1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* AF_RXRPC sendmsg() implementation.
4 * Copyright (C) 2007, 2016 Red Hat, Inc. All Rights Reserved.
5 * Written by David Howells (dhowells@redhat.com)
8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10 #include <linux/net.h>
11 #include <linux/gfp.h>
12 #include <linux/skbuff.h>
13 #include <linux/export.h>
14 #include <linux/sched/signal.h>
17 #include <net/af_rxrpc.h>
18 #include "ar-internal.h"
21 * Wait for space to appear in the Tx queue or a signal to occur.
23 static int rxrpc_wait_for_tx_window_intr(struct rxrpc_sock *rx,
24 struct rxrpc_call *call,
28 set_current_state(TASK_INTERRUPTIBLE);
29 if (call->tx_top - call->tx_hard_ack <
30 min_t(unsigned int, call->tx_winsize,
31 call->cong_cwnd + call->cong_extra))
34 if (call->state >= RXRPC_CALL_COMPLETE)
37 if (signal_pending(current))
38 return sock_intr_errno(*timeo);
40 trace_rxrpc_transmit(call, rxrpc_transmit_wait);
41 mutex_unlock(&call->user_mutex);
42 *timeo = schedule_timeout(*timeo);
43 if (mutex_lock_interruptible(&call->user_mutex) < 0)
44 return sock_intr_errno(*timeo);
49 * Wait for space to appear in the Tx queue uninterruptibly, but with
50 * a timeout of 2*RTT if no progress was made and a signal occurred.
52 static int rxrpc_wait_for_tx_window_nonintr(struct rxrpc_sock *rx,
53 struct rxrpc_call *call)
55 rxrpc_seq_t tx_start, tx_win;
56 signed long rtt2, timeout;
59 rtt = READ_ONCE(call->peer->rtt);
60 rtt2 = nsecs_to_jiffies64(rtt) * 2;
65 tx_start = READ_ONCE(call->tx_hard_ack);
68 set_current_state(TASK_UNINTERRUPTIBLE);
70 tx_win = READ_ONCE(call->tx_hard_ack);
71 if (call->tx_top - tx_win <
72 min_t(unsigned int, call->tx_winsize,
73 call->cong_cwnd + call->cong_extra))
76 if (call->state >= RXRPC_CALL_COMPLETE)
79 if (test_bit(RXRPC_CALL_IS_INTR, &call->flags) &&
81 tx_win == tx_start && signal_pending(current))
84 if (tx_win != tx_start) {
89 trace_rxrpc_transmit(call, rxrpc_transmit_wait);
90 timeout = schedule_timeout(timeout);
95 * wait for space to appear in the transmit/ACK window
96 * - caller holds the socket locked
98 static int rxrpc_wait_for_tx_window(struct rxrpc_sock *rx,
99 struct rxrpc_call *call,
103 DECLARE_WAITQUEUE(myself, current);
106 _enter(",{%u,%u,%u}",
107 call->tx_hard_ack, call->tx_top, call->tx_winsize);
109 add_wait_queue(&call->waitq, &myself);
112 ret = rxrpc_wait_for_tx_window_nonintr(rx, call);
114 ret = rxrpc_wait_for_tx_window_intr(rx, call, timeo);
116 remove_wait_queue(&call->waitq, &myself);
117 set_current_state(TASK_RUNNING);
118 _leave(" = %d", ret);
123 * Schedule an instant Tx resend.
125 static inline void rxrpc_instant_resend(struct rxrpc_call *call, int ix)
127 spin_lock_bh(&call->lock);
129 if (call->state < RXRPC_CALL_COMPLETE) {
130 call->rxtx_annotations[ix] =
131 (call->rxtx_annotations[ix] & RXRPC_TX_ANNO_LAST) |
132 RXRPC_TX_ANNO_RETRANS;
133 if (!test_and_set_bit(RXRPC_CALL_EV_RESEND, &call->events))
134 rxrpc_queue_call(call);
137 spin_unlock_bh(&call->lock);
141 * Notify the owner of the call that the transmit phase is ended and the last
142 * packet has been queued.
144 static void rxrpc_notify_end_tx(struct rxrpc_sock *rx, struct rxrpc_call *call,
145 rxrpc_notify_end_tx_t notify_end_tx)
148 notify_end_tx(&rx->sk, call, call->user_call_ID);
152 * Queue a DATA packet for transmission, set the resend timeout and send
153 * the packet immediately. Returns the error from rxrpc_send_data_packet()
154 * in case the caller wants to do something with it.
156 static int rxrpc_queue_packet(struct rxrpc_sock *rx, struct rxrpc_call *call,
157 struct sk_buff *skb, bool last,
158 rxrpc_notify_end_tx_t notify_end_tx)
160 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
162 rxrpc_seq_t seq = sp->hdr.seq;
164 u8 annotation = RXRPC_TX_ANNO_UNACK;
166 _net("queue skb %p [%d]", skb, seq);
168 ASSERTCMP(seq, ==, call->tx_top + 1);
171 annotation |= RXRPC_TX_ANNO_LAST;
173 /* We have to set the timestamp before queueing as the retransmit
174 * algorithm can see the packet as soon as we queue it.
176 skb->tstamp = ktime_get_real();
178 ix = seq & RXRPC_RXTX_BUFF_MASK;
179 rxrpc_get_skb(skb, rxrpc_skb_tx_got);
180 call->rxtx_annotations[ix] = annotation;
182 call->rxtx_buffer[ix] = skb;
185 trace_rxrpc_transmit(call, rxrpc_transmit_queue_last);
187 trace_rxrpc_transmit(call, rxrpc_transmit_queue);
189 if (last || call->state == RXRPC_CALL_SERVER_ACK_REQUEST) {
190 _debug("________awaiting reply/ACK__________");
191 write_lock_bh(&call->state_lock);
192 switch (call->state) {
193 case RXRPC_CALL_CLIENT_SEND_REQUEST:
194 call->state = RXRPC_CALL_CLIENT_AWAIT_REPLY;
195 rxrpc_notify_end_tx(rx, call, notify_end_tx);
197 case RXRPC_CALL_SERVER_ACK_REQUEST:
198 call->state = RXRPC_CALL_SERVER_SEND_REPLY;
200 WRITE_ONCE(call->ack_at, now + MAX_JIFFY_OFFSET);
201 if (call->ackr_reason == RXRPC_ACK_DELAY)
202 call->ackr_reason = 0;
203 trace_rxrpc_timer(call, rxrpc_timer_init_for_send_reply, now);
207 case RXRPC_CALL_SERVER_SEND_REPLY:
208 call->state = RXRPC_CALL_SERVER_AWAIT_ACK;
209 rxrpc_notify_end_tx(rx, call, notify_end_tx);
214 write_unlock_bh(&call->state_lock);
217 if (seq == 1 && rxrpc_is_client_call(call))
218 rxrpc_expose_client_call(call);
220 ret = rxrpc_send_data_packet(call, skb, false);
226 rxrpc_set_call_completion(call,
227 RXRPC_CALL_LOCAL_ERROR,
231 _debug("need instant resend %d", ret);
232 rxrpc_instant_resend(call, ix);
234 unsigned long now = jiffies, resend_at;
236 if (call->peer->rtt_usage > 1)
237 resend_at = nsecs_to_jiffies(call->peer->rtt * 3 / 2);
239 resend_at = rxrpc_resend_timeout;
244 WRITE_ONCE(call->resend_at, resend_at);
245 rxrpc_reduce_call_timer(call, resend_at, now,
246 rxrpc_timer_set_for_send);
250 rxrpc_free_skb(skb, rxrpc_skb_tx_freed);
251 _leave(" = %d", ret);
256 * send data through a socket
257 * - must be called in process context
258 * - The caller holds the call user access mutex, but not the socket lock.
260 static int rxrpc_send_data(struct rxrpc_sock *rx,
261 struct rxrpc_call *call,
262 struct msghdr *msg, size_t len,
263 rxrpc_notify_end_tx_t notify_end_tx)
265 struct rxrpc_skb_priv *sp;
267 struct sock *sk = &rx->sk;
272 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
274 /* this should be in poll */
275 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
277 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
280 more = msg->msg_flags & MSG_MORE;
282 if (call->tx_total_len != -1) {
283 if (len > call->tx_total_len)
285 if (!more && len != call->tx_total_len)
289 skb = call->tx_pending;
290 call->tx_pending = NULL;
291 rxrpc_see_skb(skb, rxrpc_skb_tx_seen);
295 /* Check to see if there's a ping ACK to reply to. */
296 if (call->ackr_reason == RXRPC_ACK_PING_RESPONSE)
297 rxrpc_send_ack_packet(call, false, NULL);
300 size_t size, chunk, max, space;
304 if (call->tx_top - call->tx_hard_ack >=
305 min_t(unsigned int, call->tx_winsize,
306 call->cong_cwnd + call->cong_extra)) {
308 if (msg->msg_flags & MSG_DONTWAIT)
310 ret = rxrpc_wait_for_tx_window(rx, call,
312 msg->msg_flags & MSG_WAITALL);
317 max = RXRPC_JUMBO_DATALEN;
318 max -= call->conn->security_size;
319 max &= ~(call->conn->size_align - 1UL);
322 if (chunk > msg_data_left(msg) && !more)
323 chunk = msg_data_left(msg);
325 space = chunk + call->conn->size_align;
326 space &= ~(call->conn->size_align - 1UL);
328 size = space + call->conn->security_size;
330 _debug("SIZE: %zu/%zu/%zu", chunk, space, size);
332 /* create a buffer that we can retain until it's ACK'd */
333 skb = sock_alloc_send_skb(
334 sk, size, msg->msg_flags & MSG_DONTWAIT, &ret);
338 rxrpc_new_skb(skb, rxrpc_skb_tx_new);
340 _debug("ALLOC SEND %p", skb);
342 ASSERTCMP(skb->mark, ==, 0);
344 _debug("HS: %u", call->conn->security_size);
345 skb_reserve(skb, call->conn->security_size);
346 skb->len += call->conn->security_size;
350 if (sp->remain > skb_tailroom(skb))
351 sp->remain = skb_tailroom(skb);
353 _net("skb: hr %d, tr %d, hl %d, rm %d",
359 skb->ip_summed = CHECKSUM_UNNECESSARY;
365 /* append next segment of data to the current buffer */
366 if (msg_data_left(msg) > 0) {
367 int copy = skb_tailroom(skb);
368 ASSERTCMP(copy, >, 0);
369 if (copy > msg_data_left(msg))
370 copy = msg_data_left(msg);
371 if (copy > sp->remain)
375 ret = skb_add_data(skb, &msg->msg_iter, copy);
382 if (call->tx_total_len != -1)
383 call->tx_total_len -= copy;
386 /* check for the far side aborting the call or a network error
388 if (call->state == RXRPC_CALL_COMPLETE)
389 goto call_terminated;
391 /* add the packet to the send queue if it's now full */
392 if (sp->remain <= 0 ||
393 (msg_data_left(msg) == 0 && !more)) {
394 struct rxrpc_connection *conn = call->conn;
398 /* pad out if we're using security */
399 if (conn->security_ix) {
400 pad = conn->security_size + skb->mark;
401 pad = conn->size_align - pad;
402 pad &= conn->size_align - 1;
403 _debug("pad %zu", pad);
405 skb_put_zero(skb, pad);
408 seq = call->tx_top + 1;
412 sp->hdr.flags = conn->out_clientflag;
414 if (msg_data_left(msg) == 0 && !more)
415 sp->hdr.flags |= RXRPC_LAST_PACKET;
416 else if (call->tx_top - call->tx_hard_ack <
418 sp->hdr.flags |= RXRPC_MORE_PACKETS;
420 ret = conn->security->secure_packet(
421 call, skb, skb->mark, skb->head);
425 ret = rxrpc_queue_packet(rx, call, skb,
426 !msg_data_left(msg) && !more,
428 /* Should check for failure here */
431 } while (msg_data_left(msg) > 0);
436 call->tx_pending = skb;
437 _leave(" = %d", ret);
441 rxrpc_free_skb(skb, rxrpc_skb_tx_freed);
442 _leave(" = %d", call->error);
456 * extract control messages from the sendmsg() control buffer
458 static int rxrpc_sendmsg_cmsg(struct msghdr *msg, struct rxrpc_send_params *p)
460 struct cmsghdr *cmsg;
461 bool got_user_ID = false;
464 if (msg->msg_controllen == 0)
467 for_each_cmsghdr(cmsg, msg) {
468 if (!CMSG_OK(msg, cmsg))
471 len = cmsg->cmsg_len - sizeof(struct cmsghdr);
472 _debug("CMSG %d, %d, %d",
473 cmsg->cmsg_level, cmsg->cmsg_type, len);
475 if (cmsg->cmsg_level != SOL_RXRPC)
478 switch (cmsg->cmsg_type) {
479 case RXRPC_USER_CALL_ID:
480 if (msg->msg_flags & MSG_CMSG_COMPAT) {
481 if (len != sizeof(u32))
483 p->call.user_call_ID = *(u32 *)CMSG_DATA(cmsg);
485 if (len != sizeof(unsigned long))
487 p->call.user_call_ID = *(unsigned long *)
494 if (p->command != RXRPC_CMD_SEND_DATA)
496 p->command = RXRPC_CMD_SEND_ABORT;
497 if (len != sizeof(p->abort_code))
499 p->abort_code = *(unsigned int *)CMSG_DATA(cmsg);
500 if (p->abort_code == 0)
505 if (p->command != RXRPC_CMD_SEND_DATA)
507 p->command = RXRPC_CMD_ACCEPT;
512 case RXRPC_EXCLUSIVE_CALL:
518 case RXRPC_UPGRADE_SERVICE:
524 case RXRPC_TX_LENGTH:
525 if (p->call.tx_total_len != -1 || len != sizeof(__s64))
527 p->call.tx_total_len = *(__s64 *)CMSG_DATA(cmsg);
528 if (p->call.tx_total_len < 0)
532 case RXRPC_SET_CALL_TIMEOUT:
533 if (len & 3 || len < 4 || len > 12)
535 memcpy(&p->call.timeouts, CMSG_DATA(cmsg), len);
536 p->call.nr_timeouts = len / 4;
537 if (p->call.timeouts.hard > INT_MAX / HZ)
539 if (p->call.nr_timeouts >= 2 && p->call.timeouts.idle > 60 * 60 * 1000)
541 if (p->call.nr_timeouts >= 3 && p->call.timeouts.normal > 60 * 60 * 1000)
552 if (p->call.tx_total_len != -1 && p->command != RXRPC_CMD_SEND_DATA)
559 * Create a new client call for sendmsg().
560 * - Called with the socket lock held, which it must release.
561 * - If it returns a call, the call's lock will need releasing by the caller.
563 static struct rxrpc_call *
564 rxrpc_new_client_call_for_sendmsg(struct rxrpc_sock *rx, struct msghdr *msg,
565 struct rxrpc_send_params *p)
566 __releases(&rx->sk.sk_lock.slock)
567 __acquires(&call->user_mutex)
569 struct rxrpc_conn_parameters cp;
570 struct rxrpc_call *call;
573 DECLARE_SOCKADDR(struct sockaddr_rxrpc *, srx, msg->msg_name);
577 if (!msg->msg_name) {
578 release_sock(&rx->sk);
579 return ERR_PTR(-EDESTADDRREQ);
583 if (key && !rx->key->payload.data[0])
586 memset(&cp, 0, sizeof(cp));
587 cp.local = rx->local;
589 cp.security_level = rx->min_sec_level;
590 cp.exclusive = rx->exclusive | p->exclusive;
591 cp.upgrade = p->upgrade;
592 cp.service_id = srx->srx_service;
593 call = rxrpc_new_client_call(rx, &cp, srx, &p->call, GFP_KERNEL,
594 atomic_inc_return(&rxrpc_debug_id));
595 /* The socket is now unlocked */
597 rxrpc_put_peer(cp.peer);
598 _leave(" = %p\n", call);
603 * send a message forming part of a client call through an RxRPC socket
604 * - caller holds the socket locked
605 * - the socket may be either a client socket or a server socket
607 int rxrpc_do_sendmsg(struct rxrpc_sock *rx, struct msghdr *msg, size_t len)
608 __releases(&rx->sk.sk_lock.slock)
609 __releases(&call->user_mutex)
611 enum rxrpc_call_state state;
612 struct rxrpc_call *call;
613 unsigned long now, j;
616 struct rxrpc_send_params p = {
617 .call.tx_total_len = -1,
618 .call.user_call_ID = 0,
619 .call.nr_timeouts = 0,
622 .command = RXRPC_CMD_SEND_DATA,
629 ret = rxrpc_sendmsg_cmsg(msg, &p);
631 goto error_release_sock;
633 if (p.command == RXRPC_CMD_ACCEPT) {
635 if (rx->sk.sk_state != RXRPC_SERVER_LISTENING)
636 goto error_release_sock;
637 call = rxrpc_accept_call(rx, p.call.user_call_ID, NULL);
638 /* The socket is now unlocked. */
640 return PTR_ERR(call);
645 call = rxrpc_find_call_by_user_ID(rx, p.call.user_call_ID);
648 if (p.command != RXRPC_CMD_SEND_DATA)
649 goto error_release_sock;
650 call = rxrpc_new_client_call_for_sendmsg(rx, msg, &p);
651 /* The socket is now unlocked... */
653 return PTR_ERR(call);
654 /* ... and we have the call lock. */
656 switch (READ_ONCE(call->state)) {
657 case RXRPC_CALL_UNINITIALISED:
658 case RXRPC_CALL_CLIENT_AWAIT_CONN:
659 case RXRPC_CALL_SERVER_PREALLOC:
660 case RXRPC_CALL_SERVER_SECURING:
661 case RXRPC_CALL_SERVER_ACCEPTING:
663 goto error_release_sock;
668 ret = mutex_lock_interruptible(&call->user_mutex);
669 release_sock(&rx->sk);
675 if (p.call.tx_total_len != -1) {
677 if (call->tx_total_len != -1 ||
681 call->tx_total_len = p.call.tx_total_len;
685 switch (p.call.nr_timeouts) {
687 j = msecs_to_jiffies(p.call.timeouts.normal);
688 if (p.call.timeouts.normal > 0 && j == 0)
690 WRITE_ONCE(call->next_rx_timo, j);
693 j = msecs_to_jiffies(p.call.timeouts.idle);
694 if (p.call.timeouts.idle > 0 && j == 0)
696 WRITE_ONCE(call->next_req_timo, j);
699 if (p.call.timeouts.hard > 0) {
700 j = msecs_to_jiffies(p.call.timeouts.hard);
703 WRITE_ONCE(call->expect_term_by, j);
704 rxrpc_reduce_call_timer(call, j, now,
705 rxrpc_timer_set_for_hard);
710 state = READ_ONCE(call->state);
711 _debug("CALL %d USR %lx ST %d on CONN %p",
712 call->debug_id, call->user_call_ID, state, call->conn);
714 if (state >= RXRPC_CALL_COMPLETE) {
715 /* it's too late for this call */
717 } else if (p.command == RXRPC_CMD_SEND_ABORT) {
719 if (rxrpc_abort_call("CMD", call, 0, p.abort_code, -ECONNABORTED))
720 ret = rxrpc_send_abort_packet(call);
721 } else if (p.command != RXRPC_CMD_SEND_DATA) {
723 } else if (rxrpc_is_client_call(call) &&
724 state != RXRPC_CALL_CLIENT_SEND_REQUEST) {
725 /* request phase complete for this client call */
727 } else if (rxrpc_is_service_call(call) &&
728 state != RXRPC_CALL_SERVER_ACK_REQUEST &&
729 state != RXRPC_CALL_SERVER_SEND_REPLY) {
730 /* Reply phase not begun or not complete for service call. */
733 ret = rxrpc_send_data(rx, call, msg, len, NULL);
737 mutex_unlock(&call->user_mutex);
739 rxrpc_put_call(call, rxrpc_call_put);
740 _leave(" = %d", ret);
744 release_sock(&rx->sk);
749 * rxrpc_kernel_send_data - Allow a kernel service to send data on a call
750 * @sock: The socket the call is on
751 * @call: The call to send data through
752 * @msg: The data to send
753 * @len: The amount of data to send
754 * @notify_end_tx: Notification that the last packet is queued.
756 * Allow a kernel service to send data on a call. The call must be in an state
757 * appropriate to sending data. No control data should be supplied in @msg,
758 * nor should an address be supplied. MSG_MORE should be flagged if there's
759 * more data to come, otherwise this data will end the transmission phase.
761 int rxrpc_kernel_send_data(struct socket *sock, struct rxrpc_call *call,
762 struct msghdr *msg, size_t len,
763 rxrpc_notify_end_tx_t notify_end_tx)
767 _enter("{%d,%s},", call->debug_id, rxrpc_call_states[call->state]);
769 ASSERTCMP(msg->msg_name, ==, NULL);
770 ASSERTCMP(msg->msg_control, ==, NULL);
772 mutex_lock(&call->user_mutex);
774 _debug("CALL %d USR %lx ST %d on CONN %p",
775 call->debug_id, call->user_call_ID, call->state, call->conn);
777 switch (READ_ONCE(call->state)) {
778 case RXRPC_CALL_CLIENT_SEND_REQUEST:
779 case RXRPC_CALL_SERVER_ACK_REQUEST:
780 case RXRPC_CALL_SERVER_SEND_REPLY:
781 ret = rxrpc_send_data(rxrpc_sk(sock->sk), call, msg, len,
784 case RXRPC_CALL_COMPLETE:
785 read_lock_bh(&call->state_lock);
787 read_unlock_bh(&call->state_lock);
790 /* Request phase complete for this client call */
791 trace_rxrpc_rx_eproto(call, 0, tracepoint_string("late_send"));
796 mutex_unlock(&call->user_mutex);
797 _leave(" = %d", ret);
800 EXPORT_SYMBOL(rxrpc_kernel_send_data);
803 * rxrpc_kernel_abort_call - Allow a kernel service to abort a call
804 * @sock: The socket the call is on
805 * @call: The call to be aborted
806 * @abort_code: The abort code to stick into the ABORT packet
807 * @error: Local error value
808 * @why: 3-char string indicating why.
810 * Allow a kernel service to abort a call, if it's still in an abortable state
811 * and return true if the call was aborted, false if it was already complete.
813 bool rxrpc_kernel_abort_call(struct socket *sock, struct rxrpc_call *call,
814 u32 abort_code, int error, const char *why)
818 _enter("{%d},%d,%d,%s", call->debug_id, abort_code, error, why);
820 mutex_lock(&call->user_mutex);
822 aborted = rxrpc_abort_call(why, call, 0, abort_code, error);
824 rxrpc_send_abort_packet(call);
826 mutex_unlock(&call->user_mutex);
829 EXPORT_SYMBOL(rxrpc_kernel_abort_call);
832 * rxrpc_kernel_set_tx_length - Set the total Tx length on a call
833 * @sock: The socket the call is on
834 * @call: The call to be informed
835 * @tx_total_len: The amount of data to be transmitted for this call
837 * Allow a kernel service to set the total transmit length on a call. This
838 * allows buffer-to-packet encrypt-and-copy to be performed.
840 * This function is primarily for use for setting the reply length since the
841 * request length can be set when beginning the call.
843 void rxrpc_kernel_set_tx_length(struct socket *sock, struct rxrpc_call *call,
846 WARN_ON(call->tx_total_len != -1);
847 call->tx_total_len = tx_total_len;
849 EXPORT_SYMBOL(rxrpc_kernel_set_tx_length);