Book a Demo!
CoCalc Logo Icon
StoreFeaturesDocsShareSupportNewsAboutPoliciesSign UpSign In
awilliam
GitHub Repository: awilliam/linux-vfio
Path: blob/master/net/rxrpc/ar-recvmsg.c
15109 views
1
/* RxRPC recvmsg() implementation
2
*
3
* Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
4
* Written by David Howells ([email protected])
5
*
6
* This program is free software; you can redistribute it and/or
7
* modify it under the terms of the GNU General Public License
8
* as published by the Free Software Foundation; either version
9
* 2 of the License, or (at your option) any later version.
10
*/
11
12
#include <linux/net.h>
13
#include <linux/skbuff.h>
14
#include <net/sock.h>
15
#include <net/af_rxrpc.h>
16
#include "ar-internal.h"
17
18
/*
19
* removal a call's user ID from the socket tree to make the user ID available
20
* again and so that it won't be seen again in association with that call
21
*/
22
void rxrpc_remove_user_ID(struct rxrpc_sock *rx, struct rxrpc_call *call)
23
{
24
_debug("RELEASE CALL %d", call->debug_id);
25
26
if (test_bit(RXRPC_CALL_HAS_USERID, &call->flags)) {
27
write_lock_bh(&rx->call_lock);
28
rb_erase(&call->sock_node, &call->socket->calls);
29
clear_bit(RXRPC_CALL_HAS_USERID, &call->flags);
30
write_unlock_bh(&rx->call_lock);
31
}
32
33
read_lock_bh(&call->state_lock);
34
if (!test_bit(RXRPC_CALL_RELEASED, &call->flags) &&
35
!test_and_set_bit(RXRPC_CALL_RELEASE, &call->events))
36
rxrpc_queue_call(call);
37
read_unlock_bh(&call->state_lock);
38
}
39
40
/*
41
* receive a message from an RxRPC socket
42
* - we need to be careful about two or more threads calling recvmsg
43
* simultaneously
44
*/
45
int rxrpc_recvmsg(struct kiocb *iocb, struct socket *sock,
46
struct msghdr *msg, size_t len, int flags)
47
{
48
struct rxrpc_skb_priv *sp;
49
struct rxrpc_call *call = NULL, *continue_call = NULL;
50
struct rxrpc_sock *rx = rxrpc_sk(sock->sk);
51
struct sk_buff *skb;
52
long timeo;
53
int copy, ret, ullen, offset, copied = 0;
54
u32 abort_code;
55
56
DEFINE_WAIT(wait);
57
58
_enter(",,,%zu,%d", len, flags);
59
60
if (flags & (MSG_OOB | MSG_TRUNC))
61
return -EOPNOTSUPP;
62
63
ullen = msg->msg_flags & MSG_CMSG_COMPAT ? 4 : sizeof(unsigned long);
64
65
timeo = sock_rcvtimeo(&rx->sk, flags & MSG_DONTWAIT);
66
msg->msg_flags |= MSG_MORE;
67
68
lock_sock(&rx->sk);
69
70
for (;;) {
71
/* return immediately if a client socket has no outstanding
72
* calls */
73
if (RB_EMPTY_ROOT(&rx->calls)) {
74
if (copied)
75
goto out;
76
if (rx->sk.sk_state != RXRPC_SERVER_LISTENING) {
77
release_sock(&rx->sk);
78
if (continue_call)
79
rxrpc_put_call(continue_call);
80
return -ENODATA;
81
}
82
}
83
84
/* get the next message on the Rx queue */
85
skb = skb_peek(&rx->sk.sk_receive_queue);
86
if (!skb) {
87
/* nothing remains on the queue */
88
if (copied &&
89
(msg->msg_flags & MSG_PEEK || timeo == 0))
90
goto out;
91
92
/* wait for a message to turn up */
93
release_sock(&rx->sk);
94
prepare_to_wait_exclusive(sk_sleep(&rx->sk), &wait,
95
TASK_INTERRUPTIBLE);
96
ret = sock_error(&rx->sk);
97
if (ret)
98
goto wait_error;
99
100
if (skb_queue_empty(&rx->sk.sk_receive_queue)) {
101
if (signal_pending(current))
102
goto wait_interrupted;
103
timeo = schedule_timeout(timeo);
104
}
105
finish_wait(sk_sleep(&rx->sk), &wait);
106
lock_sock(&rx->sk);
107
continue;
108
}
109
110
peek_next_packet:
111
sp = rxrpc_skb(skb);
112
call = sp->call;
113
ASSERT(call != NULL);
114
115
_debug("next pkt %s", rxrpc_pkts[sp->hdr.type]);
116
117
/* make sure we wait for the state to be updated in this call */
118
spin_lock_bh(&call->lock);
119
spin_unlock_bh(&call->lock);
120
121
if (test_bit(RXRPC_CALL_RELEASED, &call->flags)) {
122
_debug("packet from released call");
123
if (skb_dequeue(&rx->sk.sk_receive_queue) != skb)
124
BUG();
125
rxrpc_free_skb(skb);
126
continue;
127
}
128
129
/* determine whether to continue last data receive */
130
if (continue_call) {
131
_debug("maybe cont");
132
if (call != continue_call ||
133
skb->mark != RXRPC_SKB_MARK_DATA) {
134
release_sock(&rx->sk);
135
rxrpc_put_call(continue_call);
136
_leave(" = %d [noncont]", copied);
137
return copied;
138
}
139
}
140
141
rxrpc_get_call(call);
142
143
/* copy the peer address and timestamp */
144
if (!continue_call) {
145
if (msg->msg_name && msg->msg_namelen > 0)
146
memcpy(msg->msg_name,
147
&call->conn->trans->peer->srx,
148
sizeof(call->conn->trans->peer->srx));
149
sock_recv_ts_and_drops(msg, &rx->sk, skb);
150
}
151
152
/* receive the message */
153
if (skb->mark != RXRPC_SKB_MARK_DATA)
154
goto receive_non_data_message;
155
156
_debug("recvmsg DATA #%u { %d, %d }",
157
ntohl(sp->hdr.seq), skb->len, sp->offset);
158
159
if (!continue_call) {
160
/* only set the control data once per recvmsg() */
161
ret = put_cmsg(msg, SOL_RXRPC, RXRPC_USER_CALL_ID,
162
ullen, &call->user_call_ID);
163
if (ret < 0)
164
goto copy_error;
165
ASSERT(test_bit(RXRPC_CALL_HAS_USERID, &call->flags));
166
}
167
168
ASSERTCMP(ntohl(sp->hdr.seq), >=, call->rx_data_recv);
169
ASSERTCMP(ntohl(sp->hdr.seq), <=, call->rx_data_recv + 1);
170
call->rx_data_recv = ntohl(sp->hdr.seq);
171
172
ASSERTCMP(ntohl(sp->hdr.seq), >, call->rx_data_eaten);
173
174
offset = sp->offset;
175
copy = skb->len - offset;
176
if (copy > len - copied)
177
copy = len - copied;
178
179
if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
180
ret = skb_copy_datagram_iovec(skb, offset,
181
msg->msg_iov, copy);
182
} else {
183
ret = skb_copy_and_csum_datagram_iovec(skb, offset,
184
msg->msg_iov);
185
if (ret == -EINVAL)
186
goto csum_copy_error;
187
}
188
189
if (ret < 0)
190
goto copy_error;
191
192
/* handle piecemeal consumption of data packets */
193
_debug("copied %d+%d", copy, copied);
194
195
offset += copy;
196
copied += copy;
197
198
if (!(flags & MSG_PEEK))
199
sp->offset = offset;
200
201
if (sp->offset < skb->len) {
202
_debug("buffer full");
203
ASSERTCMP(copied, ==, len);
204
break;
205
}
206
207
/* we transferred the whole data packet */
208
if (sp->hdr.flags & RXRPC_LAST_PACKET) {
209
_debug("last");
210
if (call->conn->out_clientflag) {
211
/* last byte of reply received */
212
ret = copied;
213
goto terminal_message;
214
}
215
216
/* last bit of request received */
217
if (!(flags & MSG_PEEK)) {
218
_debug("eat packet");
219
if (skb_dequeue(&rx->sk.sk_receive_queue) !=
220
skb)
221
BUG();
222
rxrpc_free_skb(skb);
223
}
224
msg->msg_flags &= ~MSG_MORE;
225
break;
226
}
227
228
/* move on to the next data message */
229
_debug("next");
230
if (!continue_call)
231
continue_call = sp->call;
232
else
233
rxrpc_put_call(call);
234
call = NULL;
235
236
if (flags & MSG_PEEK) {
237
_debug("peek next");
238
skb = skb->next;
239
if (skb == (struct sk_buff *) &rx->sk.sk_receive_queue)
240
break;
241
goto peek_next_packet;
242
}
243
244
_debug("eat packet");
245
if (skb_dequeue(&rx->sk.sk_receive_queue) != skb)
246
BUG();
247
rxrpc_free_skb(skb);
248
}
249
250
/* end of non-terminal data packet reception for the moment */
251
_debug("end rcv data");
252
out:
253
release_sock(&rx->sk);
254
if (call)
255
rxrpc_put_call(call);
256
if (continue_call)
257
rxrpc_put_call(continue_call);
258
_leave(" = %d [data]", copied);
259
return copied;
260
261
/* handle non-DATA messages such as aborts, incoming connections and
262
* final ACKs */
263
receive_non_data_message:
264
_debug("non-data");
265
266
if (skb->mark == RXRPC_SKB_MARK_NEW_CALL) {
267
_debug("RECV NEW CALL");
268
ret = put_cmsg(msg, SOL_RXRPC, RXRPC_NEW_CALL, 0, &abort_code);
269
if (ret < 0)
270
goto copy_error;
271
if (!(flags & MSG_PEEK)) {
272
if (skb_dequeue(&rx->sk.sk_receive_queue) != skb)
273
BUG();
274
rxrpc_free_skb(skb);
275
}
276
goto out;
277
}
278
279
ret = put_cmsg(msg, SOL_RXRPC, RXRPC_USER_CALL_ID,
280
ullen, &call->user_call_ID);
281
if (ret < 0)
282
goto copy_error;
283
ASSERT(test_bit(RXRPC_CALL_HAS_USERID, &call->flags));
284
285
switch (skb->mark) {
286
case RXRPC_SKB_MARK_DATA:
287
BUG();
288
case RXRPC_SKB_MARK_FINAL_ACK:
289
ret = put_cmsg(msg, SOL_RXRPC, RXRPC_ACK, 0, &abort_code);
290
break;
291
case RXRPC_SKB_MARK_BUSY:
292
ret = put_cmsg(msg, SOL_RXRPC, RXRPC_BUSY, 0, &abort_code);
293
break;
294
case RXRPC_SKB_MARK_REMOTE_ABORT:
295
abort_code = call->abort_code;
296
ret = put_cmsg(msg, SOL_RXRPC, RXRPC_ABORT, 4, &abort_code);
297
break;
298
case RXRPC_SKB_MARK_NET_ERROR:
299
_debug("RECV NET ERROR %d", sp->error);
300
abort_code = sp->error;
301
ret = put_cmsg(msg, SOL_RXRPC, RXRPC_NET_ERROR, 4, &abort_code);
302
break;
303
case RXRPC_SKB_MARK_LOCAL_ERROR:
304
_debug("RECV LOCAL ERROR %d", sp->error);
305
abort_code = sp->error;
306
ret = put_cmsg(msg, SOL_RXRPC, RXRPC_LOCAL_ERROR, 4,
307
&abort_code);
308
break;
309
default:
310
BUG();
311
break;
312
}
313
314
if (ret < 0)
315
goto copy_error;
316
317
terminal_message:
318
_debug("terminal");
319
msg->msg_flags &= ~MSG_MORE;
320
msg->msg_flags |= MSG_EOR;
321
322
if (!(flags & MSG_PEEK)) {
323
_net("free terminal skb %p", skb);
324
if (skb_dequeue(&rx->sk.sk_receive_queue) != skb)
325
BUG();
326
rxrpc_free_skb(skb);
327
rxrpc_remove_user_ID(rx, call);
328
}
329
330
release_sock(&rx->sk);
331
rxrpc_put_call(call);
332
if (continue_call)
333
rxrpc_put_call(continue_call);
334
_leave(" = %d", ret);
335
return ret;
336
337
copy_error:
338
_debug("copy error");
339
release_sock(&rx->sk);
340
rxrpc_put_call(call);
341
if (continue_call)
342
rxrpc_put_call(continue_call);
343
_leave(" = %d", ret);
344
return ret;
345
346
csum_copy_error:
347
_debug("csum error");
348
release_sock(&rx->sk);
349
if (continue_call)
350
rxrpc_put_call(continue_call);
351
rxrpc_kill_skb(skb);
352
skb_kill_datagram(&rx->sk, skb, flags);
353
rxrpc_put_call(call);
354
return -EAGAIN;
355
356
wait_interrupted:
357
ret = sock_intr_errno(timeo);
358
wait_error:
359
finish_wait(sk_sleep(&rx->sk), &wait);
360
if (continue_call)
361
rxrpc_put_call(continue_call);
362
if (copied)
363
copied = ret;
364
_leave(" = %d [waitfail %d]", copied, ret);
365
return copied;
366
367
}
368
369
/**
370
* rxrpc_kernel_data_delivered - Record delivery of data message
371
* @skb: Message holding data
372
*
373
* Record the delivery of a data message. This permits RxRPC to keep its
374
* tracking correct. The socket buffer will be deleted.
375
*/
376
void rxrpc_kernel_data_delivered(struct sk_buff *skb)
377
{
378
struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
379
struct rxrpc_call *call = sp->call;
380
381
ASSERTCMP(ntohl(sp->hdr.seq), >=, call->rx_data_recv);
382
ASSERTCMP(ntohl(sp->hdr.seq), <=, call->rx_data_recv + 1);
383
call->rx_data_recv = ntohl(sp->hdr.seq);
384
385
ASSERTCMP(ntohl(sp->hdr.seq), >, call->rx_data_eaten);
386
rxrpc_free_skb(skb);
387
}
388
389
EXPORT_SYMBOL(rxrpc_kernel_data_delivered);
390
391
/**
392
* rxrpc_kernel_is_data_last - Determine if data message is last one
393
* @skb: Message holding data
394
*
395
* Determine if data message is last one for the parent call.
396
*/
397
bool rxrpc_kernel_is_data_last(struct sk_buff *skb)
398
{
399
struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
400
401
ASSERTCMP(skb->mark, ==, RXRPC_SKB_MARK_DATA);
402
403
return sp->hdr.flags & RXRPC_LAST_PACKET;
404
}
405
406
EXPORT_SYMBOL(rxrpc_kernel_is_data_last);
407
408
/**
409
* rxrpc_kernel_get_abort_code - Get the abort code from an RxRPC abort message
410
* @skb: Message indicating an abort
411
*
412
* Get the abort code from an RxRPC abort message.
413
*/
414
u32 rxrpc_kernel_get_abort_code(struct sk_buff *skb)
415
{
416
struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
417
418
ASSERTCMP(skb->mark, ==, RXRPC_SKB_MARK_REMOTE_ABORT);
419
420
return sp->call->abort_code;
421
}
422
423
EXPORT_SYMBOL(rxrpc_kernel_get_abort_code);
424
425
/**
426
* rxrpc_kernel_get_error - Get the error number from an RxRPC error message
427
* @skb: Message indicating an error
428
*
429
* Get the error number from an RxRPC error message.
430
*/
431
int rxrpc_kernel_get_error_number(struct sk_buff *skb)
432
{
433
struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
434
435
return sp->error;
436
}
437
438
EXPORT_SYMBOL(rxrpc_kernel_get_error_number);
439
440