xref: /xnu-8019.80.24/bsd/netinet/mptcp.c (revision a325d9c4a84054e40bbe985afedcb50ab80993ea)
1 /*
2  * Copyright (c) 2012-2021 Apple Inc. All rights reserved.
3  *
4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5  *
6  * This file contains Original Code and/or Modifications of Original Code
7  * as defined in and that are subject to the Apple Public Source License
8  * Version 2.0 (the 'License'). You may not use this file except in
9  * compliance with the License. The rights granted to you under the License
10  * may not be used to create, or enable the creation or redistribution of,
11  * unlawful or unlicensed copies of an Apple operating system, or to
12  * circumvent, violate, or enable the circumvention or violation of, any
13  * terms of an Apple operating system software license agreement.
14  *
15  * Please obtain a copy of the License at
16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
17  *
18  * The Original Code and all software distributed under the License are
19  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23  * Please see the License for the specific language governing rights and
24  * limitations under the License.
25  *
26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27  */
28 
29 /*
30  * A note on the MPTCP/NECP-interactions:
31  *
32  * MPTCP uses NECP-callbacks to get notified of interface/policy events.
33  * MPTCP registers to these events at the MPTCP-layer for interface-events
34  * through a call to necp_client_register_multipath_cb.
35  * To get per-flow events (aka per TCP-subflow), we register to it with
36  * necp_client_register_socket_flow. Both registrations happen by using the
37  * necp-client-uuid that comes from the app.
38  *
39  * The locking is rather tricky. In general, we expect the lock-ordering to
40  * happen from necp-fd -> necp->client -> mpp_lock.
41  *
42  * There are however some subtleties.
43  *
44  * 1. When registering the multipath_cb, we are holding the mpp_lock. This is
45  * safe, because it is the very first time this MPTCP-connection goes into NECP.
46  * As we go into NECP we take the NECP-locks and thus are guaranteed that no
47  * NECP-locks will deadlock us. Because these NECP-events will also first take
48  * the NECP-locks. Either they win the race and thus won't find our
49  * MPTCP-connection. Or, MPTCP wins the race and thus it will safely install
50  * the callbacks while holding the NECP lock.
51  *
52  * 2. When registering the subflow-callbacks we must unlock the mpp_lock. This,
53  * because we have already registered callbacks and we might race against an
54  * NECP-event that will match on our socket. So, we have to unlock to be safe.
55  *
56  * 3. When removing the multipath_cb, we do it in mp_pcbdispose(). The
57  * so_usecount has reached 0. We must be careful to not remove the mpp_socket
58  * pointers before we unregistered the callback. Because, again we might be
59  * racing against an NECP-event. Unregistering must happen with an unlocked
60  * mpp_lock, because of the lock-ordering constraint. It could be that
61  * before we had a chance to unregister an NECP-event triggers. That's why
62  * we need to check for the so_usecount in mptcp_session_necp_cb. If we get
63  * there while the socket is being garbage-collected, the use-count will go
64  * down to 0 and we exit. Removal of the multipath_cb again happens by taking
65  * the NECP-locks so any running NECP-events will finish first and exit cleanly.
66  *
67  * 4. When removing the subflow-callback, we do it in in_pcbdispose(). Again,
68  * the socket-lock must be unlocked for lock-ordering constraints. This gets a
69  * bit tricky here, as in tcp_garbage_collect we hold the mp_so and so lock.
70  * So, we drop the mp_so-lock as soon as the subflow is unlinked with
71  * mptcp_subflow_del. Then, in in_pcbdispose we drop the subflow-lock.
72  * If an NECP-event was waiting on the lock in mptcp_subflow_necp_cb, when it
73  * gets it, it will realize that the subflow became non-MPTCP and retry (see
74  * tcp_lock). Then it waits again on the subflow-lock. When we drop this lock
75  * in in_pcbdispose, and enter necp_inpcb_dispose, this one will have to wait
76  * for the NECP-lock (held by the other thread that is taking care of the NECP-
77  * event). So, the event now finally gets the subflow-lock and then hits an
78  * so_usecount that is 0 and exits. Eventually, we can remove the subflow from
79  * the NECP callback.
80  */
81 
82 #include <sys/param.h>
83 #include <sys/systm.h>
84 #include <sys/kernel.h>
85 #include <sys/mbuf.h>
86 #include <sys/mcache.h>
87 #include <sys/socket.h>
88 #include <sys/socketvar.h>
89 #include <sys/syslog.h>
90 #include <sys/protosw.h>
91 
92 #include <kern/zalloc.h>
93 #include <kern/locks.h>
94 
95 #include <mach/sdt.h>
96 
97 #include <net/if.h>
98 #include <netinet/in.h>
99 #include <netinet/in_var.h>
100 #include <netinet/tcp.h>
101 #include <netinet/tcp_fsm.h>
102 #include <netinet/tcp_seq.h>
103 #include <netinet/tcp_var.h>
104 #include <netinet/mptcp_var.h>
105 #include <netinet/mptcp.h>
106 #include <netinet/mptcp_seq.h>
107 #include <netinet/mptcp_opt.h>
108 #include <netinet/mptcp_timer.h>
109 
110 int mptcp_enable = 1;
111 SYSCTL_INT(_net_inet_mptcp, OID_AUTO, enable, CTLFLAG_RW | CTLFLAG_LOCKED,
112     &mptcp_enable, 0, "Enable Multipath TCP Support");
113 
114 /*
115  * Number of times to try negotiating MPTCP on SYN retransmissions.
116  * We haven't seen any reports of a middlebox that is dropping all SYN-segments
117  * that have an MPTCP-option. Thus, let's be generous and retransmit it 4 times.
118  */
119 int mptcp_mpcap_retries = 4;
120 SYSCTL_INT(_net_inet_mptcp, OID_AUTO, mptcp_cap_retr,
121     CTLFLAG_RW | CTLFLAG_LOCKED,
122     &mptcp_mpcap_retries, 0, "Number of MP Capable SYN Retries");
123 
124 /*
125  * By default, DSS checksum is turned off, revisit if we ever do
126  * MPTCP for non SSL Traffic.
127  */
128 int mptcp_dss_csum = 0;
129 SYSCTL_INT(_net_inet_mptcp, OID_AUTO, dss_csum, CTLFLAG_RW | CTLFLAG_LOCKED,
130     &mptcp_dss_csum, 0, "Enable DSS checksum");
131 
132 /*
133  * When mptcp_fail_thresh number of retransmissions are sent, subflow failover
134  * is attempted on a different path.
135  */
136 int mptcp_fail_thresh = 1;
137 SYSCTL_INT(_net_inet_mptcp, OID_AUTO, fail, CTLFLAG_RW | CTLFLAG_LOCKED,
138     &mptcp_fail_thresh, 0, "Failover threshold");
139 
140 /*
141  * MPTCP subflows have TCP keepalives set to ON. Set a conservative keeptime
142  * as carrier networks mostly have a 30 minute to 60 minute NAT Timeout.
143  * Some carrier networks have a timeout of 10 or 15 minutes.
144  */
145 int mptcp_subflow_keeptime = 60 * 14;
146 SYSCTL_INT(_net_inet_mptcp, OID_AUTO, keepalive, CTLFLAG_RW | CTLFLAG_LOCKED,
147     &mptcp_subflow_keeptime, 0, "Keepalive in seconds");
148 
149 int mptcp_rtthist_rtthresh = 600;
150 SYSCTL_INT(_net_inet_mptcp, OID_AUTO, rtthist_thresh, CTLFLAG_RW | CTLFLAG_LOCKED,
151     &mptcp_rtthist_rtthresh, 0, "Rtt threshold");
152 
153 int mptcp_rtothresh = 1500;
154 SYSCTL_INT(_net_inet_mptcp, OID_AUTO, rto_thresh, CTLFLAG_RW | CTLFLAG_LOCKED,
155     &mptcp_rtothresh, 0, "RTO threshold");
156 
157 /*
158  * Probe the preferred path, when it is not in use
159  */
160 uint32_t mptcp_probeto = 1000;
161 SYSCTL_UINT(_net_inet_mptcp, OID_AUTO, probeto, CTLFLAG_RW | CTLFLAG_LOCKED,
162     &mptcp_probeto, 0, "Disable probing by setting to 0");
163 
164 uint32_t mptcp_probecnt = 5;
165 SYSCTL_UINT(_net_inet_mptcp, OID_AUTO, probecnt, CTLFLAG_RW | CTLFLAG_LOCKED,
166     &mptcp_probecnt, 0, "Number of probe writes");
167 
168 uint32_t mptcp_enable_v1 = 1;
169 SYSCTL_UINT(_net_inet_mptcp, OID_AUTO, enable_v1, CTLFLAG_RW | CTLFLAG_LOCKED,
170     &mptcp_enable_v1, 0, "Enable or disable v1");
171 
172 static int
173 sysctl_mptcp_version_check SYSCTL_HANDLER_ARGS
174 {
175 #pragma unused(arg1, arg2)
176 	int error;
177 	int new_value = *(int *)oidp->oid_arg1;
178 	int old_value = *(int *)oidp->oid_arg1;
179 
180 	error = sysctl_handle_int(oidp, &new_value, 0, req);
181 	if (!error) {
182 		if (new_value != MPTCP_VERSION_0 && new_value != MPTCP_VERSION_1) {
183 			return EINVAL;
184 		}
185 		*(int *)oidp->oid_arg1 = new_value;
186 	}
187 
188 	os_log(OS_LOG_DEFAULT,
189 	    "%s:%u sysctl net.inet.tcp.mptcp_preferred_version: %d -> %d)",
190 	    proc_best_name(current_proc()), proc_selfpid(),
191 	    old_value, *(int *)oidp->oid_arg1);
192 
193 	return error;
194 }
195 
196 int mptcp_preferred_version = MPTCP_VERSION_0;
197 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, mptcp_preferred_version,
198     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
199     &mptcp_preferred_version, 0, &sysctl_mptcp_version_check, "I", "");
200 
201 
202 static int
mptcp_reass_present(struct socket * mp_so)203 mptcp_reass_present(struct socket *mp_so)
204 {
205 	struct mptses *mpte = mpsotompte(mp_so);
206 	struct mptcb *mp_tp = mpte->mpte_mptcb;
207 	struct tseg_qent *q;
208 	int dowakeup = 0;
209 	int flags = 0;
210 
211 	/*
212 	 * Present data to user, advancing rcv_nxt through
213 	 * completed sequence space.
214 	 */
215 	if (mp_tp->mpt_state < MPTCPS_ESTABLISHED) {
216 		return flags;
217 	}
218 	q = LIST_FIRST(&mp_tp->mpt_segq);
219 	if (!q || q->tqe_m->m_pkthdr.mp_dsn != mp_tp->mpt_rcvnxt) {
220 		return flags;
221 	}
222 
223 	/*
224 	 * If there is already another thread doing reassembly for this
225 	 * connection, it is better to let it finish the job --
226 	 * (radar 16316196)
227 	 */
228 	if (mp_tp->mpt_flags & MPTCPF_REASS_INPROG) {
229 		return flags;
230 	}
231 
232 	mp_tp->mpt_flags |= MPTCPF_REASS_INPROG;
233 
234 	do {
235 		mp_tp->mpt_rcvnxt += q->tqe_len;
236 		LIST_REMOVE(q, tqe_q);
237 		if (mp_so->so_state & SS_CANTRCVMORE) {
238 			m_freem(q->tqe_m);
239 		} else {
240 			flags = !!(q->tqe_m->m_pkthdr.pkt_flags & PKTF_MPTCP_DFIN);
241 			if (sbappendstream_rcvdemux(mp_so, q->tqe_m)) {
242 				dowakeup = 1;
243 			}
244 		}
245 		zfree(tcp_reass_zone, q);
246 		mp_tp->mpt_reassqlen--;
247 		q = LIST_FIRST(&mp_tp->mpt_segq);
248 	} while (q && q->tqe_m->m_pkthdr.mp_dsn == mp_tp->mpt_rcvnxt);
249 	mp_tp->mpt_flags &= ~MPTCPF_REASS_INPROG;
250 
251 	if (dowakeup) {
252 		sorwakeup(mp_so); /* done with socket lock held */
253 	}
254 	return flags;
255 }
256 
257 static int
mptcp_reass(struct socket * mp_so,struct pkthdr * phdr,int * tlenp,struct mbuf * m)258 mptcp_reass(struct socket *mp_so, struct pkthdr *phdr, int *tlenp, struct mbuf *m)
259 {
260 	struct mptcb *mp_tp = mpsotomppcb(mp_so)->mpp_pcbe->mpte_mptcb;
261 	u_int64_t mb_dsn = phdr->mp_dsn;
262 	struct tseg_qent *q;
263 	struct tseg_qent *p = NULL;
264 	struct tseg_qent *nq;
265 	struct tseg_qent *te = NULL;
266 	uint32_t qlimit;
267 
268 	/*
269 	 * Limit the number of segments in the reassembly queue to prevent
270 	 * holding on to too many segments (and thus running out of mbufs).
271 	 * Make sure to let the missing segment through which caused this
272 	 * queue.  Always keep one global queue entry spare to be able to
273 	 * process the missing segment.
274 	 */
275 	qlimit = MIN(MAX(100, mp_so->so_rcv.sb_hiwat >> 10),
276 	    (tcp_autorcvbuf_max >> 10));
277 	if (mb_dsn != mp_tp->mpt_rcvnxt &&
278 	    (mp_tp->mpt_reassqlen + 1) >= qlimit) {
279 		tcpstat.tcps_mptcp_rcvmemdrop++;
280 		m_freem(m);
281 		*tlenp = 0;
282 		return 0;
283 	}
284 
285 	/* Allocate a new queue entry. If we can't, just drop the pkt. XXX */
286 	te = zalloc_flags(tcp_reass_zone, Z_WAITOK | Z_NOFAIL);
287 
288 	mp_tp->mpt_reassqlen++;
289 
290 	/*
291 	 * Find a segment which begins after this one does.
292 	 */
293 	LIST_FOREACH(q, &mp_tp->mpt_segq, tqe_q) {
294 		if (MPTCP_SEQ_GT(q->tqe_m->m_pkthdr.mp_dsn, mb_dsn)) {
295 			break;
296 		}
297 		p = q;
298 	}
299 
300 	/*
301 	 * If there is a preceding segment, it may provide some of
302 	 * our data already.  If so, drop the data from the incoming
303 	 * segment.  If it provides all of our data, drop us.
304 	 */
305 	if (p != NULL) {
306 		int64_t i;
307 		/* conversion to int (in i) handles seq wraparound */
308 		i = p->tqe_m->m_pkthdr.mp_dsn + p->tqe_len - mb_dsn;
309 		if (i > 0) {
310 			if (i >= *tlenp) {
311 				tcpstat.tcps_mptcp_rcvduppack++;
312 				m_freem(m);
313 				zfree(tcp_reass_zone, te);
314 				te = NULL;
315 				mp_tp->mpt_reassqlen--;
316 				/*
317 				 * Try to present any queued data
318 				 * at the left window edge to the user.
319 				 * This is needed after the 3-WHS
320 				 * completes.
321 				 */
322 				goto out;
323 			}
324 			VERIFY(i <= INT_MAX);
325 			m_adj(m, (int)i);
326 			*tlenp -= i;
327 			phdr->mp_dsn += i;
328 		}
329 	}
330 
331 	tcpstat.tcps_mp_oodata++;
332 
333 	/*
334 	 * While we overlap succeeding segments trim them or,
335 	 * if they are completely covered, dequeue them.
336 	 */
337 	while (q) {
338 		int64_t i = (mb_dsn + *tlenp) - q->tqe_m->m_pkthdr.mp_dsn;
339 		if (i <= 0) {
340 			break;
341 		}
342 
343 		if (i < q->tqe_len) {
344 			q->tqe_m->m_pkthdr.mp_dsn += i;
345 			q->tqe_len -= i;
346 
347 			VERIFY(i <= INT_MAX);
348 			m_adj(q->tqe_m, (int)i);
349 			break;
350 		}
351 
352 		nq = LIST_NEXT(q, tqe_q);
353 		LIST_REMOVE(q, tqe_q);
354 		m_freem(q->tqe_m);
355 		zfree(tcp_reass_zone, q);
356 		mp_tp->mpt_reassqlen--;
357 		q = nq;
358 	}
359 
360 	/* Insert the new segment queue entry into place. */
361 	te->tqe_m = m;
362 	te->tqe_th = NULL;
363 	te->tqe_len = *tlenp;
364 
365 	if (p == NULL) {
366 		LIST_INSERT_HEAD(&mp_tp->mpt_segq, te, tqe_q);
367 	} else {
368 		LIST_INSERT_AFTER(p, te, tqe_q);
369 	}
370 
371 out:
372 	return mptcp_reass_present(mp_so);
373 }
374 
375 /*
376  * MPTCP input, called when data has been read from a subflow socket.
377  */
378 void
mptcp_input(struct mptses * mpte,struct mbuf * m)379 mptcp_input(struct mptses *mpte, struct mbuf *m)
380 {
381 	struct socket *mp_so;
382 	struct mptcb *mp_tp = NULL;
383 	int count = 0, wakeup = 0;
384 	struct mbuf *save = NULL, *prev = NULL;
385 	struct mbuf *freelist = NULL, *tail = NULL;
386 
387 	VERIFY(m->m_flags & M_PKTHDR);
388 
389 	mp_so = mptetoso(mpte);
390 	mp_tp = mpte->mpte_mptcb;
391 
392 	socket_lock_assert_owned(mp_so);
393 
394 	DTRACE_MPTCP(input);
395 
396 	mp_tp->mpt_rcvwnd = mptcp_sbspace(mp_tp);
397 
398 	/*
399 	 * Each mbuf contains MPTCP Data Sequence Map
400 	 * Process the data for reassembly, delivery to MPTCP socket
401 	 * client, etc.
402 	 *
403 	 */
404 	count = mp_so->so_rcv.sb_cc;
405 
406 	/*
407 	 * In the degraded fallback case, data is accepted without DSS map
408 	 */
409 	if (mp_tp->mpt_flags & MPTCPF_FALLBACK_TO_TCP) {
410 		struct mbuf *iter;
411 		int mb_dfin;
412 fallback:
413 		mb_dfin = 0;
414 		mptcp_sbrcv_grow(mp_tp);
415 
416 		iter = m;
417 		while (iter) {
418 			if ((iter->m_flags & M_PKTHDR) &&
419 			    (iter->m_pkthdr.pkt_flags & PKTF_MPTCP_DFIN)) {
420 				mb_dfin = 1;
421 			}
422 
423 			if ((iter->m_flags & M_PKTHDR) && m_pktlen(iter) == 0) {
424 				/* Don't add zero-length packets, so jump it! */
425 				if (prev == NULL) {
426 					m = iter->m_next;
427 					m_free(iter);
428 					iter = m;
429 				} else {
430 					prev->m_next = iter->m_next;
431 					m_free(iter);
432 					iter = prev->m_next;
433 				}
434 
435 				/* It was a zero-length packet so next one must be a pkthdr */
436 				VERIFY(iter == NULL || iter->m_flags & M_PKTHDR);
437 			} else {
438 				prev = iter;
439 				iter = iter->m_next;
440 			}
441 		}
442 
443 		/*
444 		 * assume degraded flow as this may be the first packet
445 		 * without DSS, and the subflow state is not updated yet.
446 		 */
447 		if (sbappendstream_rcvdemux(mp_so, m)) {
448 			sorwakeup(mp_so);
449 		}
450 
451 		DTRACE_MPTCP5(receive__degraded, struct mbuf *, m,
452 		    struct socket *, mp_so,
453 		    struct sockbuf *, &mp_so->so_rcv,
454 		    struct sockbuf *, &mp_so->so_snd,
455 		    struct mptses *, mpte);
456 		count = mp_so->so_rcv.sb_cc - count;
457 
458 		mp_tp->mpt_rcvnxt += count;
459 
460 		if (mb_dfin) {
461 			mptcp_close_fsm(mp_tp, MPCE_RECV_DATA_FIN);
462 			socantrcvmore(mp_so);
463 		}
464 		return;
465 	}
466 
467 	do {
468 		u_int64_t mb_dsn;
469 		int32_t mb_datalen;
470 		int64_t todrop;
471 		int mb_dfin = 0;
472 
473 		VERIFY(m->m_flags & M_PKTHDR);
474 
475 		/* If fallback occurs, mbufs will not have PKTF_MPTCP set */
476 		if (!(m->m_pkthdr.pkt_flags & PKTF_MPTCP)) {
477 			goto fallback;
478 		}
479 
480 		save = m->m_next;
481 		/*
482 		 * A single TCP packet formed of multiple mbufs
483 		 * holds DSS mapping in the first mbuf of the chain.
484 		 * Other mbufs in the chain may have M_PKTHDR set
485 		 * even though they belong to the same TCP packet
486 		 * and therefore use the DSS mapping stored in the
487 		 * first mbuf of the mbuf chain. mptcp_input() can
488 		 * get an mbuf chain with multiple TCP packets.
489 		 */
490 		while (save && (!(save->m_flags & M_PKTHDR) ||
491 		    !(save->m_pkthdr.pkt_flags & PKTF_MPTCP))) {
492 			prev = save;
493 			save = save->m_next;
494 		}
495 		if (prev) {
496 			prev->m_next = NULL;
497 		} else {
498 			m->m_next = NULL;
499 		}
500 
501 		mb_dsn = m->m_pkthdr.mp_dsn;
502 		mb_datalen = m->m_pkthdr.mp_rlen;
503 
504 		todrop = (mb_dsn + mb_datalen) - (mp_tp->mpt_rcvnxt + mp_tp->mpt_rcvwnd);
505 		if (todrop > 0) {
506 			tcpstat.tcps_mptcp_rcvpackafterwin++;
507 
508 			os_log_info(mptcp_log_handle, "%s - %lx: dropping dsn %u dlen %u rcvnxt %u rcvwnd %u todrop %lld\n",
509 			    __func__, (unsigned long)VM_KERNEL_ADDRPERM(mpte),
510 			    (uint32_t)mb_dsn, mb_datalen, (uint32_t)mp_tp->mpt_rcvnxt,
511 			    mp_tp->mpt_rcvwnd, todrop);
512 
513 			if (todrop >= mb_datalen) {
514 				if (freelist == NULL) {
515 					freelist = m;
516 				} else {
517 					tail->m_next = m;
518 				}
519 
520 				if (prev != NULL) {
521 					tail = prev;
522 				} else {
523 					tail = m;
524 				}
525 
526 				m = save;
527 				prev = save = NULL;
528 				continue;
529 			} else {
530 				VERIFY(todrop <= INT_MAX);
531 				m_adj(m, (int)-todrop);
532 				mb_datalen -= todrop;
533 				m->m_pkthdr.mp_rlen -= todrop;
534 			}
535 
536 			/*
537 			 * We drop from the right edge of the mbuf, thus the
538 			 * DATA_FIN is dropped as well
539 			 */
540 			m->m_pkthdr.pkt_flags &= ~PKTF_MPTCP_DFIN;
541 		}
542 
543 		if (MPTCP_SEQ_LT(mb_dsn, mp_tp->mpt_rcvnxt)) {
544 			if (MPTCP_SEQ_LEQ((mb_dsn + mb_datalen),
545 			    mp_tp->mpt_rcvnxt)) {
546 				if (freelist == NULL) {
547 					freelist = m;
548 				} else {
549 					tail->m_next = m;
550 				}
551 
552 				if (prev != NULL) {
553 					tail = prev;
554 				} else {
555 					tail = m;
556 				}
557 
558 				m = save;
559 				prev = save = NULL;
560 				continue;
561 			} else {
562 				VERIFY((mp_tp->mpt_rcvnxt - mb_dsn) <= INT_MAX);
563 				m_adj(m, (int)(mp_tp->mpt_rcvnxt - mb_dsn));
564 				mb_datalen -= (mp_tp->mpt_rcvnxt - mb_dsn);
565 				mb_dsn = mp_tp->mpt_rcvnxt;
566 				VERIFY(mb_datalen >= 0 && mb_datalen <= USHRT_MAX);
567 				m->m_pkthdr.mp_rlen = (uint16_t)mb_datalen;
568 				m->m_pkthdr.mp_dsn = mb_dsn;
569 			}
570 		}
571 
572 		if (MPTCP_SEQ_GT(mb_dsn, mp_tp->mpt_rcvnxt) ||
573 		    !LIST_EMPTY(&mp_tp->mpt_segq)) {
574 			mb_dfin = mptcp_reass(mp_so, &m->m_pkthdr, &mb_datalen, m);
575 
576 			goto next;
577 		}
578 		mb_dfin = !!(m->m_pkthdr.pkt_flags & PKTF_MPTCP_DFIN);
579 
580 		mptcp_sbrcv_grow(mp_tp);
581 
582 		if (sbappendstream_rcvdemux(mp_so, m)) {
583 			wakeup = 1;
584 		}
585 
586 		DTRACE_MPTCP6(receive, struct mbuf *, m, struct socket *, mp_so,
587 		    struct sockbuf *, &mp_so->so_rcv,
588 		    struct sockbuf *, &mp_so->so_snd,
589 		    struct mptses *, mpte,
590 		    struct mptcb *, mp_tp);
591 		count = mp_so->so_rcv.sb_cc - count;
592 		tcpstat.tcps_mp_rcvtotal++;
593 		tcpstat.tcps_mp_rcvbytes += count;
594 
595 		mp_tp->mpt_rcvnxt += count;
596 
597 next:
598 		if (mb_dfin) {
599 			mptcp_close_fsm(mp_tp, MPCE_RECV_DATA_FIN);
600 			socantrcvmore(mp_so);
601 		}
602 		m = save;
603 		prev = save = NULL;
604 		count = mp_so->so_rcv.sb_cc;
605 	} while (m);
606 
607 	if (freelist) {
608 		m_freem(freelist);
609 	}
610 
611 	if (wakeup) {
612 		sorwakeup(mp_so);
613 	}
614 }
615 
616 boolean_t
mptcp_can_send_more(struct mptcb * mp_tp,boolean_t ignore_reinject)617 mptcp_can_send_more(struct mptcb *mp_tp, boolean_t ignore_reinject)
618 {
619 	struct socket *mp_so = mptetoso(mp_tp->mpt_mpte);
620 
621 	/*
622 	 * Always send if there is data in the reinject-queue.
623 	 */
624 	if (!ignore_reinject && mp_tp->mpt_mpte->mpte_reinjectq) {
625 		return TRUE;
626 	}
627 
628 	/*
629 	 * Don't send, if:
630 	 *
631 	 * 1. snd_nxt >= snd_max : Means, basically everything has been sent.
632 	 *    Except when using TFO, we might be doing a 0-byte write.
633 	 * 2. snd_una + snd_wnd <= snd_nxt: No space in the receiver's window
634 	 * 3. snd_nxt + 1 == snd_max and we are closing: A DATA_FIN is scheduled.
635 	 */
636 
637 	if (!(mp_so->so_flags1 & SOF1_PRECONNECT_DATA) && MPTCP_SEQ_GEQ(mp_tp->mpt_sndnxt, mp_tp->mpt_sndmax)) {
638 		return FALSE;
639 	}
640 
641 	if (MPTCP_SEQ_LEQ(mp_tp->mpt_snduna + mp_tp->mpt_sndwnd, mp_tp->mpt_sndnxt)) {
642 		return FALSE;
643 	}
644 
645 	if (mp_tp->mpt_sndnxt + 1 == mp_tp->mpt_sndmax && mp_tp->mpt_state > MPTCPS_CLOSE_WAIT) {
646 		return FALSE;
647 	}
648 
649 	if (mp_tp->mpt_state >= MPTCPS_FIN_WAIT_2) {
650 		return FALSE;
651 	}
652 
653 	return TRUE;
654 }
655 
656 /*
657  * MPTCP output.
658  */
659 int
mptcp_output(struct mptses * mpte)660 mptcp_output(struct mptses *mpte)
661 {
662 	struct mptcb *mp_tp;
663 	struct mptsub *mpts;
664 	struct mptsub *mpts_tried = NULL;
665 	struct socket *mp_so;
666 	struct mptsub *preferred_mpts = NULL;
667 	uint64_t old_snd_nxt;
668 	int error = 0;
669 
670 	mp_so = mptetoso(mpte);
671 	mp_tp = mpte->mpte_mptcb;
672 
673 	socket_lock_assert_owned(mp_so);
674 
675 	if (mp_so->so_flags & SOF_DEFUNCT) {
676 		return 0;
677 	}
678 
679 	VERIFY(!(mpte->mpte_mppcb->mpp_flags & MPP_WUPCALL));
680 	mpte->mpte_mppcb->mpp_flags |= MPP_WUPCALL;
681 
682 	old_snd_nxt = mp_tp->mpt_sndnxt;
683 	while (mptcp_can_send_more(mp_tp, FALSE)) {
684 		/* get the "best" subflow to be used for transmission */
685 		mpts = mptcp_get_subflow(mpte, &preferred_mpts);
686 		if (mpts == NULL) {
687 			mptcplog((LOG_INFO, "%s: no subflow\n", __func__),
688 			    MPTCP_SENDER_DBG, MPTCP_LOGLVL_LOG);
689 			break;
690 		}
691 
692 		/* In case there's just one flow, we reattempt later */
693 		if (mpts_tried != NULL &&
694 		    (mpts == mpts_tried || (mpts->mpts_flags & MPTSF_FAILINGOVER))) {
695 			mpts_tried->mpts_flags &= ~MPTSF_FAILINGOVER;
696 			mpts_tried->mpts_flags |= MPTSF_ACTIVE;
697 			mptcp_start_timer(mpte, MPTT_REXMT);
698 			break;
699 		}
700 
701 		/*
702 		 * Automatic sizing of send socket buffer. Increase the send
703 		 * socket buffer size if all of the following criteria are met
704 		 *	1. the receiver has enough buffer space for this data
705 		 *	2. send buffer is filled to 7/8th with data (so we actually
706 		 *	   have data to make use of it);
707 		 */
708 		if ((mp_so->so_snd.sb_flags & (SB_AUTOSIZE | SB_TRIM)) == SB_AUTOSIZE &&
709 		    tcp_cansbgrow(&mp_so->so_snd)) {
710 			if ((mp_tp->mpt_sndwnd / 4 * 5) >= mp_so->so_snd.sb_hiwat &&
711 			    mp_so->so_snd.sb_cc >= (mp_so->so_snd.sb_hiwat / 8 * 7)) {
712 				if (sbreserve(&mp_so->so_snd,
713 				    min(mp_so->so_snd.sb_hiwat + tcp_autosndbuf_inc,
714 				    tcp_autosndbuf_max)) == 1) {
715 					mp_so->so_snd.sb_idealsize = mp_so->so_snd.sb_hiwat;
716 				}
717 			}
718 		}
719 
720 		DTRACE_MPTCP3(output, struct mptses *, mpte, struct mptsub *, mpts,
721 		    struct socket *, mp_so);
722 		error = mptcp_subflow_output(mpte, mpts, 0);
723 		if (error) {
724 			/* can be a temporary loss of source address or other error */
725 			mpts->mpts_flags |= MPTSF_FAILINGOVER;
726 			mpts->mpts_flags &= ~MPTSF_ACTIVE;
727 			mpts_tried = mpts;
728 			if (error != ECANCELED) {
729 				os_log_error(mptcp_log_handle, "%s - %lx: Error = %d mpts_flags %#x\n",
730 				    __func__, (unsigned long)VM_KERNEL_ADDRPERM(mpte),
731 				    error, mpts->mpts_flags);
732 			}
733 			break;
734 		}
735 		/* The model is to have only one active flow at a time */
736 		mpts->mpts_flags |= MPTSF_ACTIVE;
737 		mpts->mpts_probesoon = mpts->mpts_probecnt = 0;
738 
739 		/* Allows us to update the smoothed rtt */
740 		if (mptcp_probeto && mpts != preferred_mpts && preferred_mpts != NULL) {
741 			if (preferred_mpts->mpts_probesoon) {
742 				if ((tcp_now - preferred_mpts->mpts_probesoon) > mptcp_probeto) {
743 					mptcp_subflow_output(mpte, preferred_mpts, MPTCP_SUBOUT_PROBING);
744 					if (preferred_mpts->mpts_probecnt >= mptcp_probecnt) {
745 						preferred_mpts->mpts_probesoon = 0;
746 						preferred_mpts->mpts_probecnt = 0;
747 					}
748 				}
749 			} else {
750 				preferred_mpts->mpts_probesoon = tcp_now;
751 				preferred_mpts->mpts_probecnt = 0;
752 			}
753 		}
754 
755 		if (mpte->mpte_active_sub == NULL) {
756 			mpte->mpte_active_sub = mpts;
757 		} else if (mpte->mpte_active_sub != mpts) {
758 			mpte->mpte_active_sub->mpts_flags &= ~MPTSF_ACTIVE;
759 			mpte->mpte_active_sub = mpts;
760 
761 			mptcpstats_inc_switch(mpte, mpts);
762 		}
763 	}
764 
765 	if (mp_tp->mpt_state > MPTCPS_CLOSE_WAIT) {
766 		if (mp_tp->mpt_sndnxt + 1 == mp_tp->mpt_sndmax &&
767 		    mp_tp->mpt_snduna == mp_tp->mpt_sndnxt) {
768 			mptcp_finish_usrclosed(mpte);
769 		}
770 	}
771 
772 	mptcp_handle_deferred_upcalls(mpte->mpte_mppcb, MPP_WUPCALL);
773 
774 	/* subflow errors should not be percolated back up */
775 	return 0;
776 }
777 
778 
779 static struct mptsub *
mptcp_choose_subflow(struct mptsub * mpts,struct mptsub * curbest,int * currtt)780 mptcp_choose_subflow(struct mptsub *mpts, struct mptsub *curbest, int *currtt)
781 {
782 	struct tcpcb *tp = sototcpcb(mpts->mpts_socket);
783 
784 	/*
785 	 * Lower RTT? Take it, if it's our first one, or
786 	 * it doesn't has any loss, or the current one has
787 	 * loss as well.
788 	 */
789 	if (tp->t_srtt && *currtt > tp->t_srtt &&
790 	    (curbest == NULL || tp->t_rxtshift == 0 ||
791 	    sototcpcb(curbest->mpts_socket)->t_rxtshift)) {
792 		*currtt = tp->t_srtt;
793 		return mpts;
794 	}
795 
796 	/*
797 	 * If we find a subflow without loss, take it always!
798 	 */
799 	if (curbest &&
800 	    sototcpcb(curbest->mpts_socket)->t_rxtshift &&
801 	    tp->t_rxtshift == 0) {
802 		*currtt = tp->t_srtt;
803 		return mpts;
804 	}
805 
806 	return curbest != NULL ? curbest : mpts;
807 }
808 
809 static struct mptsub *
mptcp_return_subflow(struct mptsub * mpts)810 mptcp_return_subflow(struct mptsub *mpts)
811 {
812 	if (mpts && mptcp_subflow_cwnd_space(mpts->mpts_socket) <= 0) {
813 		return NULL;
814 	}
815 
816 	return mpts;
817 }
818 
819 static boolean_t
mptcp_subflow_is_slow(struct mptses * mpte,struct mptsub * mpts)820 mptcp_subflow_is_slow(struct mptses *mpte, struct mptsub *mpts)
821 {
822 	struct tcpcb *tp = sototcpcb(mpts->mpts_socket);
823 	int fail_thresh = mptcp_fail_thresh;
824 
825 	if (mpte->mpte_svctype == MPTCP_SVCTYPE_HANDOVER || mpte->mpte_svctype == MPTCP_SVCTYPE_PURE_HANDOVER) {
826 		fail_thresh *= 2;
827 	}
828 
829 	return tp->t_rxtshift >= fail_thresh &&
830 	       (mptetoso(mpte)->so_snd.sb_cc || mpte->mpte_reinjectq);
831 }
832 
833 /*
834  * Return the most eligible subflow to be used for sending data.
835  */
836 struct mptsub *
mptcp_get_subflow(struct mptses * mpte,struct mptsub ** preferred)837 mptcp_get_subflow(struct mptses *mpte, struct mptsub **preferred)
838 {
839 	struct tcpcb *besttp, *secondtp;
840 	struct inpcb *bestinp, *secondinp;
841 	struct mptsub *mpts;
842 	struct mptsub *best = NULL;
843 	struct mptsub *second_best = NULL;
844 	int exp_rtt = INT_MAX, cheap_rtt = INT_MAX;
845 
846 	/*
847 	 * First Step:
848 	 * Choose the best subflow for cellular and non-cellular interfaces.
849 	 */
850 
851 	TAILQ_FOREACH(mpts, &mpte->mpte_subflows, mpts_entry) {
852 		struct socket *so = mpts->mpts_socket;
853 		struct tcpcb *tp = sototcpcb(so);
854 		struct inpcb *inp = sotoinpcb(so);
855 
856 		mptcplog((LOG_DEBUG, "%s mpts %u mpts_flags %#x, suspended %u sostate %#x tpstate %u cellular %d rtt %u rxtshift %u cheap %u exp %u cwnd %d\n",
857 		    __func__, mpts->mpts_connid, mpts->mpts_flags,
858 		    INP_WAIT_FOR_IF_FEEDBACK(inp), so->so_state, tp->t_state,
859 		    inp->inp_last_outifp ? IFNET_IS_CELLULAR(inp->inp_last_outifp) : -1,
860 		    tp->t_srtt, tp->t_rxtshift, cheap_rtt, exp_rtt,
861 		    mptcp_subflow_cwnd_space(so)),
862 		    MPTCP_SOCKET_DBG, MPTCP_LOGLVL_VERBOSE);
863 
864 		/*
865 		 * First, the hard conditions to reject subflows
866 		 * (e.g., not connected,...)
867 		 */
868 		if (inp->inp_last_outifp == NULL) {
869 			continue;
870 		}
871 
872 		if (INP_WAIT_FOR_IF_FEEDBACK(inp)) {
873 			continue;
874 		}
875 
876 		/* There can only be one subflow in degraded state */
877 		if (mpts->mpts_flags & MPTSF_MP_DEGRADED) {
878 			best = mpts;
879 			break;
880 		}
881 
882 		/*
883 		 * If this subflow is waiting to finally send, do it!
884 		 */
885 		if (so->so_flags1 & SOF1_PRECONNECT_DATA) {
886 			return mptcp_return_subflow(mpts);
887 		}
888 
889 		/*
890 		 * Only send if the subflow is MP_CAPABLE. The exceptions to
891 		 * this rule (degraded or TFO) have been taken care of above.
892 		 */
893 		if (!(mpts->mpts_flags & MPTSF_MP_CAPABLE)) {
894 			continue;
895 		}
896 
897 		if ((so->so_state & SS_ISDISCONNECTED) ||
898 		    !(so->so_state & SS_ISCONNECTED) ||
899 		    !TCPS_HAVEESTABLISHED(tp->t_state) ||
900 		    tp->t_state > TCPS_CLOSE_WAIT) {
901 			continue;
902 		}
903 
904 		/*
905 		 * Second, the soft conditions to find the subflow with best
906 		 * conditions for each set (aka cellular vs non-cellular)
907 		 */
908 		if (IFNET_IS_CELLULAR(inp->inp_last_outifp)) {
909 			second_best = mptcp_choose_subflow(mpts, second_best,
910 			    &exp_rtt);
911 		} else {
912 			best = mptcp_choose_subflow(mpts, best, &cheap_rtt);
913 		}
914 	}
915 
916 	/*
917 	 * If there is no preferred or backup subflow, and there is no active
918 	 * subflow use the last usable subflow.
919 	 */
920 	if (best == NULL) {
921 		return mptcp_return_subflow(second_best);
922 	}
923 
924 	if (second_best == NULL) {
925 		return mptcp_return_subflow(best);
926 	}
927 
928 	besttp = sototcpcb(best->mpts_socket);
929 	bestinp = sotoinpcb(best->mpts_socket);
930 	secondtp = sototcpcb(second_best->mpts_socket);
931 	secondinp = sotoinpcb(second_best->mpts_socket);
932 
933 	if (preferred != NULL) {
934 		*preferred = mptcp_return_subflow(best);
935 	}
936 
937 	/*
938 	 * Second Step: Among best and second_best. Choose the one that is
939 	 * most appropriate for this particular service-type.
940 	 */
941 	if (mpte->mpte_svctype == MPTCP_SVCTYPE_PURE_HANDOVER) {
942 		return mptcp_return_subflow(best);
943 	} else if (mpte->mpte_svctype == MPTCP_SVCTYPE_HANDOVER) {
944 		/*
945 		 * Only handover if Symptoms tells us to do so.
946 		 */
947 		if (!IFNET_IS_CELLULAR(bestinp->inp_last_outifp) &&
948 		    mptcp_is_wifi_unusable_for_session(mpte) != 0 && mptcp_subflow_is_slow(mpte, best)) {
949 			return mptcp_return_subflow(second_best);
950 		}
951 
952 		return mptcp_return_subflow(best);
953 	} else if (mpte->mpte_svctype == MPTCP_SVCTYPE_INTERACTIVE) {
954 		int rtt_thresh = mptcp_rtthist_rtthresh << TCP_RTT_SHIFT;
955 		int rto_thresh = mptcp_rtothresh;
956 
957 		/* Adjust with symptoms information */
958 		if (!IFNET_IS_CELLULAR(bestinp->inp_last_outifp) &&
959 		    mptcp_is_wifi_unusable_for_session(mpte) != 0) {
960 			rtt_thresh /= 2;
961 			rto_thresh /= 2;
962 		}
963 
964 		if (besttp->t_srtt && secondtp->t_srtt &&
965 		    besttp->t_srtt >= rtt_thresh &&
966 		    secondtp->t_srtt < rtt_thresh) {
967 			tcpstat.tcps_mp_sel_rtt++;
968 			mptcplog((LOG_DEBUG, "%s: best cid %d at rtt %d,  second cid %d at rtt %d\n", __func__,
969 			    best->mpts_connid, besttp->t_srtt >> TCP_RTT_SHIFT,
970 			    second_best->mpts_connid,
971 			    secondtp->t_srtt >> TCP_RTT_SHIFT),
972 			    MPTCP_SENDER_DBG, MPTCP_LOGLVL_LOG);
973 			return mptcp_return_subflow(second_best);
974 		}
975 
976 		if (mptcp_subflow_is_slow(mpte, best) &&
977 		    secondtp->t_rxtshift == 0) {
978 			return mptcp_return_subflow(second_best);
979 		}
980 
981 		/* Compare RTOs, select second_best if best's rto exceeds rtothresh */
982 		if (besttp->t_rxtcur && secondtp->t_rxtcur &&
983 		    besttp->t_rxtcur >= rto_thresh &&
984 		    secondtp->t_rxtcur < rto_thresh) {
985 			tcpstat.tcps_mp_sel_rto++;
986 			mptcplog((LOG_DEBUG, "%s: best cid %d at rto %d, second cid %d at rto %d\n", __func__,
987 			    best->mpts_connid, besttp->t_rxtcur,
988 			    second_best->mpts_connid, secondtp->t_rxtcur),
989 			    MPTCP_SENDER_DBG, MPTCP_LOGLVL_LOG);
990 
991 			return mptcp_return_subflow(second_best);
992 		}
993 
994 		/*
995 		 * None of the above conditions for sending on the secondary
996 		 * were true. So, let's schedule on the best one, if he still
997 		 * has some space in the congestion-window.
998 		 */
999 		return mptcp_return_subflow(best);
1000 	} else if (mpte->mpte_svctype >= MPTCP_SVCTYPE_AGGREGATE) {
1001 		struct mptsub *tmp;
1002 
1003 		/*
1004 		 * We only care about RTT when aggregating
1005 		 */
1006 		if (besttp->t_srtt > secondtp->t_srtt) {
1007 			tmp = best;
1008 			best = second_best;
1009 			besttp = secondtp;
1010 			bestinp = secondinp;
1011 
1012 			second_best = tmp;
1013 			secondtp = sototcpcb(second_best->mpts_socket);
1014 			secondinp = sotoinpcb(second_best->mpts_socket);
1015 		}
1016 
1017 		/* Is there still space in the congestion window? */
1018 		if (mptcp_subflow_cwnd_space(bestinp->inp_socket) <= 0) {
1019 			return mptcp_return_subflow(second_best);
1020 		}
1021 
1022 		return mptcp_return_subflow(best);
1023 	} else {
1024 		panic("Unknown service-type configured for MPTCP");
1025 	}
1026 
1027 	return NULL;
1028 }
1029 
1030 static const char *
mptcp_event_to_str(uint32_t event)1031 mptcp_event_to_str(uint32_t event)
1032 {
1033 	const char *c = "UNDEFINED";
1034 	switch (event) {
1035 	case MPCE_CLOSE:
1036 		c = "MPCE_CLOSE";
1037 		break;
1038 	case MPCE_RECV_DATA_ACK:
1039 		c = "MPCE_RECV_DATA_ACK";
1040 		break;
1041 	case MPCE_RECV_DATA_FIN:
1042 		c = "MPCE_RECV_DATA_FIN";
1043 		break;
1044 	}
1045 	return c;
1046 }
1047 
1048 static const char *
mptcp_state_to_str(mptcp_state_t state)1049 mptcp_state_to_str(mptcp_state_t state)
1050 {
1051 	const char *c = "UNDEFINED";
1052 	switch (state) {
1053 	case MPTCPS_CLOSED:
1054 		c = "MPTCPS_CLOSED";
1055 		break;
1056 	case MPTCPS_LISTEN:
1057 		c = "MPTCPS_LISTEN";
1058 		break;
1059 	case MPTCPS_ESTABLISHED:
1060 		c = "MPTCPS_ESTABLISHED";
1061 		break;
1062 	case MPTCPS_CLOSE_WAIT:
1063 		c = "MPTCPS_CLOSE_WAIT";
1064 		break;
1065 	case MPTCPS_FIN_WAIT_1:
1066 		c = "MPTCPS_FIN_WAIT_1";
1067 		break;
1068 	case MPTCPS_CLOSING:
1069 		c = "MPTCPS_CLOSING";
1070 		break;
1071 	case MPTCPS_LAST_ACK:
1072 		c = "MPTCPS_LAST_ACK";
1073 		break;
1074 	case MPTCPS_FIN_WAIT_2:
1075 		c = "MPTCPS_FIN_WAIT_2";
1076 		break;
1077 	case MPTCPS_TIME_WAIT:
1078 		c = "MPTCPS_TIME_WAIT";
1079 		break;
1080 	case MPTCPS_TERMINATE:
1081 		c = "MPTCPS_TERMINATE";
1082 		break;
1083 	}
1084 	return c;
1085 }
1086 
1087 void
mptcp_close_fsm(struct mptcb * mp_tp,uint32_t event)1088 mptcp_close_fsm(struct mptcb *mp_tp, uint32_t event)
1089 {
1090 	struct socket *mp_so = mptetoso(mp_tp->mpt_mpte);
1091 
1092 	socket_lock_assert_owned(mp_so);
1093 
1094 	mptcp_state_t old_state = mp_tp->mpt_state;
1095 
1096 	DTRACE_MPTCP2(state__change, struct mptcb *, mp_tp,
1097 	    uint32_t, event);
1098 
1099 	switch (mp_tp->mpt_state) {
1100 	case MPTCPS_CLOSED:
1101 	case MPTCPS_LISTEN:
1102 		mp_tp->mpt_state = MPTCPS_TERMINATE;
1103 		break;
1104 
1105 	case MPTCPS_ESTABLISHED:
1106 		if (event == MPCE_CLOSE) {
1107 			mp_tp->mpt_state = MPTCPS_FIN_WAIT_1;
1108 			mp_tp->mpt_sndmax += 1; /* adjust for Data FIN */
1109 		} else if (event == MPCE_RECV_DATA_FIN) {
1110 			mp_tp->mpt_rcvnxt += 1; /* adj remote data FIN */
1111 			mp_tp->mpt_state = MPTCPS_CLOSE_WAIT;
1112 		}
1113 		break;
1114 
1115 	case MPTCPS_CLOSE_WAIT:
1116 		if (event == MPCE_CLOSE) {
1117 			mp_tp->mpt_state = MPTCPS_LAST_ACK;
1118 			mp_tp->mpt_sndmax += 1; /* adjust for Data FIN */
1119 		}
1120 		break;
1121 
1122 	case MPTCPS_FIN_WAIT_1:
1123 		if (event == MPCE_RECV_DATA_ACK) {
1124 			mp_tp->mpt_state = MPTCPS_FIN_WAIT_2;
1125 		} else if (event == MPCE_RECV_DATA_FIN) {
1126 			mp_tp->mpt_rcvnxt += 1; /* adj remote data FIN */
1127 			mp_tp->mpt_state = MPTCPS_CLOSING;
1128 		}
1129 		break;
1130 
1131 	case MPTCPS_CLOSING:
1132 		if (event == MPCE_RECV_DATA_ACK) {
1133 			mp_tp->mpt_state = MPTCPS_TIME_WAIT;
1134 		}
1135 		break;
1136 
1137 	case MPTCPS_LAST_ACK:
1138 		if (event == MPCE_RECV_DATA_ACK) {
1139 			mptcp_close(mp_tp->mpt_mpte, mp_tp);
1140 		}
1141 		break;
1142 
1143 	case MPTCPS_FIN_WAIT_2:
1144 		if (event == MPCE_RECV_DATA_FIN) {
1145 			mp_tp->mpt_rcvnxt += 1; /* adj remote data FIN */
1146 			mp_tp->mpt_state = MPTCPS_TIME_WAIT;
1147 		}
1148 		break;
1149 
1150 	case MPTCPS_TIME_WAIT:
1151 	case MPTCPS_TERMINATE:
1152 		break;
1153 
1154 	default:
1155 		VERIFY(0);
1156 		/* NOTREACHED */
1157 	}
1158 	DTRACE_MPTCP2(state__change, struct mptcb *, mp_tp,
1159 	    uint32_t, event);
1160 	mptcplog((LOG_INFO, "%s: %s to %s on event %s\n", __func__,
1161 	    mptcp_state_to_str(old_state),
1162 	    mptcp_state_to_str(mp_tp->mpt_state),
1163 	    mptcp_event_to_str(event)),
1164 	    MPTCP_STATE_DBG, MPTCP_LOGLVL_LOG);
1165 }
1166 
1167 /* If you change this function, match up mptcp_update_rcv_state_f */
1168 void
mptcp_update_dss_rcv_state(struct mptcp_dsn_opt * dss_info,struct tcpcb * tp,uint16_t csum)1169 mptcp_update_dss_rcv_state(struct mptcp_dsn_opt *dss_info, struct tcpcb *tp,
1170     uint16_t csum)
1171 {
1172 	struct mptcb *mp_tp = tptomptp(tp);
1173 	u_int64_t full_dsn = 0;
1174 
1175 	NTOHL(dss_info->mdss_dsn);
1176 	NTOHL(dss_info->mdss_subflow_seqn);
1177 	NTOHS(dss_info->mdss_data_len);
1178 
1179 	/* XXX for autosndbuf grow sb here */
1180 	MPTCP_EXTEND_DSN(mp_tp->mpt_rcvnxt, dss_info->mdss_dsn, full_dsn);
1181 	mptcp_update_rcv_state_meat(mp_tp, tp,
1182 	    full_dsn, dss_info->mdss_subflow_seqn, dss_info->mdss_data_len,
1183 	    csum);
1184 }
1185 
1186 void
mptcp_update_rcv_state_meat(struct mptcb * mp_tp,struct tcpcb * tp,u_int64_t full_dsn,u_int32_t seqn,u_int16_t mdss_data_len,uint16_t csum)1187 mptcp_update_rcv_state_meat(struct mptcb *mp_tp, struct tcpcb *tp,
1188     u_int64_t full_dsn, u_int32_t seqn, u_int16_t mdss_data_len,
1189     uint16_t csum)
1190 {
1191 	if (mdss_data_len == 0) {
1192 		os_log_error(mptcp_log_handle, "%s - %lx: Infinite Mapping.\n",
1193 		    __func__, (unsigned long)VM_KERNEL_ADDRPERM(mp_tp->mpt_mpte));
1194 
1195 		if ((mp_tp->mpt_flags & MPTCPF_CHECKSUM) && (csum != 0)) {
1196 			os_log_error(mptcp_log_handle, "%s - %lx: Bad checksum %x \n",
1197 			    __func__, (unsigned long)VM_KERNEL_ADDRPERM(mp_tp->mpt_mpte), csum);
1198 		}
1199 		mptcp_notify_mpfail(tp->t_inpcb->inp_socket);
1200 		return;
1201 	}
1202 
1203 	mptcp_notify_mpready(tp->t_inpcb->inp_socket);
1204 
1205 	tp->t_rcv_map.mpt_dsn = full_dsn;
1206 	tp->t_rcv_map.mpt_sseq = seqn;
1207 	tp->t_rcv_map.mpt_len = mdss_data_len;
1208 	tp->t_rcv_map.mpt_csum = csum;
1209 	tp->t_mpflags |= TMPF_EMBED_DSN;
1210 }
1211 
1212 
1213 static int
mptcp_validate_dss_map(struct socket * so,struct tcpcb * tp,struct mbuf * m,int hdrlen)1214 mptcp_validate_dss_map(struct socket *so, struct tcpcb *tp, struct mbuf *m,
1215     int hdrlen)
1216 {
1217 	u_int32_t datalen;
1218 
1219 	if (!(m->m_pkthdr.pkt_flags & PKTF_MPTCP)) {
1220 		return 0;
1221 	}
1222 
1223 	datalen = m->m_pkthdr.mp_rlen;
1224 
1225 	/* unacceptable DSS option, fallback to TCP */
1226 	if (m->m_pkthdr.len > ((int) datalen + hdrlen)) {
1227 		os_log_error(mptcp_log_handle, "%s - %lx: mbuf len %d, MPTCP expected %d",
1228 		    __func__, (unsigned long)VM_KERNEL_ADDRPERM(tptomptp(tp)->mpt_mpte), m->m_pkthdr.len, datalen);
1229 	} else {
1230 		return 0;
1231 	}
1232 	tp->t_mpflags |= TMPF_SND_MPFAIL;
1233 	mptcp_notify_mpfail(so);
1234 	m_freem(m);
1235 	return -1;
1236 }
1237 
1238 int
mptcp_input_preproc(struct tcpcb * tp,struct mbuf * m,struct tcphdr * th,int drop_hdrlen)1239 mptcp_input_preproc(struct tcpcb *tp, struct mbuf *m, struct tcphdr *th,
1240     int drop_hdrlen)
1241 {
1242 	mptcp_insert_rmap(tp, m, th);
1243 	if (mptcp_validate_dss_map(tp->t_inpcb->inp_socket, tp, m,
1244 	    drop_hdrlen) != 0) {
1245 		return -1;
1246 	}
1247 	return 0;
1248 }
1249 
1250 static uint16_t
mptcp_input_csum(struct tcpcb * tp,struct mbuf * m,uint64_t dsn,uint32_t sseq,uint16_t dlen,uint16_t csum,int dfin)1251 mptcp_input_csum(struct tcpcb *tp, struct mbuf *m, uint64_t dsn, uint32_t sseq,
1252     uint16_t dlen, uint16_t csum, int dfin)
1253 {
1254 	struct mptcb *mp_tp = tptomptp(tp);
1255 	int real_len = dlen - dfin;
1256 	uint32_t sum = 0;
1257 
1258 	VERIFY(real_len >= 0);
1259 
1260 	if (mp_tp == NULL) {
1261 		return 0;
1262 	}
1263 
1264 	if (!(mp_tp->mpt_flags & MPTCPF_CHECKSUM)) {
1265 		return 0;
1266 	}
1267 
1268 	if (tp->t_mpflags & TMPF_TCP_FALLBACK) {
1269 		return 0;
1270 	}
1271 
1272 	/*
1273 	 * The remote side may send a packet with fewer bytes than the
1274 	 * claimed DSS checksum length.
1275 	 */
1276 	if ((int)m_length2(m, NULL) < real_len) {
1277 		return 0xffff;
1278 	}
1279 
1280 	if (real_len != 0) {
1281 		sum = m_sum16(m, 0, real_len);
1282 	}
1283 
1284 	sum += in_pseudo64(htonll(dsn), htonl(sseq), htons(dlen) + csum);
1285 	ADDCARRY(sum);
1286 
1287 	DTRACE_MPTCP3(checksum__result, struct tcpcb *, tp, struct mbuf *, m,
1288 	    uint32_t, sum);
1289 
1290 	return ~sum & 0xffff;
1291 }
1292 
1293 /*
1294  * MPTCP Checksum support
1295  * The checksum is calculated whenever the MPTCP DSS option is included
1296  * in the TCP packet. The checksum includes the sum of the MPTCP psuedo
1297  * header and the actual data indicated by the length specified in the
1298  * DSS option.
1299  */
1300 
1301 int
mptcp_validate_csum(struct tcpcb * tp,struct mbuf * m,uint64_t dsn,uint32_t sseq,uint16_t dlen,uint16_t csum,int dfin)1302 mptcp_validate_csum(struct tcpcb *tp, struct mbuf *m, uint64_t dsn,
1303     uint32_t sseq, uint16_t dlen, uint16_t csum, int dfin)
1304 {
1305 	uint16_t mptcp_csum;
1306 
1307 	mptcp_csum = mptcp_input_csum(tp, m, dsn, sseq, dlen, csum, dfin);
1308 	if (mptcp_csum) {
1309 		tp->t_mpflags |= TMPF_SND_MPFAIL;
1310 		mptcp_notify_mpfail(tp->t_inpcb->inp_socket);
1311 		m_freem(m);
1312 		tcpstat.tcps_mp_badcsum++;
1313 		return -1;
1314 	}
1315 	return 0;
1316 }
1317 
1318 uint16_t
mptcp_output_csum(struct mbuf * m,uint64_t dss_val,uint32_t sseq,uint16_t dlen)1319 mptcp_output_csum(struct mbuf *m, uint64_t dss_val, uint32_t sseq, uint16_t dlen)
1320 {
1321 	uint32_t sum = 0;
1322 
1323 	if (dlen) {
1324 		sum = m_sum16(m, 0, dlen);
1325 	}
1326 
1327 	dss_val = mptcp_hton64(dss_val);
1328 	sseq = htonl(sseq);
1329 	dlen = htons(dlen);
1330 	sum += in_pseudo64(dss_val, sseq, dlen);
1331 
1332 	ADDCARRY(sum);
1333 	sum = ~sum & 0xffff;
1334 	DTRACE_MPTCP2(checksum__result, struct mbuf *, m, uint32_t, sum);
1335 	mptcplog((LOG_DEBUG, "%s: sum = %x \n", __func__, sum),
1336 	    MPTCP_SENDER_DBG, MPTCP_LOGLVL_VERBOSE);
1337 
1338 	return (uint16_t)sum;
1339 }
1340 
1341 /*
1342  * When WiFi signal starts fading, there's more loss and RTT spikes.
1343  * Check if there has been a large spike by comparing against
1344  * a tolerable RTT spike threshold.
1345  */
1346 boolean_t
mptcp_no_rto_spike(struct socket * so)1347 mptcp_no_rto_spike(struct socket *so)
1348 {
1349 	struct tcpcb *tp = intotcpcb(sotoinpcb(so));
1350 	int32_t spike = 0;
1351 
1352 	if (tp->t_rxtcur > mptcp_rtothresh) {
1353 		spike = tp->t_rxtcur - mptcp_rtothresh;
1354 
1355 		mptcplog((LOG_DEBUG, "%s: spike = %d rto = %d best = %d cur = %d\n",
1356 		    __func__, spike,
1357 		    tp->t_rxtcur, tp->t_rttbest >> TCP_RTT_SHIFT,
1358 		    tp->t_rttcur),
1359 		    (MPTCP_SOCKET_DBG | MPTCP_SENDER_DBG), MPTCP_LOGLVL_LOG);
1360 	}
1361 
1362 	if (spike > 0) {
1363 		return FALSE;
1364 	} else {
1365 		return TRUE;
1366 	}
1367 }
1368 
1369 void
mptcp_handle_deferred_upcalls(struct mppcb * mpp,uint32_t flag)1370 mptcp_handle_deferred_upcalls(struct mppcb *mpp, uint32_t flag)
1371 {
1372 	VERIFY(mpp->mpp_flags & flag);
1373 	mpp->mpp_flags &= ~flag;
1374 
1375 	if (mptcp_should_defer_upcall(mpp)) {
1376 		return;
1377 	}
1378 
1379 	if (mpp->mpp_flags & MPP_SHOULD_WORKLOOP) {
1380 		mpp->mpp_flags &= ~MPP_SHOULD_WORKLOOP;
1381 
1382 		mptcp_subflow_workloop(mpp->mpp_pcbe);
1383 	}
1384 
1385 	if (mpp->mpp_flags & MPP_SHOULD_RWAKEUP) {
1386 		mpp->mpp_flags &= ~MPP_SHOULD_RWAKEUP;
1387 
1388 		sorwakeup(mpp->mpp_socket);
1389 	}
1390 
1391 	if (mpp->mpp_flags & MPP_SHOULD_WWAKEUP) {
1392 		mpp->mpp_flags &= ~MPP_SHOULD_WWAKEUP;
1393 
1394 		sowwakeup(mpp->mpp_socket);
1395 	}
1396 }
1397 
1398 static void
mptcp_reset_itfinfo(struct mpt_itf_info * info)1399 mptcp_reset_itfinfo(struct mpt_itf_info *info)
1400 {
1401 	memset(info, 0, sizeof(*info));
1402 }
1403 
1404 void
mptcp_session_necp_cb(void * handle,int action,uint32_t interface_index,uint32_t necp_flags,__unused bool * viable)1405 mptcp_session_necp_cb(void *handle, int action, uint32_t interface_index,
1406     uint32_t necp_flags, __unused bool *viable)
1407 {
1408 	boolean_t has_v4 = !!(necp_flags & NECP_CLIENT_RESULT_FLAG_HAS_IPV4);
1409 	boolean_t has_v6 = !!(necp_flags & NECP_CLIENT_RESULT_FLAG_HAS_IPV6);
1410 	boolean_t has_nat64 = !!(necp_flags & NECP_CLIENT_RESULT_FLAG_HAS_NAT64);
1411 	boolean_t low_power = !!(necp_flags & NECP_CLIENT_RESULT_FLAG_INTERFACE_LOW_POWER);
1412 	struct mppcb *mp = (struct mppcb *)handle;
1413 	struct mptses *mpte = mptompte(mp);
1414 	struct socket *mp_so;
1415 	struct mptcb *mp_tp;
1416 	uint32_t i, ifindex;
1417 	struct ifnet *ifp;
1418 	int locked = 0;
1419 
1420 	ifindex = interface_index;
1421 	VERIFY(ifindex != IFSCOPE_NONE);
1422 
1423 	/* About to be garbage-collected (see note about MPTCP/NECP interactions) */
1424 	if (mp->mpp_socket->so_usecount == 0) {
1425 		return;
1426 	}
1427 
1428 	mp_so = mptetoso(mpte);
1429 
1430 	if (action != NECP_CLIENT_CBACTION_INITIAL) {
1431 		socket_lock(mp_so, 1);
1432 		locked = 1;
1433 
1434 		/* Check again, because it might have changed while waiting */
1435 		if (mp->mpp_socket->so_usecount == 0) {
1436 			goto out;
1437 		}
1438 	}
1439 
1440 	socket_lock_assert_owned(mp_so);
1441 
1442 	mp_tp = mpte->mpte_mptcb;
1443 
1444 	ifnet_head_lock_shared();
1445 	ifp = ifindex2ifnet[ifindex];
1446 	ifnet_head_done();
1447 
1448 	os_log(mptcp_log_handle, "%s - %lx: action: %u ifindex %u delegated to %u usecount %u mpt_flags %#x state %u v4 %u v6 %u nat64 %u power %u\n",
1449 	    __func__, (unsigned long)VM_KERNEL_ADDRPERM(mpte), action, ifindex,
1450 	    ifp && ifp->if_delegated.ifp ? ifp->if_delegated.ifp->if_index : IFSCOPE_NONE,
1451 	    mp->mpp_socket->so_usecount, mp_tp->mpt_flags, mp_tp->mpt_state,
1452 	    has_v4, has_v6, has_nat64, low_power);
1453 
1454 	/* No need on fallen back sockets */
1455 	if (mp_tp->mpt_flags & MPTCPF_FALLBACK_TO_TCP) {
1456 		goto out;
1457 	}
1458 
1459 	/*
1460 	 * When the interface goes in low-power mode we don't want to establish
1461 	 * new subflows on it. Thus, mark it internally as non-viable.
1462 	 */
1463 	if (low_power) {
1464 		action = NECP_CLIENT_CBACTION_NONVIABLE;
1465 	}
1466 
1467 	if (action == NECP_CLIENT_CBACTION_NONVIABLE) {
1468 		for (i = 0; i < mpte->mpte_itfinfo_size; i++) {
1469 			if (mpte->mpte_itfinfo[i].ifindex == IFSCOPE_NONE) {
1470 				continue;
1471 			}
1472 
1473 			if (mpte->mpte_itfinfo[i].ifindex == ifindex) {
1474 				mptcp_reset_itfinfo(&mpte->mpte_itfinfo[i]);
1475 			}
1476 		}
1477 
1478 		mptcp_sched_create_subflows(mpte);
1479 	} else if (action == NECP_CLIENT_CBACTION_VIABLE ||
1480 	    action == NECP_CLIENT_CBACTION_INITIAL) {
1481 		int found_slot = 0, slot_index = -1;
1482 		struct sockaddr *dst;
1483 
1484 		if (ifp == NULL) {
1485 			goto out;
1486 		}
1487 
1488 		if (IFNET_IS_COMPANION_LINK(ifp)) {
1489 			goto out;
1490 		}
1491 
1492 		if (IFNET_IS_EXPENSIVE(ifp) &&
1493 		    (mp_so->so_restrictions & SO_RESTRICT_DENY_EXPENSIVE)) {
1494 			goto out;
1495 		}
1496 
1497 		if (IFNET_IS_CONSTRAINED(ifp) &&
1498 		    (mp_so->so_restrictions & SO_RESTRICT_DENY_CONSTRAINED)) {
1499 			goto out;
1500 		}
1501 
1502 		if (IFNET_IS_CELLULAR(ifp) &&
1503 		    (mp_so->so_restrictions & SO_RESTRICT_DENY_CELLULAR)) {
1504 			goto out;
1505 		}
1506 
1507 		if (IS_INTF_CLAT46(ifp)) {
1508 			has_v4 = FALSE;
1509 		}
1510 
1511 		/* Look for the slot on where to store/update the interface-info. */
1512 		for (i = 0; i < mpte->mpte_itfinfo_size; i++) {
1513 			/* Found a potential empty slot where we can put it */
1514 			if (mpte->mpte_itfinfo[i].ifindex == 0) {
1515 				found_slot = 1;
1516 				slot_index = i;
1517 			}
1518 
1519 			/*
1520 			 * The interface is already in our array. Check if we
1521 			 * need to update it.
1522 			 */
1523 			if (mpte->mpte_itfinfo[i].ifindex == ifindex &&
1524 			    (mpte->mpte_itfinfo[i].has_v4_conn != has_v4 ||
1525 			    mpte->mpte_itfinfo[i].has_v6_conn != has_v6 ||
1526 			    mpte->mpte_itfinfo[i].has_nat64_conn != has_nat64)) {
1527 				found_slot = 1;
1528 				slot_index = i;
1529 				break;
1530 			}
1531 
1532 			if (mpte->mpte_itfinfo[i].ifindex == ifindex) {
1533 				/*
1534 				 * Ok, it's already there and we don't need
1535 				 * to update it
1536 				 */
1537 				goto out;
1538 			}
1539 		}
1540 
1541 		dst = mptcp_get_session_dst(mpte, has_v6, has_v4);
1542 		if (dst && dst->sa_family == AF_INET &&
1543 		    has_v6 && !has_nat64 && !has_v4) {
1544 			if (found_slot) {
1545 				mpte->mpte_itfinfo[slot_index].ifindex = ifindex;
1546 				mpte->mpte_itfinfo[slot_index].has_v4_conn = has_v4;
1547 				mpte->mpte_itfinfo[slot_index].has_v6_conn = has_v6;
1548 				mpte->mpte_itfinfo[slot_index].has_nat64_conn = has_nat64;
1549 			}
1550 			goto out;
1551 		}
1552 
1553 		if (found_slot == 0) {
1554 			int new_size = mpte->mpte_itfinfo_size * 2;
1555 			struct mpt_itf_info *info = kalloc_data(sizeof(*info) * new_size, Z_ZERO);
1556 
1557 			if (info == NULL) {
1558 				os_log_error(mptcp_log_handle, "%s - %lx: malloc failed for %u\n",
1559 				    __func__, (unsigned long)VM_KERNEL_ADDRPERM(mpte), new_size);
1560 				goto out;
1561 			}
1562 
1563 			memcpy(info, mpte->mpte_itfinfo, mpte->mpte_itfinfo_size * sizeof(*info));
1564 
1565 			if (mpte->mpte_itfinfo_size > MPTE_ITFINFO_SIZE) {
1566 				kfree_data(mpte->mpte_itfinfo,
1567 				    sizeof(*info) * mpte->mpte_itfinfo_size);
1568 			}
1569 
1570 			/* We allocated a new one, thus the first must be empty */
1571 			slot_index = mpte->mpte_itfinfo_size;
1572 
1573 			mpte->mpte_itfinfo = info;
1574 			mpte->mpte_itfinfo_size = new_size;
1575 		}
1576 
1577 		VERIFY(slot_index >= 0 && slot_index < (int)mpte->mpte_itfinfo_size);
1578 		mpte->mpte_itfinfo[slot_index].ifindex = ifindex;
1579 		mpte->mpte_itfinfo[slot_index].has_v4_conn = has_v4;
1580 		mpte->mpte_itfinfo[slot_index].has_v6_conn = has_v6;
1581 		mpte->mpte_itfinfo[slot_index].has_nat64_conn = has_nat64;
1582 
1583 		mptcp_sched_create_subflows(mpte);
1584 	}
1585 
1586 out:
1587 	if (locked) {
1588 		socket_unlock(mp_so, 1);
1589 	}
1590 }
1591 
1592 void
mptcp_set_restrictions(struct socket * mp_so)1593 mptcp_set_restrictions(struct socket *mp_so)
1594 {
1595 	struct mptses *mpte = mpsotompte(mp_so);
1596 	uint32_t i;
1597 
1598 	socket_lock_assert_owned(mp_so);
1599 
1600 	ifnet_head_lock_shared();
1601 
1602 	for (i = 0; i < mpte->mpte_itfinfo_size; i++) {
1603 		struct mpt_itf_info *info = &mpte->mpte_itfinfo[i];
1604 		uint32_t ifindex = info->ifindex;
1605 		struct ifnet *ifp;
1606 
1607 		if (ifindex == IFSCOPE_NONE) {
1608 			continue;
1609 		}
1610 
1611 		ifp = ifindex2ifnet[ifindex];
1612 		if (ifp == NULL) {
1613 			continue;
1614 		}
1615 
1616 		if (IFNET_IS_EXPENSIVE(ifp) &&
1617 		    (mp_so->so_restrictions & SO_RESTRICT_DENY_EXPENSIVE)) {
1618 			info->ifindex = IFSCOPE_NONE;
1619 		}
1620 
1621 		if (IFNET_IS_CONSTRAINED(ifp) &&
1622 		    (mp_so->so_restrictions & SO_RESTRICT_DENY_CONSTRAINED)) {
1623 			info->ifindex = IFSCOPE_NONE;
1624 		}
1625 
1626 		if (IFNET_IS_CELLULAR(ifp) &&
1627 		    (mp_so->so_restrictions & SO_RESTRICT_DENY_CELLULAR)) {
1628 			info->ifindex = IFSCOPE_NONE;
1629 		}
1630 	}
1631 
1632 	ifnet_head_done();
1633 }
1634