2 * linux/net/sunrpc/xprtsock.c
4 * Client-side transport implementation for sockets.
6 * TCP callback races fixes (C) 1998 Red Hat
7 * TCP send fixes (C) 1998 Red Hat
8 * TCP NFS related read + write fixes
9 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
11 * Rewrite of larges part of the code in order to stabilize TCP stuff.
12 * Fix behaviour when socket buffer is full.
13 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
15 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
17 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
18 * <gilles.quillard@bull.net>
21 #include <linux/types.h>
22 #include <linux/string.h>
23 #include <linux/slab.h>
24 #include <linux/module.h>
25 #include <linux/capability.h>
26 #include <linux/pagemap.h>
27 #include <linux/errno.h>
28 #include <linux/socket.h>
30 #include <linux/net.h>
33 #include <linux/udp.h>
34 #include <linux/tcp.h>
35 #include <linux/sunrpc/clnt.h>
36 #include <linux/sunrpc/addr.h>
37 #include <linux/sunrpc/sched.h>
38 #include <linux/sunrpc/svcsock.h>
39 #include <linux/sunrpc/xprtsock.h>
40 #include <linux/file.h>
41 #ifdef CONFIG_SUNRPC_BACKCHANNEL
42 #include <linux/sunrpc/bc_xprt.h>
46 #include <net/checksum.h>
50 #include <trace/events/sunrpc.h>
54 static void xs_close(struct rpc_xprt
*xprt
);
59 static unsigned int xprt_udp_slot_table_entries
= RPC_DEF_SLOT_TABLE
;
60 static unsigned int xprt_tcp_slot_table_entries
= RPC_MIN_SLOT_TABLE
;
61 static unsigned int xprt_max_tcp_slot_table_entries
= RPC_MAX_SLOT_TABLE
;
63 static unsigned int xprt_min_resvport
= RPC_DEF_MIN_RESVPORT
;
64 static unsigned int xprt_max_resvport
= RPC_DEF_MAX_RESVPORT
;
66 #define XS_TCP_LINGER_TO (15U * HZ)
67 static unsigned int xs_tcp_fin_timeout __read_mostly
= XS_TCP_LINGER_TO
;
70 * We can register our own files under /proc/sys/sunrpc by
71 * calling register_sysctl_table() again. The files in that
72 * directory become the union of all files registered there.
74 * We simply need to make sure that we don't collide with
75 * someone else's file names!
80 static unsigned int min_slot_table_size
= RPC_MIN_SLOT_TABLE
;
81 static unsigned int max_slot_table_size
= RPC_MAX_SLOT_TABLE
;
82 static unsigned int max_tcp_slot_table_limit
= RPC_MAX_SLOT_TABLE_LIMIT
;
83 static unsigned int xprt_min_resvport_limit
= RPC_MIN_RESVPORT
;
84 static unsigned int xprt_max_resvport_limit
= RPC_MAX_RESVPORT
;
86 static struct ctl_table_header
*sunrpc_table_header
;
89 * FIXME: changing the UDP slot table size should also resize the UDP
90 * socket buffers for existing UDP transports
92 static struct ctl_table xs_tunables_table
[] = {
94 .procname
= "udp_slot_table_entries",
95 .data
= &xprt_udp_slot_table_entries
,
96 .maxlen
= sizeof(unsigned int),
98 .proc_handler
= proc_dointvec_minmax
,
99 .extra1
= &min_slot_table_size
,
100 .extra2
= &max_slot_table_size
103 .procname
= "tcp_slot_table_entries",
104 .data
= &xprt_tcp_slot_table_entries
,
105 .maxlen
= sizeof(unsigned int),
107 .proc_handler
= proc_dointvec_minmax
,
108 .extra1
= &min_slot_table_size
,
109 .extra2
= &max_slot_table_size
112 .procname
= "tcp_max_slot_table_entries",
113 .data
= &xprt_max_tcp_slot_table_entries
,
114 .maxlen
= sizeof(unsigned int),
116 .proc_handler
= proc_dointvec_minmax
,
117 .extra1
= &min_slot_table_size
,
118 .extra2
= &max_tcp_slot_table_limit
121 .procname
= "min_resvport",
122 .data
= &xprt_min_resvport
,
123 .maxlen
= sizeof(unsigned int),
125 .proc_handler
= proc_dointvec_minmax
,
126 .extra1
= &xprt_min_resvport_limit
,
127 .extra2
= &xprt_max_resvport_limit
130 .procname
= "max_resvport",
131 .data
= &xprt_max_resvport
,
132 .maxlen
= sizeof(unsigned int),
134 .proc_handler
= proc_dointvec_minmax
,
135 .extra1
= &xprt_min_resvport_limit
,
136 .extra2
= &xprt_max_resvport_limit
139 .procname
= "tcp_fin_timeout",
140 .data
= &xs_tcp_fin_timeout
,
141 .maxlen
= sizeof(xs_tcp_fin_timeout
),
143 .proc_handler
= proc_dointvec_jiffies
,
148 static struct ctl_table sunrpc_table
[] = {
150 .procname
= "sunrpc",
152 .child
= xs_tunables_table
160 * Wait duration for a reply from the RPC portmapper.
162 #define XS_BIND_TO (60U * HZ)
165 * Delay if a UDP socket connect error occurs. This is most likely some
166 * kind of resource problem on the local host.
168 #define XS_UDP_REEST_TO (2U * HZ)
171 * The reestablish timeout allows clients to delay for a bit before attempting
172 * to reconnect to a server that just dropped our connection.
174 * We implement an exponential backoff when trying to reestablish a TCP
175 * transport connection with the server. Some servers like to drop a TCP
176 * connection when they are overworked, so we start with a short timeout and
177 * increase over time if the server is down or not responding.
179 #define XS_TCP_INIT_REEST_TO (3U * HZ)
180 #define XS_TCP_MAX_REEST_TO (5U * 60 * HZ)
183 * TCP idle timeout; client drops the transport socket if it is idle
184 * for this long. Note that we also timeout UDP sockets to prevent
185 * holding port numbers when there is no RPC traffic.
187 #define XS_IDLE_DISC_TO (5U * 60 * HZ)
190 # undef RPC_DEBUG_DATA
191 # define RPCDBG_FACILITY RPCDBG_TRANS
194 #ifdef RPC_DEBUG_DATA
195 static void xs_pktdump(char *msg
, u32
*packet
, unsigned int count
)
197 u8
*buf
= (u8
*) packet
;
200 dprintk("RPC: %s\n", msg
);
201 for (j
= 0; j
< count
&& j
< 128; j
+= 4) {
205 dprintk("0x%04x ", j
);
207 dprintk("%02x%02x%02x%02x ",
208 buf
[j
], buf
[j
+1], buf
[j
+2], buf
[j
+3]);
213 static inline void xs_pktdump(char *msg
, u32
*packet
, unsigned int count
)
220 struct rpc_xprt xprt
;
225 struct socket
* sock
;
229 * State of TCP reply receive
238 unsigned long tcp_copied
,
242 * Connection of transports
244 struct delayed_work connect_worker
;
245 struct sockaddr_storage srcaddr
;
246 unsigned short srcport
;
249 * UDP socket buffer size parameters
255 * Saved socket callback addresses
257 void (*old_data_ready
)(struct sock
*, int);
258 void (*old_state_change
)(struct sock
*);
259 void (*old_write_space
)(struct sock
*);
263 * TCP receive state flags
265 #define TCP_RCV_LAST_FRAG (1UL << 0)
266 #define TCP_RCV_COPY_FRAGHDR (1UL << 1)
267 #define TCP_RCV_COPY_XID (1UL << 2)
268 #define TCP_RCV_COPY_DATA (1UL << 3)
269 #define TCP_RCV_READ_CALLDIR (1UL << 4)
270 #define TCP_RCV_COPY_CALLDIR (1UL << 5)
275 #define TCP_RPC_REPLY (1UL << 6)
277 static inline struct sockaddr
*xs_addr(struct rpc_xprt
*xprt
)
279 return (struct sockaddr
*) &xprt
->addr
;
282 static inline struct sockaddr_un
*xs_addr_un(struct rpc_xprt
*xprt
)
284 return (struct sockaddr_un
*) &xprt
->addr
;
287 static inline struct sockaddr_in
*xs_addr_in(struct rpc_xprt
*xprt
)
289 return (struct sockaddr_in
*) &xprt
->addr
;
292 static inline struct sockaddr_in6
*xs_addr_in6(struct rpc_xprt
*xprt
)
294 return (struct sockaddr_in6
*) &xprt
->addr
;
297 static void xs_format_common_peer_addresses(struct rpc_xprt
*xprt
)
299 struct sockaddr
*sap
= xs_addr(xprt
);
300 struct sockaddr_in6
*sin6
;
301 struct sockaddr_in
*sin
;
302 struct sockaddr_un
*sun
;
305 switch (sap
->sa_family
) {
307 sun
= xs_addr_un(xprt
);
308 strlcpy(buf
, sun
->sun_path
, sizeof(buf
));
309 xprt
->address_strings
[RPC_DISPLAY_ADDR
] =
310 kstrdup(buf
, GFP_KERNEL
);
313 (void)rpc_ntop(sap
, buf
, sizeof(buf
));
314 xprt
->address_strings
[RPC_DISPLAY_ADDR
] =
315 kstrdup(buf
, GFP_KERNEL
);
316 sin
= xs_addr_in(xprt
);
317 snprintf(buf
, sizeof(buf
), "%08x", ntohl(sin
->sin_addr
.s_addr
));
320 (void)rpc_ntop(sap
, buf
, sizeof(buf
));
321 xprt
->address_strings
[RPC_DISPLAY_ADDR
] =
322 kstrdup(buf
, GFP_KERNEL
);
323 sin6
= xs_addr_in6(xprt
);
324 snprintf(buf
, sizeof(buf
), "%pi6", &sin6
->sin6_addr
);
330 xprt
->address_strings
[RPC_DISPLAY_HEX_ADDR
] = kstrdup(buf
, GFP_KERNEL
);
333 static void xs_format_common_peer_ports(struct rpc_xprt
*xprt
)
335 struct sockaddr
*sap
= xs_addr(xprt
);
338 snprintf(buf
, sizeof(buf
), "%u", rpc_get_port(sap
));
339 xprt
->address_strings
[RPC_DISPLAY_PORT
] = kstrdup(buf
, GFP_KERNEL
);
341 snprintf(buf
, sizeof(buf
), "%4hx", rpc_get_port(sap
));
342 xprt
->address_strings
[RPC_DISPLAY_HEX_PORT
] = kstrdup(buf
, GFP_KERNEL
);
345 static void xs_format_peer_addresses(struct rpc_xprt
*xprt
,
346 const char *protocol
,
349 xprt
->address_strings
[RPC_DISPLAY_PROTO
] = protocol
;
350 xprt
->address_strings
[RPC_DISPLAY_NETID
] = netid
;
351 xs_format_common_peer_addresses(xprt
);
352 xs_format_common_peer_ports(xprt
);
355 static void xs_update_peer_port(struct rpc_xprt
*xprt
)
357 kfree(xprt
->address_strings
[RPC_DISPLAY_HEX_PORT
]);
358 kfree(xprt
->address_strings
[RPC_DISPLAY_PORT
]);
360 xs_format_common_peer_ports(xprt
);
363 static void xs_free_peer_addresses(struct rpc_xprt
*xprt
)
367 for (i
= 0; i
< RPC_DISPLAY_MAX
; i
++)
369 case RPC_DISPLAY_PROTO
:
370 case RPC_DISPLAY_NETID
:
373 kfree(xprt
->address_strings
[i
]);
377 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
379 static int xs_send_kvec(struct socket
*sock
, struct sockaddr
*addr
, int addrlen
, struct kvec
*vec
, unsigned int base
, int more
)
381 struct msghdr msg
= {
383 .msg_namelen
= addrlen
,
384 .msg_flags
= XS_SENDMSG_FLAGS
| (more
? MSG_MORE
: 0),
387 .iov_base
= vec
->iov_base
+ base
,
388 .iov_len
= vec
->iov_len
- base
,
391 if (iov
.iov_len
!= 0)
392 return kernel_sendmsg(sock
, &msg
, &iov
, 1, iov
.iov_len
);
393 return kernel_sendmsg(sock
, &msg
, NULL
, 0, 0);
396 static int xs_send_pagedata(struct socket
*sock
, struct xdr_buf
*xdr
, unsigned int base
, int more
, bool zerocopy
)
398 ssize_t (*do_sendpage
)(struct socket
*sock
, struct page
*page
,
399 int offset
, size_t size
, int flags
);
401 unsigned int remainder
;
404 remainder
= xdr
->page_len
- base
;
405 base
+= xdr
->page_base
;
406 ppage
= xdr
->pages
+ (base
>> PAGE_SHIFT
);
408 do_sendpage
= sock
->ops
->sendpage
;
410 do_sendpage
= sock_no_sendpage
;
412 unsigned int len
= min_t(unsigned int, PAGE_SIZE
- base
, remainder
);
413 int flags
= XS_SENDMSG_FLAGS
;
416 if (remainder
!= 0 || more
)
418 err
= do_sendpage(sock
, *ppage
, base
, len
, flags
);
419 if (remainder
== 0 || err
!= len
)
433 * xs_sendpages - write pages directly to a socket
434 * @sock: socket to send on
435 * @addr: UDP only -- address of destination
436 * @addrlen: UDP only -- length of destination address
437 * @xdr: buffer containing this request
438 * @base: starting position in the buffer
439 * @zerocopy: true if it is safe to use sendpage()
442 static int xs_sendpages(struct socket
*sock
, struct sockaddr
*addr
, int addrlen
, struct xdr_buf
*xdr
, unsigned int base
, bool zerocopy
)
444 unsigned int remainder
= xdr
->len
- base
;
450 clear_bit(SOCK_ASYNC_NOSPACE
, &sock
->flags
);
456 if (base
< xdr
->head
[0].iov_len
|| addr
!= NULL
) {
457 unsigned int len
= xdr
->head
[0].iov_len
- base
;
459 err
= xs_send_kvec(sock
, addr
, addrlen
, &xdr
->head
[0], base
, remainder
!= 0);
460 if (remainder
== 0 || err
!= len
)
465 base
-= xdr
->head
[0].iov_len
;
467 if (base
< xdr
->page_len
) {
468 unsigned int len
= xdr
->page_len
- base
;
470 err
= xs_send_pagedata(sock
, xdr
, base
, remainder
!= 0, zerocopy
);
471 if (remainder
== 0 || err
!= len
)
476 base
-= xdr
->page_len
;
478 if (base
>= xdr
->tail
[0].iov_len
)
480 err
= xs_send_kvec(sock
, NULL
, 0, &xdr
->tail
[0], base
, 0);
489 static void xs_nospace_callback(struct rpc_task
*task
)
491 struct sock_xprt
*transport
= container_of(task
->tk_rqstp
->rq_xprt
, struct sock_xprt
, xprt
);
493 transport
->inet
->sk_write_pending
--;
494 clear_bit(SOCK_ASYNC_NOSPACE
, &transport
->sock
->flags
);
498 * xs_nospace - place task on wait queue if transmit was incomplete
499 * @task: task to put to sleep
502 static int xs_nospace(struct rpc_task
*task
)
504 struct rpc_rqst
*req
= task
->tk_rqstp
;
505 struct rpc_xprt
*xprt
= req
->rq_xprt
;
506 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
509 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
510 task
->tk_pid
, req
->rq_slen
- req
->rq_bytes_sent
,
513 /* Protect against races with write_space */
514 spin_lock_bh(&xprt
->transport_lock
);
516 /* Don't race with disconnect */
517 if (xprt_connected(xprt
)) {
518 if (test_bit(SOCK_ASYNC_NOSPACE
, &transport
->sock
->flags
)) {
520 * Notify TCP that we're limited by the application
523 set_bit(SOCK_NOSPACE
, &transport
->sock
->flags
);
524 transport
->inet
->sk_write_pending
++;
525 /* ...and wait for more buffer space */
526 xprt_wait_for_buffer_space(task
, xs_nospace_callback
);
529 clear_bit(SOCK_ASYNC_NOSPACE
, &transport
->sock
->flags
);
533 spin_unlock_bh(&xprt
->transport_lock
);
538 * Construct a stream transport record marker in @buf.
540 static inline void xs_encode_stream_record_marker(struct xdr_buf
*buf
)
542 u32 reclen
= buf
->len
- sizeof(rpc_fraghdr
);
543 rpc_fraghdr
*base
= buf
->head
[0].iov_base
;
544 *base
= cpu_to_be32(RPC_LAST_STREAM_FRAGMENT
| reclen
);
548 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
549 * @task: RPC task that manages the state of an RPC request
552 * 0: The request has been sent
553 * EAGAIN: The socket was blocked, please call again later to
554 * complete the request
555 * ENOTCONN: Caller needs to invoke connect logic then call again
556 * other: Some other error occured, the request was not sent
558 static int xs_local_send_request(struct rpc_task
*task
)
560 struct rpc_rqst
*req
= task
->tk_rqstp
;
561 struct rpc_xprt
*xprt
= req
->rq_xprt
;
562 struct sock_xprt
*transport
=
563 container_of(xprt
, struct sock_xprt
, xprt
);
564 struct xdr_buf
*xdr
= &req
->rq_snd_buf
;
567 xs_encode_stream_record_marker(&req
->rq_snd_buf
);
569 xs_pktdump("packet data:",
570 req
->rq_svec
->iov_base
, req
->rq_svec
->iov_len
);
572 status
= xs_sendpages(transport
->sock
, NULL
, 0,
573 xdr
, req
->rq_bytes_sent
, true);
574 dprintk("RPC: %s(%u) = %d\n",
575 __func__
, xdr
->len
- req
->rq_bytes_sent
, status
);
576 if (likely(status
>= 0)) {
577 req
->rq_bytes_sent
+= status
;
578 req
->rq_xmit_bytes_sent
+= status
;
579 if (likely(req
->rq_bytes_sent
>= req
->rq_slen
)) {
580 req
->rq_bytes_sent
= 0;
588 status
= xs_nospace(task
);
591 dprintk("RPC: sendmsg returned unrecognized error %d\n",
602 * xs_udp_send_request - write an RPC request to a UDP socket
603 * @task: address of RPC task that manages the state of an RPC request
606 * 0: The request has been sent
607 * EAGAIN: The socket was blocked, please call again later to
608 * complete the request
609 * ENOTCONN: Caller needs to invoke connect logic then call again
610 * other: Some other error occurred, the request was not sent
612 static int xs_udp_send_request(struct rpc_task
*task
)
614 struct rpc_rqst
*req
= task
->tk_rqstp
;
615 struct rpc_xprt
*xprt
= req
->rq_xprt
;
616 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
617 struct xdr_buf
*xdr
= &req
->rq_snd_buf
;
620 xs_pktdump("packet data:",
621 req
->rq_svec
->iov_base
,
622 req
->rq_svec
->iov_len
);
624 if (!xprt_bound(xprt
))
626 status
= xs_sendpages(transport
->sock
,
629 req
->rq_bytes_sent
, true);
631 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
632 xdr
->len
- req
->rq_bytes_sent
, status
);
635 req
->rq_xmit_bytes_sent
+= status
;
636 if (status
>= req
->rq_slen
)
638 /* Still some bytes left; set up for a retry later. */
645 /* Should we call xs_close() here? */
648 status
= xs_nospace(task
);
651 dprintk("RPC: sendmsg returned unrecognized error %d\n",
656 /* When the server has died, an ICMP port unreachable message
657 * prompts ECONNREFUSED. */
658 clear_bit(SOCK_ASYNC_NOSPACE
, &transport
->sock
->flags
);
665 * xs_tcp_shutdown - gracefully shut down a TCP socket
668 * Initiates a graceful shutdown of the TCP socket by calling the
669 * equivalent of shutdown(SHUT_WR);
671 static void xs_tcp_shutdown(struct rpc_xprt
*xprt
)
673 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
674 struct socket
*sock
= transport
->sock
;
677 kernel_sock_shutdown(sock
, SHUT_WR
);
678 trace_rpc_socket_shutdown(xprt
, sock
);
683 * xs_tcp_send_request - write an RPC request to a TCP socket
684 * @task: address of RPC task that manages the state of an RPC request
687 * 0: The request has been sent
688 * EAGAIN: The socket was blocked, please call again later to
689 * complete the request
690 * ENOTCONN: Caller needs to invoke connect logic then call again
691 * other: Some other error occurred, the request was not sent
693 * XXX: In the case of soft timeouts, should we eventually give up
694 * if sendmsg is not able to make progress?
696 static int xs_tcp_send_request(struct rpc_task
*task
)
698 struct rpc_rqst
*req
= task
->tk_rqstp
;
699 struct rpc_xprt
*xprt
= req
->rq_xprt
;
700 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
701 struct xdr_buf
*xdr
= &req
->rq_snd_buf
;
702 bool zerocopy
= true;
705 xs_encode_stream_record_marker(&req
->rq_snd_buf
);
707 xs_pktdump("packet data:",
708 req
->rq_svec
->iov_base
,
709 req
->rq_svec
->iov_len
);
710 /* Don't use zero copy if this is a resend. If the RPC call
711 * completes while the socket holds a reference to the pages,
712 * then we may end up resending corrupted data.
714 if (task
->tk_flags
& RPC_TASK_SENT
)
717 /* Continue transmitting the packet/record. We must be careful
718 * to cope with writespace callbacks arriving _after_ we have
719 * called sendmsg(). */
721 status
= xs_sendpages(transport
->sock
,
722 NULL
, 0, xdr
, req
->rq_bytes_sent
,
725 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
726 xdr
->len
- req
->rq_bytes_sent
, status
);
728 if (unlikely(status
< 0))
731 /* If we've sent the entire packet, immediately
732 * reset the count of bytes sent. */
733 req
->rq_bytes_sent
+= status
;
734 req
->rq_xmit_bytes_sent
+= status
;
735 if (likely(req
->rq_bytes_sent
>= req
->rq_slen
)) {
736 req
->rq_bytes_sent
= 0;
749 /* Should we call xs_close() here? */
752 status
= xs_nospace(task
);
755 dprintk("RPC: sendmsg returned unrecognized error %d\n",
758 xs_tcp_shutdown(xprt
);
762 clear_bit(SOCK_ASYNC_NOSPACE
, &transport
->sock
->flags
);
769 * xs_tcp_release_xprt - clean up after a tcp transmission
773 * This cleans up if an error causes us to abort the transmission of a request.
774 * In this case, the socket may need to be reset in order to avoid confusing
777 static void xs_tcp_release_xprt(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
779 struct rpc_rqst
*req
;
781 if (task
!= xprt
->snd_task
)
785 req
= task
->tk_rqstp
;
788 if (req
->rq_bytes_sent
== 0)
790 if (req
->rq_bytes_sent
== req
->rq_snd_buf
.len
)
792 set_bit(XPRT_CLOSE_WAIT
, &xprt
->state
);
794 xprt_release_xprt(xprt
, task
);
797 static void xs_save_old_callbacks(struct sock_xprt
*transport
, struct sock
*sk
)
799 transport
->old_data_ready
= sk
->sk_data_ready
;
800 transport
->old_state_change
= sk
->sk_state_change
;
801 transport
->old_write_space
= sk
->sk_write_space
;
804 static void xs_restore_old_callbacks(struct sock_xprt
*transport
, struct sock
*sk
)
806 sk
->sk_data_ready
= transport
->old_data_ready
;
807 sk
->sk_state_change
= transport
->old_state_change
;
808 sk
->sk_write_space
= transport
->old_write_space
;
811 static void xs_reset_transport(struct sock_xprt
*transport
)
813 struct socket
*sock
= transport
->sock
;
814 struct sock
*sk
= transport
->inet
;
819 transport
->srcport
= 0;
821 write_lock_bh(&sk
->sk_callback_lock
);
822 transport
->inet
= NULL
;
823 transport
->sock
= NULL
;
825 sk
->sk_user_data
= NULL
;
827 xs_restore_old_callbacks(transport
, sk
);
828 write_unlock_bh(&sk
->sk_callback_lock
);
832 trace_rpc_socket_close(&transport
->xprt
, sock
);
837 * xs_close - close a socket
840 * This is used when all requests are complete; ie, no DRC state remains
841 * on the server we want to save.
843 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
844 * xs_reset_transport() zeroing the socket from underneath a writer.
846 static void xs_close(struct rpc_xprt
*xprt
)
848 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
850 dprintk("RPC: xs_close xprt %p\n", xprt
);
852 cancel_delayed_work_sync(&transport
->connect_worker
);
854 xs_reset_transport(transport
);
855 xprt
->reestablish_timeout
= 0;
857 smp_mb__before_clear_bit();
858 clear_bit(XPRT_CONNECTION_ABORT
, &xprt
->state
);
859 clear_bit(XPRT_CLOSE_WAIT
, &xprt
->state
);
860 clear_bit(XPRT_CLOSING
, &xprt
->state
);
861 smp_mb__after_clear_bit();
862 xprt_disconnect_done(xprt
);
865 static void xs_tcp_close(struct rpc_xprt
*xprt
)
867 if (test_and_clear_bit(XPRT_CONNECTION_CLOSE
, &xprt
->state
))
870 xs_tcp_shutdown(xprt
);
874 * xs_destroy - prepare to shutdown a transport
875 * @xprt: doomed transport
878 static void xs_destroy(struct rpc_xprt
*xprt
)
880 dprintk("RPC: xs_destroy xprt %p\n", xprt
);
883 xs_free_peer_addresses(xprt
);
885 module_put(THIS_MODULE
);
888 static inline struct rpc_xprt
*xprt_from_sock(struct sock
*sk
)
890 return (struct rpc_xprt
*) sk
->sk_user_data
;
893 static int xs_local_copy_to_xdr(struct xdr_buf
*xdr
, struct sk_buff
*skb
)
895 struct xdr_skb_reader desc
= {
897 .offset
= sizeof(rpc_fraghdr
),
898 .count
= skb
->len
- sizeof(rpc_fraghdr
),
901 if (xdr_partial_copy_from_skb(xdr
, 0, &desc
, xdr_skb_read_bits
) < 0)
909 * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets
910 * @sk: socket with data to read
911 * @len: how much data to read
913 * Currently this assumes we can read the whole reply in a single gulp.
915 static void xs_local_data_ready(struct sock
*sk
, int len
)
917 struct rpc_task
*task
;
918 struct rpc_xprt
*xprt
;
919 struct rpc_rqst
*rovr
;
921 int err
, repsize
, copied
;
925 read_lock_bh(&sk
->sk_callback_lock
);
926 dprintk("RPC: %s...\n", __func__
);
927 xprt
= xprt_from_sock(sk
);
931 skb
= skb_recv_datagram(sk
, 0, 1, &err
);
935 repsize
= skb
->len
- sizeof(rpc_fraghdr
);
937 dprintk("RPC: impossible RPC reply size %d\n", repsize
);
941 /* Copy the XID from the skb... */
942 xp
= skb_header_pointer(skb
, sizeof(rpc_fraghdr
), sizeof(_xid
), &_xid
);
946 /* Look up and lock the request corresponding to the given XID */
947 spin_lock(&xprt
->transport_lock
);
948 rovr
= xprt_lookup_rqst(xprt
, *xp
);
951 task
= rovr
->rq_task
;
953 copied
= rovr
->rq_private_buf
.buflen
;
954 if (copied
> repsize
)
957 if (xs_local_copy_to_xdr(&rovr
->rq_private_buf
, skb
)) {
958 dprintk("RPC: sk_buff copy failed\n");
962 xprt_complete_rqst(task
, copied
);
965 spin_unlock(&xprt
->transport_lock
);
967 skb_free_datagram(sk
, skb
);
969 read_unlock_bh(&sk
->sk_callback_lock
);
973 * xs_udp_data_ready - "data ready" callback for UDP sockets
974 * @sk: socket with data to read
975 * @len: how much data to read
978 static void xs_udp_data_ready(struct sock
*sk
, int len
)
980 struct rpc_task
*task
;
981 struct rpc_xprt
*xprt
;
982 struct rpc_rqst
*rovr
;
984 int err
, repsize
, copied
;
988 read_lock_bh(&sk
->sk_callback_lock
);
989 dprintk("RPC: xs_udp_data_ready...\n");
990 if (!(xprt
= xprt_from_sock(sk
)))
993 if ((skb
= skb_recv_datagram(sk
, 0, 1, &err
)) == NULL
)
996 repsize
= skb
->len
- sizeof(struct udphdr
);
998 dprintk("RPC: impossible RPC reply size %d!\n", repsize
);
1002 /* Copy the XID from the skb... */
1003 xp
= skb_header_pointer(skb
, sizeof(struct udphdr
),
1004 sizeof(_xid
), &_xid
);
1008 /* Look up and lock the request corresponding to the given XID */
1009 spin_lock(&xprt
->transport_lock
);
1010 rovr
= xprt_lookup_rqst(xprt
, *xp
);
1013 task
= rovr
->rq_task
;
1015 if ((copied
= rovr
->rq_private_buf
.buflen
) > repsize
)
1018 /* Suck it into the iovec, verify checksum if not done by hw. */
1019 if (csum_partial_copy_to_xdr(&rovr
->rq_private_buf
, skb
)) {
1020 UDPX_INC_STATS_BH(sk
, UDP_MIB_INERRORS
);
1024 UDPX_INC_STATS_BH(sk
, UDP_MIB_INDATAGRAMS
);
1026 xprt_adjust_cwnd(xprt
, task
, copied
);
1027 xprt_complete_rqst(task
, copied
);
1030 spin_unlock(&xprt
->transport_lock
);
1032 skb_free_datagram(sk
, skb
);
1034 read_unlock_bh(&sk
->sk_callback_lock
);
1038 * Helper function to force a TCP close if the server is sending
1039 * junk and/or it has put us in CLOSE_WAIT
1041 static void xs_tcp_force_close(struct rpc_xprt
*xprt
)
1043 set_bit(XPRT_CONNECTION_CLOSE
, &xprt
->state
);
1044 xprt_force_disconnect(xprt
);
1047 static inline void xs_tcp_read_fraghdr(struct rpc_xprt
*xprt
, struct xdr_skb_reader
*desc
)
1049 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1053 p
= ((char *) &transport
->tcp_fraghdr
) + transport
->tcp_offset
;
1054 len
= sizeof(transport
->tcp_fraghdr
) - transport
->tcp_offset
;
1055 used
= xdr_skb_read_bits(desc
, p
, len
);
1056 transport
->tcp_offset
+= used
;
1060 transport
->tcp_reclen
= ntohl(transport
->tcp_fraghdr
);
1061 if (transport
->tcp_reclen
& RPC_LAST_STREAM_FRAGMENT
)
1062 transport
->tcp_flags
|= TCP_RCV_LAST_FRAG
;
1064 transport
->tcp_flags
&= ~TCP_RCV_LAST_FRAG
;
1065 transport
->tcp_reclen
&= RPC_FRAGMENT_SIZE_MASK
;
1067 transport
->tcp_flags
&= ~TCP_RCV_COPY_FRAGHDR
;
1068 transport
->tcp_offset
= 0;
1070 /* Sanity check of the record length */
1071 if (unlikely(transport
->tcp_reclen
< 8)) {
1072 dprintk("RPC: invalid TCP record fragment length\n");
1073 xs_tcp_force_close(xprt
);
1076 dprintk("RPC: reading TCP record fragment of length %d\n",
1077 transport
->tcp_reclen
);
1080 static void xs_tcp_check_fraghdr(struct sock_xprt
*transport
)
1082 if (transport
->tcp_offset
== transport
->tcp_reclen
) {
1083 transport
->tcp_flags
|= TCP_RCV_COPY_FRAGHDR
;
1084 transport
->tcp_offset
= 0;
1085 if (transport
->tcp_flags
& TCP_RCV_LAST_FRAG
) {
1086 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
1087 transport
->tcp_flags
|= TCP_RCV_COPY_XID
;
1088 transport
->tcp_copied
= 0;
1093 static inline void xs_tcp_read_xid(struct sock_xprt
*transport
, struct xdr_skb_reader
*desc
)
1098 len
= sizeof(transport
->tcp_xid
) - transport
->tcp_offset
;
1099 dprintk("RPC: reading XID (%Zu bytes)\n", len
);
1100 p
= ((char *) &transport
->tcp_xid
) + transport
->tcp_offset
;
1101 used
= xdr_skb_read_bits(desc
, p
, len
);
1102 transport
->tcp_offset
+= used
;
1105 transport
->tcp_flags
&= ~TCP_RCV_COPY_XID
;
1106 transport
->tcp_flags
|= TCP_RCV_READ_CALLDIR
;
1107 transport
->tcp_copied
= 4;
1108 dprintk("RPC: reading %s XID %08x\n",
1109 (transport
->tcp_flags
& TCP_RPC_REPLY
) ? "reply for"
1111 ntohl(transport
->tcp_xid
));
1112 xs_tcp_check_fraghdr(transport
);
1115 static inline void xs_tcp_read_calldir(struct sock_xprt
*transport
,
1116 struct xdr_skb_reader
*desc
)
1123 * We want transport->tcp_offset to be 8 at the end of this routine
1124 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1125 * When this function is called for the first time,
1126 * transport->tcp_offset is 4 (after having already read the xid).
1128 offset
= transport
->tcp_offset
- sizeof(transport
->tcp_xid
);
1129 len
= sizeof(transport
->tcp_calldir
) - offset
;
1130 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len
);
1131 p
= ((char *) &transport
->tcp_calldir
) + offset
;
1132 used
= xdr_skb_read_bits(desc
, p
, len
);
1133 transport
->tcp_offset
+= used
;
1136 transport
->tcp_flags
&= ~TCP_RCV_READ_CALLDIR
;
1138 * We don't yet have the XDR buffer, so we will write the calldir
1139 * out after we get the buffer from the 'struct rpc_rqst'
1141 switch (ntohl(transport
->tcp_calldir
)) {
1143 transport
->tcp_flags
|= TCP_RCV_COPY_CALLDIR
;
1144 transport
->tcp_flags
|= TCP_RCV_COPY_DATA
;
1145 transport
->tcp_flags
|= TCP_RPC_REPLY
;
1148 transport
->tcp_flags
|= TCP_RCV_COPY_CALLDIR
;
1149 transport
->tcp_flags
|= TCP_RCV_COPY_DATA
;
1150 transport
->tcp_flags
&= ~TCP_RPC_REPLY
;
1153 dprintk("RPC: invalid request message type\n");
1154 xs_tcp_force_close(&transport
->xprt
);
1156 xs_tcp_check_fraghdr(transport
);
1159 static inline void xs_tcp_read_common(struct rpc_xprt
*xprt
,
1160 struct xdr_skb_reader
*desc
,
1161 struct rpc_rqst
*req
)
1163 struct sock_xprt
*transport
=
1164 container_of(xprt
, struct sock_xprt
, xprt
);
1165 struct xdr_buf
*rcvbuf
;
1169 rcvbuf
= &req
->rq_private_buf
;
1171 if (transport
->tcp_flags
& TCP_RCV_COPY_CALLDIR
) {
1173 * Save the RPC direction in the XDR buffer
1175 memcpy(rcvbuf
->head
[0].iov_base
+ transport
->tcp_copied
,
1176 &transport
->tcp_calldir
,
1177 sizeof(transport
->tcp_calldir
));
1178 transport
->tcp_copied
+= sizeof(transport
->tcp_calldir
);
1179 transport
->tcp_flags
&= ~TCP_RCV_COPY_CALLDIR
;
1183 if (len
> transport
->tcp_reclen
- transport
->tcp_offset
) {
1184 struct xdr_skb_reader my_desc
;
1186 len
= transport
->tcp_reclen
- transport
->tcp_offset
;
1187 memcpy(&my_desc
, desc
, sizeof(my_desc
));
1188 my_desc
.count
= len
;
1189 r
= xdr_partial_copy_from_skb(rcvbuf
, transport
->tcp_copied
,
1190 &my_desc
, xdr_skb_read_bits
);
1194 r
= xdr_partial_copy_from_skb(rcvbuf
, transport
->tcp_copied
,
1195 desc
, xdr_skb_read_bits
);
1198 transport
->tcp_copied
+= r
;
1199 transport
->tcp_offset
+= r
;
1202 /* Error when copying to the receive buffer,
1203 * usually because we weren't able to allocate
1204 * additional buffer pages. All we can do now
1205 * is turn off TCP_RCV_COPY_DATA, so the request
1206 * will not receive any additional updates,
1208 * Any remaining data from this record will
1211 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
1212 dprintk("RPC: XID %08x truncated request\n",
1213 ntohl(transport
->tcp_xid
));
1214 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1215 "tcp_offset = %u, tcp_reclen = %u\n",
1216 xprt
, transport
->tcp_copied
,
1217 transport
->tcp_offset
, transport
->tcp_reclen
);
1221 dprintk("RPC: XID %08x read %Zd bytes\n",
1222 ntohl(transport
->tcp_xid
), r
);
1223 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1224 "tcp_reclen = %u\n", xprt
, transport
->tcp_copied
,
1225 transport
->tcp_offset
, transport
->tcp_reclen
);
1227 if (transport
->tcp_copied
== req
->rq_private_buf
.buflen
)
1228 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
1229 else if (transport
->tcp_offset
== transport
->tcp_reclen
) {
1230 if (transport
->tcp_flags
& TCP_RCV_LAST_FRAG
)
1231 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
1236 * Finds the request corresponding to the RPC xid and invokes the common
1237 * tcp read code to read the data.
1239 static inline int xs_tcp_read_reply(struct rpc_xprt
*xprt
,
1240 struct xdr_skb_reader
*desc
)
1242 struct sock_xprt
*transport
=
1243 container_of(xprt
, struct sock_xprt
, xprt
);
1244 struct rpc_rqst
*req
;
1246 dprintk("RPC: read reply XID %08x\n", ntohl(transport
->tcp_xid
));
1248 /* Find and lock the request corresponding to this xid */
1249 spin_lock(&xprt
->transport_lock
);
1250 req
= xprt_lookup_rqst(xprt
, transport
->tcp_xid
);
1252 dprintk("RPC: XID %08x request not found!\n",
1253 ntohl(transport
->tcp_xid
));
1254 spin_unlock(&xprt
->transport_lock
);
1258 xs_tcp_read_common(xprt
, desc
, req
);
1260 if (!(transport
->tcp_flags
& TCP_RCV_COPY_DATA
))
1261 xprt_complete_rqst(req
->rq_task
, transport
->tcp_copied
);
1263 spin_unlock(&xprt
->transport_lock
);
1267 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1269 * Obtains an rpc_rqst previously allocated and invokes the common
1270 * tcp read code to read the data. The result is placed in the callback
1272 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1273 * connection and return -1.
1275 static inline int xs_tcp_read_callback(struct rpc_xprt
*xprt
,
1276 struct xdr_skb_reader
*desc
)
1278 struct sock_xprt
*transport
=
1279 container_of(xprt
, struct sock_xprt
, xprt
);
1280 struct rpc_rqst
*req
;
1282 req
= xprt_alloc_bc_request(xprt
);
1284 printk(KERN_WARNING
"Callback slot table overflowed\n");
1285 xprt_force_disconnect(xprt
);
1289 req
->rq_xid
= transport
->tcp_xid
;
1290 dprintk("RPC: read callback XID %08x\n", ntohl(req
->rq_xid
));
1291 xs_tcp_read_common(xprt
, desc
, req
);
1293 if (!(transport
->tcp_flags
& TCP_RCV_COPY_DATA
)) {
1294 struct svc_serv
*bc_serv
= xprt
->bc_serv
;
1297 * Add callback request to callback list. The callback
1298 * service sleeps on the sv_cb_waitq waiting for new
1299 * requests. Wake it up after adding enqueing the
1302 dprintk("RPC: add callback request to list\n");
1303 spin_lock(&bc_serv
->sv_cb_lock
);
1304 list_add(&req
->rq_bc_list
, &bc_serv
->sv_cb_list
);
1305 spin_unlock(&bc_serv
->sv_cb_lock
);
1306 wake_up(&bc_serv
->sv_cb_waitq
);
1309 req
->rq_private_buf
.len
= transport
->tcp_copied
;
1314 static inline int _xs_tcp_read_data(struct rpc_xprt
*xprt
,
1315 struct xdr_skb_reader
*desc
)
1317 struct sock_xprt
*transport
=
1318 container_of(xprt
, struct sock_xprt
, xprt
);
1320 return (transport
->tcp_flags
& TCP_RPC_REPLY
) ?
1321 xs_tcp_read_reply(xprt
, desc
) :
1322 xs_tcp_read_callback(xprt
, desc
);
1325 static inline int _xs_tcp_read_data(struct rpc_xprt
*xprt
,
1326 struct xdr_skb_reader
*desc
)
1328 return xs_tcp_read_reply(xprt
, desc
);
1330 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1333 * Read data off the transport. This can be either an RPC_CALL or an
1334 * RPC_REPLY. Relay the processing to helper functions.
1336 static void xs_tcp_read_data(struct rpc_xprt
*xprt
,
1337 struct xdr_skb_reader
*desc
)
1339 struct sock_xprt
*transport
=
1340 container_of(xprt
, struct sock_xprt
, xprt
);
1342 if (_xs_tcp_read_data(xprt
, desc
) == 0)
1343 xs_tcp_check_fraghdr(transport
);
1346 * The transport_lock protects the request handling.
1347 * There's no need to hold it to update the tcp_flags.
1349 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
1353 static inline void xs_tcp_read_discard(struct sock_xprt
*transport
, struct xdr_skb_reader
*desc
)
1357 len
= transport
->tcp_reclen
- transport
->tcp_offset
;
1358 if (len
> desc
->count
)
1361 desc
->offset
+= len
;
1362 transport
->tcp_offset
+= len
;
1363 dprintk("RPC: discarded %Zu bytes\n", len
);
1364 xs_tcp_check_fraghdr(transport
);
1367 static int xs_tcp_data_recv(read_descriptor_t
*rd_desc
, struct sk_buff
*skb
, unsigned int offset
, size_t len
)
1369 struct rpc_xprt
*xprt
= rd_desc
->arg
.data
;
1370 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1371 struct xdr_skb_reader desc
= {
1377 dprintk("RPC: xs_tcp_data_recv started\n");
1379 /* Read in a new fragment marker if necessary */
1380 /* Can we ever really expect to get completely empty fragments? */
1381 if (transport
->tcp_flags
& TCP_RCV_COPY_FRAGHDR
) {
1382 xs_tcp_read_fraghdr(xprt
, &desc
);
1385 /* Read in the xid if necessary */
1386 if (transport
->tcp_flags
& TCP_RCV_COPY_XID
) {
1387 xs_tcp_read_xid(transport
, &desc
);
1390 /* Read in the call/reply flag */
1391 if (transport
->tcp_flags
& TCP_RCV_READ_CALLDIR
) {
1392 xs_tcp_read_calldir(transport
, &desc
);
1395 /* Read in the request data */
1396 if (transport
->tcp_flags
& TCP_RCV_COPY_DATA
) {
1397 xs_tcp_read_data(xprt
, &desc
);
1400 /* Skip over any trailing bytes on short reads */
1401 xs_tcp_read_discard(transport
, &desc
);
1402 } while (desc
.count
);
1403 dprintk("RPC: xs_tcp_data_recv done\n");
1404 return len
- desc
.count
;
1408 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1409 * @sk: socket with data to read
1410 * @bytes: how much data to read
1413 static void xs_tcp_data_ready(struct sock
*sk
, int bytes
)
1415 struct rpc_xprt
*xprt
;
1416 read_descriptor_t rd_desc
;
1419 dprintk("RPC: xs_tcp_data_ready...\n");
1421 read_lock_bh(&sk
->sk_callback_lock
);
1422 if (!(xprt
= xprt_from_sock(sk
)))
1424 /* Any data means we had a useful conversation, so
1425 * the we don't need to delay the next reconnect
1427 if (xprt
->reestablish_timeout
)
1428 xprt
->reestablish_timeout
= 0;
1430 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1431 rd_desc
.arg
.data
= xprt
;
1433 rd_desc
.count
= 65536;
1434 read
= tcp_read_sock(sk
, &rd_desc
, xs_tcp_data_recv
);
1437 read_unlock_bh(&sk
->sk_callback_lock
);
1441 * Do the equivalent of linger/linger2 handling for dealing with
1442 * broken servers that don't close the socket in a timely
1445 static void xs_tcp_schedule_linger_timeout(struct rpc_xprt
*xprt
,
1446 unsigned long timeout
)
1448 struct sock_xprt
*transport
;
1450 if (xprt_test_and_set_connecting(xprt
))
1452 set_bit(XPRT_CONNECTION_ABORT
, &xprt
->state
);
1453 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1454 queue_delayed_work(rpciod_workqueue
, &transport
->connect_worker
,
1458 static void xs_tcp_cancel_linger_timeout(struct rpc_xprt
*xprt
)
1460 struct sock_xprt
*transport
;
1462 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1464 if (!test_bit(XPRT_CONNECTION_ABORT
, &xprt
->state
) ||
1465 !cancel_delayed_work(&transport
->connect_worker
))
1467 clear_bit(XPRT_CONNECTION_ABORT
, &xprt
->state
);
1468 xprt_clear_connecting(xprt
);
1471 static void xs_sock_reset_connection_flags(struct rpc_xprt
*xprt
)
1473 smp_mb__before_clear_bit();
1474 clear_bit(XPRT_CONNECTION_ABORT
, &xprt
->state
);
1475 clear_bit(XPRT_CONNECTION_CLOSE
, &xprt
->state
);
1476 clear_bit(XPRT_CLOSE_WAIT
, &xprt
->state
);
1477 clear_bit(XPRT_CLOSING
, &xprt
->state
);
1478 smp_mb__after_clear_bit();
1481 static void xs_sock_mark_closed(struct rpc_xprt
*xprt
)
1483 xs_sock_reset_connection_flags(xprt
);
1484 /* Mark transport as closed and wake up all pending tasks */
1485 xprt_disconnect_done(xprt
);
1489 * xs_tcp_state_change - callback to handle TCP socket state changes
1490 * @sk: socket whose state has changed
1493 static void xs_tcp_state_change(struct sock
*sk
)
1495 struct rpc_xprt
*xprt
;
1497 read_lock_bh(&sk
->sk_callback_lock
);
1498 if (!(xprt
= xprt_from_sock(sk
)))
1500 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt
);
1501 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1502 sk
->sk_state
, xprt_connected(xprt
),
1503 sock_flag(sk
, SOCK_DEAD
),
1504 sock_flag(sk
, SOCK_ZAPPED
),
1507 trace_rpc_socket_state_change(xprt
, sk
->sk_socket
);
1508 switch (sk
->sk_state
) {
1509 case TCP_ESTABLISHED
:
1510 spin_lock(&xprt
->transport_lock
);
1511 if (!xprt_test_and_set_connected(xprt
)) {
1512 struct sock_xprt
*transport
= container_of(xprt
,
1513 struct sock_xprt
, xprt
);
1515 /* Reset TCP record info */
1516 transport
->tcp_offset
= 0;
1517 transport
->tcp_reclen
= 0;
1518 transport
->tcp_copied
= 0;
1519 transport
->tcp_flags
=
1520 TCP_RCV_COPY_FRAGHDR
| TCP_RCV_COPY_XID
;
1521 xprt
->connect_cookie
++;
1523 xprt_wake_pending_tasks(xprt
, -EAGAIN
);
1525 spin_unlock(&xprt
->transport_lock
);
1528 /* The client initiated a shutdown of the socket */
1529 xprt
->connect_cookie
++;
1530 xprt
->reestablish_timeout
= 0;
1531 set_bit(XPRT_CLOSING
, &xprt
->state
);
1532 smp_mb__before_clear_bit();
1533 clear_bit(XPRT_CONNECTED
, &xprt
->state
);
1534 clear_bit(XPRT_CLOSE_WAIT
, &xprt
->state
);
1535 smp_mb__after_clear_bit();
1536 xs_tcp_schedule_linger_timeout(xprt
, xs_tcp_fin_timeout
);
1538 case TCP_CLOSE_WAIT
:
1539 /* The server initiated a shutdown of the socket */
1540 xprt
->connect_cookie
++;
1541 clear_bit(XPRT_CONNECTED
, &xprt
->state
);
1542 xs_tcp_force_close(xprt
);
1545 * If the server closed down the connection, make sure that
1546 * we back off before reconnecting
1548 if (xprt
->reestablish_timeout
< XS_TCP_INIT_REEST_TO
)
1549 xprt
->reestablish_timeout
= XS_TCP_INIT_REEST_TO
;
1552 set_bit(XPRT_CLOSING
, &xprt
->state
);
1553 xs_tcp_schedule_linger_timeout(xprt
, xs_tcp_fin_timeout
);
1554 smp_mb__before_clear_bit();
1555 clear_bit(XPRT_CONNECTED
, &xprt
->state
);
1556 smp_mb__after_clear_bit();
1559 xs_tcp_cancel_linger_timeout(xprt
);
1560 xs_sock_mark_closed(xprt
);
1563 read_unlock_bh(&sk
->sk_callback_lock
);
1566 static void xs_write_space(struct sock
*sk
)
1568 struct socket
*sock
;
1569 struct rpc_xprt
*xprt
;
1571 if (unlikely(!(sock
= sk
->sk_socket
)))
1573 clear_bit(SOCK_NOSPACE
, &sock
->flags
);
1575 if (unlikely(!(xprt
= xprt_from_sock(sk
))))
1577 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE
, &sock
->flags
) == 0)
1580 xprt_write_space(xprt
);
1584 * xs_udp_write_space - callback invoked when socket buffer space
1586 * @sk: socket whose state has changed
1588 * Called when more output buffer space is available for this socket.
1589 * We try not to wake our writers until they can make "significant"
1590 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1591 * with a bunch of small requests.
1593 static void xs_udp_write_space(struct sock
*sk
)
1595 read_lock_bh(&sk
->sk_callback_lock
);
1597 /* from net/core/sock.c:sock_def_write_space */
1598 if (sock_writeable(sk
))
1601 read_unlock_bh(&sk
->sk_callback_lock
);
1605 * xs_tcp_write_space - callback invoked when socket buffer space
1607 * @sk: socket whose state has changed
1609 * Called when more output buffer space is available for this socket.
1610 * We try not to wake our writers until they can make "significant"
1611 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1612 * with a bunch of small requests.
1614 static void xs_tcp_write_space(struct sock
*sk
)
1616 read_lock_bh(&sk
->sk_callback_lock
);
1618 /* from net/core/stream.c:sk_stream_write_space */
1619 if (sk_stream_is_writeable(sk
))
1622 read_unlock_bh(&sk
->sk_callback_lock
);
1625 static void xs_udp_do_set_buffer_size(struct rpc_xprt
*xprt
)
1627 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1628 struct sock
*sk
= transport
->inet
;
1630 if (transport
->rcvsize
) {
1631 sk
->sk_userlocks
|= SOCK_RCVBUF_LOCK
;
1632 sk
->sk_rcvbuf
= transport
->rcvsize
* xprt
->max_reqs
* 2;
1634 if (transport
->sndsize
) {
1635 sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
;
1636 sk
->sk_sndbuf
= transport
->sndsize
* xprt
->max_reqs
* 2;
1637 sk
->sk_write_space(sk
);
1642 * xs_udp_set_buffer_size - set send and receive limits
1643 * @xprt: generic transport
1644 * @sndsize: requested size of send buffer, in bytes
1645 * @rcvsize: requested size of receive buffer, in bytes
1647 * Set socket send and receive buffer size limits.
1649 static void xs_udp_set_buffer_size(struct rpc_xprt
*xprt
, size_t sndsize
, size_t rcvsize
)
1651 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1653 transport
->sndsize
= 0;
1655 transport
->sndsize
= sndsize
+ 1024;
1656 transport
->rcvsize
= 0;
1658 transport
->rcvsize
= rcvsize
+ 1024;
1660 xs_udp_do_set_buffer_size(xprt
);
1664 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1665 * @task: task that timed out
1667 * Adjust the congestion window after a retransmit timeout has occurred.
1669 static void xs_udp_timer(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
1671 xprt_adjust_cwnd(xprt
, task
, -ETIMEDOUT
);
1674 static unsigned short xs_get_random_port(void)
1676 unsigned short range
= xprt_max_resvport
- xprt_min_resvport
;
1677 unsigned short rand
= (unsigned short) net_random() % range
;
1678 return rand
+ xprt_min_resvport
;
1682 * xs_set_port - reset the port number in the remote endpoint address
1683 * @xprt: generic transport
1684 * @port: new port number
1687 static void xs_set_port(struct rpc_xprt
*xprt
, unsigned short port
)
1689 dprintk("RPC: setting port for xprt %p to %u\n", xprt
, port
);
1691 rpc_set_port(xs_addr(xprt
), port
);
1692 xs_update_peer_port(xprt
);
1695 static unsigned short xs_get_srcport(struct sock_xprt
*transport
)
1697 unsigned short port
= transport
->srcport
;
1699 if (port
== 0 && transport
->xprt
.resvport
)
1700 port
= xs_get_random_port();
1704 static unsigned short xs_next_srcport(struct sock_xprt
*transport
, unsigned short port
)
1706 if (transport
->srcport
!= 0)
1707 transport
->srcport
= 0;
1708 if (!transport
->xprt
.resvport
)
1710 if (port
<= xprt_min_resvport
|| port
> xprt_max_resvport
)
1711 return xprt_max_resvport
;
1714 static int xs_bind(struct sock_xprt
*transport
, struct socket
*sock
)
1716 struct sockaddr_storage myaddr
;
1718 unsigned short port
= xs_get_srcport(transport
);
1719 unsigned short last
;
1721 memcpy(&myaddr
, &transport
->srcaddr
, transport
->xprt
.addrlen
);
1723 rpc_set_port((struct sockaddr
*)&myaddr
, port
);
1724 err
= kernel_bind(sock
, (struct sockaddr
*)&myaddr
,
1725 transport
->xprt
.addrlen
);
1729 transport
->srcport
= port
;
1733 port
= xs_next_srcport(transport
, port
);
1736 } while (err
== -EADDRINUSE
&& nloop
!= 2);
1738 if (myaddr
.ss_family
== AF_INET
)
1739 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__
,
1740 &((struct sockaddr_in
*)&myaddr
)->sin_addr
,
1741 port
, err
? "failed" : "ok", err
);
1743 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__
,
1744 &((struct sockaddr_in6
*)&myaddr
)->sin6_addr
,
1745 port
, err
? "failed" : "ok", err
);
1750 * We don't support autobind on AF_LOCAL sockets
1752 static void xs_local_rpcbind(struct rpc_task
*task
)
1755 xprt_set_bound(rcu_dereference(task
->tk_client
->cl_xprt
));
1759 static void xs_local_set_port(struct rpc_xprt
*xprt
, unsigned short port
)
1763 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1764 static struct lock_class_key xs_key
[2];
1765 static struct lock_class_key xs_slock_key
[2];
1767 static inline void xs_reclassify_socketu(struct socket
*sock
)
1769 struct sock
*sk
= sock
->sk
;
1771 sock_lock_init_class_and_name(sk
, "slock-AF_LOCAL-RPC",
1772 &xs_slock_key
[1], "sk_lock-AF_LOCAL-RPC", &xs_key
[1]);
1775 static inline void xs_reclassify_socket4(struct socket
*sock
)
1777 struct sock
*sk
= sock
->sk
;
1779 sock_lock_init_class_and_name(sk
, "slock-AF_INET-RPC",
1780 &xs_slock_key
[0], "sk_lock-AF_INET-RPC", &xs_key
[0]);
1783 static inline void xs_reclassify_socket6(struct socket
*sock
)
1785 struct sock
*sk
= sock
->sk
;
1787 sock_lock_init_class_and_name(sk
, "slock-AF_INET6-RPC",
1788 &xs_slock_key
[1], "sk_lock-AF_INET6-RPC", &xs_key
[1]);
1791 static inline void xs_reclassify_socket(int family
, struct socket
*sock
)
1793 WARN_ON_ONCE(sock_owned_by_user(sock
->sk
));
1794 if (sock_owned_by_user(sock
->sk
))
1799 xs_reclassify_socketu(sock
);
1802 xs_reclassify_socket4(sock
);
1805 xs_reclassify_socket6(sock
);
1810 static inline void xs_reclassify_socketu(struct socket
*sock
)
1814 static inline void xs_reclassify_socket4(struct socket
*sock
)
1818 static inline void xs_reclassify_socket6(struct socket
*sock
)
1822 static inline void xs_reclassify_socket(int family
, struct socket
*sock
)
1827 static void xs_dummy_setup_socket(struct work_struct
*work
)
1831 static struct socket
*xs_create_sock(struct rpc_xprt
*xprt
,
1832 struct sock_xprt
*transport
, int family
, int type
, int protocol
)
1834 struct socket
*sock
;
1837 err
= __sock_create(xprt
->xprt_net
, family
, type
, protocol
, &sock
, 1);
1839 dprintk("RPC: can't create %d transport socket (%d).\n",
1843 xs_reclassify_socket(family
, sock
);
1845 err
= xs_bind(transport
, sock
);
1853 return ERR_PTR(err
);
1856 static int xs_local_finish_connecting(struct rpc_xprt
*xprt
,
1857 struct socket
*sock
)
1859 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
,
1862 if (!transport
->inet
) {
1863 struct sock
*sk
= sock
->sk
;
1865 write_lock_bh(&sk
->sk_callback_lock
);
1867 xs_save_old_callbacks(transport
, sk
);
1869 sk
->sk_user_data
= xprt
;
1870 sk
->sk_data_ready
= xs_local_data_ready
;
1871 sk
->sk_write_space
= xs_udp_write_space
;
1872 sk
->sk_allocation
= GFP_ATOMIC
;
1874 xprt_clear_connected(xprt
);
1876 /* Reset to new socket */
1877 transport
->sock
= sock
;
1878 transport
->inet
= sk
;
1880 write_unlock_bh(&sk
->sk_callback_lock
);
1883 /* Tell the socket layer to start connecting... */
1884 xprt
->stat
.connect_count
++;
1885 xprt
->stat
.connect_start
= jiffies
;
1886 return kernel_connect(sock
, xs_addr(xprt
), xprt
->addrlen
, 0);
1890 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1891 * @xprt: RPC transport to connect
1892 * @transport: socket transport to connect
1893 * @create_sock: function to create a socket of the correct type
1895 static int xs_local_setup_socket(struct sock_xprt
*transport
)
1897 struct rpc_xprt
*xprt
= &transport
->xprt
;
1898 struct socket
*sock
;
1901 current
->flags
|= PF_FSTRANS
;
1903 clear_bit(XPRT_CONNECTION_ABORT
, &xprt
->state
);
1904 status
= __sock_create(xprt
->xprt_net
, AF_LOCAL
,
1905 SOCK_STREAM
, 0, &sock
, 1);
1907 dprintk("RPC: can't create AF_LOCAL "
1908 "transport socket (%d).\n", -status
);
1911 xs_reclassify_socketu(sock
);
1913 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1914 xprt
, xprt
->address_strings
[RPC_DISPLAY_ADDR
]);
1916 status
= xs_local_finish_connecting(xprt
, sock
);
1917 trace_rpc_socket_connect(xprt
, sock
, status
);
1920 dprintk("RPC: xprt %p connected to %s\n",
1921 xprt
, xprt
->address_strings
[RPC_DISPLAY_ADDR
]);
1922 xprt_set_connected(xprt
);
1925 dprintk("RPC: xprt %p: socket %s does not exist\n",
1926 xprt
, xprt
->address_strings
[RPC_DISPLAY_ADDR
]);
1929 dprintk("RPC: xprt %p: connection refused for %s\n",
1930 xprt
, xprt
->address_strings
[RPC_DISPLAY_ADDR
]);
1933 printk(KERN_ERR
"%s: unhandled error (%d) connecting to %s\n",
1935 xprt
->address_strings
[RPC_DISPLAY_ADDR
]);
1939 xprt_clear_connecting(xprt
);
1940 xprt_wake_pending_tasks(xprt
, status
);
1941 current
->flags
&= ~PF_FSTRANS
;
1945 static void xs_local_connect(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
1947 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1950 if (RPC_IS_ASYNC(task
)) {
1952 * We want the AF_LOCAL connect to be resolved in the
1953 * filesystem namespace of the process making the rpc
1954 * call. Thus we connect synchronously.
1956 * If we want to support asynchronous AF_LOCAL calls,
1957 * we'll need to figure out how to pass a namespace to
1960 rpc_exit(task
, -ENOTCONN
);
1963 ret
= xs_local_setup_socket(transport
);
1964 if (ret
&& !RPC_IS_SOFTCONN(task
))
1965 msleep_interruptible(15000);
1968 #ifdef CONFIG_SUNRPC_SWAP
1969 static void xs_set_memalloc(struct rpc_xprt
*xprt
)
1971 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
,
1975 sk_set_memalloc(transport
->inet
);
1979 * xs_swapper - Tag this transport as being used for swap.
1980 * @xprt: transport to tag
1981 * @enable: enable/disable
1984 int xs_swapper(struct rpc_xprt
*xprt
, int enable
)
1986 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
,
1992 xs_set_memalloc(xprt
);
1993 } else if (xprt
->swapper
) {
1995 sk_clear_memalloc(transport
->inet
);
2000 EXPORT_SYMBOL_GPL(xs_swapper
);
2002 static void xs_set_memalloc(struct rpc_xprt
*xprt
)
2007 static void xs_udp_finish_connecting(struct rpc_xprt
*xprt
, struct socket
*sock
)
2009 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2011 if (!transport
->inet
) {
2012 struct sock
*sk
= sock
->sk
;
2014 write_lock_bh(&sk
->sk_callback_lock
);
2016 xs_save_old_callbacks(transport
, sk
);
2018 sk
->sk_user_data
= xprt
;
2019 sk
->sk_data_ready
= xs_udp_data_ready
;
2020 sk
->sk_write_space
= xs_udp_write_space
;
2021 sk
->sk_no_check
= UDP_CSUM_NORCV
;
2022 sk
->sk_allocation
= GFP_ATOMIC
;
2024 xprt_set_connected(xprt
);
2026 /* Reset to new socket */
2027 transport
->sock
= sock
;
2028 transport
->inet
= sk
;
2030 xs_set_memalloc(xprt
);
2032 write_unlock_bh(&sk
->sk_callback_lock
);
2034 xs_udp_do_set_buffer_size(xprt
);
2037 static void xs_udp_setup_socket(struct work_struct
*work
)
2039 struct sock_xprt
*transport
=
2040 container_of(work
, struct sock_xprt
, connect_worker
.work
);
2041 struct rpc_xprt
*xprt
= &transport
->xprt
;
2042 struct socket
*sock
= transport
->sock
;
2045 current
->flags
|= PF_FSTRANS
;
2047 /* Start by resetting any existing state */
2048 xs_reset_transport(transport
);
2049 sock
= xs_create_sock(xprt
, transport
,
2050 xs_addr(xprt
)->sa_family
, SOCK_DGRAM
, IPPROTO_UDP
);
2054 dprintk("RPC: worker connecting xprt %p via %s to "
2055 "%s (port %s)\n", xprt
,
2056 xprt
->address_strings
[RPC_DISPLAY_PROTO
],
2057 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
2058 xprt
->address_strings
[RPC_DISPLAY_PORT
]);
2060 xs_udp_finish_connecting(xprt
, sock
);
2061 trace_rpc_socket_connect(xprt
, sock
, 0);
2064 xprt_clear_connecting(xprt
);
2065 xprt_wake_pending_tasks(xprt
, status
);
2066 current
->flags
&= ~PF_FSTRANS
;
2070 * We need to preserve the port number so the reply cache on the server can
2071 * find our cached RPC replies when we get around to reconnecting.
2073 static void xs_abort_connection(struct sock_xprt
*transport
)
2076 struct sockaddr any
;
2078 dprintk("RPC: disconnecting xprt %p to reuse port\n", transport
);
2081 * Disconnect the transport socket by doing a connect operation
2082 * with AF_UNSPEC. This should return immediately...
2084 memset(&any
, 0, sizeof(any
));
2085 any
.sa_family
= AF_UNSPEC
;
2086 result
= kernel_connect(transport
->sock
, &any
, sizeof(any
), 0);
2087 trace_rpc_socket_reset_connection(&transport
->xprt
,
2088 transport
->sock
, result
);
2090 xs_sock_reset_connection_flags(&transport
->xprt
);
2091 dprintk("RPC: AF_UNSPEC connect return code %d\n", result
);
2094 static void xs_tcp_reuse_connection(struct sock_xprt
*transport
)
2096 unsigned int state
= transport
->inet
->sk_state
;
2098 if (state
== TCP_CLOSE
&& transport
->sock
->state
== SS_UNCONNECTED
) {
2099 /* we don't need to abort the connection if the socket
2100 * hasn't undergone a shutdown
2102 if (transport
->inet
->sk_shutdown
== 0)
2104 dprintk("RPC: %s: TCP_CLOSEd and sk_shutdown set to %d\n",
2105 __func__
, transport
->inet
->sk_shutdown
);
2107 if ((1 << state
) & (TCPF_ESTABLISHED
|TCPF_SYN_SENT
)) {
2108 /* we don't need to abort the connection if the socket
2109 * hasn't undergone a shutdown
2111 if (transport
->inet
->sk_shutdown
== 0)
2113 dprintk("RPC: %s: ESTABLISHED/SYN_SENT "
2114 "sk_shutdown set to %d\n",
2115 __func__
, transport
->inet
->sk_shutdown
);
2117 xs_abort_connection(transport
);
2120 static int xs_tcp_finish_connecting(struct rpc_xprt
*xprt
, struct socket
*sock
)
2122 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2123 int ret
= -ENOTCONN
;
2125 if (!transport
->inet
) {
2126 struct sock
*sk
= sock
->sk
;
2127 unsigned int keepidle
= xprt
->timeout
->to_initval
/ HZ
;
2128 unsigned int keepcnt
= xprt
->timeout
->to_retries
+ 1;
2129 unsigned int opt_on
= 1;
2131 /* TCP Keepalive options */
2132 kernel_setsockopt(sock
, SOL_SOCKET
, SO_KEEPALIVE
,
2133 (char *)&opt_on
, sizeof(opt_on
));
2134 kernel_setsockopt(sock
, SOL_TCP
, TCP_KEEPIDLE
,
2135 (char *)&keepidle
, sizeof(keepidle
));
2136 kernel_setsockopt(sock
, SOL_TCP
, TCP_KEEPINTVL
,
2137 (char *)&keepidle
, sizeof(keepidle
));
2138 kernel_setsockopt(sock
, SOL_TCP
, TCP_KEEPCNT
,
2139 (char *)&keepcnt
, sizeof(keepcnt
));
2141 write_lock_bh(&sk
->sk_callback_lock
);
2143 xs_save_old_callbacks(transport
, sk
);
2145 sk
->sk_user_data
= xprt
;
2146 sk
->sk_data_ready
= xs_tcp_data_ready
;
2147 sk
->sk_state_change
= xs_tcp_state_change
;
2148 sk
->sk_write_space
= xs_tcp_write_space
;
2149 sk
->sk_allocation
= GFP_ATOMIC
;
2151 /* socket options */
2152 sk
->sk_userlocks
|= SOCK_BINDPORT_LOCK
;
2153 sock_reset_flag(sk
, SOCK_LINGER
);
2154 tcp_sk(sk
)->linger2
= 0;
2155 tcp_sk(sk
)->nonagle
|= TCP_NAGLE_OFF
;
2157 xprt_clear_connected(xprt
);
2159 /* Reset to new socket */
2160 transport
->sock
= sock
;
2161 transport
->inet
= sk
;
2163 write_unlock_bh(&sk
->sk_callback_lock
);
2166 if (!xprt_bound(xprt
))
2169 xs_set_memalloc(xprt
);
2171 /* Tell the socket layer to start connecting... */
2172 xprt
->stat
.connect_count
++;
2173 xprt
->stat
.connect_start
= jiffies
;
2174 ret
= kernel_connect(sock
, xs_addr(xprt
), xprt
->addrlen
, O_NONBLOCK
);
2179 if (xprt
->reestablish_timeout
< XS_TCP_INIT_REEST_TO
)
2180 xprt
->reestablish_timeout
= XS_TCP_INIT_REEST_TO
;
2187 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2188 * @xprt: RPC transport to connect
2189 * @transport: socket transport to connect
2190 * @create_sock: function to create a socket of the correct type
2192 * Invoked by a work queue tasklet.
2194 static void xs_tcp_setup_socket(struct work_struct
*work
)
2196 struct sock_xprt
*transport
=
2197 container_of(work
, struct sock_xprt
, connect_worker
.work
);
2198 struct socket
*sock
= transport
->sock
;
2199 struct rpc_xprt
*xprt
= &transport
->xprt
;
2202 current
->flags
|= PF_FSTRANS
;
2205 clear_bit(XPRT_CONNECTION_ABORT
, &xprt
->state
);
2206 sock
= xs_create_sock(xprt
, transport
,
2207 xs_addr(xprt
)->sa_family
, SOCK_STREAM
, IPPROTO_TCP
);
2209 status
= PTR_ERR(sock
);
2215 abort_and_exit
= test_and_clear_bit(XPRT_CONNECTION_ABORT
,
2217 /* "close" the socket, preserving the local port */
2218 xs_tcp_reuse_connection(transport
);
2224 dprintk("RPC: worker connecting xprt %p via %s to "
2225 "%s (port %s)\n", xprt
,
2226 xprt
->address_strings
[RPC_DISPLAY_PROTO
],
2227 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
2228 xprt
->address_strings
[RPC_DISPLAY_PORT
]);
2230 status
= xs_tcp_finish_connecting(xprt
, sock
);
2231 trace_rpc_socket_connect(xprt
, sock
, status
);
2232 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2233 xprt
, -status
, xprt_connected(xprt
),
2234 sock
->sk
->sk_state
);
2237 printk("%s: connect returned unhandled error %d\n",
2239 case -EADDRNOTAVAIL
:
2240 /* We're probably in TIME_WAIT. Get rid of existing socket,
2243 xs_tcp_force_close(xprt
);
2248 xprt_clear_connecting(xprt
);
2249 current
->flags
&= ~PF_FSTRANS
;
2252 /* Happens, for instance, if the user specified a link
2253 * local IPv6 address without a scope-id.
2258 /* retry with existing socket, after a delay */
2264 xprt_clear_connecting(xprt
);
2265 xprt_wake_pending_tasks(xprt
, status
);
2266 current
->flags
&= ~PF_FSTRANS
;
2270 * xs_connect - connect a socket to a remote endpoint
2271 * @xprt: pointer to transport structure
2272 * @task: address of RPC task that manages state of connect request
2274 * TCP: If the remote end dropped the connection, delay reconnecting.
2276 * UDP socket connects are synchronous, but we use a work queue anyway
2277 * to guarantee that even unprivileged user processes can set up a
2278 * socket on a privileged port.
2280 * If a UDP socket connect fails, the delay behavior here prevents
2281 * retry floods (hard mounts).
2283 static void xs_connect(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
2285 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2287 if (transport
->sock
!= NULL
&& !RPC_IS_SOFTCONN(task
)) {
2288 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2290 xprt
, xprt
->reestablish_timeout
/ HZ
);
2291 queue_delayed_work(rpciod_workqueue
,
2292 &transport
->connect_worker
,
2293 xprt
->reestablish_timeout
);
2294 xprt
->reestablish_timeout
<<= 1;
2295 if (xprt
->reestablish_timeout
< XS_TCP_INIT_REEST_TO
)
2296 xprt
->reestablish_timeout
= XS_TCP_INIT_REEST_TO
;
2297 if (xprt
->reestablish_timeout
> XS_TCP_MAX_REEST_TO
)
2298 xprt
->reestablish_timeout
= XS_TCP_MAX_REEST_TO
;
2300 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt
);
2301 queue_delayed_work(rpciod_workqueue
,
2302 &transport
->connect_worker
, 0);
2307 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2308 * @xprt: rpc_xprt struct containing statistics
2312 static void xs_local_print_stats(struct rpc_xprt
*xprt
, struct seq_file
*seq
)
2316 if (xprt_connected(xprt
))
2317 idle_time
= (long)(jiffies
- xprt
->last_used
) / HZ
;
2319 seq_printf(seq
, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2320 "%llu %llu %lu %llu %llu\n",
2321 xprt
->stat
.bind_count
,
2322 xprt
->stat
.connect_count
,
2323 xprt
->stat
.connect_time
,
2327 xprt
->stat
.bad_xids
,
2330 xprt
->stat
.max_slots
,
2331 xprt
->stat
.sending_u
,
2332 xprt
->stat
.pending_u
);
2336 * xs_udp_print_stats - display UDP socket-specifc stats
2337 * @xprt: rpc_xprt struct containing statistics
2341 static void xs_udp_print_stats(struct rpc_xprt
*xprt
, struct seq_file
*seq
)
2343 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2345 seq_printf(seq
, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2348 xprt
->stat
.bind_count
,
2351 xprt
->stat
.bad_xids
,
2354 xprt
->stat
.max_slots
,
2355 xprt
->stat
.sending_u
,
2356 xprt
->stat
.pending_u
);
2360 * xs_tcp_print_stats - display TCP socket-specifc stats
2361 * @xprt: rpc_xprt struct containing statistics
2365 static void xs_tcp_print_stats(struct rpc_xprt
*xprt
, struct seq_file
*seq
)
2367 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2370 if (xprt_connected(xprt
))
2371 idle_time
= (long)(jiffies
- xprt
->last_used
) / HZ
;
2373 seq_printf(seq
, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2374 "%llu %llu %lu %llu %llu\n",
2376 xprt
->stat
.bind_count
,
2377 xprt
->stat
.connect_count
,
2378 xprt
->stat
.connect_time
,
2382 xprt
->stat
.bad_xids
,
2385 xprt
->stat
.max_slots
,
2386 xprt
->stat
.sending_u
,
2387 xprt
->stat
.pending_u
);
2391 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2392 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2393 * to use the server side send routines.
2395 static void *bc_malloc(struct rpc_task
*task
, size_t size
)
2398 struct rpc_buffer
*buf
;
2400 WARN_ON_ONCE(size
> PAGE_SIZE
- sizeof(struct rpc_buffer
));
2401 if (size
> PAGE_SIZE
- sizeof(struct rpc_buffer
))
2404 page
= alloc_page(GFP_KERNEL
);
2408 buf
= page_address(page
);
2409 buf
->len
= PAGE_SIZE
;
2415 * Free the space allocated in the bc_alloc routine
2417 static void bc_free(void *buffer
)
2419 struct rpc_buffer
*buf
;
2424 buf
= container_of(buffer
, struct rpc_buffer
, data
);
2425 free_page((unsigned long)buf
);
2429 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2430 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2432 static int bc_sendto(struct rpc_rqst
*req
)
2435 struct xdr_buf
*xbufp
= &req
->rq_snd_buf
;
2436 struct rpc_xprt
*xprt
= req
->rq_xprt
;
2437 struct sock_xprt
*transport
=
2438 container_of(xprt
, struct sock_xprt
, xprt
);
2439 struct socket
*sock
= transport
->sock
;
2440 unsigned long headoff
;
2441 unsigned long tailoff
;
2443 xs_encode_stream_record_marker(xbufp
);
2445 tailoff
= (unsigned long)xbufp
->tail
[0].iov_base
& ~PAGE_MASK
;
2446 headoff
= (unsigned long)xbufp
->head
[0].iov_base
& ~PAGE_MASK
;
2447 len
= svc_send_common(sock
, xbufp
,
2448 virt_to_page(xbufp
->head
[0].iov_base
), headoff
,
2449 xbufp
->tail
[0].iov_base
, tailoff
);
2451 if (len
!= xbufp
->len
) {
2452 printk(KERN_NOTICE
"Error sending entire callback!\n");
2460 * The send routine. Borrows from svc_send
2462 static int bc_send_request(struct rpc_task
*task
)
2464 struct rpc_rqst
*req
= task
->tk_rqstp
;
2465 struct svc_xprt
*xprt
;
2468 dprintk("sending request with xid: %08x\n", ntohl(req
->rq_xid
));
2470 * Get the server socket associated with this callback xprt
2472 xprt
= req
->rq_xprt
->bc_xprt
;
2475 * Grab the mutex to serialize data as the connection is shared
2476 * with the fore channel
2478 if (!mutex_trylock(&xprt
->xpt_mutex
)) {
2479 rpc_sleep_on(&xprt
->xpt_bc_pending
, task
, NULL
);
2480 if (!mutex_trylock(&xprt
->xpt_mutex
))
2482 rpc_wake_up_queued_task(&xprt
->xpt_bc_pending
, task
);
2484 if (test_bit(XPT_DEAD
, &xprt
->xpt_flags
))
2487 len
= bc_sendto(req
);
2488 mutex_unlock(&xprt
->xpt_mutex
);
2497 * The close routine. Since this is client initiated, we do nothing
2500 static void bc_close(struct rpc_xprt
*xprt
)
2505 * The xprt destroy routine. Again, because this connection is client
2506 * initiated, we do nothing
2509 static void bc_destroy(struct rpc_xprt
*xprt
)
2513 static struct rpc_xprt_ops xs_local_ops
= {
2514 .reserve_xprt
= xprt_reserve_xprt
,
2515 .release_xprt
= xs_tcp_release_xprt
,
2516 .alloc_slot
= xprt_alloc_slot
,
2517 .rpcbind
= xs_local_rpcbind
,
2518 .set_port
= xs_local_set_port
,
2519 .connect
= xs_local_connect
,
2520 .buf_alloc
= rpc_malloc
,
2521 .buf_free
= rpc_free
,
2522 .send_request
= xs_local_send_request
,
2523 .set_retrans_timeout
= xprt_set_retrans_timeout_def
,
2525 .destroy
= xs_destroy
,
2526 .print_stats
= xs_local_print_stats
,
2529 static struct rpc_xprt_ops xs_udp_ops
= {
2530 .set_buffer_size
= xs_udp_set_buffer_size
,
2531 .reserve_xprt
= xprt_reserve_xprt_cong
,
2532 .release_xprt
= xprt_release_xprt_cong
,
2533 .alloc_slot
= xprt_alloc_slot
,
2534 .rpcbind
= rpcb_getport_async
,
2535 .set_port
= xs_set_port
,
2536 .connect
= xs_connect
,
2537 .buf_alloc
= rpc_malloc
,
2538 .buf_free
= rpc_free
,
2539 .send_request
= xs_udp_send_request
,
2540 .set_retrans_timeout
= xprt_set_retrans_timeout_rtt
,
2541 .timer
= xs_udp_timer
,
2542 .release_request
= xprt_release_rqst_cong
,
2544 .destroy
= xs_destroy
,
2545 .print_stats
= xs_udp_print_stats
,
2548 static struct rpc_xprt_ops xs_tcp_ops
= {
2549 .reserve_xprt
= xprt_reserve_xprt
,
2550 .release_xprt
= xs_tcp_release_xprt
,
2551 .alloc_slot
= xprt_lock_and_alloc_slot
,
2552 .rpcbind
= rpcb_getport_async
,
2553 .set_port
= xs_set_port
,
2554 .connect
= xs_connect
,
2555 .buf_alloc
= rpc_malloc
,
2556 .buf_free
= rpc_free
,
2557 .send_request
= xs_tcp_send_request
,
2558 .set_retrans_timeout
= xprt_set_retrans_timeout_def
,
2559 .close
= xs_tcp_close
,
2560 .destroy
= xs_destroy
,
2561 .print_stats
= xs_tcp_print_stats
,
2565 * The rpc_xprt_ops for the server backchannel
2568 static struct rpc_xprt_ops bc_tcp_ops
= {
2569 .reserve_xprt
= xprt_reserve_xprt
,
2570 .release_xprt
= xprt_release_xprt
,
2571 .alloc_slot
= xprt_alloc_slot
,
2572 .buf_alloc
= bc_malloc
,
2573 .buf_free
= bc_free
,
2574 .send_request
= bc_send_request
,
2575 .set_retrans_timeout
= xprt_set_retrans_timeout_def
,
2577 .destroy
= bc_destroy
,
2578 .print_stats
= xs_tcp_print_stats
,
2581 static int xs_init_anyaddr(const int family
, struct sockaddr
*sap
)
2583 static const struct sockaddr_in sin
= {
2584 .sin_family
= AF_INET
,
2585 .sin_addr
.s_addr
= htonl(INADDR_ANY
),
2587 static const struct sockaddr_in6 sin6
= {
2588 .sin6_family
= AF_INET6
,
2589 .sin6_addr
= IN6ADDR_ANY_INIT
,
2596 memcpy(sap
, &sin
, sizeof(sin
));
2599 memcpy(sap
, &sin6
, sizeof(sin6
));
2602 dprintk("RPC: %s: Bad address family\n", __func__
);
2603 return -EAFNOSUPPORT
;
2608 static struct rpc_xprt
*xs_setup_xprt(struct xprt_create
*args
,
2609 unsigned int slot_table_size
,
2610 unsigned int max_slot_table_size
)
2612 struct rpc_xprt
*xprt
;
2613 struct sock_xprt
*new;
2615 if (args
->addrlen
> sizeof(xprt
->addr
)) {
2616 dprintk("RPC: xs_setup_xprt: address too large\n");
2617 return ERR_PTR(-EBADF
);
2620 xprt
= xprt_alloc(args
->net
, sizeof(*new), slot_table_size
,
2621 max_slot_table_size
);
2623 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2625 return ERR_PTR(-ENOMEM
);
2628 new = container_of(xprt
, struct sock_xprt
, xprt
);
2629 memcpy(&xprt
->addr
, args
->dstaddr
, args
->addrlen
);
2630 xprt
->addrlen
= args
->addrlen
;
2632 memcpy(&new->srcaddr
, args
->srcaddr
, args
->addrlen
);
2635 err
= xs_init_anyaddr(args
->dstaddr
->sa_family
,
2636 (struct sockaddr
*)&new->srcaddr
);
2639 return ERR_PTR(err
);
2646 static const struct rpc_timeout xs_local_default_timeout
= {
2647 .to_initval
= 10 * HZ
,
2648 .to_maxval
= 10 * HZ
,
2653 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2654 * @args: rpc transport creation arguments
2656 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2658 static struct rpc_xprt
*xs_setup_local(struct xprt_create
*args
)
2660 struct sockaddr_un
*sun
= (struct sockaddr_un
*)args
->dstaddr
;
2661 struct sock_xprt
*transport
;
2662 struct rpc_xprt
*xprt
;
2663 struct rpc_xprt
*ret
;
2665 xprt
= xs_setup_xprt(args
, xprt_tcp_slot_table_entries
,
2666 xprt_max_tcp_slot_table_entries
);
2669 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2672 xprt
->tsh_size
= sizeof(rpc_fraghdr
) / sizeof(u32
);
2673 xprt
->max_payload
= RPC_MAX_FRAGMENT_SIZE
;
2675 xprt
->bind_timeout
= XS_BIND_TO
;
2676 xprt
->reestablish_timeout
= XS_TCP_INIT_REEST_TO
;
2677 xprt
->idle_timeout
= XS_IDLE_DISC_TO
;
2679 xprt
->ops
= &xs_local_ops
;
2680 xprt
->timeout
= &xs_local_default_timeout
;
2682 INIT_DELAYED_WORK(&transport
->connect_worker
,
2683 xs_dummy_setup_socket
);
2685 switch (sun
->sun_family
) {
2687 if (sun
->sun_path
[0] != '/') {
2688 dprintk("RPC: bad AF_LOCAL address: %s\n",
2690 ret
= ERR_PTR(-EINVAL
);
2693 xprt_set_bound(xprt
);
2694 xs_format_peer_addresses(xprt
, "local", RPCBIND_NETID_LOCAL
);
2695 ret
= ERR_PTR(xs_local_setup_socket(transport
));
2700 ret
= ERR_PTR(-EAFNOSUPPORT
);
2704 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2705 xprt
->address_strings
[RPC_DISPLAY_ADDR
]);
2707 if (try_module_get(THIS_MODULE
))
2709 ret
= ERR_PTR(-EINVAL
);
2715 static const struct rpc_timeout xs_udp_default_timeout
= {
2716 .to_initval
= 5 * HZ
,
2717 .to_maxval
= 30 * HZ
,
2718 .to_increment
= 5 * HZ
,
2723 * xs_setup_udp - Set up transport to use a UDP socket
2724 * @args: rpc transport creation arguments
2727 static struct rpc_xprt
*xs_setup_udp(struct xprt_create
*args
)
2729 struct sockaddr
*addr
= args
->dstaddr
;
2730 struct rpc_xprt
*xprt
;
2731 struct sock_xprt
*transport
;
2732 struct rpc_xprt
*ret
;
2734 xprt
= xs_setup_xprt(args
, xprt_udp_slot_table_entries
,
2735 xprt_udp_slot_table_entries
);
2738 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2740 xprt
->prot
= IPPROTO_UDP
;
2742 /* XXX: header size can vary due to auth type, IPv6, etc. */
2743 xprt
->max_payload
= (1U << 16) - (MAX_HEADER
<< 3);
2745 xprt
->bind_timeout
= XS_BIND_TO
;
2746 xprt
->reestablish_timeout
= XS_UDP_REEST_TO
;
2747 xprt
->idle_timeout
= XS_IDLE_DISC_TO
;
2749 xprt
->ops
= &xs_udp_ops
;
2751 xprt
->timeout
= &xs_udp_default_timeout
;
2753 switch (addr
->sa_family
) {
2755 if (((struct sockaddr_in
*)addr
)->sin_port
!= htons(0))
2756 xprt_set_bound(xprt
);
2758 INIT_DELAYED_WORK(&transport
->connect_worker
,
2759 xs_udp_setup_socket
);
2760 xs_format_peer_addresses(xprt
, "udp", RPCBIND_NETID_UDP
);
2763 if (((struct sockaddr_in6
*)addr
)->sin6_port
!= htons(0))
2764 xprt_set_bound(xprt
);
2766 INIT_DELAYED_WORK(&transport
->connect_worker
,
2767 xs_udp_setup_socket
);
2768 xs_format_peer_addresses(xprt
, "udp", RPCBIND_NETID_UDP6
);
2771 ret
= ERR_PTR(-EAFNOSUPPORT
);
2775 if (xprt_bound(xprt
))
2776 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2777 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
2778 xprt
->address_strings
[RPC_DISPLAY_PORT
],
2779 xprt
->address_strings
[RPC_DISPLAY_PROTO
]);
2781 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2782 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
2783 xprt
->address_strings
[RPC_DISPLAY_PROTO
]);
2785 if (try_module_get(THIS_MODULE
))
2787 ret
= ERR_PTR(-EINVAL
);
2793 static const struct rpc_timeout xs_tcp_default_timeout
= {
2794 .to_initval
= 60 * HZ
,
2795 .to_maxval
= 60 * HZ
,
2800 * xs_setup_tcp - Set up transport to use a TCP socket
2801 * @args: rpc transport creation arguments
2804 static struct rpc_xprt
*xs_setup_tcp(struct xprt_create
*args
)
2806 struct sockaddr
*addr
= args
->dstaddr
;
2807 struct rpc_xprt
*xprt
;
2808 struct sock_xprt
*transport
;
2809 struct rpc_xprt
*ret
;
2810 unsigned int max_slot_table_size
= xprt_max_tcp_slot_table_entries
;
2812 if (args
->flags
& XPRT_CREATE_INFINITE_SLOTS
)
2813 max_slot_table_size
= RPC_MAX_SLOT_TABLE_LIMIT
;
2815 xprt
= xs_setup_xprt(args
, xprt_tcp_slot_table_entries
,
2816 max_slot_table_size
);
2819 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2821 xprt
->prot
= IPPROTO_TCP
;
2822 xprt
->tsh_size
= sizeof(rpc_fraghdr
) / sizeof(u32
);
2823 xprt
->max_payload
= RPC_MAX_FRAGMENT_SIZE
;
2825 xprt
->bind_timeout
= XS_BIND_TO
;
2826 xprt
->reestablish_timeout
= XS_TCP_INIT_REEST_TO
;
2827 xprt
->idle_timeout
= XS_IDLE_DISC_TO
;
2829 xprt
->ops
= &xs_tcp_ops
;
2830 xprt
->timeout
= &xs_tcp_default_timeout
;
2832 switch (addr
->sa_family
) {
2834 if (((struct sockaddr_in
*)addr
)->sin_port
!= htons(0))
2835 xprt_set_bound(xprt
);
2837 INIT_DELAYED_WORK(&transport
->connect_worker
,
2838 xs_tcp_setup_socket
);
2839 xs_format_peer_addresses(xprt
, "tcp", RPCBIND_NETID_TCP
);
2842 if (((struct sockaddr_in6
*)addr
)->sin6_port
!= htons(0))
2843 xprt_set_bound(xprt
);
2845 INIT_DELAYED_WORK(&transport
->connect_worker
,
2846 xs_tcp_setup_socket
);
2847 xs_format_peer_addresses(xprt
, "tcp", RPCBIND_NETID_TCP6
);
2850 ret
= ERR_PTR(-EAFNOSUPPORT
);
2854 if (xprt_bound(xprt
))
2855 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2856 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
2857 xprt
->address_strings
[RPC_DISPLAY_PORT
],
2858 xprt
->address_strings
[RPC_DISPLAY_PROTO
]);
2860 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2861 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
2862 xprt
->address_strings
[RPC_DISPLAY_PROTO
]);
2865 if (try_module_get(THIS_MODULE
))
2867 ret
= ERR_PTR(-EINVAL
);
2874 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2875 * @args: rpc transport creation arguments
2878 static struct rpc_xprt
*xs_setup_bc_tcp(struct xprt_create
*args
)
2880 struct sockaddr
*addr
= args
->dstaddr
;
2881 struct rpc_xprt
*xprt
;
2882 struct sock_xprt
*transport
;
2883 struct svc_sock
*bc_sock
;
2884 struct rpc_xprt
*ret
;
2886 if (args
->bc_xprt
->xpt_bc_xprt
) {
2888 * This server connection already has a backchannel
2889 * transport; we can't create a new one, as we wouldn't
2890 * be able to match replies based on xid any more. So,
2891 * reuse the already-existing one:
2893 return args
->bc_xprt
->xpt_bc_xprt
;
2895 xprt
= xs_setup_xprt(args
, xprt_tcp_slot_table_entries
,
2896 xprt_tcp_slot_table_entries
);
2899 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2901 xprt
->prot
= IPPROTO_TCP
;
2902 xprt
->tsh_size
= sizeof(rpc_fraghdr
) / sizeof(u32
);
2903 xprt
->max_payload
= RPC_MAX_FRAGMENT_SIZE
;
2904 xprt
->timeout
= &xs_tcp_default_timeout
;
2907 xprt_set_bound(xprt
);
2908 xprt
->bind_timeout
= 0;
2909 xprt
->reestablish_timeout
= 0;
2910 xprt
->idle_timeout
= 0;
2912 xprt
->ops
= &bc_tcp_ops
;
2914 switch (addr
->sa_family
) {
2916 xs_format_peer_addresses(xprt
, "tcp",
2920 xs_format_peer_addresses(xprt
, "tcp",
2921 RPCBIND_NETID_TCP6
);
2924 ret
= ERR_PTR(-EAFNOSUPPORT
);
2928 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2929 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
2930 xprt
->address_strings
[RPC_DISPLAY_PORT
],
2931 xprt
->address_strings
[RPC_DISPLAY_PROTO
]);
2934 * Once we've associated a backchannel xprt with a connection,
2935 * we want to keep it around as long as long as the connection
2936 * lasts, in case we need to start using it for a backchannel
2937 * again; this reference won't be dropped until bc_xprt is
2941 args
->bc_xprt
->xpt_bc_xprt
= xprt
;
2942 xprt
->bc_xprt
= args
->bc_xprt
;
2943 bc_sock
= container_of(args
->bc_xprt
, struct svc_sock
, sk_xprt
);
2944 transport
->sock
= bc_sock
->sk_sock
;
2945 transport
->inet
= bc_sock
->sk_sk
;
2948 * Since we don't want connections for the backchannel, we set
2949 * the xprt status to connected
2951 xprt_set_connected(xprt
);
2954 if (try_module_get(THIS_MODULE
))
2957 ret
= ERR_PTR(-EINVAL
);
2963 static struct xprt_class xs_local_transport
= {
2964 .list
= LIST_HEAD_INIT(xs_local_transport
.list
),
2965 .name
= "named UNIX socket",
2966 .owner
= THIS_MODULE
,
2967 .ident
= XPRT_TRANSPORT_LOCAL
,
2968 .setup
= xs_setup_local
,
2971 static struct xprt_class xs_udp_transport
= {
2972 .list
= LIST_HEAD_INIT(xs_udp_transport
.list
),
2974 .owner
= THIS_MODULE
,
2975 .ident
= XPRT_TRANSPORT_UDP
,
2976 .setup
= xs_setup_udp
,
2979 static struct xprt_class xs_tcp_transport
= {
2980 .list
= LIST_HEAD_INIT(xs_tcp_transport
.list
),
2982 .owner
= THIS_MODULE
,
2983 .ident
= XPRT_TRANSPORT_TCP
,
2984 .setup
= xs_setup_tcp
,
2987 static struct xprt_class xs_bc_tcp_transport
= {
2988 .list
= LIST_HEAD_INIT(xs_bc_tcp_transport
.list
),
2989 .name
= "tcp NFSv4.1 backchannel",
2990 .owner
= THIS_MODULE
,
2991 .ident
= XPRT_TRANSPORT_BC_TCP
,
2992 .setup
= xs_setup_bc_tcp
,
2996 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2999 int init_socket_xprt(void)
3002 if (!sunrpc_table_header
)
3003 sunrpc_table_header
= register_sysctl_table(sunrpc_table
);
3006 xprt_register_transport(&xs_local_transport
);
3007 xprt_register_transport(&xs_udp_transport
);
3008 xprt_register_transport(&xs_tcp_transport
);
3009 xprt_register_transport(&xs_bc_tcp_transport
);
3015 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3018 void cleanup_socket_xprt(void)
3021 if (sunrpc_table_header
) {
3022 unregister_sysctl_table(sunrpc_table_header
);
3023 sunrpc_table_header
= NULL
;
3027 xprt_unregister_transport(&xs_local_transport
);
3028 xprt_unregister_transport(&xs_udp_transport
);
3029 xprt_unregister_transport(&xs_tcp_transport
);
3030 xprt_unregister_transport(&xs_bc_tcp_transport
);
3033 static int param_set_uint_minmax(const char *val
,
3034 const struct kernel_param
*kp
,
3035 unsigned int min
, unsigned int max
)
3042 ret
= strict_strtoul(val
, 0, &num
);
3043 if (ret
== -EINVAL
|| num
< min
|| num
> max
)
3045 *((unsigned int *)kp
->arg
) = num
;
3049 static int param_set_portnr(const char *val
, const struct kernel_param
*kp
)
3051 return param_set_uint_minmax(val
, kp
,
3056 static struct kernel_param_ops param_ops_portnr
= {
3057 .set
= param_set_portnr
,
3058 .get
= param_get_uint
,
3061 #define param_check_portnr(name, p) \
3062 __param_check(name, p, unsigned int);
3064 module_param_named(min_resvport
, xprt_min_resvport
, portnr
, 0644);
3065 module_param_named(max_resvport
, xprt_max_resvport
, portnr
, 0644);
3067 static int param_set_slot_table_size(const char *val
,
3068 const struct kernel_param
*kp
)
3070 return param_set_uint_minmax(val
, kp
,
3072 RPC_MAX_SLOT_TABLE
);
3075 static struct kernel_param_ops param_ops_slot_table_size
= {
3076 .set
= param_set_slot_table_size
,
3077 .get
= param_get_uint
,
3080 #define param_check_slot_table_size(name, p) \
3081 __param_check(name, p, unsigned int);
3083 static int param_set_max_slot_table_size(const char *val
,
3084 const struct kernel_param
*kp
)
3086 return param_set_uint_minmax(val
, kp
,
3088 RPC_MAX_SLOT_TABLE_LIMIT
);
3091 static struct kernel_param_ops param_ops_max_slot_table_size
= {
3092 .set
= param_set_max_slot_table_size
,
3093 .get
= param_get_uint
,
3096 #define param_check_max_slot_table_size(name, p) \
3097 __param_check(name, p, unsigned int);
3099 module_param_named(tcp_slot_table_entries
, xprt_tcp_slot_table_entries
,
3100 slot_table_size
, 0644);
3101 module_param_named(tcp_max_slot_table_entries
, xprt_max_tcp_slot_table_entries
,
3102 max_slot_table_size
, 0644);
3103 module_param_named(udp_slot_table_entries
, xprt_udp_slot_table_entries
,
3104 slot_table_size
, 0644);