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,
229 rxrpc_notify_socket(call);
232 _debug("need instant resend %d", ret);
233 rxrpc_instant_resend(call, ix);
235 unsigned long now = jiffies, resend_at;
237 if (call->peer->rtt_usage > 1)
238 resend_at = nsecs_to_jiffies(call->peer->rtt * 3 / 2);
240 resend_at = rxrpc_resend_timeout;
245 WRITE_ONCE(call->resend_at, resend_at);
246 rxrpc_reduce_call_timer(call, resend_at, now,
247 rxrpc_timer_set_for_send);
251 rxrpc_free_skb(skb, rxrpc_skb_tx_freed);
252 _leave(" = %d", ret);
257 * send data through a socket
258 * - must be called in process context
259 * - The caller holds the call user access mutex, but not the socket lock.
261 static int rxrpc_send_data(struct rxrpc_sock *rx,
262 struct rxrpc_call *call,
263 struct msghdr *msg, size_t len,
264 rxrpc_notify_end_tx_t notify_end_tx)
266 struct rxrpc_skb_priv *sp;
268 struct sock *sk = &rx->sk;
273 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
275 /* this should be in poll */
276 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
278 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
281 more = msg->msg_flags & MSG_MORE;
283 if (call->tx_total_len != -1) {
284 if (len > call->tx_total_len)
286 if (!more && len != call->tx_total_len)
290 skb = call->tx_pending;
291 call->tx_pending = NULL;
292 rxrpc_see_skb(skb, rxrpc_skb_tx_seen);
296 /* Check to see if there's a ping ACK to reply to. */
297 if (call->ackr_reason == RXRPC_ACK_PING_RESPONSE)
298 rxrpc_send_ack_packet(call, false, NULL);
301 size_t size, chunk, max, space;
305 if (call->tx_top - call->tx_hard_ack >=
306 min_t(unsigned int, call->tx_winsize,
307 call->cong_cwnd + call->cong_extra)) {
309 if (msg->msg_flags & MSG_DONTWAIT)
311 ret = rxrpc_wait_for_tx_window(rx, call,
313 msg->msg_flags & MSG_WAITALL);
318 max = RXRPC_JUMBO_DATALEN;
319 max -= call->conn->security_size;
320 max &= ~(call->conn->size_align - 1UL);
323 if (chunk > msg_data_left(msg) && !more)
324 chunk = msg_data_left(msg);
326 space = chunk + call->conn->size_align;
327 space &= ~(call->conn->size_align - 1UL);
329 size = space + call->conn->security_size;
331 _debug("SIZE: %zu/%zu/%zu", chunk, space, size);
333 /* create a buffer that we can retain until it's ACK'd */
334 skb = sock_alloc_send_skb(
335 sk, size, msg->msg_flags & MSG_DONTWAIT, &ret);
339 rxrpc_new_skb(skb, rxrpc_skb_tx_new);
341 _debug("ALLOC SEND %p", skb);
343 ASSERTCMP(skb->mark, ==, 0);
345 _debug("HS: %u", call->conn->security_size);
346 skb_reserve(skb, call->conn->security_size);
347 skb->len += call->conn->security_size;
351 if (sp->remain > skb_tailroom(skb))
352 sp->remain = skb_tailroom(skb);
354 _net("skb: hr %d, tr %d, hl %d, rm %d",
360 skb->ip_summed = CHECKSUM_UNNECESSARY;
366 /* append next segment of data to the current buffer */
367 if (msg_data_left(msg) > 0) {
368 int copy = skb_tailroom(skb);
369 ASSERTCMP(copy, >, 0);
370 if (copy > msg_data_left(msg))
371 copy = msg_data_left(msg);
372 if (copy > sp->remain)
376 ret = skb_add_data(skb, &msg->msg_iter, copy);
383 if (call->tx_total_len != -1)
384 call->tx_total_len -= copy;
387 /* check for the far side aborting the call or a network error
389 if (call->state == RXRPC_CALL_COMPLETE)
390 goto call_terminated;
392 /* add the packet to the send queue if it's now full */
393 if (sp->remain <= 0 ||
394 (msg_data_left(msg) == 0 && !more)) {
395 struct rxrpc_connection *conn = call->conn;
399 /* pad out if we're using security */
400 if (conn->security_ix) {
401 pad = conn->security_size + skb->mark;
402 pad = conn->size_align - pad;
403 pad &= conn->size_align - 1;
404 _debug("pad %zu", pad);
406 skb_put_zero(skb, pad);
409 seq = call->tx_top + 1;
413 sp->hdr.flags = conn->out_clientflag;
415 if (msg_data_left(msg) == 0 && !more)
416 sp->hdr.flags |= RXRPC_LAST_PACKET;
417 else if (call->tx_top - call->tx_hard_ack <
419 sp->hdr.flags |= RXRPC_MORE_PACKETS;
421 ret = conn->security->secure_packet(
422 call, skb, skb->mark, skb->head);
426 ret = rxrpc_queue_packet(rx, call, skb,
427 !msg_data_left(msg) && !more,
429 /* Should check for failure here */
432 } while (msg_data_left(msg) > 0);
437 call->tx_pending = skb;
438 _leave(" = %d", ret);
442 rxrpc_free_skb(skb, rxrpc_skb_tx_freed);
443 _leave(" = %d", call->error);
457 * extract control messages from the sendmsg() control buffer
459 static int rxrpc_sendmsg_cmsg(struct msghdr *msg, struct rxrpc_send_params *p)
461 struct cmsghdr *cmsg;
462 bool got_user_ID = false;
465 if (msg->msg_controllen == 0)
468 for_each_cmsghdr(cmsg, msg) {
469 if (!CMSG_OK(msg, cmsg))
472 len = cmsg->cmsg_len - sizeof(struct cmsghdr);
473 _debug("CMSG %d, %d, %d",
474 cmsg->cmsg_level, cmsg->cmsg_type, len);
476 if (cmsg->cmsg_level != SOL_RXRPC)
479 switch (cmsg->cmsg_type) {
480 case RXRPC_USER_CALL_ID:
481 if (msg->msg_flags & MSG_CMSG_COMPAT) {
482 if (len != sizeof(u32))
484 p->call.user_call_ID = *(u32 *)CMSG_DATA(cmsg);
486 if (len != sizeof(unsigned long))
488 p->call.user_call_ID = *(unsigned long *)
495 if (p->command != RXRPC_CMD_SEND_DATA)
497 p->command = RXRPC_CMD_SEND_ABORT;
498 if (len != sizeof(p->abort_code))
500 p->abort_code = *(unsigned int *)CMSG_DATA(cmsg);
501 if (p->abort_code == 0)
506 if (p->command != RXRPC_CMD_SEND_DATA)
508 p->command = RXRPC_CMD_ACCEPT;
513 case RXRPC_EXCLUSIVE_CALL:
519 case RXRPC_UPGRADE_SERVICE:
525 case RXRPC_TX_LENGTH:
526 if (p->call.tx_total_len != -1 || len != sizeof(__s64))
528 p->call.tx_total_len = *(__s64 *)CMSG_DATA(cmsg);
529 if (p->call.tx_total_len < 0)
533 case RXRPC_SET_CALL_TIMEOUT:
534 if (len & 3 || len < 4 || len > 12)
536 memcpy(&p->call.timeouts, CMSG_DATA(cmsg), len);
537 p->call.nr_timeouts = len / 4;
538 if (p->call.timeouts.hard > INT_MAX / HZ)
540 if (p->call.nr_timeouts >= 2 && p->call.timeouts.idle > 60 * 60 * 1000)
542 if (p->call.nr_timeouts >= 3 && p->call.timeouts.normal > 60 * 60 * 1000)
553 if (p->call.tx_total_len != -1 && p->command != RXRPC_CMD_SEND_DATA)
560 * Create a new client call for sendmsg().
561 * - Called with the socket lock held, which it must release.
562 * - If it returns a call, the call's lock will need releasing by the caller.
564 static struct rxrpc_call *
565 rxrpc_new_client_call_for_sendmsg(struct rxrpc_sock *rx, struct msghdr *msg,
566 struct rxrpc_send_params *p)
567 __releases(&rx->sk.sk_lock.slock)
568 __acquires(&call->user_mutex)
570 struct rxrpc_conn_parameters cp;
571 struct rxrpc_call *call;
574 DECLARE_SOCKADDR(struct sockaddr_rxrpc *, srx, msg->msg_name);
578 if (!msg->msg_name) {
579 release_sock(&rx->sk);
580 return ERR_PTR(-EDESTADDRREQ);
584 if (key && !rx->key->payload.data[0])
587 memset(&cp, 0, sizeof(cp));
588 cp.local = rx->local;
590 cp.security_level = rx->min_sec_level;
591 cp.exclusive = rx->exclusive | p->exclusive;
592 cp.upgrade = p->upgrade;
593 cp.service_id = srx->srx_service;
594 call = rxrpc_new_client_call(rx, &cp, srx, &p->call, GFP_KERNEL,
595 atomic_inc_return(&rxrpc_debug_id));
596 /* The socket is now unlocked */
598 rxrpc_put_peer(cp.peer);
599 _leave(" = %p\n", call);
604 * send a message forming part of a client call through an RxRPC socket
605 * - caller holds the socket locked
606 * - the socket may be either a client socket or a server socket
608 int rxrpc_do_sendmsg(struct rxrpc_sock *rx, struct msghdr *msg, size_t len)
609 __releases(&rx->sk.sk_lock.slock)
610 __releases(&call->user_mutex)
612 enum rxrpc_call_state state;
613 struct rxrpc_call *call;
614 unsigned long now, j;
617 struct rxrpc_send_params p = {
618 .call.tx_total_len = -1,
619 .call.user_call_ID = 0,
620 .call.nr_timeouts = 0,
623 .command = RXRPC_CMD_SEND_DATA,
630 ret = rxrpc_sendmsg_cmsg(msg, &p);
632 goto error_release_sock;
634 if (p.command == RXRPC_CMD_ACCEPT) {
636 if (rx->sk.sk_state != RXRPC_SERVER_LISTENING)
637 goto error_release_sock;
638 call = rxrpc_accept_call(rx, p.call.user_call_ID, NULL);
639 /* The socket is now unlocked. */
641 return PTR_ERR(call);
646 call = rxrpc_find_call_by_user_ID(rx, p.call.user_call_ID);
649 if (p.command != RXRPC_CMD_SEND_DATA)
650 goto error_release_sock;
651 call = rxrpc_new_client_call_for_sendmsg(rx, msg, &p);
652 /* The socket is now unlocked... */
654 return PTR_ERR(call);
655 /* ... and we have the call lock. */
657 switch (READ_ONCE(call->state)) {
658 case RXRPC_CALL_UNINITIALISED:
659 case RXRPC_CALL_CLIENT_AWAIT_CONN:
660 case RXRPC_CALL_SERVER_PREALLOC:
661 case RXRPC_CALL_SERVER_SECURING:
662 case RXRPC_CALL_SERVER_ACCEPTING:
664 goto error_release_sock;
669 ret = mutex_lock_interruptible(&call->user_mutex);
670 release_sock(&rx->sk);
676 if (p.call.tx_total_len != -1) {
678 if (call->tx_total_len != -1 ||
682 call->tx_total_len = p.call.tx_total_len;
686 switch (p.call.nr_timeouts) {
688 j = msecs_to_jiffies(p.call.timeouts.normal);
689 if (p.call.timeouts.normal > 0 && j == 0)
691 WRITE_ONCE(call->next_rx_timo, j);
694 j = msecs_to_jiffies(p.call.timeouts.idle);
695 if (p.call.timeouts.idle > 0 && j == 0)
697 WRITE_ONCE(call->next_req_timo, j);
700 if (p.call.timeouts.hard > 0) {
701 j = msecs_to_jiffies(p.call.timeouts.hard);
704 WRITE_ONCE(call->expect_term_by, j);
705 rxrpc_reduce_call_timer(call, j, now,
706 rxrpc_timer_set_for_hard);
711 state = READ_ONCE(call->state);
712 _debug("CALL %d USR %lx ST %d on CONN %p",
713 call->debug_id, call->user_call_ID, state, call->conn);
715 if (state >= RXRPC_CALL_COMPLETE) {
716 /* it's too late for this call */
718 } else if (p.command == RXRPC_CMD_SEND_ABORT) {
720 if (rxrpc_abort_call("CMD", call, 0, p.abort_code, -ECONNABORTED))
721 ret = rxrpc_send_abort_packet(call);
722 } else if (p.command != RXRPC_CMD_SEND_DATA) {
724 } else if (rxrpc_is_client_call(call) &&
725 state != RXRPC_CALL_CLIENT_SEND_REQUEST) {
726 /* request phase complete for this client call */
728 } else if (rxrpc_is_service_call(call) &&
729 state != RXRPC_CALL_SERVER_ACK_REQUEST &&
730 state != RXRPC_CALL_SERVER_SEND_REPLY) {
731 /* Reply phase not begun or not complete for service call. */
734 ret = rxrpc_send_data(rx, call, msg, len, NULL);
738 mutex_unlock(&call->user_mutex);
740 rxrpc_put_call(call, rxrpc_call_put);
741 _leave(" = %d", ret);
745 release_sock(&rx->sk);
750 * rxrpc_kernel_send_data - Allow a kernel service to send data on a call
751 * @sock: The socket the call is on
752 * @call: The call to send data through
753 * @msg: The data to send
754 * @len: The amount of data to send
755 * @notify_end_tx: Notification that the last packet is queued.
757 * Allow a kernel service to send data on a call. The call must be in an state
758 * appropriate to sending data. No control data should be supplied in @msg,
759 * nor should an address be supplied. MSG_MORE should be flagged if there's
760 * more data to come, otherwise this data will end the transmission phase.
762 int rxrpc_kernel_send_data(struct socket *sock, struct rxrpc_call *call,
763 struct msghdr *msg, size_t len,
764 rxrpc_notify_end_tx_t notify_end_tx)
768 _enter("{%d,%s},", call->debug_id, rxrpc_call_states[call->state]);
770 ASSERTCMP(msg->msg_name, ==, NULL);
771 ASSERTCMP(msg->msg_control, ==, NULL);
773 mutex_lock(&call->user_mutex);
775 _debug("CALL %d USR %lx ST %d on CONN %p",
776 call->debug_id, call->user_call_ID, call->state, call->conn);
778 switch (READ_ONCE(call->state)) {
779 case RXRPC_CALL_CLIENT_SEND_REQUEST:
780 case RXRPC_CALL_SERVER_ACK_REQUEST:
781 case RXRPC_CALL_SERVER_SEND_REPLY:
782 ret = rxrpc_send_data(rxrpc_sk(sock->sk), call, msg, len,
785 case RXRPC_CALL_COMPLETE:
786 read_lock_bh(&call->state_lock);
788 read_unlock_bh(&call->state_lock);
791 /* Request phase complete for this client call */
792 trace_rxrpc_rx_eproto(call, 0, tracepoint_string("late_send"));
797 mutex_unlock(&call->user_mutex);
798 _leave(" = %d", ret);
801 EXPORT_SYMBOL(rxrpc_kernel_send_data);
804 * rxrpc_kernel_abort_call - Allow a kernel service to abort a call
805 * @sock: The socket the call is on
806 * @call: The call to be aborted
807 * @abort_code: The abort code to stick into the ABORT packet
808 * @error: Local error value
809 * @why: 3-char string indicating why.
811 * Allow a kernel service to abort a call, if it's still in an abortable state
812 * and return true if the call was aborted, false if it was already complete.
814 bool rxrpc_kernel_abort_call(struct socket *sock, struct rxrpc_call *call,
815 u32 abort_code, int error, const char *why)
819 _enter("{%d},%d,%d,%s", call->debug_id, abort_code, error, why);
821 mutex_lock(&call->user_mutex);
823 aborted = rxrpc_abort_call(why, call, 0, abort_code, error);
825 rxrpc_send_abort_packet(call);
827 mutex_unlock(&call->user_mutex);
830 EXPORT_SYMBOL(rxrpc_kernel_abort_call);
833 * rxrpc_kernel_set_tx_length - Set the total Tx length on a call
834 * @sock: The socket the call is on
835 * @call: The call to be informed
836 * @tx_total_len: The amount of data to be transmitted for this call
838 * Allow a kernel service to set the total transmit length on a call. This
839 * allows buffer-to-packet encrypt-and-copy to be performed.
841 * This function is primarily for use for setting the reply length since the
842 * request length can be set when beginning the call.
844 void rxrpc_kernel_set_tx_length(struct socket *sock, struct rxrpc_call *call,
847 WARN_ON(call->tx_total_len != -1);
848 call->tx_total_len = tx_total_len;
850 EXPORT_SYMBOL(rxrpc_kernel_set_tx_length);