Merge branch 'v6v7' into devel
[linux/fpc-iii.git] / fs / afs / rxrpc.c
blobe45a323aebb4468711de395fd06e5f0f9d2833d0
1 /* Maintain an RxRPC server socket to do AFS communications through
3 * Copyright (C) 2007 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 License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
12 #include <linux/slab.h>
13 #include <net/sock.h>
14 #include <net/af_rxrpc.h>
15 #include <rxrpc/packet.h>
16 #include "internal.h"
17 #include "afs_cm.h"
19 static struct socket *afs_socket; /* my RxRPC socket */
20 static struct workqueue_struct *afs_async_calls;
21 static atomic_t afs_outstanding_calls;
22 static atomic_t afs_outstanding_skbs;
24 static void afs_wake_up_call_waiter(struct afs_call *);
25 static int afs_wait_for_call_to_complete(struct afs_call *);
26 static void afs_wake_up_async_call(struct afs_call *);
27 static int afs_dont_wait_for_call_to_complete(struct afs_call *);
28 static void afs_process_async_call(struct work_struct *);
29 static void afs_rx_interceptor(struct sock *, unsigned long, struct sk_buff *);
30 static int afs_deliver_cm_op_id(struct afs_call *, struct sk_buff *, bool);
32 /* synchronous call management */
33 const struct afs_wait_mode afs_sync_call = {
34 .rx_wakeup = afs_wake_up_call_waiter,
35 .wait = afs_wait_for_call_to_complete,
38 /* asynchronous call management */
39 const struct afs_wait_mode afs_async_call = {
40 .rx_wakeup = afs_wake_up_async_call,
41 .wait = afs_dont_wait_for_call_to_complete,
44 /* asynchronous incoming call management */
45 static const struct afs_wait_mode afs_async_incoming_call = {
46 .rx_wakeup = afs_wake_up_async_call,
49 /* asynchronous incoming call initial processing */
50 static const struct afs_call_type afs_RXCMxxxx = {
51 .name = "CB.xxxx",
52 .deliver = afs_deliver_cm_op_id,
53 .abort_to_error = afs_abort_to_error,
56 static void afs_collect_incoming_call(struct work_struct *);
58 static struct sk_buff_head afs_incoming_calls;
59 static DECLARE_WORK(afs_collect_incoming_call_work, afs_collect_incoming_call);
62 * open an RxRPC socket and bind it to be a server for callback notifications
63 * - the socket is left in blocking mode and non-blocking ops use MSG_DONTWAIT
65 int afs_open_socket(void)
67 struct sockaddr_rxrpc srx;
68 struct socket *socket;
69 int ret;
71 _enter("");
73 skb_queue_head_init(&afs_incoming_calls);
75 afs_async_calls = create_singlethread_workqueue("kafsd");
76 if (!afs_async_calls) {
77 _leave(" = -ENOMEM [wq]");
78 return -ENOMEM;
81 ret = sock_create_kern(AF_RXRPC, SOCK_DGRAM, PF_INET, &socket);
82 if (ret < 0) {
83 destroy_workqueue(afs_async_calls);
84 _leave(" = %d [socket]", ret);
85 return ret;
88 socket->sk->sk_allocation = GFP_NOFS;
90 /* bind the callback manager's address to make this a server socket */
91 srx.srx_family = AF_RXRPC;
92 srx.srx_service = CM_SERVICE;
93 srx.transport_type = SOCK_DGRAM;
94 srx.transport_len = sizeof(srx.transport.sin);
95 srx.transport.sin.sin_family = AF_INET;
96 srx.transport.sin.sin_port = htons(AFS_CM_PORT);
97 memset(&srx.transport.sin.sin_addr, 0,
98 sizeof(srx.transport.sin.sin_addr));
100 ret = kernel_bind(socket, (struct sockaddr *) &srx, sizeof(srx));
101 if (ret < 0) {
102 sock_release(socket);
103 destroy_workqueue(afs_async_calls);
104 _leave(" = %d [bind]", ret);
105 return ret;
108 rxrpc_kernel_intercept_rx_messages(socket, afs_rx_interceptor);
110 afs_socket = socket;
111 _leave(" = 0");
112 return 0;
116 * close the RxRPC socket AFS was using
118 void afs_close_socket(void)
120 _enter("");
122 sock_release(afs_socket);
124 _debug("dework");
125 destroy_workqueue(afs_async_calls);
127 ASSERTCMP(atomic_read(&afs_outstanding_skbs), ==, 0);
128 ASSERTCMP(atomic_read(&afs_outstanding_calls), ==, 0);
129 _leave("");
133 * note that the data in a socket buffer is now delivered and that the buffer
134 * should be freed
136 static void afs_data_delivered(struct sk_buff *skb)
138 if (!skb) {
139 _debug("DLVR NULL [%d]", atomic_read(&afs_outstanding_skbs));
140 dump_stack();
141 } else {
142 _debug("DLVR %p{%u} [%d]",
143 skb, skb->mark, atomic_read(&afs_outstanding_skbs));
144 if (atomic_dec_return(&afs_outstanding_skbs) == -1)
145 BUG();
146 rxrpc_kernel_data_delivered(skb);
151 * free a socket buffer
153 static void afs_free_skb(struct sk_buff *skb)
155 if (!skb) {
156 _debug("FREE NULL [%d]", atomic_read(&afs_outstanding_skbs));
157 dump_stack();
158 } else {
159 _debug("FREE %p{%u} [%d]",
160 skb, skb->mark, atomic_read(&afs_outstanding_skbs));
161 if (atomic_dec_return(&afs_outstanding_skbs) == -1)
162 BUG();
163 rxrpc_kernel_free_skb(skb);
168 * free a call
170 static void afs_free_call(struct afs_call *call)
172 _debug("DONE %p{%s} [%d]",
173 call, call->type->name, atomic_read(&afs_outstanding_calls));
174 if (atomic_dec_return(&afs_outstanding_calls) == -1)
175 BUG();
177 ASSERTCMP(call->rxcall, ==, NULL);
178 ASSERT(!work_pending(&call->async_work));
179 ASSERT(skb_queue_empty(&call->rx_queue));
180 ASSERT(call->type->name != NULL);
182 kfree(call->request);
183 kfree(call);
187 * allocate a call with flat request and reply buffers
189 struct afs_call *afs_alloc_flat_call(const struct afs_call_type *type,
190 size_t request_size, size_t reply_size)
192 struct afs_call *call;
194 call = kzalloc(sizeof(*call), GFP_NOFS);
195 if (!call)
196 goto nomem_call;
198 _debug("CALL %p{%s} [%d]",
199 call, type->name, atomic_read(&afs_outstanding_calls));
200 atomic_inc(&afs_outstanding_calls);
202 call->type = type;
203 call->request_size = request_size;
204 call->reply_max = reply_size;
206 if (request_size) {
207 call->request = kmalloc(request_size, GFP_NOFS);
208 if (!call->request)
209 goto nomem_free;
212 if (reply_size) {
213 call->buffer = kmalloc(reply_size, GFP_NOFS);
214 if (!call->buffer)
215 goto nomem_free;
218 init_waitqueue_head(&call->waitq);
219 skb_queue_head_init(&call->rx_queue);
220 return call;
222 nomem_free:
223 afs_free_call(call);
224 nomem_call:
225 return NULL;
229 * clean up a call with flat buffer
231 void afs_flat_call_destructor(struct afs_call *call)
233 _enter("");
235 kfree(call->request);
236 call->request = NULL;
237 kfree(call->buffer);
238 call->buffer = NULL;
242 * attach the data from a bunch of pages on an inode to a call
244 static int afs_send_pages(struct afs_call *call, struct msghdr *msg,
245 struct kvec *iov)
247 struct page *pages[8];
248 unsigned count, n, loop, offset, to;
249 pgoff_t first = call->first, last = call->last;
250 int ret;
252 _enter("");
254 offset = call->first_offset;
255 call->first_offset = 0;
257 do {
258 _debug("attach %lx-%lx", first, last);
260 count = last - first + 1;
261 if (count > ARRAY_SIZE(pages))
262 count = ARRAY_SIZE(pages);
263 n = find_get_pages_contig(call->mapping, first, count, pages);
264 ASSERTCMP(n, ==, count);
266 loop = 0;
267 do {
268 msg->msg_flags = 0;
269 to = PAGE_SIZE;
270 if (first + loop >= last)
271 to = call->last_to;
272 else
273 msg->msg_flags = MSG_MORE;
274 iov->iov_base = kmap(pages[loop]) + offset;
275 iov->iov_len = to - offset;
276 offset = 0;
278 _debug("- range %u-%u%s",
279 offset, to, msg->msg_flags ? " [more]" : "");
280 msg->msg_iov = (struct iovec *) iov;
281 msg->msg_iovlen = 1;
283 /* have to change the state *before* sending the last
284 * packet as RxRPC might give us the reply before it
285 * returns from sending the request */
286 if (first + loop >= last)
287 call->state = AFS_CALL_AWAIT_REPLY;
288 ret = rxrpc_kernel_send_data(call->rxcall, msg,
289 to - offset);
290 kunmap(pages[loop]);
291 if (ret < 0)
292 break;
293 } while (++loop < count);
294 first += count;
296 for (loop = 0; loop < count; loop++)
297 put_page(pages[loop]);
298 if (ret < 0)
299 break;
300 } while (first <= last);
302 _leave(" = %d", ret);
303 return ret;
307 * initiate a call
309 int afs_make_call(struct in_addr *addr, struct afs_call *call, gfp_t gfp,
310 const struct afs_wait_mode *wait_mode)
312 struct sockaddr_rxrpc srx;
313 struct rxrpc_call *rxcall;
314 struct msghdr msg;
315 struct kvec iov[1];
316 int ret;
318 _enter("%x,{%d},", addr->s_addr, ntohs(call->port));
320 ASSERT(call->type != NULL);
321 ASSERT(call->type->name != NULL);
323 _debug("____MAKE %p{%s,%x} [%d]____",
324 call, call->type->name, key_serial(call->key),
325 atomic_read(&afs_outstanding_calls));
327 call->wait_mode = wait_mode;
328 INIT_WORK(&call->async_work, afs_process_async_call);
330 memset(&srx, 0, sizeof(srx));
331 srx.srx_family = AF_RXRPC;
332 srx.srx_service = call->service_id;
333 srx.transport_type = SOCK_DGRAM;
334 srx.transport_len = sizeof(srx.transport.sin);
335 srx.transport.sin.sin_family = AF_INET;
336 srx.transport.sin.sin_port = call->port;
337 memcpy(&srx.transport.sin.sin_addr, addr, 4);
339 /* create a call */
340 rxcall = rxrpc_kernel_begin_call(afs_socket, &srx, call->key,
341 (unsigned long) call, gfp);
342 call->key = NULL;
343 if (IS_ERR(rxcall)) {
344 ret = PTR_ERR(rxcall);
345 goto error_kill_call;
348 call->rxcall = rxcall;
350 /* send the request */
351 iov[0].iov_base = call->request;
352 iov[0].iov_len = call->request_size;
354 msg.msg_name = NULL;
355 msg.msg_namelen = 0;
356 msg.msg_iov = (struct iovec *) iov;
357 msg.msg_iovlen = 1;
358 msg.msg_control = NULL;
359 msg.msg_controllen = 0;
360 msg.msg_flags = (call->send_pages ? MSG_MORE : 0);
362 /* have to change the state *before* sending the last packet as RxRPC
363 * might give us the reply before it returns from sending the
364 * request */
365 if (!call->send_pages)
366 call->state = AFS_CALL_AWAIT_REPLY;
367 ret = rxrpc_kernel_send_data(rxcall, &msg, call->request_size);
368 if (ret < 0)
369 goto error_do_abort;
371 if (call->send_pages) {
372 ret = afs_send_pages(call, &msg, iov);
373 if (ret < 0)
374 goto error_do_abort;
377 /* at this point, an async call may no longer exist as it may have
378 * already completed */
379 return wait_mode->wait(call);
381 error_do_abort:
382 rxrpc_kernel_abort_call(rxcall, RX_USER_ABORT);
383 rxrpc_kernel_end_call(rxcall);
384 call->rxcall = NULL;
385 error_kill_call:
386 call->type->destructor(call);
387 afs_free_call(call);
388 _leave(" = %d", ret);
389 return ret;
393 * handles intercepted messages that were arriving in the socket's Rx queue
394 * - called with the socket receive queue lock held to ensure message ordering
395 * - called with softirqs disabled
397 static void afs_rx_interceptor(struct sock *sk, unsigned long user_call_ID,
398 struct sk_buff *skb)
400 struct afs_call *call = (struct afs_call *) user_call_ID;
402 _enter("%p,,%u", call, skb->mark);
404 _debug("ICPT %p{%u} [%d]",
405 skb, skb->mark, atomic_read(&afs_outstanding_skbs));
407 ASSERTCMP(sk, ==, afs_socket->sk);
408 atomic_inc(&afs_outstanding_skbs);
410 if (!call) {
411 /* its an incoming call for our callback service */
412 skb_queue_tail(&afs_incoming_calls, skb);
413 queue_work(afs_wq, &afs_collect_incoming_call_work);
414 } else {
415 /* route the messages directly to the appropriate call */
416 skb_queue_tail(&call->rx_queue, skb);
417 call->wait_mode->rx_wakeup(call);
420 _leave("");
424 * deliver messages to a call
426 static void afs_deliver_to_call(struct afs_call *call)
428 struct sk_buff *skb;
429 bool last;
430 u32 abort_code;
431 int ret;
433 _enter("");
435 while ((call->state == AFS_CALL_AWAIT_REPLY ||
436 call->state == AFS_CALL_AWAIT_OP_ID ||
437 call->state == AFS_CALL_AWAIT_REQUEST ||
438 call->state == AFS_CALL_AWAIT_ACK) &&
439 (skb = skb_dequeue(&call->rx_queue))) {
440 switch (skb->mark) {
441 case RXRPC_SKB_MARK_DATA:
442 _debug("Rcv DATA");
443 last = rxrpc_kernel_is_data_last(skb);
444 ret = call->type->deliver(call, skb, last);
445 switch (ret) {
446 case 0:
447 if (last &&
448 call->state == AFS_CALL_AWAIT_REPLY)
449 call->state = AFS_CALL_COMPLETE;
450 break;
451 case -ENOTCONN:
452 abort_code = RX_CALL_DEAD;
453 goto do_abort;
454 case -ENOTSUPP:
455 abort_code = RX_INVALID_OPERATION;
456 goto do_abort;
457 default:
458 abort_code = RXGEN_CC_UNMARSHAL;
459 if (call->state != AFS_CALL_AWAIT_REPLY)
460 abort_code = RXGEN_SS_UNMARSHAL;
461 do_abort:
462 rxrpc_kernel_abort_call(call->rxcall,
463 abort_code);
464 call->error = ret;
465 call->state = AFS_CALL_ERROR;
466 break;
468 afs_data_delivered(skb);
469 skb = NULL;
470 continue;
471 case RXRPC_SKB_MARK_FINAL_ACK:
472 _debug("Rcv ACK");
473 call->state = AFS_CALL_COMPLETE;
474 break;
475 case RXRPC_SKB_MARK_BUSY:
476 _debug("Rcv BUSY");
477 call->error = -EBUSY;
478 call->state = AFS_CALL_BUSY;
479 break;
480 case RXRPC_SKB_MARK_REMOTE_ABORT:
481 abort_code = rxrpc_kernel_get_abort_code(skb);
482 call->error = call->type->abort_to_error(abort_code);
483 call->state = AFS_CALL_ABORTED;
484 _debug("Rcv ABORT %u -> %d", abort_code, call->error);
485 break;
486 case RXRPC_SKB_MARK_NET_ERROR:
487 call->error = -rxrpc_kernel_get_error_number(skb);
488 call->state = AFS_CALL_ERROR;
489 _debug("Rcv NET ERROR %d", call->error);
490 break;
491 case RXRPC_SKB_MARK_LOCAL_ERROR:
492 call->error = -rxrpc_kernel_get_error_number(skb);
493 call->state = AFS_CALL_ERROR;
494 _debug("Rcv LOCAL ERROR %d", call->error);
495 break;
496 default:
497 BUG();
498 break;
501 afs_free_skb(skb);
504 /* make sure the queue is empty if the call is done with (we might have
505 * aborted the call early because of an unmarshalling error) */
506 if (call->state >= AFS_CALL_COMPLETE) {
507 while ((skb = skb_dequeue(&call->rx_queue)))
508 afs_free_skb(skb);
509 if (call->incoming) {
510 rxrpc_kernel_end_call(call->rxcall);
511 call->rxcall = NULL;
512 call->type->destructor(call);
513 afs_free_call(call);
517 _leave("");
521 * wait synchronously for a call to complete
523 static int afs_wait_for_call_to_complete(struct afs_call *call)
525 struct sk_buff *skb;
526 int ret;
528 DECLARE_WAITQUEUE(myself, current);
530 _enter("");
532 add_wait_queue(&call->waitq, &myself);
533 for (;;) {
534 set_current_state(TASK_INTERRUPTIBLE);
536 /* deliver any messages that are in the queue */
537 if (!skb_queue_empty(&call->rx_queue)) {
538 __set_current_state(TASK_RUNNING);
539 afs_deliver_to_call(call);
540 continue;
543 ret = call->error;
544 if (call->state >= AFS_CALL_COMPLETE)
545 break;
546 ret = -EINTR;
547 if (signal_pending(current))
548 break;
549 schedule();
552 remove_wait_queue(&call->waitq, &myself);
553 __set_current_state(TASK_RUNNING);
555 /* kill the call */
556 if (call->state < AFS_CALL_COMPLETE) {
557 _debug("call incomplete");
558 rxrpc_kernel_abort_call(call->rxcall, RX_CALL_DEAD);
559 while ((skb = skb_dequeue(&call->rx_queue)))
560 afs_free_skb(skb);
563 _debug("call complete");
564 rxrpc_kernel_end_call(call->rxcall);
565 call->rxcall = NULL;
566 call->type->destructor(call);
567 afs_free_call(call);
568 _leave(" = %d", ret);
569 return ret;
573 * wake up a waiting call
575 static void afs_wake_up_call_waiter(struct afs_call *call)
577 wake_up(&call->waitq);
581 * wake up an asynchronous call
583 static void afs_wake_up_async_call(struct afs_call *call)
585 _enter("");
586 queue_work(afs_async_calls, &call->async_work);
590 * put a call into asynchronous mode
591 * - mustn't touch the call descriptor as the call my have completed by the
592 * time we get here
594 static int afs_dont_wait_for_call_to_complete(struct afs_call *call)
596 _enter("");
597 return -EINPROGRESS;
601 * delete an asynchronous call
603 static void afs_delete_async_call(struct work_struct *work)
605 struct afs_call *call =
606 container_of(work, struct afs_call, async_work);
608 _enter("");
610 afs_free_call(call);
612 _leave("");
616 * perform processing on an asynchronous call
617 * - on a multiple-thread workqueue this work item may try to run on several
618 * CPUs at the same time
620 static void afs_process_async_call(struct work_struct *work)
622 struct afs_call *call =
623 container_of(work, struct afs_call, async_work);
625 _enter("");
627 if (!skb_queue_empty(&call->rx_queue))
628 afs_deliver_to_call(call);
630 if (call->state >= AFS_CALL_COMPLETE && call->wait_mode) {
631 if (call->wait_mode->async_complete)
632 call->wait_mode->async_complete(call->reply,
633 call->error);
634 call->reply = NULL;
636 /* kill the call */
637 rxrpc_kernel_end_call(call->rxcall);
638 call->rxcall = NULL;
639 if (call->type->destructor)
640 call->type->destructor(call);
642 /* we can't just delete the call because the work item may be
643 * queued */
644 PREPARE_WORK(&call->async_work, afs_delete_async_call);
645 queue_work(afs_async_calls, &call->async_work);
648 _leave("");
652 * empty a socket buffer into a flat reply buffer
654 void afs_transfer_reply(struct afs_call *call, struct sk_buff *skb)
656 size_t len = skb->len;
658 if (skb_copy_bits(skb, 0, call->buffer + call->reply_size, len) < 0)
659 BUG();
660 call->reply_size += len;
664 * accept the backlog of incoming calls
666 static void afs_collect_incoming_call(struct work_struct *work)
668 struct rxrpc_call *rxcall;
669 struct afs_call *call = NULL;
670 struct sk_buff *skb;
672 while ((skb = skb_dequeue(&afs_incoming_calls))) {
673 _debug("new call");
675 /* don't need the notification */
676 afs_free_skb(skb);
678 if (!call) {
679 call = kzalloc(sizeof(struct afs_call), GFP_KERNEL);
680 if (!call) {
681 rxrpc_kernel_reject_call(afs_socket);
682 return;
685 INIT_WORK(&call->async_work, afs_process_async_call);
686 call->wait_mode = &afs_async_incoming_call;
687 call->type = &afs_RXCMxxxx;
688 init_waitqueue_head(&call->waitq);
689 skb_queue_head_init(&call->rx_queue);
690 call->state = AFS_CALL_AWAIT_OP_ID;
692 _debug("CALL %p{%s} [%d]",
693 call, call->type->name,
694 atomic_read(&afs_outstanding_calls));
695 atomic_inc(&afs_outstanding_calls);
698 rxcall = rxrpc_kernel_accept_call(afs_socket,
699 (unsigned long) call);
700 if (!IS_ERR(rxcall)) {
701 call->rxcall = rxcall;
702 call = NULL;
706 if (call)
707 afs_free_call(call);
711 * grab the operation ID from an incoming cache manager call
713 static int afs_deliver_cm_op_id(struct afs_call *call, struct sk_buff *skb,
714 bool last)
716 size_t len = skb->len;
717 void *oibuf = (void *) &call->operation_ID;
719 _enter("{%u},{%zu},%d", call->offset, len, last);
721 ASSERTCMP(call->offset, <, 4);
723 /* the operation ID forms the first four bytes of the request data */
724 len = min_t(size_t, len, 4 - call->offset);
725 if (skb_copy_bits(skb, 0, oibuf + call->offset, len) < 0)
726 BUG();
727 if (!pskb_pull(skb, len))
728 BUG();
729 call->offset += len;
731 if (call->offset < 4) {
732 if (last) {
733 _leave(" = -EBADMSG [op ID short]");
734 return -EBADMSG;
736 _leave(" = 0 [incomplete]");
737 return 0;
740 call->state = AFS_CALL_AWAIT_REQUEST;
742 /* ask the cache manager to route the call (it'll change the call type
743 * if successful) */
744 if (!afs_cm_incoming_call(call))
745 return -ENOTSUPP;
747 /* pass responsibility for the remainer of this message off to the
748 * cache manager op */
749 return call->type->deliver(call, skb, last);
753 * send an empty reply
755 void afs_send_empty_reply(struct afs_call *call)
757 struct msghdr msg;
758 struct iovec iov[1];
760 _enter("");
762 iov[0].iov_base = NULL;
763 iov[0].iov_len = 0;
764 msg.msg_name = NULL;
765 msg.msg_namelen = 0;
766 msg.msg_iov = iov;
767 msg.msg_iovlen = 0;
768 msg.msg_control = NULL;
769 msg.msg_controllen = 0;
770 msg.msg_flags = 0;
772 call->state = AFS_CALL_AWAIT_ACK;
773 switch (rxrpc_kernel_send_data(call->rxcall, &msg, 0)) {
774 case 0:
775 _leave(" [replied]");
776 return;
778 case -ENOMEM:
779 _debug("oom");
780 rxrpc_kernel_abort_call(call->rxcall, RX_USER_ABORT);
781 default:
782 rxrpc_kernel_end_call(call->rxcall);
783 call->rxcall = NULL;
784 call->type->destructor(call);
785 afs_free_call(call);
786 _leave(" [error]");
787 return;
792 * send a simple reply
794 void afs_send_simple_reply(struct afs_call *call, const void *buf, size_t len)
796 struct msghdr msg;
797 struct iovec iov[1];
798 int n;
800 _enter("");
802 iov[0].iov_base = (void *) buf;
803 iov[0].iov_len = len;
804 msg.msg_name = NULL;
805 msg.msg_namelen = 0;
806 msg.msg_iov = iov;
807 msg.msg_iovlen = 1;
808 msg.msg_control = NULL;
809 msg.msg_controllen = 0;
810 msg.msg_flags = 0;
812 call->state = AFS_CALL_AWAIT_ACK;
813 n = rxrpc_kernel_send_data(call->rxcall, &msg, len);
814 if (n >= 0) {
815 _leave(" [replied]");
816 return;
818 if (n == -ENOMEM) {
819 _debug("oom");
820 rxrpc_kernel_abort_call(call->rxcall, RX_USER_ABORT);
822 rxrpc_kernel_end_call(call->rxcall);
823 call->rxcall = NULL;
824 call->type->destructor(call);
825 afs_free_call(call);
826 _leave(" [error]");
830 * extract a piece of data from the received data socket buffers
832 int afs_extract_data(struct afs_call *call, struct sk_buff *skb,
833 bool last, void *buf, size_t count)
835 size_t len = skb->len;
837 _enter("{%u},{%zu},%d,,%zu", call->offset, len, last, count);
839 ASSERTCMP(call->offset, <, count);
841 len = min_t(size_t, len, count - call->offset);
842 if (skb_copy_bits(skb, 0, buf + call->offset, len) < 0 ||
843 !pskb_pull(skb, len))
844 BUG();
845 call->offset += len;
847 if (call->offset < count) {
848 if (last) {
849 _leave(" = -EBADMSG [%d < %zu]", call->offset, count);
850 return -EBADMSG;
852 _leave(" = -EAGAIN");
853 return -EAGAIN;
855 return 0;