1 /* AF_RXRPC sendmsg() implementation.
3 * Copyright (C) 2007, 2016 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public Licence
8 * as published by the Free Software Foundation; either version
9 * 2 of the Licence, or (at your option) any later version.
12 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14 #include <linux/net.h>
15 #include <linux/gfp.h>
16 #include <linux/skbuff.h>
17 #include <linux/export.h>
18 #include <linux/sched/signal.h>
21 #include <net/af_rxrpc.h>
22 #include "ar-internal.h"
25 * Wait for space to appear in the Tx queue or a signal to occur.
27 static int rxrpc_wait_for_tx_window_intr(struct rxrpc_sock *rx,
28 struct rxrpc_call *call,
32 set_current_state(TASK_INTERRUPTIBLE);
33 if (call->tx_top - call->tx_hard_ack <
34 min_t(unsigned int, call->tx_winsize,
35 call->cong_cwnd + call->cong_extra))
38 if (call->state >= RXRPC_CALL_COMPLETE)
41 if (signal_pending(current))
42 return sock_intr_errno(*timeo);
44 trace_rxrpc_transmit(call, rxrpc_transmit_wait);
45 mutex_unlock(&call->user_mutex);
46 *timeo = schedule_timeout(*timeo);
47 if (mutex_lock_interruptible(&call->user_mutex) < 0)
48 return sock_intr_errno(*timeo);
53 * Wait for space to appear in the Tx queue uninterruptibly, but with
54 * a timeout of 2*RTT if no progress was made and a signal occurred.
56 static int rxrpc_wait_for_tx_window_nonintr(struct rxrpc_sock *rx,
57 struct rxrpc_call *call)
59 rxrpc_seq_t tx_start, tx_win;
60 signed long rtt2, timeout;
63 rtt = READ_ONCE(call->peer->rtt);
64 rtt2 = nsecs_to_jiffies64(rtt) * 2;
69 tx_start = READ_ONCE(call->tx_hard_ack);
72 set_current_state(TASK_UNINTERRUPTIBLE);
74 tx_win = READ_ONCE(call->tx_hard_ack);
75 if (call->tx_top - tx_win <
76 min_t(unsigned int, call->tx_winsize,
77 call->cong_cwnd + call->cong_extra))
80 if (call->state >= RXRPC_CALL_COMPLETE)
84 tx_win == tx_start && signal_pending(current))
87 if (tx_win != tx_start) {
92 trace_rxrpc_transmit(call, rxrpc_transmit_wait);
93 timeout = schedule_timeout(timeout);
98 * wait for space to appear in the transmit/ACK window
99 * - caller holds the socket locked
101 static int rxrpc_wait_for_tx_window(struct rxrpc_sock *rx,
102 struct rxrpc_call *call,
106 DECLARE_WAITQUEUE(myself, current);
109 _enter(",{%u,%u,%u}",
110 call->tx_hard_ack, call->tx_top, call->tx_winsize);
112 add_wait_queue(&call->waitq, &myself);
115 ret = rxrpc_wait_for_tx_window_nonintr(rx, call);
117 ret = rxrpc_wait_for_tx_window_intr(rx, call, timeo);
119 remove_wait_queue(&call->waitq, &myself);
120 set_current_state(TASK_RUNNING);
121 _leave(" = %d", ret);
126 * Schedule an instant Tx resend.
128 static inline void rxrpc_instant_resend(struct rxrpc_call *call, int ix)
130 spin_lock_bh(&call->lock);
132 if (call->state < RXRPC_CALL_COMPLETE) {
133 call->rxtx_annotations[ix] = RXRPC_TX_ANNO_RETRANS;
134 if (!test_and_set_bit(RXRPC_CALL_EV_RESEND, &call->events))
135 rxrpc_queue_call(call);
138 spin_unlock_bh(&call->lock);
142 * Notify the owner of the call that the transmit phase is ended and the last
143 * packet has been queued.
145 static void rxrpc_notify_end_tx(struct rxrpc_sock *rx, struct rxrpc_call *call,
146 rxrpc_notify_end_tx_t notify_end_tx)
149 notify_end_tx(&rx->sk, call, call->user_call_ID);
153 * Queue a DATA packet for transmission, set the resend timeout and send the
156 static void rxrpc_queue_packet(struct rxrpc_sock *rx, struct rxrpc_call *call,
157 struct sk_buff *skb, bool last,
158 rxrpc_notify_end_tx_t notify_end_tx)
160 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
162 rxrpc_seq_t seq = sp->hdr.seq;
164 u8 annotation = RXRPC_TX_ANNO_UNACK;
166 _net("queue skb %p [%d]", skb, seq);
168 ASSERTCMP(seq, ==, call->tx_top + 1);
171 annotation |= RXRPC_TX_ANNO_LAST;
172 set_bit(RXRPC_CALL_TX_LASTQ, &call->flags);
175 /* We have to set the timestamp before queueing as the retransmit
176 * algorithm can see the packet as soon as we queue it.
178 skb->tstamp = ktime_get_real();
180 ix = seq & RXRPC_RXTX_BUFF_MASK;
181 rxrpc_get_skb(skb, rxrpc_skb_tx_got);
182 call->rxtx_annotations[ix] = annotation;
184 call->rxtx_buffer[ix] = skb;
187 trace_rxrpc_transmit(call, rxrpc_transmit_queue_last);
189 trace_rxrpc_transmit(call, rxrpc_transmit_queue);
191 if (last || call->state == RXRPC_CALL_SERVER_ACK_REQUEST) {
192 _debug("________awaiting reply/ACK__________");
193 write_lock_bh(&call->state_lock);
194 switch (call->state) {
195 case RXRPC_CALL_CLIENT_SEND_REQUEST:
196 call->state = RXRPC_CALL_CLIENT_AWAIT_REPLY;
197 rxrpc_notify_end_tx(rx, call, notify_end_tx);
199 case RXRPC_CALL_SERVER_ACK_REQUEST:
200 call->state = RXRPC_CALL_SERVER_SEND_REPLY;
202 WRITE_ONCE(call->ack_at, now + MAX_JIFFY_OFFSET);
203 if (call->ackr_reason == RXRPC_ACK_DELAY)
204 call->ackr_reason = 0;
205 trace_rxrpc_timer(call, rxrpc_timer_init_for_send_reply, now);
209 case RXRPC_CALL_SERVER_SEND_REPLY:
210 call->state = RXRPC_CALL_SERVER_AWAIT_ACK;
211 rxrpc_notify_end_tx(rx, call, notify_end_tx);
216 write_unlock_bh(&call->state_lock);
219 if (seq == 1 && rxrpc_is_client_call(call))
220 rxrpc_expose_client_call(call);
222 ret = rxrpc_send_data_packet(call, skb, false);
224 _debug("need instant resend %d", ret);
225 rxrpc_instant_resend(call, ix);
227 unsigned long now = jiffies, resend_at;
229 if (call->peer->rtt_usage > 1)
230 resend_at = nsecs_to_jiffies(call->peer->rtt * 3 / 2);
232 resend_at = rxrpc_resend_timeout;
237 WRITE_ONCE(call->resend_at, resend_at);
238 rxrpc_reduce_call_timer(call, resend_at, now,
239 rxrpc_timer_set_for_send);
242 rxrpc_free_skb(skb, rxrpc_skb_tx_freed);
247 * send data through a socket
248 * - must be called in process context
249 * - The caller holds the call user access mutex, but not the socket lock.
251 static int rxrpc_send_data(struct rxrpc_sock *rx,
252 struct rxrpc_call *call,
253 struct msghdr *msg, size_t len,
254 rxrpc_notify_end_tx_t notify_end_tx)
256 struct rxrpc_skb_priv *sp;
258 struct sock *sk = &rx->sk;
263 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
265 /* this should be in poll */
266 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
268 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
271 more = msg->msg_flags & MSG_MORE;
273 if (call->tx_total_len != -1) {
274 if (len > call->tx_total_len)
276 if (!more && len != call->tx_total_len)
280 skb = call->tx_pending;
281 call->tx_pending = NULL;
282 rxrpc_see_skb(skb, rxrpc_skb_tx_seen);
286 /* Check to see if there's a ping ACK to reply to. */
287 if (call->ackr_reason == RXRPC_ACK_PING_RESPONSE)
288 rxrpc_send_ack_packet(call, false, NULL);
291 size_t size, chunk, max, space;
295 if (call->tx_top - call->tx_hard_ack >=
296 min_t(unsigned int, call->tx_winsize,
297 call->cong_cwnd + call->cong_extra)) {
299 if (msg->msg_flags & MSG_DONTWAIT)
301 ret = rxrpc_wait_for_tx_window(rx, call,
303 msg->msg_flags & MSG_WAITALL);
308 max = RXRPC_JUMBO_DATALEN;
309 max -= call->conn->security_size;
310 max &= ~(call->conn->size_align - 1UL);
313 if (chunk > msg_data_left(msg) && !more)
314 chunk = msg_data_left(msg);
316 space = chunk + call->conn->size_align;
317 space &= ~(call->conn->size_align - 1UL);
319 size = space + call->conn->security_size;
321 _debug("SIZE: %zu/%zu/%zu", chunk, space, size);
323 /* create a buffer that we can retain until it's ACK'd */
324 skb = sock_alloc_send_skb(
325 sk, size, msg->msg_flags & MSG_DONTWAIT, &ret);
329 rxrpc_new_skb(skb, rxrpc_skb_tx_new);
331 _debug("ALLOC SEND %p", skb);
333 ASSERTCMP(skb->mark, ==, 0);
335 _debug("HS: %u", call->conn->security_size);
336 skb_reserve(skb, call->conn->security_size);
337 skb->len += call->conn->security_size;
341 if (sp->remain > skb_tailroom(skb))
342 sp->remain = skb_tailroom(skb);
344 _net("skb: hr %d, tr %d, hl %d, rm %d",
350 skb->ip_summed = CHECKSUM_UNNECESSARY;
356 /* append next segment of data to the current buffer */
357 if (msg_data_left(msg) > 0) {
358 int copy = skb_tailroom(skb);
359 ASSERTCMP(copy, >, 0);
360 if (copy > msg_data_left(msg))
361 copy = msg_data_left(msg);
362 if (copy > sp->remain)
366 ret = skb_add_data(skb, &msg->msg_iter, copy);
373 if (call->tx_total_len != -1)
374 call->tx_total_len -= copy;
377 /* add the packet to the send queue if it's now full */
378 if (sp->remain <= 0 ||
379 (msg_data_left(msg) == 0 && !more)) {
380 struct rxrpc_connection *conn = call->conn;
384 /* pad out if we're using security */
385 if (conn->security_ix) {
386 pad = conn->security_size + skb->mark;
387 pad = conn->size_align - pad;
388 pad &= conn->size_align - 1;
389 _debug("pad %zu", pad);
391 skb_put_zero(skb, pad);
394 seq = call->tx_top + 1;
398 sp->hdr.flags = conn->out_clientflag;
400 if (msg_data_left(msg) == 0 && !more)
401 sp->hdr.flags |= RXRPC_LAST_PACKET;
402 else if (call->tx_top - call->tx_hard_ack <
404 sp->hdr.flags |= RXRPC_MORE_PACKETS;
406 ret = conn->security->secure_packet(
407 call, skb, skb->mark, skb->head);
411 rxrpc_queue_packet(rx, call, skb,
412 !msg_data_left(msg) && !more,
417 /* Check for the far side aborting the call or a network error
418 * occurring. If this happens, save any packet that was under
419 * construction so that in the case of a network error, the
420 * call can be retried or redirected.
422 if (call->state == RXRPC_CALL_COMPLETE) {
426 } while (msg_data_left(msg) > 0);
431 call->tx_pending = skb;
432 _leave(" = %d", ret);
446 * extract control messages from the sendmsg() control buffer
448 static int rxrpc_sendmsg_cmsg(struct msghdr *msg, struct rxrpc_send_params *p)
450 struct cmsghdr *cmsg;
451 bool got_user_ID = false;
454 if (msg->msg_controllen == 0)
457 for_each_cmsghdr(cmsg, msg) {
458 if (!CMSG_OK(msg, cmsg))
461 len = cmsg->cmsg_len - sizeof(struct cmsghdr);
462 _debug("CMSG %d, %d, %d",
463 cmsg->cmsg_level, cmsg->cmsg_type, len);
465 if (cmsg->cmsg_level != SOL_RXRPC)
468 switch (cmsg->cmsg_type) {
469 case RXRPC_USER_CALL_ID:
470 if (msg->msg_flags & MSG_CMSG_COMPAT) {
471 if (len != sizeof(u32))
473 p->call.user_call_ID = *(u32 *)CMSG_DATA(cmsg);
475 if (len != sizeof(unsigned long))
477 p->call.user_call_ID = *(unsigned long *)
484 if (p->command != RXRPC_CMD_SEND_DATA)
486 p->command = RXRPC_CMD_SEND_ABORT;
487 if (len != sizeof(p->abort_code))
489 p->abort_code = *(unsigned int *)CMSG_DATA(cmsg);
490 if (p->abort_code == 0)
495 if (p->command != RXRPC_CMD_SEND_DATA)
497 p->command = RXRPC_CMD_ACCEPT;
502 case RXRPC_EXCLUSIVE_CALL:
508 case RXRPC_UPGRADE_SERVICE:
514 case RXRPC_TX_LENGTH:
515 if (p->call.tx_total_len != -1 || len != sizeof(__s64))
517 p->call.tx_total_len = *(__s64 *)CMSG_DATA(cmsg);
518 if (p->call.tx_total_len < 0)
522 case RXRPC_SET_CALL_TIMEOUT:
523 if (len & 3 || len < 4 || len > 12)
525 memcpy(&p->call.timeouts, CMSG_DATA(cmsg), len);
526 p->call.nr_timeouts = len / 4;
527 if (p->call.timeouts.hard > INT_MAX / HZ)
529 if (p->call.nr_timeouts >= 2 && p->call.timeouts.idle > 60 * 60 * 1000)
531 if (p->call.nr_timeouts >= 3 && p->call.timeouts.normal > 60 * 60 * 1000)
542 if (p->call.tx_total_len != -1 && p->command != RXRPC_CMD_SEND_DATA)
549 * Create a new client call for sendmsg().
550 * - Called with the socket lock held, which it must release.
551 * - If it returns a call, the call's lock will need releasing by the caller.
553 static struct rxrpc_call *
554 rxrpc_new_client_call_for_sendmsg(struct rxrpc_sock *rx, struct msghdr *msg,
555 struct rxrpc_send_params *p)
556 __releases(&rx->sk.sk_lock.slock)
558 struct rxrpc_conn_parameters cp;
559 struct rxrpc_call *call;
562 DECLARE_SOCKADDR(struct sockaddr_rxrpc *, srx, msg->msg_name);
566 if (!msg->msg_name) {
567 release_sock(&rx->sk);
568 return ERR_PTR(-EDESTADDRREQ);
572 if (key && !rx->key->payload.data[0])
575 memset(&cp, 0, sizeof(cp));
576 cp.local = rx->local;
578 cp.security_level = rx->min_sec_level;
579 cp.exclusive = rx->exclusive | p->exclusive;
580 cp.upgrade = p->upgrade;
581 cp.service_id = srx->srx_service;
582 call = rxrpc_new_client_call(rx, &cp, srx, &p->call, GFP_KERNEL);
583 /* The socket is now unlocked */
585 _leave(" = %p\n", call);
590 * send a message forming part of a client call through an RxRPC socket
591 * - caller holds the socket locked
592 * - the socket may be either a client socket or a server socket
594 int rxrpc_do_sendmsg(struct rxrpc_sock *rx, struct msghdr *msg, size_t len)
595 __releases(&rx->sk.sk_lock.slock)
597 enum rxrpc_call_state state;
598 struct rxrpc_call *call;
599 unsigned long now, j;
602 struct rxrpc_send_params p = {
603 .call.tx_total_len = -1,
604 .call.user_call_ID = 0,
605 .call.nr_timeouts = 0,
607 .command = RXRPC_CMD_SEND_DATA,
614 ret = rxrpc_sendmsg_cmsg(msg, &p);
616 goto error_release_sock;
618 if (p.command == RXRPC_CMD_ACCEPT) {
620 if (rx->sk.sk_state != RXRPC_SERVER_LISTENING)
621 goto error_release_sock;
622 call = rxrpc_accept_call(rx, p.call.user_call_ID, NULL);
623 /* The socket is now unlocked. */
625 return PTR_ERR(call);
630 call = rxrpc_find_call_by_user_ID(rx, p.call.user_call_ID);
633 if (p.command != RXRPC_CMD_SEND_DATA)
634 goto error_release_sock;
635 call = rxrpc_new_client_call_for_sendmsg(rx, msg, &p);
636 /* The socket is now unlocked... */
638 return PTR_ERR(call);
639 /* ... and we have the call lock. */
641 switch (READ_ONCE(call->state)) {
642 case RXRPC_CALL_UNINITIALISED:
643 case RXRPC_CALL_CLIENT_AWAIT_CONN:
644 case RXRPC_CALL_SERVER_PREALLOC:
645 case RXRPC_CALL_SERVER_SECURING:
646 case RXRPC_CALL_SERVER_ACCEPTING:
648 goto error_release_sock;
653 ret = mutex_lock_interruptible(&call->user_mutex);
654 release_sock(&rx->sk);
660 if (p.call.tx_total_len != -1) {
662 if (call->tx_total_len != -1 ||
666 call->tx_total_len = p.call.tx_total_len;
670 switch (p.call.nr_timeouts) {
672 j = msecs_to_jiffies(p.call.timeouts.normal);
673 if (p.call.timeouts.normal > 0 && j == 0)
675 WRITE_ONCE(call->next_rx_timo, j);
678 j = msecs_to_jiffies(p.call.timeouts.idle);
679 if (p.call.timeouts.idle > 0 && j == 0)
681 WRITE_ONCE(call->next_req_timo, j);
684 if (p.call.timeouts.hard > 0) {
685 j = msecs_to_jiffies(p.call.timeouts.hard);
688 WRITE_ONCE(call->expect_term_by, j);
689 rxrpc_reduce_call_timer(call, j, now,
690 rxrpc_timer_set_for_hard);
695 state = READ_ONCE(call->state);
696 _debug("CALL %d USR %lx ST %d on CONN %p",
697 call->debug_id, call->user_call_ID, state, call->conn);
699 if (state >= RXRPC_CALL_COMPLETE) {
700 /* it's too late for this call */
702 } else if (p.command == RXRPC_CMD_SEND_ABORT) {
704 if (rxrpc_abort_call("CMD", call, 0, p.abort_code, -ECONNABORTED))
705 ret = rxrpc_send_abort_packet(call);
706 } else if (p.command != RXRPC_CMD_SEND_DATA) {
708 } else if (rxrpc_is_client_call(call) &&
709 state != RXRPC_CALL_CLIENT_SEND_REQUEST) {
710 /* request phase complete for this client call */
712 } else if (rxrpc_is_service_call(call) &&
713 state != RXRPC_CALL_SERVER_ACK_REQUEST &&
714 state != RXRPC_CALL_SERVER_SEND_REPLY) {
715 /* Reply phase not begun or not complete for service call. */
718 ret = rxrpc_send_data(rx, call, msg, len, NULL);
722 mutex_unlock(&call->user_mutex);
724 rxrpc_put_call(call, rxrpc_call_put);
725 _leave(" = %d", ret);
729 release_sock(&rx->sk);
734 * rxrpc_kernel_send_data - Allow a kernel service to send data on a call
735 * @sock: The socket the call is on
736 * @call: The call to send data through
737 * @msg: The data to send
738 * @len: The amount of data to send
739 * @notify_end_tx: Notification that the last packet is queued.
741 * Allow a kernel service to send data on a call. The call must be in an state
742 * appropriate to sending data. No control data should be supplied in @msg,
743 * nor should an address be supplied. MSG_MORE should be flagged if there's
744 * more data to come, otherwise this data will end the transmission phase.
746 int rxrpc_kernel_send_data(struct socket *sock, struct rxrpc_call *call,
747 struct msghdr *msg, size_t len,
748 rxrpc_notify_end_tx_t notify_end_tx)
752 _enter("{%d,%s},", call->debug_id, rxrpc_call_states[call->state]);
754 ASSERTCMP(msg->msg_name, ==, NULL);
755 ASSERTCMP(msg->msg_control, ==, NULL);
757 mutex_lock(&call->user_mutex);
759 _debug("CALL %d USR %lx ST %d on CONN %p",
760 call->debug_id, call->user_call_ID, call->state, call->conn);
762 switch (READ_ONCE(call->state)) {
763 case RXRPC_CALL_CLIENT_SEND_REQUEST:
764 case RXRPC_CALL_SERVER_ACK_REQUEST:
765 case RXRPC_CALL_SERVER_SEND_REPLY:
766 ret = rxrpc_send_data(rxrpc_sk(sock->sk), call, msg, len,
769 case RXRPC_CALL_COMPLETE:
770 read_lock_bh(&call->state_lock);
772 read_unlock_bh(&call->state_lock);
775 /* Request phase complete for this client call */
776 trace_rxrpc_rx_eproto(call, 0, tracepoint_string("late_send"));
781 mutex_unlock(&call->user_mutex);
782 _leave(" = %d", ret);
785 EXPORT_SYMBOL(rxrpc_kernel_send_data);
788 * rxrpc_kernel_abort_call - Allow a kernel service to abort a call
789 * @sock: The socket the call is on
790 * @call: The call to be aborted
791 * @abort_code: The abort code to stick into the ABORT packet
792 * @error: Local error value
793 * @why: 3-char string indicating why.
795 * Allow a kernel service to abort a call, if it's still in an abortable state
796 * and return true if the call was aborted, false if it was already complete.
798 bool rxrpc_kernel_abort_call(struct socket *sock, struct rxrpc_call *call,
799 u32 abort_code, int error, const char *why)
803 _enter("{%d},%d,%d,%s", call->debug_id, abort_code, error, why);
805 mutex_lock(&call->user_mutex);
807 aborted = rxrpc_abort_call(why, call, 0, abort_code, error);
809 rxrpc_send_abort_packet(call);
811 mutex_unlock(&call->user_mutex);
814 EXPORT_SYMBOL(rxrpc_kernel_abort_call);
817 * rxrpc_kernel_set_tx_length - Set the total Tx length on a call
818 * @sock: The socket the call is on
819 * @call: The call to be informed
820 * @tx_total_len: The amount of data to be transmitted for this call
822 * Allow a kernel service to set the total transmit length on a call. This
823 * allows buffer-to-packet encrypt-and-copy to be performed.
825 * This function is primarily for use for setting the reply length since the
826 * request length can be set when beginning the call.
828 void rxrpc_kernel_set_tx_length(struct socket *sock, struct rxrpc_call *call,
831 WARN_ON(call->tx_total_len != -1);
832 call->tx_total_len = tx_total_len;
834 EXPORT_SYMBOL(rxrpc_kernel_set_tx_length);