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.
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)
25 unsigned int win_size =
26 min_t(unsigned int, call->tx_winsize,
27 call->cong_cwnd + call->cong_extra);
28 rxrpc_seq_t tx_win = READ_ONCE(call->tx_hard_ack);
32 return call->tx_top - tx_win < win_size;
36 * Wait for space to appear in the Tx queue or a signal to occur.
38 static int rxrpc_wait_for_tx_window_intr(struct rxrpc_sock *rx,
39 struct rxrpc_call *call,
43 set_current_state(TASK_INTERRUPTIBLE);
44 if (rxrpc_check_tx_space(call, NULL))
47 if (call->state >= RXRPC_CALL_COMPLETE)
50 if (signal_pending(current))
51 return sock_intr_errno(*timeo);
53 trace_rxrpc_transmit(call, rxrpc_transmit_wait);
54 mutex_unlock(&call->user_mutex);
55 *timeo = schedule_timeout(*timeo);
56 if (mutex_lock_interruptible(&call->user_mutex) < 0)
57 return sock_intr_errno(*timeo);
62 * Wait for space to appear in the Tx queue uninterruptibly, but with
63 * a timeout of 2*RTT if no progress was made and a signal occurred.
65 static int rxrpc_wait_for_tx_window_waitall(struct rxrpc_sock *rx,
66 struct rxrpc_call *call)
68 rxrpc_seq_t tx_start, tx_win;
69 signed long rtt, timeout;
71 rtt = READ_ONCE(call->peer->srtt_us) >> 3;
72 rtt = usecs_to_jiffies(rtt) * 2;
77 tx_start = READ_ONCE(call->tx_hard_ack);
80 set_current_state(TASK_UNINTERRUPTIBLE);
82 tx_win = READ_ONCE(call->tx_hard_ack);
83 if (rxrpc_check_tx_space(call, &tx_win))
86 if (call->state >= RXRPC_CALL_COMPLETE)
90 tx_win == tx_start && signal_pending(current))
93 if (tx_win != tx_start) {
98 trace_rxrpc_transmit(call, rxrpc_transmit_wait);
99 timeout = schedule_timeout(timeout);
104 * Wait for space to appear in the Tx queue uninterruptibly.
106 static int rxrpc_wait_for_tx_window_nonintr(struct rxrpc_sock *rx,
107 struct rxrpc_call *call,
111 set_current_state(TASK_UNINTERRUPTIBLE);
112 if (rxrpc_check_tx_space(call, NULL))
115 if (call->state >= RXRPC_CALL_COMPLETE)
118 trace_rxrpc_transmit(call, rxrpc_transmit_wait);
119 *timeo = schedule_timeout(*timeo);
124 * wait for space to appear in the transmit/ACK window
125 * - caller holds the socket locked
127 static int rxrpc_wait_for_tx_window(struct rxrpc_sock *rx,
128 struct rxrpc_call *call,
132 DECLARE_WAITQUEUE(myself, current);
135 _enter(",{%u,%u,%u}",
136 call->tx_hard_ack, call->tx_top, call->tx_winsize);
138 add_wait_queue(&call->waitq, &myself);
140 switch (call->interruptibility) {
141 case RXRPC_INTERRUPTIBLE:
143 ret = rxrpc_wait_for_tx_window_waitall(rx, call);
145 ret = rxrpc_wait_for_tx_window_intr(rx, call, timeo);
147 case RXRPC_PREINTERRUPTIBLE:
148 case RXRPC_UNINTERRUPTIBLE:
150 ret = rxrpc_wait_for_tx_window_nonintr(rx, call, timeo);
154 remove_wait_queue(&call->waitq, &myself);
155 set_current_state(TASK_RUNNING);
156 _leave(" = %d", ret);
161 * Schedule an instant Tx resend.
163 static inline void rxrpc_instant_resend(struct rxrpc_call *call, int ix)
165 spin_lock_bh(&call->lock);
167 if (call->state < RXRPC_CALL_COMPLETE) {
168 call->rxtx_annotations[ix] =
169 (call->rxtx_annotations[ix] & RXRPC_TX_ANNO_LAST) |
170 RXRPC_TX_ANNO_RETRANS;
171 if (!test_and_set_bit(RXRPC_CALL_EV_RESEND, &call->events))
172 rxrpc_queue_call(call);
175 spin_unlock_bh(&call->lock);
179 * Notify the owner of the call that the transmit phase is ended and the last
180 * packet has been queued.
182 static void rxrpc_notify_end_tx(struct rxrpc_sock *rx, struct rxrpc_call *call,
183 rxrpc_notify_end_tx_t notify_end_tx)
186 notify_end_tx(&rx->sk, call, call->user_call_ID);
190 * Queue a DATA packet for transmission, set the resend timeout and send
191 * the packet immediately. Returns the error from rxrpc_send_data_packet()
192 * in case the caller wants to do something with it.
194 static int rxrpc_queue_packet(struct rxrpc_sock *rx, struct rxrpc_call *call,
195 struct sk_buff *skb, bool last,
196 rxrpc_notify_end_tx_t notify_end_tx)
198 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
200 rxrpc_seq_t seq = sp->hdr.seq;
202 u8 annotation = RXRPC_TX_ANNO_UNACK;
204 _net("queue skb %p [%d]", skb, seq);
206 ASSERTCMP(seq, ==, call->tx_top + 1);
209 annotation |= RXRPC_TX_ANNO_LAST;
211 /* We have to set the timestamp before queueing as the retransmit
212 * algorithm can see the packet as soon as we queue it.
214 skb->tstamp = ktime_get_real();
216 ix = seq & RXRPC_RXTX_BUFF_MASK;
217 rxrpc_get_skb(skb, rxrpc_skb_got);
218 call->rxtx_annotations[ix] = annotation;
220 call->rxtx_buffer[ix] = skb;
223 trace_rxrpc_transmit(call, rxrpc_transmit_queue_last);
225 trace_rxrpc_transmit(call, rxrpc_transmit_queue);
227 if (last || call->state == RXRPC_CALL_SERVER_ACK_REQUEST) {
228 _debug("________awaiting reply/ACK__________");
229 write_lock_bh(&call->state_lock);
230 switch (call->state) {
231 case RXRPC_CALL_CLIENT_SEND_REQUEST:
232 call->state = RXRPC_CALL_CLIENT_AWAIT_REPLY;
233 rxrpc_notify_end_tx(rx, call, notify_end_tx);
235 case RXRPC_CALL_SERVER_ACK_REQUEST:
236 call->state = RXRPC_CALL_SERVER_SEND_REPLY;
238 WRITE_ONCE(call->ack_at, now + MAX_JIFFY_OFFSET);
239 if (call->ackr_reason == RXRPC_ACK_DELAY)
240 call->ackr_reason = 0;
241 trace_rxrpc_timer(call, rxrpc_timer_init_for_send_reply, now);
245 case RXRPC_CALL_SERVER_SEND_REPLY:
246 call->state = RXRPC_CALL_SERVER_AWAIT_ACK;
247 rxrpc_notify_end_tx(rx, call, notify_end_tx);
252 write_unlock_bh(&call->state_lock);
255 if (seq == 1 && rxrpc_is_client_call(call))
256 rxrpc_expose_client_call(call);
258 ret = rxrpc_send_data_packet(call, skb, false);
264 rxrpc_set_call_completion(call, RXRPC_CALL_LOCAL_ERROR,
268 _debug("need instant resend %d", ret);
269 rxrpc_instant_resend(call, ix);
271 unsigned long now = jiffies;
272 unsigned long resend_at = now + call->peer->rto_j;
274 WRITE_ONCE(call->resend_at, resend_at);
275 rxrpc_reduce_call_timer(call, resend_at, now,
276 rxrpc_timer_set_for_send);
280 rxrpc_free_skb(skb, rxrpc_skb_freed);
281 _leave(" = %d", ret);
286 * send data through a socket
287 * - must be called in process context
288 * - The caller holds the call user access mutex, but not the socket lock.
290 static int rxrpc_send_data(struct rxrpc_sock *rx,
291 struct rxrpc_call *call,
292 struct msghdr *msg, size_t len,
293 rxrpc_notify_end_tx_t notify_end_tx)
295 struct rxrpc_skb_priv *sp;
297 struct sock *sk = &rx->sk;
302 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
304 /* this should be in poll */
305 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
307 if (sk->sk_shutdown & SEND_SHUTDOWN)
310 more = msg->msg_flags & MSG_MORE;
312 if (call->tx_total_len != -1) {
313 if (len > call->tx_total_len)
315 if (!more && len != call->tx_total_len)
319 skb = call->tx_pending;
320 call->tx_pending = NULL;
321 rxrpc_see_skb(skb, rxrpc_skb_seen);
325 /* Check to see if there's a ping ACK to reply to. */
326 if (call->ackr_reason == RXRPC_ACK_PING_RESPONSE)
327 rxrpc_send_ack_packet(call, false, NULL);
330 size_t remain, bufsize, chunk, offset;
334 if (!rxrpc_check_tx_space(call, NULL)) {
336 if (msg->msg_flags & MSG_DONTWAIT)
338 ret = rxrpc_wait_for_tx_window(rx, call,
340 msg->msg_flags & MSG_WAITALL);
345 /* Work out the maximum size of a packet. Assume that
346 * the security header is going to be in the padded
347 * region (enc blocksize), but the trailer is not.
349 remain = more ? INT_MAX : msg_data_left(msg);
350 ret = call->conn->security->how_much_data(call, remain,
351 &bufsize, &chunk, &offset);
355 _debug("SIZE: %zu/%zu @%zu", chunk, bufsize, offset);
357 /* create a buffer that we can retain until it's ACK'd */
358 skb = sock_alloc_send_skb(
359 sk, bufsize, msg->msg_flags & MSG_DONTWAIT, &ret);
364 sp->rx_flags |= RXRPC_SKB_TX_BUFFER;
365 rxrpc_new_skb(skb, rxrpc_skb_new);
367 _debug("ALLOC SEND %p", skb);
369 ASSERTCMP(skb->mark, ==, 0);
371 __skb_put(skb, offset);
374 if (sp->remain > skb_tailroom(skb))
375 sp->remain = skb_tailroom(skb);
377 _net("skb: hr %d, tr %d, hl %d, rm %d",
383 skb->ip_summed = CHECKSUM_UNNECESSARY;
389 /* append next segment of data to the current buffer */
390 if (msg_data_left(msg) > 0) {
391 int copy = skb_tailroom(skb);
392 ASSERTCMP(copy, >, 0);
393 if (copy > msg_data_left(msg))
394 copy = msg_data_left(msg);
395 if (copy > sp->remain)
399 ret = skb_add_data(skb, &msg->msg_iter, copy);
406 if (call->tx_total_len != -1)
407 call->tx_total_len -= copy;
410 /* check for the far side aborting the call or a network error
412 if (call->state == RXRPC_CALL_COMPLETE)
413 goto call_terminated;
415 /* add the packet to the send queue if it's now full */
416 if (sp->remain <= 0 ||
417 (msg_data_left(msg) == 0 && !more)) {
418 struct rxrpc_connection *conn = call->conn;
421 seq = call->tx_top + 1;
425 sp->hdr.flags = conn->out_clientflag;
427 if (msg_data_left(msg) == 0 && !more)
428 sp->hdr.flags |= RXRPC_LAST_PACKET;
429 else if (call->tx_top - call->tx_hard_ack <
431 sp->hdr.flags |= RXRPC_MORE_PACKETS;
433 ret = call->security->secure_packet(call, skb, skb->mark);
437 ret = rxrpc_queue_packet(rx, call, skb,
438 !msg_data_left(msg) && !more,
440 /* Should check for failure here */
443 } while (msg_data_left(msg) > 0);
447 if (READ_ONCE(call->state) == RXRPC_CALL_COMPLETE) {
448 read_lock_bh(&call->state_lock);
451 read_unlock_bh(&call->state_lock);
454 call->tx_pending = skb;
455 _leave(" = %d", ret);
459 rxrpc_free_skb(skb, rxrpc_skb_freed);
460 _leave(" = %d", call->error);
474 * extract control messages from the sendmsg() control buffer
476 static int rxrpc_sendmsg_cmsg(struct msghdr *msg, struct rxrpc_send_params *p)
478 struct cmsghdr *cmsg;
479 bool got_user_ID = false;
482 if (msg->msg_controllen == 0)
485 for_each_cmsghdr(cmsg, msg) {
486 if (!CMSG_OK(msg, cmsg))
489 len = cmsg->cmsg_len - sizeof(struct cmsghdr);
490 _debug("CMSG %d, %d, %d",
491 cmsg->cmsg_level, cmsg->cmsg_type, len);
493 if (cmsg->cmsg_level != SOL_RXRPC)
496 switch (cmsg->cmsg_type) {
497 case RXRPC_USER_CALL_ID:
498 if (msg->msg_flags & MSG_CMSG_COMPAT) {
499 if (len != sizeof(u32))
501 p->call.user_call_ID = *(u32 *)CMSG_DATA(cmsg);
503 if (len != sizeof(unsigned long))
505 p->call.user_call_ID = *(unsigned long *)
512 if (p->command != RXRPC_CMD_SEND_DATA)
514 p->command = RXRPC_CMD_SEND_ABORT;
515 if (len != sizeof(p->abort_code))
517 p->abort_code = *(unsigned int *)CMSG_DATA(cmsg);
518 if (p->abort_code == 0)
522 case RXRPC_CHARGE_ACCEPT:
523 if (p->command != RXRPC_CMD_SEND_DATA)
525 p->command = RXRPC_CMD_CHARGE_ACCEPT;
530 case RXRPC_EXCLUSIVE_CALL:
536 case RXRPC_UPGRADE_SERVICE:
542 case RXRPC_TX_LENGTH:
543 if (p->call.tx_total_len != -1 || len != sizeof(__s64))
545 p->call.tx_total_len = *(__s64 *)CMSG_DATA(cmsg);
546 if (p->call.tx_total_len < 0)
550 case RXRPC_SET_CALL_TIMEOUT:
551 if (len & 3 || len < 4 || len > 12)
553 memcpy(&p->call.timeouts, CMSG_DATA(cmsg), len);
554 p->call.nr_timeouts = len / 4;
555 if (p->call.timeouts.hard > INT_MAX / HZ)
557 if (p->call.nr_timeouts >= 2 && p->call.timeouts.idle > 60 * 60 * 1000)
559 if (p->call.nr_timeouts >= 3 && p->call.timeouts.normal > 60 * 60 * 1000)
570 if (p->call.tx_total_len != -1 && p->command != RXRPC_CMD_SEND_DATA)
577 * Create a new client call for sendmsg().
578 * - Called with the socket lock held, which it must release.
579 * - If it returns a call, the call's lock will need releasing by the caller.
581 static struct rxrpc_call *
582 rxrpc_new_client_call_for_sendmsg(struct rxrpc_sock *rx, struct msghdr *msg,
583 struct rxrpc_send_params *p)
584 __releases(&rx->sk.sk_lock.slock)
585 __acquires(&call->user_mutex)
587 struct rxrpc_conn_parameters cp;
588 struct rxrpc_call *call;
591 DECLARE_SOCKADDR(struct sockaddr_rxrpc *, srx, msg->msg_name);
595 if (!msg->msg_name) {
596 release_sock(&rx->sk);
597 return ERR_PTR(-EDESTADDRREQ);
601 if (key && !rx->key->payload.data[0])
604 memset(&cp, 0, sizeof(cp));
605 cp.local = rx->local;
607 cp.security_level = rx->min_sec_level;
608 cp.exclusive = rx->exclusive | p->exclusive;
609 cp.upgrade = p->upgrade;
610 cp.service_id = srx->srx_service;
611 call = rxrpc_new_client_call(rx, &cp, srx, &p->call, GFP_KERNEL,
612 atomic_inc_return(&rxrpc_debug_id));
613 /* The socket is now unlocked */
615 rxrpc_put_peer(cp.peer);
616 _leave(" = %p\n", call);
621 * send a message forming part of a client call through an RxRPC socket
622 * - caller holds the socket locked
623 * - the socket may be either a client socket or a server socket
625 int rxrpc_do_sendmsg(struct rxrpc_sock *rx, struct msghdr *msg, size_t len)
626 __releases(&rx->sk.sk_lock.slock)
627 __releases(&call->user_mutex)
629 enum rxrpc_call_state state;
630 struct rxrpc_call *call;
631 unsigned long now, j;
634 struct rxrpc_send_params p = {
635 .call.tx_total_len = -1,
636 .call.user_call_ID = 0,
637 .call.nr_timeouts = 0,
638 .call.interruptibility = RXRPC_INTERRUPTIBLE,
640 .command = RXRPC_CMD_SEND_DATA,
647 ret = rxrpc_sendmsg_cmsg(msg, &p);
649 goto error_release_sock;
651 if (p.command == RXRPC_CMD_CHARGE_ACCEPT) {
653 if (rx->sk.sk_state != RXRPC_SERVER_LISTENING)
654 goto error_release_sock;
655 ret = rxrpc_user_charge_accept(rx, p.call.user_call_ID);
656 goto error_release_sock;
659 call = rxrpc_find_call_by_user_ID(rx, p.call.user_call_ID);
662 if (p.command != RXRPC_CMD_SEND_DATA)
663 goto error_release_sock;
664 call = rxrpc_new_client_call_for_sendmsg(rx, msg, &p);
665 /* The socket is now unlocked... */
667 return PTR_ERR(call);
668 /* ... and we have the call lock. */
670 if (READ_ONCE(call->state) == RXRPC_CALL_COMPLETE)
673 switch (READ_ONCE(call->state)) {
674 case RXRPC_CALL_UNINITIALISED:
675 case RXRPC_CALL_CLIENT_AWAIT_CONN:
676 case RXRPC_CALL_SERVER_PREALLOC:
677 case RXRPC_CALL_SERVER_SECURING:
678 rxrpc_put_call(call, rxrpc_call_put);
680 goto error_release_sock;
685 ret = mutex_lock_interruptible(&call->user_mutex);
686 release_sock(&rx->sk);
692 if (p.call.tx_total_len != -1) {
694 if (call->tx_total_len != -1 ||
698 call->tx_total_len = p.call.tx_total_len;
702 switch (p.call.nr_timeouts) {
704 j = msecs_to_jiffies(p.call.timeouts.normal);
705 if (p.call.timeouts.normal > 0 && j == 0)
707 WRITE_ONCE(call->next_rx_timo, j);
710 j = msecs_to_jiffies(p.call.timeouts.idle);
711 if (p.call.timeouts.idle > 0 && j == 0)
713 WRITE_ONCE(call->next_req_timo, j);
716 if (p.call.timeouts.hard > 0) {
717 j = msecs_to_jiffies(p.call.timeouts.hard);
720 WRITE_ONCE(call->expect_term_by, j);
721 rxrpc_reduce_call_timer(call, j, now,
722 rxrpc_timer_set_for_hard);
727 state = READ_ONCE(call->state);
728 _debug("CALL %d USR %lx ST %d on CONN %p",
729 call->debug_id, call->user_call_ID, state, call->conn);
731 if (state >= RXRPC_CALL_COMPLETE) {
732 /* it's too late for this call */
734 } else if (p.command == RXRPC_CMD_SEND_ABORT) {
736 if (rxrpc_abort_call("CMD", call, 0, p.abort_code, -ECONNABORTED))
737 ret = rxrpc_send_abort_packet(call);
738 } else if (p.command != RXRPC_CMD_SEND_DATA) {
740 } else if (rxrpc_is_client_call(call) &&
741 state != RXRPC_CALL_CLIENT_SEND_REQUEST) {
742 /* request phase complete for this client call */
744 } else if (rxrpc_is_service_call(call) &&
745 state != RXRPC_CALL_SERVER_ACK_REQUEST &&
746 state != RXRPC_CALL_SERVER_SEND_REPLY) {
747 /* Reply phase not begun or not complete for service call. */
750 ret = rxrpc_send_data(rx, call, msg, len, NULL);
754 mutex_unlock(&call->user_mutex);
756 rxrpc_put_call(call, rxrpc_call_put);
757 _leave(" = %d", ret);
761 release_sock(&rx->sk);
766 * rxrpc_kernel_send_data - Allow a kernel service to send data on a call
767 * @sock: The socket the call is on
768 * @call: The call to send data through
769 * @msg: The data to send
770 * @len: The amount of data to send
771 * @notify_end_tx: Notification that the last packet is queued.
773 * Allow a kernel service to send data on a call. The call must be in an state
774 * appropriate to sending data. No control data should be supplied in @msg,
775 * nor should an address be supplied. MSG_MORE should be flagged if there's
776 * more data to come, otherwise this data will end the transmission phase.
778 int rxrpc_kernel_send_data(struct socket *sock, struct rxrpc_call *call,
779 struct msghdr *msg, size_t len,
780 rxrpc_notify_end_tx_t notify_end_tx)
784 _enter("{%d,%s},", call->debug_id, rxrpc_call_states[call->state]);
786 ASSERTCMP(msg->msg_name, ==, NULL);
787 ASSERTCMP(msg->msg_control, ==, NULL);
789 mutex_lock(&call->user_mutex);
791 _debug("CALL %d USR %lx ST %d on CONN %p",
792 call->debug_id, call->user_call_ID, call->state, call->conn);
794 switch (READ_ONCE(call->state)) {
795 case RXRPC_CALL_CLIENT_SEND_REQUEST:
796 case RXRPC_CALL_SERVER_ACK_REQUEST:
797 case RXRPC_CALL_SERVER_SEND_REPLY:
798 ret = rxrpc_send_data(rxrpc_sk(sock->sk), call, msg, len,
801 case RXRPC_CALL_COMPLETE:
802 read_lock_bh(&call->state_lock);
804 read_unlock_bh(&call->state_lock);
807 /* Request phase complete for this client call */
808 trace_rxrpc_rx_eproto(call, 0, tracepoint_string("late_send"));
813 mutex_unlock(&call->user_mutex);
814 _leave(" = %d", ret);
817 EXPORT_SYMBOL(rxrpc_kernel_send_data);
820 * rxrpc_kernel_abort_call - Allow a kernel service to abort a call
821 * @sock: The socket the call is on
822 * @call: The call to be aborted
823 * @abort_code: The abort code to stick into the ABORT packet
824 * @error: Local error value
825 * @why: 3-char string indicating why.
827 * Allow a kernel service to abort a call, if it's still in an abortable state
828 * and return true if the call was aborted, false if it was already complete.
830 bool rxrpc_kernel_abort_call(struct socket *sock, struct rxrpc_call *call,
831 u32 abort_code, int error, const char *why)
835 _enter("{%d},%d,%d,%s", call->debug_id, abort_code, error, why);
837 mutex_lock(&call->user_mutex);
839 aborted = rxrpc_abort_call(why, call, 0, abort_code, error);
841 rxrpc_send_abort_packet(call);
843 mutex_unlock(&call->user_mutex);
846 EXPORT_SYMBOL(rxrpc_kernel_abort_call);
849 * rxrpc_kernel_set_tx_length - Set the total Tx length on a call
850 * @sock: The socket the call is on
851 * @call: The call to be informed
852 * @tx_total_len: The amount of data to be transmitted for this call
854 * Allow a kernel service to set the total transmit length on a call. This
855 * allows buffer-to-packet encrypt-and-copy to be performed.
857 * This function is primarily for use for setting the reply length since the
858 * request length can be set when beginning the call.
860 void rxrpc_kernel_set_tx_length(struct socket *sock, struct rxrpc_call *call,
863 WARN_ON(call->tx_total_len != -1);
864 call->tx_total_len = tx_total_len;
866 EXPORT_SYMBOL(rxrpc_kernel_set_tx_length);