xref: /openbmc/linux/net/rds/recv.c (revision 3ce311af)
1 /*
2  * Copyright (c) 2006, 2019 Oracle and/or its affiliates. All rights reserved.
3  *
4  * This software is available to you under a choice of one of two
5  * licenses.  You may choose to be licensed under the terms of the GNU
6  * General Public License (GPL) Version 2, available from the file
7  * COPYING in the main directory of this source tree, or the
8  * OpenIB.org BSD license below:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *        copyright notice, this list of conditions and the following
16  *        disclaimer.
17  *
18  *      - Redistributions in binary form must reproduce the above
19  *        copyright notice, this list of conditions and the following
20  *        disclaimer in the documentation and/or other materials
21  *        provided with the distribution.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30  * SOFTWARE.
31  *
32  */
33 #include <linux/kernel.h>
34 #include <linux/slab.h>
35 #include <net/sock.h>
36 #include <linux/in.h>
37 #include <linux/export.h>
38 #include <linux/time.h>
39 #include <linux/rds.h>
40 
41 #include "rds.h"
42 
43 void rds_inc_init(struct rds_incoming *inc, struct rds_connection *conn,
44 		 struct in6_addr *saddr)
45 {
46 	refcount_set(&inc->i_refcount, 1);
47 	INIT_LIST_HEAD(&inc->i_item);
48 	inc->i_conn = conn;
49 	inc->i_saddr = *saddr;
50 	inc->i_usercopy.rdma_cookie = 0;
51 	inc->i_usercopy.rx_tstamp = ktime_set(0, 0);
52 
53 	memset(inc->i_rx_lat_trace, 0, sizeof(inc->i_rx_lat_trace));
54 }
55 EXPORT_SYMBOL_GPL(rds_inc_init);
56 
57 void rds_inc_path_init(struct rds_incoming *inc, struct rds_conn_path *cp,
58 		       struct in6_addr  *saddr)
59 {
60 	refcount_set(&inc->i_refcount, 1);
61 	INIT_LIST_HEAD(&inc->i_item);
62 	inc->i_conn = cp->cp_conn;
63 	inc->i_conn_path = cp;
64 	inc->i_saddr = *saddr;
65 	inc->i_usercopy.rdma_cookie = 0;
66 	inc->i_usercopy.rx_tstamp = ktime_set(0, 0);
67 }
68 EXPORT_SYMBOL_GPL(rds_inc_path_init);
69 
70 static void rds_inc_addref(struct rds_incoming *inc)
71 {
72 	rdsdebug("addref inc %p ref %d\n", inc, refcount_read(&inc->i_refcount));
73 	refcount_inc(&inc->i_refcount);
74 }
75 
76 void rds_inc_put(struct rds_incoming *inc)
77 {
78 	rdsdebug("put inc %p ref %d\n", inc, refcount_read(&inc->i_refcount));
79 	if (refcount_dec_and_test(&inc->i_refcount)) {
80 		BUG_ON(!list_empty(&inc->i_item));
81 
82 		inc->i_conn->c_trans->inc_free(inc);
83 	}
84 }
85 EXPORT_SYMBOL_GPL(rds_inc_put);
86 
87 static void rds_recv_rcvbuf_delta(struct rds_sock *rs, struct sock *sk,
88 				  struct rds_cong_map *map,
89 				  int delta, __be16 port)
90 {
91 	int now_congested;
92 
93 	if (delta == 0)
94 		return;
95 
96 	rs->rs_rcv_bytes += delta;
97 	if (delta > 0)
98 		rds_stats_add(s_recv_bytes_added_to_socket, delta);
99 	else
100 		rds_stats_add(s_recv_bytes_removed_from_socket, -delta);
101 
102 	/* loop transport doesn't send/recv congestion updates */
103 	if (rs->rs_transport->t_type == RDS_TRANS_LOOP)
104 		return;
105 
106 	now_congested = rs->rs_rcv_bytes > rds_sk_rcvbuf(rs);
107 
108 	rdsdebug("rs %p (%pI6c:%u) recv bytes %d buf %d "
109 	  "now_cong %d delta %d\n",
110 	  rs, &rs->rs_bound_addr,
111 	  ntohs(rs->rs_bound_port), rs->rs_rcv_bytes,
112 	  rds_sk_rcvbuf(rs), now_congested, delta);
113 
114 	/* wasn't -> am congested */
115 	if (!rs->rs_congested && now_congested) {
116 		rs->rs_congested = 1;
117 		rds_cong_set_bit(map, port);
118 		rds_cong_queue_updates(map);
119 	}
120 	/* was -> aren't congested */
121 	/* Require more free space before reporting uncongested to prevent
122 	   bouncing cong/uncong state too often */
123 	else if (rs->rs_congested && (rs->rs_rcv_bytes < (rds_sk_rcvbuf(rs)/2))) {
124 		rs->rs_congested = 0;
125 		rds_cong_clear_bit(map, port);
126 		rds_cong_queue_updates(map);
127 	}
128 
129 	/* do nothing if no change in cong state */
130 }
131 
132 static void rds_conn_peer_gen_update(struct rds_connection *conn,
133 				     u32 peer_gen_num)
134 {
135 	int i;
136 	struct rds_message *rm, *tmp;
137 	unsigned long flags;
138 
139 	WARN_ON(conn->c_trans->t_type != RDS_TRANS_TCP);
140 	if (peer_gen_num != 0) {
141 		if (conn->c_peer_gen_num != 0 &&
142 		    peer_gen_num != conn->c_peer_gen_num) {
143 			for (i = 0; i < RDS_MPATH_WORKERS; i++) {
144 				struct rds_conn_path *cp;
145 
146 				cp = &conn->c_path[i];
147 				spin_lock_irqsave(&cp->cp_lock, flags);
148 				cp->cp_next_tx_seq = 1;
149 				cp->cp_next_rx_seq = 0;
150 				list_for_each_entry_safe(rm, tmp,
151 							 &cp->cp_retrans,
152 							 m_conn_item) {
153 					set_bit(RDS_MSG_FLUSH, &rm->m_flags);
154 				}
155 				spin_unlock_irqrestore(&cp->cp_lock, flags);
156 			}
157 		}
158 		conn->c_peer_gen_num = peer_gen_num;
159 	}
160 }
161 
162 /*
163  * Process all extension headers that come with this message.
164  */
165 static void rds_recv_incoming_exthdrs(struct rds_incoming *inc, struct rds_sock *rs)
166 {
167 	struct rds_header *hdr = &inc->i_hdr;
168 	unsigned int pos = 0, type, len;
169 	union {
170 		struct rds_ext_header_version version;
171 		struct rds_ext_header_rdma rdma;
172 		struct rds_ext_header_rdma_dest rdma_dest;
173 	} buffer;
174 
175 	while (1) {
176 		len = sizeof(buffer);
177 		type = rds_message_next_extension(hdr, &pos, &buffer, &len);
178 		if (type == RDS_EXTHDR_NONE)
179 			break;
180 		/* Process extension header here */
181 		switch (type) {
182 		case RDS_EXTHDR_RDMA:
183 			rds_rdma_unuse(rs, be32_to_cpu(buffer.rdma.h_rdma_rkey), 0);
184 			break;
185 
186 		case RDS_EXTHDR_RDMA_DEST:
187 			/* We ignore the size for now. We could stash it
188 			 * somewhere and use it for error checking. */
189 			inc->i_usercopy.rdma_cookie = rds_rdma_make_cookie(
190 					be32_to_cpu(buffer.rdma_dest.h_rdma_rkey),
191 					be32_to_cpu(buffer.rdma_dest.h_rdma_offset));
192 
193 			break;
194 		}
195 	}
196 }
197 
198 static void rds_recv_hs_exthdrs(struct rds_header *hdr,
199 				struct rds_connection *conn)
200 {
201 	unsigned int pos = 0, type, len;
202 	union {
203 		struct rds_ext_header_version version;
204 		u16 rds_npaths;
205 		u32 rds_gen_num;
206 	} buffer;
207 	u32 new_peer_gen_num = 0;
208 
209 	while (1) {
210 		len = sizeof(buffer);
211 		type = rds_message_next_extension(hdr, &pos, &buffer, &len);
212 		if (type == RDS_EXTHDR_NONE)
213 			break;
214 		/* Process extension header here */
215 		switch (type) {
216 		case RDS_EXTHDR_NPATHS:
217 			conn->c_npaths = min_t(int, RDS_MPATH_WORKERS,
218 					       be16_to_cpu(buffer.rds_npaths));
219 			break;
220 		case RDS_EXTHDR_GEN_NUM:
221 			new_peer_gen_num = be32_to_cpu(buffer.rds_gen_num);
222 			break;
223 		default:
224 			pr_warn_ratelimited("ignoring unknown exthdr type "
225 					     "0x%x\n", type);
226 		}
227 	}
228 	/* if RDS_EXTHDR_NPATHS was not found, default to a single-path */
229 	conn->c_npaths = max_t(int, conn->c_npaths, 1);
230 	conn->c_ping_triggered = 0;
231 	rds_conn_peer_gen_update(conn, new_peer_gen_num);
232 }
233 
234 /* rds_start_mprds() will synchronously start multiple paths when appropriate.
235  * The scheme is based on the following rules:
236  *
237  * 1. rds_sendmsg on first connect attempt sends the probe ping, with the
238  *    sender's npaths (s_npaths)
239  * 2. rcvr of probe-ping knows the mprds_paths = min(s_npaths, r_npaths). It
240  *    sends back a probe-pong with r_npaths. After that, if rcvr is the
241  *    smaller ip addr, it starts rds_conn_path_connect_if_down on all
242  *    mprds_paths.
243  * 3. sender gets woken up, and can move to rds_conn_path_connect_if_down.
244  *    If it is the smaller ipaddr, rds_conn_path_connect_if_down can be
245  *    called after reception of the probe-pong on all mprds_paths.
246  *    Otherwise (sender of probe-ping is not the smaller ip addr): just call
247  *    rds_conn_path_connect_if_down on the hashed path. (see rule 4)
248  * 4. rds_connect_worker must only trigger a connection if laddr < faddr.
249  * 5. sender may end up queuing the packet on the cp. will get sent out later.
250  *    when connection is completed.
251  */
252 static void rds_start_mprds(struct rds_connection *conn)
253 {
254 	int i;
255 	struct rds_conn_path *cp;
256 
257 	if (conn->c_npaths > 1 &&
258 	    rds_addr_cmp(&conn->c_laddr, &conn->c_faddr) < 0) {
259 		for (i = 0; i < conn->c_npaths; i++) {
260 			cp = &conn->c_path[i];
261 			rds_conn_path_connect_if_down(cp);
262 		}
263 	}
264 }
265 
266 /*
267  * The transport must make sure that this is serialized against other
268  * rx and conn reset on this specific conn.
269  *
270  * We currently assert that only one fragmented message will be sent
271  * down a connection at a time.  This lets us reassemble in the conn
272  * instead of per-flow which means that we don't have to go digging through
273  * flows to tear down partial reassembly progress on conn failure and
274  * we save flow lookup and locking for each frag arrival.  It does mean
275  * that small messages will wait behind large ones.  Fragmenting at all
276  * is only to reduce the memory consumption of pre-posted buffers.
277  *
278  * The caller passes in saddr and daddr instead of us getting it from the
279  * conn.  This lets loopback, who only has one conn for both directions,
280  * tell us which roles the addrs in the conn are playing for this message.
281  */
282 void rds_recv_incoming(struct rds_connection *conn, struct in6_addr *saddr,
283 		       struct in6_addr *daddr,
284 		       struct rds_incoming *inc, gfp_t gfp)
285 {
286 	struct rds_sock *rs = NULL;
287 	struct sock *sk;
288 	unsigned long flags;
289 	struct rds_conn_path *cp;
290 
291 	inc->i_conn = conn;
292 	inc->i_rx_jiffies = jiffies;
293 	if (conn->c_trans->t_mp_capable)
294 		cp = inc->i_conn_path;
295 	else
296 		cp = &conn->c_path[0];
297 
298 	rdsdebug("conn %p next %llu inc %p seq %llu len %u sport %u dport %u "
299 		 "flags 0x%x rx_jiffies %lu\n", conn,
300 		 (unsigned long long)cp->cp_next_rx_seq,
301 		 inc,
302 		 (unsigned long long)be64_to_cpu(inc->i_hdr.h_sequence),
303 		 be32_to_cpu(inc->i_hdr.h_len),
304 		 be16_to_cpu(inc->i_hdr.h_sport),
305 		 be16_to_cpu(inc->i_hdr.h_dport),
306 		 inc->i_hdr.h_flags,
307 		 inc->i_rx_jiffies);
308 
309 	/*
310 	 * Sequence numbers should only increase.  Messages get their
311 	 * sequence number as they're queued in a sending conn.  They
312 	 * can be dropped, though, if the sending socket is closed before
313 	 * they hit the wire.  So sequence numbers can skip forward
314 	 * under normal operation.  They can also drop back in the conn
315 	 * failover case as previously sent messages are resent down the
316 	 * new instance of a conn.  We drop those, otherwise we have
317 	 * to assume that the next valid seq does not come after a
318 	 * hole in the fragment stream.
319 	 *
320 	 * The headers don't give us a way to realize if fragments of
321 	 * a message have been dropped.  We assume that frags that arrive
322 	 * to a flow are part of the current message on the flow that is
323 	 * being reassembled.  This means that senders can't drop messages
324 	 * from the sending conn until all their frags are sent.
325 	 *
326 	 * XXX we could spend more on the wire to get more robust failure
327 	 * detection, arguably worth it to avoid data corruption.
328 	 */
329 	if (be64_to_cpu(inc->i_hdr.h_sequence) < cp->cp_next_rx_seq &&
330 	    (inc->i_hdr.h_flags & RDS_FLAG_RETRANSMITTED)) {
331 		rds_stats_inc(s_recv_drop_old_seq);
332 		goto out;
333 	}
334 	cp->cp_next_rx_seq = be64_to_cpu(inc->i_hdr.h_sequence) + 1;
335 
336 	if (rds_sysctl_ping_enable && inc->i_hdr.h_dport == 0) {
337 		if (inc->i_hdr.h_sport == 0) {
338 			rdsdebug("ignore ping with 0 sport from %pI6c\n",
339 				 saddr);
340 			goto out;
341 		}
342 		rds_stats_inc(s_recv_ping);
343 		rds_send_pong(cp, inc->i_hdr.h_sport);
344 		/* if this is a handshake ping, start multipath if necessary */
345 		if (RDS_HS_PROBE(be16_to_cpu(inc->i_hdr.h_sport),
346 				 be16_to_cpu(inc->i_hdr.h_dport))) {
347 			rds_recv_hs_exthdrs(&inc->i_hdr, cp->cp_conn);
348 			rds_start_mprds(cp->cp_conn);
349 		}
350 		goto out;
351 	}
352 
353 	if (be16_to_cpu(inc->i_hdr.h_dport) ==  RDS_FLAG_PROBE_PORT &&
354 	    inc->i_hdr.h_sport == 0) {
355 		rds_recv_hs_exthdrs(&inc->i_hdr, cp->cp_conn);
356 		/* if this is a handshake pong, start multipath if necessary */
357 		rds_start_mprds(cp->cp_conn);
358 		wake_up(&cp->cp_conn->c_hs_waitq);
359 		goto out;
360 	}
361 
362 	rs = rds_find_bound(daddr, inc->i_hdr.h_dport, conn->c_bound_if);
363 	if (!rs) {
364 		rds_stats_inc(s_recv_drop_no_sock);
365 		goto out;
366 	}
367 
368 	/* Process extension headers */
369 	rds_recv_incoming_exthdrs(inc, rs);
370 
371 	/* We can be racing with rds_release() which marks the socket dead. */
372 	sk = rds_rs_to_sk(rs);
373 
374 	/* serialize with rds_release -> sock_orphan */
375 	write_lock_irqsave(&rs->rs_recv_lock, flags);
376 	if (!sock_flag(sk, SOCK_DEAD)) {
377 		rdsdebug("adding inc %p to rs %p's recv queue\n", inc, rs);
378 		rds_stats_inc(s_recv_queued);
379 		rds_recv_rcvbuf_delta(rs, sk, inc->i_conn->c_lcong,
380 				      be32_to_cpu(inc->i_hdr.h_len),
381 				      inc->i_hdr.h_dport);
382 		if (sock_flag(sk, SOCK_RCVTSTAMP))
383 			inc->i_usercopy.rx_tstamp = ktime_get_real();
384 		rds_inc_addref(inc);
385 		inc->i_rx_lat_trace[RDS_MSG_RX_END] = local_clock();
386 		list_add_tail(&inc->i_item, &rs->rs_recv_queue);
387 		__rds_wake_sk_sleep(sk);
388 	} else {
389 		rds_stats_inc(s_recv_drop_dead_sock);
390 	}
391 	write_unlock_irqrestore(&rs->rs_recv_lock, flags);
392 
393 out:
394 	if (rs)
395 		rds_sock_put(rs);
396 }
397 EXPORT_SYMBOL_GPL(rds_recv_incoming);
398 
399 /*
400  * be very careful here.  This is being called as the condition in
401  * wait_event_*() needs to cope with being called many times.
402  */
403 static int rds_next_incoming(struct rds_sock *rs, struct rds_incoming **inc)
404 {
405 	unsigned long flags;
406 
407 	if (!*inc) {
408 		read_lock_irqsave(&rs->rs_recv_lock, flags);
409 		if (!list_empty(&rs->rs_recv_queue)) {
410 			*inc = list_entry(rs->rs_recv_queue.next,
411 					  struct rds_incoming,
412 					  i_item);
413 			rds_inc_addref(*inc);
414 		}
415 		read_unlock_irqrestore(&rs->rs_recv_lock, flags);
416 	}
417 
418 	return *inc != NULL;
419 }
420 
421 static int rds_still_queued(struct rds_sock *rs, struct rds_incoming *inc,
422 			    int drop)
423 {
424 	struct sock *sk = rds_rs_to_sk(rs);
425 	int ret = 0;
426 	unsigned long flags;
427 
428 	write_lock_irqsave(&rs->rs_recv_lock, flags);
429 	if (!list_empty(&inc->i_item)) {
430 		ret = 1;
431 		if (drop) {
432 			/* XXX make sure this i_conn is reliable */
433 			rds_recv_rcvbuf_delta(rs, sk, inc->i_conn->c_lcong,
434 					      -be32_to_cpu(inc->i_hdr.h_len),
435 					      inc->i_hdr.h_dport);
436 			list_del_init(&inc->i_item);
437 			rds_inc_put(inc);
438 		}
439 	}
440 	write_unlock_irqrestore(&rs->rs_recv_lock, flags);
441 
442 	rdsdebug("inc %p rs %p still %d dropped %d\n", inc, rs, ret, drop);
443 	return ret;
444 }
445 
446 /*
447  * Pull errors off the error queue.
448  * If msghdr is NULL, we will just purge the error queue.
449  */
450 int rds_notify_queue_get(struct rds_sock *rs, struct msghdr *msghdr)
451 {
452 	struct rds_notifier *notifier;
453 	struct rds_rdma_notify cmsg = { 0 }; /* fill holes with zero */
454 	unsigned int count = 0, max_messages = ~0U;
455 	unsigned long flags;
456 	LIST_HEAD(copy);
457 	int err = 0;
458 
459 
460 	/* put_cmsg copies to user space and thus may sleep. We can't do this
461 	 * with rs_lock held, so first grab as many notifications as we can stuff
462 	 * in the user provided cmsg buffer. We don't try to copy more, to avoid
463 	 * losing notifications - except when the buffer is so small that it wouldn't
464 	 * even hold a single notification. Then we give him as much of this single
465 	 * msg as we can squeeze in, and set MSG_CTRUNC.
466 	 */
467 	if (msghdr) {
468 		max_messages = msghdr->msg_controllen / CMSG_SPACE(sizeof(cmsg));
469 		if (!max_messages)
470 			max_messages = 1;
471 	}
472 
473 	spin_lock_irqsave(&rs->rs_lock, flags);
474 	while (!list_empty(&rs->rs_notify_queue) && count < max_messages) {
475 		notifier = list_entry(rs->rs_notify_queue.next,
476 				struct rds_notifier, n_list);
477 		list_move(&notifier->n_list, &copy);
478 		count++;
479 	}
480 	spin_unlock_irqrestore(&rs->rs_lock, flags);
481 
482 	if (!count)
483 		return 0;
484 
485 	while (!list_empty(&copy)) {
486 		notifier = list_entry(copy.next, struct rds_notifier, n_list);
487 
488 		if (msghdr) {
489 			cmsg.user_token = notifier->n_user_token;
490 			cmsg.status = notifier->n_status;
491 
492 			err = put_cmsg(msghdr, SOL_RDS, RDS_CMSG_RDMA_STATUS,
493 				       sizeof(cmsg), &cmsg);
494 			if (err)
495 				break;
496 		}
497 
498 		list_del_init(&notifier->n_list);
499 		kfree(notifier);
500 	}
501 
502 	/* If we bailed out because of an error in put_cmsg,
503 	 * we may be left with one or more notifications that we
504 	 * didn't process. Return them to the head of the list. */
505 	if (!list_empty(&copy)) {
506 		spin_lock_irqsave(&rs->rs_lock, flags);
507 		list_splice(&copy, &rs->rs_notify_queue);
508 		spin_unlock_irqrestore(&rs->rs_lock, flags);
509 	}
510 
511 	return err;
512 }
513 
514 /*
515  * Queue a congestion notification
516  */
517 static int rds_notify_cong(struct rds_sock *rs, struct msghdr *msghdr)
518 {
519 	uint64_t notify = rs->rs_cong_notify;
520 	unsigned long flags;
521 	int err;
522 
523 	err = put_cmsg(msghdr, SOL_RDS, RDS_CMSG_CONG_UPDATE,
524 			sizeof(notify), &notify);
525 	if (err)
526 		return err;
527 
528 	spin_lock_irqsave(&rs->rs_lock, flags);
529 	rs->rs_cong_notify &= ~notify;
530 	spin_unlock_irqrestore(&rs->rs_lock, flags);
531 
532 	return 0;
533 }
534 
535 /*
536  * Receive any control messages.
537  */
538 static int rds_cmsg_recv(struct rds_incoming *inc, struct msghdr *msg,
539 			 struct rds_sock *rs)
540 {
541 	int ret = 0;
542 
543 	if (inc->i_usercopy.rdma_cookie) {
544 		ret = put_cmsg(msg, SOL_RDS, RDS_CMSG_RDMA_DEST,
545 				sizeof(inc->i_usercopy.rdma_cookie),
546 				&inc->i_usercopy.rdma_cookie);
547 		if (ret)
548 			goto out;
549 	}
550 
551 	if ((inc->i_usercopy.rx_tstamp != 0) &&
552 	    sock_flag(rds_rs_to_sk(rs), SOCK_RCVTSTAMP)) {
553 		struct __kernel_old_timeval tv =
554 			ns_to_kernel_old_timeval(inc->i_usercopy.rx_tstamp);
555 
556 		if (!sock_flag(rds_rs_to_sk(rs), SOCK_TSTAMP_NEW)) {
557 			ret = put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_OLD,
558 				       sizeof(tv), &tv);
559 		} else {
560 			struct __kernel_sock_timeval sk_tv;
561 
562 			sk_tv.tv_sec = tv.tv_sec;
563 			sk_tv.tv_usec = tv.tv_usec;
564 
565 			ret = put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_NEW,
566 				       sizeof(sk_tv), &sk_tv);
567 		}
568 
569 		if (ret)
570 			goto out;
571 	}
572 
573 	if (rs->rs_rx_traces) {
574 		struct rds_cmsg_rx_trace t;
575 		int i, j;
576 
577 		memset(&t, 0, sizeof(t));
578 		inc->i_rx_lat_trace[RDS_MSG_RX_CMSG] = local_clock();
579 		t.rx_traces =  rs->rs_rx_traces;
580 		for (i = 0; i < rs->rs_rx_traces; i++) {
581 			j = rs->rs_rx_trace[i];
582 			t.rx_trace_pos[i] = j;
583 			t.rx_trace[i] = inc->i_rx_lat_trace[j + 1] -
584 					  inc->i_rx_lat_trace[j];
585 		}
586 
587 		ret = put_cmsg(msg, SOL_RDS, RDS_CMSG_RXPATH_LATENCY,
588 			       sizeof(t), &t);
589 		if (ret)
590 			goto out;
591 	}
592 
593 out:
594 	return ret;
595 }
596 
597 static bool rds_recvmsg_zcookie(struct rds_sock *rs, struct msghdr *msg)
598 {
599 	struct rds_msg_zcopy_queue *q = &rs->rs_zcookie_queue;
600 	struct rds_msg_zcopy_info *info = NULL;
601 	struct rds_zcopy_cookies *done;
602 	unsigned long flags;
603 
604 	if (!msg->msg_control)
605 		return false;
606 
607 	if (!sock_flag(rds_rs_to_sk(rs), SOCK_ZEROCOPY) ||
608 	    msg->msg_controllen < CMSG_SPACE(sizeof(*done)))
609 		return false;
610 
611 	spin_lock_irqsave(&q->lock, flags);
612 	if (!list_empty(&q->zcookie_head)) {
613 		info = list_entry(q->zcookie_head.next,
614 				  struct rds_msg_zcopy_info, rs_zcookie_next);
615 		list_del(&info->rs_zcookie_next);
616 	}
617 	spin_unlock_irqrestore(&q->lock, flags);
618 	if (!info)
619 		return false;
620 	done = &info->zcookies;
621 	if (put_cmsg(msg, SOL_RDS, RDS_CMSG_ZCOPY_COMPLETION, sizeof(*done),
622 		     done)) {
623 		spin_lock_irqsave(&q->lock, flags);
624 		list_add(&info->rs_zcookie_next, &q->zcookie_head);
625 		spin_unlock_irqrestore(&q->lock, flags);
626 		return false;
627 	}
628 	kfree(info);
629 	return true;
630 }
631 
632 int rds_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
633 		int msg_flags)
634 {
635 	struct sock *sk = sock->sk;
636 	struct rds_sock *rs = rds_sk_to_rs(sk);
637 	long timeo;
638 	int ret = 0, nonblock = msg_flags & MSG_DONTWAIT;
639 	DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
640 	DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
641 	struct rds_incoming *inc = NULL;
642 
643 	/* udp_recvmsg()->sock_recvtimeo() gets away without locking too.. */
644 	timeo = sock_rcvtimeo(sk, nonblock);
645 
646 	rdsdebug("size %zu flags 0x%x timeo %ld\n", size, msg_flags, timeo);
647 
648 	if (msg_flags & MSG_OOB)
649 		goto out;
650 	if (msg_flags & MSG_ERRQUEUE)
651 		return sock_recv_errqueue(sk, msg, size, SOL_IP, IP_RECVERR);
652 
653 	while (1) {
654 		/* If there are pending notifications, do those - and nothing else */
655 		if (!list_empty(&rs->rs_notify_queue)) {
656 			ret = rds_notify_queue_get(rs, msg);
657 			break;
658 		}
659 
660 		if (rs->rs_cong_notify) {
661 			ret = rds_notify_cong(rs, msg);
662 			break;
663 		}
664 
665 		if (!rds_next_incoming(rs, &inc)) {
666 			if (nonblock) {
667 				bool reaped = rds_recvmsg_zcookie(rs, msg);
668 
669 				ret = reaped ?  0 : -EAGAIN;
670 				break;
671 			}
672 
673 			timeo = wait_event_interruptible_timeout(*sk_sleep(sk),
674 					(!list_empty(&rs->rs_notify_queue) ||
675 					 rs->rs_cong_notify ||
676 					 rds_next_incoming(rs, &inc)), timeo);
677 			rdsdebug("recvmsg woke inc %p timeo %ld\n", inc,
678 				 timeo);
679 			if (timeo > 0 || timeo == MAX_SCHEDULE_TIMEOUT)
680 				continue;
681 
682 			ret = timeo;
683 			if (ret == 0)
684 				ret = -ETIMEDOUT;
685 			break;
686 		}
687 
688 		rdsdebug("copying inc %p from %pI6c:%u to user\n", inc,
689 			 &inc->i_conn->c_faddr,
690 			 ntohs(inc->i_hdr.h_sport));
691 		ret = inc->i_conn->c_trans->inc_copy_to_user(inc, &msg->msg_iter);
692 		if (ret < 0)
693 			break;
694 
695 		/*
696 		 * if the message we just copied isn't at the head of the
697 		 * recv queue then someone else raced us to return it, try
698 		 * to get the next message.
699 		 */
700 		if (!rds_still_queued(rs, inc, !(msg_flags & MSG_PEEK))) {
701 			rds_inc_put(inc);
702 			inc = NULL;
703 			rds_stats_inc(s_recv_deliver_raced);
704 			iov_iter_revert(&msg->msg_iter, ret);
705 			continue;
706 		}
707 
708 		if (ret < be32_to_cpu(inc->i_hdr.h_len)) {
709 			if (msg_flags & MSG_TRUNC)
710 				ret = be32_to_cpu(inc->i_hdr.h_len);
711 			msg->msg_flags |= MSG_TRUNC;
712 		}
713 
714 		if (rds_cmsg_recv(inc, msg, rs)) {
715 			ret = -EFAULT;
716 			goto out;
717 		}
718 		rds_recvmsg_zcookie(rs, msg);
719 
720 		rds_stats_inc(s_recv_delivered);
721 
722 		if (msg->msg_name) {
723 			if (ipv6_addr_v4mapped(&inc->i_saddr)) {
724 				sin = (struct sockaddr_in *)msg->msg_name;
725 
726 				sin->sin_family = AF_INET;
727 				sin->sin_port = inc->i_hdr.h_sport;
728 				sin->sin_addr.s_addr =
729 				    inc->i_saddr.s6_addr32[3];
730 				memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
731 				msg->msg_namelen = sizeof(*sin);
732 			} else {
733 				sin6 = (struct sockaddr_in6 *)msg->msg_name;
734 
735 				sin6->sin6_family = AF_INET6;
736 				sin6->sin6_port = inc->i_hdr.h_sport;
737 				sin6->sin6_addr = inc->i_saddr;
738 				sin6->sin6_flowinfo = 0;
739 				sin6->sin6_scope_id = rs->rs_bound_scope_id;
740 				msg->msg_namelen = sizeof(*sin6);
741 			}
742 		}
743 		break;
744 	}
745 
746 	if (inc)
747 		rds_inc_put(inc);
748 
749 out:
750 	return ret;
751 }
752 
753 /*
754  * The socket is being shut down and we're asked to drop messages that were
755  * queued for recvmsg.  The caller has unbound the socket so the receive path
756  * won't queue any more incoming fragments or messages on the socket.
757  */
758 void rds_clear_recv_queue(struct rds_sock *rs)
759 {
760 	struct sock *sk = rds_rs_to_sk(rs);
761 	struct rds_incoming *inc, *tmp;
762 	unsigned long flags;
763 
764 	write_lock_irqsave(&rs->rs_recv_lock, flags);
765 	list_for_each_entry_safe(inc, tmp, &rs->rs_recv_queue, i_item) {
766 		rds_recv_rcvbuf_delta(rs, sk, inc->i_conn->c_lcong,
767 				      -be32_to_cpu(inc->i_hdr.h_len),
768 				      inc->i_hdr.h_dport);
769 		list_del_init(&inc->i_item);
770 		rds_inc_put(inc);
771 	}
772 	write_unlock_irqrestore(&rs->rs_recv_lock, flags);
773 }
774 
775 /*
776  * inc->i_saddr isn't used here because it is only set in the receive
777  * path.
778  */
779 void rds_inc_info_copy(struct rds_incoming *inc,
780 		       struct rds_info_iterator *iter,
781 		       __be32 saddr, __be32 daddr, int flip)
782 {
783 	struct rds_info_message minfo;
784 
785 	minfo.seq = be64_to_cpu(inc->i_hdr.h_sequence);
786 	minfo.len = be32_to_cpu(inc->i_hdr.h_len);
787 	minfo.tos = inc->i_conn->c_tos;
788 
789 	if (flip) {
790 		minfo.laddr = daddr;
791 		minfo.faddr = saddr;
792 		minfo.lport = inc->i_hdr.h_dport;
793 		minfo.fport = inc->i_hdr.h_sport;
794 	} else {
795 		minfo.laddr = saddr;
796 		minfo.faddr = daddr;
797 		minfo.lport = inc->i_hdr.h_sport;
798 		minfo.fport = inc->i_hdr.h_dport;
799 	}
800 
801 	minfo.flags = 0;
802 
803 	rds_info_copy(iter, &minfo, sizeof(minfo));
804 }
805 
806 #if IS_ENABLED(CONFIG_IPV6)
807 void rds6_inc_info_copy(struct rds_incoming *inc,
808 			struct rds_info_iterator *iter,
809 			struct in6_addr *saddr, struct in6_addr *daddr,
810 			int flip)
811 {
812 	struct rds6_info_message minfo6;
813 
814 	minfo6.seq = be64_to_cpu(inc->i_hdr.h_sequence);
815 	minfo6.len = be32_to_cpu(inc->i_hdr.h_len);
816 	minfo6.tos = inc->i_conn->c_tos;
817 
818 	if (flip) {
819 		minfo6.laddr = *daddr;
820 		minfo6.faddr = *saddr;
821 		minfo6.lport = inc->i_hdr.h_dport;
822 		minfo6.fport = inc->i_hdr.h_sport;
823 	} else {
824 		minfo6.laddr = *saddr;
825 		minfo6.faddr = *daddr;
826 		minfo6.lport = inc->i_hdr.h_sport;
827 		minfo6.fport = inc->i_hdr.h_dport;
828 	}
829 
830 	minfo6.flags = 0;
831 
832 	rds_info_copy(iter, &minfo6, sizeof(minfo6));
833 }
834 #endif
835