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 last = test_bit(RXRPC_TXBUF_LAST, &txb->flags), poke;
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);
339 size_t remain, bufsize, chunk, offset;
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 ret = call->conn->security->how_much_data(call, remain,
352 &bufsize, &chunk, &offset);
356 _debug("SIZE: %zu/%zu @%zu", chunk, bufsize, offset);
358 /* create a buffer that we can retain until it's ACK'd */
360 txb = rxrpc_alloc_txbuf(call, RXRPC_PACKET_TYPE_DATA,
365 txb->offset = offset;
366 txb->space -= offset;
367 txb->space = min_t(size_t, chunk, txb->space);
372 /* append next segment of data to the current buffer */
373 if (msg_data_left(msg) > 0) {
374 size_t copy = min_t(size_t, txb->space, msg_data_left(msg));
376 _debug("add %zu", copy);
377 if (!copy_from_iter_full(txb->data + txb->offset, copy,
385 if (call->tx_total_len != -1)
386 call->tx_total_len -= copy;
389 /* check for the far side aborting the call or a network error
391 if (rxrpc_call_is_complete(call))
392 goto call_terminated;
394 /* add the packet to the send queue if it's now full */
396 (msg_data_left(msg) == 0 && !more)) {
397 if (msg_data_left(msg) == 0 && !more) {
398 txb->wire.flags |= RXRPC_LAST_PACKET;
399 __set_bit(RXRPC_TXBUF_LAST, &txb->flags);
401 else if (call->tx_top - call->acks_hard_ack <
403 txb->wire.flags |= RXRPC_MORE_PACKETS;
405 ret = call->security->secure_packet(call, txb);
409 rxrpc_queue_packet(rx, call, txb, notify_end_tx);
412 } while (msg_data_left(msg) > 0);
416 if (rxrpc_call_is_complete(call) &&
420 call->tx_pending = txb;
421 _leave(" = %d", ret);
425 rxrpc_put_txbuf(txb, rxrpc_txbuf_put_send_aborted);
426 _leave(" = %d", call->error);
440 if (msg->msg_flags & MSG_DONTWAIT)
442 mutex_unlock(&call->user_mutex);
443 *_dropped_lock = true;
444 ret = rxrpc_wait_for_tx_window(rx, call, &timeo,
445 msg->msg_flags & MSG_WAITALL);
448 if (call->interruptibility == RXRPC_INTERRUPTIBLE) {
449 if (mutex_lock_interruptible(&call->user_mutex) < 0) {
450 ret = sock_intr_errno(timeo);
454 mutex_lock(&call->user_mutex);
456 *_dropped_lock = false;
461 * extract control messages from the sendmsg() control buffer
463 static int rxrpc_sendmsg_cmsg(struct msghdr *msg, struct rxrpc_send_params *p)
465 struct cmsghdr *cmsg;
466 bool got_user_ID = false;
469 if (msg->msg_controllen == 0)
472 for_each_cmsghdr(cmsg, msg) {
473 if (!CMSG_OK(msg, cmsg))
476 len = cmsg->cmsg_len - sizeof(struct cmsghdr);
477 _debug("CMSG %d, %d, %d",
478 cmsg->cmsg_level, cmsg->cmsg_type, len);
480 if (cmsg->cmsg_level != SOL_RXRPC)
483 switch (cmsg->cmsg_type) {
484 case RXRPC_USER_CALL_ID:
485 if (msg->msg_flags & MSG_CMSG_COMPAT) {
486 if (len != sizeof(u32))
488 p->call.user_call_ID = *(u32 *)CMSG_DATA(cmsg);
490 if (len != sizeof(unsigned long))
492 p->call.user_call_ID = *(unsigned long *)
499 if (p->command != RXRPC_CMD_SEND_DATA)
501 p->command = RXRPC_CMD_SEND_ABORT;
502 if (len != sizeof(p->abort_code))
504 p->abort_code = *(unsigned int *)CMSG_DATA(cmsg);
505 if (p->abort_code == 0)
509 case RXRPC_CHARGE_ACCEPT:
510 if (p->command != RXRPC_CMD_SEND_DATA)
512 p->command = RXRPC_CMD_CHARGE_ACCEPT;
517 case RXRPC_EXCLUSIVE_CALL:
523 case RXRPC_UPGRADE_SERVICE:
529 case RXRPC_TX_LENGTH:
530 if (p->call.tx_total_len != -1 || len != sizeof(__s64))
532 p->call.tx_total_len = *(__s64 *)CMSG_DATA(cmsg);
533 if (p->call.tx_total_len < 0)
537 case RXRPC_SET_CALL_TIMEOUT:
538 if (len & 3 || len < 4 || len > 12)
540 memcpy(&p->call.timeouts, CMSG_DATA(cmsg), len);
541 p->call.nr_timeouts = len / 4;
542 if (p->call.timeouts.hard > INT_MAX / HZ)
544 if (p->call.nr_timeouts >= 2 && p->call.timeouts.idle > 60 * 60 * 1000)
546 if (p->call.nr_timeouts >= 3 && p->call.timeouts.normal > 60 * 60 * 1000)
557 if (p->call.tx_total_len != -1 && p->command != RXRPC_CMD_SEND_DATA)
564 * Create a new client call for sendmsg().
565 * - Called with the socket lock held, which it must release.
566 * - If it returns a call, the call's lock will need releasing by the caller.
568 static struct rxrpc_call *
569 rxrpc_new_client_call_for_sendmsg(struct rxrpc_sock *rx, struct msghdr *msg,
570 struct rxrpc_send_params *p)
571 __releases(&rx->sk.sk_lock.slock)
572 __acquires(&call->user_mutex)
574 struct rxrpc_conn_parameters cp;
575 struct rxrpc_peer *peer;
576 struct rxrpc_call *call;
579 DECLARE_SOCKADDR(struct sockaddr_rxrpc *, srx, msg->msg_name);
583 if (!msg->msg_name) {
584 release_sock(&rx->sk);
585 return ERR_PTR(-EDESTADDRREQ);
588 peer = rxrpc_lookup_peer(rx->local, srx, GFP_KERNEL);
590 release_sock(&rx->sk);
591 return ERR_PTR(-ENOMEM);
595 if (key && !rx->key->payload.data[0])
598 memset(&cp, 0, sizeof(cp));
599 cp.local = rx->local;
602 cp.security_level = rx->min_sec_level;
603 cp.exclusive = rx->exclusive | p->exclusive;
604 cp.upgrade = p->upgrade;
605 cp.service_id = srx->srx_service;
606 call = rxrpc_new_client_call(rx, &cp, &p->call, GFP_KERNEL,
607 atomic_inc_return(&rxrpc_debug_id));
608 /* The socket is now unlocked */
610 rxrpc_put_peer(peer, rxrpc_peer_put_application);
611 _leave(" = %p\n", call);
616 * send a message forming part of a client call through an RxRPC socket
617 * - caller holds the socket locked
618 * - the socket may be either a client socket or a server socket
620 int rxrpc_do_sendmsg(struct rxrpc_sock *rx, struct msghdr *msg, size_t len)
621 __releases(&rx->sk.sk_lock.slock)
623 struct rxrpc_call *call;
624 unsigned long now, j;
625 bool dropped_lock = false;
628 struct rxrpc_send_params p = {
629 .call.tx_total_len = -1,
630 .call.user_call_ID = 0,
631 .call.nr_timeouts = 0,
632 .call.interruptibility = RXRPC_INTERRUPTIBLE,
634 .command = RXRPC_CMD_SEND_DATA,
641 ret = rxrpc_sendmsg_cmsg(msg, &p);
643 goto error_release_sock;
645 if (p.command == RXRPC_CMD_CHARGE_ACCEPT) {
647 if (rx->sk.sk_state != RXRPC_SERVER_LISTENING)
648 goto error_release_sock;
649 ret = rxrpc_user_charge_accept(rx, p.call.user_call_ID);
650 goto error_release_sock;
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. */
663 p.call.nr_timeouts = 0;
665 if (rxrpc_call_is_complete(call))
668 switch (rxrpc_call_state(call)) {
669 case RXRPC_CALL_CLIENT_AWAIT_CONN:
670 case RXRPC_CALL_SERVER_SECURING:
671 if (p.command == RXRPC_CMD_SEND_ABORT)
674 case RXRPC_CALL_UNINITIALISED:
675 case RXRPC_CALL_SERVER_PREALLOC:
676 rxrpc_put_call(call, rxrpc_call_put_sendmsg);
678 goto error_release_sock;
683 ret = mutex_lock_interruptible(&call->user_mutex);
684 release_sock(&rx->sk);
690 if (p.call.tx_total_len != -1) {
692 if (call->tx_total_len != -1 ||
696 call->tx_total_len = p.call.tx_total_len;
700 switch (p.call.nr_timeouts) {
702 j = msecs_to_jiffies(p.call.timeouts.normal);
703 if (p.call.timeouts.normal > 0 && j == 0)
705 WRITE_ONCE(call->next_rx_timo, j);
708 j = msecs_to_jiffies(p.call.timeouts.idle);
709 if (p.call.timeouts.idle > 0 && j == 0)
711 WRITE_ONCE(call->next_req_timo, j);
714 if (p.call.timeouts.hard > 0) {
715 j = p.call.timeouts.hard * HZ;
718 WRITE_ONCE(call->expect_term_by, j);
719 rxrpc_reduce_call_timer(call, j, now,
720 rxrpc_timer_set_for_hard);
725 if (rxrpc_call_is_complete(call)) {
726 /* it's too late for this call */
728 } else if (p.command == RXRPC_CMD_SEND_ABORT) {
729 rxrpc_propose_abort(call, p.abort_code, -ECONNABORTED,
730 rxrpc_abort_call_sendmsg);
732 } else if (p.command != RXRPC_CMD_SEND_DATA) {
735 ret = rxrpc_send_data(rx, call, msg, len, NULL, &dropped_lock);
740 mutex_unlock(&call->user_mutex);
742 rxrpc_put_call(call, rxrpc_call_put_sendmsg);
743 _leave(" = %d", ret);
747 release_sock(&rx->sk);
752 * rxrpc_kernel_send_data - Allow a kernel service to send data on a call
753 * @sock: The socket the call is on
754 * @call: The call to send data through
755 * @msg: The data to send
756 * @len: The amount of data to send
757 * @notify_end_tx: Notification that the last packet is queued.
759 * Allow a kernel service to send data on a call. The call must be in an state
760 * appropriate to sending data. No control data should be supplied in @msg,
761 * nor should an address be supplied. MSG_MORE should be flagged if there's
762 * more data to come, otherwise this data will end the transmission phase.
764 int rxrpc_kernel_send_data(struct socket *sock, struct rxrpc_call *call,
765 struct msghdr *msg, size_t len,
766 rxrpc_notify_end_tx_t notify_end_tx)
768 bool dropped_lock = false;
771 _enter("{%d},", call->debug_id);
773 ASSERTCMP(msg->msg_name, ==, NULL);
774 ASSERTCMP(msg->msg_control, ==, NULL);
776 mutex_lock(&call->user_mutex);
778 ret = rxrpc_send_data(rxrpc_sk(sock->sk), call, msg, len,
779 notify_end_tx, &dropped_lock);
780 if (ret == -ESHUTDOWN)
784 mutex_unlock(&call->user_mutex);
785 _leave(" = %d", ret);
788 EXPORT_SYMBOL(rxrpc_kernel_send_data);
791 * rxrpc_kernel_abort_call - Allow a kernel service to abort a call
792 * @sock: The socket the call is on
793 * @call: The call to be aborted
794 * @abort_code: The abort code to stick into the ABORT packet
795 * @error: Local error value
796 * @why: Indication as to why.
798 * Allow a kernel service to abort a call, if it's still in an abortable state
799 * and return true if the call was aborted, false if it was already complete.
801 bool rxrpc_kernel_abort_call(struct socket *sock, struct rxrpc_call *call,
802 u32 abort_code, int error, enum rxrpc_abort_reason why)
806 _enter("{%d},%d,%d,%u", call->debug_id, abort_code, error, why);
808 mutex_lock(&call->user_mutex);
809 aborted = rxrpc_propose_abort(call, abort_code, error, why);
810 mutex_unlock(&call->user_mutex);
813 EXPORT_SYMBOL(rxrpc_kernel_abort_call);
816 * rxrpc_kernel_set_tx_length - Set the total Tx length on a call
817 * @sock: The socket the call is on
818 * @call: The call to be informed
819 * @tx_total_len: The amount of data to be transmitted for this call
821 * Allow a kernel service to set the total transmit length on a call. This
822 * allows buffer-to-packet encrypt-and-copy to be performed.
824 * This function is primarily for use for setting the reply length since the
825 * request length can be set when beginning the call.
827 void rxrpc_kernel_set_tx_length(struct socket *sock, struct rxrpc_call *call,
830 WARN_ON(call->tx_total_len != -1);
831 call->tx_total_len = tx_total_len;
833 EXPORT_SYMBOL(rxrpc_kernel_set_tx_length);