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 * Return true if there's sufficient Tx queue space.
23 static bool rxrpc_check_tx_space(struct rxrpc_call *call, rxrpc_seq_t *_tx_win)
26 *_tx_win = call->tx_bottom;
27 return call->tx_prepared - call->tx_bottom < 256;
31 * Wait for space to appear in the Tx queue or a signal to occur.
33 static int rxrpc_wait_for_tx_window_intr(struct rxrpc_sock *rx,
34 struct rxrpc_call *call,
38 set_current_state(TASK_INTERRUPTIBLE);
39 if (rxrpc_check_tx_space(call, NULL))
42 if (call->state >= RXRPC_CALL_COMPLETE)
45 if (signal_pending(current))
46 return sock_intr_errno(*timeo);
48 trace_rxrpc_txqueue(call, rxrpc_txqueue_wait);
49 *timeo = schedule_timeout(*timeo);
54 * Wait for space to appear in the Tx queue uninterruptibly, but with
55 * a timeout of 2*RTT if no progress was made and a signal occurred.
57 static int rxrpc_wait_for_tx_window_waitall(struct rxrpc_sock *rx,
58 struct rxrpc_call *call)
60 rxrpc_seq_t tx_start, tx_win;
61 signed long rtt, timeout;
63 rtt = READ_ONCE(call->peer->srtt_us) >> 3;
64 rtt = usecs_to_jiffies(rtt) * 2;
69 tx_start = smp_load_acquire(&call->acks_hard_ack);
72 set_current_state(TASK_UNINTERRUPTIBLE);
74 if (rxrpc_check_tx_space(call, &tx_win))
77 if (call->state >= RXRPC_CALL_COMPLETE)
81 tx_win == tx_start && signal_pending(current))
84 if (tx_win != tx_start) {
89 trace_rxrpc_txqueue(call, rxrpc_txqueue_wait);
90 timeout = schedule_timeout(timeout);
95 * Wait for space to appear in the Tx queue uninterruptibly.
97 static int rxrpc_wait_for_tx_window_nonintr(struct rxrpc_sock *rx,
98 struct rxrpc_call *call,
102 set_current_state(TASK_UNINTERRUPTIBLE);
103 if (rxrpc_check_tx_space(call, NULL))
106 if (call->state >= RXRPC_CALL_COMPLETE)
109 trace_rxrpc_txqueue(call, rxrpc_txqueue_wait);
110 *timeo = schedule_timeout(*timeo);
115 * wait for space to appear in the transmit/ACK window
116 * - caller holds the socket locked
118 static int rxrpc_wait_for_tx_window(struct rxrpc_sock *rx,
119 struct rxrpc_call *call,
123 DECLARE_WAITQUEUE(myself, current);
126 _enter(",{%u,%u,%u,%u}",
127 call->tx_bottom, call->acks_hard_ack, call->tx_top, call->tx_winsize);
129 add_wait_queue(&call->waitq, &myself);
131 switch (call->interruptibility) {
132 case RXRPC_INTERRUPTIBLE:
134 ret = rxrpc_wait_for_tx_window_waitall(rx, call);
136 ret = rxrpc_wait_for_tx_window_intr(rx, call, timeo);
138 case RXRPC_PREINTERRUPTIBLE:
139 case RXRPC_UNINTERRUPTIBLE:
141 ret = rxrpc_wait_for_tx_window_nonintr(rx, call, timeo);
145 remove_wait_queue(&call->waitq, &myself);
146 set_current_state(TASK_RUNNING);
147 _leave(" = %d", ret);
152 * Notify the owner of the call that the transmit phase is ended and the last
153 * packet has been queued.
155 static void rxrpc_notify_end_tx(struct rxrpc_sock *rx, struct rxrpc_call *call,
156 rxrpc_notify_end_tx_t notify_end_tx)
159 notify_end_tx(&rx->sk, call, call->user_call_ID);
163 * Queue a DATA packet for transmission, set the resend timeout and send
164 * the packet immediately. Returns the error from rxrpc_send_data_packet()
165 * in case the caller wants to do something with it.
167 static void rxrpc_queue_packet(struct rxrpc_sock *rx, struct rxrpc_call *call,
168 struct rxrpc_txbuf *txb,
169 rxrpc_notify_end_tx_t notify_end_tx)
172 rxrpc_seq_t seq = txb->seq;
173 bool last = test_bit(RXRPC_TXBUF_LAST, &txb->flags), poke;
175 rxrpc_inc_stat(call->rxnet, stat_tx_data);
177 ASSERTCMP(txb->seq, ==, call->tx_prepared + 1);
179 /* We have to set the timestamp before queueing as the retransmit
180 * algorithm can see the packet as soon as we queue it.
182 txb->last_sent = ktime_get_real();
185 trace_rxrpc_txqueue(call, rxrpc_txqueue_queue_last);
187 trace_rxrpc_txqueue(call, rxrpc_txqueue_queue);
189 /* Add the packet to the call's output buffer */
190 spin_lock(&call->tx_lock);
191 poke = list_empty(&call->tx_sendmsg);
192 list_add_tail(&txb->call_link, &call->tx_sendmsg);
193 call->tx_prepared = seq;
194 spin_unlock(&call->tx_lock);
196 if (last || call->state == RXRPC_CALL_SERVER_ACK_REQUEST) {
197 _debug("________awaiting reply/ACK__________");
198 write_lock_bh(&call->state_lock);
199 switch (call->state) {
200 case RXRPC_CALL_CLIENT_SEND_REQUEST:
201 call->state = RXRPC_CALL_CLIENT_AWAIT_REPLY;
202 rxrpc_notify_end_tx(rx, call, notify_end_tx);
204 case RXRPC_CALL_SERVER_ACK_REQUEST:
205 call->state = RXRPC_CALL_SERVER_SEND_REPLY;
207 WRITE_ONCE(call->delay_ack_at, now + MAX_JIFFY_OFFSET);
208 if (call->ackr_reason == RXRPC_ACK_DELAY)
209 call->ackr_reason = 0;
210 trace_rxrpc_timer(call, rxrpc_timer_init_for_send_reply, now);
214 case RXRPC_CALL_SERVER_SEND_REPLY:
215 call->state = RXRPC_CALL_SERVER_AWAIT_ACK;
216 rxrpc_notify_end_tx(rx, call, notify_end_tx);
221 write_unlock_bh(&call->state_lock);
225 rxrpc_poke_call(call, rxrpc_call_poke_start);
229 * send data through a socket
230 * - must be called in process context
231 * - The caller holds the call user access mutex, but not the socket lock.
233 static int rxrpc_send_data(struct rxrpc_sock *rx,
234 struct rxrpc_call *call,
235 struct msghdr *msg, size_t len,
236 rxrpc_notify_end_tx_t notify_end_tx,
239 struct rxrpc_txbuf *txb;
240 struct sock *sk = &rx->sk;
241 enum rxrpc_call_state state;
243 bool more = msg->msg_flags & MSG_MORE;
246 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
248 /* this should be in poll */
249 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
253 if (sk->sk_shutdown & SEND_SHUTDOWN)
255 state = READ_ONCE(call->state);
257 if (state >= RXRPC_CALL_COMPLETE)
260 if (state != RXRPC_CALL_CLIENT_SEND_REQUEST &&
261 state != RXRPC_CALL_SERVER_ACK_REQUEST &&
262 state != RXRPC_CALL_SERVER_SEND_REPLY)
266 if (call->tx_total_len != -1) {
267 if (len - copied > call->tx_total_len)
269 if (!more && len - copied != call->tx_total_len)
273 txb = call->tx_pending;
274 call->tx_pending = NULL;
276 rxrpc_see_txbuf(txb, rxrpc_txbuf_see_send_more);
279 rxrpc_transmit_ack_packets(call->peer->local);
282 size_t remain, bufsize, chunk, offset;
286 if (!rxrpc_check_tx_space(call, NULL))
289 /* Work out the maximum size of a packet. Assume that
290 * the security header is going to be in the padded
291 * region (enc blocksize), but the trailer is not.
293 remain = more ? INT_MAX : msg_data_left(msg);
294 ret = call->conn->security->how_much_data(call, remain,
295 &bufsize, &chunk, &offset);
299 _debug("SIZE: %zu/%zu @%zu", chunk, bufsize, offset);
301 /* create a buffer that we can retain until it's ACK'd */
303 txb = rxrpc_alloc_txbuf(call, RXRPC_PACKET_TYPE_DATA,
308 txb->offset = offset;
309 txb->space -= offset;
310 txb->space = min_t(size_t, chunk, txb->space);
315 /* append next segment of data to the current buffer */
316 if (msg_data_left(msg) > 0) {
317 size_t copy = min_t(size_t, txb->space, msg_data_left(msg));
319 _debug("add %zu", copy);
320 if (!copy_from_iter_full(txb->data + txb->offset, copy,
328 if (call->tx_total_len != -1)
329 call->tx_total_len -= copy;
332 /* check for the far side aborting the call or a network error
334 if (call->state == RXRPC_CALL_COMPLETE)
335 goto call_terminated;
337 /* add the packet to the send queue if it's now full */
339 (msg_data_left(msg) == 0 && !more)) {
340 if (msg_data_left(msg) == 0 && !more) {
341 txb->wire.flags |= RXRPC_LAST_PACKET;
342 __set_bit(RXRPC_TXBUF_LAST, &txb->flags);
344 else if (call->tx_top - call->acks_hard_ack <
346 txb->wire.flags |= RXRPC_MORE_PACKETS;
348 ret = call->security->secure_packet(call, txb);
352 rxrpc_queue_packet(rx, call, txb, notify_end_tx);
355 } while (msg_data_left(msg) > 0);
359 if (READ_ONCE(call->state) == RXRPC_CALL_COMPLETE) {
360 read_lock_bh(&call->state_lock);
363 read_unlock_bh(&call->state_lock);
366 call->tx_pending = txb;
367 _leave(" = %d", ret);
371 rxrpc_put_txbuf(txb, rxrpc_txbuf_put_send_aborted);
372 _leave(" = %d", call->error);
386 if (msg->msg_flags & MSG_DONTWAIT)
388 mutex_unlock(&call->user_mutex);
389 *_dropped_lock = true;
390 ret = rxrpc_wait_for_tx_window(rx, call, &timeo,
391 msg->msg_flags & MSG_WAITALL);
394 if (call->interruptibility == RXRPC_INTERRUPTIBLE) {
395 if (mutex_lock_interruptible(&call->user_mutex) < 0) {
396 ret = sock_intr_errno(timeo);
400 mutex_lock(&call->user_mutex);
402 *_dropped_lock = false;
407 * extract control messages from the sendmsg() control buffer
409 static int rxrpc_sendmsg_cmsg(struct msghdr *msg, struct rxrpc_send_params *p)
411 struct cmsghdr *cmsg;
412 bool got_user_ID = false;
415 if (msg->msg_controllen == 0)
418 for_each_cmsghdr(cmsg, msg) {
419 if (!CMSG_OK(msg, cmsg))
422 len = cmsg->cmsg_len - sizeof(struct cmsghdr);
423 _debug("CMSG %d, %d, %d",
424 cmsg->cmsg_level, cmsg->cmsg_type, len);
426 if (cmsg->cmsg_level != SOL_RXRPC)
429 switch (cmsg->cmsg_type) {
430 case RXRPC_USER_CALL_ID:
431 if (msg->msg_flags & MSG_CMSG_COMPAT) {
432 if (len != sizeof(u32))
434 p->call.user_call_ID = *(u32 *)CMSG_DATA(cmsg);
436 if (len != sizeof(unsigned long))
438 p->call.user_call_ID = *(unsigned long *)
445 if (p->command != RXRPC_CMD_SEND_DATA)
447 p->command = RXRPC_CMD_SEND_ABORT;
448 if (len != sizeof(p->abort_code))
450 p->abort_code = *(unsigned int *)CMSG_DATA(cmsg);
451 if (p->abort_code == 0)
455 case RXRPC_CHARGE_ACCEPT:
456 if (p->command != RXRPC_CMD_SEND_DATA)
458 p->command = RXRPC_CMD_CHARGE_ACCEPT;
463 case RXRPC_EXCLUSIVE_CALL:
469 case RXRPC_UPGRADE_SERVICE:
475 case RXRPC_TX_LENGTH:
476 if (p->call.tx_total_len != -1 || len != sizeof(__s64))
478 p->call.tx_total_len = *(__s64 *)CMSG_DATA(cmsg);
479 if (p->call.tx_total_len < 0)
483 case RXRPC_SET_CALL_TIMEOUT:
484 if (len & 3 || len < 4 || len > 12)
486 memcpy(&p->call.timeouts, CMSG_DATA(cmsg), len);
487 p->call.nr_timeouts = len / 4;
488 if (p->call.timeouts.hard > INT_MAX / HZ)
490 if (p->call.nr_timeouts >= 2 && p->call.timeouts.idle > 60 * 60 * 1000)
492 if (p->call.nr_timeouts >= 3 && p->call.timeouts.normal > 60 * 60 * 1000)
503 if (p->call.tx_total_len != -1 && p->command != RXRPC_CMD_SEND_DATA)
510 * Create a new client call for sendmsg().
511 * - Called with the socket lock held, which it must release.
512 * - If it returns a call, the call's lock will need releasing by the caller.
514 static struct rxrpc_call *
515 rxrpc_new_client_call_for_sendmsg(struct rxrpc_sock *rx, struct msghdr *msg,
516 struct rxrpc_send_params *p)
517 __releases(&rx->sk.sk_lock.slock)
518 __acquires(&call->user_mutex)
520 struct rxrpc_conn_parameters cp;
521 struct rxrpc_call *call;
524 DECLARE_SOCKADDR(struct sockaddr_rxrpc *, srx, msg->msg_name);
528 if (!msg->msg_name) {
529 release_sock(&rx->sk);
530 return ERR_PTR(-EDESTADDRREQ);
534 if (key && !rx->key->payload.data[0])
537 memset(&cp, 0, sizeof(cp));
538 cp.local = rx->local;
540 cp.security_level = rx->min_sec_level;
541 cp.exclusive = rx->exclusive | p->exclusive;
542 cp.upgrade = p->upgrade;
543 cp.service_id = srx->srx_service;
544 call = rxrpc_new_client_call(rx, &cp, srx, &p->call, GFP_KERNEL,
545 atomic_inc_return(&rxrpc_debug_id));
546 /* The socket is now unlocked */
548 rxrpc_put_peer(cp.peer, rxrpc_peer_put_discard_tmp);
549 _leave(" = %p\n", call);
554 * send a message forming part of a client call through an RxRPC socket
555 * - caller holds the socket locked
556 * - the socket may be either a client socket or a server socket
558 int rxrpc_do_sendmsg(struct rxrpc_sock *rx, struct msghdr *msg, size_t len)
559 __releases(&rx->sk.sk_lock.slock)
561 enum rxrpc_call_state state;
562 struct rxrpc_call *call;
563 unsigned long now, j;
564 bool dropped_lock = false;
567 struct rxrpc_send_params p = {
568 .call.tx_total_len = -1,
569 .call.user_call_ID = 0,
570 .call.nr_timeouts = 0,
571 .call.interruptibility = RXRPC_INTERRUPTIBLE,
573 .command = RXRPC_CMD_SEND_DATA,
580 ret = rxrpc_sendmsg_cmsg(msg, &p);
582 goto error_release_sock;
584 if (p.command == RXRPC_CMD_CHARGE_ACCEPT) {
586 if (rx->sk.sk_state != RXRPC_SERVER_LISTENING)
587 goto error_release_sock;
588 ret = rxrpc_user_charge_accept(rx, p.call.user_call_ID);
589 goto error_release_sock;
592 call = rxrpc_find_call_by_user_ID(rx, p.call.user_call_ID);
595 if (p.command != RXRPC_CMD_SEND_DATA)
596 goto error_release_sock;
597 call = rxrpc_new_client_call_for_sendmsg(rx, msg, &p);
598 /* The socket is now unlocked... */
600 return PTR_ERR(call);
601 /* ... and we have the call lock. */
603 if (READ_ONCE(call->state) == RXRPC_CALL_COMPLETE)
606 switch (READ_ONCE(call->state)) {
607 case RXRPC_CALL_UNINITIALISED:
608 case RXRPC_CALL_CLIENT_AWAIT_CONN:
609 case RXRPC_CALL_SERVER_PREALLOC:
610 case RXRPC_CALL_SERVER_SECURING:
611 rxrpc_put_call(call, rxrpc_call_put_sendmsg);
613 goto error_release_sock;
618 ret = mutex_lock_interruptible(&call->user_mutex);
619 release_sock(&rx->sk);
625 if (p.call.tx_total_len != -1) {
627 if (call->tx_total_len != -1 ||
631 call->tx_total_len = p.call.tx_total_len;
635 switch (p.call.nr_timeouts) {
637 j = msecs_to_jiffies(p.call.timeouts.normal);
638 if (p.call.timeouts.normal > 0 && j == 0)
640 WRITE_ONCE(call->next_rx_timo, j);
643 j = msecs_to_jiffies(p.call.timeouts.idle);
644 if (p.call.timeouts.idle > 0 && j == 0)
646 WRITE_ONCE(call->next_req_timo, j);
649 if (p.call.timeouts.hard > 0) {
650 j = msecs_to_jiffies(p.call.timeouts.hard);
653 WRITE_ONCE(call->expect_term_by, j);
654 rxrpc_reduce_call_timer(call, j, now,
655 rxrpc_timer_set_for_hard);
660 state = READ_ONCE(call->state);
661 _debug("CALL %d USR %lx ST %d on CONN %p",
662 call->debug_id, call->user_call_ID, state, call->conn);
664 if (state >= RXRPC_CALL_COMPLETE) {
665 /* it's too late for this call */
667 } else if (p.command == RXRPC_CMD_SEND_ABORT) {
669 if (rxrpc_abort_call("CMD", call, 0, p.abort_code, -ECONNABORTED))
670 ret = rxrpc_send_abort_packet(call);
671 } else if (p.command != RXRPC_CMD_SEND_DATA) {
674 ret = rxrpc_send_data(rx, call, msg, len, NULL, &dropped_lock);
679 mutex_unlock(&call->user_mutex);
681 rxrpc_put_call(call, rxrpc_call_put_sendmsg);
682 _leave(" = %d", ret);
686 release_sock(&rx->sk);
691 * rxrpc_kernel_send_data - Allow a kernel service to send data on a call
692 * @sock: The socket the call is on
693 * @call: The call to send data through
694 * @msg: The data to send
695 * @len: The amount of data to send
696 * @notify_end_tx: Notification that the last packet is queued.
698 * Allow a kernel service to send data on a call. The call must be in an state
699 * appropriate to sending data. No control data should be supplied in @msg,
700 * nor should an address be supplied. MSG_MORE should be flagged if there's
701 * more data to come, otherwise this data will end the transmission phase.
703 int rxrpc_kernel_send_data(struct socket *sock, struct rxrpc_call *call,
704 struct msghdr *msg, size_t len,
705 rxrpc_notify_end_tx_t notify_end_tx)
707 bool dropped_lock = false;
710 _enter("{%d,%s},", call->debug_id, rxrpc_call_states[call->state]);
712 ASSERTCMP(msg->msg_name, ==, NULL);
713 ASSERTCMP(msg->msg_control, ==, NULL);
715 mutex_lock(&call->user_mutex);
717 _debug("CALL %d USR %lx ST %d on CONN %p",
718 call->debug_id, call->user_call_ID, call->state, call->conn);
720 switch (READ_ONCE(call->state)) {
721 case RXRPC_CALL_CLIENT_SEND_REQUEST:
722 case RXRPC_CALL_SERVER_ACK_REQUEST:
723 case RXRPC_CALL_SERVER_SEND_REPLY:
724 ret = rxrpc_send_data(rxrpc_sk(sock->sk), call, msg, len,
725 notify_end_tx, &dropped_lock);
727 case RXRPC_CALL_COMPLETE:
728 read_lock_bh(&call->state_lock);
730 read_unlock_bh(&call->state_lock);
733 /* Request phase complete for this client call */
734 trace_rxrpc_rx_eproto(call, 0, tracepoint_string("late_send"));
740 mutex_unlock(&call->user_mutex);
741 _leave(" = %d", ret);
744 EXPORT_SYMBOL(rxrpc_kernel_send_data);
747 * rxrpc_kernel_abort_call - Allow a kernel service to abort a call
748 * @sock: The socket the call is on
749 * @call: The call to be aborted
750 * @abort_code: The abort code to stick into the ABORT packet
751 * @error: Local error value
752 * @why: 3-char string indicating why.
754 * Allow a kernel service to abort a call, if it's still in an abortable state
755 * and return true if the call was aborted, false if it was already complete.
757 bool rxrpc_kernel_abort_call(struct socket *sock, struct rxrpc_call *call,
758 u32 abort_code, int error, const char *why)
762 _enter("{%d},%d,%d,%s", call->debug_id, abort_code, error, why);
764 mutex_lock(&call->user_mutex);
766 aborted = rxrpc_abort_call(why, call, 0, abort_code, error);
768 rxrpc_send_abort_packet(call);
770 mutex_unlock(&call->user_mutex);
773 EXPORT_SYMBOL(rxrpc_kernel_abort_call);
776 * rxrpc_kernel_set_tx_length - Set the total Tx length on a call
777 * @sock: The socket the call is on
778 * @call: The call to be informed
779 * @tx_total_len: The amount of data to be transmitted for this call
781 * Allow a kernel service to set the total transmit length on a call. This
782 * allows buffer-to-packet encrypt-and-copy to be performed.
784 * This function is primarily for use for setting the reply length since the
785 * request length can be set when beginning the call.
787 void rxrpc_kernel_set_tx_length(struct socket *sock, struct rxrpc_call *call,
790 WARN_ON(call->tx_total_len != -1);
791 call->tx_total_len = tx_total_len;
793 EXPORT_SYMBOL(rxrpc_kernel_set_tx_length);