2 * Copyright (c) 2006, 2019 Oracle and/or its affiliates. All rights reserved.
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:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
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.
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
33 #include <linux/kernel.h>
34 #include <linux/slab.h>
37 #include <linux/export.h>
38 #include <linux/time.h>
39 #include <linux/rds.h>
43 void rds_inc_init(struct rds_incoming *inc, struct rds_connection *conn,
44 struct in6_addr *saddr)
48 refcount_set(&inc->i_refcount, 1);
49 INIT_LIST_HEAD(&inc->i_item);
51 inc->i_saddr = *saddr;
52 inc->i_rdma_cookie = 0;
53 inc->i_rx_tstamp.tv_sec = 0;
54 inc->i_rx_tstamp.tv_usec = 0;
56 for (i = 0; i < RDS_RX_MAX_TRACES; i++)
57 inc->i_rx_lat_trace[i] = 0;
59 EXPORT_SYMBOL_GPL(rds_inc_init);
61 void rds_inc_path_init(struct rds_incoming *inc, struct rds_conn_path *cp,
62 struct in6_addr *saddr)
64 refcount_set(&inc->i_refcount, 1);
65 INIT_LIST_HEAD(&inc->i_item);
66 inc->i_conn = cp->cp_conn;
67 inc->i_conn_path = cp;
68 inc->i_saddr = *saddr;
69 inc->i_rdma_cookie = 0;
70 inc->i_rx_tstamp.tv_sec = 0;
71 inc->i_rx_tstamp.tv_usec = 0;
73 EXPORT_SYMBOL_GPL(rds_inc_path_init);
75 static void rds_inc_addref(struct rds_incoming *inc)
77 rdsdebug("addref inc %p ref %d\n", inc, refcount_read(&inc->i_refcount));
78 refcount_inc(&inc->i_refcount);
81 void rds_inc_put(struct rds_incoming *inc)
83 rdsdebug("put inc %p ref %d\n", inc, refcount_read(&inc->i_refcount));
84 if (refcount_dec_and_test(&inc->i_refcount)) {
85 BUG_ON(!list_empty(&inc->i_item));
87 inc->i_conn->c_trans->inc_free(inc);
90 EXPORT_SYMBOL_GPL(rds_inc_put);
92 static void rds_recv_rcvbuf_delta(struct rds_sock *rs, struct sock *sk,
93 struct rds_cong_map *map,
94 int delta, __be16 port)
101 rs->rs_rcv_bytes += delta;
103 rds_stats_add(s_recv_bytes_added_to_socket, delta);
105 rds_stats_add(s_recv_bytes_removed_from_socket, -delta);
107 /* loop transport doesn't send/recv congestion updates */
108 if (rs->rs_transport->t_type == RDS_TRANS_LOOP)
111 now_congested = rs->rs_rcv_bytes > rds_sk_rcvbuf(rs);
113 rdsdebug("rs %p (%pI6c:%u) recv bytes %d buf %d "
114 "now_cong %d delta %d\n",
115 rs, &rs->rs_bound_addr,
116 ntohs(rs->rs_bound_port), rs->rs_rcv_bytes,
117 rds_sk_rcvbuf(rs), now_congested, delta);
119 /* wasn't -> am congested */
120 if (!rs->rs_congested && now_congested) {
121 rs->rs_congested = 1;
122 rds_cong_set_bit(map, port);
123 rds_cong_queue_updates(map);
125 /* was -> aren't congested */
126 /* Require more free space before reporting uncongested to prevent
127 bouncing cong/uncong state too often */
128 else if (rs->rs_congested && (rs->rs_rcv_bytes < (rds_sk_rcvbuf(rs)/2))) {
129 rs->rs_congested = 0;
130 rds_cong_clear_bit(map, port);
131 rds_cong_queue_updates(map);
134 /* do nothing if no change in cong state */
137 static void rds_conn_peer_gen_update(struct rds_connection *conn,
141 struct rds_message *rm, *tmp;
144 WARN_ON(conn->c_trans->t_type != RDS_TRANS_TCP);
145 if (peer_gen_num != 0) {
146 if (conn->c_peer_gen_num != 0 &&
147 peer_gen_num != conn->c_peer_gen_num) {
148 for (i = 0; i < RDS_MPATH_WORKERS; i++) {
149 struct rds_conn_path *cp;
151 cp = &conn->c_path[i];
152 spin_lock_irqsave(&cp->cp_lock, flags);
153 cp->cp_next_tx_seq = 1;
154 cp->cp_next_rx_seq = 0;
155 list_for_each_entry_safe(rm, tmp,
158 set_bit(RDS_MSG_FLUSH, &rm->m_flags);
160 spin_unlock_irqrestore(&cp->cp_lock, flags);
163 conn->c_peer_gen_num = peer_gen_num;
168 * Process all extension headers that come with this message.
170 static void rds_recv_incoming_exthdrs(struct rds_incoming *inc, struct rds_sock *rs)
172 struct rds_header *hdr = &inc->i_hdr;
173 unsigned int pos = 0, type, len;
175 struct rds_ext_header_version version;
176 struct rds_ext_header_rdma rdma;
177 struct rds_ext_header_rdma_dest rdma_dest;
181 len = sizeof(buffer);
182 type = rds_message_next_extension(hdr, &pos, &buffer, &len);
183 if (type == RDS_EXTHDR_NONE)
185 /* Process extension header here */
187 case RDS_EXTHDR_RDMA:
188 rds_rdma_unuse(rs, be32_to_cpu(buffer.rdma.h_rdma_rkey), 0);
191 case RDS_EXTHDR_RDMA_DEST:
192 /* We ignore the size for now. We could stash it
193 * somewhere and use it for error checking. */
194 inc->i_rdma_cookie = rds_rdma_make_cookie(
195 be32_to_cpu(buffer.rdma_dest.h_rdma_rkey),
196 be32_to_cpu(buffer.rdma_dest.h_rdma_offset));
203 static void rds_recv_hs_exthdrs(struct rds_header *hdr,
204 struct rds_connection *conn)
206 unsigned int pos = 0, type, len;
208 struct rds_ext_header_version version;
212 u32 new_peer_gen_num = 0;
215 len = sizeof(buffer);
216 type = rds_message_next_extension(hdr, &pos, &buffer, &len);
217 if (type == RDS_EXTHDR_NONE)
219 /* Process extension header here */
221 case RDS_EXTHDR_NPATHS:
222 conn->c_npaths = min_t(int, RDS_MPATH_WORKERS,
223 be16_to_cpu(buffer.rds_npaths));
225 case RDS_EXTHDR_GEN_NUM:
226 new_peer_gen_num = be32_to_cpu(buffer.rds_gen_num);
229 pr_warn_ratelimited("ignoring unknown exthdr type "
233 /* if RDS_EXTHDR_NPATHS was not found, default to a single-path */
234 conn->c_npaths = max_t(int, conn->c_npaths, 1);
235 conn->c_ping_triggered = 0;
236 rds_conn_peer_gen_update(conn, new_peer_gen_num);
239 /* rds_start_mprds() will synchronously start multiple paths when appropriate.
240 * The scheme is based on the following rules:
242 * 1. rds_sendmsg on first connect attempt sends the probe ping, with the
243 * sender's npaths (s_npaths)
244 * 2. rcvr of probe-ping knows the mprds_paths = min(s_npaths, r_npaths). It
245 * sends back a probe-pong with r_npaths. After that, if rcvr is the
246 * smaller ip addr, it starts rds_conn_path_connect_if_down on all
248 * 3. sender gets woken up, and can move to rds_conn_path_connect_if_down.
249 * If it is the smaller ipaddr, rds_conn_path_connect_if_down can be
250 * called after reception of the probe-pong on all mprds_paths.
251 * Otherwise (sender of probe-ping is not the smaller ip addr): just call
252 * rds_conn_path_connect_if_down on the hashed path. (see rule 4)
253 * 4. rds_connect_worker must only trigger a connection if laddr < faddr.
254 * 5. sender may end up queuing the packet on the cp. will get sent out later.
255 * when connection is completed.
257 static void rds_start_mprds(struct rds_connection *conn)
260 struct rds_conn_path *cp;
262 if (conn->c_npaths > 1 &&
263 rds_addr_cmp(&conn->c_laddr, &conn->c_faddr) < 0) {
264 for (i = 0; i < conn->c_npaths; i++) {
265 cp = &conn->c_path[i];
266 rds_conn_path_connect_if_down(cp);
272 * The transport must make sure that this is serialized against other
273 * rx and conn reset on this specific conn.
275 * We currently assert that only one fragmented message will be sent
276 * down a connection at a time. This lets us reassemble in the conn
277 * instead of per-flow which means that we don't have to go digging through
278 * flows to tear down partial reassembly progress on conn failure and
279 * we save flow lookup and locking for each frag arrival. It does mean
280 * that small messages will wait behind large ones. Fragmenting at all
281 * is only to reduce the memory consumption of pre-posted buffers.
283 * The caller passes in saddr and daddr instead of us getting it from the
284 * conn. This lets loopback, who only has one conn for both directions,
285 * tell us which roles the addrs in the conn are playing for this message.
287 void rds_recv_incoming(struct rds_connection *conn, struct in6_addr *saddr,
288 struct in6_addr *daddr,
289 struct rds_incoming *inc, gfp_t gfp)
291 struct rds_sock *rs = NULL;
294 struct rds_conn_path *cp;
297 inc->i_rx_jiffies = jiffies;
298 if (conn->c_trans->t_mp_capable)
299 cp = inc->i_conn_path;
301 cp = &conn->c_path[0];
303 rdsdebug("conn %p next %llu inc %p seq %llu len %u sport %u dport %u "
304 "flags 0x%x rx_jiffies %lu\n", conn,
305 (unsigned long long)cp->cp_next_rx_seq,
307 (unsigned long long)be64_to_cpu(inc->i_hdr.h_sequence),
308 be32_to_cpu(inc->i_hdr.h_len),
309 be16_to_cpu(inc->i_hdr.h_sport),
310 be16_to_cpu(inc->i_hdr.h_dport),
315 * Sequence numbers should only increase. Messages get their
316 * sequence number as they're queued in a sending conn. They
317 * can be dropped, though, if the sending socket is closed before
318 * they hit the wire. So sequence numbers can skip forward
319 * under normal operation. They can also drop back in the conn
320 * failover case as previously sent messages are resent down the
321 * new instance of a conn. We drop those, otherwise we have
322 * to assume that the next valid seq does not come after a
323 * hole in the fragment stream.
325 * The headers don't give us a way to realize if fragments of
326 * a message have been dropped. We assume that frags that arrive
327 * to a flow are part of the current message on the flow that is
328 * being reassembled. This means that senders can't drop messages
329 * from the sending conn until all their frags are sent.
331 * XXX we could spend more on the wire to get more robust failure
332 * detection, arguably worth it to avoid data corruption.
334 if (be64_to_cpu(inc->i_hdr.h_sequence) < cp->cp_next_rx_seq &&
335 (inc->i_hdr.h_flags & RDS_FLAG_RETRANSMITTED)) {
336 rds_stats_inc(s_recv_drop_old_seq);
339 cp->cp_next_rx_seq = be64_to_cpu(inc->i_hdr.h_sequence) + 1;
341 if (rds_sysctl_ping_enable && inc->i_hdr.h_dport == 0) {
342 if (inc->i_hdr.h_sport == 0) {
343 rdsdebug("ignore ping with 0 sport from %pI6c\n",
347 rds_stats_inc(s_recv_ping);
348 rds_send_pong(cp, inc->i_hdr.h_sport);
349 /* if this is a handshake ping, start multipath if necessary */
350 if (RDS_HS_PROBE(be16_to_cpu(inc->i_hdr.h_sport),
351 be16_to_cpu(inc->i_hdr.h_dport))) {
352 rds_recv_hs_exthdrs(&inc->i_hdr, cp->cp_conn);
353 rds_start_mprds(cp->cp_conn);
358 if (be16_to_cpu(inc->i_hdr.h_dport) == RDS_FLAG_PROBE_PORT &&
359 inc->i_hdr.h_sport == 0) {
360 rds_recv_hs_exthdrs(&inc->i_hdr, cp->cp_conn);
361 /* if this is a handshake pong, start multipath if necessary */
362 rds_start_mprds(cp->cp_conn);
363 wake_up(&cp->cp_conn->c_hs_waitq);
367 rs = rds_find_bound(daddr, inc->i_hdr.h_dport, conn->c_bound_if);
369 rds_stats_inc(s_recv_drop_no_sock);
373 /* Process extension headers */
374 rds_recv_incoming_exthdrs(inc, rs);
376 /* We can be racing with rds_release() which marks the socket dead. */
377 sk = rds_rs_to_sk(rs);
379 /* serialize with rds_release -> sock_orphan */
380 write_lock_irqsave(&rs->rs_recv_lock, flags);
381 if (!sock_flag(sk, SOCK_DEAD)) {
382 rdsdebug("adding inc %p to rs %p's recv queue\n", inc, rs);
383 rds_stats_inc(s_recv_queued);
384 rds_recv_rcvbuf_delta(rs, sk, inc->i_conn->c_lcong,
385 be32_to_cpu(inc->i_hdr.h_len),
387 if (sock_flag(sk, SOCK_RCVTSTAMP))
388 do_gettimeofday(&inc->i_rx_tstamp);
390 inc->i_rx_lat_trace[RDS_MSG_RX_END] = local_clock();
391 list_add_tail(&inc->i_item, &rs->rs_recv_queue);
392 __rds_wake_sk_sleep(sk);
394 rds_stats_inc(s_recv_drop_dead_sock);
396 write_unlock_irqrestore(&rs->rs_recv_lock, flags);
402 EXPORT_SYMBOL_GPL(rds_recv_incoming);
405 * be very careful here. This is being called as the condition in
406 * wait_event_*() needs to cope with being called many times.
408 static int rds_next_incoming(struct rds_sock *rs, struct rds_incoming **inc)
413 read_lock_irqsave(&rs->rs_recv_lock, flags);
414 if (!list_empty(&rs->rs_recv_queue)) {
415 *inc = list_entry(rs->rs_recv_queue.next,
418 rds_inc_addref(*inc);
420 read_unlock_irqrestore(&rs->rs_recv_lock, flags);
426 static int rds_still_queued(struct rds_sock *rs, struct rds_incoming *inc,
429 struct sock *sk = rds_rs_to_sk(rs);
433 write_lock_irqsave(&rs->rs_recv_lock, flags);
434 if (!list_empty(&inc->i_item)) {
437 /* XXX make sure this i_conn is reliable */
438 rds_recv_rcvbuf_delta(rs, sk, inc->i_conn->c_lcong,
439 -be32_to_cpu(inc->i_hdr.h_len),
441 list_del_init(&inc->i_item);
445 write_unlock_irqrestore(&rs->rs_recv_lock, flags);
447 rdsdebug("inc %p rs %p still %d dropped %d\n", inc, rs, ret, drop);
452 * Pull errors off the error queue.
453 * If msghdr is NULL, we will just purge the error queue.
455 int rds_notify_queue_get(struct rds_sock *rs, struct msghdr *msghdr)
457 struct rds_notifier *notifier;
458 struct rds_rdma_notify cmsg;
459 unsigned int count = 0, max_messages = ~0U;
464 memset(&cmsg, 0, sizeof(cmsg)); /* fill holes with zero */
466 /* put_cmsg copies to user space and thus may sleep. We can't do this
467 * with rs_lock held, so first grab as many notifications as we can stuff
468 * in the user provided cmsg buffer. We don't try to copy more, to avoid
469 * losing notifications - except when the buffer is so small that it wouldn't
470 * even hold a single notification. Then we give him as much of this single
471 * msg as we can squeeze in, and set MSG_CTRUNC.
474 max_messages = msghdr->msg_controllen / CMSG_SPACE(sizeof(cmsg));
479 spin_lock_irqsave(&rs->rs_lock, flags);
480 while (!list_empty(&rs->rs_notify_queue) && count < max_messages) {
481 notifier = list_entry(rs->rs_notify_queue.next,
482 struct rds_notifier, n_list);
483 list_move(¬ifier->n_list, ©);
486 spin_unlock_irqrestore(&rs->rs_lock, flags);
491 while (!list_empty(©)) {
492 notifier = list_entry(copy.next, struct rds_notifier, n_list);
495 cmsg.user_token = notifier->n_user_token;
496 cmsg.status = notifier->n_status;
498 err = put_cmsg(msghdr, SOL_RDS, RDS_CMSG_RDMA_STATUS,
499 sizeof(cmsg), &cmsg);
504 list_del_init(¬ifier->n_list);
508 /* If we bailed out because of an error in put_cmsg,
509 * we may be left with one or more notifications that we
510 * didn't process. Return them to the head of the list. */
511 if (!list_empty(©)) {
512 spin_lock_irqsave(&rs->rs_lock, flags);
513 list_splice(©, &rs->rs_notify_queue);
514 spin_unlock_irqrestore(&rs->rs_lock, flags);
521 * Queue a congestion notification
523 static int rds_notify_cong(struct rds_sock *rs, struct msghdr *msghdr)
525 uint64_t notify = rs->rs_cong_notify;
529 err = put_cmsg(msghdr, SOL_RDS, RDS_CMSG_CONG_UPDATE,
530 sizeof(notify), ¬ify);
534 spin_lock_irqsave(&rs->rs_lock, flags);
535 rs->rs_cong_notify &= ~notify;
536 spin_unlock_irqrestore(&rs->rs_lock, flags);
542 * Receive any control messages.
544 static int rds_cmsg_recv(struct rds_incoming *inc, struct msghdr *msg,
549 if (inc->i_rdma_cookie) {
550 ret = put_cmsg(msg, SOL_RDS, RDS_CMSG_RDMA_DEST,
551 sizeof(inc->i_rdma_cookie), &inc->i_rdma_cookie);
556 if ((inc->i_rx_tstamp.tv_sec != 0) &&
557 sock_flag(rds_rs_to_sk(rs), SOCK_RCVTSTAMP)) {
558 ret = put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMP,
559 sizeof(struct timeval),
565 if (rs->rs_rx_traces) {
566 struct rds_cmsg_rx_trace t;
569 memset(&t, 0, sizeof(t));
570 inc->i_rx_lat_trace[RDS_MSG_RX_CMSG] = local_clock();
571 t.rx_traces = rs->rs_rx_traces;
572 for (i = 0; i < rs->rs_rx_traces; i++) {
573 j = rs->rs_rx_trace[i];
574 t.rx_trace_pos[i] = j;
575 t.rx_trace[i] = inc->i_rx_lat_trace[j + 1] -
576 inc->i_rx_lat_trace[j];
579 ret = put_cmsg(msg, SOL_RDS, RDS_CMSG_RXPATH_LATENCY,
589 static bool rds_recvmsg_zcookie(struct rds_sock *rs, struct msghdr *msg)
591 struct rds_msg_zcopy_queue *q = &rs->rs_zcookie_queue;
592 struct rds_msg_zcopy_info *info = NULL;
593 struct rds_zcopy_cookies *done;
596 if (!msg->msg_control)
599 if (!sock_flag(rds_rs_to_sk(rs), SOCK_ZEROCOPY) ||
600 msg->msg_controllen < CMSG_SPACE(sizeof(*done)))
603 spin_lock_irqsave(&q->lock, flags);
604 if (!list_empty(&q->zcookie_head)) {
605 info = list_entry(q->zcookie_head.next,
606 struct rds_msg_zcopy_info, rs_zcookie_next);
607 list_del(&info->rs_zcookie_next);
609 spin_unlock_irqrestore(&q->lock, flags);
612 done = &info->zcookies;
613 if (put_cmsg(msg, SOL_RDS, RDS_CMSG_ZCOPY_COMPLETION, sizeof(*done),
615 spin_lock_irqsave(&q->lock, flags);
616 list_add(&info->rs_zcookie_next, &q->zcookie_head);
617 spin_unlock_irqrestore(&q->lock, flags);
624 int rds_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
627 struct sock *sk = sock->sk;
628 struct rds_sock *rs = rds_sk_to_rs(sk);
630 int ret = 0, nonblock = msg_flags & MSG_DONTWAIT;
631 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
632 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
633 struct rds_incoming *inc = NULL;
635 /* udp_recvmsg()->sock_recvtimeo() gets away without locking too.. */
636 timeo = sock_rcvtimeo(sk, nonblock);
638 rdsdebug("size %zu flags 0x%x timeo %ld\n", size, msg_flags, timeo);
640 if (msg_flags & MSG_OOB)
642 if (msg_flags & MSG_ERRQUEUE)
643 return sock_recv_errqueue(sk, msg, size, SOL_IP, IP_RECVERR);
646 /* If there are pending notifications, do those - and nothing else */
647 if (!list_empty(&rs->rs_notify_queue)) {
648 ret = rds_notify_queue_get(rs, msg);
652 if (rs->rs_cong_notify) {
653 ret = rds_notify_cong(rs, msg);
657 if (!rds_next_incoming(rs, &inc)) {
659 bool reaped = rds_recvmsg_zcookie(rs, msg);
661 ret = reaped ? 0 : -EAGAIN;
665 timeo = wait_event_interruptible_timeout(*sk_sleep(sk),
666 (!list_empty(&rs->rs_notify_queue) ||
667 rs->rs_cong_notify ||
668 rds_next_incoming(rs, &inc)), timeo);
669 rdsdebug("recvmsg woke inc %p timeo %ld\n", inc,
671 if (timeo > 0 || timeo == MAX_SCHEDULE_TIMEOUT)
680 rdsdebug("copying inc %p from %pI6c:%u to user\n", inc,
681 &inc->i_conn->c_faddr,
682 ntohs(inc->i_hdr.h_sport));
683 ret = inc->i_conn->c_trans->inc_copy_to_user(inc, &msg->msg_iter);
688 * if the message we just copied isn't at the head of the
689 * recv queue then someone else raced us to return it, try
690 * to get the next message.
692 if (!rds_still_queued(rs, inc, !(msg_flags & MSG_PEEK))) {
695 rds_stats_inc(s_recv_deliver_raced);
696 iov_iter_revert(&msg->msg_iter, ret);
700 if (ret < be32_to_cpu(inc->i_hdr.h_len)) {
701 if (msg_flags & MSG_TRUNC)
702 ret = be32_to_cpu(inc->i_hdr.h_len);
703 msg->msg_flags |= MSG_TRUNC;
706 if (rds_cmsg_recv(inc, msg, rs)) {
710 rds_recvmsg_zcookie(rs, msg);
712 rds_stats_inc(s_recv_delivered);
715 if (ipv6_addr_v4mapped(&inc->i_saddr)) {
716 sin = (struct sockaddr_in *)msg->msg_name;
718 sin->sin_family = AF_INET;
719 sin->sin_port = inc->i_hdr.h_sport;
720 sin->sin_addr.s_addr =
721 inc->i_saddr.s6_addr32[3];
722 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
723 msg->msg_namelen = sizeof(*sin);
725 sin6 = (struct sockaddr_in6 *)msg->msg_name;
727 sin6->sin6_family = AF_INET6;
728 sin6->sin6_port = inc->i_hdr.h_sport;
729 sin6->sin6_addr = inc->i_saddr;
730 sin6->sin6_flowinfo = 0;
731 sin6->sin6_scope_id = rs->rs_bound_scope_id;
732 msg->msg_namelen = sizeof(*sin6);
746 * The socket is being shut down and we're asked to drop messages that were
747 * queued for recvmsg. The caller has unbound the socket so the receive path
748 * won't queue any more incoming fragments or messages on the socket.
750 void rds_clear_recv_queue(struct rds_sock *rs)
752 struct sock *sk = rds_rs_to_sk(rs);
753 struct rds_incoming *inc, *tmp;
756 write_lock_irqsave(&rs->rs_recv_lock, flags);
757 list_for_each_entry_safe(inc, tmp, &rs->rs_recv_queue, i_item) {
758 rds_recv_rcvbuf_delta(rs, sk, inc->i_conn->c_lcong,
759 -be32_to_cpu(inc->i_hdr.h_len),
761 list_del_init(&inc->i_item);
764 write_unlock_irqrestore(&rs->rs_recv_lock, flags);
768 * inc->i_saddr isn't used here because it is only set in the receive
771 void rds_inc_info_copy(struct rds_incoming *inc,
772 struct rds_info_iterator *iter,
773 __be32 saddr, __be32 daddr, int flip)
775 struct rds_info_message minfo;
777 minfo.seq = be64_to_cpu(inc->i_hdr.h_sequence);
778 minfo.len = be32_to_cpu(inc->i_hdr.h_len);
783 minfo.lport = inc->i_hdr.h_dport;
784 minfo.fport = inc->i_hdr.h_sport;
788 minfo.lport = inc->i_hdr.h_sport;
789 minfo.fport = inc->i_hdr.h_dport;
794 rds_info_copy(iter, &minfo, sizeof(minfo));
797 #if IS_ENABLED(CONFIG_IPV6)
798 void rds6_inc_info_copy(struct rds_incoming *inc,
799 struct rds_info_iterator *iter,
800 struct in6_addr *saddr, struct in6_addr *daddr,
803 struct rds6_info_message minfo6;
805 minfo6.seq = be64_to_cpu(inc->i_hdr.h_sequence);
806 minfo6.len = be32_to_cpu(inc->i_hdr.h_len);
810 minfo6.laddr = *daddr;
811 minfo6.faddr = *saddr;
812 minfo6.lport = inc->i_hdr.h_dport;
813 minfo6.fport = inc->i_hdr.h_sport;
815 minfo6.laddr = *saddr;
816 minfo6.faddr = *daddr;
817 minfo6.lport = inc->i_hdr.h_sport;
818 minfo6.fport = inc->i_hdr.h_dport;
823 rds_info_copy(iter, &minfo6, sizeof(minfo6));