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 <linux/sched/signal.h>
16 #include <net/af_rxrpc.h>
20 struct socket
*afs_socket
; /* my RxRPC socket */
21 static struct workqueue_struct
*afs_async_calls
;
22 static struct afs_call
*afs_spare_incoming_call
;
23 atomic_t afs_outstanding_calls
;
25 static void afs_wake_up_call_waiter(struct sock
*, struct rxrpc_call
*, unsigned long);
26 static int afs_wait_for_call_to_complete(struct afs_call
*);
27 static void afs_wake_up_async_call(struct sock
*, struct rxrpc_call
*, unsigned long);
28 static void afs_process_async_call(struct work_struct
*);
29 static void afs_rx_new_call(struct sock
*, struct rxrpc_call
*, unsigned long);
30 static void afs_rx_discard_new_call(struct rxrpc_call
*, unsigned long);
31 static int afs_deliver_cm_op_id(struct afs_call
*);
33 /* asynchronous incoming call initial processing */
34 static const struct afs_call_type afs_RXCMxxxx
= {
36 .deliver
= afs_deliver_cm_op_id
,
37 .abort_to_error
= afs_abort_to_error
,
40 static void afs_charge_preallocation(struct work_struct
*);
42 static DECLARE_WORK(afs_charge_preallocation_work
, afs_charge_preallocation
);
44 static int afs_wait_atomic_t(atomic_t
*p
)
51 * open an RxRPC socket and bind it to be a server for callback notifications
52 * - the socket is left in blocking mode and non-blocking ops use MSG_DONTWAIT
54 int afs_open_socket(void)
56 struct sockaddr_rxrpc srx
;
57 struct socket
*socket
;
63 afs_async_calls
= alloc_workqueue("kafsd", WQ_MEM_RECLAIM
, 0);
67 ret
= sock_create_kern(&init_net
, AF_RXRPC
, SOCK_DGRAM
, PF_INET
, &socket
);
71 socket
->sk
->sk_allocation
= GFP_NOFS
;
73 /* bind the callback manager's address to make this a server socket */
74 srx
.srx_family
= AF_RXRPC
;
75 srx
.srx_service
= CM_SERVICE
;
76 srx
.transport_type
= SOCK_DGRAM
;
77 srx
.transport_len
= sizeof(srx
.transport
.sin
);
78 srx
.transport
.sin
.sin_family
= AF_INET
;
79 srx
.transport
.sin
.sin_port
= htons(AFS_CM_PORT
);
80 memset(&srx
.transport
.sin
.sin_addr
, 0,
81 sizeof(srx
.transport
.sin
.sin_addr
));
83 ret
= kernel_bind(socket
, (struct sockaddr
*) &srx
, sizeof(srx
));
87 rxrpc_kernel_new_call_notification(socket
, afs_rx_new_call
,
88 afs_rx_discard_new_call
);
90 ret
= kernel_listen(socket
, INT_MAX
);
95 afs_charge_preallocation(NULL
);
100 sock_release(socket
);
102 destroy_workqueue(afs_async_calls
);
104 _leave(" = %d", ret
);
109 * close the RxRPC socket AFS was using
111 void afs_close_socket(void)
115 kernel_listen(afs_socket
, 0);
116 flush_workqueue(afs_async_calls
);
118 if (afs_spare_incoming_call
) {
119 afs_put_call(afs_spare_incoming_call
);
120 afs_spare_incoming_call
= NULL
;
123 _debug("outstanding %u", atomic_read(&afs_outstanding_calls
));
124 wait_on_atomic_t(&afs_outstanding_calls
, afs_wait_atomic_t
,
125 TASK_UNINTERRUPTIBLE
);
126 _debug("no outstanding calls");
128 kernel_sock_shutdown(afs_socket
, SHUT_RDWR
);
129 flush_workqueue(afs_async_calls
);
130 sock_release(afs_socket
);
133 destroy_workqueue(afs_async_calls
);
140 static struct afs_call
*afs_alloc_call(const struct afs_call_type
*type
,
143 struct afs_call
*call
;
146 call
= kzalloc(sizeof(*call
), gfp
);
151 atomic_set(&call
->usage
, 1);
152 INIT_WORK(&call
->async_work
, afs_process_async_call
);
153 init_waitqueue_head(&call
->waitq
);
155 o
= atomic_inc_return(&afs_outstanding_calls
);
156 trace_afs_call(call
, afs_call_trace_alloc
, 1, o
,
157 __builtin_return_address(0));
162 * Dispose of a reference on a call.
164 void afs_put_call(struct afs_call
*call
)
166 int n
= atomic_dec_return(&call
->usage
);
167 int o
= atomic_read(&afs_outstanding_calls
);
169 trace_afs_call(call
, afs_call_trace_put
, n
+ 1, o
,
170 __builtin_return_address(0));
174 ASSERT(!work_pending(&call
->async_work
));
175 ASSERT(call
->type
->name
!= NULL
);
178 rxrpc_kernel_end_call(afs_socket
, call
->rxcall
);
181 if (call
->type
->destructor
)
182 call
->type
->destructor(call
);
184 kfree(call
->request
);
187 o
= atomic_dec_return(&afs_outstanding_calls
);
188 trace_afs_call(call
, afs_call_trace_free
, 0, o
,
189 __builtin_return_address(0));
191 wake_up_atomic_t(&afs_outstanding_calls
);
196 * Queue the call for actual work. Returns 0 unconditionally for convenience.
198 int afs_queue_call_work(struct afs_call
*call
)
200 int u
= atomic_inc_return(&call
->usage
);
202 trace_afs_call(call
, afs_call_trace_work
, u
,
203 atomic_read(&afs_outstanding_calls
),
204 __builtin_return_address(0));
206 INIT_WORK(&call
->work
, call
->type
->work
);
208 if (!queue_work(afs_wq
, &call
->work
))
214 * allocate a call with flat request and reply buffers
216 struct afs_call
*afs_alloc_flat_call(const struct afs_call_type
*type
,
217 size_t request_size
, size_t reply_max
)
219 struct afs_call
*call
;
221 call
= afs_alloc_call(type
, GFP_NOFS
);
226 call
->request_size
= request_size
;
227 call
->request
= kmalloc(request_size
, GFP_NOFS
);
233 call
->reply_max
= reply_max
;
234 call
->buffer
= kmalloc(reply_max
, GFP_NOFS
);
239 init_waitqueue_head(&call
->waitq
);
249 * clean up a call with flat buffer
251 void afs_flat_call_destructor(struct afs_call
*call
)
255 kfree(call
->request
);
256 call
->request
= NULL
;
261 #define AFS_BVEC_MAX 8
264 * Load the given bvec with the next few pages.
266 static void afs_load_bvec(struct afs_call
*call
, struct msghdr
*msg
,
267 struct bio_vec
*bv
, pgoff_t first
, pgoff_t last
,
270 struct page
*pages
[AFS_BVEC_MAX
];
271 unsigned int nr
, n
, i
, to
, bytes
= 0;
273 nr
= min_t(pgoff_t
, last
- first
+ 1, AFS_BVEC_MAX
);
274 n
= find_get_pages_contig(call
->mapping
, first
, nr
, pages
);
275 ASSERTCMP(n
, ==, nr
);
277 msg
->msg_flags
|= MSG_MORE
;
278 for (i
= 0; i
< nr
; i
++) {
280 if (first
+ i
>= last
) {
282 msg
->msg_flags
&= ~MSG_MORE
;
284 bv
[i
].bv_page
= pages
[i
];
285 bv
[i
].bv_len
= to
- offset
;
286 bv
[i
].bv_offset
= offset
;
287 bytes
+= to
- offset
;
291 iov_iter_bvec(&msg
->msg_iter
, WRITE
| ITER_BVEC
, bv
, nr
, bytes
);
295 * Advance the AFS call state when the RxRPC call ends the transmit phase.
297 static void afs_notify_end_request_tx(struct sock
*sock
,
298 struct rxrpc_call
*rxcall
,
299 unsigned long call_user_ID
)
301 struct afs_call
*call
= (struct afs_call
*)call_user_ID
;
303 if (call
->state
== AFS_CALL_REQUESTING
)
304 call
->state
= AFS_CALL_AWAIT_REPLY
;
308 * attach the data from a bunch of pages on an inode to a call
310 static int afs_send_pages(struct afs_call
*call
, struct msghdr
*msg
)
312 struct bio_vec bv
[AFS_BVEC_MAX
];
313 unsigned int bytes
, nr
, loop
, offset
;
314 pgoff_t first
= call
->first
, last
= call
->last
;
317 offset
= call
->first_offset
;
318 call
->first_offset
= 0;
321 afs_load_bvec(call
, msg
, bv
, first
, last
, offset
);
323 bytes
= msg
->msg_iter
.count
;
324 nr
= msg
->msg_iter
.nr_segs
;
326 ret
= rxrpc_kernel_send_data(afs_socket
, call
->rxcall
, msg
,
327 bytes
, afs_notify_end_request_tx
);
328 for (loop
= 0; loop
< nr
; loop
++)
329 put_page(bv
[loop
].bv_page
);
334 } while (first
<= last
);
342 int afs_make_call(struct in_addr
*addr
, struct afs_call
*call
, gfp_t gfp
,
345 struct sockaddr_rxrpc srx
;
346 struct rxrpc_call
*rxcall
;
354 _enter("%x,{%d},", addr
->s_addr
, ntohs(call
->port
));
356 ASSERT(call
->type
!= NULL
);
357 ASSERT(call
->type
->name
!= NULL
);
359 _debug("____MAKE %p{%s,%x} [%d]____",
360 call
, call
->type
->name
, key_serial(call
->key
),
361 atomic_read(&afs_outstanding_calls
));
365 memset(&srx
, 0, sizeof(srx
));
366 srx
.srx_family
= AF_RXRPC
;
367 srx
.srx_service
= call
->service_id
;
368 srx
.transport_type
= SOCK_DGRAM
;
369 srx
.transport_len
= sizeof(srx
.transport
.sin
);
370 srx
.transport
.sin
.sin_family
= AF_INET
;
371 srx
.transport
.sin
.sin_port
= call
->port
;
372 memcpy(&srx
.transport
.sin
.sin_addr
, addr
, 4);
374 /* Work out the length we're going to transmit. This is awkward for
375 * calls such as FS.StoreData where there's an extra injection of data
376 * after the initial fixed part.
378 tx_total_len
= call
->request_size
;
379 if (call
->send_pages
) {
380 tx_total_len
+= call
->last_to
- call
->first_offset
;
381 tx_total_len
+= (call
->last
- call
->first
) * PAGE_SIZE
;
385 rxcall
= rxrpc_kernel_begin_call(afs_socket
, &srx
, call
->key
,
389 afs_wake_up_async_call
:
390 afs_wake_up_call_waiter
));
392 if (IS_ERR(rxcall
)) {
393 ret
= PTR_ERR(rxcall
);
394 goto error_kill_call
;
397 call
->rxcall
= rxcall
;
399 /* send the request */
400 iov
[0].iov_base
= call
->request
;
401 iov
[0].iov_len
= call
->request_size
;
405 iov_iter_kvec(&msg
.msg_iter
, WRITE
| ITER_KVEC
, iov
, 1,
407 msg
.msg_control
= NULL
;
408 msg
.msg_controllen
= 0;
409 msg
.msg_flags
= (call
->send_pages
? MSG_MORE
: 0);
411 /* We have to change the state *before* sending the last packet as
412 * rxrpc might give us the reply before it returns from sending the
413 * request. Further, if the send fails, we may already have been given
414 * a notification and may have collected it.
416 if (!call
->send_pages
)
417 call
->state
= AFS_CALL_AWAIT_REPLY
;
418 ret
= rxrpc_kernel_send_data(afs_socket
, rxcall
,
419 &msg
, call
->request_size
,
420 afs_notify_end_request_tx
);
424 if (call
->send_pages
) {
425 ret
= afs_send_pages(call
, &msg
);
430 /* at this point, an async call may no longer exist as it may have
431 * already completed */
435 return afs_wait_for_call_to_complete(call
);
438 call
->state
= AFS_CALL_COMPLETE
;
439 if (ret
!= -ECONNABORTED
) {
440 rxrpc_kernel_abort_call(afs_socket
, rxcall
, RX_USER_ABORT
,
445 rxrpc_kernel_recv_data(afs_socket
, rxcall
, NULL
, 0, &offset
,
447 ret
= call
->type
->abort_to_error(abort_code
);
451 _leave(" = %d", ret
);
456 * deliver messages to a call
458 static void afs_deliver_to_call(struct afs_call
*call
)
463 _enter("%s", call
->type
->name
);
465 while (call
->state
== AFS_CALL_AWAIT_REPLY
||
466 call
->state
== AFS_CALL_AWAIT_OP_ID
||
467 call
->state
== AFS_CALL_AWAIT_REQUEST
||
468 call
->state
== AFS_CALL_AWAIT_ACK
470 if (call
->state
== AFS_CALL_AWAIT_ACK
) {
472 ret
= rxrpc_kernel_recv_data(afs_socket
, call
->rxcall
,
473 NULL
, 0, &offset
, false,
475 trace_afs_recv_data(call
, 0, offset
, false, ret
);
477 if (ret
== -EINPROGRESS
|| ret
== -EAGAIN
)
479 if (ret
== 1 || ret
< 0) {
480 call
->state
= AFS_CALL_COMPLETE
;
486 ret
= call
->type
->deliver(call
);
489 if (call
->state
== AFS_CALL_AWAIT_REPLY
)
490 call
->state
= AFS_CALL_COMPLETE
;
498 abort_code
= RX_CALL_DEAD
;
499 rxrpc_kernel_abort_call(afs_socket
, call
->rxcall
,
500 abort_code
, ret
, "KNC");
503 abort_code
= RXGEN_OPCODE
;
504 rxrpc_kernel_abort_call(afs_socket
, call
->rxcall
,
505 abort_code
, ret
, "KIV");
511 abort_code
= RXGEN_CC_UNMARSHAL
;
512 if (call
->state
!= AFS_CALL_AWAIT_REPLY
)
513 abort_code
= RXGEN_SS_UNMARSHAL
;
514 rxrpc_kernel_abort_call(afs_socket
, call
->rxcall
,
515 abort_code
, -EBADMSG
, "KUM");
521 if (call
->state
== AFS_CALL_COMPLETE
&& call
->incoming
)
530 call
->state
= AFS_CALL_COMPLETE
;
535 * wait synchronously for a call to complete
537 static int afs_wait_for_call_to_complete(struct afs_call
*call
)
541 DECLARE_WAITQUEUE(myself
, current
);
545 add_wait_queue(&call
->waitq
, &myself
);
547 set_current_state(TASK_INTERRUPTIBLE
);
549 /* deliver any messages that are in the queue */
550 if (call
->state
< AFS_CALL_COMPLETE
&& call
->need_attention
) {
551 call
->need_attention
= false;
552 __set_current_state(TASK_RUNNING
);
553 afs_deliver_to_call(call
);
557 if (call
->state
== AFS_CALL_COMPLETE
||
558 signal_pending(current
))
563 remove_wait_queue(&call
->waitq
, &myself
);
564 __set_current_state(TASK_RUNNING
);
566 /* Kill off the call if it's still live. */
567 if (call
->state
< AFS_CALL_COMPLETE
) {
568 _debug("call interrupted");
569 rxrpc_kernel_abort_call(afs_socket
, call
->rxcall
,
570 RX_USER_ABORT
, -EINTR
, "KWI");
574 _debug("call complete");
576 _leave(" = %d", ret
);
581 * wake up a waiting call
583 static void afs_wake_up_call_waiter(struct sock
*sk
, struct rxrpc_call
*rxcall
,
584 unsigned long call_user_ID
)
586 struct afs_call
*call
= (struct afs_call
*)call_user_ID
;
588 call
->need_attention
= true;
589 wake_up(&call
->waitq
);
593 * wake up an asynchronous call
595 static void afs_wake_up_async_call(struct sock
*sk
, struct rxrpc_call
*rxcall
,
596 unsigned long call_user_ID
)
598 struct afs_call
*call
= (struct afs_call
*)call_user_ID
;
601 trace_afs_notify_call(rxcall
, call
);
602 call
->need_attention
= true;
604 u
= __atomic_add_unless(&call
->usage
, 1, 0);
606 trace_afs_call(call
, afs_call_trace_wake
, u
,
607 atomic_read(&afs_outstanding_calls
),
608 __builtin_return_address(0));
610 if (!queue_work(afs_async_calls
, &call
->async_work
))
616 * Delete an asynchronous call. The work item carries a ref to the call struct
617 * that we need to release.
619 static void afs_delete_async_call(struct work_struct
*work
)
621 struct afs_call
*call
= container_of(work
, struct afs_call
, async_work
);
631 * Perform I/O processing on an asynchronous call. The work item carries a ref
632 * to the call struct that we either need to release or to pass on.
634 static void afs_process_async_call(struct work_struct
*work
)
636 struct afs_call
*call
= container_of(work
, struct afs_call
, async_work
);
640 if (call
->state
< AFS_CALL_COMPLETE
&& call
->need_attention
) {
641 call
->need_attention
= false;
642 afs_deliver_to_call(call
);
645 if (call
->state
== AFS_CALL_COMPLETE
) {
648 /* We have two refs to release - one from the alloc and one
649 * queued with the work item - and we can't just deallocate the
650 * call because the work item may be queued again.
652 call
->async_work
.func
= afs_delete_async_call
;
653 if (!queue_work(afs_async_calls
, &call
->async_work
))
661 static void afs_rx_attach(struct rxrpc_call
*rxcall
, unsigned long user_call_ID
)
663 struct afs_call
*call
= (struct afs_call
*)user_call_ID
;
665 call
->rxcall
= rxcall
;
669 * Charge the incoming call preallocation.
671 static void afs_charge_preallocation(struct work_struct
*work
)
673 struct afs_call
*call
= afs_spare_incoming_call
;
677 call
= afs_alloc_call(&afs_RXCMxxxx
, GFP_KERNEL
);
682 call
->state
= AFS_CALL_AWAIT_OP_ID
;
683 init_waitqueue_head(&call
->waitq
);
686 if (rxrpc_kernel_charge_accept(afs_socket
,
687 afs_wake_up_async_call
,
694 afs_spare_incoming_call
= call
;
698 * Discard a preallocated call when a socket is shut down.
700 static void afs_rx_discard_new_call(struct rxrpc_call
*rxcall
,
701 unsigned long user_call_ID
)
703 struct afs_call
*call
= (struct afs_call
*)user_call_ID
;
710 * Notification of an incoming call.
712 static void afs_rx_new_call(struct sock
*sk
, struct rxrpc_call
*rxcall
,
713 unsigned long user_call_ID
)
715 queue_work(afs_wq
, &afs_charge_preallocation_work
);
719 * Grab the operation ID from an incoming cache manager call. The socket
720 * buffer is discarded on error or if we don't yet have sufficient data.
722 static int afs_deliver_cm_op_id(struct afs_call
*call
)
726 _enter("{%zu}", call
->offset
);
728 ASSERTCMP(call
->offset
, <, 4);
730 /* the operation ID forms the first four bytes of the request data */
731 ret
= afs_extract_data(call
, &call
->tmp
, 4, true);
735 call
->operation_ID
= ntohl(call
->tmp
);
736 call
->state
= AFS_CALL_AWAIT_REQUEST
;
739 /* ask the cache manager to route the call (it'll change the call type
741 if (!afs_cm_incoming_call(call
))
744 trace_afs_cb_call(call
);
746 /* pass responsibility for the remainer of this message off to the
747 * cache manager op */
748 return call
->type
->deliver(call
);
752 * Advance the AFS call state when an RxRPC service call ends the transmit
755 static void afs_notify_end_reply_tx(struct sock
*sock
,
756 struct rxrpc_call
*rxcall
,
757 unsigned long call_user_ID
)
759 struct afs_call
*call
= (struct afs_call
*)call_user_ID
;
761 if (call
->state
== AFS_CALL_REPLYING
)
762 call
->state
= AFS_CALL_AWAIT_ACK
;
766 * send an empty reply
768 void afs_send_empty_reply(struct afs_call
*call
)
774 rxrpc_kernel_set_tx_length(afs_socket
, call
->rxcall
, 0);
778 iov_iter_kvec(&msg
.msg_iter
, WRITE
| ITER_KVEC
, NULL
, 0, 0);
779 msg
.msg_control
= NULL
;
780 msg
.msg_controllen
= 0;
783 call
->state
= AFS_CALL_AWAIT_ACK
;
784 switch (rxrpc_kernel_send_data(afs_socket
, call
->rxcall
, &msg
, 0,
785 afs_notify_end_reply_tx
)) {
787 _leave(" [replied]");
792 rxrpc_kernel_abort_call(afs_socket
, call
->rxcall
,
793 RX_USER_ABORT
, -ENOMEM
, "KOO");
801 * send a simple reply
803 void afs_send_simple_reply(struct afs_call
*call
, const void *buf
, size_t len
)
811 rxrpc_kernel_set_tx_length(afs_socket
, call
->rxcall
, len
);
813 iov
[0].iov_base
= (void *) buf
;
814 iov
[0].iov_len
= len
;
817 iov_iter_kvec(&msg
.msg_iter
, WRITE
| ITER_KVEC
, iov
, 1, len
);
818 msg
.msg_control
= NULL
;
819 msg
.msg_controllen
= 0;
822 call
->state
= AFS_CALL_AWAIT_ACK
;
823 n
= rxrpc_kernel_send_data(afs_socket
, call
->rxcall
, &msg
, len
,
824 afs_notify_end_reply_tx
);
827 _leave(" [replied]");
833 rxrpc_kernel_abort_call(afs_socket
, call
->rxcall
,
834 RX_USER_ABORT
, -ENOMEM
, "KOO");
840 * Extract a piece of data from the received data socket buffers.
842 int afs_extract_data(struct afs_call
*call
, void *buf
, size_t count
,
847 _enter("{%s,%zu},,%zu,%d",
848 call
->type
->name
, call
->offset
, count
, want_more
);
850 ASSERTCMP(call
->offset
, <=, count
);
852 ret
= rxrpc_kernel_recv_data(afs_socket
, call
->rxcall
,
853 buf
, count
, &call
->offset
,
854 want_more
, &call
->abort_code
);
855 trace_afs_recv_data(call
, count
, call
->offset
, want_more
, ret
);
856 if (ret
== 0 || ret
== -EAGAIN
)
860 switch (call
->state
) {
861 case AFS_CALL_AWAIT_REPLY
:
862 call
->state
= AFS_CALL_COMPLETE
;
864 case AFS_CALL_AWAIT_REQUEST
:
865 call
->state
= AFS_CALL_REPLYING
;
873 if (ret
== -ECONNABORTED
)
874 call
->error
= call
->type
->abort_to_error(call
->abort_code
);
877 call
->state
= AFS_CALL_COMPLETE
;