1 /* AF_RXRPC sendmsg() implementation.
3 * Copyright (C) 2007, 2016 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public Licence
8 * as published by the Free Software Foundation; either version
9 * 2 of the Licence, or (at your option) any later version.
12 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14 #include <linux/net.h>
15 #include <linux/gfp.h>
16 #include <linux/skbuff.h>
17 #include <linux/export.h>
18 #include <linux/sched/signal.h>
21 #include <net/af_rxrpc.h>
22 #include "ar-internal.h"
25 * Wait for space to appear in the Tx queue or a signal to occur.
27 static int rxrpc_wait_for_tx_window_intr(struct rxrpc_sock *rx,
28 struct rxrpc_call *call,
32 set_current_state(TASK_INTERRUPTIBLE);
33 if (call->tx_top - call->tx_hard_ack <
34 min_t(unsigned int, call->tx_winsize,
35 call->cong_cwnd + call->cong_extra))
38 if (call->state >= RXRPC_CALL_COMPLETE)
41 if (signal_pending(current))
42 return sock_intr_errno(*timeo);
44 trace_rxrpc_transmit(call, rxrpc_transmit_wait);
45 mutex_unlock(&call->user_mutex);
46 *timeo = schedule_timeout(*timeo);
47 if (mutex_lock_interruptible(&call->user_mutex) < 0)
48 return sock_intr_errno(*timeo);
53 * Wait for space to appear in the Tx queue uninterruptibly, but with
54 * a timeout of 2*RTT if no progress was made and a signal occurred.
56 static int rxrpc_wait_for_tx_window_nonintr(struct rxrpc_sock *rx,
57 struct rxrpc_call *call)
59 rxrpc_seq_t tx_start, tx_win;
60 signed long rtt2, timeout;
63 rtt = READ_ONCE(call->peer->rtt);
64 rtt2 = nsecs_to_jiffies64(rtt) * 2;
69 tx_start = READ_ONCE(call->tx_hard_ack);
72 set_current_state(TASK_UNINTERRUPTIBLE);
74 tx_win = READ_ONCE(call->tx_hard_ack);
75 if (call->tx_top - tx_win <
76 min_t(unsigned int, call->tx_winsize,
77 call->cong_cwnd + call->cong_extra))
80 if (call->state >= RXRPC_CALL_COMPLETE)
84 tx_win == tx_start && signal_pending(current))
87 if (tx_win != tx_start) {
92 trace_rxrpc_transmit(call, rxrpc_transmit_wait);
93 timeout = schedule_timeout(timeout);
98 * wait for space to appear in the transmit/ACK window
99 * - caller holds the socket locked
101 static int rxrpc_wait_for_tx_window(struct rxrpc_sock *rx,
102 struct rxrpc_call *call,
106 DECLARE_WAITQUEUE(myself, current);
109 _enter(",{%u,%u,%u}",
110 call->tx_hard_ack, call->tx_top, call->tx_winsize);
112 add_wait_queue(&call->waitq, &myself);
115 ret = rxrpc_wait_for_tx_window_nonintr(rx, call);
117 ret = rxrpc_wait_for_tx_window_intr(rx, call, timeo);
119 remove_wait_queue(&call->waitq, &myself);
120 set_current_state(TASK_RUNNING);
121 _leave(" = %d", ret);
126 * Schedule an instant Tx resend.
128 static inline void rxrpc_instant_resend(struct rxrpc_call *call, int ix)
130 spin_lock_bh(&call->lock);
132 if (call->state < RXRPC_CALL_COMPLETE) {
133 call->rxtx_annotations[ix] =
134 (call->rxtx_annotations[ix] & RXRPC_TX_ANNO_LAST) |
135 RXRPC_TX_ANNO_RETRANS;
136 if (!test_and_set_bit(RXRPC_CALL_EV_RESEND, &call->events))
137 rxrpc_queue_call(call);
140 spin_unlock_bh(&call->lock);
144 * Notify the owner of the call that the transmit phase is ended and the last
145 * packet has been queued.
147 static void rxrpc_notify_end_tx(struct rxrpc_sock *rx, struct rxrpc_call *call,
148 rxrpc_notify_end_tx_t notify_end_tx)
151 notify_end_tx(&rx->sk, call, call->user_call_ID);
155 * Queue a DATA packet for transmission, set the resend timeout and send the
158 static void rxrpc_queue_packet(struct rxrpc_sock *rx, struct rxrpc_call *call,
159 struct sk_buff *skb, bool last,
160 rxrpc_notify_end_tx_t notify_end_tx)
162 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
164 rxrpc_seq_t seq = sp->hdr.seq;
166 u8 annotation = RXRPC_TX_ANNO_UNACK;
168 _net("queue skb %p [%d]", skb, seq);
170 ASSERTCMP(seq, ==, call->tx_top + 1);
173 annotation |= RXRPC_TX_ANNO_LAST;
174 set_bit(RXRPC_CALL_TX_LASTQ, &call->flags);
177 /* We have to set the timestamp before queueing as the retransmit
178 * algorithm can see the packet as soon as we queue it.
180 skb->tstamp = ktime_get_real();
182 ix = seq & RXRPC_RXTX_BUFF_MASK;
183 rxrpc_get_skb(skb, rxrpc_skb_tx_got);
184 call->rxtx_annotations[ix] = annotation;
186 call->rxtx_buffer[ix] = skb;
189 trace_rxrpc_transmit(call, rxrpc_transmit_queue_last);
191 trace_rxrpc_transmit(call, rxrpc_transmit_queue);
193 if (last || call->state == RXRPC_CALL_SERVER_ACK_REQUEST) {
194 _debug("________awaiting reply/ACK__________");
195 write_lock_bh(&call->state_lock);
196 switch (call->state) {
197 case RXRPC_CALL_CLIENT_SEND_REQUEST:
198 call->state = RXRPC_CALL_CLIENT_AWAIT_REPLY;
199 rxrpc_notify_end_tx(rx, call, notify_end_tx);
201 case RXRPC_CALL_SERVER_ACK_REQUEST:
202 call->state = RXRPC_CALL_SERVER_SEND_REPLY;
204 WRITE_ONCE(call->ack_at, now + MAX_JIFFY_OFFSET);
205 if (call->ackr_reason == RXRPC_ACK_DELAY)
206 call->ackr_reason = 0;
207 trace_rxrpc_timer(call, rxrpc_timer_init_for_send_reply, now);
211 case RXRPC_CALL_SERVER_SEND_REPLY:
212 call->state = RXRPC_CALL_SERVER_AWAIT_ACK;
213 rxrpc_notify_end_tx(rx, call, notify_end_tx);
218 write_unlock_bh(&call->state_lock);
221 if (seq == 1 && rxrpc_is_client_call(call))
222 rxrpc_expose_client_call(call);
224 ret = rxrpc_send_data_packet(call, skb, false);
230 rxrpc_set_call_completion(call,
231 RXRPC_CALL_LOCAL_ERROR,
233 rxrpc_notify_socket(call);
236 _debug("need instant resend %d", ret);
237 rxrpc_instant_resend(call, ix);
239 unsigned long now = jiffies, resend_at;
241 if (call->peer->rtt_usage > 1)
242 resend_at = nsecs_to_jiffies(call->peer->rtt * 3 / 2);
244 resend_at = rxrpc_resend_timeout;
249 WRITE_ONCE(call->resend_at, resend_at);
250 rxrpc_reduce_call_timer(call, resend_at, now,
251 rxrpc_timer_set_for_send);
255 rxrpc_free_skb(skb, rxrpc_skb_tx_freed);
260 * send data through a socket
261 * - must be called in process context
262 * - The caller holds the call user access mutex, but not the socket lock.
264 static int rxrpc_send_data(struct rxrpc_sock *rx,
265 struct rxrpc_call *call,
266 struct msghdr *msg, size_t len,
267 rxrpc_notify_end_tx_t notify_end_tx)
269 struct rxrpc_skb_priv *sp;
271 struct sock *sk = &rx->sk;
276 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
278 /* this should be in poll */
279 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
281 if (sk->sk_shutdown & SEND_SHUTDOWN)
284 more = msg->msg_flags & MSG_MORE;
286 if (call->tx_total_len != -1) {
287 if (len > call->tx_total_len)
289 if (!more && len != call->tx_total_len)
293 skb = call->tx_pending;
294 call->tx_pending = NULL;
295 rxrpc_see_skb(skb, rxrpc_skb_tx_seen);
299 /* Check to see if there's a ping ACK to reply to. */
300 if (call->ackr_reason == RXRPC_ACK_PING_RESPONSE)
301 rxrpc_send_ack_packet(call, false, NULL);
304 size_t size, chunk, max, space;
308 if (call->tx_top - call->tx_hard_ack >=
309 min_t(unsigned int, call->tx_winsize,
310 call->cong_cwnd + call->cong_extra)) {
312 if (msg->msg_flags & MSG_DONTWAIT)
314 ret = rxrpc_wait_for_tx_window(rx, call,
316 msg->msg_flags & MSG_WAITALL);
321 max = RXRPC_JUMBO_DATALEN;
322 max -= call->conn->security_size;
323 max &= ~(call->conn->size_align - 1UL);
326 if (chunk > msg_data_left(msg) && !more)
327 chunk = msg_data_left(msg);
329 space = chunk + call->conn->size_align;
330 space &= ~(call->conn->size_align - 1UL);
332 size = space + call->conn->security_size;
334 _debug("SIZE: %zu/%zu/%zu", chunk, space, size);
336 /* create a buffer that we can retain until it's ACK'd */
337 skb = sock_alloc_send_skb(
338 sk, size, msg->msg_flags & MSG_DONTWAIT, &ret);
342 rxrpc_new_skb(skb, rxrpc_skb_tx_new);
344 _debug("ALLOC SEND %p", skb);
346 ASSERTCMP(skb->mark, ==, 0);
348 _debug("HS: %u", call->conn->security_size);
349 skb_reserve(skb, call->conn->security_size);
350 skb->len += call->conn->security_size;
354 if (sp->remain > skb_tailroom(skb))
355 sp->remain = skb_tailroom(skb);
357 _net("skb: hr %d, tr %d, hl %d, rm %d",
363 skb->ip_summed = CHECKSUM_UNNECESSARY;
369 /* append next segment of data to the current buffer */
370 if (msg_data_left(msg) > 0) {
371 int copy = skb_tailroom(skb);
372 ASSERTCMP(copy, >, 0);
373 if (copy > msg_data_left(msg))
374 copy = msg_data_left(msg);
375 if (copy > sp->remain)
379 ret = skb_add_data(skb, &msg->msg_iter, copy);
386 if (call->tx_total_len != -1)
387 call->tx_total_len -= copy;
390 /* add the packet to the send queue if it's now full */
391 if (sp->remain <= 0 ||
392 (msg_data_left(msg) == 0 && !more)) {
393 struct rxrpc_connection *conn = call->conn;
397 /* pad out if we're using security */
398 if (conn->security_ix) {
399 pad = conn->security_size + skb->mark;
400 pad = conn->size_align - pad;
401 pad &= conn->size_align - 1;
402 _debug("pad %zu", pad);
404 skb_put_zero(skb, pad);
407 seq = call->tx_top + 1;
411 sp->hdr.flags = conn->out_clientflag;
413 if (msg_data_left(msg) == 0 && !more)
414 sp->hdr.flags |= RXRPC_LAST_PACKET;
415 else if (call->tx_top - call->tx_hard_ack <
417 sp->hdr.flags |= RXRPC_MORE_PACKETS;
419 ret = conn->security->secure_packet(
420 call, skb, skb->mark, skb->head);
424 rxrpc_queue_packet(rx, call, skb,
425 !msg_data_left(msg) && !more,
430 /* Check for the far side aborting the call or a network error
431 * occurring. If this happens, save any packet that was under
432 * construction so that in the case of a network error, the
433 * call can be retried or redirected.
435 if (call->state == RXRPC_CALL_COMPLETE) {
439 } while (msg_data_left(msg) > 0);
443 if (READ_ONCE(call->state) == RXRPC_CALL_COMPLETE) {
444 read_lock_bh(&call->state_lock);
447 read_unlock_bh(&call->state_lock);
450 call->tx_pending = skb;
451 _leave(" = %d", ret);
465 * extract control messages from the sendmsg() control buffer
467 static int rxrpc_sendmsg_cmsg(struct msghdr *msg, struct rxrpc_send_params *p)
469 struct cmsghdr *cmsg;
470 bool got_user_ID = false;
473 if (msg->msg_controllen == 0)
476 for_each_cmsghdr(cmsg, msg) {
477 if (!CMSG_OK(msg, cmsg))
480 len = cmsg->cmsg_len - sizeof(struct cmsghdr);
481 _debug("CMSG %d, %d, %d",
482 cmsg->cmsg_level, cmsg->cmsg_type, len);
484 if (cmsg->cmsg_level != SOL_RXRPC)
487 switch (cmsg->cmsg_type) {
488 case RXRPC_USER_CALL_ID:
489 if (msg->msg_flags & MSG_CMSG_COMPAT) {
490 if (len != sizeof(u32))
492 p->call.user_call_ID = *(u32 *)CMSG_DATA(cmsg);
494 if (len != sizeof(unsigned long))
496 p->call.user_call_ID = *(unsigned long *)
503 if (p->command != RXRPC_CMD_SEND_DATA)
505 p->command = RXRPC_CMD_SEND_ABORT;
506 if (len != sizeof(p->abort_code))
508 p->abort_code = *(unsigned int *)CMSG_DATA(cmsg);
509 if (p->abort_code == 0)
514 if (p->command != RXRPC_CMD_SEND_DATA)
516 p->command = RXRPC_CMD_ACCEPT;
521 case RXRPC_EXCLUSIVE_CALL:
527 case RXRPC_UPGRADE_SERVICE:
533 case RXRPC_TX_LENGTH:
534 if (p->call.tx_total_len != -1 || len != sizeof(__s64))
536 p->call.tx_total_len = *(__s64 *)CMSG_DATA(cmsg);
537 if (p->call.tx_total_len < 0)
541 case RXRPC_SET_CALL_TIMEOUT:
542 if (len & 3 || len < 4 || len > 12)
544 memcpy(&p->call.timeouts, CMSG_DATA(cmsg), len);
545 p->call.nr_timeouts = len / 4;
546 if (p->call.timeouts.hard > INT_MAX / HZ)
548 if (p->call.nr_timeouts >= 2 && p->call.timeouts.idle > 60 * 60 * 1000)
550 if (p->call.nr_timeouts >= 3 && p->call.timeouts.normal > 60 * 60 * 1000)
561 if (p->call.tx_total_len != -1 && p->command != RXRPC_CMD_SEND_DATA)
568 * Create a new client call for sendmsg().
569 * - Called with the socket lock held, which it must release.
570 * - If it returns a call, the call's lock will need releasing by the caller.
572 static struct rxrpc_call *
573 rxrpc_new_client_call_for_sendmsg(struct rxrpc_sock *rx, struct msghdr *msg,
574 struct rxrpc_send_params *p)
575 __releases(&rx->sk.sk_lock.slock)
576 __acquires(&call->user_mutex)
578 struct rxrpc_conn_parameters cp;
579 struct rxrpc_call *call;
582 DECLARE_SOCKADDR(struct sockaddr_rxrpc *, srx, msg->msg_name);
586 if (!msg->msg_name) {
587 release_sock(&rx->sk);
588 return ERR_PTR(-EDESTADDRREQ);
592 if (key && !rx->key->payload.data[0])
595 memset(&cp, 0, sizeof(cp));
596 cp.local = rx->local;
598 cp.security_level = rx->min_sec_level;
599 cp.exclusive = rx->exclusive | p->exclusive;
600 cp.upgrade = p->upgrade;
601 cp.service_id = srx->srx_service;
602 call = rxrpc_new_client_call(rx, &cp, srx, &p->call, GFP_KERNEL,
603 atomic_inc_return(&rxrpc_debug_id));
604 /* The socket is now unlocked */
606 rxrpc_put_peer(cp.peer);
607 _leave(" = %p\n", call);
612 * send a message forming part of a client call through an RxRPC socket
613 * - caller holds the socket locked
614 * - the socket may be either a client socket or a server socket
616 int rxrpc_do_sendmsg(struct rxrpc_sock *rx, struct msghdr *msg, size_t len)
617 __releases(&rx->sk.sk_lock.slock)
618 __releases(&call->user_mutex)
620 enum rxrpc_call_state state;
621 struct rxrpc_call *call;
622 unsigned long now, j;
625 struct rxrpc_send_params p = {
626 .call.tx_total_len = -1,
627 .call.user_call_ID = 0,
628 .call.nr_timeouts = 0,
630 .command = RXRPC_CMD_SEND_DATA,
637 ret = rxrpc_sendmsg_cmsg(msg, &p);
639 goto error_release_sock;
641 if (p.command == RXRPC_CMD_ACCEPT) {
643 if (rx->sk.sk_state != RXRPC_SERVER_LISTENING)
644 goto error_release_sock;
645 call = rxrpc_accept_call(rx, p.call.user_call_ID, NULL);
646 /* The socket is now unlocked. */
648 return PTR_ERR(call);
653 call = rxrpc_find_call_by_user_ID(rx, p.call.user_call_ID);
656 if (p.command != RXRPC_CMD_SEND_DATA)
657 goto error_release_sock;
658 call = rxrpc_new_client_call_for_sendmsg(rx, msg, &p);
659 /* The socket is now unlocked... */
661 return PTR_ERR(call);
662 /* ... and we have the call lock. */
664 if (READ_ONCE(call->state) == RXRPC_CALL_COMPLETE)
667 switch (READ_ONCE(call->state)) {
668 case RXRPC_CALL_UNINITIALISED:
669 case RXRPC_CALL_CLIENT_AWAIT_CONN:
670 case RXRPC_CALL_SERVER_PREALLOC:
671 case RXRPC_CALL_SERVER_SECURING:
672 case RXRPC_CALL_SERVER_ACCEPTING:
673 rxrpc_put_call(call, rxrpc_call_put);
675 goto error_release_sock;
680 ret = mutex_lock_interruptible(&call->user_mutex);
681 release_sock(&rx->sk);
687 if (p.call.tx_total_len != -1) {
689 if (call->tx_total_len != -1 ||
693 call->tx_total_len = p.call.tx_total_len;
697 switch (p.call.nr_timeouts) {
699 j = msecs_to_jiffies(p.call.timeouts.normal);
700 if (p.call.timeouts.normal > 0 && j == 0)
702 WRITE_ONCE(call->next_rx_timo, j);
705 j = msecs_to_jiffies(p.call.timeouts.idle);
706 if (p.call.timeouts.idle > 0 && j == 0)
708 WRITE_ONCE(call->next_req_timo, j);
711 if (p.call.timeouts.hard > 0) {
712 j = p.call.timeouts.hard * HZ;
715 WRITE_ONCE(call->expect_term_by, j);
716 rxrpc_reduce_call_timer(call, j, now,
717 rxrpc_timer_set_for_hard);
722 state = READ_ONCE(call->state);
723 _debug("CALL %d USR %lx ST %d on CONN %p",
724 call->debug_id, call->user_call_ID, state, call->conn);
726 if (state >= RXRPC_CALL_COMPLETE) {
727 /* it's too late for this call */
729 } else if (p.command == RXRPC_CMD_SEND_ABORT) {
731 if (rxrpc_abort_call("CMD", call, 0, p.abort_code, -ECONNABORTED))
732 ret = rxrpc_send_abort_packet(call);
733 } else if (p.command != RXRPC_CMD_SEND_DATA) {
735 } else if (rxrpc_is_client_call(call) &&
736 state != RXRPC_CALL_CLIENT_SEND_REQUEST) {
737 /* request phase complete for this client call */
739 } else if (rxrpc_is_service_call(call) &&
740 state != RXRPC_CALL_SERVER_ACK_REQUEST &&
741 state != RXRPC_CALL_SERVER_SEND_REPLY) {
742 /* Reply phase not begun or not complete for service call. */
745 ret = rxrpc_send_data(rx, call, msg, len, NULL);
749 mutex_unlock(&call->user_mutex);
751 rxrpc_put_call(call, rxrpc_call_put);
752 _leave(" = %d", ret);
756 release_sock(&rx->sk);
761 * rxrpc_kernel_send_data - Allow a kernel service to send data on a call
762 * @sock: The socket the call is on
763 * @call: The call to send data through
764 * @msg: The data to send
765 * @len: The amount of data to send
766 * @notify_end_tx: Notification that the last packet is queued.
768 * Allow a kernel service to send data on a call. The call must be in an state
769 * appropriate to sending data. No control data should be supplied in @msg,
770 * nor should an address be supplied. MSG_MORE should be flagged if there's
771 * more data to come, otherwise this data will end the transmission phase.
773 int rxrpc_kernel_send_data(struct socket *sock, struct rxrpc_call *call,
774 struct msghdr *msg, size_t len,
775 rxrpc_notify_end_tx_t notify_end_tx)
779 _enter("{%d,%s},", call->debug_id, rxrpc_call_states[call->state]);
781 ASSERTCMP(msg->msg_name, ==, NULL);
782 ASSERTCMP(msg->msg_control, ==, NULL);
784 mutex_lock(&call->user_mutex);
786 _debug("CALL %d USR %lx ST %d on CONN %p",
787 call->debug_id, call->user_call_ID, call->state, call->conn);
789 switch (READ_ONCE(call->state)) {
790 case RXRPC_CALL_CLIENT_SEND_REQUEST:
791 case RXRPC_CALL_SERVER_ACK_REQUEST:
792 case RXRPC_CALL_SERVER_SEND_REPLY:
793 ret = rxrpc_send_data(rxrpc_sk(sock->sk), call, msg, len,
796 case RXRPC_CALL_COMPLETE:
797 read_lock_bh(&call->state_lock);
799 read_unlock_bh(&call->state_lock);
802 /* Request phase complete for this client call */
803 trace_rxrpc_rx_eproto(call, 0, tracepoint_string("late_send"));
808 mutex_unlock(&call->user_mutex);
809 _leave(" = %d", ret);
812 EXPORT_SYMBOL(rxrpc_kernel_send_data);
815 * rxrpc_kernel_abort_call - Allow a kernel service to abort a call
816 * @sock: The socket the call is on
817 * @call: The call to be aborted
818 * @abort_code: The abort code to stick into the ABORT packet
819 * @error: Local error value
820 * @why: 3-char string indicating why.
822 * Allow a kernel service to abort a call, if it's still in an abortable state
823 * and return true if the call was aborted, false if it was already complete.
825 bool rxrpc_kernel_abort_call(struct socket *sock, struct rxrpc_call *call,
826 u32 abort_code, int error, const char *why)
830 _enter("{%d},%d,%d,%s", call->debug_id, abort_code, error, why);
832 mutex_lock(&call->user_mutex);
834 aborted = rxrpc_abort_call(why, call, 0, abort_code, error);
836 rxrpc_send_abort_packet(call);
838 mutex_unlock(&call->user_mutex);
841 EXPORT_SYMBOL(rxrpc_kernel_abort_call);
844 * rxrpc_kernel_set_tx_length - Set the total Tx length on a call
845 * @sock: The socket the call is on
846 * @call: The call to be informed
847 * @tx_total_len: The amount of data to be transmitted for this call
849 * Allow a kernel service to set the total transmit length on a call. This
850 * allows buffer-to-packet encrypt-and-copy to be performed.
852 * This function is primarily for use for setting the reply length since the
853 * request length can be set when beginning the call.
855 void rxrpc_kernel_set_tx_length(struct socket *sock, struct rxrpc_call *call,
858 WARN_ON(call->tx_total_len != -1);
859 call->tx_total_len = tx_total_len;
861 EXPORT_SYMBOL(rxrpc_kernel_set_tx_length);