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 RXRPC_CMD_SEND_DATA, /* send data message */
26 RXRPC_CMD_SEND_ABORT, /* request abort generation */
27 RXRPC_CMD_ACCEPT, /* [server] accept incoming call */
28 RXRPC_CMD_REJECT_BUSY, /* [server] reject a call as busy */
31 struct rxrpc_send_params {
32 s64 tx_total_len; /* Total Tx data length (if send data) */
33 unsigned long user_call_ID; /* User's call ID */
34 u32 abort_code; /* Abort code to Tx (if abort) */
35 enum rxrpc_command command : 8; /* The command to implement */
36 bool exclusive; /* Shared or exclusive call */
37 bool upgrade; /* If the connection is upgradeable */
41 * wait for space to appear in the transmit/ACK window
42 * - caller holds the socket locked
44 static int rxrpc_wait_for_tx_window(struct rxrpc_sock *rx,
45 struct rxrpc_call *call,
48 DECLARE_WAITQUEUE(myself, current);
52 call->tx_hard_ack, call->tx_top, call->tx_winsize);
54 add_wait_queue(&call->waitq, &myself);
57 set_current_state(TASK_INTERRUPTIBLE);
59 if (call->tx_top - call->tx_hard_ack <
60 min_t(unsigned int, call->tx_winsize,
61 call->cong_cwnd + call->cong_extra))
63 if (call->state >= RXRPC_CALL_COMPLETE) {
67 if (signal_pending(current)) {
68 ret = sock_intr_errno(*timeo);
72 trace_rxrpc_transmit(call, rxrpc_transmit_wait);
73 mutex_unlock(&call->user_mutex);
74 *timeo = schedule_timeout(*timeo);
75 if (mutex_lock_interruptible(&call->user_mutex) < 0) {
76 ret = sock_intr_errno(*timeo);
81 remove_wait_queue(&call->waitq, &myself);
82 set_current_state(TASK_RUNNING);
88 * Schedule an instant Tx resend.
90 static inline void rxrpc_instant_resend(struct rxrpc_call *call, int ix)
92 spin_lock_bh(&call->lock);
94 if (call->state < RXRPC_CALL_COMPLETE) {
95 call->rxtx_annotations[ix] =
96 (call->rxtx_annotations[ix] & RXRPC_TX_ANNO_LAST) |
97 RXRPC_TX_ANNO_RETRANS;
98 if (!test_and_set_bit(RXRPC_CALL_EV_RESEND, &call->events))
99 rxrpc_queue_call(call);
102 spin_unlock_bh(&call->lock);
106 * Notify the owner of the call that the transmit phase is ended and the last
107 * packet has been queued.
109 static void rxrpc_notify_end_tx(struct rxrpc_sock *rx, struct rxrpc_call *call,
110 rxrpc_notify_end_tx_t notify_end_tx)
113 notify_end_tx(&rx->sk, call, call->user_call_ID);
117 * Queue a DATA packet for transmission, set the resend timeout and send the
120 static void rxrpc_queue_packet(struct rxrpc_sock *rx, struct rxrpc_call *call,
121 struct sk_buff *skb, bool last,
122 rxrpc_notify_end_tx_t notify_end_tx)
124 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
125 rxrpc_seq_t seq = sp->hdr.seq;
127 u8 annotation = RXRPC_TX_ANNO_UNACK;
129 _net("queue skb %p [%d]", skb, seq);
131 ASSERTCMP(seq, ==, call->tx_top + 1);
134 annotation |= RXRPC_TX_ANNO_LAST;
135 set_bit(RXRPC_CALL_TX_LASTQ, &call->flags);
138 /* We have to set the timestamp before queueing as the retransmit
139 * algorithm can see the packet as soon as we queue it.
141 skb->tstamp = ktime_get_real();
143 ix = seq & RXRPC_RXTX_BUFF_MASK;
144 rxrpc_get_skb(skb, rxrpc_skb_tx_got);
145 call->rxtx_annotations[ix] = annotation;
147 call->rxtx_buffer[ix] = skb;
150 trace_rxrpc_transmit(call, rxrpc_transmit_queue_last);
152 trace_rxrpc_transmit(call, rxrpc_transmit_queue);
154 if (last || call->state == RXRPC_CALL_SERVER_ACK_REQUEST) {
155 _debug("________awaiting reply/ACK__________");
156 write_lock_bh(&call->state_lock);
157 switch (call->state) {
158 case RXRPC_CALL_CLIENT_SEND_REQUEST:
159 call->state = RXRPC_CALL_CLIENT_AWAIT_REPLY;
160 rxrpc_notify_end_tx(rx, call, notify_end_tx);
162 case RXRPC_CALL_SERVER_ACK_REQUEST:
163 call->state = RXRPC_CALL_SERVER_SEND_REPLY;
164 call->ack_at = call->expire_at;
165 if (call->ackr_reason == RXRPC_ACK_DELAY)
166 call->ackr_reason = 0;
167 __rxrpc_set_timer(call, rxrpc_timer_init_for_send_reply,
171 case RXRPC_CALL_SERVER_SEND_REPLY:
172 call->state = RXRPC_CALL_SERVER_AWAIT_ACK;
173 rxrpc_notify_end_tx(rx, call, notify_end_tx);
178 write_unlock_bh(&call->state_lock);
181 if (seq == 1 && rxrpc_is_client_call(call))
182 rxrpc_expose_client_call(call);
184 ret = rxrpc_send_data_packet(call, skb, false);
186 _debug("need instant resend %d", ret);
187 rxrpc_instant_resend(call, ix);
189 ktime_t now = ktime_get_real(), resend_at;
191 resend_at = ktime_add_ms(now, rxrpc_resend_timeout);
193 if (ktime_before(resend_at, call->resend_at)) {
194 call->resend_at = resend_at;
195 rxrpc_set_timer(call, rxrpc_timer_set_for_send, now);
199 rxrpc_free_skb(skb, rxrpc_skb_tx_freed);
204 * send data through a socket
205 * - must be called in process context
206 * - The caller holds the call user access mutex, but not the socket lock.
208 static int rxrpc_send_data(struct rxrpc_sock *rx,
209 struct rxrpc_call *call,
210 struct msghdr *msg, size_t len,
211 rxrpc_notify_end_tx_t notify_end_tx)
213 struct rxrpc_skb_priv *sp;
215 struct sock *sk = &rx->sk;
220 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
222 /* this should be in poll */
223 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
225 if (sk->sk_shutdown & SEND_SHUTDOWN)
228 more = msg->msg_flags & MSG_MORE;
230 if (call->tx_total_len != -1) {
231 if (len > call->tx_total_len)
233 if (!more && len != call->tx_total_len)
237 skb = call->tx_pending;
238 call->tx_pending = NULL;
239 rxrpc_see_skb(skb, rxrpc_skb_tx_seen);
243 /* Check to see if there's a ping ACK to reply to. */
244 if (call->ackr_reason == RXRPC_ACK_PING_RESPONSE)
245 rxrpc_send_ack_packet(call, false);
248 size_t size, chunk, max, space;
252 if (call->tx_top - call->tx_hard_ack >=
253 min_t(unsigned int, call->tx_winsize,
254 call->cong_cwnd + call->cong_extra)) {
256 if (msg->msg_flags & MSG_DONTWAIT)
258 ret = rxrpc_wait_for_tx_window(rx, call,
264 max = RXRPC_JUMBO_DATALEN;
265 max -= call->conn->security_size;
266 max &= ~(call->conn->size_align - 1UL);
269 if (chunk > msg_data_left(msg) && !more)
270 chunk = msg_data_left(msg);
272 space = chunk + call->conn->size_align;
273 space &= ~(call->conn->size_align - 1UL);
275 size = space + call->conn->security_size;
277 _debug("SIZE: %zu/%zu/%zu", chunk, space, size);
279 /* create a buffer that we can retain until it's ACK'd */
280 skb = sock_alloc_send_skb(
281 sk, size, msg->msg_flags & MSG_DONTWAIT, &ret);
285 rxrpc_new_skb(skb, rxrpc_skb_tx_new);
287 _debug("ALLOC SEND %p", skb);
289 ASSERTCMP(skb->mark, ==, 0);
291 _debug("HS: %u", call->conn->security_size);
292 skb_reserve(skb, call->conn->security_size);
293 skb->len += call->conn->security_size;
297 if (sp->remain > skb_tailroom(skb))
298 sp->remain = skb_tailroom(skb);
300 _net("skb: hr %d, tr %d, hl %d, rm %d",
306 skb->ip_summed = CHECKSUM_UNNECESSARY;
312 /* append next segment of data to the current buffer */
313 if (msg_data_left(msg) > 0) {
314 int copy = skb_tailroom(skb);
315 ASSERTCMP(copy, >, 0);
316 if (copy > msg_data_left(msg))
317 copy = msg_data_left(msg);
318 if (copy > sp->remain)
322 ret = skb_add_data(skb, &msg->msg_iter, copy);
329 if (call->tx_total_len != -1)
330 call->tx_total_len -= copy;
333 /* add the packet to the send queue if it's now full */
334 if (sp->remain <= 0 ||
335 (msg_data_left(msg) == 0 && !more)) {
336 struct rxrpc_connection *conn = call->conn;
340 /* pad out if we're using security */
341 if (conn->security_ix) {
342 pad = conn->security_size + skb->mark;
343 pad = conn->size_align - pad;
344 pad &= conn->size_align - 1;
345 _debug("pad %zu", pad);
347 skb_put_zero(skb, pad);
350 seq = call->tx_top + 1;
354 sp->hdr.flags = conn->out_clientflag;
356 if (msg_data_left(msg) == 0 && !more)
357 sp->hdr.flags |= RXRPC_LAST_PACKET;
358 else if (call->tx_top - call->tx_hard_ack <
360 sp->hdr.flags |= RXRPC_MORE_PACKETS;
362 ret = conn->security->secure_packet(
363 call, skb, skb->mark, skb->head);
367 rxrpc_queue_packet(rx, call, skb,
368 !msg_data_left(msg) && !more,
373 /* Check for the far side aborting the call or a network error
374 * occurring. If this happens, save any packet that was under
375 * construction so that in the case of a network error, the
376 * call can be retried or redirected.
378 if (call->state == RXRPC_CALL_COMPLETE) {
382 } while (msg_data_left(msg) > 0);
387 call->tx_pending = skb;
388 _leave(" = %d", ret);
402 * extract control messages from the sendmsg() control buffer
404 static int rxrpc_sendmsg_cmsg(struct msghdr *msg, struct rxrpc_send_params *p)
406 struct cmsghdr *cmsg;
407 bool got_user_ID = false;
410 if (msg->msg_controllen == 0)
413 for_each_cmsghdr(cmsg, msg) {
414 if (!CMSG_OK(msg, cmsg))
417 len = cmsg->cmsg_len - sizeof(struct cmsghdr);
418 _debug("CMSG %d, %d, %d",
419 cmsg->cmsg_level, cmsg->cmsg_type, len);
421 if (cmsg->cmsg_level != SOL_RXRPC)
424 switch (cmsg->cmsg_type) {
425 case RXRPC_USER_CALL_ID:
426 if (msg->msg_flags & MSG_CMSG_COMPAT) {
427 if (len != sizeof(u32))
429 p->user_call_ID = *(u32 *)CMSG_DATA(cmsg);
431 if (len != sizeof(unsigned long))
433 p->user_call_ID = *(unsigned long *)
440 if (p->command != RXRPC_CMD_SEND_DATA)
442 p->command = RXRPC_CMD_SEND_ABORT;
443 if (len != sizeof(p->abort_code))
445 p->abort_code = *(unsigned int *)CMSG_DATA(cmsg);
446 if (p->abort_code == 0)
451 if (p->command != RXRPC_CMD_SEND_DATA)
453 p->command = RXRPC_CMD_ACCEPT;
458 case RXRPC_EXCLUSIVE_CALL:
464 case RXRPC_UPGRADE_SERVICE:
470 case RXRPC_TX_LENGTH:
471 if (p->tx_total_len != -1 || len != sizeof(__s64))
473 p->tx_total_len = *(__s64 *)CMSG_DATA(cmsg);
474 if (p->tx_total_len < 0)
485 if (p->tx_total_len != -1 && p->command != RXRPC_CMD_SEND_DATA)
492 * Create a new client call for sendmsg().
493 * - Called with the socket lock held, which it must release.
494 * - If it returns a call, the call's lock will need releasing by the caller.
496 static struct rxrpc_call *
497 rxrpc_new_client_call_for_sendmsg(struct rxrpc_sock *rx, struct msghdr *msg,
498 struct rxrpc_send_params *p)
499 __releases(&rx->sk.sk_lock.slock)
501 struct rxrpc_conn_parameters cp;
502 struct rxrpc_call *call;
505 DECLARE_SOCKADDR(struct sockaddr_rxrpc *, srx, msg->msg_name);
509 if (!msg->msg_name) {
510 release_sock(&rx->sk);
511 return ERR_PTR(-EDESTADDRREQ);
515 if (key && !rx->key->payload.data[0])
518 memset(&cp, 0, sizeof(cp));
519 cp.local = rx->local;
521 cp.security_level = rx->min_sec_level;
522 cp.exclusive = rx->exclusive | p->exclusive;
523 cp.upgrade = p->upgrade;
524 cp.service_id = srx->srx_service;
525 call = rxrpc_new_client_call(rx, &cp, srx, p->user_call_ID,
526 p->tx_total_len, GFP_KERNEL);
527 /* The socket is now unlocked */
529 _leave(" = %p\n", call);
534 * send a message forming part of a client call through an RxRPC socket
535 * - caller holds the socket locked
536 * - the socket may be either a client socket or a server socket
538 int rxrpc_do_sendmsg(struct rxrpc_sock *rx, struct msghdr *msg, size_t len)
539 __releases(&rx->sk.sk_lock.slock)
541 enum rxrpc_call_state state;
542 struct rxrpc_call *call;
545 struct rxrpc_send_params p = {
549 .command = RXRPC_CMD_SEND_DATA,
556 ret = rxrpc_sendmsg_cmsg(msg, &p);
558 goto error_release_sock;
560 if (p.command == RXRPC_CMD_ACCEPT) {
562 if (rx->sk.sk_state != RXRPC_SERVER_LISTENING)
563 goto error_release_sock;
564 call = rxrpc_accept_call(rx, p.user_call_ID, NULL);
565 /* The socket is now unlocked. */
567 return PTR_ERR(call);
572 call = rxrpc_find_call_by_user_ID(rx, p.user_call_ID);
575 if (p.command != RXRPC_CMD_SEND_DATA)
576 goto error_release_sock;
577 call = rxrpc_new_client_call_for_sendmsg(rx, msg, &p);
578 /* The socket is now unlocked... */
580 return PTR_ERR(call);
581 /* ... and we have the call lock. */
583 if (READ_ONCE(call->state) == RXRPC_CALL_COMPLETE)
586 switch (READ_ONCE(call->state)) {
587 case RXRPC_CALL_UNINITIALISED:
588 case RXRPC_CALL_CLIENT_AWAIT_CONN:
589 case RXRPC_CALL_SERVER_PREALLOC:
590 case RXRPC_CALL_SERVER_SECURING:
591 case RXRPC_CALL_SERVER_ACCEPTING:
592 rxrpc_put_call(call, rxrpc_call_put);
594 goto error_release_sock;
599 ret = mutex_lock_interruptible(&call->user_mutex);
600 release_sock(&rx->sk);
606 if (p.tx_total_len != -1) {
608 if (call->tx_total_len != -1 ||
612 call->tx_total_len = p.tx_total_len;
616 state = READ_ONCE(call->state);
617 _debug("CALL %d USR %lx ST %d on CONN %p",
618 call->debug_id, call->user_call_ID, state, call->conn);
620 if (state >= RXRPC_CALL_COMPLETE) {
621 /* it's too late for this call */
623 } else if (p.command == RXRPC_CMD_SEND_ABORT) {
625 if (rxrpc_abort_call("CMD", call, 0, p.abort_code, -ECONNABORTED))
626 ret = rxrpc_send_abort_packet(call);
627 } else if (p.command != RXRPC_CMD_SEND_DATA) {
629 } else if (rxrpc_is_client_call(call) &&
630 state != RXRPC_CALL_CLIENT_SEND_REQUEST) {
631 /* request phase complete for this client call */
633 } else if (rxrpc_is_service_call(call) &&
634 state != RXRPC_CALL_SERVER_ACK_REQUEST &&
635 state != RXRPC_CALL_SERVER_SEND_REPLY) {
636 /* Reply phase not begun or not complete for service call. */
639 ret = rxrpc_send_data(rx, call, msg, len, NULL);
643 mutex_unlock(&call->user_mutex);
645 rxrpc_put_call(call, rxrpc_call_put);
646 _leave(" = %d", ret);
650 release_sock(&rx->sk);
655 * rxrpc_kernel_send_data - Allow a kernel service to send data on a call
656 * @sock: The socket the call is on
657 * @call: The call to send data through
658 * @msg: The data to send
659 * @len: The amount of data to send
660 * @notify_end_tx: Notification that the last packet is queued.
662 * Allow a kernel service to send data on a call. The call must be in an state
663 * appropriate to sending data. No control data should be supplied in @msg,
664 * nor should an address be supplied. MSG_MORE should be flagged if there's
665 * more data to come, otherwise this data will end the transmission phase.
667 int rxrpc_kernel_send_data(struct socket *sock, struct rxrpc_call *call,
668 struct msghdr *msg, size_t len,
669 rxrpc_notify_end_tx_t notify_end_tx)
673 _enter("{%d,%s},", call->debug_id, rxrpc_call_states[call->state]);
675 ASSERTCMP(msg->msg_name, ==, NULL);
676 ASSERTCMP(msg->msg_control, ==, NULL);
678 mutex_lock(&call->user_mutex);
680 _debug("CALL %d USR %lx ST %d on CONN %p",
681 call->debug_id, call->user_call_ID, call->state, call->conn);
683 switch (READ_ONCE(call->state)) {
684 case RXRPC_CALL_CLIENT_SEND_REQUEST:
685 case RXRPC_CALL_SERVER_ACK_REQUEST:
686 case RXRPC_CALL_SERVER_SEND_REPLY:
687 ret = rxrpc_send_data(rxrpc_sk(sock->sk), call, msg, len,
690 case RXRPC_CALL_COMPLETE:
691 read_lock_bh(&call->state_lock);
693 read_unlock_bh(&call->state_lock);
696 /* Request phase complete for this client call */
697 trace_rxrpc_rx_eproto(call, 0, tracepoint_string("late_send"));
702 mutex_unlock(&call->user_mutex);
703 _leave(" = %d", ret);
706 EXPORT_SYMBOL(rxrpc_kernel_send_data);
709 * rxrpc_kernel_abort_call - Allow a kernel service to abort a call
710 * @sock: The socket the call is on
711 * @call: The call to be aborted
712 * @abort_code: The abort code to stick into the ABORT packet
713 * @error: Local error value
714 * @why: 3-char string indicating why.
716 * Allow a kernel service to abort a call, if it's still in an abortable state
717 * and return true if the call was aborted, false if it was already complete.
719 bool rxrpc_kernel_abort_call(struct socket *sock, struct rxrpc_call *call,
720 u32 abort_code, int error, const char *why)
724 _enter("{%d},%d,%d,%s", call->debug_id, abort_code, error, why);
726 mutex_lock(&call->user_mutex);
728 aborted = rxrpc_abort_call(why, call, 0, abort_code, error);
730 rxrpc_send_abort_packet(call);
732 mutex_unlock(&call->user_mutex);
735 EXPORT_SYMBOL(rxrpc_kernel_abort_call);
738 * rxrpc_kernel_set_tx_length - Set the total Tx length on a call
739 * @sock: The socket the call is on
740 * @call: The call to be informed
741 * @tx_total_len: The amount of data to be transmitted for this call
743 * Allow a kernel service to set the total transmit length on a call. This
744 * allows buffer-to-packet encrypt-and-copy to be performed.
746 * This function is primarily for use for setting the reply length since the
747 * request length can be set when beginning the call.
749 void rxrpc_kernel_set_tx_length(struct socket *sock, struct rxrpc_call *call,
752 WARN_ON(call->tx_total_len != -1);
753 call->tx_total_len = tx_total_len;
755 EXPORT_SYMBOL(rxrpc_kernel_set_tx_length);