xref: /xnu-10002.61.3/bsd/netinet/mptcp.c (revision 0f4c859e951fba394238ab619495c4e1d54d0f34)
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_1;
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 int mptcp_reass_total_qlen = 0;
202 SYSCTL_INT(_net_inet_mptcp, OID_AUTO, reass_qlen,
203     CTLFLAG_RD | CTLFLAG_LOCKED, &mptcp_reass_total_qlen, 0,
204     "Total number of MPTCP segments in reassembly queues");
205 
206 static int
mptcp_reass_present(struct socket * mp_so)207 mptcp_reass_present(struct socket *mp_so)
208 {
209 	struct mptses *mpte = mpsotompte(mp_so);
210 	struct mptcb *mp_tp = mpte->mpte_mptcb;
211 	struct tseg_qent *q;
212 	int dowakeup = 0;
213 	int flags = 0;
214 	int count = 0;
215 
216 	/*
217 	 * Present data to user, advancing rcv_nxt through
218 	 * completed sequence space.
219 	 */
220 	if (mp_tp->mpt_state < MPTCPS_ESTABLISHED) {
221 		return flags;
222 	}
223 	q = LIST_FIRST(&mp_tp->mpt_segq);
224 	if (!q || q->tqe_m->m_pkthdr.mp_dsn != mp_tp->mpt_rcvnxt) {
225 		return flags;
226 	}
227 
228 	/*
229 	 * If there is already another thread doing reassembly for this
230 	 * connection, it is better to let it finish the job --
231 	 * (radar 16316196)
232 	 */
233 	if (mp_tp->mpt_flags & MPTCPF_REASS_INPROG) {
234 		return flags;
235 	}
236 
237 	mp_tp->mpt_flags |= MPTCPF_REASS_INPROG;
238 
239 	do {
240 		mp_tp->mpt_rcvnxt += q->tqe_len;
241 		LIST_REMOVE(q, tqe_q);
242 		if (mp_so->so_state & SS_CANTRCVMORE) {
243 			m_freem(q->tqe_m);
244 		} else {
245 			flags = !!(q->tqe_m->m_pkthdr.pkt_flags & PKTF_MPTCP_DFIN);
246 			if (sbappendstream_rcvdemux(mp_so, q->tqe_m)) {
247 				dowakeup = 1;
248 			}
249 		}
250 		zfree(tcp_reass_zone, q);
251 		mp_tp->mpt_reassqlen--;
252 		count++;
253 		q = LIST_FIRST(&mp_tp->mpt_segq);
254 	} while (q && q->tqe_m->m_pkthdr.mp_dsn == mp_tp->mpt_rcvnxt);
255 	mp_tp->mpt_flags &= ~MPTCPF_REASS_INPROG;
256 
257 	if (count > 0) {
258 		OSAddAtomic(-count, &mptcp_reass_total_qlen);
259 	}
260 	if (dowakeup) {
261 		sorwakeup(mp_so); /* done with socket lock held */
262 	}
263 	return flags;
264 }
265 
266 static int
mptcp_reass(struct socket * mp_so,struct pkthdr * phdr,int * tlenp,struct mbuf * m)267 mptcp_reass(struct socket *mp_so, struct pkthdr *phdr, int *tlenp, struct mbuf *m)
268 {
269 	struct mptcb *mp_tp = mpsotomppcb(mp_so)->mpp_pcbe->mpte_mptcb;
270 	u_int64_t mb_dsn = phdr->mp_dsn;
271 	struct tseg_qent *q;
272 	struct tseg_qent *p = NULL;
273 	struct tseg_qent *nq;
274 	struct tseg_qent *te = NULL;
275 	uint32_t qlimit;
276 
277 	/*
278 	 * Limit the number of segments in the reassembly queue to prevent
279 	 * holding on to too many segments (and thus running out of mbufs).
280 	 * Make sure to let the missing segment through which caused this
281 	 * queue.  Always keep one global queue entry spare to be able to
282 	 * process the missing segment.
283 	 */
284 	qlimit = MIN(MAX(100, mp_so->so_rcv.sb_hiwat >> 10),
285 	    (tcp_autorcvbuf_max >> 10));
286 	if (mb_dsn != mp_tp->mpt_rcvnxt &&
287 	    (mp_tp->mpt_reassqlen + 1) >= qlimit) {
288 		tcpstat.tcps_mptcp_rcvmemdrop++;
289 		m_freem(m);
290 		*tlenp = 0;
291 		return 0;
292 	}
293 
294 	/* Allocate a new queue entry. If we can't, just drop the pkt. XXX */
295 	te = zalloc_flags(tcp_reass_zone, Z_WAITOK | Z_NOFAIL);
296 
297 	mp_tp->mpt_reassqlen++;
298 	OSIncrementAtomic(&mptcp_reass_total_qlen);
299 
300 	/*
301 	 * Find a segment which begins after this one does.
302 	 */
303 	LIST_FOREACH(q, &mp_tp->mpt_segq, tqe_q) {
304 		if (MPTCP_SEQ_GT(q->tqe_m->m_pkthdr.mp_dsn, mb_dsn)) {
305 			break;
306 		}
307 		p = q;
308 	}
309 
310 	/*
311 	 * If there is a preceding segment, it may provide some of
312 	 * our data already.  If so, drop the data from the incoming
313 	 * segment.  If it provides all of our data, drop us.
314 	 */
315 	if (p != NULL) {
316 		int64_t i;
317 		/* conversion to int (in i) handles seq wraparound */
318 		i = p->tqe_m->m_pkthdr.mp_dsn + p->tqe_len - mb_dsn;
319 		if (i > 0) {
320 			if (i >= *tlenp) {
321 				tcpstat.tcps_mptcp_rcvduppack++;
322 				m_freem(m);
323 				zfree(tcp_reass_zone, te);
324 				te = NULL;
325 				mp_tp->mpt_reassqlen--;
326 				OSDecrementAtomic(&mptcp_reass_total_qlen);
327 				/*
328 				 * Try to present any queued data
329 				 * at the left window edge to the user.
330 				 * This is needed after the 3-WHS
331 				 * completes.
332 				 */
333 				goto out;
334 			}
335 			VERIFY(i <= INT_MAX);
336 			m_adj(m, (int)i);
337 			*tlenp -= i;
338 			phdr->mp_dsn += i;
339 		}
340 	}
341 
342 	tcpstat.tcps_mp_oodata++;
343 
344 	/*
345 	 * While we overlap succeeding segments trim them or,
346 	 * if they are completely covered, dequeue them.
347 	 */
348 	while (q) {
349 		int64_t i = (mb_dsn + *tlenp) - q->tqe_m->m_pkthdr.mp_dsn;
350 		if (i <= 0) {
351 			break;
352 		}
353 
354 		if (i < q->tqe_len) {
355 			q->tqe_m->m_pkthdr.mp_dsn += i;
356 			q->tqe_len -= i;
357 
358 			VERIFY(i <= INT_MAX);
359 			m_adj(q->tqe_m, (int)i);
360 			break;
361 		}
362 
363 		nq = LIST_NEXT(q, tqe_q);
364 		LIST_REMOVE(q, tqe_q);
365 		m_freem(q->tqe_m);
366 		zfree(tcp_reass_zone, q);
367 		mp_tp->mpt_reassqlen--;
368 		OSDecrementAtomic(&mptcp_reass_total_qlen);
369 		q = nq;
370 	}
371 
372 	/* Insert the new segment queue entry into place. */
373 	te->tqe_m = m;
374 	te->tqe_th = NULL;
375 	te->tqe_len = *tlenp;
376 
377 	if (p == NULL) {
378 		LIST_INSERT_HEAD(&mp_tp->mpt_segq, te, tqe_q);
379 	} else {
380 		LIST_INSERT_AFTER(p, te, tqe_q);
381 	}
382 
383 out:
384 	return mptcp_reass_present(mp_so);
385 }
386 
387 /*
388  * MPTCP input, called when data has been read from a subflow socket.
389  */
390 void
mptcp_input(struct mptses * mpte,struct mbuf * m)391 mptcp_input(struct mptses *mpte, struct mbuf *m)
392 {
393 	struct socket *mp_so;
394 	struct mptcb *mp_tp = NULL;
395 	int count = 0, wakeup = 0;
396 	struct mbuf *save = NULL, *prev = NULL;
397 	struct mbuf *freelist = NULL, *tail = NULL;
398 
399 	if (__improbable((m->m_flags & M_PKTHDR) == 0)) {
400 		panic("mbuf invalid: %p", m);
401 	}
402 
403 	mp_so = mptetoso(mpte);
404 	mp_tp = mpte->mpte_mptcb;
405 
406 	socket_lock_assert_owned(mp_so);
407 
408 	DTRACE_MPTCP(input);
409 
410 	mp_tp->mpt_rcvwnd = mptcp_sbspace(mp_tp);
411 
412 	/*
413 	 * Each mbuf contains MPTCP Data Sequence Map
414 	 * Process the data for reassembly, delivery to MPTCP socket
415 	 * client, etc.
416 	 *
417 	 */
418 	count = mp_so->so_rcv.sb_cc;
419 
420 	/*
421 	 * In the degraded fallback case, data is accepted without DSS map
422 	 */
423 	if (mp_tp->mpt_flags & MPTCPF_FALLBACK_TO_TCP) {
424 		struct mbuf *iter;
425 		int mb_dfin;
426 fallback:
427 		mb_dfin = 0;
428 		mptcp_sbrcv_grow(mp_tp);
429 
430 		iter = m;
431 		while (iter) {
432 			if ((iter->m_flags & M_PKTHDR) &&
433 			    (iter->m_pkthdr.pkt_flags & PKTF_MPTCP_DFIN)) {
434 				mb_dfin = 1;
435 			}
436 
437 			if ((iter->m_flags & M_PKTHDR) && m_pktlen(iter) == 0) {
438 				/* Don't add zero-length packets, so jump it! */
439 				if (prev == NULL) {
440 					m = iter->m_next;
441 					m_free(iter);
442 					iter = m;
443 				} else {
444 					prev->m_next = iter->m_next;
445 					m_free(iter);
446 					iter = prev->m_next;
447 				}
448 
449 				/* It was a zero-length packet so next one must be a pkthdr */
450 				VERIFY(iter == NULL || iter->m_flags & M_PKTHDR);
451 			} else {
452 				prev = iter;
453 				iter = iter->m_next;
454 			}
455 		}
456 
457 		/*
458 		 * assume degraded flow as this may be the first packet
459 		 * without DSS, and the subflow state is not updated yet.
460 		 */
461 		if (sbappendstream_rcvdemux(mp_so, m)) {
462 			sorwakeup(mp_so);
463 		}
464 
465 		DTRACE_MPTCP5(receive__degraded, struct mbuf *, m,
466 		    struct socket *, mp_so,
467 		    struct sockbuf *, &mp_so->so_rcv,
468 		    struct sockbuf *, &mp_so->so_snd,
469 		    struct mptses *, mpte);
470 		count = mp_so->so_rcv.sb_cc - count;
471 
472 		mp_tp->mpt_rcvnxt += count;
473 
474 		if (mb_dfin) {
475 			mptcp_close_fsm(mp_tp, MPCE_RECV_DATA_FIN);
476 			socantrcvmore(mp_so);
477 		}
478 		return;
479 	}
480 
481 	do {
482 		u_int64_t mb_dsn;
483 		int32_t mb_datalen;
484 		int64_t todrop;
485 		int mb_dfin = 0;
486 
487 		VERIFY(m->m_flags & M_PKTHDR);
488 
489 		/* If fallback occurs, mbufs will not have PKTF_MPTCP set */
490 		if (!(m->m_pkthdr.pkt_flags & PKTF_MPTCP)) {
491 			goto fallback;
492 		}
493 
494 		save = m->m_next;
495 		/*
496 		 * A single TCP packet formed of multiple mbufs
497 		 * holds DSS mapping in the first mbuf of the chain.
498 		 * Other mbufs in the chain may have M_PKTHDR set
499 		 * even though they belong to the same TCP packet
500 		 * and therefore use the DSS mapping stored in the
501 		 * first mbuf of the mbuf chain. mptcp_input() can
502 		 * get an mbuf chain with multiple TCP packets.
503 		 */
504 		while (save && (!(save->m_flags & M_PKTHDR) ||
505 		    !(save->m_pkthdr.pkt_flags & PKTF_MPTCP))) {
506 			prev = save;
507 			save = save->m_next;
508 		}
509 		if (prev) {
510 			prev->m_next = NULL;
511 		} else {
512 			m->m_next = NULL;
513 		}
514 
515 		mb_dsn = m->m_pkthdr.mp_dsn;
516 		mb_datalen = m->m_pkthdr.mp_rlen;
517 
518 		todrop = (mb_dsn + mb_datalen) - (mp_tp->mpt_rcvnxt + mp_tp->mpt_rcvwnd);
519 		if (todrop > 0) {
520 			tcpstat.tcps_mptcp_rcvpackafterwin++;
521 
522 			os_log_info(mptcp_log_handle, "%s - %lx: dropping dsn %u dlen %u rcvnxt %u rcvwnd %u todrop %lld\n",
523 			    __func__, (unsigned long)VM_KERNEL_ADDRPERM(mpte),
524 			    (uint32_t)mb_dsn, mb_datalen, (uint32_t)mp_tp->mpt_rcvnxt,
525 			    mp_tp->mpt_rcvwnd, todrop);
526 
527 			if (todrop >= mb_datalen) {
528 				if (freelist == NULL) {
529 					freelist = m;
530 				} else {
531 					tail->m_next = m;
532 				}
533 
534 				if (prev != NULL) {
535 					tail = prev;
536 				} else {
537 					tail = m;
538 				}
539 
540 				m = save;
541 				prev = save = NULL;
542 				continue;
543 			} else {
544 				VERIFY(todrop <= INT_MAX);
545 				m_adj(m, (int)-todrop);
546 				mb_datalen -= todrop;
547 				m->m_pkthdr.mp_rlen -= todrop;
548 			}
549 
550 			/*
551 			 * We drop from the right edge of the mbuf, thus the
552 			 * DATA_FIN is dropped as well
553 			 */
554 			m->m_pkthdr.pkt_flags &= ~PKTF_MPTCP_DFIN;
555 		}
556 
557 		if (MPTCP_SEQ_LT(mb_dsn, mp_tp->mpt_rcvnxt)) {
558 			if (MPTCP_SEQ_LEQ((mb_dsn + mb_datalen),
559 			    mp_tp->mpt_rcvnxt)) {
560 				if (freelist == NULL) {
561 					freelist = m;
562 				} else {
563 					tail->m_next = m;
564 				}
565 
566 				if (prev != NULL) {
567 					tail = prev;
568 				} else {
569 					tail = m;
570 				}
571 
572 				m = save;
573 				prev = save = NULL;
574 				continue;
575 			} else {
576 				VERIFY((mp_tp->mpt_rcvnxt - mb_dsn) <= INT_MAX);
577 				m_adj(m, (int)(mp_tp->mpt_rcvnxt - mb_dsn));
578 				mb_datalen -= (mp_tp->mpt_rcvnxt - mb_dsn);
579 				mb_dsn = mp_tp->mpt_rcvnxt;
580 				VERIFY(mb_datalen >= 0 && mb_datalen <= USHRT_MAX);
581 				m->m_pkthdr.mp_rlen = (uint16_t)mb_datalen;
582 				m->m_pkthdr.mp_dsn = mb_dsn;
583 			}
584 		}
585 
586 		if (MPTCP_SEQ_GT(mb_dsn, mp_tp->mpt_rcvnxt) ||
587 		    !LIST_EMPTY(&mp_tp->mpt_segq)) {
588 			mb_dfin = mptcp_reass(mp_so, &m->m_pkthdr, &mb_datalen, m);
589 
590 			goto next;
591 		}
592 		mb_dfin = !!(m->m_pkthdr.pkt_flags & PKTF_MPTCP_DFIN);
593 
594 		mptcp_sbrcv_grow(mp_tp);
595 
596 		if (sbappendstream_rcvdemux(mp_so, m)) {
597 			wakeup = 1;
598 		}
599 
600 		DTRACE_MPTCP6(receive, struct mbuf *, m, struct socket *, mp_so,
601 		    struct sockbuf *, &mp_so->so_rcv,
602 		    struct sockbuf *, &mp_so->so_snd,
603 		    struct mptses *, mpte,
604 		    struct mptcb *, mp_tp);
605 		count = mp_so->so_rcv.sb_cc - count;
606 		tcpstat.tcps_mp_rcvtotal++;
607 		tcpstat.tcps_mp_rcvbytes += count;
608 
609 		mp_tp->mpt_rcvnxt += count;
610 
611 next:
612 		if (mb_dfin) {
613 			mptcp_close_fsm(mp_tp, MPCE_RECV_DATA_FIN);
614 			socantrcvmore(mp_so);
615 		}
616 		m = save;
617 		prev = save = NULL;
618 		count = mp_so->so_rcv.sb_cc;
619 	} while (m);
620 
621 	if (freelist) {
622 		m_freem(freelist);
623 	}
624 
625 	if (wakeup) {
626 		sorwakeup(mp_so);
627 	}
628 }
629 
630 boolean_t
mptcp_can_send_more(struct mptcb * mp_tp,boolean_t ignore_reinject)631 mptcp_can_send_more(struct mptcb *mp_tp, boolean_t ignore_reinject)
632 {
633 	struct socket *mp_so = mptetoso(mp_tp->mpt_mpte);
634 
635 	/*
636 	 * Always send if there is data in the reinject-queue.
637 	 */
638 	if (!ignore_reinject && mp_tp->mpt_mpte->mpte_reinjectq) {
639 		return TRUE;
640 	}
641 
642 	/*
643 	 * Don't send, if:
644 	 *
645 	 * 1. snd_nxt >= snd_max : Means, basically everything has been sent.
646 	 *    Except when using TFO, we might be doing a 0-byte write.
647 	 * 2. snd_una + snd_wnd <= snd_nxt: No space in the receiver's window
648 	 * 3. snd_nxt + 1 == snd_max and we are closing: A DATA_FIN is scheduled.
649 	 */
650 
651 	if (!(mp_so->so_flags1 & SOF1_PRECONNECT_DATA) && MPTCP_SEQ_GEQ(mp_tp->mpt_sndnxt, mp_tp->mpt_sndmax)) {
652 		return FALSE;
653 	}
654 
655 	if (MPTCP_SEQ_LEQ(mp_tp->mpt_snduna + mp_tp->mpt_sndwnd, mp_tp->mpt_sndnxt)) {
656 		return FALSE;
657 	}
658 
659 	if (mp_tp->mpt_sndnxt + 1 == mp_tp->mpt_sndmax && mp_tp->mpt_state > MPTCPS_CLOSE_WAIT) {
660 		return FALSE;
661 	}
662 
663 	if (mp_tp->mpt_state >= MPTCPS_FIN_WAIT_2) {
664 		return FALSE;
665 	}
666 
667 	return TRUE;
668 }
669 
670 /*
671  * MPTCP output.
672  */
673 int
mptcp_output(struct mptses * mpte)674 mptcp_output(struct mptses *mpte)
675 {
676 	struct mptcb *mp_tp;
677 	struct mptsub *mpts;
678 	struct mptsub *mpts_tried = NULL;
679 	struct socket *mp_so;
680 	struct mptsub *preferred_mpts = NULL;
681 	uint64_t old_snd_nxt;
682 	int error = 0;
683 
684 	mp_so = mptetoso(mpte);
685 	mp_tp = mpte->mpte_mptcb;
686 
687 	socket_lock_assert_owned(mp_so);
688 
689 	if (mp_so->so_flags & SOF_DEFUNCT) {
690 		return 0;
691 	}
692 
693 	VERIFY(!(mpte->mpte_mppcb->mpp_flags & MPP_WUPCALL));
694 	mpte->mpte_mppcb->mpp_flags |= MPP_WUPCALL;
695 
696 	old_snd_nxt = mp_tp->mpt_sndnxt;
697 	while (mptcp_can_send_more(mp_tp, FALSE)) {
698 		/* get the "best" subflow to be used for transmission */
699 		mpts = mptcp_get_subflow(mpte, &preferred_mpts);
700 		if (mpts == NULL) {
701 			break;
702 		}
703 
704 		/* In case there's just one flow, we reattempt later */
705 		if (mpts_tried != NULL &&
706 		    (mpts == mpts_tried || (mpts->mpts_flags & MPTSF_FAILINGOVER))) {
707 			mpts_tried->mpts_flags &= ~MPTSF_FAILINGOVER;
708 			mpts_tried->mpts_flags |= MPTSF_ACTIVE;
709 			mptcp_start_timer(mpte, MPTT_REXMT);
710 			break;
711 		}
712 
713 		/*
714 		 * Automatic sizing of send socket buffer. Increase the send
715 		 * socket buffer size if all of the following criteria are met
716 		 *	1. the receiver has enough buffer space for this data
717 		 *	2. send buffer is filled to 7/8th with data (so we actually
718 		 *	   have data to make use of it);
719 		 */
720 		if ((mp_so->so_snd.sb_flags & (SB_AUTOSIZE | SB_TRIM)) == SB_AUTOSIZE &&
721 		    tcp_cansbgrow(&mp_so->so_snd)) {
722 			if ((mp_tp->mpt_sndwnd / 4 * 5) >= mp_so->so_snd.sb_hiwat &&
723 			    mp_so->so_snd.sb_cc >= (mp_so->so_snd.sb_hiwat / 8 * 7)) {
724 				if (sbreserve(&mp_so->so_snd,
725 				    min(mp_so->so_snd.sb_hiwat + tcp_autosndbuf_inc,
726 				    tcp_autosndbuf_max)) == 1) {
727 					mp_so->so_snd.sb_idealsize = mp_so->so_snd.sb_hiwat;
728 				}
729 			}
730 		}
731 
732 		DTRACE_MPTCP3(output, struct mptses *, mpte, struct mptsub *, mpts,
733 		    struct socket *, mp_so);
734 		error = mptcp_subflow_output(mpte, mpts, 0);
735 		if (error) {
736 			/* can be a temporary loss of source address or other error */
737 			mpts->mpts_flags |= MPTSF_FAILINGOVER;
738 			mpts->mpts_flags &= ~MPTSF_ACTIVE;
739 			mpts_tried = mpts;
740 			if (error != ECANCELED) {
741 				os_log_error(mptcp_log_handle, "%s - %lx: Error = %d mpts_flags %#x\n",
742 				    __func__, (unsigned long)VM_KERNEL_ADDRPERM(mpte),
743 				    error, mpts->mpts_flags);
744 			}
745 			break;
746 		}
747 		/* The model is to have only one active flow at a time */
748 		mpts->mpts_flags |= MPTSF_ACTIVE;
749 		mpts->mpts_probesoon = mpts->mpts_probecnt = 0;
750 
751 		/* Allows us to update the smoothed rtt */
752 		if (mptcp_probeto && mpts != preferred_mpts && preferred_mpts != NULL) {
753 			if (preferred_mpts->mpts_probesoon) {
754 				if ((tcp_now - preferred_mpts->mpts_probesoon) > mptcp_probeto) {
755 					mptcp_subflow_output(mpte, preferred_mpts, MPTCP_SUBOUT_PROBING);
756 					if (preferred_mpts->mpts_probecnt >= mptcp_probecnt) {
757 						preferred_mpts->mpts_probesoon = 0;
758 						preferred_mpts->mpts_probecnt = 0;
759 					}
760 				}
761 			} else {
762 				preferred_mpts->mpts_probesoon = tcp_now;
763 				preferred_mpts->mpts_probecnt = 0;
764 			}
765 		}
766 
767 		if (mpte->mpte_active_sub == NULL) {
768 			mpte->mpte_active_sub = mpts;
769 		} else if (mpte->mpte_active_sub != mpts) {
770 			mpte->mpte_active_sub->mpts_flags &= ~MPTSF_ACTIVE;
771 			mpte->mpte_active_sub = mpts;
772 
773 			mptcpstats_inc_switch(mpte, mpts);
774 		}
775 	}
776 
777 	if (mp_tp->mpt_state > MPTCPS_CLOSE_WAIT) {
778 		if (mp_tp->mpt_sndnxt + 1 == mp_tp->mpt_sndmax &&
779 		    mp_tp->mpt_snduna == mp_tp->mpt_sndnxt) {
780 			mptcp_finish_usrclosed(mpte);
781 		}
782 	}
783 
784 	mptcp_handle_deferred_upcalls(mpte->mpte_mppcb, MPP_WUPCALL);
785 
786 	/* subflow errors should not be percolated back up */
787 	return 0;
788 }
789 
790 
791 static struct mptsub *
mptcp_choose_subflow(struct mptsub * mpts,struct mptsub * curbest,int * currtt)792 mptcp_choose_subflow(struct mptsub *mpts, struct mptsub *curbest, int *currtt)
793 {
794 	struct tcpcb *tp = sototcpcb(mpts->mpts_socket);
795 
796 	/*
797 	 * Lower RTT? Take it, if it's our first one, or
798 	 * it doesn't has any loss, or the current one has
799 	 * loss as well.
800 	 */
801 	if (tp->t_srtt && *currtt > tp->t_srtt &&
802 	    (curbest == NULL || tp->t_rxtshift == 0 ||
803 	    sototcpcb(curbest->mpts_socket)->t_rxtshift)) {
804 		*currtt = tp->t_srtt;
805 		return mpts;
806 	}
807 
808 	/*
809 	 * If we find a subflow without loss, take it always!
810 	 */
811 	if (curbest &&
812 	    sototcpcb(curbest->mpts_socket)->t_rxtshift &&
813 	    tp->t_rxtshift == 0) {
814 		*currtt = tp->t_srtt;
815 		return mpts;
816 	}
817 
818 	return curbest != NULL ? curbest : mpts;
819 }
820 
821 static struct mptsub *
mptcp_return_subflow(struct mptsub * mpts)822 mptcp_return_subflow(struct mptsub *mpts)
823 {
824 	if (mpts && mptcp_subflow_cwnd_space(mpts->mpts_socket) <= 0) {
825 		return NULL;
826 	}
827 
828 	return mpts;
829 }
830 
831 static boolean_t
mptcp_subflow_is_slow(struct mptses * mpte,struct mptsub * mpts)832 mptcp_subflow_is_slow(struct mptses *mpte, struct mptsub *mpts)
833 {
834 	struct tcpcb *tp = sototcpcb(mpts->mpts_socket);
835 	int fail_thresh = mptcp_fail_thresh;
836 
837 	if (mpte->mpte_svctype == MPTCP_SVCTYPE_HANDOVER || mpte->mpte_svctype == MPTCP_SVCTYPE_PURE_HANDOVER) {
838 		fail_thresh *= 2;
839 	}
840 
841 	return tp->t_rxtshift >= fail_thresh &&
842 	       (mptetoso(mpte)->so_snd.sb_cc || mpte->mpte_reinjectq);
843 }
844 
845 /*
846  * Return the most eligible subflow to be used for sending data.
847  */
848 struct mptsub *
mptcp_get_subflow(struct mptses * mpte,struct mptsub ** preferred)849 mptcp_get_subflow(struct mptses *mpte, struct mptsub **preferred)
850 {
851 	struct tcpcb *besttp, *secondtp;
852 	struct inpcb *bestinp, *secondinp;
853 	struct mptsub *mpts;
854 	struct mptsub *best = NULL;
855 	struct mptsub *second_best = NULL;
856 	int exp_rtt = INT_MAX, cheap_rtt = INT_MAX;
857 
858 	/*
859 	 * First Step:
860 	 * Choose the best subflow for cellular and non-cellular interfaces.
861 	 */
862 
863 	TAILQ_FOREACH(mpts, &mpte->mpte_subflows, mpts_entry) {
864 		struct socket *so = mpts->mpts_socket;
865 		struct tcpcb *tp = sototcpcb(so);
866 		struct inpcb *inp = sotoinpcb(so);
867 
868 		/*
869 		 * First, the hard conditions to reject subflows
870 		 * (e.g., not connected,...)
871 		 */
872 		if (inp->inp_last_outifp == NULL) {
873 			continue;
874 		}
875 
876 		if (INP_WAIT_FOR_IF_FEEDBACK(inp)) {
877 			continue;
878 		}
879 
880 		/* There can only be one subflow in degraded state */
881 		if (mpts->mpts_flags & MPTSF_MP_DEGRADED) {
882 			best = mpts;
883 			break;
884 		}
885 
886 		/*
887 		 * If this subflow is waiting to finally send, do it!
888 		 */
889 		if (so->so_flags1 & SOF1_PRECONNECT_DATA) {
890 			return mptcp_return_subflow(mpts);
891 		}
892 
893 		/*
894 		 * Only send if the subflow is MP_CAPABLE. The exceptions to
895 		 * this rule (degraded or TFO) have been taken care of above.
896 		 */
897 		if (!(mpts->mpts_flags & MPTSF_MP_CAPABLE)) {
898 			continue;
899 		}
900 
901 		if ((so->so_state & SS_ISDISCONNECTED) ||
902 		    !(so->so_state & SS_ISCONNECTED) ||
903 		    !TCPS_HAVEESTABLISHED(tp->t_state) ||
904 		    tp->t_state > TCPS_CLOSE_WAIT) {
905 			continue;
906 		}
907 
908 		/*
909 		 * Second, the soft conditions to find the subflow with best
910 		 * conditions for each set (aka cellular vs non-cellular)
911 		 */
912 		if (IFNET_IS_CELLULAR(inp->inp_last_outifp)) {
913 			second_best = mptcp_choose_subflow(mpts, second_best,
914 			    &exp_rtt);
915 		} else {
916 			best = mptcp_choose_subflow(mpts, best, &cheap_rtt);
917 		}
918 	}
919 
920 	/*
921 	 * If there is no preferred or backup subflow, and there is no active
922 	 * subflow use the last usable subflow.
923 	 */
924 	if (best == NULL) {
925 		return mptcp_return_subflow(second_best);
926 	}
927 
928 	if (second_best == NULL) {
929 		return mptcp_return_subflow(best);
930 	}
931 
932 	besttp = sototcpcb(best->mpts_socket);
933 	bestinp = sotoinpcb(best->mpts_socket);
934 	secondtp = sototcpcb(second_best->mpts_socket);
935 	secondinp = sotoinpcb(second_best->mpts_socket);
936 
937 	if (preferred != NULL) {
938 		*preferred = mptcp_return_subflow(best);
939 	}
940 
941 	/*
942 	 * Second Step: Among best and second_best. Choose the one that is
943 	 * most appropriate for this particular service-type.
944 	 */
945 	if (mpte->mpte_svctype == MPTCP_SVCTYPE_PURE_HANDOVER) {
946 		return mptcp_return_subflow(best);
947 	} else if (mpte->mpte_svctype == MPTCP_SVCTYPE_HANDOVER) {
948 		/*
949 		 * Only handover if Symptoms tells us to do so.
950 		 */
951 		if (!IFNET_IS_CELLULAR(bestinp->inp_last_outifp) &&
952 		    mptcp_wifi_quality_for_session(mpte) != MPTCP_WIFI_QUALITY_GOOD &&
953 		    mptcp_subflow_is_slow(mpte, best)) {
954 			return mptcp_return_subflow(second_best);
955 		}
956 
957 		return mptcp_return_subflow(best);
958 	} else if (mpte->mpte_svctype == MPTCP_SVCTYPE_INTERACTIVE) {
959 		int rtt_thresh = mptcp_rtthist_rtthresh << TCP_RTT_SHIFT;
960 		int rto_thresh = mptcp_rtothresh;
961 
962 		/* Adjust with symptoms information */
963 		if (!IFNET_IS_CELLULAR(bestinp->inp_last_outifp) &&
964 		    mptcp_wifi_quality_for_session(mpte) != MPTCP_WIFI_QUALITY_GOOD) {
965 			rtt_thresh /= 2;
966 			rto_thresh /= 2;
967 		}
968 
969 		if (besttp->t_srtt && secondtp->t_srtt &&
970 		    besttp->t_srtt >= rtt_thresh &&
971 		    secondtp->t_srtt < rtt_thresh) {
972 			tcpstat.tcps_mp_sel_rtt++;
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 
987 			return mptcp_return_subflow(second_best);
988 		}
989 
990 		/*
991 		 * None of the above conditions for sending on the secondary
992 		 * were true. So, let's schedule on the best one, if he still
993 		 * has some space in the congestion-window.
994 		 */
995 		return mptcp_return_subflow(best);
996 	} else if (mpte->mpte_svctype >= MPTCP_SVCTYPE_AGGREGATE) {
997 		struct mptsub *tmp;
998 
999 		/*
1000 		 * We only care about RTT when aggregating
1001 		 */
1002 		if (besttp->t_srtt > secondtp->t_srtt) {
1003 			tmp = best;
1004 			best = second_best;
1005 			besttp = secondtp;
1006 			bestinp = secondinp;
1007 
1008 			second_best = tmp;
1009 			secondtp = sototcpcb(second_best->mpts_socket);
1010 			secondinp = sotoinpcb(second_best->mpts_socket);
1011 		}
1012 
1013 		/* Is there still space in the congestion window? */
1014 		if (mptcp_subflow_cwnd_space(bestinp->inp_socket) <= 0) {
1015 			return mptcp_return_subflow(second_best);
1016 		}
1017 
1018 		return mptcp_return_subflow(best);
1019 	} else {
1020 		panic("Unknown service-type configured for MPTCP");
1021 	}
1022 
1023 	return NULL;
1024 }
1025 
1026 void
mptcp_close_fsm(struct mptcb * mp_tp,uint32_t event)1027 mptcp_close_fsm(struct mptcb *mp_tp, uint32_t event)
1028 {
1029 	struct socket *mp_so = mptetoso(mp_tp->mpt_mpte);
1030 
1031 	socket_lock_assert_owned(mp_so);
1032 
1033 	DTRACE_MPTCP2(state__change, struct mptcb *, mp_tp,
1034 	    uint32_t, event);
1035 
1036 	switch (mp_tp->mpt_state) {
1037 	case MPTCPS_CLOSED:
1038 	case MPTCPS_LISTEN:
1039 		mp_tp->mpt_state = MPTCPS_TERMINATE;
1040 		break;
1041 
1042 	case MPTCPS_ESTABLISHED:
1043 		if (event == MPCE_CLOSE) {
1044 			mp_tp->mpt_state = MPTCPS_FIN_WAIT_1;
1045 			mp_tp->mpt_sndmax += 1; /* adjust for Data FIN */
1046 		} else if (event == MPCE_RECV_DATA_FIN) {
1047 			mp_tp->mpt_rcvnxt += 1; /* adj remote data FIN */
1048 			mp_tp->mpt_state = MPTCPS_CLOSE_WAIT;
1049 		}
1050 		break;
1051 
1052 	case MPTCPS_CLOSE_WAIT:
1053 		if (event == MPCE_CLOSE) {
1054 			mp_tp->mpt_state = MPTCPS_LAST_ACK;
1055 			mp_tp->mpt_sndmax += 1; /* adjust for Data FIN */
1056 		}
1057 		break;
1058 
1059 	case MPTCPS_FIN_WAIT_1:
1060 		if (event == MPCE_RECV_DATA_ACK) {
1061 			mp_tp->mpt_state = MPTCPS_FIN_WAIT_2;
1062 		} else if (event == MPCE_RECV_DATA_FIN) {
1063 			mp_tp->mpt_rcvnxt += 1; /* adj remote data FIN */
1064 			mp_tp->mpt_state = MPTCPS_CLOSING;
1065 		}
1066 		break;
1067 
1068 	case MPTCPS_CLOSING:
1069 		if (event == MPCE_RECV_DATA_ACK) {
1070 			mp_tp->mpt_state = MPTCPS_TIME_WAIT;
1071 		}
1072 		break;
1073 
1074 	case MPTCPS_LAST_ACK:
1075 		if (event == MPCE_RECV_DATA_ACK) {
1076 			mptcp_close(mp_tp->mpt_mpte, mp_tp);
1077 		}
1078 		break;
1079 
1080 	case MPTCPS_FIN_WAIT_2:
1081 		if (event == MPCE_RECV_DATA_FIN) {
1082 			mp_tp->mpt_rcvnxt += 1; /* adj remote data FIN */
1083 			mp_tp->mpt_state = MPTCPS_TIME_WAIT;
1084 		}
1085 		break;
1086 
1087 	case MPTCPS_TIME_WAIT:
1088 	case MPTCPS_TERMINATE:
1089 		break;
1090 
1091 	default:
1092 		VERIFY(0);
1093 		/* NOTREACHED */
1094 	}
1095 	DTRACE_MPTCP2(state__change, struct mptcb *, mp_tp,
1096 	    uint32_t, event);
1097 }
1098 
1099 /* If you change this function, match up mptcp_update_rcv_state_f */
1100 void
mptcp_update_dss_rcv_state(struct mptcp_dsn_opt * dss_info,struct tcpcb * tp,uint16_t csum)1101 mptcp_update_dss_rcv_state(struct mptcp_dsn_opt *dss_info, struct tcpcb *tp,
1102     uint16_t csum)
1103 {
1104 	struct mptcb *mp_tp = tptomptp(tp);
1105 	u_int64_t full_dsn = 0;
1106 
1107 	NTOHL(dss_info->mdss_dsn);
1108 	NTOHL(dss_info->mdss_subflow_seqn);
1109 	NTOHS(dss_info->mdss_data_len);
1110 
1111 	/* XXX for autosndbuf grow sb here */
1112 	MPTCP_EXTEND_DSN(mp_tp->mpt_rcvnxt, dss_info->mdss_dsn, full_dsn);
1113 	mptcp_update_rcv_state_meat(mp_tp, tp,
1114 	    full_dsn, dss_info->mdss_subflow_seqn, dss_info->mdss_data_len,
1115 	    csum);
1116 }
1117 
1118 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)1119 mptcp_update_rcv_state_meat(struct mptcb *mp_tp, struct tcpcb *tp,
1120     u_int64_t full_dsn, u_int32_t seqn, u_int16_t mdss_data_len,
1121     uint16_t csum)
1122 {
1123 	if (mdss_data_len == 0) {
1124 		os_log_error(mptcp_log_handle, "%s - %lx: Infinite Mapping.\n",
1125 		    __func__, (unsigned long)VM_KERNEL_ADDRPERM(mp_tp->mpt_mpte));
1126 
1127 		if ((mp_tp->mpt_flags & MPTCPF_CHECKSUM) && (csum != 0)) {
1128 			os_log_error(mptcp_log_handle, "%s - %lx: Bad checksum %x \n",
1129 			    __func__, (unsigned long)VM_KERNEL_ADDRPERM(mp_tp->mpt_mpte), csum);
1130 		}
1131 		mptcp_notify_mpfail(tp->t_inpcb->inp_socket);
1132 		return;
1133 	}
1134 
1135 	mptcp_notify_mpready(tp->t_inpcb->inp_socket);
1136 
1137 	tp->t_rcv_map.mpt_dsn = full_dsn;
1138 	tp->t_rcv_map.mpt_sseq = seqn;
1139 	tp->t_rcv_map.mpt_len = mdss_data_len;
1140 	tp->t_rcv_map.mpt_csum = csum;
1141 	tp->t_mpflags |= TMPF_EMBED_DSN;
1142 }
1143 
1144 
1145 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)1146 mptcp_input_csum(struct tcpcb *tp, struct mbuf *m, uint64_t dsn, uint32_t sseq,
1147     uint16_t dlen, uint16_t csum, int dfin)
1148 {
1149 	struct mptcb *mp_tp = tptomptp(tp);
1150 	int real_len = dlen - dfin;
1151 	uint32_t sum = 0;
1152 
1153 	VERIFY(real_len >= 0);
1154 
1155 	if (mp_tp == NULL) {
1156 		return 0;
1157 	}
1158 
1159 	if (!(mp_tp->mpt_flags & MPTCPF_CHECKSUM)) {
1160 		return 0;
1161 	}
1162 
1163 	if (tp->t_mpflags & TMPF_TCP_FALLBACK) {
1164 		return 0;
1165 	}
1166 
1167 	/*
1168 	 * The remote side may send a packet with fewer bytes than the
1169 	 * claimed DSS checksum length.
1170 	 */
1171 	if ((int)m_length2(m, NULL) < real_len) {
1172 		return 0xffff;
1173 	}
1174 
1175 	if (real_len != 0) {
1176 		sum = m_sum16(m, 0, real_len);
1177 	}
1178 
1179 	sum += in_pseudo64(htonll(dsn), htonl(sseq), htons(dlen) + csum);
1180 	ADDCARRY(sum);
1181 
1182 	DTRACE_MPTCP3(checksum__result, struct tcpcb *, tp, struct mbuf *, m,
1183 	    uint32_t, sum);
1184 
1185 	return ~sum & 0xffff;
1186 }
1187 
1188 /*
1189  * MPTCP Checksum support
1190  * The checksum is calculated whenever the MPTCP DSS option is included
1191  * in the TCP packet. The checksum includes the sum of the MPTCP psuedo
1192  * header and the actual data indicated by the length specified in the
1193  * DSS option.
1194  */
1195 
1196 int
mptcp_validate_csum(struct tcpcb * tp,struct mbuf * m,uint64_t dsn,uint32_t sseq,uint16_t dlen,uint16_t csum,int dfin)1197 mptcp_validate_csum(struct tcpcb *tp, struct mbuf *m, uint64_t dsn,
1198     uint32_t sseq, uint16_t dlen, uint16_t csum, int dfin)
1199 {
1200 	uint16_t mptcp_csum;
1201 
1202 	mptcp_csum = mptcp_input_csum(tp, m, dsn, sseq, dlen, csum, dfin);
1203 	if (mptcp_csum) {
1204 		tp->t_mpflags |= TMPF_SND_MPFAIL;
1205 		mptcp_notify_mpfail(tp->t_inpcb->inp_socket);
1206 		m_freem(m);
1207 		tcpstat.tcps_mp_badcsum++;
1208 		return -1;
1209 	}
1210 	return 0;
1211 }
1212 
1213 uint16_t
mptcp_output_csum(struct mbuf * m,uint64_t dss_val,uint32_t sseq,uint16_t dlen)1214 mptcp_output_csum(struct mbuf *m, uint64_t dss_val, uint32_t sseq, uint16_t dlen)
1215 {
1216 	uint32_t sum = 0;
1217 
1218 	if (dlen) {
1219 		sum = m_sum16(m, 0, dlen);
1220 	}
1221 
1222 	dss_val = mptcp_hton64(dss_val);
1223 	sseq = htonl(sseq);
1224 	dlen = htons(dlen);
1225 	sum += in_pseudo64(dss_val, sseq, dlen);
1226 
1227 	ADDCARRY(sum);
1228 	sum = ~sum & 0xffff;
1229 	DTRACE_MPTCP2(checksum__result, struct mbuf *, m, uint32_t, sum);
1230 
1231 	return (uint16_t)sum;
1232 }
1233 
1234 /*
1235  * When WiFi signal starts fading, there's more loss and RTT spikes.
1236  * Check if there has been a large spike by comparing against
1237  * a tolerable RTT spike threshold.
1238  */
1239 boolean_t
mptcp_no_rto_spike(struct socket * so)1240 mptcp_no_rto_spike(struct socket *so)
1241 {
1242 	struct tcpcb *tp = intotcpcb(sotoinpcb(so));
1243 	int32_t spike = 0;
1244 
1245 	if (tp->t_rxtcur > mptcp_rtothresh) {
1246 		spike = tp->t_rxtcur - mptcp_rtothresh;
1247 	}
1248 
1249 	if (spike > 0) {
1250 		return FALSE;
1251 	} else {
1252 		return TRUE;
1253 	}
1254 }
1255 
1256 void
mptcp_handle_deferred_upcalls(struct mppcb * mpp,uint32_t flag)1257 mptcp_handle_deferred_upcalls(struct mppcb *mpp, uint32_t flag)
1258 {
1259 	VERIFY(mpp->mpp_flags & flag);
1260 	mpp->mpp_flags &= ~flag;
1261 
1262 	if (mptcp_should_defer_upcall(mpp)) {
1263 		return;
1264 	}
1265 
1266 	if (mpp->mpp_flags & MPP_SHOULD_WORKLOOP) {
1267 		mpp->mpp_flags &= ~MPP_SHOULD_WORKLOOP;
1268 
1269 		mptcp_subflow_workloop(mpp->mpp_pcbe);
1270 	}
1271 
1272 	if (mpp->mpp_flags & MPP_SHOULD_RWAKEUP) {
1273 		mpp->mpp_flags &= ~MPP_SHOULD_RWAKEUP;
1274 
1275 		sorwakeup(mpp->mpp_socket);
1276 	}
1277 
1278 	if (mpp->mpp_flags & MPP_SHOULD_WWAKEUP) {
1279 		mpp->mpp_flags &= ~MPP_SHOULD_WWAKEUP;
1280 
1281 		sowwakeup(mpp->mpp_socket);
1282 	}
1283 }
1284 
1285 static void
mptcp_reset_itfinfo(struct mpt_itf_info * info)1286 mptcp_reset_itfinfo(struct mpt_itf_info *info)
1287 {
1288 	memset(info, 0, sizeof(*info));
1289 }
1290 
1291 void
mptcp_session_necp_cb(void * handle,int action,uint32_t interface_index,uint32_t necp_flags,__unused bool * viable)1292 mptcp_session_necp_cb(void *handle, int action, uint32_t interface_index,
1293     uint32_t necp_flags, __unused bool *viable)
1294 {
1295 	boolean_t has_v4 = !!(necp_flags & NECP_CLIENT_RESULT_FLAG_HAS_IPV4);
1296 	boolean_t has_v6 = !!(necp_flags & NECP_CLIENT_RESULT_FLAG_HAS_IPV6);
1297 	boolean_t has_nat64 = !!(necp_flags & NECP_CLIENT_RESULT_FLAG_HAS_NAT64);
1298 	boolean_t low_power = !!(necp_flags & NECP_CLIENT_RESULT_FLAG_INTERFACE_LOW_POWER);
1299 	struct mppcb *mp = (struct mppcb *)handle;
1300 	struct mptses *mpte = mptompte(mp);
1301 	struct socket *mp_so;
1302 	struct mptcb *mp_tp;
1303 	uint32_t i, ifindex;
1304 	struct ifnet *ifp;
1305 	int locked = 0;
1306 
1307 	ifindex = interface_index;
1308 	VERIFY(ifindex != IFSCOPE_NONE);
1309 
1310 	/* About to be garbage-collected (see note about MPTCP/NECP interactions) */
1311 	if (mp->mpp_socket->so_usecount == 0) {
1312 		return;
1313 	}
1314 
1315 	mp_so = mptetoso(mpte);
1316 
1317 	if (action != NECP_CLIENT_CBACTION_INITIAL) {
1318 		socket_lock(mp_so, 1);
1319 		locked = 1;
1320 
1321 		/* Check again, because it might have changed while waiting */
1322 		if (mp->mpp_socket->so_usecount == 0) {
1323 			goto out;
1324 		}
1325 	}
1326 
1327 	socket_lock_assert_owned(mp_so);
1328 
1329 	mp_tp = mpte->mpte_mptcb;
1330 
1331 	ifnet_head_lock_shared();
1332 	ifp = ifindex2ifnet[ifindex];
1333 	ifnet_head_done();
1334 
1335 	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",
1336 	    __func__, (unsigned long)VM_KERNEL_ADDRPERM(mpte), action, ifindex,
1337 	    ifp && ifp->if_delegated.ifp ? ifp->if_delegated.ifp->if_index : IFSCOPE_NONE,
1338 	    mp->mpp_socket->so_usecount, mp_tp->mpt_flags, mp_tp->mpt_state,
1339 	    has_v4, has_v6, has_nat64, low_power);
1340 
1341 	/* No need on fallen back sockets */
1342 	if (mp_tp->mpt_flags & MPTCPF_FALLBACK_TO_TCP) {
1343 		goto out;
1344 	}
1345 
1346 	/*
1347 	 * When the interface goes in low-power mode we don't want to establish
1348 	 * new subflows on it. Thus, mark it internally as non-viable.
1349 	 */
1350 	if (low_power) {
1351 		action = NECP_CLIENT_CBACTION_NONVIABLE;
1352 	}
1353 
1354 	if (action == NECP_CLIENT_CBACTION_INITIAL) {
1355 		mpte->mpte_flags |= MPTE_ITFINFO_INIT;
1356 	}
1357 
1358 	if (action == NECP_CLIENT_CBACTION_NONVIABLE) {
1359 		for (i = 0; i < mpte->mpte_itfinfo_size; i++) {
1360 			if (mpte->mpte_itfinfo[i].ifindex == IFSCOPE_NONE) {
1361 				continue;
1362 			}
1363 
1364 			if (mpte->mpte_itfinfo[i].ifindex == ifindex) {
1365 				mptcp_reset_itfinfo(&mpte->mpte_itfinfo[i]);
1366 			}
1367 		}
1368 
1369 		mptcp_sched_create_subflows(mpte);
1370 	} else if (action == NECP_CLIENT_CBACTION_VIABLE ||
1371 	    action == NECP_CLIENT_CBACTION_INITIAL) {
1372 		int found_slot = 0, slot_index = -1;
1373 		struct sockaddr *dst;
1374 
1375 		if (ifp == NULL) {
1376 			goto out;
1377 		}
1378 
1379 		if (IFNET_IS_COMPANION_LINK(ifp)) {
1380 			goto out;
1381 		}
1382 
1383 		if (IFNET_IS_EXPENSIVE(ifp) &&
1384 		    (mp_so->so_restrictions & SO_RESTRICT_DENY_EXPENSIVE)) {
1385 			goto out;
1386 		}
1387 
1388 		if (IFNET_IS_CONSTRAINED(ifp) &&
1389 		    (mp_so->so_restrictions & SO_RESTRICT_DENY_CONSTRAINED)) {
1390 			goto out;
1391 		}
1392 
1393 		if (IFNET_IS_CELLULAR(ifp) &&
1394 		    (mp_so->so_restrictions & SO_RESTRICT_DENY_CELLULAR)) {
1395 			goto out;
1396 		}
1397 
1398 		if (IS_INTF_CLAT46(ifp)) {
1399 			has_v4 = FALSE;
1400 		}
1401 
1402 		/* Look for the slot on where to store/update the interface-info. */
1403 		for (i = 0; i < mpte->mpte_itfinfo_size; i++) {
1404 			/* Found a potential empty slot where we can put it */
1405 			if (mpte->mpte_itfinfo[i].ifindex == 0) {
1406 				found_slot = 1;
1407 				slot_index = i;
1408 			}
1409 
1410 			/*
1411 			 * The interface is already in our array. Check if we
1412 			 * need to update it.
1413 			 */
1414 			if (mpte->mpte_itfinfo[i].ifindex == ifindex &&
1415 			    (mpte->mpte_itfinfo[i].has_v4_conn != has_v4 ||
1416 			    mpte->mpte_itfinfo[i].has_v6_conn != has_v6 ||
1417 			    mpte->mpte_itfinfo[i].has_nat64_conn != has_nat64)) {
1418 				found_slot = 1;
1419 				slot_index = i;
1420 				break;
1421 			}
1422 
1423 			if (mpte->mpte_itfinfo[i].ifindex == ifindex) {
1424 				/*
1425 				 * Ok, it's already there and we don't need
1426 				 * to update it
1427 				 */
1428 				goto out;
1429 			}
1430 		}
1431 
1432 		dst = mptcp_get_session_dst(mpte, has_v6, has_v4);
1433 		if (dst && dst->sa_family == AF_INET &&
1434 		    has_v6 && !has_nat64 && !has_v4) {
1435 			if (found_slot) {
1436 				mpte->mpte_itfinfo[slot_index].ifindex = ifindex;
1437 				mpte->mpte_itfinfo[slot_index].has_v4_conn = has_v4;
1438 				mpte->mpte_itfinfo[slot_index].has_v6_conn = has_v6;
1439 				mpte->mpte_itfinfo[slot_index].has_nat64_conn = has_nat64;
1440 			}
1441 			goto out;
1442 		}
1443 
1444 		if (found_slot == 0) {
1445 			int new_size = mpte->mpte_itfinfo_size * 2;
1446 			struct mpt_itf_info *info = kalloc_data(sizeof(*info) * new_size, Z_ZERO);
1447 
1448 			if (info == NULL) {
1449 				os_log_error(mptcp_log_handle, "%s - %lx: malloc failed for %u\n",
1450 				    __func__, (unsigned long)VM_KERNEL_ADDRPERM(mpte), new_size);
1451 				goto out;
1452 			}
1453 
1454 			memcpy(info, mpte->mpte_itfinfo, mpte->mpte_itfinfo_size * sizeof(*info));
1455 
1456 			if (mpte->mpte_itfinfo_size > MPTE_ITFINFO_SIZE) {
1457 				kfree_data(mpte->mpte_itfinfo,
1458 				    sizeof(*info) * mpte->mpte_itfinfo_size);
1459 			}
1460 
1461 			/* We allocated a new one, thus the first must be empty */
1462 			slot_index = mpte->mpte_itfinfo_size;
1463 
1464 			mpte->mpte_itfinfo = info;
1465 			mpte->mpte_itfinfo_size = new_size;
1466 		}
1467 
1468 		VERIFY(slot_index >= 0 && slot_index < (int)mpte->mpte_itfinfo_size);
1469 		mpte->mpte_itfinfo[slot_index].ifindex = ifindex;
1470 		mpte->mpte_itfinfo[slot_index].has_v4_conn = has_v4;
1471 		mpte->mpte_itfinfo[slot_index].has_v6_conn = has_v6;
1472 		mpte->mpte_itfinfo[slot_index].has_nat64_conn = has_nat64;
1473 
1474 		mptcp_sched_create_subflows(mpte);
1475 	}
1476 
1477 out:
1478 	if (locked) {
1479 		socket_unlock(mp_so, 1);
1480 	}
1481 }
1482 
1483 void
mptcp_set_restrictions(struct socket * mp_so)1484 mptcp_set_restrictions(struct socket *mp_so)
1485 {
1486 	struct mptses *mpte = mpsotompte(mp_so);
1487 	uint32_t i;
1488 
1489 	socket_lock_assert_owned(mp_so);
1490 
1491 	ifnet_head_lock_shared();
1492 
1493 	for (i = 0; i < mpte->mpte_itfinfo_size; i++) {
1494 		struct mpt_itf_info *info = &mpte->mpte_itfinfo[i];
1495 		uint32_t ifindex = info->ifindex;
1496 		struct ifnet *ifp;
1497 
1498 		if (ifindex == IFSCOPE_NONE) {
1499 			continue;
1500 		}
1501 
1502 		ifp = ifindex2ifnet[ifindex];
1503 		if (ifp == NULL) {
1504 			continue;
1505 		}
1506 
1507 		if (IFNET_IS_EXPENSIVE(ifp) &&
1508 		    (mp_so->so_restrictions & SO_RESTRICT_DENY_EXPENSIVE)) {
1509 			info->ifindex = IFSCOPE_NONE;
1510 		}
1511 
1512 		if (IFNET_IS_CONSTRAINED(ifp) &&
1513 		    (mp_so->so_restrictions & SO_RESTRICT_DENY_CONSTRAINED)) {
1514 			info->ifindex = IFSCOPE_NONE;
1515 		}
1516 
1517 		if (IFNET_IS_CELLULAR(ifp) &&
1518 		    (mp_so->so_restrictions & SO_RESTRICT_DENY_CELLULAR)) {
1519 			info->ifindex = IFSCOPE_NONE;
1520 		}
1521 	}
1522 
1523 	ifnet_head_done();
1524 }
1525 
1526 #define DUMP_BUF_CHK() {        \
1527 	clen -= k;              \
1528 	if (clen < 1)           \
1529 	        goto done;      \
1530 	c += k;                 \
1531 }
1532 
1533 int
dump_mptcp_reass_qlen(char * str,int str_len)1534 dump_mptcp_reass_qlen(char *str, int str_len)
1535 {
1536 	char *c = str;
1537 	int k, clen = str_len;
1538 
1539 	if (mptcp_reass_total_qlen != 0) {
1540 		k = scnprintf(c, clen, "\nmptcp reass qlen %d\n", mptcp_reass_total_qlen);
1541 		DUMP_BUF_CHK();
1542 	}
1543 
1544 done:
1545 	return str_len - clen;
1546 }
1547