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);
62 switch (call->interruptibility) {
63 case RXRPC_INTERRUPTIBLE:
64 case RXRPC_PREINTERRUPTIBLE:
65 set_current_state(TASK_INTERRUPTIBLE);
67 case RXRPC_UNINTERRUPTIBLE:
69 set_current_state(TASK_UNINTERRUPTIBLE);
72 if (rxrpc_call_state(call) != RXRPC_CALL_CLIENT_AWAIT_CONN) {
76 if ((call->interruptibility == RXRPC_INTERRUPTIBLE ||
77 call->interruptibility == RXRPC_PREINTERRUPTIBLE) &&
78 signal_pending(current)) {
79 ret = sock_intr_errno(*timeo);
82 *timeo = schedule_timeout(*timeo);
85 remove_wait_queue(&call->waitq, &myself);
86 __set_current_state(TASK_RUNNING);
88 if (ret == 0 && rxrpc_call_is_complete(call))
96 * Return true if there's sufficient Tx queue space.
98 static bool rxrpc_check_tx_space(struct rxrpc_call *call, rxrpc_seq_t *_tx_win)
101 *_tx_win = call->tx_bottom;
102 return call->tx_prepared - call->tx_bottom < 256;
106 * Wait for space to appear in the Tx queue or a signal to occur.
108 static int rxrpc_wait_for_tx_window_intr(struct rxrpc_sock *rx,
109 struct rxrpc_call *call,
113 set_current_state(TASK_INTERRUPTIBLE);
114 if (rxrpc_check_tx_space(call, NULL))
117 if (rxrpc_call_is_complete(call))
120 if (signal_pending(current))
121 return sock_intr_errno(*timeo);
123 trace_rxrpc_txqueue(call, rxrpc_txqueue_wait);
124 *timeo = schedule_timeout(*timeo);
129 * Wait for space to appear in the Tx queue uninterruptibly, but with
130 * a timeout of 2*RTT if no progress was made and a signal occurred.
132 static int rxrpc_wait_for_tx_window_waitall(struct rxrpc_sock *rx,
133 struct rxrpc_call *call)
135 rxrpc_seq_t tx_start, tx_win;
136 signed long rtt, timeout;
138 rtt = READ_ONCE(call->peer->srtt_us) >> 3;
139 rtt = usecs_to_jiffies(rtt) * 2;
144 tx_start = smp_load_acquire(&call->acks_hard_ack);
147 set_current_state(TASK_UNINTERRUPTIBLE);
149 if (rxrpc_check_tx_space(call, &tx_win))
152 if (rxrpc_call_is_complete(call))
156 tx_win == tx_start && signal_pending(current))
159 if (tx_win != tx_start) {
164 trace_rxrpc_txqueue(call, rxrpc_txqueue_wait);
165 timeout = schedule_timeout(timeout);
170 * Wait for space to appear in the Tx queue uninterruptibly.
172 static int rxrpc_wait_for_tx_window_nonintr(struct rxrpc_sock *rx,
173 struct rxrpc_call *call,
177 set_current_state(TASK_UNINTERRUPTIBLE);
178 if (rxrpc_check_tx_space(call, NULL))
181 if (rxrpc_call_is_complete(call))
184 trace_rxrpc_txqueue(call, rxrpc_txqueue_wait);
185 *timeo = schedule_timeout(*timeo);
190 * wait for space to appear in the transmit/ACK window
191 * - caller holds the socket locked
193 static int rxrpc_wait_for_tx_window(struct rxrpc_sock *rx,
194 struct rxrpc_call *call,
198 DECLARE_WAITQUEUE(myself, current);
201 _enter(",{%u,%u,%u,%u}",
202 call->tx_bottom, call->acks_hard_ack, call->tx_top, call->tx_winsize);
204 add_wait_queue(&call->waitq, &myself);
206 switch (call->interruptibility) {
207 case RXRPC_INTERRUPTIBLE:
209 ret = rxrpc_wait_for_tx_window_waitall(rx, call);
211 ret = rxrpc_wait_for_tx_window_intr(rx, call, timeo);
213 case RXRPC_PREINTERRUPTIBLE:
214 case RXRPC_UNINTERRUPTIBLE:
216 ret = rxrpc_wait_for_tx_window_nonintr(rx, call, timeo);
220 remove_wait_queue(&call->waitq, &myself);
221 set_current_state(TASK_RUNNING);
222 _leave(" = %d", ret);
227 * Notify the owner of the call that the transmit phase is ended and the last
228 * packet has been queued.
230 static void rxrpc_notify_end_tx(struct rxrpc_sock *rx, struct rxrpc_call *call,
231 rxrpc_notify_end_tx_t notify_end_tx)
234 notify_end_tx(&rx->sk, call, call->user_call_ID);
238 * Queue a DATA packet for transmission, set the resend timeout and send
239 * the packet immediately. Returns the error from rxrpc_send_data_packet()
240 * in case the caller wants to do something with it.
242 static void rxrpc_queue_packet(struct rxrpc_sock *rx, struct rxrpc_call *call,
243 struct rxrpc_txbuf *txb,
244 rxrpc_notify_end_tx_t notify_end_tx)
246 rxrpc_seq_t seq = txb->seq;
247 bool last = test_bit(RXRPC_TXBUF_LAST, &txb->flags), poke;
249 rxrpc_inc_stat(call->rxnet, stat_tx_data);
251 ASSERTCMP(txb->seq, ==, call->tx_prepared + 1);
253 /* We have to set the timestamp before queueing as the retransmit
254 * algorithm can see the packet as soon as we queue it.
256 txb->last_sent = ktime_get_real();
259 trace_rxrpc_txqueue(call, rxrpc_txqueue_queue_last);
261 trace_rxrpc_txqueue(call, rxrpc_txqueue_queue);
263 /* Add the packet to the call's output buffer */
264 spin_lock(&call->tx_lock);
265 poke = list_empty(&call->tx_sendmsg);
266 list_add_tail(&txb->call_link, &call->tx_sendmsg);
267 call->tx_prepared = seq;
269 rxrpc_notify_end_tx(rx, call, notify_end_tx);
270 spin_unlock(&call->tx_lock);
273 rxrpc_poke_call(call, rxrpc_call_poke_start);
277 * send data through a socket
278 * - must be called in process context
279 * - The caller holds the call user access mutex, but not the socket lock.
281 static int rxrpc_send_data(struct rxrpc_sock *rx,
282 struct rxrpc_call *call,
283 struct msghdr *msg, size_t len,
284 rxrpc_notify_end_tx_t notify_end_tx,
287 struct rxrpc_txbuf *txb;
288 struct sock *sk = &rx->sk;
289 enum rxrpc_call_state state;
291 bool more = msg->msg_flags & MSG_MORE;
294 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
296 ret = rxrpc_wait_to_be_connected(call, &timeo);
300 if (call->conn->state == RXRPC_CONN_CLIENT_UNSECURED) {
301 ret = rxrpc_init_client_conn_security(call->conn);
306 /* this should be in poll */
307 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
311 if (sk->sk_shutdown & SEND_SHUTDOWN)
313 state = rxrpc_call_state(call);
315 if (state >= RXRPC_CALL_COMPLETE)
318 if (state != RXRPC_CALL_CLIENT_SEND_REQUEST &&
319 state != RXRPC_CALL_SERVER_ACK_REQUEST &&
320 state != RXRPC_CALL_SERVER_SEND_REPLY) {
321 /* Request phase complete for this client call */
322 trace_rxrpc_abort(call->debug_id, rxrpc_sendmsg_late_send,
323 call->cid, call->call_id, call->rx_consumed,
329 if (call->tx_total_len != -1) {
330 if (len - copied > call->tx_total_len)
332 if (!more && len - copied != call->tx_total_len)
336 txb = call->tx_pending;
337 call->tx_pending = NULL;
339 rxrpc_see_txbuf(txb, rxrpc_txbuf_see_send_more);
343 size_t remain, bufsize, chunk, offset;
347 if (!rxrpc_check_tx_space(call, NULL))
350 /* Work out the maximum size of a packet. Assume that
351 * the security header is going to be in the padded
352 * region (enc blocksize), but the trailer is not.
354 remain = more ? INT_MAX : msg_data_left(msg);
355 ret = call->conn->security->how_much_data(call, remain,
356 &bufsize, &chunk, &offset);
360 _debug("SIZE: %zu/%zu @%zu", chunk, bufsize, offset);
362 /* create a buffer that we can retain until it's ACK'd */
364 txb = rxrpc_alloc_txbuf(call, RXRPC_PACKET_TYPE_DATA,
369 txb->offset = offset;
370 txb->space -= offset;
371 txb->space = min_t(size_t, chunk, txb->space);
376 /* append next segment of data to the current buffer */
377 if (msg_data_left(msg) > 0) {
378 size_t copy = min_t(size_t, txb->space, msg_data_left(msg));
380 _debug("add %zu", copy);
381 if (!copy_from_iter_full(txb->data + txb->offset, copy,
389 if (call->tx_total_len != -1)
390 call->tx_total_len -= copy;
393 /* check for the far side aborting the call or a network error
395 if (rxrpc_call_is_complete(call))
396 goto call_terminated;
398 /* add the packet to the send queue if it's now full */
400 (msg_data_left(msg) == 0 && !more)) {
401 if (msg_data_left(msg) == 0 && !more) {
402 txb->wire.flags |= RXRPC_LAST_PACKET;
403 __set_bit(RXRPC_TXBUF_LAST, &txb->flags);
405 else if (call->tx_top - call->acks_hard_ack <
407 txb->wire.flags |= RXRPC_MORE_PACKETS;
409 ret = call->security->secure_packet(call, txb);
413 rxrpc_queue_packet(rx, call, txb, notify_end_tx);
416 } while (msg_data_left(msg) > 0);
420 if (rxrpc_call_is_complete(call) &&
424 call->tx_pending = txb;
425 _leave(" = %d", ret);
429 rxrpc_put_txbuf(txb, rxrpc_txbuf_put_send_aborted);
430 _leave(" = %d", call->error);
444 if (msg->msg_flags & MSG_DONTWAIT)
446 mutex_unlock(&call->user_mutex);
447 *_dropped_lock = true;
448 ret = rxrpc_wait_for_tx_window(rx, call, &timeo,
449 msg->msg_flags & MSG_WAITALL);
452 if (call->interruptibility == RXRPC_INTERRUPTIBLE) {
453 if (mutex_lock_interruptible(&call->user_mutex) < 0) {
454 ret = sock_intr_errno(timeo);
458 mutex_lock(&call->user_mutex);
460 *_dropped_lock = false;
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)
513 case RXRPC_CHARGE_ACCEPT:
514 if (p->command != RXRPC_CMD_SEND_DATA)
516 p->command = RXRPC_CMD_CHARGE_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 _leave(" = %p\n", call);
611 * send a message forming part of a client call through an RxRPC socket
612 * - caller holds the socket locked
613 * - the socket may be either a client socket or a server socket
615 int rxrpc_do_sendmsg(struct rxrpc_sock *rx, struct msghdr *msg, size_t len)
616 __releases(&rx->sk.sk_lock.slock)
618 struct rxrpc_call *call;
619 unsigned long now, j;
620 bool dropped_lock = false;
623 struct rxrpc_send_params p = {
624 .call.tx_total_len = -1,
625 .call.user_call_ID = 0,
626 .call.nr_timeouts = 0,
627 .call.interruptibility = RXRPC_INTERRUPTIBLE,
629 .command = RXRPC_CMD_SEND_DATA,
636 ret = rxrpc_sendmsg_cmsg(msg, &p);
638 goto error_release_sock;
640 if (p.command == RXRPC_CMD_CHARGE_ACCEPT) {
642 if (rx->sk.sk_state != RXRPC_SERVER_LISTENING)
643 goto error_release_sock;
644 ret = rxrpc_user_charge_accept(rx, p.call.user_call_ID);
645 goto error_release_sock;
648 call = rxrpc_find_call_by_user_ID(rx, p.call.user_call_ID);
651 if (p.command != RXRPC_CMD_SEND_DATA)
652 goto error_release_sock;
653 call = rxrpc_new_client_call_for_sendmsg(rx, msg, &p);
654 /* The socket is now unlocked... */
656 return PTR_ERR(call);
657 /* ... and we have the call lock. */
659 if (rxrpc_call_is_complete(call))
662 switch (rxrpc_call_state(call)) {
663 case RXRPC_CALL_UNINITIALISED:
664 case RXRPC_CALL_CLIENT_AWAIT_CONN:
665 case RXRPC_CALL_SERVER_PREALLOC:
666 case RXRPC_CALL_SERVER_SECURING:
667 rxrpc_put_call(call, rxrpc_call_put_sendmsg);
669 goto error_release_sock;
674 ret = mutex_lock_interruptible(&call->user_mutex);
675 release_sock(&rx->sk);
681 if (p.call.tx_total_len != -1) {
683 if (call->tx_total_len != -1 ||
687 call->tx_total_len = p.call.tx_total_len;
691 switch (p.call.nr_timeouts) {
693 j = msecs_to_jiffies(p.call.timeouts.normal);
694 if (p.call.timeouts.normal > 0 && j == 0)
696 WRITE_ONCE(call->next_rx_timo, j);
699 j = msecs_to_jiffies(p.call.timeouts.idle);
700 if (p.call.timeouts.idle > 0 && j == 0)
702 WRITE_ONCE(call->next_req_timo, j);
705 if (p.call.timeouts.hard > 0) {
706 j = msecs_to_jiffies(p.call.timeouts.hard);
709 WRITE_ONCE(call->expect_term_by, j);
710 rxrpc_reduce_call_timer(call, j, now,
711 rxrpc_timer_set_for_hard);
716 if (rxrpc_call_is_complete(call)) {
717 /* it's too late for this call */
719 } else if (p.command == RXRPC_CMD_SEND_ABORT) {
720 rxrpc_propose_abort(call, p.abort_code, -ECONNABORTED,
721 rxrpc_abort_call_sendmsg);
723 } else if (p.command != RXRPC_CMD_SEND_DATA) {
726 ret = rxrpc_send_data(rx, call, msg, len, NULL, &dropped_lock);
731 mutex_unlock(&call->user_mutex);
733 rxrpc_put_call(call, rxrpc_call_put_sendmsg);
734 _leave(" = %d", ret);
738 release_sock(&rx->sk);
743 * rxrpc_kernel_send_data - Allow a kernel service to send data on a call
744 * @sock: The socket the call is on
745 * @call: The call to send data through
746 * @msg: The data to send
747 * @len: The amount of data to send
748 * @notify_end_tx: Notification that the last packet is queued.
750 * Allow a kernel service to send data on a call. The call must be in an state
751 * appropriate to sending data. No control data should be supplied in @msg,
752 * nor should an address be supplied. MSG_MORE should be flagged if there's
753 * more data to come, otherwise this data will end the transmission phase.
755 int rxrpc_kernel_send_data(struct socket *sock, struct rxrpc_call *call,
756 struct msghdr *msg, size_t len,
757 rxrpc_notify_end_tx_t notify_end_tx)
759 bool dropped_lock = false;
762 _enter("{%d},", call->debug_id);
764 ASSERTCMP(msg->msg_name, ==, NULL);
765 ASSERTCMP(msg->msg_control, ==, NULL);
767 mutex_lock(&call->user_mutex);
769 ret = rxrpc_send_data(rxrpc_sk(sock->sk), call, msg, len,
770 notify_end_tx, &dropped_lock);
771 if (ret == -ESHUTDOWN)
775 mutex_unlock(&call->user_mutex);
776 _leave(" = %d", ret);
779 EXPORT_SYMBOL(rxrpc_kernel_send_data);
782 * rxrpc_kernel_abort_call - Allow a kernel service to abort a call
783 * @sock: The socket the call is on
784 * @call: The call to be aborted
785 * @abort_code: The abort code to stick into the ABORT packet
786 * @error: Local error value
787 * @why: Indication as to why.
789 * Allow a kernel service to abort a call, if it's still in an abortable state
790 * and return true if the call was aborted, false if it was already complete.
792 bool rxrpc_kernel_abort_call(struct socket *sock, struct rxrpc_call *call,
793 u32 abort_code, int error, enum rxrpc_abort_reason why)
797 _enter("{%d},%d,%d,%u", call->debug_id, abort_code, error, why);
799 mutex_lock(&call->user_mutex);
800 aborted = rxrpc_propose_abort(call, abort_code, error, why);
801 mutex_unlock(&call->user_mutex);
804 EXPORT_SYMBOL(rxrpc_kernel_abort_call);
807 * rxrpc_kernel_set_tx_length - Set the total Tx length on a call
808 * @sock: The socket the call is on
809 * @call: The call to be informed
810 * @tx_total_len: The amount of data to be transmitted for this call
812 * Allow a kernel service to set the total transmit length on a call. This
813 * allows buffer-to-packet encrypt-and-copy to be performed.
815 * This function is primarily for use for setting the reply length since the
816 * request length can be set when beginning the call.
818 void rxrpc_kernel_set_tx_length(struct socket *sock, struct rxrpc_call *call,
821 WARN_ON(call->tx_total_len != -1);
822 call->tx_total_len = tx_total_len;
824 EXPORT_SYMBOL(rxrpc_kernel_set_tx_length);