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/moduleparam.h>
35 #include <linux/gfp.h>
38 #include <linux/list.h>
39 #include <linux/ratelimit.h>
40 #include <linux/export.h>
41 #include <linux/sizes.h>
45 /* When transmitting messages in rds_send_xmit, we need to emerge from
46 * time to time and briefly release the CPU. Otherwise the softlock watchdog
48 * Also, it seems fairer to not let one busy connection stall all the
51 * send_batch_count is the number of times we'll loop in send_xmit. Setting
52 * it to 0 will restore the old behavior (where we looped until we had
55 static int send_batch_count
= SZ_1K
;
56 module_param(send_batch_count
, int, 0444);
57 MODULE_PARM_DESC(send_batch_count
, " batch factor when working the send queue");
59 static void rds_send_remove_from_sock(struct list_head
*messages
, int status
);
62 * Reset the send state. Callers must ensure that this doesn't race with
65 void rds_send_path_reset(struct rds_conn_path
*cp
)
67 struct rds_message
*rm
, *tmp
;
72 cp
->cp_xmit_rm
= NULL
;
73 /* Tell the user the RDMA op is no longer mapped by the
74 * transport. This isn't entirely true (it's flushed out
75 * independently) but as the connection is down, there's
76 * no ongoing RDMA to/from that memory */
77 rds_message_unmapped(rm
);
82 cp
->cp_xmit_hdr_off
= 0;
83 cp
->cp_xmit_data_off
= 0;
84 cp
->cp_xmit_atomic_sent
= 0;
85 cp
->cp_xmit_rdma_sent
= 0;
86 cp
->cp_xmit_data_sent
= 0;
88 cp
->cp_conn
->c_map_queued
= 0;
90 cp
->cp_unacked_packets
= rds_sysctl_max_unacked_packets
;
91 cp
->cp_unacked_bytes
= rds_sysctl_max_unacked_bytes
;
93 /* Mark messages as retransmissions, and move them to the send q */
94 spin_lock_irqsave(&cp
->cp_lock
, flags
);
95 list_for_each_entry_safe(rm
, tmp
, &cp
->cp_retrans
, m_conn_item
) {
96 set_bit(RDS_MSG_ACK_REQUIRED
, &rm
->m_flags
);
97 set_bit(RDS_MSG_RETRANSMITTED
, &rm
->m_flags
);
99 list_splice_init(&cp
->cp_retrans
, &cp
->cp_send_queue
);
100 spin_unlock_irqrestore(&cp
->cp_lock
, flags
);
102 EXPORT_SYMBOL_GPL(rds_send_path_reset
);
104 static int acquire_in_xmit(struct rds_conn_path
*cp
)
106 return test_and_set_bit(RDS_IN_XMIT
, &cp
->cp_flags
) == 0;
109 static void release_in_xmit(struct rds_conn_path
*cp
)
111 clear_bit(RDS_IN_XMIT
, &cp
->cp_flags
);
112 smp_mb__after_atomic();
114 * We don't use wait_on_bit()/wake_up_bit() because our waking is in a
115 * hot path and finding waiters is very rare. We don't want to walk
116 * the system-wide hashed waitqueue buckets in the fast path only to
117 * almost never find waiters.
119 if (waitqueue_active(&cp
->cp_waitq
))
120 wake_up_all(&cp
->cp_waitq
);
124 * We're making the conscious trade-off here to only send one message
125 * down the connection at a time.
127 * - tx queueing is a simple fifo list
128 * - reassembly is optional and easily done by transports per conn
129 * - no per flow rx lookup at all, straight to the socket
130 * - less per-frag memory and wire overhead
132 * - queued acks can be delayed behind large messages
134 * - small message latency is higher behind queued large messages
135 * - large message latency isn't starved by intervening small sends
137 int rds_send_xmit(struct rds_conn_path
*cp
)
139 struct rds_connection
*conn
= cp
->cp_conn
;
140 struct rds_message
*rm
;
143 struct scatterlist
*sg
;
145 LIST_HEAD(to_be_dropped
);
147 unsigned long send_gen
= 0;
153 * sendmsg calls here after having queued its message on the send
154 * queue. We only have one task feeding the connection at a time. If
155 * another thread is already feeding the queue then we back off. This
156 * avoids blocking the caller and trading per-connection data between
157 * caches per message.
159 if (!acquire_in_xmit(cp
)) {
160 rds_stats_inc(s_send_lock_contention
);
166 * we record the send generation after doing the xmit acquire.
167 * if someone else manages to jump in and do some work, we'll use
168 * this to avoid a goto restart farther down.
170 * The acquire_in_xmit() check above ensures that only one
171 * caller can increment c_send_gen at any time.
174 send_gen
= cp
->cp_send_gen
;
177 * rds_conn_shutdown() sets the conn state and then tests RDS_IN_XMIT,
178 * we do the opposite to avoid races.
180 if (!rds_conn_path_up(cp
)) {
186 if (conn
->c_trans
->xmit_path_prepare
)
187 conn
->c_trans
->xmit_path_prepare(cp
);
190 * spin trying to push headers and data down the connection until
191 * the connection doesn't make forward progress.
198 * If between sending messages, we can send a pending congestion
201 if (!rm
&& test_and_clear_bit(0, &conn
->c_map_queued
)) {
202 rm
= rds_cong_update_alloc(conn
);
207 rm
->data
.op_active
= 1;
208 rm
->m_inc
.i_conn_path
= cp
;
209 rm
->m_inc
.i_conn
= cp
->cp_conn
;
215 * If not already working on one, grab the next message.
217 * cp_xmit_rm holds a ref while we're sending this message down
218 * the connction. We can use this ref while holding the
219 * send_sem.. rds_send_reset() is serialized with it.
226 /* we want to process as big a batch as we can, but
227 * we also want to avoid softlockups. If we've been
228 * through a lot of messages, lets back off and see
229 * if anyone else jumps in
231 if (batch_count
>= send_batch_count
)
234 spin_lock_irqsave(&cp
->cp_lock
, flags
);
236 if (!list_empty(&cp
->cp_send_queue
)) {
237 rm
= list_entry(cp
->cp_send_queue
.next
,
240 rds_message_addref(rm
);
243 * Move the message from the send queue to the retransmit
246 list_move_tail(&rm
->m_conn_item
,
250 spin_unlock_irqrestore(&cp
->cp_lock
, flags
);
255 /* Unfortunately, the way Infiniband deals with
256 * RDMA to a bad MR key is by moving the entire
257 * queue pair to error state. We cold possibly
258 * recover from that, but right now we drop the
260 * Therefore, we never retransmit messages with RDMA ops.
262 if (rm
->rdma
.op_active
&&
263 test_bit(RDS_MSG_RETRANSMITTED
, &rm
->m_flags
)) {
264 spin_lock_irqsave(&cp
->cp_lock
, flags
);
265 if (test_and_clear_bit(RDS_MSG_ON_CONN
, &rm
->m_flags
))
266 list_move(&rm
->m_conn_item
, &to_be_dropped
);
267 spin_unlock_irqrestore(&cp
->cp_lock
, flags
);
271 /* Require an ACK every once in a while */
272 len
= ntohl(rm
->m_inc
.i_hdr
.h_len
);
273 if (cp
->cp_unacked_packets
== 0 ||
274 cp
->cp_unacked_bytes
< len
) {
275 __set_bit(RDS_MSG_ACK_REQUIRED
, &rm
->m_flags
);
277 cp
->cp_unacked_packets
=
278 rds_sysctl_max_unacked_packets
;
279 cp
->cp_unacked_bytes
=
280 rds_sysctl_max_unacked_bytes
;
281 rds_stats_inc(s_send_ack_required
);
283 cp
->cp_unacked_bytes
-= len
;
284 cp
->cp_unacked_packets
--;
290 /* The transport either sends the whole rdma or none of it */
291 if (rm
->rdma
.op_active
&& !cp
->cp_xmit_rdma_sent
) {
292 rm
->m_final_op
= &rm
->rdma
;
293 /* The transport owns the mapped memory for now.
294 * You can't unmap it while it's on the send queue
296 set_bit(RDS_MSG_MAPPED
, &rm
->m_flags
);
297 ret
= conn
->c_trans
->xmit_rdma(conn
, &rm
->rdma
);
299 clear_bit(RDS_MSG_MAPPED
, &rm
->m_flags
);
300 wake_up_interruptible(&rm
->m_flush_wait
);
303 cp
->cp_xmit_rdma_sent
= 1;
307 if (rm
->atomic
.op_active
&& !cp
->cp_xmit_atomic_sent
) {
308 rm
->m_final_op
= &rm
->atomic
;
309 /* The transport owns the mapped memory for now.
310 * You can't unmap it while it's on the send queue
312 set_bit(RDS_MSG_MAPPED
, &rm
->m_flags
);
313 ret
= conn
->c_trans
->xmit_atomic(conn
, &rm
->atomic
);
315 clear_bit(RDS_MSG_MAPPED
, &rm
->m_flags
);
316 wake_up_interruptible(&rm
->m_flush_wait
);
319 cp
->cp_xmit_atomic_sent
= 1;
324 * A number of cases require an RDS header to be sent
325 * even if there is no data.
326 * We permit 0-byte sends; rds-ping depends on this.
327 * However, if there are exclusively attached silent ops,
328 * we skip the hdr/data send, to enable silent operation.
330 if (rm
->data
.op_nents
== 0) {
332 int all_ops_are_silent
= 1;
334 ops_present
= (rm
->atomic
.op_active
|| rm
->rdma
.op_active
);
335 if (rm
->atomic
.op_active
&& !rm
->atomic
.op_silent
)
336 all_ops_are_silent
= 0;
337 if (rm
->rdma
.op_active
&& !rm
->rdma
.op_silent
)
338 all_ops_are_silent
= 0;
340 if (ops_present
&& all_ops_are_silent
341 && !rm
->m_rdma_cookie
)
342 rm
->data
.op_active
= 0;
345 if (rm
->data
.op_active
&& !cp
->cp_xmit_data_sent
) {
346 rm
->m_final_op
= &rm
->data
;
348 ret
= conn
->c_trans
->xmit(conn
, rm
,
351 cp
->cp_xmit_data_off
);
355 if (cp
->cp_xmit_hdr_off
< sizeof(struct rds_header
)) {
356 tmp
= min_t(int, ret
,
357 sizeof(struct rds_header
) -
358 cp
->cp_xmit_hdr_off
);
359 cp
->cp_xmit_hdr_off
+= tmp
;
363 sg
= &rm
->data
.op_sg
[cp
->cp_xmit_sg
];
365 tmp
= min_t(int, ret
, sg
->length
-
366 cp
->cp_xmit_data_off
);
367 cp
->cp_xmit_data_off
+= tmp
;
369 if (cp
->cp_xmit_data_off
== sg
->length
) {
370 cp
->cp_xmit_data_off
= 0;
373 BUG_ON(ret
!= 0 && cp
->cp_xmit_sg
==
378 if (cp
->cp_xmit_hdr_off
== sizeof(struct rds_header
) &&
379 (cp
->cp_xmit_sg
== rm
->data
.op_nents
))
380 cp
->cp_xmit_data_sent
= 1;
384 * A rm will only take multiple times through this loop
385 * if there is a data op. Thus, if the data is sent (or there was
386 * none), then we're done with the rm.
388 if (!rm
->data
.op_active
|| cp
->cp_xmit_data_sent
) {
389 cp
->cp_xmit_rm
= NULL
;
391 cp
->cp_xmit_hdr_off
= 0;
392 cp
->cp_xmit_data_off
= 0;
393 cp
->cp_xmit_rdma_sent
= 0;
394 cp
->cp_xmit_atomic_sent
= 0;
395 cp
->cp_xmit_data_sent
= 0;
402 if (conn
->c_trans
->xmit_path_complete
)
403 conn
->c_trans
->xmit_path_complete(cp
);
406 /* Nuke any messages we decided not to retransmit. */
407 if (!list_empty(&to_be_dropped
)) {
408 /* irqs on here, so we can put(), unlike above */
409 list_for_each_entry(rm
, &to_be_dropped
, m_conn_item
)
411 rds_send_remove_from_sock(&to_be_dropped
, RDS_RDMA_DROPPED
);
415 * Other senders can queue a message after we last test the send queue
416 * but before we clear RDS_IN_XMIT. In that case they'd back off and
417 * not try and send their newly queued message. We need to check the
418 * send queue after having cleared RDS_IN_XMIT so that their message
419 * doesn't get stuck on the send queue.
421 * If the transport cannot continue (i.e ret != 0), then it must
422 * call us when more room is available, such as from the tx
423 * completion handler.
425 * We have an extra generation check here so that if someone manages
426 * to jump in after our release_in_xmit, we'll see that they have done
427 * some work and we will skip our goto
431 if ((test_bit(0, &conn
->c_map_queued
) ||
432 !list_empty(&cp
->cp_send_queue
)) &&
433 send_gen
== cp
->cp_send_gen
) {
434 rds_stats_inc(s_send_lock_queue_raced
);
435 if (batch_count
< send_batch_count
)
437 queue_delayed_work(rds_wq
, &cp
->cp_send_w
, 1);
443 EXPORT_SYMBOL_GPL(rds_send_xmit
);
445 static void rds_send_sndbuf_remove(struct rds_sock
*rs
, struct rds_message
*rm
)
447 u32 len
= be32_to_cpu(rm
->m_inc
.i_hdr
.h_len
);
449 assert_spin_locked(&rs
->rs_lock
);
451 BUG_ON(rs
->rs_snd_bytes
< len
);
452 rs
->rs_snd_bytes
-= len
;
454 if (rs
->rs_snd_bytes
== 0)
455 rds_stats_inc(s_send_queue_empty
);
458 static inline int rds_send_is_acked(struct rds_message
*rm
, u64 ack
,
459 is_acked_func is_acked
)
462 return is_acked(rm
, ack
);
463 return be64_to_cpu(rm
->m_inc
.i_hdr
.h_sequence
) <= ack
;
467 * This is pretty similar to what happens below in the ACK
468 * handling code - except that we call here as soon as we get
469 * the IB send completion on the RDMA op and the accompanying
472 void rds_rdma_send_complete(struct rds_message
*rm
, int status
)
474 struct rds_sock
*rs
= NULL
;
475 struct rm_rdma_op
*ro
;
476 struct rds_notifier
*notifier
;
479 spin_lock_irqsave(&rm
->m_rs_lock
, flags
);
482 if (test_bit(RDS_MSG_ON_SOCK
, &rm
->m_flags
) &&
483 ro
->op_active
&& ro
->op_notify
&& ro
->op_notifier
) {
484 notifier
= ro
->op_notifier
;
486 sock_hold(rds_rs_to_sk(rs
));
488 notifier
->n_status
= status
;
489 spin_lock(&rs
->rs_lock
);
490 list_add_tail(¬ifier
->n_list
, &rs
->rs_notify_queue
);
491 spin_unlock(&rs
->rs_lock
);
493 ro
->op_notifier
= NULL
;
496 spin_unlock_irqrestore(&rm
->m_rs_lock
, flags
);
499 rds_wake_sk_sleep(rs
);
500 sock_put(rds_rs_to_sk(rs
));
503 EXPORT_SYMBOL_GPL(rds_rdma_send_complete
);
506 * Just like above, except looks at atomic op
508 void rds_atomic_send_complete(struct rds_message
*rm
, int status
)
510 struct rds_sock
*rs
= NULL
;
511 struct rm_atomic_op
*ao
;
512 struct rds_notifier
*notifier
;
515 spin_lock_irqsave(&rm
->m_rs_lock
, flags
);
518 if (test_bit(RDS_MSG_ON_SOCK
, &rm
->m_flags
)
519 && ao
->op_active
&& ao
->op_notify
&& ao
->op_notifier
) {
520 notifier
= ao
->op_notifier
;
522 sock_hold(rds_rs_to_sk(rs
));
524 notifier
->n_status
= status
;
525 spin_lock(&rs
->rs_lock
);
526 list_add_tail(¬ifier
->n_list
, &rs
->rs_notify_queue
);
527 spin_unlock(&rs
->rs_lock
);
529 ao
->op_notifier
= NULL
;
532 spin_unlock_irqrestore(&rm
->m_rs_lock
, flags
);
535 rds_wake_sk_sleep(rs
);
536 sock_put(rds_rs_to_sk(rs
));
539 EXPORT_SYMBOL_GPL(rds_atomic_send_complete
);
542 * This is the same as rds_rdma_send_complete except we
543 * don't do any locking - we have all the ingredients (message,
544 * socket, socket lock) and can just move the notifier.
547 __rds_send_complete(struct rds_sock
*rs
, struct rds_message
*rm
, int status
)
549 struct rm_rdma_op
*ro
;
550 struct rm_atomic_op
*ao
;
553 if (ro
->op_active
&& ro
->op_notify
&& ro
->op_notifier
) {
554 ro
->op_notifier
->n_status
= status
;
555 list_add_tail(&ro
->op_notifier
->n_list
, &rs
->rs_notify_queue
);
556 ro
->op_notifier
= NULL
;
560 if (ao
->op_active
&& ao
->op_notify
&& ao
->op_notifier
) {
561 ao
->op_notifier
->n_status
= status
;
562 list_add_tail(&ao
->op_notifier
->n_list
, &rs
->rs_notify_queue
);
563 ao
->op_notifier
= NULL
;
566 /* No need to wake the app - caller does this */
570 * This removes messages from the socket's list if they're on it. The list
571 * argument must be private to the caller, we must be able to modify it
572 * without locks. The messages must have a reference held for their
573 * position on the list. This function will drop that reference after
574 * removing the messages from the 'messages' list regardless of if it found
575 * the messages on the socket list or not.
577 static void rds_send_remove_from_sock(struct list_head
*messages
, int status
)
580 struct rds_sock
*rs
= NULL
;
581 struct rds_message
*rm
;
583 while (!list_empty(messages
)) {
586 rm
= list_entry(messages
->next
, struct rds_message
,
588 list_del_init(&rm
->m_conn_item
);
591 * If we see this flag cleared then we're *sure* that someone
592 * else beat us to removing it from the sock. If we race
593 * with their flag update we'll get the lock and then really
594 * see that the flag has been cleared.
596 * The message spinlock makes sure nobody clears rm->m_rs
597 * while we're messing with it. It does not prevent the
598 * message from being removed from the socket, though.
600 spin_lock_irqsave(&rm
->m_rs_lock
, flags
);
601 if (!test_bit(RDS_MSG_ON_SOCK
, &rm
->m_flags
))
602 goto unlock_and_drop
;
604 if (rs
!= rm
->m_rs
) {
606 rds_wake_sk_sleep(rs
);
607 sock_put(rds_rs_to_sk(rs
));
611 sock_hold(rds_rs_to_sk(rs
));
614 goto unlock_and_drop
;
615 spin_lock(&rs
->rs_lock
);
617 if (test_and_clear_bit(RDS_MSG_ON_SOCK
, &rm
->m_flags
)) {
618 struct rm_rdma_op
*ro
= &rm
->rdma
;
619 struct rds_notifier
*notifier
;
621 list_del_init(&rm
->m_sock_item
);
622 rds_send_sndbuf_remove(rs
, rm
);
624 if (ro
->op_active
&& ro
->op_notifier
&&
625 (ro
->op_notify
|| (ro
->op_recverr
&& status
))) {
626 notifier
= ro
->op_notifier
;
627 list_add_tail(¬ifier
->n_list
,
628 &rs
->rs_notify_queue
);
629 if (!notifier
->n_status
)
630 notifier
->n_status
= status
;
631 rm
->rdma
.op_notifier
= NULL
;
636 spin_unlock(&rs
->rs_lock
);
639 spin_unlock_irqrestore(&rm
->m_rs_lock
, flags
);
646 rds_wake_sk_sleep(rs
);
647 sock_put(rds_rs_to_sk(rs
));
652 * Transports call here when they've determined that the receiver queued
653 * messages up to, and including, the given sequence number. Messages are
654 * moved to the retrans queue when rds_send_xmit picks them off the send
655 * queue. This means that in the TCP case, the message may not have been
656 * assigned the m_ack_seq yet - but that's fine as long as tcp_is_acked
657 * checks the RDS_MSG_HAS_ACK_SEQ bit.
659 void rds_send_path_drop_acked(struct rds_conn_path
*cp
, u64 ack
,
660 is_acked_func is_acked
)
662 struct rds_message
*rm
, *tmp
;
666 spin_lock_irqsave(&cp
->cp_lock
, flags
);
668 list_for_each_entry_safe(rm
, tmp
, &cp
->cp_retrans
, m_conn_item
) {
669 if (!rds_send_is_acked(rm
, ack
, is_acked
))
672 list_move(&rm
->m_conn_item
, &list
);
673 clear_bit(RDS_MSG_ON_CONN
, &rm
->m_flags
);
676 /* order flag updates with spin locks */
677 if (!list_empty(&list
))
678 smp_mb__after_atomic();
680 spin_unlock_irqrestore(&cp
->cp_lock
, flags
);
682 /* now remove the messages from the sock list as needed */
683 rds_send_remove_from_sock(&list
, RDS_RDMA_SUCCESS
);
685 EXPORT_SYMBOL_GPL(rds_send_path_drop_acked
);
687 void rds_send_drop_acked(struct rds_connection
*conn
, u64 ack
,
688 is_acked_func is_acked
)
690 WARN_ON(conn
->c_trans
->t_mp_capable
);
691 rds_send_path_drop_acked(&conn
->c_path
[0], ack
, is_acked
);
693 EXPORT_SYMBOL_GPL(rds_send_drop_acked
);
695 void rds_send_drop_to(struct rds_sock
*rs
, struct sockaddr_in
*dest
)
697 struct rds_message
*rm
, *tmp
;
698 struct rds_connection
*conn
;
699 struct rds_conn_path
*cp
;
703 /* get all the messages we're dropping under the rs lock */
704 spin_lock_irqsave(&rs
->rs_lock
, flags
);
706 list_for_each_entry_safe(rm
, tmp
, &rs
->rs_send_queue
, m_sock_item
) {
707 if (dest
&& (dest
->sin_addr
.s_addr
!= rm
->m_daddr
||
708 dest
->sin_port
!= rm
->m_inc
.i_hdr
.h_dport
))
711 list_move(&rm
->m_sock_item
, &list
);
712 rds_send_sndbuf_remove(rs
, rm
);
713 clear_bit(RDS_MSG_ON_SOCK
, &rm
->m_flags
);
716 /* order flag updates with the rs lock */
717 smp_mb__after_atomic();
719 spin_unlock_irqrestore(&rs
->rs_lock
, flags
);
721 if (list_empty(&list
))
724 /* Remove the messages from the conn */
725 list_for_each_entry(rm
, &list
, m_sock_item
) {
727 conn
= rm
->m_inc
.i_conn
;
728 if (conn
->c_trans
->t_mp_capable
)
729 cp
= rm
->m_inc
.i_conn_path
;
731 cp
= &conn
->c_path
[0];
733 spin_lock_irqsave(&cp
->cp_lock
, flags
);
735 * Maybe someone else beat us to removing rm from the conn.
736 * If we race with their flag update we'll get the lock and
737 * then really see that the flag has been cleared.
739 if (!test_and_clear_bit(RDS_MSG_ON_CONN
, &rm
->m_flags
)) {
740 spin_unlock_irqrestore(&cp
->cp_lock
, flags
);
741 spin_lock_irqsave(&rm
->m_rs_lock
, flags
);
743 spin_unlock_irqrestore(&rm
->m_rs_lock
, flags
);
746 list_del_init(&rm
->m_conn_item
);
747 spin_unlock_irqrestore(&cp
->cp_lock
, flags
);
750 * Couldn't grab m_rs_lock in top loop (lock ordering),
753 spin_lock_irqsave(&rm
->m_rs_lock
, flags
);
755 spin_lock(&rs
->rs_lock
);
756 __rds_send_complete(rs
, rm
, RDS_RDMA_CANCELED
);
757 spin_unlock(&rs
->rs_lock
);
760 spin_unlock_irqrestore(&rm
->m_rs_lock
, flags
);
765 rds_wake_sk_sleep(rs
);
767 while (!list_empty(&list
)) {
768 rm
= list_entry(list
.next
, struct rds_message
, m_sock_item
);
769 list_del_init(&rm
->m_sock_item
);
770 rds_message_wait(rm
);
772 /* just in case the code above skipped this message
773 * because RDS_MSG_ON_CONN wasn't set, run it again here
774 * taking m_rs_lock is the only thing that keeps us
775 * from racing with ack processing.
777 spin_lock_irqsave(&rm
->m_rs_lock
, flags
);
779 spin_lock(&rs
->rs_lock
);
780 __rds_send_complete(rs
, rm
, RDS_RDMA_CANCELED
);
781 spin_unlock(&rs
->rs_lock
);
784 spin_unlock_irqrestore(&rm
->m_rs_lock
, flags
);
791 * we only want this to fire once so we use the callers 'queued'. It's
792 * possible that another thread can race with us and remove the
793 * message from the flow with RDS_CANCEL_SENT_TO.
795 static int rds_send_queue_rm(struct rds_sock
*rs
, struct rds_connection
*conn
,
796 struct rds_conn_path
*cp
,
797 struct rds_message
*rm
, __be16 sport
,
798 __be16 dport
, int *queued
)
806 len
= be32_to_cpu(rm
->m_inc
.i_hdr
.h_len
);
808 /* this is the only place which holds both the socket's rs_lock
809 * and the connection's c_lock */
810 spin_lock_irqsave(&rs
->rs_lock
, flags
);
813 * If there is a little space in sndbuf, we don't queue anything,
814 * and userspace gets -EAGAIN. But poll() indicates there's send
815 * room. This can lead to bad behavior (spinning) if snd_bytes isn't
816 * freed up by incoming acks. So we check the *old* value of
817 * rs_snd_bytes here to allow the last msg to exceed the buffer,
818 * and poll() now knows no more data can be sent.
820 if (rs
->rs_snd_bytes
< rds_sk_sndbuf(rs
)) {
821 rs
->rs_snd_bytes
+= len
;
823 /* let recv side know we are close to send space exhaustion.
824 * This is probably not the optimal way to do it, as this
825 * means we set the flag on *all* messages as soon as our
826 * throughput hits a certain threshold.
828 if (rs
->rs_snd_bytes
>= rds_sk_sndbuf(rs
) / 2)
829 __set_bit(RDS_MSG_ACK_REQUIRED
, &rm
->m_flags
);
831 list_add_tail(&rm
->m_sock_item
, &rs
->rs_send_queue
);
832 set_bit(RDS_MSG_ON_SOCK
, &rm
->m_flags
);
833 rds_message_addref(rm
);
836 /* The code ordering is a little weird, but we're
837 trying to minimize the time we hold c_lock */
838 rds_message_populate_header(&rm
->m_inc
.i_hdr
, sport
, dport
, 0);
839 rm
->m_inc
.i_conn
= conn
;
840 rm
->m_inc
.i_conn_path
= cp
;
841 rds_message_addref(rm
);
843 spin_lock(&cp
->cp_lock
);
844 rm
->m_inc
.i_hdr
.h_sequence
= cpu_to_be64(cp
->cp_next_tx_seq
++);
845 list_add_tail(&rm
->m_conn_item
, &cp
->cp_send_queue
);
846 set_bit(RDS_MSG_ON_CONN
, &rm
->m_flags
);
847 spin_unlock(&cp
->cp_lock
);
849 rdsdebug("queued msg %p len %d, rs %p bytes %d seq %llu\n",
850 rm
, len
, rs
, rs
->rs_snd_bytes
,
851 (unsigned long long)be64_to_cpu(rm
->m_inc
.i_hdr
.h_sequence
));
856 spin_unlock_irqrestore(&rs
->rs_lock
, flags
);
862 * rds_message is getting to be quite complicated, and we'd like to allocate
863 * it all in one go. This figures out how big it needs to be up front.
865 static int rds_rm_size(struct msghdr
*msg
, int data_len
)
867 struct cmsghdr
*cmsg
;
872 for_each_cmsghdr(cmsg
, msg
) {
873 if (!CMSG_OK(msg
, cmsg
))
876 if (cmsg
->cmsg_level
!= SOL_RDS
)
879 switch (cmsg
->cmsg_type
) {
880 case RDS_CMSG_RDMA_ARGS
:
882 retval
= rds_rdma_extra_size(CMSG_DATA(cmsg
));
889 case RDS_CMSG_RDMA_DEST
:
890 case RDS_CMSG_RDMA_MAP
:
892 /* these are valid but do no add any size */
895 case RDS_CMSG_ATOMIC_CSWP
:
896 case RDS_CMSG_ATOMIC_FADD
:
897 case RDS_CMSG_MASKED_ATOMIC_CSWP
:
898 case RDS_CMSG_MASKED_ATOMIC_FADD
:
900 size
+= sizeof(struct scatterlist
);
909 size
+= ceil(data_len
, PAGE_SIZE
) * sizeof(struct scatterlist
);
911 /* Ensure (DEST, MAP) are never used with (ARGS, ATOMIC) */
912 if (cmsg_groups
== 3)
918 static int rds_cmsg_send(struct rds_sock
*rs
, struct rds_message
*rm
,
919 struct msghdr
*msg
, int *allocated_mr
)
921 struct cmsghdr
*cmsg
;
924 for_each_cmsghdr(cmsg
, msg
) {
925 if (!CMSG_OK(msg
, cmsg
))
928 if (cmsg
->cmsg_level
!= SOL_RDS
)
931 /* As a side effect, RDMA_DEST and RDMA_MAP will set
932 * rm->rdma.m_rdma_cookie and rm->rdma.m_rdma_mr.
934 switch (cmsg
->cmsg_type
) {
935 case RDS_CMSG_RDMA_ARGS
:
936 ret
= rds_cmsg_rdma_args(rs
, rm
, cmsg
);
939 case RDS_CMSG_RDMA_DEST
:
940 ret
= rds_cmsg_rdma_dest(rs
, rm
, cmsg
);
943 case RDS_CMSG_RDMA_MAP
:
944 ret
= rds_cmsg_rdma_map(rs
, rm
, cmsg
);
948 case RDS_CMSG_ATOMIC_CSWP
:
949 case RDS_CMSG_ATOMIC_FADD
:
950 case RDS_CMSG_MASKED_ATOMIC_CSWP
:
951 case RDS_CMSG_MASKED_ATOMIC_FADD
:
952 ret
= rds_cmsg_atomic(rs
, rm
, cmsg
);
966 static void rds_send_ping(struct rds_connection
*conn
);
968 static int rds_send_mprds_hash(struct rds_sock
*rs
, struct rds_connection
*conn
)
972 if (conn
->c_npaths
== 0)
973 hash
= RDS_MPATH_HASH(rs
, RDS_MPATH_WORKERS
);
975 hash
= RDS_MPATH_HASH(rs
, conn
->c_npaths
);
976 if (conn
->c_npaths
== 0 && hash
!= 0) {
979 if (conn
->c_npaths
== 0) {
980 wait_event_interruptible(conn
->c_hs_waitq
,
981 (conn
->c_npaths
!= 0));
983 if (conn
->c_npaths
== 1)
989 int rds_sendmsg(struct socket
*sock
, struct msghdr
*msg
, size_t payload_len
)
991 struct sock
*sk
= sock
->sk
;
992 struct rds_sock
*rs
= rds_sk_to_rs(sk
);
993 DECLARE_SOCKADDR(struct sockaddr_in
*, usin
, msg
->msg_name
);
996 struct rds_message
*rm
= NULL
;
997 struct rds_connection
*conn
;
999 int queued
= 0, allocated_mr
= 0;
1000 int nonblock
= msg
->msg_flags
& MSG_DONTWAIT
;
1001 long timeo
= sock_sndtimeo(sk
, nonblock
);
1002 struct rds_conn_path
*cpath
;
1004 /* Mirror Linux UDP mirror of BSD error message compatibility */
1005 /* XXX: Perhaps MSG_MORE someday */
1006 if (msg
->msg_flags
& ~(MSG_DONTWAIT
| MSG_CMSG_COMPAT
)) {
1011 if (msg
->msg_namelen
) {
1012 /* XXX fail non-unicast destination IPs? */
1013 if (msg
->msg_namelen
< sizeof(*usin
) || usin
->sin_family
!= AF_INET
) {
1017 daddr
= usin
->sin_addr
.s_addr
;
1018 dport
= usin
->sin_port
;
1020 /* We only care about consistency with ->connect() */
1022 daddr
= rs
->rs_conn_addr
;
1023 dport
= rs
->rs_conn_port
;
1028 if (daddr
== 0 || rs
->rs_bound_addr
== 0) {
1030 ret
= -ENOTCONN
; /* XXX not a great errno */
1035 if (payload_len
> rds_sk_sndbuf(rs
)) {
1040 /* size of rm including all sgs */
1041 ret
= rds_rm_size(msg
, payload_len
);
1045 rm
= rds_message_alloc(ret
, GFP_KERNEL
);
1051 /* Attach data to the rm */
1053 rm
->data
.op_sg
= rds_message_alloc_sgs(rm
, ceil(payload_len
, PAGE_SIZE
));
1054 if (!rm
->data
.op_sg
) {
1058 ret
= rds_message_copy_from_user(rm
, &msg
->msg_iter
);
1062 rm
->data
.op_active
= 1;
1064 rm
->m_daddr
= daddr
;
1066 /* rds_conn_create has a spinlock that runs with IRQ off.
1067 * Caching the conn in the socket helps a lot. */
1068 if (rs
->rs_conn
&& rs
->rs_conn
->c_faddr
== daddr
)
1071 conn
= rds_conn_create_outgoing(sock_net(sock
->sk
),
1072 rs
->rs_bound_addr
, daddr
,
1074 sock
->sk
->sk_allocation
);
1076 ret
= PTR_ERR(conn
);
1082 /* Parse any control messages the user may have included. */
1083 ret
= rds_cmsg_send(rs
, rm
, msg
, &allocated_mr
);
1087 if (rm
->rdma
.op_active
&& !conn
->c_trans
->xmit_rdma
) {
1088 printk_ratelimited(KERN_NOTICE
"rdma_op %p conn xmit_rdma %p\n",
1089 &rm
->rdma
, conn
->c_trans
->xmit_rdma
);
1094 if (rm
->atomic
.op_active
&& !conn
->c_trans
->xmit_atomic
) {
1095 printk_ratelimited(KERN_NOTICE
"atomic_op %p conn xmit_atomic %p\n",
1096 &rm
->atomic
, conn
->c_trans
->xmit_atomic
);
1101 if (conn
->c_trans
->t_mp_capable
)
1102 cpath
= &conn
->c_path
[rds_send_mprds_hash(rs
, conn
)];
1104 cpath
= &conn
->c_path
[0];
1106 rds_conn_path_connect_if_down(cpath
);
1108 ret
= rds_cong_wait(conn
->c_fcong
, dport
, nonblock
, rs
);
1110 rs
->rs_seen_congestion
= 1;
1113 while (!rds_send_queue_rm(rs
, conn
, cpath
, rm
, rs
->rs_bound_port
,
1115 rds_stats_inc(s_send_queue_full
);
1122 timeo
= wait_event_interruptible_timeout(*sk_sleep(sk
),
1123 rds_send_queue_rm(rs
, conn
, cpath
, rm
,
1128 rdsdebug("sendmsg woke queued %d timeo %ld\n", queued
, timeo
);
1129 if (timeo
> 0 || timeo
== MAX_SCHEDULE_TIMEOUT
)
1139 * By now we've committed to the send. We reuse rds_send_worker()
1140 * to retry sends in the rds thread if the transport asks us to.
1142 rds_stats_inc(s_send_queued
);
1144 ret
= rds_send_xmit(cpath
);
1145 if (ret
== -ENOMEM
|| ret
== -EAGAIN
)
1146 queue_delayed_work(rds_wq
, &cpath
->cp_send_w
, 1);
1148 rds_message_put(rm
);
1152 /* If the user included a RDMA_MAP cmsg, we allocated a MR on the fly.
1153 * If the sendmsg goes through, we keep the MR. If it fails with EAGAIN
1154 * or in any other way, we need to destroy the MR again */
1156 rds_rdma_unuse(rs
, rds_rdma_cookie_key(rm
->m_rdma_cookie
), 1);
1159 rds_message_put(rm
);
1164 * send out a probe. Can be shared by rds_send_ping,
1165 * rds_send_pong, rds_send_hb.
1166 * rds_send_hb should use h_flags
1167 * RDS_FLAG_HB_PING|RDS_FLAG_ACK_REQUIRED
1169 * RDS_FLAG_HB_PONG|RDS_FLAG_ACK_REQUIRED
1172 rds_send_probe(struct rds_conn_path
*cp
, __be16 sport
,
1173 __be16 dport
, u8 h_flags
)
1175 struct rds_message
*rm
;
1176 unsigned long flags
;
1179 rm
= rds_message_alloc(0, GFP_ATOMIC
);
1185 rm
->m_daddr
= cp
->cp_conn
->c_faddr
;
1186 rm
->data
.op_active
= 1;
1188 rds_conn_path_connect_if_down(cp
);
1190 ret
= rds_cong_wait(cp
->cp_conn
->c_fcong
, dport
, 1, NULL
);
1194 spin_lock_irqsave(&cp
->cp_lock
, flags
);
1195 list_add_tail(&rm
->m_conn_item
, &cp
->cp_send_queue
);
1196 set_bit(RDS_MSG_ON_CONN
, &rm
->m_flags
);
1197 rds_message_addref(rm
);
1198 rm
->m_inc
.i_conn
= cp
->cp_conn
;
1199 rm
->m_inc
.i_conn_path
= cp
;
1201 rds_message_populate_header(&rm
->m_inc
.i_hdr
, sport
, dport
,
1202 cp
->cp_next_tx_seq
);
1203 rm
->m_inc
.i_hdr
.h_flags
|= h_flags
;
1204 cp
->cp_next_tx_seq
++;
1206 if (RDS_HS_PROBE(sport
, dport
) && cp
->cp_conn
->c_trans
->t_mp_capable
) {
1207 u16 npaths
= RDS_MPATH_WORKERS
;
1209 rds_message_add_extension(&rm
->m_inc
.i_hdr
,
1210 RDS_EXTHDR_NPATHS
, &npaths
,
1213 spin_unlock_irqrestore(&cp
->cp_lock
, flags
);
1215 rds_stats_inc(s_send_queued
);
1216 rds_stats_inc(s_send_pong
);
1218 /* schedule the send work on rds_wq */
1219 queue_delayed_work(rds_wq
, &cp
->cp_send_w
, 1);
1221 rds_message_put(rm
);
1226 rds_message_put(rm
);
1231 rds_send_pong(struct rds_conn_path
*cp
, __be16 dport
)
1233 return rds_send_probe(cp
, 0, dport
, 0);
1237 rds_send_ping(struct rds_connection
*conn
)
1239 unsigned long flags
;
1240 struct rds_conn_path
*cp
= &conn
->c_path
[0];
1242 spin_lock_irqsave(&cp
->cp_lock
, flags
);
1243 if (conn
->c_ping_triggered
) {
1244 spin_unlock_irqrestore(&cp
->cp_lock
, flags
);
1247 conn
->c_ping_triggered
= 1;
1248 spin_unlock_irqrestore(&cp
->cp_lock
, flags
);
1249 rds_send_probe(&conn
->c_path
[0], RDS_FLAG_PROBE_PORT
, 0, 0);