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;
40 unsigned int min_level;
45 ret = sock_create_kern(net->net, AF_RXRPC, SOCK_DGRAM, PF_INET6, &socket);
49 socket->sk->sk_allocation = GFP_NOFS;
51 /* bind the callback manager's address to make this a server socket */
52 memset(&srx, 0, sizeof(srx));
53 srx.srx_family = AF_RXRPC;
54 srx.srx_service = CM_SERVICE;
55 srx.transport_type = SOCK_DGRAM;
56 srx.transport_len = sizeof(srx.transport.sin6);
57 srx.transport.sin6.sin6_family = AF_INET6;
58 srx.transport.sin6.sin6_port = htons(AFS_CM_PORT);
60 min_level = RXRPC_SECURITY_ENCRYPT;
61 ret = kernel_setsockopt(socket, SOL_RXRPC, RXRPC_MIN_SECURITY_LEVEL,
62 (void *)&min_level, sizeof(min_level));
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 atomic_set(&call->usage, 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, 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 int n = atomic_dec_return(&call->usage);
169 int o = atomic_read(&net->nr_outstanding_calls);
171 trace_afs_call(call, afs_call_trace_put, n, o,
172 __builtin_return_address(0));
176 ASSERT(!work_pending(&call->async_work));
177 ASSERT(call->type->name != NULL);
180 rxrpc_kernel_end_call(net->socket, call->rxcall);
183 if (call->type->destructor)
184 call->type->destructor(call);
186 afs_put_server(call->net, call->server, afs_server_trace_put_call);
187 afs_put_cb_interest(call->net, call->cbi);
188 afs_put_addrlist(call->alist);
189 kfree(call->request);
191 trace_afs_call(call, afs_call_trace_free, 0, o,
192 __builtin_return_address(0));
195 o = atomic_dec_return(&net->nr_outstanding_calls);
197 wake_up_var(&net->nr_outstanding_calls);
201 static struct afs_call *afs_get_call(struct afs_call *call,
202 enum afs_call_trace why)
204 int u = atomic_inc_return(&call->usage);
206 trace_afs_call(call, why, u,
207 atomic_read(&call->net->nr_outstanding_calls),
208 __builtin_return_address(0));
213 * Queue the call for actual work.
215 static void afs_queue_call_work(struct afs_call *call)
217 if (call->type->work) {
218 INIT_WORK(&call->work, call->type->work);
220 afs_get_call(call, afs_call_trace_work);
221 if (!queue_work(afs_wq, &call->work))
227 * allocate a call with flat request and reply buffers
229 struct afs_call *afs_alloc_flat_call(struct afs_net *net,
230 const struct afs_call_type *type,
231 size_t request_size, size_t reply_max)
233 struct afs_call *call;
235 call = afs_alloc_call(net, type, GFP_NOFS);
240 call->request_size = request_size;
241 call->request = kmalloc(request_size, GFP_NOFS);
247 call->reply_max = reply_max;
248 call->buffer = kmalloc(reply_max, GFP_NOFS);
253 afs_extract_to_buf(call, call->reply_max);
254 call->operation_ID = type->op;
255 init_waitqueue_head(&call->waitq);
265 * clean up a call with flat buffer
267 void afs_flat_call_destructor(struct afs_call *call)
271 kfree(call->request);
272 call->request = NULL;
277 #define AFS_BVEC_MAX 8
280 * Load the given bvec with the next few pages.
282 static void afs_load_bvec(struct afs_call *call, struct msghdr *msg,
283 struct bio_vec *bv, pgoff_t first, pgoff_t last,
286 struct page *pages[AFS_BVEC_MAX];
287 unsigned int nr, n, i, to, bytes = 0;
289 nr = min_t(pgoff_t, last - first + 1, AFS_BVEC_MAX);
290 n = find_get_pages_contig(call->mapping, first, nr, pages);
291 ASSERTCMP(n, ==, nr);
293 msg->msg_flags |= MSG_MORE;
294 for (i = 0; i < nr; i++) {
296 if (first + i >= last) {
298 msg->msg_flags &= ~MSG_MORE;
300 bv[i].bv_page = pages[i];
301 bv[i].bv_len = to - offset;
302 bv[i].bv_offset = offset;
303 bytes += to - offset;
307 iov_iter_bvec(&msg->msg_iter, WRITE, bv, nr, bytes);
311 * Advance the AFS call state when the RxRPC call ends the transmit phase.
313 static void afs_notify_end_request_tx(struct sock *sock,
314 struct rxrpc_call *rxcall,
315 unsigned long call_user_ID)
317 struct afs_call *call = (struct afs_call *)call_user_ID;
319 afs_set_call_state(call, AFS_CALL_CL_REQUESTING, AFS_CALL_CL_AWAIT_REPLY);
323 * attach the data from a bunch of pages on an inode to a call
325 static int afs_send_pages(struct afs_call *call, struct msghdr *msg)
327 struct bio_vec bv[AFS_BVEC_MAX];
328 unsigned int bytes, nr, loop, offset;
329 pgoff_t first = call->first, last = call->last;
332 offset = call->first_offset;
333 call->first_offset = 0;
336 afs_load_bvec(call, msg, bv, first, last, offset);
337 trace_afs_send_pages(call, msg, first, last, offset);
340 bytes = msg->msg_iter.count;
341 nr = msg->msg_iter.nr_segs;
343 ret = rxrpc_kernel_send_data(call->net->socket, call->rxcall, msg,
344 bytes, afs_notify_end_request_tx);
345 for (loop = 0; loop < nr; loop++)
346 put_page(bv[loop].bv_page);
351 } while (first <= last);
353 trace_afs_sent_pages(call, call->first, last, first, ret);
358 * Initiate a call and synchronously queue up the parameters for dispatch. Any
359 * error is stored into the call struct, which the caller must check for.
361 void afs_make_call(struct afs_addr_cursor *ac, struct afs_call *call, gfp_t gfp)
363 struct sockaddr_rxrpc *srx = &ac->alist->addrs[ac->index];
364 struct rxrpc_call *rxcall;
370 _enter(",{%pISp},", &srx->transport);
372 ASSERT(call->type != NULL);
373 ASSERT(call->type->name != NULL);
375 _debug("____MAKE %p{%s,%x} [%d]____",
376 call, call->type->name, key_serial(call->key),
377 atomic_read(&call->net->nr_outstanding_calls));
379 call->addr_ix = ac->index;
380 call->alist = afs_get_addrlist(ac->alist);
382 /* Work out the length we're going to transmit. This is awkward for
383 * calls such as FS.StoreData where there's an extra injection of data
384 * after the initial fixed part.
386 tx_total_len = call->request_size;
387 if (call->send_pages) {
388 if (call->last == call->first) {
389 tx_total_len += call->last_to - call->first_offset;
391 /* It looks mathematically like you should be able to
392 * combine the following lines with the ones above, but
393 * unsigned arithmetic is fun when it wraps...
395 tx_total_len += PAGE_SIZE - call->first_offset;
396 tx_total_len += call->last_to;
397 tx_total_len += (call->last - call->first - 1) * PAGE_SIZE;
401 /* If the call is going to be asynchronous, we need an extra ref for
402 * the call to hold itself so the caller need not hang on to its ref.
405 afs_get_call(call, afs_call_trace_get);
406 call->drop_ref = true;
410 rxcall = rxrpc_kernel_begin_call(call->net->socket, srx, call->key,
414 afs_wake_up_async_call :
415 afs_wake_up_call_waiter),
417 (call->intr ? RXRPC_PREINTERRUPTIBLE :
418 RXRPC_UNINTERRUPTIBLE),
420 if (IS_ERR(rxcall)) {
421 ret = PTR_ERR(rxcall);
423 goto error_kill_call;
426 call->rxcall = rxcall;
428 if (call->max_lifespan)
429 rxrpc_kernel_set_max_life(call->net->socket, rxcall,
432 /* send the request */
433 iov[0].iov_base = call->request;
434 iov[0].iov_len = call->request_size;
438 iov_iter_kvec(&msg.msg_iter, WRITE, iov, 1, call->request_size);
439 msg.msg_control = NULL;
440 msg.msg_controllen = 0;
441 msg.msg_flags = MSG_WAITALL | (call->send_pages ? MSG_MORE : 0);
443 ret = rxrpc_kernel_send_data(call->net->socket, rxcall,
444 &msg, call->request_size,
445 afs_notify_end_request_tx);
449 if (call->send_pages) {
450 ret = afs_send_pages(call, &msg);
455 /* Note that at this point, we may have received the reply or an abort
456 * - and an asynchronous call may already have completed.
458 * afs_wait_for_call_to_complete(call, ac)
459 * must be called to synchronously clean up.
464 if (ret != -ECONNABORTED) {
465 rxrpc_kernel_abort_call(call->net->socket, rxcall,
466 RX_USER_ABORT, ret, "KSD");
468 iov_iter_kvec(&msg.msg_iter, READ, NULL, 0, 0);
469 rxrpc_kernel_recv_data(call->net->socket, rxcall,
470 &msg.msg_iter, false,
471 &call->abort_code, &call->service_id);
472 ac->abort_code = call->abort_code;
473 ac->responded = true;
476 trace_afs_call_done(call);
478 if (call->type->done)
479 call->type->done(call);
481 /* We need to dispose of the extra ref we grabbed for an async call.
482 * The call, however, might be queued on afs_async_calls and we need to
483 * make sure we don't get any more notifications that might requeue it.
486 rxrpc_kernel_end_call(call->net->socket, call->rxcall);
490 if (cancel_work_sync(&call->async_work))
496 call->state = AFS_CALL_COMPLETE;
497 _leave(" = %d", ret);
501 * deliver messages to a call
503 static void afs_deliver_to_call(struct afs_call *call)
505 enum afs_call_state state;
506 u32 abort_code, remote_abort = 0;
509 _enter("%s", call->type->name);
511 while (state = READ_ONCE(call->state),
512 state == AFS_CALL_CL_AWAIT_REPLY ||
513 state == AFS_CALL_SV_AWAIT_OP_ID ||
514 state == AFS_CALL_SV_AWAIT_REQUEST ||
515 state == AFS_CALL_SV_AWAIT_ACK
517 if (state == AFS_CALL_SV_AWAIT_ACK) {
518 iov_iter_kvec(&call->def_iter, READ, NULL, 0, 0);
519 ret = rxrpc_kernel_recv_data(call->net->socket,
520 call->rxcall, &call->def_iter,
521 false, &remote_abort,
523 trace_afs_receive_data(call, &call->def_iter, false, ret);
525 if (ret == -EINPROGRESS || ret == -EAGAIN)
527 if (ret < 0 || ret == 1) {
535 if (!call->have_reply_time &&
536 rxrpc_kernel_get_reply_time(call->net->socket,
539 call->have_reply_time = true;
541 ret = call->type->deliver(call);
542 state = READ_ONCE(call->state);
545 afs_queue_call_work(call);
546 if (state == AFS_CALL_CL_PROC_REPLY) {
548 set_bit(AFS_SERVER_FL_MAY_HAVE_CB,
549 &call->cbi->server->flags);
552 ASSERTCMP(state, >, AFS_CALL_CL_PROC_REPLY);
558 ASSERTCMP(state, ==, AFS_CALL_COMPLETE);
561 abort_code = RXGEN_OPCODE;
562 rxrpc_kernel_abort_call(call->net->socket, call->rxcall,
563 abort_code, ret, "KIV");
566 pr_err("kAFS: Call %u in bad state %u\n",
567 call->debug_id, state);
572 abort_code = RXGEN_CC_UNMARSHAL;
573 if (state != AFS_CALL_CL_AWAIT_REPLY)
574 abort_code = RXGEN_SS_UNMARSHAL;
575 rxrpc_kernel_abort_call(call->net->socket, call->rxcall,
576 abort_code, ret, "KUM");
579 abort_code = RX_USER_ABORT;
580 rxrpc_kernel_abort_call(call->net->socket, call->rxcall,
581 abort_code, ret, "KER");
587 if (call->type->done)
588 call->type->done(call);
596 afs_set_call_complete(call, ret, remote_abort);
597 state = AFS_CALL_COMPLETE;
602 * Wait synchronously for a call to complete and clean up the call struct.
604 long afs_wait_for_call_to_complete(struct afs_call *call,
605 struct afs_addr_cursor *ac)
608 bool rxrpc_complete = false;
610 DECLARE_WAITQUEUE(myself, current);
618 add_wait_queue(&call->waitq, &myself);
620 set_current_state(TASK_UNINTERRUPTIBLE);
622 /* deliver any messages that are in the queue */
623 if (!afs_check_call_state(call, AFS_CALL_COMPLETE) &&
624 call->need_attention) {
625 call->need_attention = false;
626 __set_current_state(TASK_RUNNING);
627 afs_deliver_to_call(call);
631 if (afs_check_call_state(call, AFS_CALL_COMPLETE))
634 if (!rxrpc_kernel_check_life(call->net->socket, call->rxcall)) {
635 /* rxrpc terminated the call. */
636 rxrpc_complete = true;
643 remove_wait_queue(&call->waitq, &myself);
644 __set_current_state(TASK_RUNNING);
646 if (!afs_check_call_state(call, AFS_CALL_COMPLETE)) {
647 if (rxrpc_complete) {
648 afs_set_call_complete(call, call->error, call->abort_code);
650 /* Kill off the call if it's still live. */
651 _debug("call interrupted");
652 if (rxrpc_kernel_abort_call(call->net->socket, call->rxcall,
653 RX_USER_ABORT, -EINTR, "KWI"))
654 afs_set_call_complete(call, -EINTR, 0);
658 spin_lock_bh(&call->state_lock);
659 ac->abort_code = call->abort_code;
660 ac->error = call->error;
661 spin_unlock_bh(&call->state_lock);
671 ac->responded = true;
676 _debug("call complete");
678 _leave(" = %p", (void *)ret);
683 * wake up a waiting call
685 static void afs_wake_up_call_waiter(struct sock *sk, struct rxrpc_call *rxcall,
686 unsigned long call_user_ID)
688 struct afs_call *call = (struct afs_call *)call_user_ID;
690 call->need_attention = true;
691 wake_up(&call->waitq);
695 * wake up an asynchronous call
697 static void afs_wake_up_async_call(struct sock *sk, struct rxrpc_call *rxcall,
698 unsigned long call_user_ID)
700 struct afs_call *call = (struct afs_call *)call_user_ID;
703 trace_afs_notify_call(rxcall, call);
704 call->need_attention = true;
706 u = atomic_fetch_add_unless(&call->usage, 1, 0);
708 trace_afs_call(call, afs_call_trace_wake, u + 1,
709 atomic_read(&call->net->nr_outstanding_calls),
710 __builtin_return_address(0));
712 if (!queue_work(afs_async_calls, &call->async_work))
718 * Perform I/O processing on an asynchronous call. The work item carries a ref
719 * to the call struct that we either need to release or to pass on.
721 static void afs_process_async_call(struct work_struct *work)
723 struct afs_call *call = container_of(work, struct afs_call, async_work);
727 if (call->state < AFS_CALL_COMPLETE && call->need_attention) {
728 call->need_attention = false;
729 afs_deliver_to_call(call);
736 static void afs_rx_attach(struct rxrpc_call *rxcall, unsigned long user_call_ID)
738 struct afs_call *call = (struct afs_call *)user_call_ID;
740 call->rxcall = rxcall;
744 * Charge the incoming call preallocation.
746 void afs_charge_preallocation(struct work_struct *work)
748 struct afs_net *net =
749 container_of(work, struct afs_net, charge_preallocation_work);
750 struct afs_call *call = net->spare_incoming_call;
754 call = afs_alloc_call(net, &afs_RXCMxxxx, GFP_KERNEL);
758 call->drop_ref = true;
760 call->state = AFS_CALL_SV_AWAIT_OP_ID;
761 init_waitqueue_head(&call->waitq);
762 afs_extract_to_tmp(call);
765 if (rxrpc_kernel_charge_accept(net->socket,
766 afs_wake_up_async_call,
774 net->spare_incoming_call = call;
778 * Discard a preallocated call when a socket is shut down.
780 static void afs_rx_discard_new_call(struct rxrpc_call *rxcall,
781 unsigned long user_call_ID)
783 struct afs_call *call = (struct afs_call *)user_call_ID;
790 * Notification of an incoming call.
792 static void afs_rx_new_call(struct sock *sk, struct rxrpc_call *rxcall,
793 unsigned long user_call_ID)
795 struct afs_net *net = afs_sock2net(sk);
797 queue_work(afs_wq, &net->charge_preallocation_work);
801 * Grab the operation ID from an incoming cache manager call. The socket
802 * buffer is discarded on error or if we don't yet have sufficient data.
804 static int afs_deliver_cm_op_id(struct afs_call *call)
808 _enter("{%zu}", iov_iter_count(call->iter));
810 /* the operation ID forms the first four bytes of the request data */
811 ret = afs_extract_data(call, true);
815 call->operation_ID = ntohl(call->tmp);
816 afs_set_call_state(call, AFS_CALL_SV_AWAIT_OP_ID, AFS_CALL_SV_AWAIT_REQUEST);
818 /* ask the cache manager to route the call (it'll change the call type
820 if (!afs_cm_incoming_call(call))
823 trace_afs_cb_call(call);
825 /* pass responsibility for the remainer of this message off to the
826 * cache manager op */
827 return call->type->deliver(call);
831 * Advance the AFS call state when an RxRPC service call ends the transmit
834 static void afs_notify_end_reply_tx(struct sock *sock,
835 struct rxrpc_call *rxcall,
836 unsigned long call_user_ID)
838 struct afs_call *call = (struct afs_call *)call_user_ID;
840 afs_set_call_state(call, AFS_CALL_SV_REPLYING, AFS_CALL_SV_AWAIT_ACK);
844 * send an empty reply
846 void afs_send_empty_reply(struct afs_call *call)
848 struct afs_net *net = call->net;
853 rxrpc_kernel_set_tx_length(net->socket, call->rxcall, 0);
857 iov_iter_kvec(&msg.msg_iter, WRITE, NULL, 0, 0);
858 msg.msg_control = NULL;
859 msg.msg_controllen = 0;
862 switch (rxrpc_kernel_send_data(net->socket, call->rxcall, &msg, 0,
863 afs_notify_end_reply_tx)) {
865 _leave(" [replied]");
870 rxrpc_kernel_abort_call(net->socket, call->rxcall,
871 RX_USER_ABORT, -ENOMEM, "KOO");
880 * send a simple reply
882 void afs_send_simple_reply(struct afs_call *call, const void *buf, size_t len)
884 struct afs_net *net = call->net;
891 rxrpc_kernel_set_tx_length(net->socket, call->rxcall, len);
893 iov[0].iov_base = (void *) buf;
894 iov[0].iov_len = len;
897 iov_iter_kvec(&msg.msg_iter, WRITE, iov, 1, len);
898 msg.msg_control = NULL;
899 msg.msg_controllen = 0;
902 n = rxrpc_kernel_send_data(net->socket, call->rxcall, &msg, len,
903 afs_notify_end_reply_tx);
906 _leave(" [replied]");
912 rxrpc_kernel_abort_call(net->socket, call->rxcall,
913 RX_USER_ABORT, -ENOMEM, "KOO");
919 * Extract a piece of data from the received data socket buffers.
921 int afs_extract_data(struct afs_call *call, bool want_more)
923 struct afs_net *net = call->net;
924 struct iov_iter *iter = call->iter;
925 enum afs_call_state state;
926 u32 remote_abort = 0;
929 _enter("{%s,%zu},%d", call->type->name, iov_iter_count(iter), want_more);
931 ret = rxrpc_kernel_recv_data(net->socket, call->rxcall, iter,
932 want_more, &remote_abort,
934 if (ret == 0 || ret == -EAGAIN)
937 state = READ_ONCE(call->state);
940 case AFS_CALL_CL_AWAIT_REPLY:
941 afs_set_call_state(call, state, AFS_CALL_CL_PROC_REPLY);
943 case AFS_CALL_SV_AWAIT_REQUEST:
944 afs_set_call_state(call, state, AFS_CALL_SV_REPLYING);
946 case AFS_CALL_COMPLETE:
947 kdebug("prem complete %d", call->error);
948 return afs_io_error(call, afs_io_error_extract);
955 afs_set_call_complete(call, ret, remote_abort);
960 * Log protocol error production.
962 noinline int afs_protocol_error(struct afs_call *call, int error,
963 enum afs_eproto_cause cause)
965 trace_afs_protocol_error(call, error, cause);