2 * Copyright (c) 2006 Oracle. 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>
41 void rds_inc_init(struct rds_incoming *inc, struct rds_connection *conn,
44 atomic_set(&inc->i_refcount, 1);
45 INIT_LIST_HEAD(&inc->i_item);
48 inc->i_rdma_cookie = 0;
50 EXPORT_SYMBOL_GPL(rds_inc_init);
52 static void rds_inc_addref(struct rds_incoming *inc)
54 rdsdebug("addref inc %p ref %d\n", inc, atomic_read(&inc->i_refcount));
55 atomic_inc(&inc->i_refcount);
58 void rds_inc_put(struct rds_incoming *inc)
60 rdsdebug("put inc %p ref %d\n", inc, atomic_read(&inc->i_refcount));
61 if (atomic_dec_and_test(&inc->i_refcount)) {
62 BUG_ON(!list_empty(&inc->i_item));
64 inc->i_conn->c_trans->inc_free(inc);
67 EXPORT_SYMBOL_GPL(rds_inc_put);
69 static void rds_recv_rcvbuf_delta(struct rds_sock *rs, struct sock *sk,
70 struct rds_cong_map *map,
71 int delta, __be16 port)
78 rs->rs_rcv_bytes += delta;
79 now_congested = rs->rs_rcv_bytes > rds_sk_rcvbuf(rs);
81 rdsdebug("rs %p (%pI4:%u) recv bytes %d buf %d "
82 "now_cong %d delta %d\n",
83 rs, &rs->rs_bound_addr,
84 ntohs(rs->rs_bound_port), rs->rs_rcv_bytes,
85 rds_sk_rcvbuf(rs), now_congested, delta);
87 /* wasn't -> am congested */
88 if (!rs->rs_congested && now_congested) {
90 rds_cong_set_bit(map, port);
91 rds_cong_queue_updates(map);
93 /* was -> aren't congested */
94 /* Require more free space before reporting uncongested to prevent
95 bouncing cong/uncong state too often */
96 else if (rs->rs_congested && (rs->rs_rcv_bytes < (rds_sk_rcvbuf(rs)/2))) {
98 rds_cong_clear_bit(map, port);
99 rds_cong_queue_updates(map);
102 /* do nothing if no change in cong state */
106 * Process all extension headers that come with this message.
108 static void rds_recv_incoming_exthdrs(struct rds_incoming *inc, struct rds_sock *rs)
110 struct rds_header *hdr = &inc->i_hdr;
111 unsigned int pos = 0, type, len;
113 struct rds_ext_header_version version;
114 struct rds_ext_header_rdma rdma;
115 struct rds_ext_header_rdma_dest rdma_dest;
119 len = sizeof(buffer);
120 type = rds_message_next_extension(hdr, &pos, &buffer, &len);
121 if (type == RDS_EXTHDR_NONE)
123 /* Process extension header here */
125 case RDS_EXTHDR_RDMA:
126 rds_rdma_unuse(rs, be32_to_cpu(buffer.rdma.h_rdma_rkey), 0);
129 case RDS_EXTHDR_RDMA_DEST:
130 /* We ignore the size for now. We could stash it
131 * somewhere and use it for error checking. */
132 inc->i_rdma_cookie = rds_rdma_make_cookie(
133 be32_to_cpu(buffer.rdma_dest.h_rdma_rkey),
134 be32_to_cpu(buffer.rdma_dest.h_rdma_offset));
142 * The transport must make sure that this is serialized against other
143 * rx and conn reset on this specific conn.
145 * We currently assert that only one fragmented message will be sent
146 * down a connection at a time. This lets us reassemble in the conn
147 * instead of per-flow which means that we don't have to go digging through
148 * flows to tear down partial reassembly progress on conn failure and
149 * we save flow lookup and locking for each frag arrival. It does mean
150 * that small messages will wait behind large ones. Fragmenting at all
151 * is only to reduce the memory consumption of pre-posted buffers.
153 * The caller passes in saddr and daddr instead of us getting it from the
154 * conn. This lets loopback, who only has one conn for both directions,
155 * tell us which roles the addrs in the conn are playing for this message.
157 void rds_recv_incoming(struct rds_connection *conn, __be32 saddr, __be32 daddr,
158 struct rds_incoming *inc, gfp_t gfp)
160 struct rds_sock *rs = NULL;
165 inc->i_rx_jiffies = jiffies;
167 rdsdebug("conn %p next %llu inc %p seq %llu len %u sport %u dport %u "
168 "flags 0x%x rx_jiffies %lu\n", conn,
169 (unsigned long long)conn->c_next_rx_seq,
171 (unsigned long long)be64_to_cpu(inc->i_hdr.h_sequence),
172 be32_to_cpu(inc->i_hdr.h_len),
173 be16_to_cpu(inc->i_hdr.h_sport),
174 be16_to_cpu(inc->i_hdr.h_dport),
179 * Sequence numbers should only increase. Messages get their
180 * sequence number as they're queued in a sending conn. They
181 * can be dropped, though, if the sending socket is closed before
182 * they hit the wire. So sequence numbers can skip forward
183 * under normal operation. They can also drop back in the conn
184 * failover case as previously sent messages are resent down the
185 * new instance of a conn. We drop those, otherwise we have
186 * to assume that the next valid seq does not come after a
187 * hole in the fragment stream.
189 * The headers don't give us a way to realize if fragments of
190 * a message have been dropped. We assume that frags that arrive
191 * to a flow are part of the current message on the flow that is
192 * being reassembled. This means that senders can't drop messages
193 * from the sending conn until all their frags are sent.
195 * XXX we could spend more on the wire to get more robust failure
196 * detection, arguably worth it to avoid data corruption.
198 if (be64_to_cpu(inc->i_hdr.h_sequence) < conn->c_next_rx_seq &&
199 (inc->i_hdr.h_flags & RDS_FLAG_RETRANSMITTED)) {
200 rds_stats_inc(s_recv_drop_old_seq);
203 conn->c_next_rx_seq = be64_to_cpu(inc->i_hdr.h_sequence) + 1;
205 if (rds_sysctl_ping_enable && inc->i_hdr.h_dport == 0) {
206 rds_stats_inc(s_recv_ping);
207 rds_send_pong(conn, inc->i_hdr.h_sport);
211 rs = rds_find_bound(daddr, inc->i_hdr.h_dport);
213 rds_stats_inc(s_recv_drop_no_sock);
217 /* Process extension headers */
218 rds_recv_incoming_exthdrs(inc, rs);
220 /* We can be racing with rds_release() which marks the socket dead. */
221 sk = rds_rs_to_sk(rs);
223 /* serialize with rds_release -> sock_orphan */
224 write_lock_irqsave(&rs->rs_recv_lock, flags);
225 if (!sock_flag(sk, SOCK_DEAD)) {
226 rdsdebug("adding inc %p to rs %p's recv queue\n", inc, rs);
227 rds_stats_inc(s_recv_queued);
228 rds_recv_rcvbuf_delta(rs, sk, inc->i_conn->c_lcong,
229 be32_to_cpu(inc->i_hdr.h_len),
232 list_add_tail(&inc->i_item, &rs->rs_recv_queue);
233 __rds_wake_sk_sleep(sk);
235 rds_stats_inc(s_recv_drop_dead_sock);
237 write_unlock_irqrestore(&rs->rs_recv_lock, flags);
243 EXPORT_SYMBOL_GPL(rds_recv_incoming);
246 * be very careful here. This is being called as the condition in
247 * wait_event_*() needs to cope with being called many times.
249 static int rds_next_incoming(struct rds_sock *rs, struct rds_incoming **inc)
254 read_lock_irqsave(&rs->rs_recv_lock, flags);
255 if (!list_empty(&rs->rs_recv_queue)) {
256 *inc = list_entry(rs->rs_recv_queue.next,
259 rds_inc_addref(*inc);
261 read_unlock_irqrestore(&rs->rs_recv_lock, flags);
267 static int rds_still_queued(struct rds_sock *rs, struct rds_incoming *inc,
270 struct sock *sk = rds_rs_to_sk(rs);
274 write_lock_irqsave(&rs->rs_recv_lock, flags);
275 if (!list_empty(&inc->i_item)) {
278 /* XXX make sure this i_conn is reliable */
279 rds_recv_rcvbuf_delta(rs, sk, inc->i_conn->c_lcong,
280 -be32_to_cpu(inc->i_hdr.h_len),
282 list_del_init(&inc->i_item);
286 write_unlock_irqrestore(&rs->rs_recv_lock, flags);
288 rdsdebug("inc %p rs %p still %d dropped %d\n", inc, rs, ret, drop);
293 * Pull errors off the error queue.
294 * If msghdr is NULL, we will just purge the error queue.
296 int rds_notify_queue_get(struct rds_sock *rs, struct msghdr *msghdr)
298 struct rds_notifier *notifier;
299 struct rds_rdma_notify cmsg = { 0 }; /* fill holes with zero */
300 unsigned int count = 0, max_messages = ~0U;
306 /* put_cmsg copies to user space and thus may sleep. We can't do this
307 * with rs_lock held, so first grab as many notifications as we can stuff
308 * in the user provided cmsg buffer. We don't try to copy more, to avoid
309 * losing notifications - except when the buffer is so small that it wouldn't
310 * even hold a single notification. Then we give him as much of this single
311 * msg as we can squeeze in, and set MSG_CTRUNC.
314 max_messages = msghdr->msg_controllen / CMSG_SPACE(sizeof(cmsg));
319 spin_lock_irqsave(&rs->rs_lock, flags);
320 while (!list_empty(&rs->rs_notify_queue) && count < max_messages) {
321 notifier = list_entry(rs->rs_notify_queue.next,
322 struct rds_notifier, n_list);
323 list_move(¬ifier->n_list, ©);
326 spin_unlock_irqrestore(&rs->rs_lock, flags);
331 while (!list_empty(©)) {
332 notifier = list_entry(copy.next, struct rds_notifier, n_list);
335 cmsg.user_token = notifier->n_user_token;
336 cmsg.status = notifier->n_status;
338 err = put_cmsg(msghdr, SOL_RDS, RDS_CMSG_RDMA_STATUS,
339 sizeof(cmsg), &cmsg);
344 list_del_init(¬ifier->n_list);
348 /* If we bailed out because of an error in put_cmsg,
349 * we may be left with one or more notifications that we
350 * didn't process. Return them to the head of the list. */
351 if (!list_empty(©)) {
352 spin_lock_irqsave(&rs->rs_lock, flags);
353 list_splice(©, &rs->rs_notify_queue);
354 spin_unlock_irqrestore(&rs->rs_lock, flags);
361 * Queue a congestion notification
363 static int rds_notify_cong(struct rds_sock *rs, struct msghdr *msghdr)
365 uint64_t notify = rs->rs_cong_notify;
369 err = put_cmsg(msghdr, SOL_RDS, RDS_CMSG_CONG_UPDATE,
370 sizeof(notify), ¬ify);
374 spin_lock_irqsave(&rs->rs_lock, flags);
375 rs->rs_cong_notify &= ~notify;
376 spin_unlock_irqrestore(&rs->rs_lock, flags);
382 * Receive any control messages.
384 static int rds_cmsg_recv(struct rds_incoming *inc, struct msghdr *msg)
388 if (inc->i_rdma_cookie) {
389 ret = put_cmsg(msg, SOL_RDS, RDS_CMSG_RDMA_DEST,
390 sizeof(inc->i_rdma_cookie), &inc->i_rdma_cookie);
398 int rds_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
401 struct sock *sk = sock->sk;
402 struct rds_sock *rs = rds_sk_to_rs(sk);
404 int ret = 0, nonblock = msg_flags & MSG_DONTWAIT;
405 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
406 struct rds_incoming *inc = NULL;
408 /* udp_recvmsg()->sock_recvtimeo() gets away without locking too.. */
409 timeo = sock_rcvtimeo(sk, nonblock);
411 rdsdebug("size %zu flags 0x%x timeo %ld\n", size, msg_flags, timeo);
413 if (msg_flags & MSG_OOB)
417 struct iov_iter save;
418 /* If there are pending notifications, do those - and nothing else */
419 if (!list_empty(&rs->rs_notify_queue)) {
420 ret = rds_notify_queue_get(rs, msg);
424 if (rs->rs_cong_notify) {
425 ret = rds_notify_cong(rs, msg);
429 if (!rds_next_incoming(rs, &inc)) {
435 timeo = wait_event_interruptible_timeout(*sk_sleep(sk),
436 (!list_empty(&rs->rs_notify_queue) ||
437 rs->rs_cong_notify ||
438 rds_next_incoming(rs, &inc)), timeo);
439 rdsdebug("recvmsg woke inc %p timeo %ld\n", inc,
441 if (timeo > 0 || timeo == MAX_SCHEDULE_TIMEOUT)
450 rdsdebug("copying inc %p from %pI4:%u to user\n", inc,
451 &inc->i_conn->c_faddr,
452 ntohs(inc->i_hdr.h_sport));
453 save = msg->msg_iter;
454 ret = inc->i_conn->c_trans->inc_copy_to_user(inc, &msg->msg_iter);
459 * if the message we just copied isn't at the head of the
460 * recv queue then someone else raced us to return it, try
461 * to get the next message.
463 if (!rds_still_queued(rs, inc, !(msg_flags & MSG_PEEK))) {
466 rds_stats_inc(s_recv_deliver_raced);
467 msg->msg_iter = save;
471 if (ret < be32_to_cpu(inc->i_hdr.h_len)) {
472 if (msg_flags & MSG_TRUNC)
473 ret = be32_to_cpu(inc->i_hdr.h_len);
474 msg->msg_flags |= MSG_TRUNC;
477 if (rds_cmsg_recv(inc, msg)) {
482 rds_stats_inc(s_recv_delivered);
485 sin->sin_family = AF_INET;
486 sin->sin_port = inc->i_hdr.h_sport;
487 sin->sin_addr.s_addr = inc->i_saddr;
488 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
489 msg->msg_namelen = sizeof(*sin);
502 * The socket is being shut down and we're asked to drop messages that were
503 * queued for recvmsg. The caller has unbound the socket so the receive path
504 * won't queue any more incoming fragments or messages on the socket.
506 void rds_clear_recv_queue(struct rds_sock *rs)
508 struct sock *sk = rds_rs_to_sk(rs);
509 struct rds_incoming *inc, *tmp;
512 write_lock_irqsave(&rs->rs_recv_lock, flags);
513 list_for_each_entry_safe(inc, tmp, &rs->rs_recv_queue, i_item) {
514 rds_recv_rcvbuf_delta(rs, sk, inc->i_conn->c_lcong,
515 -be32_to_cpu(inc->i_hdr.h_len),
517 list_del_init(&inc->i_item);
520 write_unlock_irqrestore(&rs->rs_recv_lock, flags);
524 * inc->i_saddr isn't used here because it is only set in the receive
527 void rds_inc_info_copy(struct rds_incoming *inc,
528 struct rds_info_iterator *iter,
529 __be32 saddr, __be32 daddr, int flip)
531 struct rds_info_message minfo;
533 minfo.seq = be64_to_cpu(inc->i_hdr.h_sequence);
534 minfo.len = be32_to_cpu(inc->i_hdr.h_len);
539 minfo.lport = inc->i_hdr.h_dport;
540 minfo.fport = inc->i_hdr.h_sport;
544 minfo.lport = inc->i_hdr.h_sport;
545 minfo.fport = inc->i_hdr.h_dport;
548 rds_info_copy(iter, &minfo, sizeof(minfo));