Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
[cascardo/linux.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 <net/sock.h>
19 #include <net/af_rxrpc.h>
20 #include "ar-internal.h"
21
22 enum rxrpc_command {
23         RXRPC_CMD_SEND_DATA,            /* send data message */
24         RXRPC_CMD_SEND_ABORT,           /* request abort generation */
25         RXRPC_CMD_ACCEPT,               /* [server] accept incoming call */
26         RXRPC_CMD_REJECT_BUSY,          /* [server] reject a call as busy */
27 };
28
29 /*
30  * wait for space to appear in the transmit/ACK window
31  * - caller holds the socket locked
32  */
33 static int rxrpc_wait_for_tx_window(struct rxrpc_sock *rx,
34                                     struct rxrpc_call *call,
35                                     long *timeo)
36 {
37         DECLARE_WAITQUEUE(myself, current);
38         int ret;
39
40         _enter(",{%u,%u,%u}",
41                call->tx_hard_ack, call->tx_top, call->tx_winsize);
42
43         add_wait_queue(&call->waitq, &myself);
44
45         for (;;) {
46                 set_current_state(TASK_INTERRUPTIBLE);
47                 ret = 0;
48                 if (call->tx_top - call->tx_hard_ack < call->tx_winsize)
49                         break;
50                 if (call->state >= RXRPC_CALL_COMPLETE) {
51                         ret = -call->error;
52                         break;
53                 }
54                 if (signal_pending(current)) {
55                         ret = sock_intr_errno(*timeo);
56                         break;
57                 }
58
59                 trace_rxrpc_transmit(call, rxrpc_transmit_wait);
60                 release_sock(&rx->sk);
61                 *timeo = schedule_timeout(*timeo);
62                 lock_sock(&rx->sk);
63         }
64
65         remove_wait_queue(&call->waitq, &myself);
66         set_current_state(TASK_RUNNING);
67         _leave(" = %d", ret);
68         return ret;
69 }
70
71 /*
72  * Schedule an instant Tx resend.
73  */
74 static inline void rxrpc_instant_resend(struct rxrpc_call *call, int ix)
75 {
76         spin_lock_bh(&call->lock);
77
78         if (call->state < RXRPC_CALL_COMPLETE) {
79                 call->rxtx_annotations[ix] = RXRPC_TX_ANNO_RETRANS;
80                 if (!test_and_set_bit(RXRPC_CALL_EV_RESEND, &call->events))
81                         rxrpc_queue_call(call);
82         }
83
84         spin_unlock_bh(&call->lock);
85 }
86
87 /*
88  * Queue a DATA packet for transmission, set the resend timeout and send the
89  * packet immediately
90  */
91 static void rxrpc_queue_packet(struct rxrpc_call *call, struct sk_buff *skb,
92                                bool last)
93 {
94         struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
95         rxrpc_seq_t seq = sp->hdr.seq;
96         int ret, ix;
97
98         _net("queue skb %p [%d]", skb, seq);
99
100         ASSERTCMP(seq, ==, call->tx_top + 1);
101
102         ix = seq & RXRPC_RXTX_BUFF_MASK;
103         rxrpc_get_skb(skb, rxrpc_skb_tx_got);
104         call->rxtx_annotations[ix] = RXRPC_TX_ANNO_UNACK;
105         smp_wmb();
106         call->rxtx_buffer[ix] = skb;
107         call->tx_top = seq;
108         if (last) {
109                 set_bit(RXRPC_CALL_TX_LAST, &call->flags);
110                 trace_rxrpc_transmit(call, rxrpc_transmit_queue_last);
111         } else if (sp->hdr.flags & RXRPC_REQUEST_ACK) {
112                 trace_rxrpc_transmit(call, rxrpc_transmit_queue_reqack);
113         } else {
114                 trace_rxrpc_transmit(call, rxrpc_transmit_queue);
115         }
116
117         if (last || call->state == RXRPC_CALL_SERVER_ACK_REQUEST) {
118                 _debug("________awaiting reply/ACK__________");
119                 write_lock_bh(&call->state_lock);
120                 switch (call->state) {
121                 case RXRPC_CALL_CLIENT_SEND_REQUEST:
122                         call->state = RXRPC_CALL_CLIENT_AWAIT_REPLY;
123                         break;
124                 case RXRPC_CALL_SERVER_ACK_REQUEST:
125                         call->state = RXRPC_CALL_SERVER_SEND_REPLY;
126                         if (!last)
127                                 break;
128                 case RXRPC_CALL_SERVER_SEND_REPLY:
129                         call->state = RXRPC_CALL_SERVER_AWAIT_ACK;
130                         break;
131                 default:
132                         break;
133                 }
134                 write_unlock_bh(&call->state_lock);
135         }
136
137         if (seq == 1 && rxrpc_is_client_call(call))
138                 rxrpc_expose_client_call(call);
139
140         ret = rxrpc_send_data_packet(call, skb);
141         if (ret < 0) {
142                 _debug("need instant resend %d", ret);
143                 rxrpc_instant_resend(call, ix);
144         }
145
146         rxrpc_free_skb(skb, rxrpc_skb_tx_freed);
147         _leave("");
148 }
149
150 /*
151  * send data through a socket
152  * - must be called in process context
153  * - caller holds the socket locked
154  */
155 static int rxrpc_send_data(struct rxrpc_sock *rx,
156                            struct rxrpc_call *call,
157                            struct msghdr *msg, size_t len)
158 {
159         struct rxrpc_skb_priv *sp;
160         struct sk_buff *skb;
161         struct sock *sk = &rx->sk;
162         long timeo;
163         bool more;
164         int ret, copied;
165
166         timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
167
168         /* this should be in poll */
169         sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
170
171         if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
172                 return -EPIPE;
173
174         more = msg->msg_flags & MSG_MORE;
175
176         skb = call->tx_pending;
177         call->tx_pending = NULL;
178         rxrpc_see_skb(skb, rxrpc_skb_tx_seen);
179
180         copied = 0;
181         do {
182                 /* Check to see if there's a ping ACK to reply to. */
183                 if (call->ackr_reason == RXRPC_ACK_PING_RESPONSE)
184                         rxrpc_send_call_packet(call, RXRPC_PACKET_TYPE_ACK);
185
186                 if (!skb) {
187                         size_t size, chunk, max, space;
188
189                         _debug("alloc");
190
191                         if (call->tx_top - call->tx_hard_ack >=
192                             call->tx_winsize) {
193                                 ret = -EAGAIN;
194                                 if (msg->msg_flags & MSG_DONTWAIT)
195                                         goto maybe_error;
196                                 ret = rxrpc_wait_for_tx_window(rx, call,
197                                                                &timeo);
198                                 if (ret < 0)
199                                         goto maybe_error;
200                         }
201
202                         max = RXRPC_JUMBO_DATALEN;
203                         max -= call->conn->security_size;
204                         max &= ~(call->conn->size_align - 1UL);
205
206                         chunk = max;
207                         if (chunk > msg_data_left(msg) && !more)
208                                 chunk = msg_data_left(msg);
209
210                         space = chunk + call->conn->size_align;
211                         space &= ~(call->conn->size_align - 1UL);
212
213                         size = space + call->conn->security_size;
214
215                         _debug("SIZE: %zu/%zu/%zu", chunk, space, size);
216
217                         /* create a buffer that we can retain until it's ACK'd */
218                         skb = sock_alloc_send_skb(
219                                 sk, size, msg->msg_flags & MSG_DONTWAIT, &ret);
220                         if (!skb)
221                                 goto maybe_error;
222
223                         rxrpc_new_skb(skb, rxrpc_skb_tx_new);
224
225                         _debug("ALLOC SEND %p", skb);
226
227                         ASSERTCMP(skb->mark, ==, 0);
228
229                         _debug("HS: %u", call->conn->security_size);
230                         skb_reserve(skb, call->conn->security_size);
231                         skb->len += call->conn->security_size;
232
233                         sp = rxrpc_skb(skb);
234                         sp->remain = chunk;
235                         if (sp->remain > skb_tailroom(skb))
236                                 sp->remain = skb_tailroom(skb);
237
238                         _net("skb: hr %d, tr %d, hl %d, rm %d",
239                                skb_headroom(skb),
240                                skb_tailroom(skb),
241                                skb_headlen(skb),
242                                sp->remain);
243
244                         skb->ip_summed = CHECKSUM_UNNECESSARY;
245                 }
246
247                 _debug("append");
248                 sp = rxrpc_skb(skb);
249
250                 /* append next segment of data to the current buffer */
251                 if (msg_data_left(msg) > 0) {
252                         int copy = skb_tailroom(skb);
253                         ASSERTCMP(copy, >, 0);
254                         if (copy > msg_data_left(msg))
255                                 copy = msg_data_left(msg);
256                         if (copy > sp->remain)
257                                 copy = sp->remain;
258
259                         _debug("add");
260                         ret = skb_add_data(skb, &msg->msg_iter, copy);
261                         _debug("added");
262                         if (ret < 0)
263                                 goto efault;
264                         sp->remain -= copy;
265                         skb->mark += copy;
266                         copied += copy;
267                 }
268
269                 /* check for the far side aborting the call or a network error
270                  * occurring */
271                 if (call->state == RXRPC_CALL_COMPLETE)
272                         goto call_terminated;
273
274                 /* add the packet to the send queue if it's now full */
275                 if (sp->remain <= 0 ||
276                     (msg_data_left(msg) == 0 && !more)) {
277                         struct rxrpc_connection *conn = call->conn;
278                         uint32_t seq;
279                         size_t pad;
280
281                         /* pad out if we're using security */
282                         if (conn->security_ix) {
283                                 pad = conn->security_size + skb->mark;
284                                 pad = conn->size_align - pad;
285                                 pad &= conn->size_align - 1;
286                                 _debug("pad %zu", pad);
287                                 if (pad)
288                                         memset(skb_put(skb, pad), 0, pad);
289                         }
290
291                         seq = call->tx_top + 1;
292
293                         sp->hdr.seq     = seq;
294                         sp->hdr._rsvd   = 0;
295                         sp->hdr.flags   = conn->out_clientflag;
296
297                         if (msg_data_left(msg) == 0 && !more)
298                                 sp->hdr.flags |= RXRPC_LAST_PACKET;
299                         else if (call->tx_top - call->tx_hard_ack <
300                                  call->tx_winsize)
301                                 sp->hdr.flags |= RXRPC_MORE_PACKETS;
302
303                         ret = conn->security->secure_packet(
304                                 call, skb, skb->mark, skb->head);
305                         if (ret < 0)
306                                 goto out;
307
308                         rxrpc_queue_packet(call, skb, !msg_data_left(msg) && !more);
309                         skb = NULL;
310                 }
311         } while (msg_data_left(msg) > 0);
312
313 success:
314         ret = copied;
315 out:
316         call->tx_pending = skb;
317         _leave(" = %d", ret);
318         return ret;
319
320 call_terminated:
321         rxrpc_free_skb(skb, rxrpc_skb_tx_freed);
322         _leave(" = %d", -call->error);
323         return -call->error;
324
325 maybe_error:
326         if (copied)
327                 goto success;
328         goto out;
329
330 efault:
331         ret = -EFAULT;
332         goto out;
333 }
334
335 /*
336  * extract control messages from the sendmsg() control buffer
337  */
338 static int rxrpc_sendmsg_cmsg(struct msghdr *msg,
339                               unsigned long *user_call_ID,
340                               enum rxrpc_command *command,
341                               u32 *abort_code,
342                               bool *_exclusive)
343 {
344         struct cmsghdr *cmsg;
345         bool got_user_ID = false;
346         int len;
347
348         *command = RXRPC_CMD_SEND_DATA;
349
350         if (msg->msg_controllen == 0)
351                 return -EINVAL;
352
353         for_each_cmsghdr(cmsg, msg) {
354                 if (!CMSG_OK(msg, cmsg))
355                         return -EINVAL;
356
357                 len = cmsg->cmsg_len - CMSG_ALIGN(sizeof(struct cmsghdr));
358                 _debug("CMSG %d, %d, %d",
359                        cmsg->cmsg_level, cmsg->cmsg_type, len);
360
361                 if (cmsg->cmsg_level != SOL_RXRPC)
362                         continue;
363
364                 switch (cmsg->cmsg_type) {
365                 case RXRPC_USER_CALL_ID:
366                         if (msg->msg_flags & MSG_CMSG_COMPAT) {
367                                 if (len != sizeof(u32))
368                                         return -EINVAL;
369                                 *user_call_ID = *(u32 *) CMSG_DATA(cmsg);
370                         } else {
371                                 if (len != sizeof(unsigned long))
372                                         return -EINVAL;
373                                 *user_call_ID = *(unsigned long *)
374                                         CMSG_DATA(cmsg);
375                         }
376                         _debug("User Call ID %lx", *user_call_ID);
377                         got_user_ID = true;
378                         break;
379
380                 case RXRPC_ABORT:
381                         if (*command != RXRPC_CMD_SEND_DATA)
382                                 return -EINVAL;
383                         *command = RXRPC_CMD_SEND_ABORT;
384                         if (len != sizeof(*abort_code))
385                                 return -EINVAL;
386                         *abort_code = *(unsigned int *) CMSG_DATA(cmsg);
387                         _debug("Abort %x", *abort_code);
388                         if (*abort_code == 0)
389                                 return -EINVAL;
390                         break;
391
392                 case RXRPC_ACCEPT:
393                         if (*command != RXRPC_CMD_SEND_DATA)
394                                 return -EINVAL;
395                         *command = RXRPC_CMD_ACCEPT;
396                         if (len != 0)
397                                 return -EINVAL;
398                         break;
399
400                 case RXRPC_EXCLUSIVE_CALL:
401                         *_exclusive = true;
402                         if (len != 0)
403                                 return -EINVAL;
404                         break;
405                 default:
406                         return -EINVAL;
407                 }
408         }
409
410         if (!got_user_ID)
411                 return -EINVAL;
412         _leave(" = 0");
413         return 0;
414 }
415
416 /*
417  * Create a new client call for sendmsg().
418  */
419 static struct rxrpc_call *
420 rxrpc_new_client_call_for_sendmsg(struct rxrpc_sock *rx, struct msghdr *msg,
421                                   unsigned long user_call_ID, bool exclusive)
422 {
423         struct rxrpc_conn_parameters cp;
424         struct rxrpc_call *call;
425         struct key *key;
426
427         DECLARE_SOCKADDR(struct sockaddr_rxrpc *, srx, msg->msg_name);
428
429         _enter("");
430
431         if (!msg->msg_name)
432                 return ERR_PTR(-EDESTADDRREQ);
433
434         key = rx->key;
435         if (key && !rx->key->payload.data[0])
436                 key = NULL;
437
438         memset(&cp, 0, sizeof(cp));
439         cp.local                = rx->local;
440         cp.key                  = rx->key;
441         cp.security_level       = rx->min_sec_level;
442         cp.exclusive            = rx->exclusive | exclusive;
443         cp.service_id           = srx->srx_service;
444         call = rxrpc_new_client_call(rx, &cp, srx, user_call_ID, GFP_KERNEL);
445
446         _leave(" = %p\n", call);
447         return call;
448 }
449
450 /*
451  * send a message forming part of a client call through an RxRPC socket
452  * - caller holds the socket locked
453  * - the socket may be either a client socket or a server socket
454  */
455 int rxrpc_do_sendmsg(struct rxrpc_sock *rx, struct msghdr *msg, size_t len)
456 {
457         enum rxrpc_command cmd;
458         struct rxrpc_call *call;
459         unsigned long user_call_ID = 0;
460         bool exclusive = false;
461         u32 abort_code = 0;
462         int ret;
463
464         _enter("");
465
466         ret = rxrpc_sendmsg_cmsg(msg, &user_call_ID, &cmd, &abort_code,
467                                  &exclusive);
468         if (ret < 0)
469                 return ret;
470
471         if (cmd == RXRPC_CMD_ACCEPT) {
472                 if (rx->sk.sk_state != RXRPC_SERVER_LISTENING)
473                         return -EINVAL;
474                 call = rxrpc_accept_call(rx, user_call_ID, NULL);
475                 if (IS_ERR(call))
476                         return PTR_ERR(call);
477                 rxrpc_put_call(call, rxrpc_call_put);
478                 return 0;
479         }
480
481         call = rxrpc_find_call_by_user_ID(rx, user_call_ID);
482         if (!call) {
483                 if (cmd != RXRPC_CMD_SEND_DATA)
484                         return -EBADSLT;
485                 call = rxrpc_new_client_call_for_sendmsg(rx, msg, user_call_ID,
486                                                          exclusive);
487                 if (IS_ERR(call))
488                         return PTR_ERR(call);
489         }
490
491         _debug("CALL %d USR %lx ST %d on CONN %p",
492                call->debug_id, call->user_call_ID, call->state, call->conn);
493
494         if (call->state >= RXRPC_CALL_COMPLETE) {
495                 /* it's too late for this call */
496                 ret = -ESHUTDOWN;
497         } else if (cmd == RXRPC_CMD_SEND_ABORT) {
498                 ret = 0;
499                 if (rxrpc_abort_call("CMD", call, 0, abort_code, ECONNABORTED))
500                         ret = rxrpc_send_call_packet(call,
501                                                      RXRPC_PACKET_TYPE_ABORT);
502         } else if (cmd != RXRPC_CMD_SEND_DATA) {
503                 ret = -EINVAL;
504         } else if (rxrpc_is_client_call(call) &&
505                    call->state != RXRPC_CALL_CLIENT_SEND_REQUEST) {
506                 /* request phase complete for this client call */
507                 ret = -EPROTO;
508         } else if (rxrpc_is_service_call(call) &&
509                    call->state != RXRPC_CALL_SERVER_ACK_REQUEST &&
510                    call->state != RXRPC_CALL_SERVER_SEND_REPLY) {
511                 /* Reply phase not begun or not complete for service call. */
512                 ret = -EPROTO;
513         } else {
514                 ret = rxrpc_send_data(rx, call, msg, len);
515         }
516
517         rxrpc_put_call(call, rxrpc_call_put);
518         _leave(" = %d", ret);
519         return ret;
520 }
521
522 /**
523  * rxrpc_kernel_send_data - Allow a kernel service to send data on a call
524  * @sock: The socket the call is on
525  * @call: The call to send data through
526  * @msg: The data to send
527  * @len: The amount of data to send
528  *
529  * Allow a kernel service to send data on a call.  The call must be in an state
530  * appropriate to sending data.  No control data should be supplied in @msg,
531  * nor should an address be supplied.  MSG_MORE should be flagged if there's
532  * more data to come, otherwise this data will end the transmission phase.
533  */
534 int rxrpc_kernel_send_data(struct socket *sock, struct rxrpc_call *call,
535                            struct msghdr *msg, size_t len)
536 {
537         int ret;
538
539         _enter("{%d,%s},", call->debug_id, rxrpc_call_states[call->state]);
540
541         ASSERTCMP(msg->msg_name, ==, NULL);
542         ASSERTCMP(msg->msg_control, ==, NULL);
543
544         lock_sock(sock->sk);
545
546         _debug("CALL %d USR %lx ST %d on CONN %p",
547                call->debug_id, call->user_call_ID, call->state, call->conn);
548
549         if (call->state >= RXRPC_CALL_COMPLETE) {
550                 ret = -ESHUTDOWN; /* it's too late for this call */
551         } else if (call->state != RXRPC_CALL_CLIENT_SEND_REQUEST &&
552                    call->state != RXRPC_CALL_SERVER_ACK_REQUEST &&
553                    call->state != RXRPC_CALL_SERVER_SEND_REPLY) {
554                 ret = -EPROTO; /* request phase complete for this client call */
555         } else {
556                 ret = rxrpc_send_data(rxrpc_sk(sock->sk), call, msg, len);
557         }
558
559         release_sock(sock->sk);
560         _leave(" = %d", ret);
561         return ret;
562 }
563 EXPORT_SYMBOL(rxrpc_kernel_send_data);
564
565 /**
566  * rxrpc_kernel_abort_call - Allow a kernel service to abort a call
567  * @sock: The socket the call is on
568  * @call: The call to be aborted
569  * @abort_code: The abort code to stick into the ABORT packet
570  * @error: Local error value
571  * @why: 3-char string indicating why.
572  *
573  * Allow a kernel service to abort a call, if it's still in an abortable state.
574  */
575 void rxrpc_kernel_abort_call(struct socket *sock, struct rxrpc_call *call,
576                              u32 abort_code, int error, const char *why)
577 {
578         _enter("{%d},%d,%d,%s", call->debug_id, abort_code, error, why);
579
580         lock_sock(sock->sk);
581
582         if (rxrpc_abort_call(why, call, 0, abort_code, error))
583                 rxrpc_send_call_packet(call, RXRPC_PACKET_TYPE_ABORT);
584
585         release_sock(sock->sk);
586         _leave("");
587 }
588
589 EXPORT_SYMBOL(rxrpc_kernel_abort_call);