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.
5 * Written by David Howells (dhowells@redhat.com)
8 #include <linux/slab.h>
9 #include <linux/sched/signal.h>
12 #include <net/af_rxrpc.h>
15 #include "protocol_yfs.h"
17 struct workqueue_struct *afs_async_calls;
19 static void afs_wake_up_call_waiter(struct sock *, struct rxrpc_call *, unsigned long);
20 static void afs_wake_up_async_call(struct sock *, struct rxrpc_call *, unsigned long);
21 static void afs_process_async_call(struct work_struct *);
22 static void afs_rx_new_call(struct sock *, struct rxrpc_call *, unsigned long);
23 static void afs_rx_discard_new_call(struct rxrpc_call *, unsigned long);
24 static int afs_deliver_cm_op_id(struct afs_call *);
26 /* asynchronous incoming call initial processing */
27 static const struct afs_call_type afs_RXCMxxxx = {
29 .deliver = afs_deliver_cm_op_id,
33 * open an RxRPC socket and bind it to be a server for callback notifications
34 * - the socket is left in blocking mode and non-blocking ops use MSG_DONTWAIT
36 int afs_open_socket(struct afs_net *net)
38 struct sockaddr_rxrpc srx;
39 struct socket *socket;
44 ret = sock_create_kern(net->net, AF_RXRPC, SOCK_DGRAM, PF_INET6, &socket);
48 socket->sk->sk_allocation = GFP_NOFS;
50 /* bind the callback manager's address to make this a server socket */
51 memset(&srx, 0, sizeof(srx));
52 srx.srx_family = AF_RXRPC;
53 srx.srx_service = CM_SERVICE;
54 srx.transport_type = SOCK_DGRAM;
55 srx.transport_len = sizeof(srx.transport.sin6);
56 srx.transport.sin6.sin6_family = AF_INET6;
57 srx.transport.sin6.sin6_port = htons(AFS_CM_PORT);
59 ret = rxrpc_sock_set_min_security_level(socket->sk,
60 RXRPC_SECURITY_ENCRYPT);
64 ret = kernel_bind(socket, (struct sockaddr *) &srx, sizeof(srx));
65 if (ret == -EADDRINUSE) {
66 srx.transport.sin6.sin6_port = 0;
67 ret = kernel_bind(socket, (struct sockaddr *) &srx, sizeof(srx));
72 srx.srx_service = YFS_CM_SERVICE;
73 ret = kernel_bind(socket, (struct sockaddr *) &srx, sizeof(srx));
77 /* Ideally, we'd turn on service upgrade here, but we can't because
78 * OpenAFS is buggy and leaks the userStatus field from packet to
79 * packet and between FS packets and CB packets - so if we try to do an
80 * upgrade on an FS packet, OpenAFS will leak that into the CB packet
81 * it sends back to us.
84 rxrpc_kernel_new_call_notification(socket, afs_rx_new_call,
85 afs_rx_discard_new_call);
87 ret = kernel_listen(socket, INT_MAX);
92 afs_charge_preallocation(&net->charge_preallocation_work);
104 * close the RxRPC socket AFS was using
106 void afs_close_socket(struct afs_net *net)
110 kernel_listen(net->socket, 0);
111 flush_workqueue(afs_async_calls);
113 if (net->spare_incoming_call) {
114 afs_put_call(net->spare_incoming_call);
115 net->spare_incoming_call = NULL;
118 _debug("outstanding %u", atomic_read(&net->nr_outstanding_calls));
119 wait_var_event(&net->nr_outstanding_calls,
120 !atomic_read(&net->nr_outstanding_calls));
121 _debug("no outstanding calls");
123 kernel_sock_shutdown(net->socket, SHUT_RDWR);
124 flush_workqueue(afs_async_calls);
125 sock_release(net->socket);
134 static struct afs_call *afs_alloc_call(struct afs_net *net,
135 const struct afs_call_type *type,
138 struct afs_call *call;
141 call = kzalloc(sizeof(*call), gfp);
147 call->debug_id = atomic_inc_return(&rxrpc_debug_id);
148 atomic_set(&call->usage, 1);
149 INIT_WORK(&call->async_work, afs_process_async_call);
150 init_waitqueue_head(&call->waitq);
151 spin_lock_init(&call->state_lock);
152 call->iter = &call->def_iter;
154 o = atomic_inc_return(&net->nr_outstanding_calls);
155 trace_afs_call(call, afs_call_trace_alloc, 1, o,
156 __builtin_return_address(0));
161 * Dispose of a reference on a call.
163 void afs_put_call(struct afs_call *call)
165 struct afs_net *net = call->net;
166 int n = atomic_dec_return(&call->usage);
167 int o = atomic_read(&net->nr_outstanding_calls);
169 trace_afs_call(call, afs_call_trace_put, n, o,
170 __builtin_return_address(0));
174 ASSERT(!work_pending(&call->async_work));
175 ASSERT(call->type->name != NULL);
178 rxrpc_kernel_end_call(net->socket, call->rxcall);
181 if (call->type->destructor)
182 call->type->destructor(call);
184 afs_put_server(call->net, call->server, afs_server_trace_put_call);
185 afs_put_cb_interest(call->net, call->cbi);
186 afs_put_addrlist(call->alist);
187 kfree(call->request);
189 trace_afs_call(call, afs_call_trace_free, 0, o,
190 __builtin_return_address(0));
193 o = atomic_dec_return(&net->nr_outstanding_calls);
195 wake_up_var(&net->nr_outstanding_calls);
199 static struct afs_call *afs_get_call(struct afs_call *call,
200 enum afs_call_trace why)
202 int u = atomic_inc_return(&call->usage);
204 trace_afs_call(call, why, u,
205 atomic_read(&call->net->nr_outstanding_calls),
206 __builtin_return_address(0));
211 * Queue the call for actual work.
213 static void afs_queue_call_work(struct afs_call *call)
215 if (call->type->work) {
216 INIT_WORK(&call->work, call->type->work);
218 afs_get_call(call, afs_call_trace_work);
219 if (!queue_work(afs_wq, &call->work))
225 * allocate a call with flat request and reply buffers
227 struct afs_call *afs_alloc_flat_call(struct afs_net *net,
228 const struct afs_call_type *type,
229 size_t request_size, size_t reply_max)
231 struct afs_call *call;
233 call = afs_alloc_call(net, type, GFP_NOFS);
238 call->request_size = request_size;
239 call->request = kmalloc(request_size, GFP_NOFS);
245 call->reply_max = reply_max;
246 call->buffer = kmalloc(reply_max, GFP_NOFS);
251 afs_extract_to_buf(call, call->reply_max);
252 call->operation_ID = type->op;
253 init_waitqueue_head(&call->waitq);
263 * clean up a call with flat buffer
265 void afs_flat_call_destructor(struct afs_call *call)
269 kfree(call->request);
270 call->request = NULL;
275 #define AFS_BVEC_MAX 8
278 * Load the given bvec with the next few pages.
280 static void afs_load_bvec(struct afs_call *call, struct msghdr *msg,
281 struct bio_vec *bv, pgoff_t first, pgoff_t last,
284 struct page *pages[AFS_BVEC_MAX];
285 unsigned int nr, n, i, to, bytes = 0;
287 nr = min_t(pgoff_t, last - first + 1, AFS_BVEC_MAX);
288 n = find_get_pages_contig(call->mapping, first, nr, pages);
289 ASSERTCMP(n, ==, nr);
291 msg->msg_flags |= MSG_MORE;
292 for (i = 0; i < nr; i++) {
294 if (first + i >= last) {
296 msg->msg_flags &= ~MSG_MORE;
298 bv[i].bv_page = pages[i];
299 bv[i].bv_len = to - offset;
300 bv[i].bv_offset = offset;
301 bytes += to - offset;
305 iov_iter_bvec(&msg->msg_iter, WRITE, bv, nr, bytes);
309 * Advance the AFS call state when the RxRPC call ends the transmit phase.
311 static void afs_notify_end_request_tx(struct sock *sock,
312 struct rxrpc_call *rxcall,
313 unsigned long call_user_ID)
315 struct afs_call *call = (struct afs_call *)call_user_ID;
317 afs_set_call_state(call, AFS_CALL_CL_REQUESTING, AFS_CALL_CL_AWAIT_REPLY);
321 * attach the data from a bunch of pages on an inode to a call
323 static int afs_send_pages(struct afs_call *call, struct msghdr *msg)
325 struct bio_vec bv[AFS_BVEC_MAX];
326 unsigned int bytes, nr, loop, offset;
327 pgoff_t first = call->first, last = call->last;
330 offset = call->first_offset;
331 call->first_offset = 0;
334 afs_load_bvec(call, msg, bv, first, last, offset);
335 trace_afs_send_pages(call, msg, first, last, offset);
338 bytes = msg->msg_iter.count;
339 nr = msg->msg_iter.nr_segs;
341 ret = rxrpc_kernel_send_data(call->net->socket, call->rxcall, msg,
342 bytes, afs_notify_end_request_tx);
343 for (loop = 0; loop < nr; loop++)
344 put_page(bv[loop].bv_page);
349 } while (first <= last);
351 trace_afs_sent_pages(call, call->first, last, first, ret);
356 * Initiate a call and synchronously queue up the parameters for dispatch. Any
357 * error is stored into the call struct, which the caller must check for.
359 void afs_make_call(struct afs_addr_cursor *ac, struct afs_call *call, gfp_t gfp)
361 struct sockaddr_rxrpc *srx = &ac->alist->addrs[ac->index];
362 struct rxrpc_call *rxcall;
368 _enter(",{%pISp},", &srx->transport);
370 ASSERT(call->type != NULL);
371 ASSERT(call->type->name != NULL);
373 _debug("____MAKE %p{%s,%x} [%d]____",
374 call, call->type->name, key_serial(call->key),
375 atomic_read(&call->net->nr_outstanding_calls));
377 call->addr_ix = ac->index;
378 call->alist = afs_get_addrlist(ac->alist);
380 /* Work out the length we're going to transmit. This is awkward for
381 * calls such as FS.StoreData where there's an extra injection of data
382 * after the initial fixed part.
384 tx_total_len = call->request_size;
385 if (call->send_pages) {
386 if (call->last == call->first) {
387 tx_total_len += call->last_to - call->first_offset;
389 /* It looks mathematically like you should be able to
390 * combine the following lines with the ones above, but
391 * unsigned arithmetic is fun when it wraps...
393 tx_total_len += PAGE_SIZE - call->first_offset;
394 tx_total_len += call->last_to;
395 tx_total_len += (call->last - call->first - 1) * PAGE_SIZE;
399 /* If the call is going to be asynchronous, we need an extra ref for
400 * the call to hold itself so the caller need not hang on to its ref.
403 afs_get_call(call, afs_call_trace_get);
404 call->drop_ref = true;
408 rxcall = rxrpc_kernel_begin_call(call->net->socket, srx, call->key,
412 afs_wake_up_async_call :
413 afs_wake_up_call_waiter),
415 (call->intr ? RXRPC_PREINTERRUPTIBLE :
416 RXRPC_UNINTERRUPTIBLE),
418 if (IS_ERR(rxcall)) {
419 ret = PTR_ERR(rxcall);
421 goto error_kill_call;
424 call->rxcall = rxcall;
426 if (call->max_lifespan)
427 rxrpc_kernel_set_max_life(call->net->socket, rxcall,
430 /* send the request */
431 iov[0].iov_base = call->request;
432 iov[0].iov_len = call->request_size;
436 iov_iter_kvec(&msg.msg_iter, WRITE, iov, 1, call->request_size);
437 msg.msg_control = NULL;
438 msg.msg_controllen = 0;
439 msg.msg_flags = MSG_WAITALL | (call->send_pages ? MSG_MORE : 0);
441 ret = rxrpc_kernel_send_data(call->net->socket, rxcall,
442 &msg, call->request_size,
443 afs_notify_end_request_tx);
447 if (call->send_pages) {
448 ret = afs_send_pages(call, &msg);
453 /* Note that at this point, we may have received the reply or an abort
454 * - and an asynchronous call may already have completed.
456 * afs_wait_for_call_to_complete(call, ac)
457 * must be called to synchronously clean up.
462 if (ret != -ECONNABORTED) {
463 rxrpc_kernel_abort_call(call->net->socket, rxcall,
464 RX_USER_ABORT, ret, "KSD");
466 iov_iter_kvec(&msg.msg_iter, READ, NULL, 0, 0);
467 rxrpc_kernel_recv_data(call->net->socket, rxcall,
468 &msg.msg_iter, false,
469 &call->abort_code, &call->service_id);
470 ac->abort_code = call->abort_code;
471 ac->responded = true;
474 trace_afs_call_done(call);
476 if (call->type->done)
477 call->type->done(call);
479 /* We need to dispose of the extra ref we grabbed for an async call.
480 * The call, however, might be queued on afs_async_calls and we need to
481 * make sure we don't get any more notifications that might requeue it.
484 rxrpc_kernel_end_call(call->net->socket, call->rxcall);
488 if (cancel_work_sync(&call->async_work))
494 call->state = AFS_CALL_COMPLETE;
495 _leave(" = %d", ret);
499 * deliver messages to a call
501 static void afs_deliver_to_call(struct afs_call *call)
503 enum afs_call_state state;
504 u32 abort_code, remote_abort = 0;
507 _enter("%s", call->type->name);
509 while (state = READ_ONCE(call->state),
510 state == AFS_CALL_CL_AWAIT_REPLY ||
511 state == AFS_CALL_SV_AWAIT_OP_ID ||
512 state == AFS_CALL_SV_AWAIT_REQUEST ||
513 state == AFS_CALL_SV_AWAIT_ACK
515 if (state == AFS_CALL_SV_AWAIT_ACK) {
516 iov_iter_kvec(&call->def_iter, READ, NULL, 0, 0);
517 ret = rxrpc_kernel_recv_data(call->net->socket,
518 call->rxcall, &call->def_iter,
519 false, &remote_abort,
521 trace_afs_receive_data(call, &call->def_iter, false, ret);
523 if (ret == -EINPROGRESS || ret == -EAGAIN)
525 if (ret < 0 || ret == 1) {
533 if (!call->have_reply_time &&
534 rxrpc_kernel_get_reply_time(call->net->socket,
537 call->have_reply_time = true;
539 ret = call->type->deliver(call);
540 state = READ_ONCE(call->state);
543 afs_queue_call_work(call);
544 if (state == AFS_CALL_CL_PROC_REPLY) {
546 set_bit(AFS_SERVER_FL_MAY_HAVE_CB,
547 &call->cbi->server->flags);
550 ASSERTCMP(state, >, AFS_CALL_CL_PROC_REPLY);
556 ASSERTCMP(state, ==, AFS_CALL_COMPLETE);
559 abort_code = RXGEN_OPCODE;
560 rxrpc_kernel_abort_call(call->net->socket, call->rxcall,
561 abort_code, ret, "KIV");
564 pr_err("kAFS: Call %u in bad state %u\n",
565 call->debug_id, state);
570 abort_code = RXGEN_CC_UNMARSHAL;
571 if (state != AFS_CALL_CL_AWAIT_REPLY)
572 abort_code = RXGEN_SS_UNMARSHAL;
573 rxrpc_kernel_abort_call(call->net->socket, call->rxcall,
574 abort_code, ret, "KUM");
577 abort_code = RX_USER_ABORT;
578 rxrpc_kernel_abort_call(call->net->socket, call->rxcall,
579 abort_code, ret, "KER");
585 if (call->type->done)
586 call->type->done(call);
594 afs_set_call_complete(call, ret, remote_abort);
595 state = AFS_CALL_COMPLETE;
600 * Wait synchronously for a call to complete and clean up the call struct.
602 long afs_wait_for_call_to_complete(struct afs_call *call,
603 struct afs_addr_cursor *ac)
606 bool rxrpc_complete = false;
608 DECLARE_WAITQUEUE(myself, current);
616 add_wait_queue(&call->waitq, &myself);
618 set_current_state(TASK_UNINTERRUPTIBLE);
620 /* deliver any messages that are in the queue */
621 if (!afs_check_call_state(call, AFS_CALL_COMPLETE) &&
622 call->need_attention) {
623 call->need_attention = false;
624 __set_current_state(TASK_RUNNING);
625 afs_deliver_to_call(call);
629 if (afs_check_call_state(call, AFS_CALL_COMPLETE))
632 if (!rxrpc_kernel_check_life(call->net->socket, call->rxcall)) {
633 /* rxrpc terminated the call. */
634 rxrpc_complete = true;
641 remove_wait_queue(&call->waitq, &myself);
642 __set_current_state(TASK_RUNNING);
644 if (!afs_check_call_state(call, AFS_CALL_COMPLETE)) {
645 if (rxrpc_complete) {
646 afs_set_call_complete(call, call->error, call->abort_code);
648 /* Kill off the call if it's still live. */
649 _debug("call interrupted");
650 if (rxrpc_kernel_abort_call(call->net->socket, call->rxcall,
651 RX_USER_ABORT, -EINTR, "KWI"))
652 afs_set_call_complete(call, -EINTR, 0);
656 spin_lock_bh(&call->state_lock);
657 ac->abort_code = call->abort_code;
658 ac->error = call->error;
659 spin_unlock_bh(&call->state_lock);
669 ac->responded = true;
674 _debug("call complete");
676 _leave(" = %p", (void *)ret);
681 * wake up a waiting call
683 static void afs_wake_up_call_waiter(struct sock *sk, struct rxrpc_call *rxcall,
684 unsigned long call_user_ID)
686 struct afs_call *call = (struct afs_call *)call_user_ID;
688 call->need_attention = true;
689 wake_up(&call->waitq);
693 * wake up an asynchronous call
695 static void afs_wake_up_async_call(struct sock *sk, struct rxrpc_call *rxcall,
696 unsigned long call_user_ID)
698 struct afs_call *call = (struct afs_call *)call_user_ID;
701 trace_afs_notify_call(rxcall, call);
702 call->need_attention = true;
704 u = atomic_fetch_add_unless(&call->usage, 1, 0);
706 trace_afs_call(call, afs_call_trace_wake, u + 1,
707 atomic_read(&call->net->nr_outstanding_calls),
708 __builtin_return_address(0));
710 if (!queue_work(afs_async_calls, &call->async_work))
716 * Perform I/O processing on an asynchronous call. The work item carries a ref
717 * to the call struct that we either need to release or to pass on.
719 static void afs_process_async_call(struct work_struct *work)
721 struct afs_call *call = container_of(work, struct afs_call, async_work);
725 if (call->state < AFS_CALL_COMPLETE && call->need_attention) {
726 call->need_attention = false;
727 afs_deliver_to_call(call);
734 static void afs_rx_attach(struct rxrpc_call *rxcall, unsigned long user_call_ID)
736 struct afs_call *call = (struct afs_call *)user_call_ID;
738 call->rxcall = rxcall;
742 * Charge the incoming call preallocation.
744 void afs_charge_preallocation(struct work_struct *work)
746 struct afs_net *net =
747 container_of(work, struct afs_net, charge_preallocation_work);
748 struct afs_call *call = net->spare_incoming_call;
752 call = afs_alloc_call(net, &afs_RXCMxxxx, GFP_KERNEL);
756 call->drop_ref = true;
758 call->state = AFS_CALL_SV_AWAIT_OP_ID;
759 init_waitqueue_head(&call->waitq);
760 afs_extract_to_tmp(call);
763 if (rxrpc_kernel_charge_accept(net->socket,
764 afs_wake_up_async_call,
772 net->spare_incoming_call = call;
776 * Discard a preallocated call when a socket is shut down.
778 static void afs_rx_discard_new_call(struct rxrpc_call *rxcall,
779 unsigned long user_call_ID)
781 struct afs_call *call = (struct afs_call *)user_call_ID;
788 * Notification of an incoming call.
790 static void afs_rx_new_call(struct sock *sk, struct rxrpc_call *rxcall,
791 unsigned long user_call_ID)
793 struct afs_net *net = afs_sock2net(sk);
795 queue_work(afs_wq, &net->charge_preallocation_work);
799 * Grab the operation ID from an incoming cache manager call. The socket
800 * buffer is discarded on error or if we don't yet have sufficient data.
802 static int afs_deliver_cm_op_id(struct afs_call *call)
806 _enter("{%zu}", iov_iter_count(call->iter));
808 /* the operation ID forms the first four bytes of the request data */
809 ret = afs_extract_data(call, true);
813 call->operation_ID = ntohl(call->tmp);
814 afs_set_call_state(call, AFS_CALL_SV_AWAIT_OP_ID, AFS_CALL_SV_AWAIT_REQUEST);
816 /* ask the cache manager to route the call (it'll change the call type
818 if (!afs_cm_incoming_call(call))
821 trace_afs_cb_call(call);
823 /* pass responsibility for the remainer of this message off to the
824 * cache manager op */
825 return call->type->deliver(call);
829 * Advance the AFS call state when an RxRPC service call ends the transmit
832 static void afs_notify_end_reply_tx(struct sock *sock,
833 struct rxrpc_call *rxcall,
834 unsigned long call_user_ID)
836 struct afs_call *call = (struct afs_call *)call_user_ID;
838 afs_set_call_state(call, AFS_CALL_SV_REPLYING, AFS_CALL_SV_AWAIT_ACK);
842 * send an empty reply
844 void afs_send_empty_reply(struct afs_call *call)
846 struct afs_net *net = call->net;
851 rxrpc_kernel_set_tx_length(net->socket, call->rxcall, 0);
855 iov_iter_kvec(&msg.msg_iter, WRITE, NULL, 0, 0);
856 msg.msg_control = NULL;
857 msg.msg_controllen = 0;
860 switch (rxrpc_kernel_send_data(net->socket, call->rxcall, &msg, 0,
861 afs_notify_end_reply_tx)) {
863 _leave(" [replied]");
868 rxrpc_kernel_abort_call(net->socket, call->rxcall,
869 RX_USER_ABORT, -ENOMEM, "KOO");
878 * send a simple reply
880 void afs_send_simple_reply(struct afs_call *call, const void *buf, size_t len)
882 struct afs_net *net = call->net;
889 rxrpc_kernel_set_tx_length(net->socket, call->rxcall, len);
891 iov[0].iov_base = (void *) buf;
892 iov[0].iov_len = len;
895 iov_iter_kvec(&msg.msg_iter, WRITE, iov, 1, len);
896 msg.msg_control = NULL;
897 msg.msg_controllen = 0;
900 n = rxrpc_kernel_send_data(net->socket, call->rxcall, &msg, len,
901 afs_notify_end_reply_tx);
904 _leave(" [replied]");
910 rxrpc_kernel_abort_call(net->socket, call->rxcall,
911 RX_USER_ABORT, -ENOMEM, "KOO");
917 * Extract a piece of data from the received data socket buffers.
919 int afs_extract_data(struct afs_call *call, bool want_more)
921 struct afs_net *net = call->net;
922 struct iov_iter *iter = call->iter;
923 enum afs_call_state state;
924 u32 remote_abort = 0;
927 _enter("{%s,%zu},%d", call->type->name, iov_iter_count(iter), want_more);
929 ret = rxrpc_kernel_recv_data(net->socket, call->rxcall, iter,
930 want_more, &remote_abort,
932 if (ret == 0 || ret == -EAGAIN)
935 state = READ_ONCE(call->state);
938 case AFS_CALL_CL_AWAIT_REPLY:
939 afs_set_call_state(call, state, AFS_CALL_CL_PROC_REPLY);
941 case AFS_CALL_SV_AWAIT_REQUEST:
942 afs_set_call_state(call, state, AFS_CALL_SV_REPLYING);
944 case AFS_CALL_COMPLETE:
945 kdebug("prem complete %d", call->error);
946 return afs_io_error(call, afs_io_error_extract);
953 afs_set_call_complete(call, ret, remote_abort);
958 * Log protocol error production.
960 noinline int afs_protocol_error(struct afs_call *call, int error,
961 enum afs_eproto_cause cause)
963 trace_afs_protocol_error(call, error, cause);