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/slab.h>
23 #include <linux/module.h>
24 #include <linux/capability.h>
25 #include <linux/pagemap.h>
26 #include <linux/errno.h>
27 #include <linux/socket.h>
29 #include <linux/net.h>
31 #include <linux/udp.h>
32 #include <linux/tcp.h>
33 #include <linux/sunrpc/clnt.h>
34 #include <linux/sunrpc/sched.h>
35 #include <linux/sunrpc/svcsock.h>
36 #include <linux/sunrpc/xprtsock.h>
37 #include <linux/file.h>
38 #ifdef CONFIG_NFS_V4_1
39 #include <linux/sunrpc/bc_xprt.h>
43 #include <net/checksum.h>
51 unsigned int xprt_udp_slot_table_entries
= RPC_DEF_SLOT_TABLE
;
52 unsigned int xprt_tcp_slot_table_entries
= RPC_DEF_SLOT_TABLE
;
54 unsigned int xprt_min_resvport
= RPC_DEF_MIN_RESVPORT
;
55 unsigned int xprt_max_resvport
= RPC_DEF_MAX_RESVPORT
;
57 #define XS_TCP_LINGER_TO (15U * HZ)
58 static unsigned int xs_tcp_fin_timeout __read_mostly
= XS_TCP_LINGER_TO
;
61 * We can register our own files under /proc/sys/sunrpc by
62 * calling register_sysctl_table() again. The files in that
63 * directory become the union of all files registered there.
65 * We simply need to make sure that we don't collide with
66 * someone else's file names!
71 static unsigned int min_slot_table_size
= RPC_MIN_SLOT_TABLE
;
72 static unsigned int max_slot_table_size
= RPC_MAX_SLOT_TABLE
;
73 static unsigned int xprt_min_resvport_limit
= RPC_MIN_RESVPORT
;
74 static unsigned int xprt_max_resvport_limit
= RPC_MAX_RESVPORT
;
76 static struct ctl_table_header
*sunrpc_table_header
;
79 * FIXME: changing the UDP slot table size should also resize the UDP
80 * socket buffers for existing UDP transports
82 static ctl_table xs_tunables_table
[] = {
84 .procname
= "udp_slot_table_entries",
85 .data
= &xprt_udp_slot_table_entries
,
86 .maxlen
= sizeof(unsigned int),
88 .proc_handler
= proc_dointvec_minmax
,
89 .extra1
= &min_slot_table_size
,
90 .extra2
= &max_slot_table_size
93 .procname
= "tcp_slot_table_entries",
94 .data
= &xprt_tcp_slot_table_entries
,
95 .maxlen
= sizeof(unsigned int),
97 .proc_handler
= proc_dointvec_minmax
,
98 .extra1
= &min_slot_table_size
,
99 .extra2
= &max_slot_table_size
102 .procname
= "min_resvport",
103 .data
= &xprt_min_resvport
,
104 .maxlen
= sizeof(unsigned int),
106 .proc_handler
= proc_dointvec_minmax
,
107 .extra1
= &xprt_min_resvport_limit
,
108 .extra2
= &xprt_max_resvport_limit
111 .procname
= "max_resvport",
112 .data
= &xprt_max_resvport
,
113 .maxlen
= sizeof(unsigned int),
115 .proc_handler
= proc_dointvec_minmax
,
116 .extra1
= &xprt_min_resvport_limit
,
117 .extra2
= &xprt_max_resvport_limit
120 .procname
= "tcp_fin_timeout",
121 .data
= &xs_tcp_fin_timeout
,
122 .maxlen
= sizeof(xs_tcp_fin_timeout
),
124 .proc_handler
= proc_dointvec_jiffies
,
129 static ctl_table sunrpc_table
[] = {
131 .procname
= "sunrpc",
133 .child
= xs_tunables_table
141 * Wait duration for a reply from the RPC portmapper.
143 #define XS_BIND_TO (60U * HZ)
146 * Delay if a UDP socket connect error occurs. This is most likely some
147 * kind of resource problem on the local host.
149 #define XS_UDP_REEST_TO (2U * HZ)
152 * The reestablish timeout allows clients to delay for a bit before attempting
153 * to reconnect to a server that just dropped our connection.
155 * We implement an exponential backoff when trying to reestablish a TCP
156 * transport connection with the server. Some servers like to drop a TCP
157 * connection when they are overworked, so we start with a short timeout and
158 * increase over time if the server is down or not responding.
160 #define XS_TCP_INIT_REEST_TO (3U * HZ)
161 #define XS_TCP_MAX_REEST_TO (5U * 60 * HZ)
164 * TCP idle timeout; client drops the transport socket if it is idle
165 * for this long. Note that we also timeout UDP sockets to prevent
166 * holding port numbers when there is no RPC traffic.
168 #define XS_IDLE_DISC_TO (5U * 60 * HZ)
171 # undef RPC_DEBUG_DATA
172 # define RPCDBG_FACILITY RPCDBG_TRANS
175 #ifdef RPC_DEBUG_DATA
176 static void xs_pktdump(char *msg
, u32
*packet
, unsigned int count
)
178 u8
*buf
= (u8
*) packet
;
181 dprintk("RPC: %s\n", msg
);
182 for (j
= 0; j
< count
&& j
< 128; j
+= 4) {
186 dprintk("0x%04x ", j
);
188 dprintk("%02x%02x%02x%02x ",
189 buf
[j
], buf
[j
+1], buf
[j
+2], buf
[j
+3]);
194 static inline void xs_pktdump(char *msg
, u32
*packet
, unsigned int count
)
201 struct rpc_xprt xprt
;
206 struct socket
* sock
;
210 * State of TCP reply receive
219 unsigned long tcp_copied
,
223 * Connection of transports
225 struct delayed_work connect_worker
;
226 struct sockaddr_storage srcaddr
;
227 unsigned short srcport
;
230 * UDP socket buffer size parameters
236 * Saved socket callback addresses
238 void (*old_data_ready
)(struct sock
*, int);
239 void (*old_state_change
)(struct sock
*);
240 void (*old_write_space
)(struct sock
*);
241 void (*old_error_report
)(struct sock
*);
245 * TCP receive state flags
247 #define TCP_RCV_LAST_FRAG (1UL << 0)
248 #define TCP_RCV_COPY_FRAGHDR (1UL << 1)
249 #define TCP_RCV_COPY_XID (1UL << 2)
250 #define TCP_RCV_COPY_DATA (1UL << 3)
251 #define TCP_RCV_READ_CALLDIR (1UL << 4)
252 #define TCP_RCV_COPY_CALLDIR (1UL << 5)
257 #define TCP_RPC_REPLY (1UL << 6)
259 static inline struct sockaddr
*xs_addr(struct rpc_xprt
*xprt
)
261 return (struct sockaddr
*) &xprt
->addr
;
264 static inline struct sockaddr_in
*xs_addr_in(struct rpc_xprt
*xprt
)
266 return (struct sockaddr_in
*) &xprt
->addr
;
269 static inline struct sockaddr_in6
*xs_addr_in6(struct rpc_xprt
*xprt
)
271 return (struct sockaddr_in6
*) &xprt
->addr
;
274 static void xs_format_common_peer_addresses(struct rpc_xprt
*xprt
)
276 struct sockaddr
*sap
= xs_addr(xprt
);
277 struct sockaddr_in6
*sin6
;
278 struct sockaddr_in
*sin
;
281 (void)rpc_ntop(sap
, buf
, sizeof(buf
));
282 xprt
->address_strings
[RPC_DISPLAY_ADDR
] = kstrdup(buf
, GFP_KERNEL
);
284 switch (sap
->sa_family
) {
286 sin
= xs_addr_in(xprt
);
287 snprintf(buf
, sizeof(buf
), "%08x", ntohl(sin
->sin_addr
.s_addr
));
290 sin6
= xs_addr_in6(xprt
);
291 snprintf(buf
, sizeof(buf
), "%pi6", &sin6
->sin6_addr
);
296 xprt
->address_strings
[RPC_DISPLAY_HEX_ADDR
] = kstrdup(buf
, GFP_KERNEL
);
299 static void xs_format_common_peer_ports(struct rpc_xprt
*xprt
)
301 struct sockaddr
*sap
= xs_addr(xprt
);
304 snprintf(buf
, sizeof(buf
), "%u", rpc_get_port(sap
));
305 xprt
->address_strings
[RPC_DISPLAY_PORT
] = kstrdup(buf
, GFP_KERNEL
);
307 snprintf(buf
, sizeof(buf
), "%4hx", rpc_get_port(sap
));
308 xprt
->address_strings
[RPC_DISPLAY_HEX_PORT
] = kstrdup(buf
, GFP_KERNEL
);
311 static void xs_format_peer_addresses(struct rpc_xprt
*xprt
,
312 const char *protocol
,
315 xprt
->address_strings
[RPC_DISPLAY_PROTO
] = protocol
;
316 xprt
->address_strings
[RPC_DISPLAY_NETID
] = netid
;
317 xs_format_common_peer_addresses(xprt
);
318 xs_format_common_peer_ports(xprt
);
321 static void xs_update_peer_port(struct rpc_xprt
*xprt
)
323 kfree(xprt
->address_strings
[RPC_DISPLAY_HEX_PORT
]);
324 kfree(xprt
->address_strings
[RPC_DISPLAY_PORT
]);
326 xs_format_common_peer_ports(xprt
);
329 static void xs_free_peer_addresses(struct rpc_xprt
*xprt
)
333 for (i
= 0; i
< RPC_DISPLAY_MAX
; i
++)
335 case RPC_DISPLAY_PROTO
:
336 case RPC_DISPLAY_NETID
:
339 kfree(xprt
->address_strings
[i
]);
343 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
345 static int xs_send_kvec(struct socket
*sock
, struct sockaddr
*addr
, int addrlen
, struct kvec
*vec
, unsigned int base
, int more
)
347 struct msghdr msg
= {
349 .msg_namelen
= addrlen
,
350 .msg_flags
= XS_SENDMSG_FLAGS
| (more
? MSG_MORE
: 0),
353 .iov_base
= vec
->iov_base
+ base
,
354 .iov_len
= vec
->iov_len
- base
,
357 if (iov
.iov_len
!= 0)
358 return kernel_sendmsg(sock
, &msg
, &iov
, 1, iov
.iov_len
);
359 return kernel_sendmsg(sock
, &msg
, NULL
, 0, 0);
362 static int xs_send_pagedata(struct socket
*sock
, struct xdr_buf
*xdr
, unsigned int base
, int more
)
365 unsigned int remainder
;
368 remainder
= xdr
->page_len
- base
;
369 base
+= xdr
->page_base
;
370 ppage
= xdr
->pages
+ (base
>> PAGE_SHIFT
);
373 unsigned int len
= min_t(unsigned int, PAGE_SIZE
- base
, remainder
);
374 int flags
= XS_SENDMSG_FLAGS
;
377 if (remainder
!= 0 || more
)
379 err
= sock
->ops
->sendpage(sock
, *ppage
, base
, len
, flags
);
380 if (remainder
== 0 || err
!= len
)
394 * xs_sendpages - write pages directly to a socket
395 * @sock: socket to send on
396 * @addr: UDP only -- address of destination
397 * @addrlen: UDP only -- length of destination address
398 * @xdr: buffer containing this request
399 * @base: starting position in the buffer
402 static int xs_sendpages(struct socket
*sock
, struct sockaddr
*addr
, int addrlen
, struct xdr_buf
*xdr
, unsigned int base
)
404 unsigned int remainder
= xdr
->len
- base
;
410 clear_bit(SOCK_ASYNC_NOSPACE
, &sock
->flags
);
416 if (base
< xdr
->head
[0].iov_len
|| addr
!= NULL
) {
417 unsigned int len
= xdr
->head
[0].iov_len
- base
;
419 err
= xs_send_kvec(sock
, addr
, addrlen
, &xdr
->head
[0], base
, remainder
!= 0);
420 if (remainder
== 0 || err
!= len
)
425 base
-= xdr
->head
[0].iov_len
;
427 if (base
< xdr
->page_len
) {
428 unsigned int len
= xdr
->page_len
- base
;
430 err
= xs_send_pagedata(sock
, xdr
, base
, remainder
!= 0);
431 if (remainder
== 0 || err
!= len
)
436 base
-= xdr
->page_len
;
438 if (base
>= xdr
->tail
[0].iov_len
)
440 err
= xs_send_kvec(sock
, NULL
, 0, &xdr
->tail
[0], base
, 0);
449 static void xs_nospace_callback(struct rpc_task
*task
)
451 struct sock_xprt
*transport
= container_of(task
->tk_rqstp
->rq_xprt
, struct sock_xprt
, xprt
);
453 transport
->inet
->sk_write_pending
--;
454 clear_bit(SOCK_ASYNC_NOSPACE
, &transport
->sock
->flags
);
458 * xs_nospace - place task on wait queue if transmit was incomplete
459 * @task: task to put to sleep
462 static int xs_nospace(struct rpc_task
*task
)
464 struct rpc_rqst
*req
= task
->tk_rqstp
;
465 struct rpc_xprt
*xprt
= req
->rq_xprt
;
466 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
469 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
470 task
->tk_pid
, req
->rq_slen
- req
->rq_bytes_sent
,
473 /* Protect against races with write_space */
474 spin_lock_bh(&xprt
->transport_lock
);
476 /* Don't race with disconnect */
477 if (xprt_connected(xprt
)) {
478 if (test_bit(SOCK_ASYNC_NOSPACE
, &transport
->sock
->flags
)) {
481 * Notify TCP that we're limited by the application
484 set_bit(SOCK_NOSPACE
, &transport
->sock
->flags
);
485 transport
->inet
->sk_write_pending
++;
486 /* ...and wait for more buffer space */
487 xprt_wait_for_buffer_space(task
, xs_nospace_callback
);
490 clear_bit(SOCK_ASYNC_NOSPACE
, &transport
->sock
->flags
);
494 spin_unlock_bh(&xprt
->transport_lock
);
499 * xs_udp_send_request - write an RPC request to a UDP socket
500 * @task: address of RPC task that manages the state of an RPC request
503 * 0: The request has been sent
504 * EAGAIN: The socket was blocked, please call again later to
505 * complete the request
506 * ENOTCONN: Caller needs to invoke connect logic then call again
507 * other: Some other error occured, the request was not sent
509 static int xs_udp_send_request(struct rpc_task
*task
)
511 struct rpc_rqst
*req
= task
->tk_rqstp
;
512 struct rpc_xprt
*xprt
= req
->rq_xprt
;
513 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
514 struct xdr_buf
*xdr
= &req
->rq_snd_buf
;
517 xs_pktdump("packet data:",
518 req
->rq_svec
->iov_base
,
519 req
->rq_svec
->iov_len
);
521 if (!xprt_bound(xprt
))
523 status
= xs_sendpages(transport
->sock
,
528 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
529 xdr
->len
- req
->rq_bytes_sent
, status
);
532 req
->rq_xmit_bytes_sent
+= status
;
533 if (status
>= req
->rq_slen
)
535 /* Still some bytes left; set up for a retry later. */
542 /* Should we call xs_close() here? */
545 status
= xs_nospace(task
);
548 dprintk("RPC: sendmsg returned unrecognized error %d\n",
553 /* When the server has died, an ICMP port unreachable message
554 * prompts ECONNREFUSED. */
555 clear_bit(SOCK_ASYNC_NOSPACE
, &transport
->sock
->flags
);
562 * xs_tcp_shutdown - gracefully shut down a TCP socket
565 * Initiates a graceful shutdown of the TCP socket by calling the
566 * equivalent of shutdown(SHUT_WR);
568 static void xs_tcp_shutdown(struct rpc_xprt
*xprt
)
570 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
571 struct socket
*sock
= transport
->sock
;
574 kernel_sock_shutdown(sock
, SHUT_WR
);
577 static inline void xs_encode_tcp_record_marker(struct xdr_buf
*buf
)
579 u32 reclen
= buf
->len
- sizeof(rpc_fraghdr
);
580 rpc_fraghdr
*base
= buf
->head
[0].iov_base
;
581 *base
= htonl(RPC_LAST_STREAM_FRAGMENT
| reclen
);
585 * xs_tcp_send_request - write an RPC request to a TCP socket
586 * @task: address of RPC task that manages the state of an RPC request
589 * 0: The request has been sent
590 * EAGAIN: The socket was blocked, please call again later to
591 * complete the request
592 * ENOTCONN: Caller needs to invoke connect logic then call again
593 * other: Some other error occured, the request was not sent
595 * XXX: In the case of soft timeouts, should we eventually give up
596 * if sendmsg is not able to make progress?
598 static int xs_tcp_send_request(struct rpc_task
*task
)
600 struct rpc_rqst
*req
= task
->tk_rqstp
;
601 struct rpc_xprt
*xprt
= req
->rq_xprt
;
602 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
603 struct xdr_buf
*xdr
= &req
->rq_snd_buf
;
606 xs_encode_tcp_record_marker(&req
->rq_snd_buf
);
608 xs_pktdump("packet data:",
609 req
->rq_svec
->iov_base
,
610 req
->rq_svec
->iov_len
);
612 /* Continue transmitting the packet/record. We must be careful
613 * to cope with writespace callbacks arriving _after_ we have
614 * called sendmsg(). */
616 status
= xs_sendpages(transport
->sock
,
617 NULL
, 0, xdr
, req
->rq_bytes_sent
);
619 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
620 xdr
->len
- req
->rq_bytes_sent
, status
);
622 if (unlikely(status
< 0))
625 /* If we've sent the entire packet, immediately
626 * reset the count of bytes sent. */
627 req
->rq_bytes_sent
+= status
;
628 req
->rq_xmit_bytes_sent
+= status
;
629 if (likely(req
->rq_bytes_sent
>= req
->rq_slen
)) {
630 req
->rq_bytes_sent
= 0;
643 /* Should we call xs_close() here? */
646 status
= xs_nospace(task
);
649 dprintk("RPC: sendmsg returned unrecognized error %d\n",
653 xs_tcp_shutdown(xprt
);
656 clear_bit(SOCK_ASYNC_NOSPACE
, &transport
->sock
->flags
);
663 * xs_tcp_release_xprt - clean up after a tcp transmission
667 * This cleans up if an error causes us to abort the transmission of a request.
668 * In this case, the socket may need to be reset in order to avoid confusing
671 static void xs_tcp_release_xprt(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
673 struct rpc_rqst
*req
;
675 if (task
!= xprt
->snd_task
)
679 req
= task
->tk_rqstp
;
680 if (req
->rq_bytes_sent
== 0)
682 if (req
->rq_bytes_sent
== req
->rq_snd_buf
.len
)
684 set_bit(XPRT_CLOSE_WAIT
, &task
->tk_xprt
->state
);
686 xprt_release_xprt(xprt
, task
);
689 static void xs_save_old_callbacks(struct sock_xprt
*transport
, struct sock
*sk
)
691 transport
->old_data_ready
= sk
->sk_data_ready
;
692 transport
->old_state_change
= sk
->sk_state_change
;
693 transport
->old_write_space
= sk
->sk_write_space
;
694 transport
->old_error_report
= sk
->sk_error_report
;
697 static void xs_restore_old_callbacks(struct sock_xprt
*transport
, struct sock
*sk
)
699 sk
->sk_data_ready
= transport
->old_data_ready
;
700 sk
->sk_state_change
= transport
->old_state_change
;
701 sk
->sk_write_space
= transport
->old_write_space
;
702 sk
->sk_error_report
= transport
->old_error_report
;
705 static void xs_reset_transport(struct sock_xprt
*transport
)
707 struct socket
*sock
= transport
->sock
;
708 struct sock
*sk
= transport
->inet
;
713 transport
->srcport
= 0;
715 write_lock_bh(&sk
->sk_callback_lock
);
716 transport
->inet
= NULL
;
717 transport
->sock
= NULL
;
719 sk
->sk_user_data
= NULL
;
721 xs_restore_old_callbacks(transport
, sk
);
722 write_unlock_bh(&sk
->sk_callback_lock
);
730 * xs_close - close a socket
733 * This is used when all requests are complete; ie, no DRC state remains
734 * on the server we want to save.
736 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
737 * xs_reset_transport() zeroing the socket from underneath a writer.
739 static void xs_close(struct rpc_xprt
*xprt
)
741 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
743 dprintk("RPC: xs_close xprt %p\n", xprt
);
745 xs_reset_transport(transport
);
746 xprt
->reestablish_timeout
= 0;
748 smp_mb__before_clear_bit();
749 clear_bit(XPRT_CONNECTION_ABORT
, &xprt
->state
);
750 clear_bit(XPRT_CLOSE_WAIT
, &xprt
->state
);
751 clear_bit(XPRT_CLOSING
, &xprt
->state
);
752 smp_mb__after_clear_bit();
753 xprt_disconnect_done(xprt
);
756 static void xs_tcp_close(struct rpc_xprt
*xprt
)
758 if (test_and_clear_bit(XPRT_CONNECTION_CLOSE
, &xprt
->state
))
761 xs_tcp_shutdown(xprt
);
765 * xs_destroy - prepare to shutdown a transport
766 * @xprt: doomed transport
769 static void xs_destroy(struct rpc_xprt
*xprt
)
771 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
773 dprintk("RPC: xs_destroy xprt %p\n", xprt
);
775 cancel_delayed_work_sync(&transport
->connect_worker
);
778 xs_free_peer_addresses(xprt
);
780 module_put(THIS_MODULE
);
783 static inline struct rpc_xprt
*xprt_from_sock(struct sock
*sk
)
785 return (struct rpc_xprt
*) sk
->sk_user_data
;
789 * xs_udp_data_ready - "data ready" callback for UDP sockets
790 * @sk: socket with data to read
791 * @len: how much data to read
794 static void xs_udp_data_ready(struct sock
*sk
, int len
)
796 struct rpc_task
*task
;
797 struct rpc_xprt
*xprt
;
798 struct rpc_rqst
*rovr
;
800 int err
, repsize
, copied
;
804 read_lock_bh(&sk
->sk_callback_lock
);
805 dprintk("RPC: xs_udp_data_ready...\n");
806 if (!(xprt
= xprt_from_sock(sk
)))
809 if ((skb
= skb_recv_datagram(sk
, 0, 1, &err
)) == NULL
)
815 repsize
= skb
->len
- sizeof(struct udphdr
);
817 dprintk("RPC: impossible RPC reply size %d!\n", repsize
);
821 /* Copy the XID from the skb... */
822 xp
= skb_header_pointer(skb
, sizeof(struct udphdr
),
823 sizeof(_xid
), &_xid
);
827 /* Look up and lock the request corresponding to the given XID */
828 spin_lock(&xprt
->transport_lock
);
829 rovr
= xprt_lookup_rqst(xprt
, *xp
);
832 task
= rovr
->rq_task
;
834 if ((copied
= rovr
->rq_private_buf
.buflen
) > repsize
)
837 /* Suck it into the iovec, verify checksum if not done by hw. */
838 if (csum_partial_copy_to_xdr(&rovr
->rq_private_buf
, skb
)) {
839 UDPX_INC_STATS_BH(sk
, UDP_MIB_INERRORS
);
843 UDPX_INC_STATS_BH(sk
, UDP_MIB_INDATAGRAMS
);
845 /* Something worked... */
846 dst_confirm(skb_dst(skb
));
848 xprt_adjust_cwnd(task
, copied
);
849 xprt_complete_rqst(task
, copied
);
852 spin_unlock(&xprt
->transport_lock
);
854 skb_free_datagram(sk
, skb
);
856 read_unlock_bh(&sk
->sk_callback_lock
);
859 static inline void xs_tcp_read_fraghdr(struct rpc_xprt
*xprt
, struct xdr_skb_reader
*desc
)
861 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
865 p
= ((char *) &transport
->tcp_fraghdr
) + transport
->tcp_offset
;
866 len
= sizeof(transport
->tcp_fraghdr
) - transport
->tcp_offset
;
867 used
= xdr_skb_read_bits(desc
, p
, len
);
868 transport
->tcp_offset
+= used
;
872 transport
->tcp_reclen
= ntohl(transport
->tcp_fraghdr
);
873 if (transport
->tcp_reclen
& RPC_LAST_STREAM_FRAGMENT
)
874 transport
->tcp_flags
|= TCP_RCV_LAST_FRAG
;
876 transport
->tcp_flags
&= ~TCP_RCV_LAST_FRAG
;
877 transport
->tcp_reclen
&= RPC_FRAGMENT_SIZE_MASK
;
879 transport
->tcp_flags
&= ~TCP_RCV_COPY_FRAGHDR
;
880 transport
->tcp_offset
= 0;
882 /* Sanity check of the record length */
883 if (unlikely(transport
->tcp_reclen
< 8)) {
884 dprintk("RPC: invalid TCP record fragment length\n");
885 xprt_force_disconnect(xprt
);
888 dprintk("RPC: reading TCP record fragment of length %d\n",
889 transport
->tcp_reclen
);
892 static void xs_tcp_check_fraghdr(struct sock_xprt
*transport
)
894 if (transport
->tcp_offset
== transport
->tcp_reclen
) {
895 transport
->tcp_flags
|= TCP_RCV_COPY_FRAGHDR
;
896 transport
->tcp_offset
= 0;
897 if (transport
->tcp_flags
& TCP_RCV_LAST_FRAG
) {
898 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
899 transport
->tcp_flags
|= TCP_RCV_COPY_XID
;
900 transport
->tcp_copied
= 0;
905 static inline void xs_tcp_read_xid(struct sock_xprt
*transport
, struct xdr_skb_reader
*desc
)
910 len
= sizeof(transport
->tcp_xid
) - transport
->tcp_offset
;
911 dprintk("RPC: reading XID (%Zu bytes)\n", len
);
912 p
= ((char *) &transport
->tcp_xid
) + transport
->tcp_offset
;
913 used
= xdr_skb_read_bits(desc
, p
, len
);
914 transport
->tcp_offset
+= used
;
917 transport
->tcp_flags
&= ~TCP_RCV_COPY_XID
;
918 transport
->tcp_flags
|= TCP_RCV_READ_CALLDIR
;
919 transport
->tcp_copied
= 4;
920 dprintk("RPC: reading %s XID %08x\n",
921 (transport
->tcp_flags
& TCP_RPC_REPLY
) ? "reply for"
923 ntohl(transport
->tcp_xid
));
924 xs_tcp_check_fraghdr(transport
);
927 static inline void xs_tcp_read_calldir(struct sock_xprt
*transport
,
928 struct xdr_skb_reader
*desc
)
935 * We want transport->tcp_offset to be 8 at the end of this routine
936 * (4 bytes for the xid and 4 bytes for the call/reply flag).
937 * When this function is called for the first time,
938 * transport->tcp_offset is 4 (after having already read the xid).
940 offset
= transport
->tcp_offset
- sizeof(transport
->tcp_xid
);
941 len
= sizeof(transport
->tcp_calldir
) - offset
;
942 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len
);
943 p
= ((char *) &transport
->tcp_calldir
) + offset
;
944 used
= xdr_skb_read_bits(desc
, p
, len
);
945 transport
->tcp_offset
+= used
;
948 transport
->tcp_flags
&= ~TCP_RCV_READ_CALLDIR
;
950 * We don't yet have the XDR buffer, so we will write the calldir
951 * out after we get the buffer from the 'struct rpc_rqst'
953 switch (ntohl(transport
->tcp_calldir
)) {
955 transport
->tcp_flags
|= TCP_RCV_COPY_CALLDIR
;
956 transport
->tcp_flags
|= TCP_RCV_COPY_DATA
;
957 transport
->tcp_flags
|= TCP_RPC_REPLY
;
960 transport
->tcp_flags
|= TCP_RCV_COPY_CALLDIR
;
961 transport
->tcp_flags
|= TCP_RCV_COPY_DATA
;
962 transport
->tcp_flags
&= ~TCP_RPC_REPLY
;
965 dprintk("RPC: invalid request message type\n");
966 xprt_force_disconnect(&transport
->xprt
);
968 xs_tcp_check_fraghdr(transport
);
971 static inline void xs_tcp_read_common(struct rpc_xprt
*xprt
,
972 struct xdr_skb_reader
*desc
,
973 struct rpc_rqst
*req
)
975 struct sock_xprt
*transport
=
976 container_of(xprt
, struct sock_xprt
, xprt
);
977 struct xdr_buf
*rcvbuf
;
981 rcvbuf
= &req
->rq_private_buf
;
983 if (transport
->tcp_flags
& TCP_RCV_COPY_CALLDIR
) {
985 * Save the RPC direction in the XDR buffer
987 memcpy(rcvbuf
->head
[0].iov_base
+ transport
->tcp_copied
,
988 &transport
->tcp_calldir
,
989 sizeof(transport
->tcp_calldir
));
990 transport
->tcp_copied
+= sizeof(transport
->tcp_calldir
);
991 transport
->tcp_flags
&= ~TCP_RCV_COPY_CALLDIR
;
995 if (len
> transport
->tcp_reclen
- transport
->tcp_offset
) {
996 struct xdr_skb_reader my_desc
;
998 len
= transport
->tcp_reclen
- transport
->tcp_offset
;
999 memcpy(&my_desc
, desc
, sizeof(my_desc
));
1000 my_desc
.count
= len
;
1001 r
= xdr_partial_copy_from_skb(rcvbuf
, transport
->tcp_copied
,
1002 &my_desc
, xdr_skb_read_bits
);
1006 r
= xdr_partial_copy_from_skb(rcvbuf
, transport
->tcp_copied
,
1007 desc
, xdr_skb_read_bits
);
1010 transport
->tcp_copied
+= r
;
1011 transport
->tcp_offset
+= r
;
1014 /* Error when copying to the receive buffer,
1015 * usually because we weren't able to allocate
1016 * additional buffer pages. All we can do now
1017 * is turn off TCP_RCV_COPY_DATA, so the request
1018 * will not receive any additional updates,
1020 * Any remaining data from this record will
1023 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
1024 dprintk("RPC: XID %08x truncated request\n",
1025 ntohl(transport
->tcp_xid
));
1026 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1027 "tcp_offset = %u, tcp_reclen = %u\n",
1028 xprt
, transport
->tcp_copied
,
1029 transport
->tcp_offset
, transport
->tcp_reclen
);
1033 dprintk("RPC: XID %08x read %Zd bytes\n",
1034 ntohl(transport
->tcp_xid
), r
);
1035 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1036 "tcp_reclen = %u\n", xprt
, transport
->tcp_copied
,
1037 transport
->tcp_offset
, transport
->tcp_reclen
);
1039 if (transport
->tcp_copied
== req
->rq_private_buf
.buflen
)
1040 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
1041 else if (transport
->tcp_offset
== transport
->tcp_reclen
) {
1042 if (transport
->tcp_flags
& TCP_RCV_LAST_FRAG
)
1043 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
1048 * Finds the request corresponding to the RPC xid and invokes the common
1049 * tcp read code to read the data.
1051 static inline int xs_tcp_read_reply(struct rpc_xprt
*xprt
,
1052 struct xdr_skb_reader
*desc
)
1054 struct sock_xprt
*transport
=
1055 container_of(xprt
, struct sock_xprt
, xprt
);
1056 struct rpc_rqst
*req
;
1058 dprintk("RPC: read reply XID %08x\n", ntohl(transport
->tcp_xid
));
1060 /* Find and lock the request corresponding to this xid */
1061 spin_lock(&xprt
->transport_lock
);
1062 req
= xprt_lookup_rqst(xprt
, transport
->tcp_xid
);
1064 dprintk("RPC: XID %08x request not found!\n",
1065 ntohl(transport
->tcp_xid
));
1066 spin_unlock(&xprt
->transport_lock
);
1070 xs_tcp_read_common(xprt
, desc
, req
);
1072 if (!(transport
->tcp_flags
& TCP_RCV_COPY_DATA
))
1073 xprt_complete_rqst(req
->rq_task
, transport
->tcp_copied
);
1075 spin_unlock(&xprt
->transport_lock
);
1079 #if defined(CONFIG_NFS_V4_1)
1081 * Obtains an rpc_rqst previously allocated and invokes the common
1082 * tcp read code to read the data. The result is placed in the callback
1084 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1085 * connection and return -1.
1087 static inline int xs_tcp_read_callback(struct rpc_xprt
*xprt
,
1088 struct xdr_skb_reader
*desc
)
1090 struct sock_xprt
*transport
=
1091 container_of(xprt
, struct sock_xprt
, xprt
);
1092 struct rpc_rqst
*req
;
1094 req
= xprt_alloc_bc_request(xprt
);
1096 printk(KERN_WARNING
"Callback slot table overflowed\n");
1097 xprt_force_disconnect(xprt
);
1101 req
->rq_xid
= transport
->tcp_xid
;
1102 dprintk("RPC: read callback XID %08x\n", ntohl(req
->rq_xid
));
1103 xs_tcp_read_common(xprt
, desc
, req
);
1105 if (!(transport
->tcp_flags
& TCP_RCV_COPY_DATA
)) {
1106 struct svc_serv
*bc_serv
= xprt
->bc_serv
;
1109 * Add callback request to callback list. The callback
1110 * service sleeps on the sv_cb_waitq waiting for new
1111 * requests. Wake it up after adding enqueing the
1114 dprintk("RPC: add callback request to list\n");
1115 spin_lock(&bc_serv
->sv_cb_lock
);
1116 list_add(&req
->rq_bc_list
, &bc_serv
->sv_cb_list
);
1117 spin_unlock(&bc_serv
->sv_cb_lock
);
1118 wake_up(&bc_serv
->sv_cb_waitq
);
1121 req
->rq_private_buf
.len
= transport
->tcp_copied
;
1126 static inline int _xs_tcp_read_data(struct rpc_xprt
*xprt
,
1127 struct xdr_skb_reader
*desc
)
1129 struct sock_xprt
*transport
=
1130 container_of(xprt
, struct sock_xprt
, xprt
);
1132 return (transport
->tcp_flags
& TCP_RPC_REPLY
) ?
1133 xs_tcp_read_reply(xprt
, desc
) :
1134 xs_tcp_read_callback(xprt
, desc
);
1137 static inline int _xs_tcp_read_data(struct rpc_xprt
*xprt
,
1138 struct xdr_skb_reader
*desc
)
1140 return xs_tcp_read_reply(xprt
, desc
);
1142 #endif /* CONFIG_NFS_V4_1 */
1145 * Read data off the transport. This can be either an RPC_CALL or an
1146 * RPC_REPLY. Relay the processing to helper functions.
1148 static void xs_tcp_read_data(struct rpc_xprt
*xprt
,
1149 struct xdr_skb_reader
*desc
)
1151 struct sock_xprt
*transport
=
1152 container_of(xprt
, struct sock_xprt
, xprt
);
1154 if (_xs_tcp_read_data(xprt
, desc
) == 0)
1155 xs_tcp_check_fraghdr(transport
);
1158 * The transport_lock protects the request handling.
1159 * There's no need to hold it to update the tcp_flags.
1161 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
1165 static inline void xs_tcp_read_discard(struct sock_xprt
*transport
, struct xdr_skb_reader
*desc
)
1169 len
= transport
->tcp_reclen
- transport
->tcp_offset
;
1170 if (len
> desc
->count
)
1173 desc
->offset
+= len
;
1174 transport
->tcp_offset
+= len
;
1175 dprintk("RPC: discarded %Zu bytes\n", len
);
1176 xs_tcp_check_fraghdr(transport
);
1179 static int xs_tcp_data_recv(read_descriptor_t
*rd_desc
, struct sk_buff
*skb
, unsigned int offset
, size_t len
)
1181 struct rpc_xprt
*xprt
= rd_desc
->arg
.data
;
1182 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1183 struct xdr_skb_reader desc
= {
1189 dprintk("RPC: xs_tcp_data_recv started\n");
1191 /* Read in a new fragment marker if necessary */
1192 /* Can we ever really expect to get completely empty fragments? */
1193 if (transport
->tcp_flags
& TCP_RCV_COPY_FRAGHDR
) {
1194 xs_tcp_read_fraghdr(xprt
, &desc
);
1197 /* Read in the xid if necessary */
1198 if (transport
->tcp_flags
& TCP_RCV_COPY_XID
) {
1199 xs_tcp_read_xid(transport
, &desc
);
1202 /* Read in the call/reply flag */
1203 if (transport
->tcp_flags
& TCP_RCV_READ_CALLDIR
) {
1204 xs_tcp_read_calldir(transport
, &desc
);
1207 /* Read in the request data */
1208 if (transport
->tcp_flags
& TCP_RCV_COPY_DATA
) {
1209 xs_tcp_read_data(xprt
, &desc
);
1212 /* Skip over any trailing bytes on short reads */
1213 xs_tcp_read_discard(transport
, &desc
);
1214 } while (desc
.count
);
1215 dprintk("RPC: xs_tcp_data_recv done\n");
1216 return len
- desc
.count
;
1220 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1221 * @sk: socket with data to read
1222 * @bytes: how much data to read
1225 static void xs_tcp_data_ready(struct sock
*sk
, int bytes
)
1227 struct rpc_xprt
*xprt
;
1228 read_descriptor_t rd_desc
;
1231 dprintk("RPC: xs_tcp_data_ready...\n");
1233 read_lock_bh(&sk
->sk_callback_lock
);
1234 if (!(xprt
= xprt_from_sock(sk
)))
1239 /* Any data means we had a useful conversation, so
1240 * the we don't need to delay the next reconnect
1242 if (xprt
->reestablish_timeout
)
1243 xprt
->reestablish_timeout
= 0;
1245 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1246 rd_desc
.arg
.data
= xprt
;
1248 rd_desc
.count
= 65536;
1249 read
= tcp_read_sock(sk
, &rd_desc
, xs_tcp_data_recv
);
1252 read_unlock_bh(&sk
->sk_callback_lock
);
1256 * Do the equivalent of linger/linger2 handling for dealing with
1257 * broken servers that don't close the socket in a timely
1260 static void xs_tcp_schedule_linger_timeout(struct rpc_xprt
*xprt
,
1261 unsigned long timeout
)
1263 struct sock_xprt
*transport
;
1265 if (xprt_test_and_set_connecting(xprt
))
1267 set_bit(XPRT_CONNECTION_ABORT
, &xprt
->state
);
1268 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1269 queue_delayed_work(rpciod_workqueue
, &transport
->connect_worker
,
1273 static void xs_tcp_cancel_linger_timeout(struct rpc_xprt
*xprt
)
1275 struct sock_xprt
*transport
;
1277 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1279 if (!test_bit(XPRT_CONNECTION_ABORT
, &xprt
->state
) ||
1280 !cancel_delayed_work(&transport
->connect_worker
))
1282 clear_bit(XPRT_CONNECTION_ABORT
, &xprt
->state
);
1283 xprt_clear_connecting(xprt
);
1286 static void xs_sock_mark_closed(struct rpc_xprt
*xprt
)
1288 smp_mb__before_clear_bit();
1289 clear_bit(XPRT_CLOSE_WAIT
, &xprt
->state
);
1290 clear_bit(XPRT_CLOSING
, &xprt
->state
);
1291 smp_mb__after_clear_bit();
1292 /* Mark transport as closed and wake up all pending tasks */
1293 xprt_disconnect_done(xprt
);
1297 * xs_tcp_state_change - callback to handle TCP socket state changes
1298 * @sk: socket whose state has changed
1301 static void xs_tcp_state_change(struct sock
*sk
)
1303 struct rpc_xprt
*xprt
;
1305 read_lock_bh(&sk
->sk_callback_lock
);
1306 if (!(xprt
= xprt_from_sock(sk
)))
1308 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt
);
1309 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1310 sk
->sk_state
, xprt_connected(xprt
),
1311 sock_flag(sk
, SOCK_DEAD
),
1312 sock_flag(sk
, SOCK_ZAPPED
),
1315 switch (sk
->sk_state
) {
1316 case TCP_ESTABLISHED
:
1317 spin_lock(&xprt
->transport_lock
);
1318 if (!xprt_test_and_set_connected(xprt
)) {
1319 struct sock_xprt
*transport
= container_of(xprt
,
1320 struct sock_xprt
, xprt
);
1322 /* Reset TCP record info */
1323 transport
->tcp_offset
= 0;
1324 transport
->tcp_reclen
= 0;
1325 transport
->tcp_copied
= 0;
1326 transport
->tcp_flags
=
1327 TCP_RCV_COPY_FRAGHDR
| TCP_RCV_COPY_XID
;
1329 xprt_wake_pending_tasks(xprt
, -EAGAIN
);
1331 spin_unlock(&xprt
->transport_lock
);
1334 /* The client initiated a shutdown of the socket */
1335 xprt
->connect_cookie
++;
1336 xprt
->reestablish_timeout
= 0;
1337 set_bit(XPRT_CLOSING
, &xprt
->state
);
1338 smp_mb__before_clear_bit();
1339 clear_bit(XPRT_CONNECTED
, &xprt
->state
);
1340 clear_bit(XPRT_CLOSE_WAIT
, &xprt
->state
);
1341 smp_mb__after_clear_bit();
1342 xs_tcp_schedule_linger_timeout(xprt
, xs_tcp_fin_timeout
);
1344 case TCP_CLOSE_WAIT
:
1345 /* The server initiated a shutdown of the socket */
1346 xprt_force_disconnect(xprt
);
1347 xprt
->connect_cookie
++;
1350 * If the server closed down the connection, make sure that
1351 * we back off before reconnecting
1353 if (xprt
->reestablish_timeout
< XS_TCP_INIT_REEST_TO
)
1354 xprt
->reestablish_timeout
= XS_TCP_INIT_REEST_TO
;
1357 set_bit(XPRT_CLOSING
, &xprt
->state
);
1358 xs_tcp_schedule_linger_timeout(xprt
, xs_tcp_fin_timeout
);
1359 smp_mb__before_clear_bit();
1360 clear_bit(XPRT_CONNECTED
, &xprt
->state
);
1361 smp_mb__after_clear_bit();
1364 xs_tcp_cancel_linger_timeout(xprt
);
1365 xs_sock_mark_closed(xprt
);
1368 read_unlock_bh(&sk
->sk_callback_lock
);
1372 * xs_error_report - callback mainly for catching socket errors
1375 static void xs_error_report(struct sock
*sk
)
1377 struct rpc_xprt
*xprt
;
1379 read_lock_bh(&sk
->sk_callback_lock
);
1380 if (!(xprt
= xprt_from_sock(sk
)))
1382 dprintk("RPC: %s client %p...\n"
1384 __func__
, xprt
, sk
->sk_err
);
1385 xprt_wake_pending_tasks(xprt
, -EAGAIN
);
1387 read_unlock_bh(&sk
->sk_callback_lock
);
1390 static void xs_write_space(struct sock
*sk
)
1392 struct socket
*sock
;
1393 struct rpc_xprt
*xprt
;
1395 if (unlikely(!(sock
= sk
->sk_socket
)))
1397 clear_bit(SOCK_NOSPACE
, &sock
->flags
);
1399 if (unlikely(!(xprt
= xprt_from_sock(sk
))))
1401 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE
, &sock
->flags
) == 0)
1404 xprt_write_space(xprt
);
1408 * xs_udp_write_space - callback invoked when socket buffer space
1410 * @sk: socket whose state has changed
1412 * Called when more output buffer space is available for this socket.
1413 * We try not to wake our writers until they can make "significant"
1414 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1415 * with a bunch of small requests.
1417 static void xs_udp_write_space(struct sock
*sk
)
1419 read_lock_bh(&sk
->sk_callback_lock
);
1421 /* from net/core/sock.c:sock_def_write_space */
1422 if (sock_writeable(sk
))
1425 read_unlock_bh(&sk
->sk_callback_lock
);
1429 * xs_tcp_write_space - callback invoked when socket buffer space
1431 * @sk: socket whose state has changed
1433 * Called when more output buffer space is available for this socket.
1434 * We try not to wake our writers until they can make "significant"
1435 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1436 * with a bunch of small requests.
1438 static void xs_tcp_write_space(struct sock
*sk
)
1440 read_lock_bh(&sk
->sk_callback_lock
);
1442 /* from net/core/stream.c:sk_stream_write_space */
1443 if (sk_stream_wspace(sk
) >= sk_stream_min_wspace(sk
))
1446 read_unlock_bh(&sk
->sk_callback_lock
);
1449 static void xs_udp_do_set_buffer_size(struct rpc_xprt
*xprt
)
1451 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1452 struct sock
*sk
= transport
->inet
;
1454 if (transport
->rcvsize
) {
1455 sk
->sk_userlocks
|= SOCK_RCVBUF_LOCK
;
1456 sk
->sk_rcvbuf
= transport
->rcvsize
* xprt
->max_reqs
* 2;
1458 if (transport
->sndsize
) {
1459 sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
;
1460 sk
->sk_sndbuf
= transport
->sndsize
* xprt
->max_reqs
* 2;
1461 sk
->sk_write_space(sk
);
1466 * xs_udp_set_buffer_size - set send and receive limits
1467 * @xprt: generic transport
1468 * @sndsize: requested size of send buffer, in bytes
1469 * @rcvsize: requested size of receive buffer, in bytes
1471 * Set socket send and receive buffer size limits.
1473 static void xs_udp_set_buffer_size(struct rpc_xprt
*xprt
, size_t sndsize
, size_t rcvsize
)
1475 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1477 transport
->sndsize
= 0;
1479 transport
->sndsize
= sndsize
+ 1024;
1480 transport
->rcvsize
= 0;
1482 transport
->rcvsize
= rcvsize
+ 1024;
1484 xs_udp_do_set_buffer_size(xprt
);
1488 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1489 * @task: task that timed out
1491 * Adjust the congestion window after a retransmit timeout has occurred.
1493 static void xs_udp_timer(struct rpc_task
*task
)
1495 xprt_adjust_cwnd(task
, -ETIMEDOUT
);
1498 static unsigned short xs_get_random_port(void)
1500 unsigned short range
= xprt_max_resvport
- xprt_min_resvport
;
1501 unsigned short rand
= (unsigned short) net_random() % range
;
1502 return rand
+ xprt_min_resvport
;
1506 * xs_set_port - reset the port number in the remote endpoint address
1507 * @xprt: generic transport
1508 * @port: new port number
1511 static void xs_set_port(struct rpc_xprt
*xprt
, unsigned short port
)
1513 dprintk("RPC: setting port for xprt %p to %u\n", xprt
, port
);
1515 rpc_set_port(xs_addr(xprt
), port
);
1516 xs_update_peer_port(xprt
);
1519 static unsigned short xs_get_srcport(struct sock_xprt
*transport
)
1521 unsigned short port
= transport
->srcport
;
1523 if (port
== 0 && transport
->xprt
.resvport
)
1524 port
= xs_get_random_port();
1528 static unsigned short xs_next_srcport(struct sock_xprt
*transport
, unsigned short port
)
1530 if (transport
->srcport
!= 0)
1531 transport
->srcport
= 0;
1532 if (!transport
->xprt
.resvport
)
1534 if (port
<= xprt_min_resvport
|| port
> xprt_max_resvport
)
1535 return xprt_max_resvport
;
1538 static int xs_bind(struct sock_xprt
*transport
, struct socket
*sock
)
1540 struct sockaddr_storage myaddr
;
1542 unsigned short port
= xs_get_srcport(transport
);
1543 unsigned short last
;
1545 memcpy(&myaddr
, &transport
->srcaddr
, transport
->xprt
.addrlen
);
1547 rpc_set_port((struct sockaddr
*)&myaddr
, port
);
1548 err
= kernel_bind(sock
, (struct sockaddr
*)&myaddr
,
1549 transport
->xprt
.addrlen
);
1553 transport
->srcport
= port
;
1557 port
= xs_next_srcport(transport
, port
);
1560 } while (err
== -EADDRINUSE
&& nloop
!= 2);
1562 if (myaddr
.ss_family
== AF_INET
)
1563 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__
,
1564 &((struct sockaddr_in
*)&myaddr
)->sin_addr
,
1565 port
, err
? "failed" : "ok", err
);
1567 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__
,
1568 &((struct sockaddr_in6
*)&myaddr
)->sin6_addr
,
1569 port
, err
? "failed" : "ok", err
);
1574 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1575 static struct lock_class_key xs_key
[2];
1576 static struct lock_class_key xs_slock_key
[2];
1578 static inline void xs_reclassify_socket4(struct socket
*sock
)
1580 struct sock
*sk
= sock
->sk
;
1582 BUG_ON(sock_owned_by_user(sk
));
1583 sock_lock_init_class_and_name(sk
, "slock-AF_INET-RPC",
1584 &xs_slock_key
[0], "sk_lock-AF_INET-RPC", &xs_key
[0]);
1587 static inline void xs_reclassify_socket6(struct socket
*sock
)
1589 struct sock
*sk
= sock
->sk
;
1591 BUG_ON(sock_owned_by_user(sk
));
1592 sock_lock_init_class_and_name(sk
, "slock-AF_INET6-RPC",
1593 &xs_slock_key
[1], "sk_lock-AF_INET6-RPC", &xs_key
[1]);
1596 static inline void xs_reclassify_socket(int family
, struct socket
*sock
)
1600 xs_reclassify_socket4(sock
);
1603 xs_reclassify_socket6(sock
);
1608 static inline void xs_reclassify_socket4(struct socket
*sock
)
1612 static inline void xs_reclassify_socket6(struct socket
*sock
)
1616 static inline void xs_reclassify_socket(int family
, struct socket
*sock
)
1621 static struct socket
*xs_create_sock(struct rpc_xprt
*xprt
,
1622 struct sock_xprt
*transport
, int family
, int type
, int protocol
)
1624 struct socket
*sock
;
1627 err
= __sock_create(xprt
->xprt_net
, family
, type
, protocol
, &sock
, 1);
1629 dprintk("RPC: can't create %d transport socket (%d).\n",
1633 xs_reclassify_socket(family
, sock
);
1635 err
= xs_bind(transport
, sock
);
1643 return ERR_PTR(err
);
1646 static void xs_udp_finish_connecting(struct rpc_xprt
*xprt
, struct socket
*sock
)
1648 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1650 if (!transport
->inet
) {
1651 struct sock
*sk
= sock
->sk
;
1653 write_lock_bh(&sk
->sk_callback_lock
);
1655 xs_save_old_callbacks(transport
, sk
);
1657 sk
->sk_user_data
= xprt
;
1658 sk
->sk_data_ready
= xs_udp_data_ready
;
1659 sk
->sk_write_space
= xs_udp_write_space
;
1660 sk
->sk_error_report
= xs_error_report
;
1661 sk
->sk_no_check
= UDP_CSUM_NORCV
;
1662 sk
->sk_allocation
= GFP_ATOMIC
;
1664 xprt_set_connected(xprt
);
1666 /* Reset to new socket */
1667 transport
->sock
= sock
;
1668 transport
->inet
= sk
;
1670 write_unlock_bh(&sk
->sk_callback_lock
);
1672 xs_udp_do_set_buffer_size(xprt
);
1675 static void xs_udp_setup_socket(struct work_struct
*work
)
1677 struct sock_xprt
*transport
=
1678 container_of(work
, struct sock_xprt
, connect_worker
.work
);
1679 struct rpc_xprt
*xprt
= &transport
->xprt
;
1680 struct socket
*sock
= transport
->sock
;
1686 /* Start by resetting any existing state */
1687 xs_reset_transport(transport
);
1688 sock
= xs_create_sock(xprt
, transport
,
1689 xs_addr(xprt
)->sa_family
, SOCK_DGRAM
, IPPROTO_UDP
);
1693 dprintk("RPC: worker connecting xprt %p via %s to "
1694 "%s (port %s)\n", xprt
,
1695 xprt
->address_strings
[RPC_DISPLAY_PROTO
],
1696 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
1697 xprt
->address_strings
[RPC_DISPLAY_PORT
]);
1699 xs_udp_finish_connecting(xprt
, sock
);
1702 xprt_clear_connecting(xprt
);
1703 xprt_wake_pending_tasks(xprt
, status
);
1707 * We need to preserve the port number so the reply cache on the server can
1708 * find our cached RPC replies when we get around to reconnecting.
1710 static void xs_abort_connection(struct sock_xprt
*transport
)
1713 struct sockaddr any
;
1715 dprintk("RPC: disconnecting xprt %p to reuse port\n", transport
);
1718 * Disconnect the transport socket by doing a connect operation
1719 * with AF_UNSPEC. This should return immediately...
1721 memset(&any
, 0, sizeof(any
));
1722 any
.sa_family
= AF_UNSPEC
;
1723 result
= kernel_connect(transport
->sock
, &any
, sizeof(any
), 0);
1725 xs_sock_mark_closed(&transport
->xprt
);
1727 dprintk("RPC: AF_UNSPEC connect return code %d\n",
1731 static void xs_tcp_reuse_connection(struct sock_xprt
*transport
)
1733 unsigned int state
= transport
->inet
->sk_state
;
1735 if (state
== TCP_CLOSE
&& transport
->sock
->state
== SS_UNCONNECTED
) {
1736 /* we don't need to abort the connection if the socket
1737 * hasn't undergone a shutdown
1739 if (transport
->inet
->sk_shutdown
== 0)
1741 dprintk("RPC: %s: TCP_CLOSEd and sk_shutdown set to %d\n",
1742 __func__
, transport
->inet
->sk_shutdown
);
1744 if ((1 << state
) & (TCPF_ESTABLISHED
|TCPF_SYN_SENT
)) {
1745 /* we don't need to abort the connection if the socket
1746 * hasn't undergone a shutdown
1748 if (transport
->inet
->sk_shutdown
== 0)
1750 dprintk("RPC: %s: ESTABLISHED/SYN_SENT "
1751 "sk_shutdown set to %d\n",
1752 __func__
, transport
->inet
->sk_shutdown
);
1754 xs_abort_connection(transport
);
1757 static int xs_tcp_finish_connecting(struct rpc_xprt
*xprt
, struct socket
*sock
)
1759 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1760 int ret
= -ENOTCONN
;
1762 if (!transport
->inet
) {
1763 struct sock
*sk
= sock
->sk
;
1765 write_lock_bh(&sk
->sk_callback_lock
);
1767 xs_save_old_callbacks(transport
, sk
);
1769 sk
->sk_user_data
= xprt
;
1770 sk
->sk_data_ready
= xs_tcp_data_ready
;
1771 sk
->sk_state_change
= xs_tcp_state_change
;
1772 sk
->sk_write_space
= xs_tcp_write_space
;
1773 sk
->sk_error_report
= xs_error_report
;
1774 sk
->sk_allocation
= GFP_ATOMIC
;
1776 /* socket options */
1777 sk
->sk_userlocks
|= SOCK_BINDPORT_LOCK
;
1778 sock_reset_flag(sk
, SOCK_LINGER
);
1779 tcp_sk(sk
)->linger2
= 0;
1780 tcp_sk(sk
)->nonagle
|= TCP_NAGLE_OFF
;
1782 xprt_clear_connected(xprt
);
1784 /* Reset to new socket */
1785 transport
->sock
= sock
;
1786 transport
->inet
= sk
;
1788 write_unlock_bh(&sk
->sk_callback_lock
);
1791 if (!xprt_bound(xprt
))
1794 /* Tell the socket layer to start connecting... */
1795 xprt
->stat
.connect_count
++;
1796 xprt
->stat
.connect_start
= jiffies
;
1797 ret
= kernel_connect(sock
, xs_addr(xprt
), xprt
->addrlen
, O_NONBLOCK
);
1802 xprt
->connect_cookie
++;
1803 if (xprt
->reestablish_timeout
< XS_TCP_INIT_REEST_TO
)
1804 xprt
->reestablish_timeout
= XS_TCP_INIT_REEST_TO
;
1811 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
1812 * @xprt: RPC transport to connect
1813 * @transport: socket transport to connect
1814 * @create_sock: function to create a socket of the correct type
1816 * Invoked by a work queue tasklet.
1818 static void xs_tcp_setup_socket(struct work_struct
*work
)
1820 struct sock_xprt
*transport
=
1821 container_of(work
, struct sock_xprt
, connect_worker
.work
);
1822 struct socket
*sock
= transport
->sock
;
1823 struct rpc_xprt
*xprt
= &transport
->xprt
;
1830 clear_bit(XPRT_CONNECTION_ABORT
, &xprt
->state
);
1831 sock
= xs_create_sock(xprt
, transport
,
1832 xs_addr(xprt
)->sa_family
, SOCK_STREAM
, IPPROTO_TCP
);
1834 status
= PTR_ERR(sock
);
1840 abort_and_exit
= test_and_clear_bit(XPRT_CONNECTION_ABORT
,
1842 /* "close" the socket, preserving the local port */
1843 xs_tcp_reuse_connection(transport
);
1849 dprintk("RPC: worker connecting xprt %p via %s to "
1850 "%s (port %s)\n", xprt
,
1851 xprt
->address_strings
[RPC_DISPLAY_PROTO
],
1852 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
1853 xprt
->address_strings
[RPC_DISPLAY_PORT
]);
1855 status
= xs_tcp_finish_connecting(xprt
, sock
);
1856 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
1857 xprt
, -status
, xprt_connected(xprt
),
1858 sock
->sk
->sk_state
);
1861 printk("%s: connect returned unhandled error %d\n",
1863 case -EADDRNOTAVAIL
:
1864 /* We're probably in TIME_WAIT. Get rid of existing socket,
1867 set_bit(XPRT_CONNECTION_CLOSE
, &xprt
->state
);
1868 xprt_force_disconnect(xprt
);
1873 /* retry with existing socket, after a delay */
1877 xprt_clear_connecting(xprt
);
1880 /* Happens, for instance, if the user specified a link
1881 * local IPv6 address without a scope-id.
1888 xprt_clear_connecting(xprt
);
1889 xprt_wake_pending_tasks(xprt
, status
);
1893 * xs_connect - connect a socket to a remote endpoint
1894 * @task: address of RPC task that manages state of connect request
1896 * TCP: If the remote end dropped the connection, delay reconnecting.
1898 * UDP socket connects are synchronous, but we use a work queue anyway
1899 * to guarantee that even unprivileged user processes can set up a
1900 * socket on a privileged port.
1902 * If a UDP socket connect fails, the delay behavior here prevents
1903 * retry floods (hard mounts).
1905 static void xs_connect(struct rpc_task
*task
)
1907 struct rpc_xprt
*xprt
= task
->tk_xprt
;
1908 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1910 if (transport
->sock
!= NULL
&& !RPC_IS_SOFTCONN(task
)) {
1911 dprintk("RPC: xs_connect delayed xprt %p for %lu "
1913 xprt
, xprt
->reestablish_timeout
/ HZ
);
1914 queue_delayed_work(rpciod_workqueue
,
1915 &transport
->connect_worker
,
1916 xprt
->reestablish_timeout
);
1917 xprt
->reestablish_timeout
<<= 1;
1918 if (xprt
->reestablish_timeout
< XS_TCP_INIT_REEST_TO
)
1919 xprt
->reestablish_timeout
= XS_TCP_INIT_REEST_TO
;
1920 if (xprt
->reestablish_timeout
> XS_TCP_MAX_REEST_TO
)
1921 xprt
->reestablish_timeout
= XS_TCP_MAX_REEST_TO
;
1923 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt
);
1924 queue_delayed_work(rpciod_workqueue
,
1925 &transport
->connect_worker
, 0);
1930 * xs_udp_print_stats - display UDP socket-specifc stats
1931 * @xprt: rpc_xprt struct containing statistics
1935 static void xs_udp_print_stats(struct rpc_xprt
*xprt
, struct seq_file
*seq
)
1937 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1939 seq_printf(seq
, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1941 xprt
->stat
.bind_count
,
1944 xprt
->stat
.bad_xids
,
1946 xprt
->stat
.bklog_u
);
1950 * xs_tcp_print_stats - display TCP socket-specifc stats
1951 * @xprt: rpc_xprt struct containing statistics
1955 static void xs_tcp_print_stats(struct rpc_xprt
*xprt
, struct seq_file
*seq
)
1957 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1960 if (xprt_connected(xprt
))
1961 idle_time
= (long)(jiffies
- xprt
->last_used
) / HZ
;
1963 seq_printf(seq
, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1965 xprt
->stat
.bind_count
,
1966 xprt
->stat
.connect_count
,
1967 xprt
->stat
.connect_time
,
1971 xprt
->stat
.bad_xids
,
1973 xprt
->stat
.bklog_u
);
1977 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
1978 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
1979 * to use the server side send routines.
1981 static void *bc_malloc(struct rpc_task
*task
, size_t size
)
1984 struct rpc_buffer
*buf
;
1986 BUG_ON(size
> PAGE_SIZE
- sizeof(struct rpc_buffer
));
1987 page
= alloc_page(GFP_KERNEL
);
1992 buf
= page_address(page
);
1993 buf
->len
= PAGE_SIZE
;
1999 * Free the space allocated in the bc_alloc routine
2001 static void bc_free(void *buffer
)
2003 struct rpc_buffer
*buf
;
2008 buf
= container_of(buffer
, struct rpc_buffer
, data
);
2009 free_page((unsigned long)buf
);
2013 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2014 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2016 static int bc_sendto(struct rpc_rqst
*req
)
2019 struct xdr_buf
*xbufp
= &req
->rq_snd_buf
;
2020 struct rpc_xprt
*xprt
= req
->rq_xprt
;
2021 struct sock_xprt
*transport
=
2022 container_of(xprt
, struct sock_xprt
, xprt
);
2023 struct socket
*sock
= transport
->sock
;
2024 unsigned long headoff
;
2025 unsigned long tailoff
;
2028 * Set up the rpc header and record marker stuff
2030 xs_encode_tcp_record_marker(xbufp
);
2032 tailoff
= (unsigned long)xbufp
->tail
[0].iov_base
& ~PAGE_MASK
;
2033 headoff
= (unsigned long)xbufp
->head
[0].iov_base
& ~PAGE_MASK
;
2034 len
= svc_send_common(sock
, xbufp
,
2035 virt_to_page(xbufp
->head
[0].iov_base
), headoff
,
2036 xbufp
->tail
[0].iov_base
, tailoff
);
2038 if (len
!= xbufp
->len
) {
2039 printk(KERN_NOTICE
"Error sending entire callback!\n");
2047 * The send routine. Borrows from svc_send
2049 static int bc_send_request(struct rpc_task
*task
)
2051 struct rpc_rqst
*req
= task
->tk_rqstp
;
2052 struct svc_xprt
*xprt
;
2053 struct svc_sock
*svsk
;
2056 dprintk("sending request with xid: %08x\n", ntohl(req
->rq_xid
));
2058 * Get the server socket associated with this callback xprt
2060 xprt
= req
->rq_xprt
->bc_xprt
;
2061 svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
2064 * Grab the mutex to serialize data as the connection is shared
2065 * with the fore channel
2067 if (!mutex_trylock(&xprt
->xpt_mutex
)) {
2068 rpc_sleep_on(&xprt
->xpt_bc_pending
, task
, NULL
);
2069 if (!mutex_trylock(&xprt
->xpt_mutex
))
2071 rpc_wake_up_queued_task(&xprt
->xpt_bc_pending
, task
);
2073 if (test_bit(XPT_DEAD
, &xprt
->xpt_flags
))
2076 len
= bc_sendto(req
);
2077 mutex_unlock(&xprt
->xpt_mutex
);
2086 * The close routine. Since this is client initiated, we do nothing
2089 static void bc_close(struct rpc_xprt
*xprt
)
2094 * The xprt destroy routine. Again, because this connection is client
2095 * initiated, we do nothing
2098 static void bc_destroy(struct rpc_xprt
*xprt
)
2102 static struct rpc_xprt_ops xs_udp_ops
= {
2103 .set_buffer_size
= xs_udp_set_buffer_size
,
2104 .reserve_xprt
= xprt_reserve_xprt_cong
,
2105 .release_xprt
= xprt_release_xprt_cong
,
2106 .rpcbind
= rpcb_getport_async
,
2107 .set_port
= xs_set_port
,
2108 .connect
= xs_connect
,
2109 .buf_alloc
= rpc_malloc
,
2110 .buf_free
= rpc_free
,
2111 .send_request
= xs_udp_send_request
,
2112 .set_retrans_timeout
= xprt_set_retrans_timeout_rtt
,
2113 .timer
= xs_udp_timer
,
2114 .release_request
= xprt_release_rqst_cong
,
2116 .destroy
= xs_destroy
,
2117 .print_stats
= xs_udp_print_stats
,
2120 static struct rpc_xprt_ops xs_tcp_ops
= {
2121 .reserve_xprt
= xprt_reserve_xprt
,
2122 .release_xprt
= xs_tcp_release_xprt
,
2123 .rpcbind
= rpcb_getport_async
,
2124 .set_port
= xs_set_port
,
2125 .connect
= xs_connect
,
2126 .buf_alloc
= rpc_malloc
,
2127 .buf_free
= rpc_free
,
2128 .send_request
= xs_tcp_send_request
,
2129 .set_retrans_timeout
= xprt_set_retrans_timeout_def
,
2130 .close
= xs_tcp_close
,
2131 .destroy
= xs_destroy
,
2132 .print_stats
= xs_tcp_print_stats
,
2136 * The rpc_xprt_ops for the server backchannel
2139 static struct rpc_xprt_ops bc_tcp_ops
= {
2140 .reserve_xprt
= xprt_reserve_xprt
,
2141 .release_xprt
= xprt_release_xprt
,
2142 .buf_alloc
= bc_malloc
,
2143 .buf_free
= bc_free
,
2144 .send_request
= bc_send_request
,
2145 .set_retrans_timeout
= xprt_set_retrans_timeout_def
,
2147 .destroy
= bc_destroy
,
2148 .print_stats
= xs_tcp_print_stats
,
2151 static int xs_init_anyaddr(const int family
, struct sockaddr
*sap
)
2153 static const struct sockaddr_in sin
= {
2154 .sin_family
= AF_INET
,
2155 .sin_addr
.s_addr
= htonl(INADDR_ANY
),
2157 static const struct sockaddr_in6 sin6
= {
2158 .sin6_family
= AF_INET6
,
2159 .sin6_addr
= IN6ADDR_ANY_INIT
,
2164 memcpy(sap
, &sin
, sizeof(sin
));
2167 memcpy(sap
, &sin6
, sizeof(sin6
));
2170 dprintk("RPC: %s: Bad address family\n", __func__
);
2171 return -EAFNOSUPPORT
;
2176 static struct rpc_xprt
*xs_setup_xprt(struct xprt_create
*args
,
2177 unsigned int slot_table_size
)
2179 struct rpc_xprt
*xprt
;
2180 struct sock_xprt
*new;
2182 if (args
->addrlen
> sizeof(xprt
->addr
)) {
2183 dprintk("RPC: xs_setup_xprt: address too large\n");
2184 return ERR_PTR(-EBADF
);
2187 xprt
= xprt_alloc(args
->net
, sizeof(*new), slot_table_size
);
2189 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2191 return ERR_PTR(-ENOMEM
);
2194 new = container_of(xprt
, struct sock_xprt
, xprt
);
2195 memcpy(&xprt
->addr
, args
->dstaddr
, args
->addrlen
);
2196 xprt
->addrlen
= args
->addrlen
;
2198 memcpy(&new->srcaddr
, args
->srcaddr
, args
->addrlen
);
2201 err
= xs_init_anyaddr(args
->dstaddr
->sa_family
,
2202 (struct sockaddr
*)&new->srcaddr
);
2204 return ERR_PTR(err
);
2210 static const struct rpc_timeout xs_udp_default_timeout
= {
2211 .to_initval
= 5 * HZ
,
2212 .to_maxval
= 30 * HZ
,
2213 .to_increment
= 5 * HZ
,
2218 * xs_setup_udp - Set up transport to use a UDP socket
2219 * @args: rpc transport creation arguments
2222 static struct rpc_xprt
*xs_setup_udp(struct xprt_create
*args
)
2224 struct sockaddr
*addr
= args
->dstaddr
;
2225 struct rpc_xprt
*xprt
;
2226 struct sock_xprt
*transport
;
2227 struct rpc_xprt
*ret
;
2229 xprt
= xs_setup_xprt(args
, xprt_udp_slot_table_entries
);
2232 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2234 xprt
->prot
= IPPROTO_UDP
;
2236 /* XXX: header size can vary due to auth type, IPv6, etc. */
2237 xprt
->max_payload
= (1U << 16) - (MAX_HEADER
<< 3);
2239 xprt
->bind_timeout
= XS_BIND_TO
;
2240 xprt
->reestablish_timeout
= XS_UDP_REEST_TO
;
2241 xprt
->idle_timeout
= XS_IDLE_DISC_TO
;
2243 xprt
->ops
= &xs_udp_ops
;
2245 xprt
->timeout
= &xs_udp_default_timeout
;
2247 switch (addr
->sa_family
) {
2249 if (((struct sockaddr_in
*)addr
)->sin_port
!= htons(0))
2250 xprt_set_bound(xprt
);
2252 INIT_DELAYED_WORK(&transport
->connect_worker
,
2253 xs_udp_setup_socket
);
2254 xs_format_peer_addresses(xprt
, "udp", RPCBIND_NETID_UDP
);
2257 if (((struct sockaddr_in6
*)addr
)->sin6_port
!= htons(0))
2258 xprt_set_bound(xprt
);
2260 INIT_DELAYED_WORK(&transport
->connect_worker
,
2261 xs_udp_setup_socket
);
2262 xs_format_peer_addresses(xprt
, "udp", RPCBIND_NETID_UDP6
);
2265 ret
= ERR_PTR(-EAFNOSUPPORT
);
2269 if (xprt_bound(xprt
))
2270 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2271 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
2272 xprt
->address_strings
[RPC_DISPLAY_PORT
],
2273 xprt
->address_strings
[RPC_DISPLAY_PROTO
]);
2275 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2276 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
2277 xprt
->address_strings
[RPC_DISPLAY_PROTO
]);
2279 if (try_module_get(THIS_MODULE
))
2281 ret
= ERR_PTR(-EINVAL
);
2287 static const struct rpc_timeout xs_tcp_default_timeout
= {
2288 .to_initval
= 60 * HZ
,
2289 .to_maxval
= 60 * HZ
,
2294 * xs_setup_tcp - Set up transport to use a TCP socket
2295 * @args: rpc transport creation arguments
2298 static struct rpc_xprt
*xs_setup_tcp(struct xprt_create
*args
)
2300 struct sockaddr
*addr
= args
->dstaddr
;
2301 struct rpc_xprt
*xprt
;
2302 struct sock_xprt
*transport
;
2303 struct rpc_xprt
*ret
;
2305 xprt
= xs_setup_xprt(args
, xprt_tcp_slot_table_entries
);
2308 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2310 xprt
->prot
= IPPROTO_TCP
;
2311 xprt
->tsh_size
= sizeof(rpc_fraghdr
) / sizeof(u32
);
2312 xprt
->max_payload
= RPC_MAX_FRAGMENT_SIZE
;
2314 xprt
->bind_timeout
= XS_BIND_TO
;
2315 xprt
->reestablish_timeout
= XS_TCP_INIT_REEST_TO
;
2316 xprt
->idle_timeout
= XS_IDLE_DISC_TO
;
2318 xprt
->ops
= &xs_tcp_ops
;
2319 xprt
->timeout
= &xs_tcp_default_timeout
;
2321 switch (addr
->sa_family
) {
2323 if (((struct sockaddr_in
*)addr
)->sin_port
!= htons(0))
2324 xprt_set_bound(xprt
);
2326 INIT_DELAYED_WORK(&transport
->connect_worker
,
2327 xs_tcp_setup_socket
);
2328 xs_format_peer_addresses(xprt
, "tcp", RPCBIND_NETID_TCP
);
2331 if (((struct sockaddr_in6
*)addr
)->sin6_port
!= htons(0))
2332 xprt_set_bound(xprt
);
2334 INIT_DELAYED_WORK(&transport
->connect_worker
,
2335 xs_tcp_setup_socket
);
2336 xs_format_peer_addresses(xprt
, "tcp", RPCBIND_NETID_TCP6
);
2339 ret
= ERR_PTR(-EAFNOSUPPORT
);
2343 if (xprt_bound(xprt
))
2344 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2345 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
2346 xprt
->address_strings
[RPC_DISPLAY_PORT
],
2347 xprt
->address_strings
[RPC_DISPLAY_PROTO
]);
2349 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2350 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
2351 xprt
->address_strings
[RPC_DISPLAY_PROTO
]);
2354 if (try_module_get(THIS_MODULE
))
2356 ret
= ERR_PTR(-EINVAL
);
2363 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2364 * @args: rpc transport creation arguments
2367 static struct rpc_xprt
*xs_setup_bc_tcp(struct xprt_create
*args
)
2369 struct sockaddr
*addr
= args
->dstaddr
;
2370 struct rpc_xprt
*xprt
;
2371 struct sock_xprt
*transport
;
2372 struct svc_sock
*bc_sock
;
2373 struct rpc_xprt
*ret
;
2375 if (args
->bc_xprt
->xpt_bc_xprt
) {
2377 * This server connection already has a backchannel
2378 * export; we can't create a new one, as we wouldn't be
2379 * able to match replies based on xid any more. So,
2380 * reuse the already-existing one:
2382 return args
->bc_xprt
->xpt_bc_xprt
;
2384 xprt
= xs_setup_xprt(args
, xprt_tcp_slot_table_entries
);
2387 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2389 xprt
->prot
= IPPROTO_TCP
;
2390 xprt
->tsh_size
= sizeof(rpc_fraghdr
) / sizeof(u32
);
2391 xprt
->max_payload
= RPC_MAX_FRAGMENT_SIZE
;
2392 xprt
->timeout
= &xs_tcp_default_timeout
;
2395 xprt_set_bound(xprt
);
2396 xprt
->bind_timeout
= 0;
2397 xprt
->reestablish_timeout
= 0;
2398 xprt
->idle_timeout
= 0;
2400 xprt
->ops
= &bc_tcp_ops
;
2402 switch (addr
->sa_family
) {
2404 xs_format_peer_addresses(xprt
, "tcp",
2408 xs_format_peer_addresses(xprt
, "tcp",
2409 RPCBIND_NETID_TCP6
);
2412 ret
= ERR_PTR(-EAFNOSUPPORT
);
2416 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2417 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
2418 xprt
->address_strings
[RPC_DISPLAY_PORT
],
2419 xprt
->address_strings
[RPC_DISPLAY_PROTO
]);
2422 * Once we've associated a backchannel xprt with a connection,
2423 * we want to keep it around as long as long as the connection
2424 * lasts, in case we need to start using it for a backchannel
2425 * again; this reference won't be dropped until bc_xprt is
2429 args
->bc_xprt
->xpt_bc_xprt
= xprt
;
2430 xprt
->bc_xprt
= args
->bc_xprt
;
2431 bc_sock
= container_of(args
->bc_xprt
, struct svc_sock
, sk_xprt
);
2432 transport
->sock
= bc_sock
->sk_sock
;
2433 transport
->inet
= bc_sock
->sk_sk
;
2436 * Since we don't want connections for the backchannel, we set
2437 * the xprt status to connected
2439 xprt_set_connected(xprt
);
2442 if (try_module_get(THIS_MODULE
))
2445 ret
= ERR_PTR(-EINVAL
);
2451 static struct xprt_class xs_udp_transport
= {
2452 .list
= LIST_HEAD_INIT(xs_udp_transport
.list
),
2454 .owner
= THIS_MODULE
,
2455 .ident
= XPRT_TRANSPORT_UDP
,
2456 .setup
= xs_setup_udp
,
2459 static struct xprt_class xs_tcp_transport
= {
2460 .list
= LIST_HEAD_INIT(xs_tcp_transport
.list
),
2462 .owner
= THIS_MODULE
,
2463 .ident
= XPRT_TRANSPORT_TCP
,
2464 .setup
= xs_setup_tcp
,
2467 static struct xprt_class xs_bc_tcp_transport
= {
2468 .list
= LIST_HEAD_INIT(xs_bc_tcp_transport
.list
),
2469 .name
= "tcp NFSv4.1 backchannel",
2470 .owner
= THIS_MODULE
,
2471 .ident
= XPRT_TRANSPORT_BC_TCP
,
2472 .setup
= xs_setup_bc_tcp
,
2476 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2479 int init_socket_xprt(void)
2482 if (!sunrpc_table_header
)
2483 sunrpc_table_header
= register_sysctl_table(sunrpc_table
);
2486 xprt_register_transport(&xs_udp_transport
);
2487 xprt_register_transport(&xs_tcp_transport
);
2488 xprt_register_transport(&xs_bc_tcp_transport
);
2494 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2497 void cleanup_socket_xprt(void)
2500 if (sunrpc_table_header
) {
2501 unregister_sysctl_table(sunrpc_table_header
);
2502 sunrpc_table_header
= NULL
;
2506 xprt_unregister_transport(&xs_udp_transport
);
2507 xprt_unregister_transport(&xs_tcp_transport
);
2508 xprt_unregister_transport(&xs_bc_tcp_transport
);
2511 static int param_set_uint_minmax(const char *val
,
2512 const struct kernel_param
*kp
,
2513 unsigned int min
, unsigned int max
)
2520 ret
= strict_strtoul(val
, 0, &num
);
2521 if (ret
== -EINVAL
|| num
< min
|| num
> max
)
2523 *((unsigned int *)kp
->arg
) = num
;
2527 static int param_set_portnr(const char *val
, const struct kernel_param
*kp
)
2529 return param_set_uint_minmax(val
, kp
,
2534 static struct kernel_param_ops param_ops_portnr
= {
2535 .set
= param_set_portnr
,
2536 .get
= param_get_uint
,
2539 #define param_check_portnr(name, p) \
2540 __param_check(name, p, unsigned int);
2542 module_param_named(min_resvport
, xprt_min_resvport
, portnr
, 0644);
2543 module_param_named(max_resvport
, xprt_max_resvport
, portnr
, 0644);
2545 static int param_set_slot_table_size(const char *val
,
2546 const struct kernel_param
*kp
)
2548 return param_set_uint_minmax(val
, kp
,
2550 RPC_MAX_SLOT_TABLE
);
2553 static struct kernel_param_ops param_ops_slot_table_size
= {
2554 .set
= param_set_slot_table_size
,
2555 .get
= param_get_uint
,
2558 #define param_check_slot_table_size(name, p) \
2559 __param_check(name, p, unsigned int);
2561 module_param_named(tcp_slot_table_entries
, xprt_tcp_slot_table_entries
,
2562 slot_table_size
, 0644);
2563 module_param_named(udp_slot_table_entries
, xprt_udp_slot_table_entries
,
2564 slot_table_size
, 0644);