1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* Maintain an RxRPC server socket to do AFS communications through
4 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
8 #include <linux/slab.h>
9 #include <linux/sched/signal.h>
12 #include <net/af_rxrpc.h>
15 #include "protocol_yfs.h"
16 #define RXRPC_TRACE_ONLY_DEFINE_ENUMS
17 #include <trace/events/rxrpc.h>
19 struct workqueue_struct *afs_async_calls;
21 static void afs_wake_up_call_waiter(struct sock *, struct rxrpc_call *, unsigned long);
22 static void afs_wake_up_async_call(struct sock *, struct rxrpc_call *, unsigned long);
23 static void afs_process_async_call(struct work_struct *);
24 static void afs_rx_new_call(struct sock *, struct rxrpc_call *, unsigned long);
25 static void afs_rx_discard_new_call(struct rxrpc_call *, unsigned long);
26 static int afs_deliver_cm_op_id(struct afs_call *);
28 /* asynchronous incoming call initial processing */
29 static const struct afs_call_type afs_RXCMxxxx = {
31 .deliver = afs_deliver_cm_op_id,
35 * open an RxRPC socket and bind it to be a server for callback notifications
36 * - the socket is left in blocking mode and non-blocking ops use MSG_DONTWAIT
38 int afs_open_socket(struct afs_net *net)
40 struct sockaddr_rxrpc srx;
41 struct socket *socket;
46 ret = sock_create_kern(net->net, AF_RXRPC, SOCK_DGRAM, PF_INET6, &socket);
50 socket->sk->sk_allocation = GFP_NOFS;
52 /* bind the callback manager's address to make this a server socket */
53 memset(&srx, 0, sizeof(srx));
54 srx.srx_family = AF_RXRPC;
55 srx.srx_service = CM_SERVICE;
56 srx.transport_type = SOCK_DGRAM;
57 srx.transport_len = sizeof(srx.transport.sin6);
58 srx.transport.sin6.sin6_family = AF_INET6;
59 srx.transport.sin6.sin6_port = htons(AFS_CM_PORT);
61 ret = rxrpc_sock_set_min_security_level(socket->sk,
62 RXRPC_SECURITY_ENCRYPT);
66 ret = kernel_bind(socket, (struct sockaddr *) &srx, sizeof(srx));
67 if (ret == -EADDRINUSE) {
68 srx.transport.sin6.sin6_port = 0;
69 ret = kernel_bind(socket, (struct sockaddr *) &srx, sizeof(srx));
74 srx.srx_service = YFS_CM_SERVICE;
75 ret = kernel_bind(socket, (struct sockaddr *) &srx, sizeof(srx));
79 /* Ideally, we'd turn on service upgrade here, but we can't because
80 * OpenAFS is buggy and leaks the userStatus field from packet to
81 * packet and between FS packets and CB packets - so if we try to do an
82 * upgrade on an FS packet, OpenAFS will leak that into the CB packet
83 * it sends back to us.
86 rxrpc_kernel_new_call_notification(socket, afs_rx_new_call,
87 afs_rx_discard_new_call);
89 ret = kernel_listen(socket, INT_MAX);
94 afs_charge_preallocation(&net->charge_preallocation_work);
101 _leave(" = %d", ret);
106 * close the RxRPC socket AFS was using
108 void afs_close_socket(struct afs_net *net)
112 kernel_listen(net->socket, 0);
113 flush_workqueue(afs_async_calls);
115 if (net->spare_incoming_call) {
116 afs_put_call(net->spare_incoming_call);
117 net->spare_incoming_call = NULL;
120 _debug("outstanding %u", atomic_read(&net->nr_outstanding_calls));
121 wait_var_event(&net->nr_outstanding_calls,
122 !atomic_read(&net->nr_outstanding_calls));
123 _debug("no outstanding calls");
125 kernel_sock_shutdown(net->socket, SHUT_RDWR);
126 flush_workqueue(afs_async_calls);
127 sock_release(net->socket);
136 static struct afs_call *afs_alloc_call(struct afs_net *net,
137 const struct afs_call_type *type,
140 struct afs_call *call;
143 call = kzalloc(sizeof(*call), gfp);
149 call->debug_id = atomic_inc_return(&rxrpc_debug_id);
150 refcount_set(&call->ref, 1);
151 INIT_WORK(&call->async_work, afs_process_async_call);
152 init_waitqueue_head(&call->waitq);
153 spin_lock_init(&call->state_lock);
154 call->iter = &call->def_iter;
156 o = atomic_inc_return(&net->nr_outstanding_calls);
157 trace_afs_call(call->debug_id, afs_call_trace_alloc, 1, o,
158 __builtin_return_address(0));
163 * Dispose of a reference on a call.
165 void afs_put_call(struct afs_call *call)
167 struct afs_net *net = call->net;
168 unsigned int debug_id = call->debug_id;
172 zero = __refcount_dec_and_test(&call->ref, &r);
173 o = atomic_read(&net->nr_outstanding_calls);
174 trace_afs_call(debug_id, afs_call_trace_put, r - 1, o,
175 __builtin_return_address(0));
178 ASSERT(!work_pending(&call->async_work));
179 ASSERT(call->type->name != NULL);
182 rxrpc_kernel_end_call(net->socket, call->rxcall);
185 if (call->type->destructor)
186 call->type->destructor(call);
188 afs_unuse_server_notime(call->net, call->server, afs_server_trace_put_call);
189 afs_put_addrlist(call->alist);
190 kfree(call->request);
192 trace_afs_call(call->debug_id, afs_call_trace_free, 0, o,
193 __builtin_return_address(0));
196 o = atomic_dec_return(&net->nr_outstanding_calls);
198 wake_up_var(&net->nr_outstanding_calls);
202 static struct afs_call *afs_get_call(struct afs_call *call,
203 enum afs_call_trace why)
207 __refcount_inc(&call->ref, &r);
209 trace_afs_call(call->debug_id, why, r + 1,
210 atomic_read(&call->net->nr_outstanding_calls),
211 __builtin_return_address(0));
216 * Queue the call for actual work.
218 static void afs_queue_call_work(struct afs_call *call)
220 if (call->type->work) {
221 INIT_WORK(&call->work, call->type->work);
223 afs_get_call(call, afs_call_trace_work);
224 if (!queue_work(afs_wq, &call->work))
230 * allocate a call with flat request and reply buffers
232 struct afs_call *afs_alloc_flat_call(struct afs_net *net,
233 const struct afs_call_type *type,
234 size_t request_size, size_t reply_max)
236 struct afs_call *call;
238 call = afs_alloc_call(net, type, GFP_NOFS);
243 call->request_size = request_size;
244 call->request = kmalloc(request_size, GFP_NOFS);
250 call->reply_max = reply_max;
251 call->buffer = kmalloc(reply_max, GFP_NOFS);
256 afs_extract_to_buf(call, call->reply_max);
257 call->operation_ID = type->op;
258 init_waitqueue_head(&call->waitq);
268 * clean up a call with flat buffer
270 void afs_flat_call_destructor(struct afs_call *call)
274 kfree(call->request);
275 call->request = NULL;
281 * Advance the AFS call state when the RxRPC call ends the transmit phase.
283 static void afs_notify_end_request_tx(struct sock *sock,
284 struct rxrpc_call *rxcall,
285 unsigned long call_user_ID)
287 struct afs_call *call = (struct afs_call *)call_user_ID;
289 afs_set_call_state(call, AFS_CALL_CL_REQUESTING, AFS_CALL_CL_AWAIT_REPLY);
293 * Initiate a call and synchronously queue up the parameters for dispatch. Any
294 * error is stored into the call struct, which the caller must check for.
296 void afs_make_call(struct afs_addr_cursor *ac, struct afs_call *call, gfp_t gfp)
298 struct sockaddr_rxrpc *srx = &ac->alist->addrs[ac->index];
299 struct rxrpc_call *rxcall;
306 _enter(",{%pISp},", &srx->transport);
308 ASSERT(call->type != NULL);
309 ASSERT(call->type->name != NULL);
311 _debug("____MAKE %p{%s,%x} [%d]____",
312 call, call->type->name, key_serial(call->key),
313 atomic_read(&call->net->nr_outstanding_calls));
315 call->addr_ix = ac->index;
316 call->alist = afs_get_addrlist(ac->alist);
318 /* Work out the length we're going to transmit. This is awkward for
319 * calls such as FS.StoreData where there's an extra injection of data
320 * after the initial fixed part.
322 tx_total_len = call->request_size;
323 if (call->write_iter)
324 tx_total_len += iov_iter_count(call->write_iter);
326 /* If the call is going to be asynchronous, we need an extra ref for
327 * the call to hold itself so the caller need not hang on to its ref.
330 afs_get_call(call, afs_call_trace_get);
331 call->drop_ref = true;
335 rxcall = rxrpc_kernel_begin_call(call->net->socket, srx, call->key,
339 afs_wake_up_async_call :
340 afs_wake_up_call_waiter),
342 (call->intr ? RXRPC_PREINTERRUPTIBLE :
343 RXRPC_UNINTERRUPTIBLE),
345 if (IS_ERR(rxcall)) {
346 ret = PTR_ERR(rxcall);
348 goto error_kill_call;
351 call->rxcall = rxcall;
353 if (call->max_lifespan)
354 rxrpc_kernel_set_max_life(call->net->socket, rxcall,
356 call->issue_time = ktime_get_real();
358 /* send the request */
359 iov[0].iov_base = call->request;
360 iov[0].iov_len = call->request_size;
364 iov_iter_kvec(&msg.msg_iter, ITER_SOURCE, iov, 1, call->request_size);
365 msg.msg_control = NULL;
366 msg.msg_controllen = 0;
367 msg.msg_flags = MSG_WAITALL | (call->write_iter ? MSG_MORE : 0);
369 ret = rxrpc_kernel_send_data(call->net->socket, rxcall,
370 &msg, call->request_size,
371 afs_notify_end_request_tx);
375 if (call->write_iter) {
376 msg.msg_iter = *call->write_iter;
377 msg.msg_flags &= ~MSG_MORE;
378 trace_afs_send_data(call, &msg);
380 ret = rxrpc_kernel_send_data(call->net->socket,
382 iov_iter_count(&msg.msg_iter),
383 afs_notify_end_request_tx);
384 *call->write_iter = msg.msg_iter;
386 trace_afs_sent_data(call, &msg, ret);
391 /* Note that at this point, we may have received the reply or an abort
392 * - and an asynchronous call may already have completed.
394 * afs_wait_for_call_to_complete(call, ac)
395 * must be called to synchronously clean up.
400 if (ret != -ECONNABORTED) {
401 rxrpc_kernel_abort_call(call->net->socket, rxcall,
403 afs_abort_send_data_error);
406 iov_iter_kvec(&msg.msg_iter, ITER_DEST, NULL, 0, 0);
407 rxrpc_kernel_recv_data(call->net->socket, rxcall,
408 &msg.msg_iter, &len, false,
409 &call->abort_code, &call->service_id);
410 ac->abort_code = call->abort_code;
411 ac->responded = true;
414 trace_afs_call_done(call);
416 if (call->type->done)
417 call->type->done(call);
419 /* We need to dispose of the extra ref we grabbed for an async call.
420 * The call, however, might be queued on afs_async_calls and we need to
421 * make sure we don't get any more notifications that might requeue it.
424 rxrpc_kernel_end_call(call->net->socket, call->rxcall);
428 if (cancel_work_sync(&call->async_work))
434 call->state = AFS_CALL_COMPLETE;
435 _leave(" = %d", ret);
439 * Log remote abort codes that indicate that we have a protocol disagreement
442 static void afs_log_error(struct afs_call *call, s32 remote_abort)
448 switch (remote_abort) {
449 case RX_EOF: msg = "unexpected EOF"; break;
450 case RXGEN_CC_MARSHAL: msg = "client marshalling"; break;
451 case RXGEN_CC_UNMARSHAL: msg = "client unmarshalling"; break;
452 case RXGEN_SS_MARSHAL: msg = "server marshalling"; break;
453 case RXGEN_SS_UNMARSHAL: msg = "server unmarshalling"; break;
454 case RXGEN_DECODE: msg = "opcode decode"; break;
455 case RXGEN_SS_XDRFREE: msg = "server XDR cleanup"; break;
456 case RXGEN_CC_XDRFREE: msg = "client XDR cleanup"; break;
457 case -32: msg = "insufficient data"; break;
465 pr_notice("kAFS: Peer reported %s failure on %s [%pISp]\n",
466 msg, call->type->name,
467 &call->alist->addrs[call->addr_ix].transport);
472 * deliver messages to a call
474 static void afs_deliver_to_call(struct afs_call *call)
476 enum afs_call_state state;
478 u32 abort_code, remote_abort = 0;
481 _enter("%s", call->type->name);
483 while (state = READ_ONCE(call->state),
484 state == AFS_CALL_CL_AWAIT_REPLY ||
485 state == AFS_CALL_SV_AWAIT_OP_ID ||
486 state == AFS_CALL_SV_AWAIT_REQUEST ||
487 state == AFS_CALL_SV_AWAIT_ACK
489 if (state == AFS_CALL_SV_AWAIT_ACK) {
491 iov_iter_kvec(&call->def_iter, ITER_DEST, NULL, 0, 0);
492 ret = rxrpc_kernel_recv_data(call->net->socket,
493 call->rxcall, &call->def_iter,
494 &len, false, &remote_abort,
496 trace_afs_receive_data(call, &call->def_iter, false, ret);
498 if (ret == -EINPROGRESS || ret == -EAGAIN)
500 if (ret < 0 || ret == 1) {
508 ret = call->type->deliver(call);
509 state = READ_ONCE(call->state);
510 if (ret == 0 && call->unmarshalling_error)
514 afs_queue_call_work(call);
515 if (state == AFS_CALL_CL_PROC_REPLY) {
517 set_bit(AFS_SERVER_FL_MAY_HAVE_CB,
518 &call->op->server->flags);
521 ASSERTCMP(state, >, AFS_CALL_CL_PROC_REPLY);
527 ASSERTCMP(state, ==, AFS_CALL_COMPLETE);
528 afs_log_error(call, call->abort_code);
531 abort_code = RXGEN_OPCODE;
532 rxrpc_kernel_abort_call(call->net->socket, call->rxcall,
534 afs_abort_op_not_supported);
537 pr_err("kAFS: Call %u in bad state %u\n",
538 call->debug_id, state);
545 abort_code = RXGEN_CC_UNMARSHAL;
546 if (state != AFS_CALL_CL_AWAIT_REPLY)
547 abort_code = RXGEN_SS_UNMARSHAL;
548 rxrpc_kernel_abort_call(call->net->socket, call->rxcall,
550 afs_abort_unmarshal_error);
553 abort_code = RX_CALL_DEAD;
554 rxrpc_kernel_abort_call(call->net->socket, call->rxcall,
556 afs_abort_general_error);
562 if (call->type->done)
563 call->type->done(call);
571 afs_set_call_complete(call, ret, remote_abort);
572 state = AFS_CALL_COMPLETE;
577 * Wait synchronously for a call to complete and clean up the call struct.
579 long afs_wait_for_call_to_complete(struct afs_call *call,
580 struct afs_addr_cursor *ac)
583 bool rxrpc_complete = false;
585 DECLARE_WAITQUEUE(myself, current);
593 add_wait_queue(&call->waitq, &myself);
595 set_current_state(TASK_UNINTERRUPTIBLE);
597 /* deliver any messages that are in the queue */
598 if (!afs_check_call_state(call, AFS_CALL_COMPLETE) &&
599 call->need_attention) {
600 call->need_attention = false;
601 __set_current_state(TASK_RUNNING);
602 afs_deliver_to_call(call);
606 if (afs_check_call_state(call, AFS_CALL_COMPLETE))
609 if (!rxrpc_kernel_check_life(call->net->socket, call->rxcall)) {
610 /* rxrpc terminated the call. */
611 rxrpc_complete = true;
618 remove_wait_queue(&call->waitq, &myself);
619 __set_current_state(TASK_RUNNING);
621 if (!afs_check_call_state(call, AFS_CALL_COMPLETE)) {
622 if (rxrpc_complete) {
623 afs_set_call_complete(call, call->error, call->abort_code);
625 /* Kill off the call if it's still live. */
626 _debug("call interrupted");
627 if (rxrpc_kernel_abort_call(call->net->socket, call->rxcall,
628 RX_USER_ABORT, -EINTR,
629 afs_abort_interrupted))
630 afs_set_call_complete(call, -EINTR, 0);
634 spin_lock_bh(&call->state_lock);
635 ac->abort_code = call->abort_code;
636 ac->error = call->error;
637 spin_unlock_bh(&call->state_lock);
647 ac->responded = true;
652 _debug("call complete");
654 _leave(" = %p", (void *)ret);
659 * wake up a waiting call
661 static void afs_wake_up_call_waiter(struct sock *sk, struct rxrpc_call *rxcall,
662 unsigned long call_user_ID)
664 struct afs_call *call = (struct afs_call *)call_user_ID;
666 call->need_attention = true;
667 wake_up(&call->waitq);
671 * wake up an asynchronous call
673 static void afs_wake_up_async_call(struct sock *sk, struct rxrpc_call *rxcall,
674 unsigned long call_user_ID)
676 struct afs_call *call = (struct afs_call *)call_user_ID;
679 trace_afs_notify_call(rxcall, call);
680 call->need_attention = true;
682 if (__refcount_inc_not_zero(&call->ref, &r)) {
683 trace_afs_call(call->debug_id, afs_call_trace_wake, r + 1,
684 atomic_read(&call->net->nr_outstanding_calls),
685 __builtin_return_address(0));
687 if (!queue_work(afs_async_calls, &call->async_work))
693 * Perform I/O processing on an asynchronous call. The work item carries a ref
694 * to the call struct that we either need to release or to pass on.
696 static void afs_process_async_call(struct work_struct *work)
698 struct afs_call *call = container_of(work, struct afs_call, async_work);
702 if (call->state < AFS_CALL_COMPLETE && call->need_attention) {
703 call->need_attention = false;
704 afs_deliver_to_call(call);
711 static void afs_rx_attach(struct rxrpc_call *rxcall, unsigned long user_call_ID)
713 struct afs_call *call = (struct afs_call *)user_call_ID;
715 call->rxcall = rxcall;
719 * Charge the incoming call preallocation.
721 void afs_charge_preallocation(struct work_struct *work)
723 struct afs_net *net =
724 container_of(work, struct afs_net, charge_preallocation_work);
725 struct afs_call *call = net->spare_incoming_call;
729 call = afs_alloc_call(net, &afs_RXCMxxxx, GFP_KERNEL);
733 call->drop_ref = true;
735 call->state = AFS_CALL_SV_AWAIT_OP_ID;
736 init_waitqueue_head(&call->waitq);
737 afs_extract_to_tmp(call);
740 if (rxrpc_kernel_charge_accept(net->socket,
741 afs_wake_up_async_call,
749 net->spare_incoming_call = call;
753 * Discard a preallocated call when a socket is shut down.
755 static void afs_rx_discard_new_call(struct rxrpc_call *rxcall,
756 unsigned long user_call_ID)
758 struct afs_call *call = (struct afs_call *)user_call_ID;
765 * Notification of an incoming call.
767 static void afs_rx_new_call(struct sock *sk, struct rxrpc_call *rxcall,
768 unsigned long user_call_ID)
770 struct afs_net *net = afs_sock2net(sk);
772 queue_work(afs_wq, &net->charge_preallocation_work);
776 * Grab the operation ID from an incoming cache manager call. The socket
777 * buffer is discarded on error or if we don't yet have sufficient data.
779 static int afs_deliver_cm_op_id(struct afs_call *call)
783 _enter("{%zu}", iov_iter_count(call->iter));
785 /* the operation ID forms the first four bytes of the request data */
786 ret = afs_extract_data(call, true);
790 call->operation_ID = ntohl(call->tmp);
791 afs_set_call_state(call, AFS_CALL_SV_AWAIT_OP_ID, AFS_CALL_SV_AWAIT_REQUEST);
793 /* ask the cache manager to route the call (it'll change the call type
795 if (!afs_cm_incoming_call(call))
798 trace_afs_cb_call(call);
800 /* pass responsibility for the remainer of this message off to the
801 * cache manager op */
802 return call->type->deliver(call);
806 * Advance the AFS call state when an RxRPC service call ends the transmit
809 static void afs_notify_end_reply_tx(struct sock *sock,
810 struct rxrpc_call *rxcall,
811 unsigned long call_user_ID)
813 struct afs_call *call = (struct afs_call *)call_user_ID;
815 afs_set_call_state(call, AFS_CALL_SV_REPLYING, AFS_CALL_SV_AWAIT_ACK);
819 * send an empty reply
821 void afs_send_empty_reply(struct afs_call *call)
823 struct afs_net *net = call->net;
828 rxrpc_kernel_set_tx_length(net->socket, call->rxcall, 0);
832 iov_iter_kvec(&msg.msg_iter, ITER_SOURCE, NULL, 0, 0);
833 msg.msg_control = NULL;
834 msg.msg_controllen = 0;
837 switch (rxrpc_kernel_send_data(net->socket, call->rxcall, &msg, 0,
838 afs_notify_end_reply_tx)) {
840 _leave(" [replied]");
845 rxrpc_kernel_abort_call(net->socket, call->rxcall,
846 RXGEN_SS_MARSHAL, -ENOMEM,
856 * send a simple reply
858 void afs_send_simple_reply(struct afs_call *call, const void *buf, size_t len)
860 struct afs_net *net = call->net;
867 rxrpc_kernel_set_tx_length(net->socket, call->rxcall, len);
869 iov[0].iov_base = (void *) buf;
870 iov[0].iov_len = len;
873 iov_iter_kvec(&msg.msg_iter, ITER_SOURCE, iov, 1, len);
874 msg.msg_control = NULL;
875 msg.msg_controllen = 0;
878 n = rxrpc_kernel_send_data(net->socket, call->rxcall, &msg, len,
879 afs_notify_end_reply_tx);
882 _leave(" [replied]");
888 rxrpc_kernel_abort_call(net->socket, call->rxcall,
889 RXGEN_SS_MARSHAL, -ENOMEM,
896 * Extract a piece of data from the received data socket buffers.
898 int afs_extract_data(struct afs_call *call, bool want_more)
900 struct afs_net *net = call->net;
901 struct iov_iter *iter = call->iter;
902 enum afs_call_state state;
903 u32 remote_abort = 0;
906 _enter("{%s,%zu,%zu},%d",
907 call->type->name, call->iov_len, iov_iter_count(iter), want_more);
909 ret = rxrpc_kernel_recv_data(net->socket, call->rxcall, iter,
910 &call->iov_len, want_more, &remote_abort,
912 trace_afs_receive_data(call, call->iter, want_more, ret);
913 if (ret == 0 || ret == -EAGAIN)
916 state = READ_ONCE(call->state);
919 case AFS_CALL_CL_AWAIT_REPLY:
920 afs_set_call_state(call, state, AFS_CALL_CL_PROC_REPLY);
922 case AFS_CALL_SV_AWAIT_REQUEST:
923 afs_set_call_state(call, state, AFS_CALL_SV_REPLYING);
925 case AFS_CALL_COMPLETE:
926 kdebug("prem complete %d", call->error);
927 return afs_io_error(call, afs_io_error_extract);
934 afs_set_call_complete(call, ret, remote_abort);
939 * Log protocol error production.
941 noinline int afs_protocol_error(struct afs_call *call,
942 enum afs_eproto_cause cause)
944 trace_afs_protocol_error(call, cause);
946 call->unmarshalling_error = true;