xref: /openbmc/linux/fs/afs/rxrpc.c (revision 06d5d6b7f9948a89543e1160ef852d57892c750d)
1 /* Maintain an RxRPC server socket to do AFS communications through
2  *
3  * Copyright (C) 2007 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 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/slab.h>
13 #include <linux/sched/signal.h>
14 
15 #include <net/sock.h>
16 #include <net/af_rxrpc.h>
17 #include "internal.h"
18 #include "afs_cm.h"
19 #include "protocol_yfs.h"
20 
21 struct workqueue_struct *afs_async_calls;
22 
23 static void afs_wake_up_call_waiter(struct sock *, struct rxrpc_call *, unsigned long);
24 static long afs_wait_for_call_to_complete(struct afs_call *, struct afs_addr_cursor *);
25 static void afs_wake_up_async_call(struct sock *, struct rxrpc_call *, unsigned long);
26 static void afs_delete_async_call(struct work_struct *);
27 static void afs_process_async_call(struct work_struct *);
28 static void afs_rx_new_call(struct sock *, struct rxrpc_call *, unsigned long);
29 static void afs_rx_discard_new_call(struct rxrpc_call *, unsigned long);
30 static int afs_deliver_cm_op_id(struct afs_call *);
31 
32 /* asynchronous incoming call initial processing */
33 static const struct afs_call_type afs_RXCMxxxx = {
34 	.name		= "CB.xxxx",
35 	.deliver	= afs_deliver_cm_op_id,
36 };
37 
38 /*
39  * open an RxRPC socket and bind it to be a server for callback notifications
40  * - the socket is left in blocking mode and non-blocking ops use MSG_DONTWAIT
41  */
42 int afs_open_socket(struct afs_net *net)
43 {
44 	struct sockaddr_rxrpc srx;
45 	struct socket *socket;
46 	unsigned int min_level;
47 	int ret;
48 
49 	_enter("");
50 
51 	ret = sock_create_kern(net->net, AF_RXRPC, SOCK_DGRAM, PF_INET6, &socket);
52 	if (ret < 0)
53 		goto error_1;
54 
55 	socket->sk->sk_allocation = GFP_NOFS;
56 
57 	/* bind the callback manager's address to make this a server socket */
58 	memset(&srx, 0, sizeof(srx));
59 	srx.srx_family			= AF_RXRPC;
60 	srx.srx_service			= CM_SERVICE;
61 	srx.transport_type		= SOCK_DGRAM;
62 	srx.transport_len		= sizeof(srx.transport.sin6);
63 	srx.transport.sin6.sin6_family	= AF_INET6;
64 	srx.transport.sin6.sin6_port	= htons(AFS_CM_PORT);
65 
66 	min_level = RXRPC_SECURITY_ENCRYPT;
67 	ret = kernel_setsockopt(socket, SOL_RXRPC, RXRPC_MIN_SECURITY_LEVEL,
68 				(void *)&min_level, sizeof(min_level));
69 	if (ret < 0)
70 		goto error_2;
71 
72 	ret = kernel_bind(socket, (struct sockaddr *) &srx, sizeof(srx));
73 	if (ret == -EADDRINUSE) {
74 		srx.transport.sin6.sin6_port = 0;
75 		ret = kernel_bind(socket, (struct sockaddr *) &srx, sizeof(srx));
76 	}
77 	if (ret < 0)
78 		goto error_2;
79 
80 	srx.srx_service = YFS_CM_SERVICE;
81 	ret = kernel_bind(socket, (struct sockaddr *) &srx, sizeof(srx));
82 	if (ret < 0)
83 		goto error_2;
84 
85 	/* Ideally, we'd turn on service upgrade here, but we can't because
86 	 * OpenAFS is buggy and leaks the userStatus field from packet to
87 	 * packet and between FS packets and CB packets - so if we try to do an
88 	 * upgrade on an FS packet, OpenAFS will leak that into the CB packet
89 	 * it sends back to us.
90 	 */
91 
92 	rxrpc_kernel_new_call_notification(socket, afs_rx_new_call,
93 					   afs_rx_discard_new_call);
94 
95 	ret = kernel_listen(socket, INT_MAX);
96 	if (ret < 0)
97 		goto error_2;
98 
99 	net->socket = socket;
100 	afs_charge_preallocation(&net->charge_preallocation_work);
101 	_leave(" = 0");
102 	return 0;
103 
104 error_2:
105 	sock_release(socket);
106 error_1:
107 	_leave(" = %d", ret);
108 	return ret;
109 }
110 
111 /*
112  * close the RxRPC socket AFS was using
113  */
114 void afs_close_socket(struct afs_net *net)
115 {
116 	_enter("");
117 
118 	kernel_listen(net->socket, 0);
119 	flush_workqueue(afs_async_calls);
120 
121 	if (net->spare_incoming_call) {
122 		afs_put_call(net->spare_incoming_call);
123 		net->spare_incoming_call = NULL;
124 	}
125 
126 	_debug("outstanding %u", atomic_read(&net->nr_outstanding_calls));
127 	wait_var_event(&net->nr_outstanding_calls,
128 		       !atomic_read(&net->nr_outstanding_calls));
129 	_debug("no outstanding calls");
130 
131 	kernel_sock_shutdown(net->socket, SHUT_RDWR);
132 	flush_workqueue(afs_async_calls);
133 	sock_release(net->socket);
134 
135 	_debug("dework");
136 	_leave("");
137 }
138 
139 /*
140  * Allocate a call.
141  */
142 static struct afs_call *afs_alloc_call(struct afs_net *net,
143 				       const struct afs_call_type *type,
144 				       gfp_t gfp)
145 {
146 	struct afs_call *call;
147 	int o;
148 
149 	call = kzalloc(sizeof(*call), gfp);
150 	if (!call)
151 		return NULL;
152 
153 	call->type = type;
154 	call->net = net;
155 	call->debug_id = atomic_inc_return(&rxrpc_debug_id);
156 	atomic_set(&call->usage, 1);
157 	INIT_WORK(&call->async_work, afs_process_async_call);
158 	init_waitqueue_head(&call->waitq);
159 	spin_lock_init(&call->state_lock);
160 	call->_iter = &call->iter;
161 
162 	o = atomic_inc_return(&net->nr_outstanding_calls);
163 	trace_afs_call(call, afs_call_trace_alloc, 1, o,
164 		       __builtin_return_address(0));
165 	return call;
166 }
167 
168 /*
169  * Dispose of a reference on a call.
170  */
171 void afs_put_call(struct afs_call *call)
172 {
173 	struct afs_net *net = call->net;
174 	int n = atomic_dec_return(&call->usage);
175 	int o = atomic_read(&net->nr_outstanding_calls);
176 
177 	trace_afs_call(call, afs_call_trace_put, n + 1, o,
178 		       __builtin_return_address(0));
179 
180 	ASSERTCMP(n, >=, 0);
181 	if (n == 0) {
182 		ASSERT(!work_pending(&call->async_work));
183 		ASSERT(call->type->name != NULL);
184 
185 		if (call->rxcall) {
186 			rxrpc_kernel_end_call(net->socket, call->rxcall);
187 			call->rxcall = NULL;
188 		}
189 		if (call->type->destructor)
190 			call->type->destructor(call);
191 
192 		afs_put_server(call->net, call->cm_server);
193 		afs_put_cb_interest(call->net, call->cbi);
194 		afs_put_addrlist(call->alist);
195 		kfree(call->request);
196 
197 		trace_afs_call(call, afs_call_trace_free, 0, o,
198 			       __builtin_return_address(0));
199 		kfree(call);
200 
201 		o = atomic_dec_return(&net->nr_outstanding_calls);
202 		if (o == 0)
203 			wake_up_var(&net->nr_outstanding_calls);
204 	}
205 }
206 
207 static struct afs_call *afs_get_call(struct afs_call *call,
208 				     enum afs_call_trace why)
209 {
210 	int u = atomic_inc_return(&call->usage);
211 
212 	trace_afs_call(call, why, u,
213 		       atomic_read(&call->net->nr_outstanding_calls),
214 		       __builtin_return_address(0));
215 	return call;
216 }
217 
218 /*
219  * Queue the call for actual work.
220  */
221 static void afs_queue_call_work(struct afs_call *call)
222 {
223 	if (call->type->work) {
224 		INIT_WORK(&call->work, call->type->work);
225 
226 		afs_get_call(call, afs_call_trace_work);
227 		if (!queue_work(afs_wq, &call->work))
228 			afs_put_call(call);
229 	}
230 }
231 
232 /*
233  * allocate a call with flat request and reply buffers
234  */
235 struct afs_call *afs_alloc_flat_call(struct afs_net *net,
236 				     const struct afs_call_type *type,
237 				     size_t request_size, size_t reply_max)
238 {
239 	struct afs_call *call;
240 
241 	call = afs_alloc_call(net, type, GFP_NOFS);
242 	if (!call)
243 		goto nomem_call;
244 
245 	if (request_size) {
246 		call->request_size = request_size;
247 		call->request = kmalloc(request_size, GFP_NOFS);
248 		if (!call->request)
249 			goto nomem_free;
250 	}
251 
252 	if (reply_max) {
253 		call->reply_max = reply_max;
254 		call->buffer = kmalloc(reply_max, GFP_NOFS);
255 		if (!call->buffer)
256 			goto nomem_free;
257 	}
258 
259 	afs_extract_to_buf(call, call->reply_max);
260 	call->operation_ID = type->op;
261 	init_waitqueue_head(&call->waitq);
262 	return call;
263 
264 nomem_free:
265 	afs_put_call(call);
266 nomem_call:
267 	return NULL;
268 }
269 
270 /*
271  * clean up a call with flat buffer
272  */
273 void afs_flat_call_destructor(struct afs_call *call)
274 {
275 	_enter("");
276 
277 	kfree(call->request);
278 	call->request = NULL;
279 	kfree(call->buffer);
280 	call->buffer = NULL;
281 }
282 
283 #define AFS_BVEC_MAX 8
284 
285 /*
286  * Load the given bvec with the next few pages.
287  */
288 static void afs_load_bvec(struct afs_call *call, struct msghdr *msg,
289 			  struct bio_vec *bv, pgoff_t first, pgoff_t last,
290 			  unsigned offset)
291 {
292 	struct page *pages[AFS_BVEC_MAX];
293 	unsigned int nr, n, i, to, bytes = 0;
294 
295 	nr = min_t(pgoff_t, last - first + 1, AFS_BVEC_MAX);
296 	n = find_get_pages_contig(call->mapping, first, nr, pages);
297 	ASSERTCMP(n, ==, nr);
298 
299 	msg->msg_flags |= MSG_MORE;
300 	for (i = 0; i < nr; i++) {
301 		to = PAGE_SIZE;
302 		if (first + i >= last) {
303 			to = call->last_to;
304 			msg->msg_flags &= ~MSG_MORE;
305 		}
306 		bv[i].bv_page = pages[i];
307 		bv[i].bv_len = to - offset;
308 		bv[i].bv_offset = offset;
309 		bytes += to - offset;
310 		offset = 0;
311 	}
312 
313 	iov_iter_bvec(&msg->msg_iter, WRITE, bv, nr, bytes);
314 }
315 
316 /*
317  * Advance the AFS call state when the RxRPC call ends the transmit phase.
318  */
319 static void afs_notify_end_request_tx(struct sock *sock,
320 				      struct rxrpc_call *rxcall,
321 				      unsigned long call_user_ID)
322 {
323 	struct afs_call *call = (struct afs_call *)call_user_ID;
324 
325 	afs_set_call_state(call, AFS_CALL_CL_REQUESTING, AFS_CALL_CL_AWAIT_REPLY);
326 }
327 
328 /*
329  * attach the data from a bunch of pages on an inode to a call
330  */
331 static int afs_send_pages(struct afs_call *call, struct msghdr *msg)
332 {
333 	struct bio_vec bv[AFS_BVEC_MAX];
334 	unsigned int bytes, nr, loop, offset;
335 	pgoff_t first = call->first, last = call->last;
336 	int ret;
337 
338 	offset = call->first_offset;
339 	call->first_offset = 0;
340 
341 	do {
342 		afs_load_bvec(call, msg, bv, first, last, offset);
343 		trace_afs_send_pages(call, msg, first, last, offset);
344 
345 		offset = 0;
346 		bytes = msg->msg_iter.count;
347 		nr = msg->msg_iter.nr_segs;
348 
349 		ret = rxrpc_kernel_send_data(call->net->socket, call->rxcall, msg,
350 					     bytes, afs_notify_end_request_tx);
351 		for (loop = 0; loop < nr; loop++)
352 			put_page(bv[loop].bv_page);
353 		if (ret < 0)
354 			break;
355 
356 		first += nr;
357 	} while (first <= last);
358 
359 	trace_afs_sent_pages(call, call->first, last, first, ret);
360 	return ret;
361 }
362 
363 /*
364  * initiate a call
365  */
366 long afs_make_call(struct afs_addr_cursor *ac, struct afs_call *call,
367 		   gfp_t gfp, bool async)
368 {
369 	struct sockaddr_rxrpc *srx = &ac->alist->addrs[ac->index];
370 	struct rxrpc_call *rxcall;
371 	struct msghdr msg;
372 	struct kvec iov[1];
373 	s64 tx_total_len;
374 	int ret;
375 
376 	_enter(",{%pISp},", &srx->transport);
377 
378 	ASSERT(call->type != NULL);
379 	ASSERT(call->type->name != NULL);
380 
381 	_debug("____MAKE %p{%s,%x} [%d]____",
382 	       call, call->type->name, key_serial(call->key),
383 	       atomic_read(&call->net->nr_outstanding_calls));
384 
385 	call->async = async;
386 	call->addr_ix = ac->index;
387 	call->alist = afs_get_addrlist(ac->alist);
388 
389 	/* Work out the length we're going to transmit.  This is awkward for
390 	 * calls such as FS.StoreData where there's an extra injection of data
391 	 * after the initial fixed part.
392 	 */
393 	tx_total_len = call->request_size;
394 	if (call->send_pages) {
395 		if (call->last == call->first) {
396 			tx_total_len += call->last_to - call->first_offset;
397 		} else {
398 			/* It looks mathematically like you should be able to
399 			 * combine the following lines with the ones above, but
400 			 * unsigned arithmetic is fun when it wraps...
401 			 */
402 			tx_total_len += PAGE_SIZE - call->first_offset;
403 			tx_total_len += call->last_to;
404 			tx_total_len += (call->last - call->first - 1) * PAGE_SIZE;
405 		}
406 	}
407 
408 	/* If the call is going to be asynchronous, we need an extra ref for
409 	 * the call to hold itself so the caller need not hang on to its ref.
410 	 */
411 	if (call->async)
412 		afs_get_call(call, afs_call_trace_get);
413 
414 	/* create a call */
415 	rxcall = rxrpc_kernel_begin_call(call->net->socket, srx, call->key,
416 					 (unsigned long)call,
417 					 tx_total_len, gfp,
418 					 (async ?
419 					  afs_wake_up_async_call :
420 					  afs_wake_up_call_waiter),
421 					 call->upgrade,
422 					 call->debug_id);
423 	if (IS_ERR(rxcall)) {
424 		ret = PTR_ERR(rxcall);
425 		call->error = ret;
426 		goto error_kill_call;
427 	}
428 
429 	call->rxcall = rxcall;
430 
431 	/* send the request */
432 	iov[0].iov_base	= call->request;
433 	iov[0].iov_len	= call->request_size;
434 
435 	msg.msg_name		= NULL;
436 	msg.msg_namelen		= 0;
437 	iov_iter_kvec(&msg.msg_iter, WRITE, iov, 1, call->request_size);
438 	msg.msg_control		= NULL;
439 	msg.msg_controllen	= 0;
440 	msg.msg_flags		= MSG_WAITALL | (call->send_pages ? MSG_MORE : 0);
441 
442 	ret = rxrpc_kernel_send_data(call->net->socket, rxcall,
443 				     &msg, call->request_size,
444 				     afs_notify_end_request_tx);
445 	if (ret < 0)
446 		goto error_do_abort;
447 
448 	if (call->send_pages) {
449 		ret = afs_send_pages(call, &msg);
450 		if (ret < 0)
451 			goto error_do_abort;
452 	}
453 
454 	/* Note that at this point, we may have received the reply or an abort
455 	 * - and an asynchronous call may already have completed.
456 	 */
457 	if (call->async) {
458 		afs_put_call(call);
459 		return -EINPROGRESS;
460 	}
461 
462 	return afs_wait_for_call_to_complete(call, ac);
463 
464 error_do_abort:
465 	if (ret != -ECONNABORTED) {
466 		rxrpc_kernel_abort_call(call->net->socket, rxcall,
467 					RX_USER_ABORT, ret, "KSD");
468 	} else {
469 		iov_iter_kvec(&msg.msg_iter, READ, NULL, 0, 0);
470 		rxrpc_kernel_recv_data(call->net->socket, rxcall,
471 				       &msg.msg_iter, false,
472 				       &call->abort_code, &call->service_id);
473 		ac->abort_code = call->abort_code;
474 		ac->responded = true;
475 	}
476 	call->error = ret;
477 	trace_afs_call_done(call);
478 error_kill_call:
479 	if (call->type->done)
480 		call->type->done(call);
481 
482 	/* We need to dispose of the extra ref we grabbed for an async call.
483 	 * The call, however, might be queued on afs_async_calls and we need to
484 	 * make sure we don't get any more notifications that might requeue it.
485 	 */
486 	if (call->rxcall) {
487 		rxrpc_kernel_end_call(call->net->socket, call->rxcall);
488 		call->rxcall = NULL;
489 	}
490 	if (call->async) {
491 		if (cancel_work_sync(&call->async_work))
492 			afs_put_call(call);
493 		afs_put_call(call);
494 	}
495 
496 	ac->error = ret;
497 	call->state = AFS_CALL_COMPLETE;
498 	afs_put_call(call);
499 	_leave(" = %d", ret);
500 	return ret;
501 }
502 
503 /*
504  * deliver messages to a call
505  */
506 static void afs_deliver_to_call(struct afs_call *call)
507 {
508 	enum afs_call_state state;
509 	u32 abort_code, remote_abort = 0;
510 	int ret;
511 
512 	_enter("%s", call->type->name);
513 
514 	while (state = READ_ONCE(call->state),
515 	       state == AFS_CALL_CL_AWAIT_REPLY ||
516 	       state == AFS_CALL_SV_AWAIT_OP_ID ||
517 	       state == AFS_CALL_SV_AWAIT_REQUEST ||
518 	       state == AFS_CALL_SV_AWAIT_ACK
519 	       ) {
520 		if (state == AFS_CALL_SV_AWAIT_ACK) {
521 			iov_iter_kvec(&call->iter, READ, NULL, 0, 0);
522 			ret = rxrpc_kernel_recv_data(call->net->socket,
523 						     call->rxcall, &call->iter,
524 						     false, &remote_abort,
525 						     &call->service_id);
526 			trace_afs_receive_data(call, &call->iter, false, ret);
527 
528 			if (ret == -EINPROGRESS || ret == -EAGAIN)
529 				return;
530 			if (ret < 0 || ret == 1) {
531 				if (ret == 1)
532 					ret = 0;
533 				goto call_complete;
534 			}
535 			return;
536 		}
537 
538 		if (call->want_reply_time &&
539 		    rxrpc_kernel_get_reply_time(call->net->socket,
540 						call->rxcall,
541 						&call->reply_time))
542 			call->want_reply_time = false;
543 
544 		ret = call->type->deliver(call);
545 		state = READ_ONCE(call->state);
546 		switch (ret) {
547 		case 0:
548 			afs_queue_call_work(call);
549 			if (state == AFS_CALL_CL_PROC_REPLY) {
550 				if (call->cbi)
551 					set_bit(AFS_SERVER_FL_MAY_HAVE_CB,
552 						&call->cbi->server->flags);
553 				goto call_complete;
554 			}
555 			ASSERTCMP(state, >, AFS_CALL_CL_PROC_REPLY);
556 			goto done;
557 		case -EINPROGRESS:
558 		case -EAGAIN:
559 			goto out;
560 		case -ECONNABORTED:
561 			ASSERTCMP(state, ==, AFS_CALL_COMPLETE);
562 			goto done;
563 		case -ENOTSUPP:
564 			abort_code = RXGEN_OPCODE;
565 			rxrpc_kernel_abort_call(call->net->socket, call->rxcall,
566 						abort_code, ret, "KIV");
567 			goto local_abort;
568 		case -EIO:
569 			pr_err("kAFS: Call %u in bad state %u\n",
570 			       call->debug_id, state);
571 			/* Fall through */
572 		case -ENODATA:
573 		case -EBADMSG:
574 		case -EMSGSIZE:
575 			abort_code = RXGEN_CC_UNMARSHAL;
576 			if (state != AFS_CALL_CL_AWAIT_REPLY)
577 				abort_code = RXGEN_SS_UNMARSHAL;
578 			rxrpc_kernel_abort_call(call->net->socket, call->rxcall,
579 						abort_code, ret, "KUM");
580 			goto local_abort;
581 		default:
582 			abort_code = RX_USER_ABORT;
583 			rxrpc_kernel_abort_call(call->net->socket, call->rxcall,
584 						abort_code, ret, "KER");
585 			goto local_abort;
586 		}
587 	}
588 
589 done:
590 	if (call->type->done)
591 		call->type->done(call);
592 	if (state == AFS_CALL_COMPLETE && call->incoming)
593 		afs_put_call(call);
594 out:
595 	_leave("");
596 	return;
597 
598 local_abort:
599 	abort_code = 0;
600 call_complete:
601 	afs_set_call_complete(call, ret, remote_abort);
602 	state = AFS_CALL_COMPLETE;
603 	goto done;
604 }
605 
606 /*
607  * wait synchronously for a call to complete
608  */
609 static long afs_wait_for_call_to_complete(struct afs_call *call,
610 					  struct afs_addr_cursor *ac)
611 {
612 	signed long rtt2, timeout;
613 	long ret;
614 	bool stalled = false;
615 	u64 rtt;
616 	u32 life, last_life;
617 	bool rxrpc_complete = false;
618 
619 	DECLARE_WAITQUEUE(myself, current);
620 
621 	_enter("");
622 
623 	rtt = rxrpc_kernel_get_rtt(call->net->socket, call->rxcall);
624 	rtt2 = nsecs_to_jiffies64(rtt) * 2;
625 	if (rtt2 < 2)
626 		rtt2 = 2;
627 
628 	timeout = rtt2;
629 	rxrpc_kernel_check_life(call->net->socket, call->rxcall, &last_life);
630 
631 	add_wait_queue(&call->waitq, &myself);
632 	for (;;) {
633 		set_current_state(TASK_UNINTERRUPTIBLE);
634 
635 		/* deliver any messages that are in the queue */
636 		if (!afs_check_call_state(call, AFS_CALL_COMPLETE) &&
637 		    call->need_attention) {
638 			call->need_attention = false;
639 			__set_current_state(TASK_RUNNING);
640 			afs_deliver_to_call(call);
641 			continue;
642 		}
643 
644 		if (afs_check_call_state(call, AFS_CALL_COMPLETE))
645 			break;
646 
647 		if (!rxrpc_kernel_check_life(call->net->socket, call->rxcall, &life)) {
648 			/* rxrpc terminated the call. */
649 			rxrpc_complete = true;
650 			break;
651 		}
652 
653 		if (timeout == 0 &&
654 		    life == last_life && signal_pending(current)) {
655 			if (stalled)
656 				break;
657 			__set_current_state(TASK_RUNNING);
658 			rxrpc_kernel_probe_life(call->net->socket, call->rxcall);
659 			timeout = rtt2;
660 			stalled = true;
661 			continue;
662 		}
663 
664 		if (life != last_life) {
665 			timeout = rtt2;
666 			last_life = life;
667 			stalled = false;
668 		}
669 
670 		timeout = schedule_timeout(timeout);
671 	}
672 
673 	remove_wait_queue(&call->waitq, &myself);
674 	__set_current_state(TASK_RUNNING);
675 
676 	if (!afs_check_call_state(call, AFS_CALL_COMPLETE)) {
677 		if (rxrpc_complete) {
678 			afs_set_call_complete(call, call->error, call->abort_code);
679 		} else {
680 			/* Kill off the call if it's still live. */
681 			_debug("call interrupted");
682 			if (rxrpc_kernel_abort_call(call->net->socket, call->rxcall,
683 						    RX_USER_ABORT, -EINTR, "KWI"))
684 				afs_set_call_complete(call, -EINTR, 0);
685 		}
686 	}
687 
688 	spin_lock_bh(&call->state_lock);
689 	ac->abort_code = call->abort_code;
690 	ac->error = call->error;
691 	spin_unlock_bh(&call->state_lock);
692 
693 	ret = ac->error;
694 	switch (ret) {
695 	case 0:
696 		if (call->ret_reply0) {
697 			ret = (long)call->reply[0];
698 			call->reply[0] = NULL;
699 		}
700 		/* Fall through */
701 	case -ECONNABORTED:
702 		ac->responded = true;
703 		break;
704 	}
705 
706 	_debug("call complete");
707 	afs_put_call(call);
708 	_leave(" = %p", (void *)ret);
709 	return ret;
710 }
711 
712 /*
713  * wake up a waiting call
714  */
715 static void afs_wake_up_call_waiter(struct sock *sk, struct rxrpc_call *rxcall,
716 				    unsigned long call_user_ID)
717 {
718 	struct afs_call *call = (struct afs_call *)call_user_ID;
719 
720 	call->need_attention = true;
721 	wake_up(&call->waitq);
722 }
723 
724 /*
725  * wake up an asynchronous call
726  */
727 static void afs_wake_up_async_call(struct sock *sk, struct rxrpc_call *rxcall,
728 				   unsigned long call_user_ID)
729 {
730 	struct afs_call *call = (struct afs_call *)call_user_ID;
731 	int u;
732 
733 	trace_afs_notify_call(rxcall, call);
734 	call->need_attention = true;
735 
736 	u = atomic_fetch_add_unless(&call->usage, 1, 0);
737 	if (u != 0) {
738 		trace_afs_call(call, afs_call_trace_wake, u,
739 			       atomic_read(&call->net->nr_outstanding_calls),
740 			       __builtin_return_address(0));
741 
742 		if (!queue_work(afs_async_calls, &call->async_work))
743 			afs_put_call(call);
744 	}
745 }
746 
747 /*
748  * Delete an asynchronous call.  The work item carries a ref to the call struct
749  * that we need to release.
750  */
751 static void afs_delete_async_call(struct work_struct *work)
752 {
753 	struct afs_call *call = container_of(work, struct afs_call, async_work);
754 
755 	_enter("");
756 
757 	afs_put_call(call);
758 
759 	_leave("");
760 }
761 
762 /*
763  * Perform I/O processing on an asynchronous call.  The work item carries a ref
764  * to the call struct that we either need to release or to pass on.
765  */
766 static void afs_process_async_call(struct work_struct *work)
767 {
768 	struct afs_call *call = container_of(work, struct afs_call, async_work);
769 
770 	_enter("");
771 
772 	if (call->state < AFS_CALL_COMPLETE && call->need_attention) {
773 		call->need_attention = false;
774 		afs_deliver_to_call(call);
775 	}
776 
777 	if (call->state == AFS_CALL_COMPLETE) {
778 		/* We have two refs to release - one from the alloc and one
779 		 * queued with the work item - and we can't just deallocate the
780 		 * call because the work item may be queued again.
781 		 */
782 		call->async_work.func = afs_delete_async_call;
783 		if (!queue_work(afs_async_calls, &call->async_work))
784 			afs_put_call(call);
785 	}
786 
787 	afs_put_call(call);
788 	_leave("");
789 }
790 
791 static void afs_rx_attach(struct rxrpc_call *rxcall, unsigned long user_call_ID)
792 {
793 	struct afs_call *call = (struct afs_call *)user_call_ID;
794 
795 	call->rxcall = rxcall;
796 }
797 
798 /*
799  * Charge the incoming call preallocation.
800  */
801 void afs_charge_preallocation(struct work_struct *work)
802 {
803 	struct afs_net *net =
804 		container_of(work, struct afs_net, charge_preallocation_work);
805 	struct afs_call *call = net->spare_incoming_call;
806 
807 	for (;;) {
808 		if (!call) {
809 			call = afs_alloc_call(net, &afs_RXCMxxxx, GFP_KERNEL);
810 			if (!call)
811 				break;
812 
813 			call->async = true;
814 			call->state = AFS_CALL_SV_AWAIT_OP_ID;
815 			init_waitqueue_head(&call->waitq);
816 			afs_extract_to_tmp(call);
817 		}
818 
819 		if (rxrpc_kernel_charge_accept(net->socket,
820 					       afs_wake_up_async_call,
821 					       afs_rx_attach,
822 					       (unsigned long)call,
823 					       GFP_KERNEL,
824 					       call->debug_id) < 0)
825 			break;
826 		call = NULL;
827 	}
828 	net->spare_incoming_call = call;
829 }
830 
831 /*
832  * Discard a preallocated call when a socket is shut down.
833  */
834 static void afs_rx_discard_new_call(struct rxrpc_call *rxcall,
835 				    unsigned long user_call_ID)
836 {
837 	struct afs_call *call = (struct afs_call *)user_call_ID;
838 
839 	call->rxcall = NULL;
840 	afs_put_call(call);
841 }
842 
843 /*
844  * Notification of an incoming call.
845  */
846 static void afs_rx_new_call(struct sock *sk, struct rxrpc_call *rxcall,
847 			    unsigned long user_call_ID)
848 {
849 	struct afs_net *net = afs_sock2net(sk);
850 
851 	queue_work(afs_wq, &net->charge_preallocation_work);
852 }
853 
854 /*
855  * Grab the operation ID from an incoming cache manager call.  The socket
856  * buffer is discarded on error or if we don't yet have sufficient data.
857  */
858 static int afs_deliver_cm_op_id(struct afs_call *call)
859 {
860 	int ret;
861 
862 	_enter("{%zu}", iov_iter_count(call->_iter));
863 
864 	/* the operation ID forms the first four bytes of the request data */
865 	ret = afs_extract_data(call, true);
866 	if (ret < 0)
867 		return ret;
868 
869 	call->operation_ID = ntohl(call->tmp);
870 	afs_set_call_state(call, AFS_CALL_SV_AWAIT_OP_ID, AFS_CALL_SV_AWAIT_REQUEST);
871 
872 	/* ask the cache manager to route the call (it'll change the call type
873 	 * if successful) */
874 	if (!afs_cm_incoming_call(call))
875 		return -ENOTSUPP;
876 
877 	trace_afs_cb_call(call);
878 
879 	/* pass responsibility for the remainer of this message off to the
880 	 * cache manager op */
881 	return call->type->deliver(call);
882 }
883 
884 /*
885  * Advance the AFS call state when an RxRPC service call ends the transmit
886  * phase.
887  */
888 static void afs_notify_end_reply_tx(struct sock *sock,
889 				    struct rxrpc_call *rxcall,
890 				    unsigned long call_user_ID)
891 {
892 	struct afs_call *call = (struct afs_call *)call_user_ID;
893 
894 	afs_set_call_state(call, AFS_CALL_SV_REPLYING, AFS_CALL_SV_AWAIT_ACK);
895 }
896 
897 /*
898  * send an empty reply
899  */
900 void afs_send_empty_reply(struct afs_call *call)
901 {
902 	struct afs_net *net = call->net;
903 	struct msghdr msg;
904 
905 	_enter("");
906 
907 	rxrpc_kernel_set_tx_length(net->socket, call->rxcall, 0);
908 
909 	msg.msg_name		= NULL;
910 	msg.msg_namelen		= 0;
911 	iov_iter_kvec(&msg.msg_iter, WRITE, NULL, 0, 0);
912 	msg.msg_control		= NULL;
913 	msg.msg_controllen	= 0;
914 	msg.msg_flags		= 0;
915 
916 	switch (rxrpc_kernel_send_data(net->socket, call->rxcall, &msg, 0,
917 				       afs_notify_end_reply_tx)) {
918 	case 0:
919 		_leave(" [replied]");
920 		return;
921 
922 	case -ENOMEM:
923 		_debug("oom");
924 		rxrpc_kernel_abort_call(net->socket, call->rxcall,
925 					RX_USER_ABORT, -ENOMEM, "KOO");
926 	default:
927 		_leave(" [error]");
928 		return;
929 	}
930 }
931 
932 /*
933  * send a simple reply
934  */
935 void afs_send_simple_reply(struct afs_call *call, const void *buf, size_t len)
936 {
937 	struct afs_net *net = call->net;
938 	struct msghdr msg;
939 	struct kvec iov[1];
940 	int n;
941 
942 	_enter("");
943 
944 	rxrpc_kernel_set_tx_length(net->socket, call->rxcall, len);
945 
946 	iov[0].iov_base		= (void *) buf;
947 	iov[0].iov_len		= len;
948 	msg.msg_name		= NULL;
949 	msg.msg_namelen		= 0;
950 	iov_iter_kvec(&msg.msg_iter, WRITE, iov, 1, len);
951 	msg.msg_control		= NULL;
952 	msg.msg_controllen	= 0;
953 	msg.msg_flags		= 0;
954 
955 	n = rxrpc_kernel_send_data(net->socket, call->rxcall, &msg, len,
956 				   afs_notify_end_reply_tx);
957 	if (n >= 0) {
958 		/* Success */
959 		_leave(" [replied]");
960 		return;
961 	}
962 
963 	if (n == -ENOMEM) {
964 		_debug("oom");
965 		rxrpc_kernel_abort_call(net->socket, call->rxcall,
966 					RX_USER_ABORT, -ENOMEM, "KOO");
967 	}
968 	_leave(" [error]");
969 }
970 
971 /*
972  * Extract a piece of data from the received data socket buffers.
973  */
974 int afs_extract_data(struct afs_call *call, bool want_more)
975 {
976 	struct afs_net *net = call->net;
977 	struct iov_iter *iter = call->_iter;
978 	enum afs_call_state state;
979 	u32 remote_abort = 0;
980 	int ret;
981 
982 	_enter("{%s,%zu},%d", call->type->name, iov_iter_count(iter), want_more);
983 
984 	ret = rxrpc_kernel_recv_data(net->socket, call->rxcall, iter,
985 				     want_more, &remote_abort,
986 				     &call->service_id);
987 	if (ret == 0 || ret == -EAGAIN)
988 		return ret;
989 
990 	state = READ_ONCE(call->state);
991 	if (ret == 1) {
992 		switch (state) {
993 		case AFS_CALL_CL_AWAIT_REPLY:
994 			afs_set_call_state(call, state, AFS_CALL_CL_PROC_REPLY);
995 			break;
996 		case AFS_CALL_SV_AWAIT_REQUEST:
997 			afs_set_call_state(call, state, AFS_CALL_SV_REPLYING);
998 			break;
999 		case AFS_CALL_COMPLETE:
1000 			kdebug("prem complete %d", call->error);
1001 			return afs_io_error(call, afs_io_error_extract);
1002 		default:
1003 			break;
1004 		}
1005 		return 0;
1006 	}
1007 
1008 	afs_set_call_complete(call, ret, remote_abort);
1009 	return ret;
1010 }
1011 
1012 /*
1013  * Log protocol error production.
1014  */
1015 noinline int afs_protocol_error(struct afs_call *call, int error,
1016 				enum afs_eproto_cause cause)
1017 {
1018 	trace_afs_protocol_error(call, error, cause);
1019 	return error;
1020 }
1021