]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - net/rxrpc/sendmsg.c
rxrpc: Add service upgrade support for client connections
[mirror_ubuntu-artful-kernel.git] / net / rxrpc / sendmsg.c
1 /* AF_RXRPC sendmsg() implementation.
2 *
3 * Copyright (C) 2007, 2016 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 *
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.
10 */
11
12 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
13
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>
19
20 #include <net/sock.h>
21 #include <net/af_rxrpc.h>
22 #include "ar-internal.h"
23
24 enum rxrpc_command {
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 */
29 };
30
31 /*
32 * wait for space to appear in the transmit/ACK window
33 * - caller holds the socket locked
34 */
35 static int rxrpc_wait_for_tx_window(struct rxrpc_sock *rx,
36 struct rxrpc_call *call,
37 long *timeo)
38 {
39 DECLARE_WAITQUEUE(myself, current);
40 int ret;
41
42 _enter(",{%u,%u,%u}",
43 call->tx_hard_ack, call->tx_top, call->tx_winsize);
44
45 add_wait_queue(&call->waitq, &myself);
46
47 for (;;) {
48 set_current_state(TASK_INTERRUPTIBLE);
49 ret = 0;
50 if (call->tx_top - call->tx_hard_ack <
51 min_t(unsigned int, call->tx_winsize,
52 call->cong_cwnd + call->cong_extra))
53 break;
54 if (call->state >= RXRPC_CALL_COMPLETE) {
55 ret = -call->error;
56 break;
57 }
58 if (signal_pending(current)) {
59 ret = sock_intr_errno(*timeo);
60 break;
61 }
62
63 trace_rxrpc_transmit(call, rxrpc_transmit_wait);
64 mutex_unlock(&call->user_mutex);
65 *timeo = schedule_timeout(*timeo);
66 if (mutex_lock_interruptible(&call->user_mutex) < 0) {
67 ret = sock_intr_errno(*timeo);
68 break;
69 }
70 }
71
72 remove_wait_queue(&call->waitq, &myself);
73 set_current_state(TASK_RUNNING);
74 _leave(" = %d", ret);
75 return ret;
76 }
77
78 /*
79 * Schedule an instant Tx resend.
80 */
81 static inline void rxrpc_instant_resend(struct rxrpc_call *call, int ix)
82 {
83 spin_lock_bh(&call->lock);
84
85 if (call->state < RXRPC_CALL_COMPLETE) {
86 call->rxtx_annotations[ix] = RXRPC_TX_ANNO_RETRANS;
87 if (!test_and_set_bit(RXRPC_CALL_EV_RESEND, &call->events))
88 rxrpc_queue_call(call);
89 }
90
91 spin_unlock_bh(&call->lock);
92 }
93
94 /*
95 * Queue a DATA packet for transmission, set the resend timeout and send the
96 * packet immediately
97 */
98 static void rxrpc_queue_packet(struct rxrpc_call *call, struct sk_buff *skb,
99 bool last)
100 {
101 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
102 rxrpc_seq_t seq = sp->hdr.seq;
103 int ret, ix;
104 u8 annotation = RXRPC_TX_ANNO_UNACK;
105
106 _net("queue skb %p [%d]", skb, seq);
107
108 ASSERTCMP(seq, ==, call->tx_top + 1);
109
110 if (last)
111 annotation |= RXRPC_TX_ANNO_LAST;
112
113 /* We have to set the timestamp before queueing as the retransmit
114 * algorithm can see the packet as soon as we queue it.
115 */
116 skb->tstamp = ktime_get_real();
117
118 ix = seq & RXRPC_RXTX_BUFF_MASK;
119 rxrpc_get_skb(skb, rxrpc_skb_tx_got);
120 call->rxtx_annotations[ix] = annotation;
121 smp_wmb();
122 call->rxtx_buffer[ix] = skb;
123 call->tx_top = seq;
124 if (last)
125 trace_rxrpc_transmit(call, rxrpc_transmit_queue_last);
126 else
127 trace_rxrpc_transmit(call, rxrpc_transmit_queue);
128
129 if (last || call->state == RXRPC_CALL_SERVER_ACK_REQUEST) {
130 _debug("________awaiting reply/ACK__________");
131 write_lock_bh(&call->state_lock);
132 switch (call->state) {
133 case RXRPC_CALL_CLIENT_SEND_REQUEST:
134 call->state = RXRPC_CALL_CLIENT_AWAIT_REPLY;
135 break;
136 case RXRPC_CALL_SERVER_ACK_REQUEST:
137 call->state = RXRPC_CALL_SERVER_SEND_REPLY;
138 call->ack_at = call->expire_at;
139 if (call->ackr_reason == RXRPC_ACK_DELAY)
140 call->ackr_reason = 0;
141 __rxrpc_set_timer(call, rxrpc_timer_init_for_send_reply,
142 ktime_get_real());
143 if (!last)
144 break;
145 case RXRPC_CALL_SERVER_SEND_REPLY:
146 call->state = RXRPC_CALL_SERVER_AWAIT_ACK;
147 break;
148 default:
149 break;
150 }
151 write_unlock_bh(&call->state_lock);
152 }
153
154 if (seq == 1 && rxrpc_is_client_call(call))
155 rxrpc_expose_client_call(call);
156
157 ret = rxrpc_send_data_packet(call, skb, false);
158 if (ret < 0) {
159 _debug("need instant resend %d", ret);
160 rxrpc_instant_resend(call, ix);
161 } else {
162 ktime_t now = ktime_get_real(), resend_at;
163
164 resend_at = ktime_add_ms(now, rxrpc_resend_timeout);
165
166 if (ktime_before(resend_at, call->resend_at)) {
167 call->resend_at = resend_at;
168 rxrpc_set_timer(call, rxrpc_timer_set_for_send, now);
169 }
170 }
171
172 rxrpc_free_skb(skb, rxrpc_skb_tx_freed);
173 _leave("");
174 }
175
176 /*
177 * send data through a socket
178 * - must be called in process context
179 * - The caller holds the call user access mutex, but not the socket lock.
180 */
181 static int rxrpc_send_data(struct rxrpc_sock *rx,
182 struct rxrpc_call *call,
183 struct msghdr *msg, size_t len)
184 {
185 struct rxrpc_skb_priv *sp;
186 struct sk_buff *skb;
187 struct sock *sk = &rx->sk;
188 long timeo;
189 bool more;
190 int ret, copied;
191
192 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
193
194 /* this should be in poll */
195 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
196
197 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
198 return -EPIPE;
199
200 more = msg->msg_flags & MSG_MORE;
201
202 skb = call->tx_pending;
203 call->tx_pending = NULL;
204 rxrpc_see_skb(skb, rxrpc_skb_tx_seen);
205
206 copied = 0;
207 do {
208 /* Check to see if there's a ping ACK to reply to. */
209 if (call->ackr_reason == RXRPC_ACK_PING_RESPONSE)
210 rxrpc_send_ack_packet(call, false);
211
212 if (!skb) {
213 size_t size, chunk, max, space;
214
215 _debug("alloc");
216
217 if (call->tx_top - call->tx_hard_ack >=
218 min_t(unsigned int, call->tx_winsize,
219 call->cong_cwnd + call->cong_extra)) {
220 ret = -EAGAIN;
221 if (msg->msg_flags & MSG_DONTWAIT)
222 goto maybe_error;
223 ret = rxrpc_wait_for_tx_window(rx, call,
224 &timeo);
225 if (ret < 0)
226 goto maybe_error;
227 }
228
229 max = RXRPC_JUMBO_DATALEN;
230 max -= call->conn->security_size;
231 max &= ~(call->conn->size_align - 1UL);
232
233 chunk = max;
234 if (chunk > msg_data_left(msg) && !more)
235 chunk = msg_data_left(msg);
236
237 space = chunk + call->conn->size_align;
238 space &= ~(call->conn->size_align - 1UL);
239
240 size = space + call->conn->security_size;
241
242 _debug("SIZE: %zu/%zu/%zu", chunk, space, size);
243
244 /* create a buffer that we can retain until it's ACK'd */
245 skb = sock_alloc_send_skb(
246 sk, size, msg->msg_flags & MSG_DONTWAIT, &ret);
247 if (!skb)
248 goto maybe_error;
249
250 rxrpc_new_skb(skb, rxrpc_skb_tx_new);
251
252 _debug("ALLOC SEND %p", skb);
253
254 ASSERTCMP(skb->mark, ==, 0);
255
256 _debug("HS: %u", call->conn->security_size);
257 skb_reserve(skb, call->conn->security_size);
258 skb->len += call->conn->security_size;
259
260 sp = rxrpc_skb(skb);
261 sp->remain = chunk;
262 if (sp->remain > skb_tailroom(skb))
263 sp->remain = skb_tailroom(skb);
264
265 _net("skb: hr %d, tr %d, hl %d, rm %d",
266 skb_headroom(skb),
267 skb_tailroom(skb),
268 skb_headlen(skb),
269 sp->remain);
270
271 skb->ip_summed = CHECKSUM_UNNECESSARY;
272 }
273
274 _debug("append");
275 sp = rxrpc_skb(skb);
276
277 /* append next segment of data to the current buffer */
278 if (msg_data_left(msg) > 0) {
279 int copy = skb_tailroom(skb);
280 ASSERTCMP(copy, >, 0);
281 if (copy > msg_data_left(msg))
282 copy = msg_data_left(msg);
283 if (copy > sp->remain)
284 copy = sp->remain;
285
286 _debug("add");
287 ret = skb_add_data(skb, &msg->msg_iter, copy);
288 _debug("added");
289 if (ret < 0)
290 goto efault;
291 sp->remain -= copy;
292 skb->mark += copy;
293 copied += copy;
294 }
295
296 /* check for the far side aborting the call or a network error
297 * occurring */
298 if (call->state == RXRPC_CALL_COMPLETE)
299 goto call_terminated;
300
301 /* add the packet to the send queue if it's now full */
302 if (sp->remain <= 0 ||
303 (msg_data_left(msg) == 0 && !more)) {
304 struct rxrpc_connection *conn = call->conn;
305 uint32_t seq;
306 size_t pad;
307
308 /* pad out if we're using security */
309 if (conn->security_ix) {
310 pad = conn->security_size + skb->mark;
311 pad = conn->size_align - pad;
312 pad &= conn->size_align - 1;
313 _debug("pad %zu", pad);
314 if (pad)
315 memset(skb_put(skb, pad), 0, pad);
316 }
317
318 seq = call->tx_top + 1;
319
320 sp->hdr.seq = seq;
321 sp->hdr._rsvd = 0;
322 sp->hdr.flags = conn->out_clientflag;
323
324 if (msg_data_left(msg) == 0 && !more)
325 sp->hdr.flags |= RXRPC_LAST_PACKET;
326 else if (call->tx_top - call->tx_hard_ack <
327 call->tx_winsize)
328 sp->hdr.flags |= RXRPC_MORE_PACKETS;
329
330 ret = conn->security->secure_packet(
331 call, skb, skb->mark, skb->head);
332 if (ret < 0)
333 goto out;
334
335 rxrpc_queue_packet(call, skb, !msg_data_left(msg) && !more);
336 skb = NULL;
337 }
338 } while (msg_data_left(msg) > 0);
339
340 success:
341 ret = copied;
342 out:
343 call->tx_pending = skb;
344 _leave(" = %d", ret);
345 return ret;
346
347 call_terminated:
348 rxrpc_free_skb(skb, rxrpc_skb_tx_freed);
349 _leave(" = %d", -call->error);
350 return -call->error;
351
352 maybe_error:
353 if (copied)
354 goto success;
355 goto out;
356
357 efault:
358 ret = -EFAULT;
359 goto out;
360 }
361
362 /*
363 * extract control messages from the sendmsg() control buffer
364 */
365 static int rxrpc_sendmsg_cmsg(struct msghdr *msg,
366 unsigned long *user_call_ID,
367 enum rxrpc_command *command,
368 u32 *abort_code,
369 bool *_exclusive,
370 bool *_upgrade)
371 {
372 struct cmsghdr *cmsg;
373 bool got_user_ID = false;
374 int len;
375
376 *command = RXRPC_CMD_SEND_DATA;
377
378 if (msg->msg_controllen == 0)
379 return -EINVAL;
380
381 for_each_cmsghdr(cmsg, msg) {
382 if (!CMSG_OK(msg, cmsg))
383 return -EINVAL;
384
385 len = cmsg->cmsg_len - sizeof(struct cmsghdr);
386 _debug("CMSG %d, %d, %d",
387 cmsg->cmsg_level, cmsg->cmsg_type, len);
388
389 if (cmsg->cmsg_level != SOL_RXRPC)
390 continue;
391
392 switch (cmsg->cmsg_type) {
393 case RXRPC_USER_CALL_ID:
394 if (msg->msg_flags & MSG_CMSG_COMPAT) {
395 if (len != sizeof(u32))
396 return -EINVAL;
397 *user_call_ID = *(u32 *) CMSG_DATA(cmsg);
398 } else {
399 if (len != sizeof(unsigned long))
400 return -EINVAL;
401 *user_call_ID = *(unsigned long *)
402 CMSG_DATA(cmsg);
403 }
404 _debug("User Call ID %lx", *user_call_ID);
405 got_user_ID = true;
406 break;
407
408 case RXRPC_ABORT:
409 if (*command != RXRPC_CMD_SEND_DATA)
410 return -EINVAL;
411 *command = RXRPC_CMD_SEND_ABORT;
412 if (len != sizeof(*abort_code))
413 return -EINVAL;
414 *abort_code = *(unsigned int *) CMSG_DATA(cmsg);
415 _debug("Abort %x", *abort_code);
416 if (*abort_code == 0)
417 return -EINVAL;
418 break;
419
420 case RXRPC_ACCEPT:
421 if (*command != RXRPC_CMD_SEND_DATA)
422 return -EINVAL;
423 *command = RXRPC_CMD_ACCEPT;
424 if (len != 0)
425 return -EINVAL;
426 break;
427
428 case RXRPC_EXCLUSIVE_CALL:
429 *_exclusive = true;
430 if (len != 0)
431 return -EINVAL;
432 break;
433
434 case RXRPC_UPGRADE_SERVICE:
435 *_upgrade = true;
436 if (len != 0)
437 return -EINVAL;
438 break;
439
440 default:
441 return -EINVAL;
442 }
443 }
444
445 if (!got_user_ID)
446 return -EINVAL;
447 _leave(" = 0");
448 return 0;
449 }
450
451 /*
452 * Create a new client call for sendmsg().
453 * - Called with the socket lock held, which it must release.
454 * - If it returns a call, the call's lock will need releasing by the caller.
455 */
456 static struct rxrpc_call *
457 rxrpc_new_client_call_for_sendmsg(struct rxrpc_sock *rx, struct msghdr *msg,
458 unsigned long user_call_ID, bool exclusive,
459 bool upgrade)
460 __releases(&rx->sk.sk_lock.slock)
461 {
462 struct rxrpc_conn_parameters cp;
463 struct rxrpc_call *call;
464 struct key *key;
465
466 DECLARE_SOCKADDR(struct sockaddr_rxrpc *, srx, msg->msg_name);
467
468 _enter("");
469
470 if (!msg->msg_name) {
471 release_sock(&rx->sk);
472 return ERR_PTR(-EDESTADDRREQ);
473 }
474
475 key = rx->key;
476 if (key && !rx->key->payload.data[0])
477 key = NULL;
478
479 memset(&cp, 0, sizeof(cp));
480 cp.local = rx->local;
481 cp.key = rx->key;
482 cp.security_level = rx->min_sec_level;
483 cp.exclusive = rx->exclusive | exclusive;
484 cp.upgrade = upgrade;
485 cp.service_id = srx->srx_service;
486 call = rxrpc_new_client_call(rx, &cp, srx, user_call_ID, GFP_KERNEL);
487 /* The socket is now unlocked */
488
489 _leave(" = %p\n", call);
490 return call;
491 }
492
493 /*
494 * send a message forming part of a client call through an RxRPC socket
495 * - caller holds the socket locked
496 * - the socket may be either a client socket or a server socket
497 */
498 int rxrpc_do_sendmsg(struct rxrpc_sock *rx, struct msghdr *msg, size_t len)
499 __releases(&rx->sk.sk_lock.slock)
500 {
501 enum rxrpc_call_state state;
502 enum rxrpc_command cmd;
503 struct rxrpc_call *call;
504 unsigned long user_call_ID = 0;
505 bool exclusive = false;
506 bool upgrade = true;
507 u32 abort_code = 0;
508 int ret;
509
510 _enter("");
511
512 ret = rxrpc_sendmsg_cmsg(msg, &user_call_ID, &cmd, &abort_code,
513 &exclusive, &upgrade);
514 if (ret < 0)
515 goto error_release_sock;
516
517 if (cmd == RXRPC_CMD_ACCEPT) {
518 ret = -EINVAL;
519 if (rx->sk.sk_state != RXRPC_SERVER_LISTENING)
520 goto error_release_sock;
521 call = rxrpc_accept_call(rx, user_call_ID, NULL);
522 /* The socket is now unlocked. */
523 if (IS_ERR(call))
524 return PTR_ERR(call);
525 rxrpc_put_call(call, rxrpc_call_put);
526 return 0;
527 }
528
529 call = rxrpc_find_call_by_user_ID(rx, user_call_ID);
530 if (!call) {
531 ret = -EBADSLT;
532 if (cmd != RXRPC_CMD_SEND_DATA)
533 goto error_release_sock;
534 call = rxrpc_new_client_call_for_sendmsg(rx, msg, user_call_ID,
535 exclusive, upgrade);
536 /* The socket is now unlocked... */
537 if (IS_ERR(call))
538 return PTR_ERR(call);
539 /* ... and we have the call lock. */
540 } else {
541 switch (READ_ONCE(call->state)) {
542 case RXRPC_CALL_UNINITIALISED:
543 case RXRPC_CALL_CLIENT_AWAIT_CONN:
544 case RXRPC_CALL_SERVER_PREALLOC:
545 case RXRPC_CALL_SERVER_SECURING:
546 case RXRPC_CALL_SERVER_ACCEPTING:
547 ret = -EBUSY;
548 goto error_release_sock;
549 default:
550 break;
551 }
552
553 ret = mutex_lock_interruptible(&call->user_mutex);
554 release_sock(&rx->sk);
555 if (ret < 0) {
556 ret = -ERESTARTSYS;
557 goto error_put;
558 }
559 }
560
561 state = READ_ONCE(call->state);
562 _debug("CALL %d USR %lx ST %d on CONN %p",
563 call->debug_id, call->user_call_ID, state, call->conn);
564
565 if (state >= RXRPC_CALL_COMPLETE) {
566 /* it's too late for this call */
567 ret = -ESHUTDOWN;
568 } else if (cmd == RXRPC_CMD_SEND_ABORT) {
569 ret = 0;
570 if (rxrpc_abort_call("CMD", call, 0, abort_code, -ECONNABORTED))
571 ret = rxrpc_send_abort_packet(call);
572 } else if (cmd != RXRPC_CMD_SEND_DATA) {
573 ret = -EINVAL;
574 } else if (rxrpc_is_client_call(call) &&
575 state != RXRPC_CALL_CLIENT_SEND_REQUEST) {
576 /* request phase complete for this client call */
577 ret = -EPROTO;
578 } else if (rxrpc_is_service_call(call) &&
579 state != RXRPC_CALL_SERVER_ACK_REQUEST &&
580 state != RXRPC_CALL_SERVER_SEND_REPLY) {
581 /* Reply phase not begun or not complete for service call. */
582 ret = -EPROTO;
583 } else {
584 ret = rxrpc_send_data(rx, call, msg, len);
585 }
586
587 mutex_unlock(&call->user_mutex);
588 error_put:
589 rxrpc_put_call(call, rxrpc_call_put);
590 _leave(" = %d", ret);
591 return ret;
592
593 error_release_sock:
594 release_sock(&rx->sk);
595 return ret;
596 }
597
598 /**
599 * rxrpc_kernel_send_data - Allow a kernel service to send data on a call
600 * @sock: The socket the call is on
601 * @call: The call to send data through
602 * @msg: The data to send
603 * @len: The amount of data to send
604 *
605 * Allow a kernel service to send data on a call. The call must be in an state
606 * appropriate to sending data. No control data should be supplied in @msg,
607 * nor should an address be supplied. MSG_MORE should be flagged if there's
608 * more data to come, otherwise this data will end the transmission phase.
609 */
610 int rxrpc_kernel_send_data(struct socket *sock, struct rxrpc_call *call,
611 struct msghdr *msg, size_t len)
612 {
613 int ret;
614
615 _enter("{%d,%s},", call->debug_id, rxrpc_call_states[call->state]);
616
617 ASSERTCMP(msg->msg_name, ==, NULL);
618 ASSERTCMP(msg->msg_control, ==, NULL);
619
620 mutex_lock(&call->user_mutex);
621
622 _debug("CALL %d USR %lx ST %d on CONN %p",
623 call->debug_id, call->user_call_ID, call->state, call->conn);
624
625 switch (READ_ONCE(call->state)) {
626 case RXRPC_CALL_CLIENT_SEND_REQUEST:
627 case RXRPC_CALL_SERVER_ACK_REQUEST:
628 case RXRPC_CALL_SERVER_SEND_REPLY:
629 ret = rxrpc_send_data(rxrpc_sk(sock->sk), call, msg, len);
630 break;
631 case RXRPC_CALL_COMPLETE:
632 read_lock_bh(&call->state_lock);
633 ret = -call->error;
634 read_unlock_bh(&call->state_lock);
635 break;
636 default:
637 /* Request phase complete for this client call */
638 trace_rxrpc_rx_eproto(call, 0, tracepoint_string("late_send"));
639 ret = -EPROTO;
640 break;
641 }
642
643 mutex_unlock(&call->user_mutex);
644 _leave(" = %d", ret);
645 return ret;
646 }
647 EXPORT_SYMBOL(rxrpc_kernel_send_data);
648
649 /**
650 * rxrpc_kernel_abort_call - Allow a kernel service to abort a call
651 * @sock: The socket the call is on
652 * @call: The call to be aborted
653 * @abort_code: The abort code to stick into the ABORT packet
654 * @error: Local error value
655 * @why: 3-char string indicating why.
656 *
657 * Allow a kernel service to abort a call, if it's still in an abortable state
658 * and return true if the call was aborted, false if it was already complete.
659 */
660 bool rxrpc_kernel_abort_call(struct socket *sock, struct rxrpc_call *call,
661 u32 abort_code, int error, const char *why)
662 {
663 bool aborted;
664
665 _enter("{%d},%d,%d,%s", call->debug_id, abort_code, error, why);
666
667 mutex_lock(&call->user_mutex);
668
669 aborted = rxrpc_abort_call(why, call, 0, abort_code, error);
670 if (aborted)
671 rxrpc_send_abort_packet(call);
672
673 mutex_unlock(&call->user_mutex);
674 return aborted;
675 }
676
677 EXPORT_SYMBOL(rxrpc_kernel_abort_call);