp_hashinfo.bhash && --bhash_order >= 0);
1182 
1183 	if (!dccp_hashinfo.bhash) {
1184 		DCCP_CRIT("Failed to allocate DCCP bind hash table");
1185 		goto out_free_dccp_locks;
1186 	}
1187 
1188 	for (i = 0; i < dccp_hashinfo.bhash_size; i++) {
1189 		spin_lock_init(&dccp_hashinfo.bhash[i].lock);
1190 		INIT_HLIST_HEAD(&dccp_hashinfo.bhash[i].chain);
1191 	}
1192 
1193 	rc = dccp_mib_init();
1194 	if (rc)
1195 		goto out_free_dccp_bhash;
1196 
1197 	rc = dccp_ackvec_init();
1198 	if (rc)
1199 		goto out_free_dccp_mib;
1200 
1201 	rc = dccp_sysctl_init();
1202 	if (rc)
1203 		goto out_ackvec_exit;
1204 
1205 	rc = ccid_initialize_builtins();
1206 	if (rc)
1207 		goto out_sysctl_exit;
1208 
1209 	dccp_timestamping_init();
1210 
1211 	return 0;
1212 
1213 out_sysctl_exit:
1214 	dccp_sysctl_exit();
1215 out_ackvec_exit:
1216 	dccp_ackvec_exit();
1217 out_free_dccp_mib:
1218 	dccp_mib_exit();
1219 out_free_dccp_bhash:
1220 	free_pages((unsigned long)dccp_hashinfo.bhash, bhash_order);
1221 out_free_dccp_locks:
1222 	inet_ehash_locks_free(&dccp_hashinfo);
1223 out_free_dccp_ehash:
1224 	free_pages((unsigned long)dccp_hashinfo.ehash, ehash_order);
1225 out_free_bind_bucket_cachep:
1226 	kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1227 out_free_hashinfo2:
1228 	inet_hashinfo2_free_mod(&dccp_hashinfo);
1229 out_free_percpu:
1230 	percpu_counter_destroy(&dccp_orphan_count);
1231 out_fail:
1232 	dccp_hashinfo.bhash = NULL;
1233 	dccp_hashinfo.ehash = NULL;
1234 	dccp_hashinfo.bind_bucket_cachep = NULL;
1235 	return rc;
1236 }
1237 
1238 static void __exit dccp_fini(void)
1239 {
1240 	ccid_cleanup_builtins();
1241 	dccp_mib_exit();
1242 	free_pages((unsigned long)dccp_hashinfo.bhash,
1243 		   get_order(dccp_hashinfo.bhash_size *
1244 			     sizeof(struct inet_bind_hashbucket)));
1245 	free_pages((unsigned long)dccp_hashinfo.ehash,
1246 		   get_order((dccp_hashinfo.ehash_mask + 1) *
1247 			     sizeof(struct inet_ehash_bucket)));
1248 	inet_ehash_locks_free(&dccp_hashinfo);
1249 	kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1250 	dccp_ackvec_exit();
1251 	dccp_sysctl_exit();
1252 	inet_hashinfo2_free_mod(&dccp_hashinfo);
1253 	percpu_counter_destroy(&dccp_orphan_count);
1254 }
1255 
1256 module_init(dccp_init);
1257 module_exit(dccp_fini);
1258 
1259 MODULE_LICENSE("GPL");
1260 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@conectiva.com.br>");
1261 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");
1262