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 * Propose an abort to be made in the I/O thread.
23 bool rxrpc_propose_abort(struct rxrpc_call *call, s32 abort_code, int error,
24 enum rxrpc_abort_reason why)
26 _enter("{%d},%d,%d,%u", call->debug_id, abort_code, error, why);
28 if (!call->send_abort && !rxrpc_call_is_complete(call)) {
29 call->send_abort_why = why;
30 call->send_abort_err = error;
31 call->send_abort_seq = 0;
32 /* Request abort locklessly vs rxrpc_input_call_event(). */
33 smp_store_release(&call->send_abort, abort_code);
34 rxrpc_poke_call(call, rxrpc_call_poke_abort);
42 * Wait for a call to become connected. Interruption here doesn't cause the
45 static int rxrpc_wait_to_be_connected(struct rxrpc_call *call, long *timeo)
47 DECLARE_WAITQUEUE(myself, current);
50 _enter("%d", call->debug_id);
52 if (rxrpc_call_state(call) != RXRPC_CALL_CLIENT_AWAIT_CONN)
55 add_wait_queue_exclusive(&call->waitq, &myself);
58 switch (call->interruptibility) {
59 case RXRPC_INTERRUPTIBLE:
60 case RXRPC_PREINTERRUPTIBLE:
61 set_current_state(TASK_INTERRUPTIBLE);
63 case RXRPC_UNINTERRUPTIBLE:
65 set_current_state(TASK_UNINTERRUPTIBLE);
69 if (rxrpc_call_state(call) != RXRPC_CALL_CLIENT_AWAIT_CONN)
71 if ((call->interruptibility == RXRPC_INTERRUPTIBLE ||
72 call->interruptibility == RXRPC_PREINTERRUPTIBLE) &&
73 signal_pending(current)) {
74 ret = sock_intr_errno(*timeo);
77 *timeo = schedule_timeout(*timeo);
80 remove_wait_queue(&call->waitq, &myself);
81 __set_current_state(TASK_RUNNING);
84 if (ret == 0 && rxrpc_call_is_complete(call))
92 * Return true if there's sufficient Tx queue space.
94 static bool rxrpc_check_tx_space(struct rxrpc_call *call, rxrpc_seq_t *_tx_win)
97 *_tx_win = call->tx_bottom;
98 return call->tx_prepared - call->tx_bottom < 256;
102 * Wait for space to appear in the Tx queue or a signal to occur.
104 static int rxrpc_wait_for_tx_window_intr(struct rxrpc_sock *rx,
105 struct rxrpc_call *call,
109 set_current_state(TASK_INTERRUPTIBLE);
110 if (rxrpc_check_tx_space(call, NULL))
113 if (rxrpc_call_is_complete(call))
116 if (signal_pending(current))
117 return sock_intr_errno(*timeo);
119 trace_rxrpc_txqueue(call, rxrpc_txqueue_wait);
120 *timeo = schedule_timeout(*timeo);
125 * Wait for space to appear in the Tx queue uninterruptibly, but with
126 * a timeout of 2*RTT if no progress was made and a signal occurred.
128 static int rxrpc_wait_for_tx_window_waitall(struct rxrpc_sock *rx,
129 struct rxrpc_call *call)
131 rxrpc_seq_t tx_start, tx_win;
132 signed long rtt, timeout;
134 rtt = READ_ONCE(call->peer->srtt_us) >> 3;
135 rtt = usecs_to_jiffies(rtt) * 2;
140 tx_start = smp_load_acquire(&call->acks_hard_ack);
143 set_current_state(TASK_UNINTERRUPTIBLE);
145 if (rxrpc_check_tx_space(call, &tx_win))
148 if (rxrpc_call_is_complete(call))
152 tx_win == tx_start && signal_pending(current))
155 if (tx_win != tx_start) {
160 trace_rxrpc_txqueue(call, rxrpc_txqueue_wait);
161 timeout = schedule_timeout(timeout);
166 * Wait for space to appear in the Tx queue uninterruptibly.
168 static int rxrpc_wait_for_tx_window_nonintr(struct rxrpc_sock *rx,
169 struct rxrpc_call *call,
173 set_current_state(TASK_UNINTERRUPTIBLE);
174 if (rxrpc_check_tx_space(call, NULL))
177 if (rxrpc_call_is_complete(call))
180 trace_rxrpc_txqueue(call, rxrpc_txqueue_wait);
181 *timeo = schedule_timeout(*timeo);
186 * wait for space to appear in the transmit/ACK window
187 * - caller holds the socket locked
189 static int rxrpc_wait_for_tx_window(struct rxrpc_sock *rx,
190 struct rxrpc_call *call,
194 DECLARE_WAITQUEUE(myself, current);
197 _enter(",{%u,%u,%u,%u}",
198 call->tx_bottom, call->acks_hard_ack, call->tx_top, call->tx_winsize);
200 add_wait_queue(&call->waitq, &myself);
202 switch (call->interruptibility) {
203 case RXRPC_INTERRUPTIBLE:
205 ret = rxrpc_wait_for_tx_window_waitall(rx, call);
207 ret = rxrpc_wait_for_tx_window_intr(rx, call, timeo);
209 case RXRPC_PREINTERRUPTIBLE:
210 case RXRPC_UNINTERRUPTIBLE:
212 ret = rxrpc_wait_for_tx_window_nonintr(rx, call, timeo);
216 remove_wait_queue(&call->waitq, &myself);
217 set_current_state(TASK_RUNNING);
218 _leave(" = %d", ret);
223 * Notify the owner of the call that the transmit phase is ended and the last
224 * packet has been queued.
226 static void rxrpc_notify_end_tx(struct rxrpc_sock *rx, struct rxrpc_call *call,
227 rxrpc_notify_end_tx_t notify_end_tx)
230 notify_end_tx(&rx->sk, call, call->user_call_ID);
234 * Queue a DATA packet for transmission, set the resend timeout and send
235 * the packet immediately. Returns the error from rxrpc_send_data_packet()
236 * in case the caller wants to do something with it.
238 static void rxrpc_queue_packet(struct rxrpc_sock *rx, struct rxrpc_call *call,
239 struct rxrpc_txbuf *txb,
240 rxrpc_notify_end_tx_t notify_end_tx)
242 rxrpc_seq_t seq = txb->seq;
243 bool poke, last = txb->flags & RXRPC_LAST_PACKET;
245 rxrpc_inc_stat(call->rxnet, stat_tx_data);
247 ASSERTCMP(txb->seq, ==, call->tx_prepared + 1);
249 /* We have to set the timestamp before queueing as the retransmit
250 * algorithm can see the packet as soon as we queue it.
252 txb->last_sent = ktime_get_real();
255 trace_rxrpc_txqueue(call, rxrpc_txqueue_queue_last);
257 trace_rxrpc_txqueue(call, rxrpc_txqueue_queue);
259 /* Add the packet to the call's output buffer */
260 spin_lock(&call->tx_lock);
261 poke = list_empty(&call->tx_sendmsg);
262 list_add_tail(&txb->call_link, &call->tx_sendmsg);
263 call->tx_prepared = seq;
265 rxrpc_notify_end_tx(rx, call, notify_end_tx);
266 spin_unlock(&call->tx_lock);
269 rxrpc_poke_call(call, rxrpc_call_poke_start);
273 * send data through a socket
274 * - must be called in process context
275 * - The caller holds the call user access mutex, but not the socket lock.
277 static int rxrpc_send_data(struct rxrpc_sock *rx,
278 struct rxrpc_call *call,
279 struct msghdr *msg, size_t len,
280 rxrpc_notify_end_tx_t notify_end_tx,
283 struct rxrpc_txbuf *txb;
284 struct sock *sk = &rx->sk;
285 enum rxrpc_call_state state;
287 bool more = msg->msg_flags & MSG_MORE;
290 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
292 ret = rxrpc_wait_to_be_connected(call, &timeo);
296 if (call->conn->state == RXRPC_CONN_CLIENT_UNSECURED) {
297 ret = rxrpc_init_client_conn_security(call->conn);
302 /* this should be in poll */
303 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
307 if (sk->sk_shutdown & SEND_SHUTDOWN)
309 state = rxrpc_call_state(call);
311 if (state >= RXRPC_CALL_COMPLETE)
314 if (state != RXRPC_CALL_CLIENT_SEND_REQUEST &&
315 state != RXRPC_CALL_SERVER_ACK_REQUEST &&
316 state != RXRPC_CALL_SERVER_SEND_REPLY) {
317 /* Request phase complete for this client call */
318 trace_rxrpc_abort(call->debug_id, rxrpc_sendmsg_late_send,
319 call->cid, call->call_id, call->rx_consumed,
325 if (call->tx_total_len != -1) {
326 if (len - copied > call->tx_total_len)
328 if (!more && len - copied != call->tx_total_len)
332 txb = call->tx_pending;
333 call->tx_pending = NULL;
335 rxrpc_see_txbuf(txb, rxrpc_txbuf_see_send_more);
343 if (!rxrpc_check_tx_space(call, NULL))
346 /* Work out the maximum size of a packet. Assume that
347 * the security header is going to be in the padded
348 * region (enc blocksize), but the trailer is not.
350 remain = more ? INT_MAX : msg_data_left(msg);
351 txb = call->conn->security->alloc_txbuf(call, remain, sk->sk_allocation);
360 /* append next segment of data to the current buffer */
361 if (msg_data_left(msg) > 0) {
362 size_t copy = min_t(size_t, txb->space, msg_data_left(msg));
364 _debug("add %zu", copy);
365 if (!copy_from_iter_full(txb->kvec[0].iov_base + txb->offset,
366 copy, &msg->msg_iter))
373 if (call->tx_total_len != -1)
374 call->tx_total_len -= copy;
377 /* check for the far side aborting the call or a network error
379 if (rxrpc_call_is_complete(call))
380 goto call_terminated;
382 /* add the packet to the send queue if it's now full */
384 (msg_data_left(msg) == 0 && !more)) {
385 if (msg_data_left(msg) == 0 && !more)
386 txb->flags |= RXRPC_LAST_PACKET;
387 else if (call->tx_top - call->acks_hard_ack <
389 txb->flags |= RXRPC_MORE_PACKETS;
391 ret = call->security->secure_packet(call, txb);
395 txb->kvec[0].iov_len += txb->len;
396 txb->len = txb->kvec[0].iov_len;
397 rxrpc_queue_packet(rx, call, txb, notify_end_tx);
400 } while (msg_data_left(msg) > 0);
404 if (rxrpc_call_is_complete(call) &&
408 call->tx_pending = txb;
409 _leave(" = %d", ret);
413 rxrpc_put_txbuf(txb, rxrpc_txbuf_put_send_aborted);
414 _leave(" = %d", call->error);
428 if (msg->msg_flags & MSG_DONTWAIT)
430 mutex_unlock(&call->user_mutex);
431 *_dropped_lock = true;
432 ret = rxrpc_wait_for_tx_window(rx, call, &timeo,
433 msg->msg_flags & MSG_WAITALL);
436 if (call->interruptibility == RXRPC_INTERRUPTIBLE) {
437 if (mutex_lock_interruptible(&call->user_mutex) < 0) {
438 ret = sock_intr_errno(timeo);
442 mutex_lock(&call->user_mutex);
444 *_dropped_lock = false;
449 * extract control messages from the sendmsg() control buffer
451 static int rxrpc_sendmsg_cmsg(struct msghdr *msg, struct rxrpc_send_params *p)
453 struct cmsghdr *cmsg;
454 bool got_user_ID = false;
457 if (msg->msg_controllen == 0)
460 for_each_cmsghdr(cmsg, msg) {
461 if (!CMSG_OK(msg, cmsg))
464 len = cmsg->cmsg_len - sizeof(struct cmsghdr);
465 _debug("CMSG %d, %d, %d",
466 cmsg->cmsg_level, cmsg->cmsg_type, len);
468 if (cmsg->cmsg_level != SOL_RXRPC)
471 switch (cmsg->cmsg_type) {
472 case RXRPC_USER_CALL_ID:
473 if (msg->msg_flags & MSG_CMSG_COMPAT) {
474 if (len != sizeof(u32))
476 p->call.user_call_ID = *(u32 *)CMSG_DATA(cmsg);
478 if (len != sizeof(unsigned long))
480 p->call.user_call_ID = *(unsigned long *)
487 if (p->command != RXRPC_CMD_SEND_DATA)
489 p->command = RXRPC_CMD_SEND_ABORT;
490 if (len != sizeof(p->abort_code))
492 p->abort_code = *(unsigned int *)CMSG_DATA(cmsg);
493 if (p->abort_code == 0)
497 case RXRPC_CHARGE_ACCEPT:
498 if (p->command != RXRPC_CMD_SEND_DATA)
500 p->command = RXRPC_CMD_CHARGE_ACCEPT;
505 case RXRPC_EXCLUSIVE_CALL:
511 case RXRPC_UPGRADE_SERVICE:
517 case RXRPC_TX_LENGTH:
518 if (p->call.tx_total_len != -1 || len != sizeof(__s64))
520 p->call.tx_total_len = *(__s64 *)CMSG_DATA(cmsg);
521 if (p->call.tx_total_len < 0)
525 case RXRPC_SET_CALL_TIMEOUT:
526 if (len & 3 || len < 4 || len > 12)
528 memcpy(&p->call.timeouts, CMSG_DATA(cmsg), len);
529 p->call.nr_timeouts = len / 4;
530 if (p->call.timeouts.hard > INT_MAX / HZ)
532 if (p->call.nr_timeouts >= 2 && p->call.timeouts.idle > 60 * 60 * 1000)
534 if (p->call.nr_timeouts >= 3 && p->call.timeouts.normal > 60 * 60 * 1000)
545 if (p->call.tx_total_len != -1 && p->command != RXRPC_CMD_SEND_DATA)
552 * Create a new client call for sendmsg().
553 * - Called with the socket lock held, which it must release.
554 * - If it returns a call, the call's lock will need releasing by the caller.
556 static struct rxrpc_call *
557 rxrpc_new_client_call_for_sendmsg(struct rxrpc_sock *rx, struct msghdr *msg,
558 struct rxrpc_send_params *p)
559 __releases(&rx->sk.sk_lock.slock)
560 __acquires(&call->user_mutex)
562 struct rxrpc_conn_parameters cp;
563 struct rxrpc_peer *peer;
564 struct rxrpc_call *call;
567 DECLARE_SOCKADDR(struct sockaddr_rxrpc *, srx, msg->msg_name);
571 if (!msg->msg_name) {
572 release_sock(&rx->sk);
573 return ERR_PTR(-EDESTADDRREQ);
576 peer = rxrpc_lookup_peer(rx->local, srx, GFP_KERNEL);
578 release_sock(&rx->sk);
579 return ERR_PTR(-ENOMEM);
583 if (key && !rx->key->payload.data[0])
586 memset(&cp, 0, sizeof(cp));
587 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, &p->call, GFP_KERNEL,
595 atomic_inc_return(&rxrpc_debug_id));
596 /* The socket is now unlocked */
598 rxrpc_put_peer(peer, rxrpc_peer_put_application);
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)
611 struct rxrpc_call *call;
612 bool dropped_lock = false;
615 struct rxrpc_send_params p = {
616 .call.tx_total_len = -1,
617 .call.user_call_ID = 0,
618 .call.nr_timeouts = 0,
619 .call.interruptibility = RXRPC_INTERRUPTIBLE,
621 .command = RXRPC_CMD_SEND_DATA,
628 ret = rxrpc_sendmsg_cmsg(msg, &p);
630 goto error_release_sock;
632 if (p.command == RXRPC_CMD_CHARGE_ACCEPT) {
634 if (rx->sk.sk_state != RXRPC_SERVER_LISTENING)
635 goto error_release_sock;
636 ret = rxrpc_user_charge_accept(rx, p.call.user_call_ID);
637 goto error_release_sock;
640 call = rxrpc_find_call_by_user_ID(rx, p.call.user_call_ID);
643 if (p.command != RXRPC_CMD_SEND_DATA)
644 goto error_release_sock;
645 call = rxrpc_new_client_call_for_sendmsg(rx, msg, &p);
646 /* The socket is now unlocked... */
648 return PTR_ERR(call);
649 /* ... and we have the call lock. */
650 p.call.nr_timeouts = 0;
652 if (rxrpc_call_is_complete(call))
655 switch (rxrpc_call_state(call)) {
656 case RXRPC_CALL_CLIENT_AWAIT_CONN:
657 case RXRPC_CALL_SERVER_SECURING:
658 if (p.command == RXRPC_CMD_SEND_ABORT)
661 case RXRPC_CALL_UNINITIALISED:
662 case RXRPC_CALL_SERVER_PREALLOC:
663 rxrpc_put_call(call, rxrpc_call_put_sendmsg);
665 goto error_release_sock;
670 ret = mutex_lock_interruptible(&call->user_mutex);
671 release_sock(&rx->sk);
677 if (p.call.tx_total_len != -1) {
679 if (call->tx_total_len != -1 ||
683 call->tx_total_len = p.call.tx_total_len;
687 switch (p.call.nr_timeouts) {
689 WRITE_ONCE(call->next_rx_timo, p.call.timeouts.normal);
692 WRITE_ONCE(call->next_req_timo, p.call.timeouts.idle);
695 if (p.call.timeouts.hard > 0) {
696 ktime_t delay = ms_to_ktime(p.call.timeouts.hard * MSEC_PER_SEC);
698 WRITE_ONCE(call->expect_term_by,
699 ktime_add(p.call.timeouts.hard,
701 trace_rxrpc_timer_set(call, delay, rxrpc_timer_trace_hard);
702 rxrpc_poke_call(call, rxrpc_call_poke_set_timeout);
708 if (rxrpc_call_is_complete(call)) {
709 /* it's too late for this call */
711 } else if (p.command == RXRPC_CMD_SEND_ABORT) {
712 rxrpc_propose_abort(call, p.abort_code, -ECONNABORTED,
713 rxrpc_abort_call_sendmsg);
715 } else if (p.command != RXRPC_CMD_SEND_DATA) {
718 ret = rxrpc_send_data(rx, call, msg, len, NULL, &dropped_lock);
723 mutex_unlock(&call->user_mutex);
725 rxrpc_put_call(call, rxrpc_call_put_sendmsg);
726 _leave(" = %d", ret);
730 release_sock(&rx->sk);
735 * rxrpc_kernel_send_data - Allow a kernel service to send data on a call
736 * @sock: The socket the call is on
737 * @call: The call to send data through
738 * @msg: The data to send
739 * @len: The amount of data to send
740 * @notify_end_tx: Notification that the last packet is queued.
742 * Allow a kernel service to send data on a call. The call must be in an state
743 * appropriate to sending data. No control data should be supplied in @msg,
744 * nor should an address be supplied. MSG_MORE should be flagged if there's
745 * more data to come, otherwise this data will end the transmission phase.
747 int rxrpc_kernel_send_data(struct socket *sock, struct rxrpc_call *call,
748 struct msghdr *msg, size_t len,
749 rxrpc_notify_end_tx_t notify_end_tx)
751 bool dropped_lock = false;
754 _enter("{%d},", call->debug_id);
756 ASSERTCMP(msg->msg_name, ==, NULL);
757 ASSERTCMP(msg->msg_control, ==, NULL);
759 mutex_lock(&call->user_mutex);
761 ret = rxrpc_send_data(rxrpc_sk(sock->sk), call, msg, len,
762 notify_end_tx, &dropped_lock);
763 if (ret == -ESHUTDOWN)
767 mutex_unlock(&call->user_mutex);
768 _leave(" = %d", ret);
771 EXPORT_SYMBOL(rxrpc_kernel_send_data);
774 * rxrpc_kernel_abort_call - Allow a kernel service to abort a call
775 * @sock: The socket the call is on
776 * @call: The call to be aborted
777 * @abort_code: The abort code to stick into the ABORT packet
778 * @error: Local error value
779 * @why: Indication as to why.
781 * Allow a kernel service to abort a call, if it's still in an abortable state
782 * and return true if the call was aborted, false if it was already complete.
784 bool rxrpc_kernel_abort_call(struct socket *sock, struct rxrpc_call *call,
785 u32 abort_code, int error, enum rxrpc_abort_reason why)
789 _enter("{%d},%d,%d,%u", call->debug_id, abort_code, error, why);
791 mutex_lock(&call->user_mutex);
792 aborted = rxrpc_propose_abort(call, abort_code, error, why);
793 mutex_unlock(&call->user_mutex);
796 EXPORT_SYMBOL(rxrpc_kernel_abort_call);
799 * rxrpc_kernel_set_tx_length - Set the total Tx length on a call
800 * @sock: The socket the call is on
801 * @call: The call to be informed
802 * @tx_total_len: The amount of data to be transmitted for this call
804 * Allow a kernel service to set the total transmit length on a call. This
805 * allows buffer-to-packet encrypt-and-copy to be performed.
807 * This function is primarily for use for setting the reply length since the
808 * request length can be set when beginning the call.
810 void rxrpc_kernel_set_tx_length(struct socket *sock, struct rxrpc_call *call,
813 WARN_ON(call->tx_total_len != -1);
814 call->tx_total_len = tx_total_len;
816 EXPORT_SYMBOL(rxrpc_kernel_set_tx_length);