Lines Matching +full:rs +full:-
14 * - Redistributions of source code must retain the above
18 * - Redistributions in binary form must reproduce the above
46 refcount_set(&inc->i_refcount, 1); in rds_inc_init()
47 INIT_LIST_HEAD(&inc->i_item); in rds_inc_init()
48 inc->i_conn = conn; in rds_inc_init()
49 inc->i_saddr = *saddr; in rds_inc_init()
50 inc->i_usercopy.rdma_cookie = 0; in rds_inc_init()
51 inc->i_usercopy.rx_tstamp = ktime_set(0, 0); in rds_inc_init()
53 memset(inc->i_rx_lat_trace, 0, sizeof(inc->i_rx_lat_trace)); in rds_inc_init()
60 refcount_set(&inc->i_refcount, 1); in rds_inc_path_init()
61 INIT_LIST_HEAD(&inc->i_item); in rds_inc_path_init()
62 inc->i_conn = cp->cp_conn; in rds_inc_path_init()
63 inc->i_conn_path = cp; in rds_inc_path_init()
64 inc->i_saddr = *saddr; in rds_inc_path_init()
65 inc->i_usercopy.rdma_cookie = 0; in rds_inc_path_init()
66 inc->i_usercopy.rx_tstamp = ktime_set(0, 0); in rds_inc_path_init()
72 rdsdebug("addref inc %p ref %d\n", inc, refcount_read(&inc->i_refcount)); in rds_inc_addref()
73 refcount_inc(&inc->i_refcount); in rds_inc_addref()
78 rdsdebug("put inc %p ref %d\n", inc, refcount_read(&inc->i_refcount)); in rds_inc_put()
79 if (refcount_dec_and_test(&inc->i_refcount)) { in rds_inc_put()
80 BUG_ON(!list_empty(&inc->i_item)); in rds_inc_put()
82 inc->i_conn->c_trans->inc_free(inc); in rds_inc_put()
87 static void rds_recv_rcvbuf_delta(struct rds_sock *rs, struct sock *sk, in rds_recv_rcvbuf_delta() argument
96 rs->rs_rcv_bytes += delta; in rds_recv_rcvbuf_delta()
100 rds_stats_add(s_recv_bytes_removed_from_socket, -delta); in rds_recv_rcvbuf_delta()
103 if (rs->rs_transport->t_type == RDS_TRANS_LOOP) in rds_recv_rcvbuf_delta()
106 now_congested = rs->rs_rcv_bytes > rds_sk_rcvbuf(rs); in rds_recv_rcvbuf_delta()
108 rdsdebug("rs %p (%pI6c:%u) recv bytes %d buf %d " in rds_recv_rcvbuf_delta()
110 rs, &rs->rs_bound_addr, in rds_recv_rcvbuf_delta()
111 ntohs(rs->rs_bound_port), rs->rs_rcv_bytes, in rds_recv_rcvbuf_delta()
112 rds_sk_rcvbuf(rs), now_congested, delta); in rds_recv_rcvbuf_delta()
114 /* wasn't -> am congested */ in rds_recv_rcvbuf_delta()
115 if (!rs->rs_congested && now_congested) { in rds_recv_rcvbuf_delta()
116 rs->rs_congested = 1; in rds_recv_rcvbuf_delta()
120 /* was -> aren't congested */ in rds_recv_rcvbuf_delta()
123 else if (rs->rs_congested && (rs->rs_rcv_bytes < (rds_sk_rcvbuf(rs)/2))) { in rds_recv_rcvbuf_delta()
124 rs->rs_congested = 0; in rds_recv_rcvbuf_delta()
139 WARN_ON(conn->c_trans->t_type != RDS_TRANS_TCP); in rds_conn_peer_gen_update()
141 if (conn->c_peer_gen_num != 0 && in rds_conn_peer_gen_update()
142 peer_gen_num != conn->c_peer_gen_num) { in rds_conn_peer_gen_update()
146 cp = &conn->c_path[i]; in rds_conn_peer_gen_update()
147 spin_lock_irqsave(&cp->cp_lock, flags); in rds_conn_peer_gen_update()
148 cp->cp_next_tx_seq = 1; in rds_conn_peer_gen_update()
149 cp->cp_next_rx_seq = 0; in rds_conn_peer_gen_update()
151 &cp->cp_retrans, in rds_conn_peer_gen_update()
153 set_bit(RDS_MSG_FLUSH, &rm->m_flags); in rds_conn_peer_gen_update()
155 spin_unlock_irqrestore(&cp->cp_lock, flags); in rds_conn_peer_gen_update()
158 conn->c_peer_gen_num = peer_gen_num; in rds_conn_peer_gen_update()
165 static void rds_recv_incoming_exthdrs(struct rds_incoming *inc, struct rds_sock *rs) in rds_recv_incoming_exthdrs() argument
167 struct rds_header *hdr = &inc->i_hdr; in rds_recv_incoming_exthdrs()
183 rds_rdma_unuse(rs, be32_to_cpu(buffer.rdma.h_rdma_rkey), 0); in rds_recv_incoming_exthdrs()
189 inc->i_usercopy.rdma_cookie = rds_rdma_make_cookie( in rds_recv_incoming_exthdrs()
217 conn->c_npaths = min_t(int, RDS_MPATH_WORKERS, in rds_recv_hs_exthdrs()
228 /* if RDS_EXTHDR_NPATHS was not found, default to a single-path */ in rds_recv_hs_exthdrs()
229 conn->c_npaths = max_t(int, conn->c_npaths, 1); in rds_recv_hs_exthdrs()
230 conn->c_ping_triggered = 0; in rds_recv_hs_exthdrs()
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
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
257 if (conn->c_npaths > 1 && in rds_start_mprds()
258 rds_addr_cmp(&conn->c_laddr, &conn->c_faddr) < 0) { in rds_start_mprds()
259 for (i = 0; i < conn->c_npaths; i++) { in rds_start_mprds()
260 cp = &conn->c_path[i]; in rds_start_mprds()
272 * instead of per-flow which means that we don't have to go digging through
276 * is only to reduce the memory consumption of pre-posted buffers.
286 struct rds_sock *rs = NULL; in rds_recv_incoming() local
291 inc->i_conn = conn; in rds_recv_incoming()
292 inc->i_rx_jiffies = jiffies; in rds_recv_incoming()
293 if (conn->c_trans->t_mp_capable) in rds_recv_incoming()
294 cp = inc->i_conn_path; in rds_recv_incoming()
296 cp = &conn->c_path[0]; in rds_recv_incoming()
300 (unsigned long long)cp->cp_next_rx_seq, in rds_recv_incoming()
302 (unsigned long long)be64_to_cpu(inc->i_hdr.h_sequence), in rds_recv_incoming()
303 be32_to_cpu(inc->i_hdr.h_len), in rds_recv_incoming()
304 be16_to_cpu(inc->i_hdr.h_sport), in rds_recv_incoming()
305 be16_to_cpu(inc->i_hdr.h_dport), in rds_recv_incoming()
306 inc->i_hdr.h_flags, in rds_recv_incoming()
307 inc->i_rx_jiffies); in rds_recv_incoming()
329 if (be64_to_cpu(inc->i_hdr.h_sequence) < cp->cp_next_rx_seq && in rds_recv_incoming()
330 (inc->i_hdr.h_flags & RDS_FLAG_RETRANSMITTED)) { in rds_recv_incoming()
334 cp->cp_next_rx_seq = be64_to_cpu(inc->i_hdr.h_sequence) + 1; in rds_recv_incoming()
336 if (rds_sysctl_ping_enable && inc->i_hdr.h_dport == 0) { in rds_recv_incoming()
337 if (inc->i_hdr.h_sport == 0) { in rds_recv_incoming()
343 rds_send_pong(cp, inc->i_hdr.h_sport); in rds_recv_incoming()
345 if (RDS_HS_PROBE(be16_to_cpu(inc->i_hdr.h_sport), in rds_recv_incoming()
346 be16_to_cpu(inc->i_hdr.h_dport))) { in rds_recv_incoming()
347 rds_recv_hs_exthdrs(&inc->i_hdr, cp->cp_conn); in rds_recv_incoming()
348 rds_start_mprds(cp->cp_conn); in rds_recv_incoming()
353 if (be16_to_cpu(inc->i_hdr.h_dport) == RDS_FLAG_PROBE_PORT && in rds_recv_incoming()
354 inc->i_hdr.h_sport == 0) { in rds_recv_incoming()
355 rds_recv_hs_exthdrs(&inc->i_hdr, cp->cp_conn); in rds_recv_incoming()
357 rds_start_mprds(cp->cp_conn); in rds_recv_incoming()
358 wake_up(&cp->cp_conn->c_hs_waitq); in rds_recv_incoming()
362 rs = rds_find_bound(daddr, inc->i_hdr.h_dport, conn->c_bound_if); in rds_recv_incoming()
363 if (!rs) { in rds_recv_incoming()
369 rds_recv_incoming_exthdrs(inc, rs); in rds_recv_incoming()
372 sk = rds_rs_to_sk(rs); in rds_recv_incoming()
374 /* serialize with rds_release -> sock_orphan */ in rds_recv_incoming()
375 write_lock_irqsave(&rs->rs_recv_lock, flags); in rds_recv_incoming()
377 rdsdebug("adding inc %p to rs %p's recv queue\n", inc, rs); in rds_recv_incoming()
379 rds_recv_rcvbuf_delta(rs, sk, inc->i_conn->c_lcong, in rds_recv_incoming()
380 be32_to_cpu(inc->i_hdr.h_len), in rds_recv_incoming()
381 inc->i_hdr.h_dport); in rds_recv_incoming()
383 inc->i_usercopy.rx_tstamp = ktime_get_real(); in rds_recv_incoming()
385 inc->i_rx_lat_trace[RDS_MSG_RX_END] = local_clock(); in rds_recv_incoming()
386 list_add_tail(&inc->i_item, &rs->rs_recv_queue); in rds_recv_incoming()
391 write_unlock_irqrestore(&rs->rs_recv_lock, flags); in rds_recv_incoming()
394 if (rs) in rds_recv_incoming()
395 rds_sock_put(rs); in rds_recv_incoming()
403 static int rds_next_incoming(struct rds_sock *rs, struct rds_incoming **inc) in rds_next_incoming() argument
408 read_lock_irqsave(&rs->rs_recv_lock, flags); in rds_next_incoming()
409 if (!list_empty(&rs->rs_recv_queue)) { in rds_next_incoming()
410 *inc = list_entry(rs->rs_recv_queue.next, in rds_next_incoming()
415 read_unlock_irqrestore(&rs->rs_recv_lock, flags); in rds_next_incoming()
421 static int rds_still_queued(struct rds_sock *rs, struct rds_incoming *inc, in rds_still_queued() argument
424 struct sock *sk = rds_rs_to_sk(rs); in rds_still_queued()
428 write_lock_irqsave(&rs->rs_recv_lock, flags); in rds_still_queued()
429 if (!list_empty(&inc->i_item)) { in rds_still_queued()
433 rds_recv_rcvbuf_delta(rs, sk, inc->i_conn->c_lcong, in rds_still_queued()
434 -be32_to_cpu(inc->i_hdr.h_len), in rds_still_queued()
435 inc->i_hdr.h_dport); in rds_still_queued()
436 list_del_init(&inc->i_item); in rds_still_queued()
440 write_unlock_irqrestore(&rs->rs_recv_lock, flags); in rds_still_queued()
442 rdsdebug("inc %p rs %p still %d dropped %d\n", inc, rs, ret, drop); in rds_still_queued()
450 int rds_notify_queue_get(struct rds_sock *rs, struct msghdr *msghdr) in rds_notify_queue_get() argument
464 * losing notifications - except when the buffer is so small that it wouldn't in rds_notify_queue_get()
469 max_messages = msghdr->msg_controllen / CMSG_SPACE(sizeof(cmsg)); in rds_notify_queue_get()
474 spin_lock_irqsave(&rs->rs_lock, flags); in rds_notify_queue_get()
475 while (!list_empty(&rs->rs_notify_queue) && count < max_messages) { in rds_notify_queue_get()
476 notifier = list_entry(rs->rs_notify_queue.next, in rds_notify_queue_get()
478 list_move(¬ifier->n_list, ©); in rds_notify_queue_get()
481 spin_unlock_irqrestore(&rs->rs_lock, flags); in rds_notify_queue_get()
490 cmsg.user_token = notifier->n_user_token; in rds_notify_queue_get()
491 cmsg.status = notifier->n_status; in rds_notify_queue_get()
499 list_del_init(¬ifier->n_list); in rds_notify_queue_get()
507 spin_lock_irqsave(&rs->rs_lock, flags); in rds_notify_queue_get()
508 list_splice(©, &rs->rs_notify_queue); in rds_notify_queue_get()
509 spin_unlock_irqrestore(&rs->rs_lock, flags); in rds_notify_queue_get()
518 static int rds_notify_cong(struct rds_sock *rs, struct msghdr *msghdr) in rds_notify_cong() argument
520 uint64_t notify = rs->rs_cong_notify; in rds_notify_cong()
529 spin_lock_irqsave(&rs->rs_lock, flags); in rds_notify_cong()
530 rs->rs_cong_notify &= ~notify; in rds_notify_cong()
531 spin_unlock_irqrestore(&rs->rs_lock, flags); in rds_notify_cong()
540 struct rds_sock *rs) in rds_cmsg_recv() argument
544 if (inc->i_usercopy.rdma_cookie) { in rds_cmsg_recv()
546 sizeof(inc->i_usercopy.rdma_cookie), in rds_cmsg_recv()
547 &inc->i_usercopy.rdma_cookie); in rds_cmsg_recv()
552 if ((inc->i_usercopy.rx_tstamp != 0) && in rds_cmsg_recv()
553 sock_flag(rds_rs_to_sk(rs), SOCK_RCVTSTAMP)) { in rds_cmsg_recv()
555 ns_to_kernel_old_timeval(inc->i_usercopy.rx_tstamp); in rds_cmsg_recv()
557 if (!sock_flag(rds_rs_to_sk(rs), SOCK_TSTAMP_NEW)) { in rds_cmsg_recv()
574 if (rs->rs_rx_traces) { in rds_cmsg_recv()
579 inc->i_rx_lat_trace[RDS_MSG_RX_CMSG] = local_clock(); in rds_cmsg_recv()
580 t.rx_traces = rs->rs_rx_traces; in rds_cmsg_recv()
581 for (i = 0; i < rs->rs_rx_traces; i++) { in rds_cmsg_recv()
582 j = rs->rs_rx_trace[i]; in rds_cmsg_recv()
584 t.rx_trace[i] = inc->i_rx_lat_trace[j + 1] - in rds_cmsg_recv()
585 inc->i_rx_lat_trace[j]; in rds_cmsg_recv()
598 static bool rds_recvmsg_zcookie(struct rds_sock *rs, struct msghdr *msg) in rds_recvmsg_zcookie() argument
600 struct rds_msg_zcopy_queue *q = &rs->rs_zcookie_queue; in rds_recvmsg_zcookie()
605 if (!msg->msg_control) in rds_recvmsg_zcookie()
608 if (!sock_flag(rds_rs_to_sk(rs), SOCK_ZEROCOPY) || in rds_recvmsg_zcookie()
609 msg->msg_controllen < CMSG_SPACE(sizeof(*done))) in rds_recvmsg_zcookie()
612 spin_lock_irqsave(&q->lock, flags); in rds_recvmsg_zcookie()
613 if (!list_empty(&q->zcookie_head)) { in rds_recvmsg_zcookie()
614 info = list_entry(q->zcookie_head.next, in rds_recvmsg_zcookie()
616 list_del(&info->rs_zcookie_next); in rds_recvmsg_zcookie()
618 spin_unlock_irqrestore(&q->lock, flags); in rds_recvmsg_zcookie()
621 done = &info->zcookies; in rds_recvmsg_zcookie()
624 spin_lock_irqsave(&q->lock, flags); in rds_recvmsg_zcookie()
625 list_add(&info->rs_zcookie_next, &q->zcookie_head); in rds_recvmsg_zcookie()
626 spin_unlock_irqrestore(&q->lock, flags); in rds_recvmsg_zcookie()
636 struct sock *sk = sock->sk; in rds_recvmsg()
637 struct rds_sock *rs = rds_sk_to_rs(sk); in rds_recvmsg() local
640 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); in rds_recvmsg()
641 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name); in rds_recvmsg()
644 /* udp_recvmsg()->sock_recvtimeo() gets away without locking too.. */ in rds_recvmsg()
655 /* If there are pending notifications, do those - and nothing else */ in rds_recvmsg()
656 if (!list_empty(&rs->rs_notify_queue)) { in rds_recvmsg()
657 ret = rds_notify_queue_get(rs, msg); in rds_recvmsg()
661 if (rs->rs_cong_notify) { in rds_recvmsg()
662 ret = rds_notify_cong(rs, msg); in rds_recvmsg()
666 if (!rds_next_incoming(rs, &inc)) { in rds_recvmsg()
668 bool reaped = rds_recvmsg_zcookie(rs, msg); in rds_recvmsg()
670 ret = reaped ? 0 : -EAGAIN; in rds_recvmsg()
675 (!list_empty(&rs->rs_notify_queue) || in rds_recvmsg()
676 rs->rs_cong_notify || in rds_recvmsg()
677 rds_next_incoming(rs, &inc)), timeo); in rds_recvmsg()
685 ret = -ETIMEDOUT; in rds_recvmsg()
690 &inc->i_conn->c_faddr, in rds_recvmsg()
691 ntohs(inc->i_hdr.h_sport)); in rds_recvmsg()
692 ret = inc->i_conn->c_trans->inc_copy_to_user(inc, &msg->msg_iter); in rds_recvmsg()
701 if (!rds_still_queued(rs, inc, !(msg_flags & MSG_PEEK))) { in rds_recvmsg()
705 iov_iter_revert(&msg->msg_iter, ret); in rds_recvmsg()
709 if (ret < be32_to_cpu(inc->i_hdr.h_len)) { in rds_recvmsg()
711 ret = be32_to_cpu(inc->i_hdr.h_len); in rds_recvmsg()
712 msg->msg_flags |= MSG_TRUNC; in rds_recvmsg()
715 if (rds_cmsg_recv(inc, msg, rs)) { in rds_recvmsg()
716 ret = -EFAULT; in rds_recvmsg()
719 rds_recvmsg_zcookie(rs, msg); in rds_recvmsg()
723 if (msg->msg_name) { in rds_recvmsg()
724 if (ipv6_addr_v4mapped(&inc->i_saddr)) { in rds_recvmsg()
725 sin->sin_family = AF_INET; in rds_recvmsg()
726 sin->sin_port = inc->i_hdr.h_sport; in rds_recvmsg()
727 sin->sin_addr.s_addr = in rds_recvmsg()
728 inc->i_saddr.s6_addr32[3]; in rds_recvmsg()
729 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); in rds_recvmsg()
730 msg->msg_namelen = sizeof(*sin); in rds_recvmsg()
732 sin6->sin6_family = AF_INET6; in rds_recvmsg()
733 sin6->sin6_port = inc->i_hdr.h_sport; in rds_recvmsg()
734 sin6->sin6_addr = inc->i_saddr; in rds_recvmsg()
735 sin6->sin6_flowinfo = 0; in rds_recvmsg()
736 sin6->sin6_scope_id = rs->rs_bound_scope_id; in rds_recvmsg()
737 msg->msg_namelen = sizeof(*sin6); in rds_recvmsg()
755 void rds_clear_recv_queue(struct rds_sock *rs) in rds_clear_recv_queue() argument
757 struct sock *sk = rds_rs_to_sk(rs); in rds_clear_recv_queue()
761 write_lock_irqsave(&rs->rs_recv_lock, flags); in rds_clear_recv_queue()
762 list_for_each_entry_safe(inc, tmp, &rs->rs_recv_queue, i_item) { in rds_clear_recv_queue()
763 rds_recv_rcvbuf_delta(rs, sk, inc->i_conn->c_lcong, in rds_clear_recv_queue()
764 -be32_to_cpu(inc->i_hdr.h_len), in rds_clear_recv_queue()
765 inc->i_hdr.h_dport); in rds_clear_recv_queue()
766 list_del_init(&inc->i_item); in rds_clear_recv_queue()
769 write_unlock_irqrestore(&rs->rs_recv_lock, flags); in rds_clear_recv_queue()
773 * inc->i_saddr isn't used here because it is only set in the receive
782 minfo.seq = be64_to_cpu(inc->i_hdr.h_sequence); in rds_inc_info_copy()
783 minfo.len = be32_to_cpu(inc->i_hdr.h_len); in rds_inc_info_copy()
784 minfo.tos = inc->i_conn->c_tos; in rds_inc_info_copy()
789 minfo.lport = inc->i_hdr.h_dport; in rds_inc_info_copy()
790 minfo.fport = inc->i_hdr.h_sport; in rds_inc_info_copy()
794 minfo.lport = inc->i_hdr.h_sport; in rds_inc_info_copy()
795 minfo.fport = inc->i_hdr.h_dport; in rds_inc_info_copy()
811 minfo6.seq = be64_to_cpu(inc->i_hdr.h_sequence); in rds6_inc_info_copy()
812 minfo6.len = be32_to_cpu(inc->i_hdr.h_len); in rds6_inc_info_copy()
813 minfo6.tos = inc->i_conn->c_tos; in rds6_inc_info_copy()
818 minfo6.lport = inc->i_hdr.h_dport; in rds6_inc_info_copy()
819 minfo6.fport = inc->i_hdr.h_sport; in rds6_inc_info_copy()
823 minfo6.lport = inc->i_hdr.h_sport; in rds6_inc_info_copy()
824 minfo6.fport = inc->i_hdr.h_dport; in rds6_inc_info_copy()