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
;
78 static ctl_table xs_tunables_table
[] = {
80 .procname
= "udp_slot_table_entries",
81 .data
= &xprt_udp_slot_table_entries
,
82 .maxlen
= sizeof(unsigned int),
84 .proc_handler
= proc_dointvec_minmax
,
85 .extra1
= &min_slot_table_size
,
86 .extra2
= &max_slot_table_size
89 .procname
= "tcp_slot_table_entries",
90 .data
= &xprt_tcp_slot_table_entries
,
91 .maxlen
= sizeof(unsigned int),
93 .proc_handler
= proc_dointvec_minmax
,
94 .extra1
= &min_slot_table_size
,
95 .extra2
= &max_slot_table_size
98 .procname
= "min_resvport",
99 .data
= &xprt_min_resvport
,
100 .maxlen
= sizeof(unsigned int),
102 .proc_handler
= proc_dointvec_minmax
,
103 .extra1
= &xprt_min_resvport_limit
,
104 .extra2
= &xprt_max_resvport_limit
107 .procname
= "max_resvport",
108 .data
= &xprt_max_resvport
,
109 .maxlen
= sizeof(unsigned int),
111 .proc_handler
= proc_dointvec_minmax
,
112 .extra1
= &xprt_min_resvport_limit
,
113 .extra2
= &xprt_max_resvport_limit
116 .procname
= "tcp_fin_timeout",
117 .data
= &xs_tcp_fin_timeout
,
118 .maxlen
= sizeof(xs_tcp_fin_timeout
),
120 .proc_handler
= proc_dointvec_jiffies
,
125 static ctl_table sunrpc_table
[] = {
127 .procname
= "sunrpc",
129 .child
= xs_tunables_table
137 * Wait duration for a reply from the RPC portmapper.
139 #define XS_BIND_TO (60U * HZ)
142 * Delay if a UDP socket connect error occurs. This is most likely some
143 * kind of resource problem on the local host.
145 #define XS_UDP_REEST_TO (2U * HZ)
148 * The reestablish timeout allows clients to delay for a bit before attempting
149 * to reconnect to a server that just dropped our connection.
151 * We implement an exponential backoff when trying to reestablish a TCP
152 * transport connection with the server. Some servers like to drop a TCP
153 * connection when they are overworked, so we start with a short timeout and
154 * increase over time if the server is down or not responding.
156 #define XS_TCP_INIT_REEST_TO (3U * HZ)
157 #define XS_TCP_MAX_REEST_TO (5U * 60 * HZ)
160 * TCP idle timeout; client drops the transport socket if it is idle
161 * for this long. Note that we also timeout UDP sockets to prevent
162 * holding port numbers when there is no RPC traffic.
164 #define XS_IDLE_DISC_TO (5U * 60 * HZ)
167 # undef RPC_DEBUG_DATA
168 # define RPCDBG_FACILITY RPCDBG_TRANS
171 #ifdef RPC_DEBUG_DATA
172 static void xs_pktdump(char *msg
, u32
*packet
, unsigned int count
)
174 u8
*buf
= (u8
*) packet
;
177 dprintk("RPC: %s\n", msg
);
178 for (j
= 0; j
< count
&& j
< 128; j
+= 4) {
182 dprintk("0x%04x ", j
);
184 dprintk("%02x%02x%02x%02x ",
185 buf
[j
], buf
[j
+1], buf
[j
+2], buf
[j
+3]);
190 static inline void xs_pktdump(char *msg
, u32
*packet
, unsigned int count
)
197 struct rpc_xprt xprt
;
202 struct socket
* sock
;
206 * State of TCP reply receive
215 unsigned long tcp_copied
,
219 * Connection of transports
221 struct delayed_work connect_worker
;
222 struct sockaddr_storage srcaddr
;
223 unsigned short srcport
;
226 * UDP socket buffer size parameters
232 * Saved socket callback addresses
234 void (*old_data_ready
)(struct sock
*, int);
235 void (*old_state_change
)(struct sock
*);
236 void (*old_write_space
)(struct sock
*);
237 void (*old_error_report
)(struct sock
*);
241 * TCP receive state flags
243 #define TCP_RCV_LAST_FRAG (1UL << 0)
244 #define TCP_RCV_COPY_FRAGHDR (1UL << 1)
245 #define TCP_RCV_COPY_XID (1UL << 2)
246 #define TCP_RCV_COPY_DATA (1UL << 3)
247 #define TCP_RCV_READ_CALLDIR (1UL << 4)
248 #define TCP_RCV_COPY_CALLDIR (1UL << 5)
253 #define TCP_RPC_REPLY (1UL << 6)
255 static inline struct sockaddr
*xs_addr(struct rpc_xprt
*xprt
)
257 return (struct sockaddr
*) &xprt
->addr
;
260 static inline struct sockaddr_in
*xs_addr_in(struct rpc_xprt
*xprt
)
262 return (struct sockaddr_in
*) &xprt
->addr
;
265 static inline struct sockaddr_in6
*xs_addr_in6(struct rpc_xprt
*xprt
)
267 return (struct sockaddr_in6
*) &xprt
->addr
;
270 static void xs_format_common_peer_addresses(struct rpc_xprt
*xprt
)
272 struct sockaddr
*sap
= xs_addr(xprt
);
273 struct sockaddr_in6
*sin6
;
274 struct sockaddr_in
*sin
;
277 (void)rpc_ntop(sap
, buf
, sizeof(buf
));
278 xprt
->address_strings
[RPC_DISPLAY_ADDR
] = kstrdup(buf
, GFP_KERNEL
);
280 switch (sap
->sa_family
) {
282 sin
= xs_addr_in(xprt
);
283 snprintf(buf
, sizeof(buf
), "%08x", ntohl(sin
->sin_addr
.s_addr
));
286 sin6
= xs_addr_in6(xprt
);
287 snprintf(buf
, sizeof(buf
), "%pi6", &sin6
->sin6_addr
);
292 xprt
->address_strings
[RPC_DISPLAY_HEX_ADDR
] = kstrdup(buf
, GFP_KERNEL
);
295 static void xs_format_common_peer_ports(struct rpc_xprt
*xprt
)
297 struct sockaddr
*sap
= xs_addr(xprt
);
300 snprintf(buf
, sizeof(buf
), "%u", rpc_get_port(sap
));
301 xprt
->address_strings
[RPC_DISPLAY_PORT
] = kstrdup(buf
, GFP_KERNEL
);
303 snprintf(buf
, sizeof(buf
), "%4hx", rpc_get_port(sap
));
304 xprt
->address_strings
[RPC_DISPLAY_HEX_PORT
] = kstrdup(buf
, GFP_KERNEL
);
307 static void xs_format_peer_addresses(struct rpc_xprt
*xprt
,
308 const char *protocol
,
311 xprt
->address_strings
[RPC_DISPLAY_PROTO
] = protocol
;
312 xprt
->address_strings
[RPC_DISPLAY_NETID
] = netid
;
313 xs_format_common_peer_addresses(xprt
);
314 xs_format_common_peer_ports(xprt
);
317 static void xs_update_peer_port(struct rpc_xprt
*xprt
)
319 kfree(xprt
->address_strings
[RPC_DISPLAY_HEX_PORT
]);
320 kfree(xprt
->address_strings
[RPC_DISPLAY_PORT
]);
322 xs_format_common_peer_ports(xprt
);
325 static void xs_free_peer_addresses(struct rpc_xprt
*xprt
)
329 for (i
= 0; i
< RPC_DISPLAY_MAX
; i
++)
331 case RPC_DISPLAY_PROTO
:
332 case RPC_DISPLAY_NETID
:
335 kfree(xprt
->address_strings
[i
]);
339 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
341 static int xs_send_kvec(struct socket
*sock
, struct sockaddr
*addr
, int addrlen
, struct kvec
*vec
, unsigned int base
, int more
)
343 struct msghdr msg
= {
345 .msg_namelen
= addrlen
,
346 .msg_flags
= XS_SENDMSG_FLAGS
| (more
? MSG_MORE
: 0),
349 .iov_base
= vec
->iov_base
+ base
,
350 .iov_len
= vec
->iov_len
- base
,
353 if (iov
.iov_len
!= 0)
354 return kernel_sendmsg(sock
, &msg
, &iov
, 1, iov
.iov_len
);
355 return kernel_sendmsg(sock
, &msg
, NULL
, 0, 0);
358 static int xs_send_pagedata(struct socket
*sock
, struct xdr_buf
*xdr
, unsigned int base
, int more
)
361 unsigned int remainder
;
364 remainder
= xdr
->page_len
- base
;
365 base
+= xdr
->page_base
;
366 ppage
= xdr
->pages
+ (base
>> PAGE_SHIFT
);
369 unsigned int len
= min_t(unsigned int, PAGE_SIZE
- base
, remainder
);
370 int flags
= XS_SENDMSG_FLAGS
;
373 if (remainder
!= 0 || more
)
375 err
= sock
->ops
->sendpage(sock
, *ppage
, base
, len
, flags
);
376 if (remainder
== 0 || err
!= len
)
390 * xs_sendpages - write pages directly to a socket
391 * @sock: socket to send on
392 * @addr: UDP only -- address of destination
393 * @addrlen: UDP only -- length of destination address
394 * @xdr: buffer containing this request
395 * @base: starting position in the buffer
398 static int xs_sendpages(struct socket
*sock
, struct sockaddr
*addr
, int addrlen
, struct xdr_buf
*xdr
, unsigned int base
)
400 unsigned int remainder
= xdr
->len
- base
;
406 clear_bit(SOCK_ASYNC_NOSPACE
, &sock
->flags
);
412 if (base
< xdr
->head
[0].iov_len
|| addr
!= NULL
) {
413 unsigned int len
= xdr
->head
[0].iov_len
- base
;
415 err
= xs_send_kvec(sock
, addr
, addrlen
, &xdr
->head
[0], base
, remainder
!= 0);
416 if (remainder
== 0 || err
!= len
)
421 base
-= xdr
->head
[0].iov_len
;
423 if (base
< xdr
->page_len
) {
424 unsigned int len
= xdr
->page_len
- base
;
426 err
= xs_send_pagedata(sock
, xdr
, base
, remainder
!= 0);
427 if (remainder
== 0 || err
!= len
)
432 base
-= xdr
->page_len
;
434 if (base
>= xdr
->tail
[0].iov_len
)
436 err
= xs_send_kvec(sock
, NULL
, 0, &xdr
->tail
[0], base
, 0);
445 static void xs_nospace_callback(struct rpc_task
*task
)
447 struct sock_xprt
*transport
= container_of(task
->tk_rqstp
->rq_xprt
, struct sock_xprt
, xprt
);
449 transport
->inet
->sk_write_pending
--;
450 clear_bit(SOCK_ASYNC_NOSPACE
, &transport
->sock
->flags
);
454 * xs_nospace - place task on wait queue if transmit was incomplete
455 * @task: task to put to sleep
458 static int xs_nospace(struct rpc_task
*task
)
460 struct rpc_rqst
*req
= task
->tk_rqstp
;
461 struct rpc_xprt
*xprt
= req
->rq_xprt
;
462 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
465 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
466 task
->tk_pid
, req
->rq_slen
- req
->rq_bytes_sent
,
469 /* Protect against races with write_space */
470 spin_lock_bh(&xprt
->transport_lock
);
472 /* Don't race with disconnect */
473 if (xprt_connected(xprt
)) {
474 if (test_bit(SOCK_ASYNC_NOSPACE
, &transport
->sock
->flags
)) {
477 * Notify TCP that we're limited by the application
480 set_bit(SOCK_NOSPACE
, &transport
->sock
->flags
);
481 transport
->inet
->sk_write_pending
++;
482 /* ...and wait for more buffer space */
483 xprt_wait_for_buffer_space(task
, xs_nospace_callback
);
486 clear_bit(SOCK_ASYNC_NOSPACE
, &transport
->sock
->flags
);
490 spin_unlock_bh(&xprt
->transport_lock
);
495 * xs_udp_send_request - write an RPC request to a UDP socket
496 * @task: address of RPC task that manages the state of an RPC request
499 * 0: The request has been sent
500 * EAGAIN: The socket was blocked, please call again later to
501 * complete the request
502 * ENOTCONN: Caller needs to invoke connect logic then call again
503 * other: Some other error occured, the request was not sent
505 static int xs_udp_send_request(struct rpc_task
*task
)
507 struct rpc_rqst
*req
= task
->tk_rqstp
;
508 struct rpc_xprt
*xprt
= req
->rq_xprt
;
509 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
510 struct xdr_buf
*xdr
= &req
->rq_snd_buf
;
513 xs_pktdump("packet data:",
514 req
->rq_svec
->iov_base
,
515 req
->rq_svec
->iov_len
);
517 if (!xprt_bound(xprt
))
519 status
= xs_sendpages(transport
->sock
,
524 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
525 xdr
->len
- req
->rq_bytes_sent
, status
);
528 req
->rq_xmit_bytes_sent
+= status
;
529 if (status
>= req
->rq_slen
)
531 /* Still some bytes left; set up for a retry later. */
538 /* Should we call xs_close() here? */
541 status
= xs_nospace(task
);
544 dprintk("RPC: sendmsg returned unrecognized error %d\n",
549 /* When the server has died, an ICMP port unreachable message
550 * prompts ECONNREFUSED. */
551 clear_bit(SOCK_ASYNC_NOSPACE
, &transport
->sock
->flags
);
558 * xs_tcp_shutdown - gracefully shut down a TCP socket
561 * Initiates a graceful shutdown of the TCP socket by calling the
562 * equivalent of shutdown(SHUT_WR);
564 static void xs_tcp_shutdown(struct rpc_xprt
*xprt
)
566 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
567 struct socket
*sock
= transport
->sock
;
570 kernel_sock_shutdown(sock
, SHUT_WR
);
573 static inline void xs_encode_tcp_record_marker(struct xdr_buf
*buf
)
575 u32 reclen
= buf
->len
- sizeof(rpc_fraghdr
);
576 rpc_fraghdr
*base
= buf
->head
[0].iov_base
;
577 *base
= htonl(RPC_LAST_STREAM_FRAGMENT
| reclen
);
580 static int xs_tcp_send_request(struct rpc_task
*task
)
582 struct rpc_rqst
*req
= task
->tk_rqstp
;
583 struct rpc_xprt
*xprt
= req
->rq_xprt
;
584 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
585 struct xdr_buf
*xdr
= &req
->rq_snd_buf
;
588 xs_encode_tcp_record_marker(&req
->rq_snd_buf
);
590 xs_pktdump("packet data:",
591 req
->rq_svec
->iov_base
,
592 req
->rq_svec
->iov_len
);
594 /* Continue transmitting the packet/record. We must be careful
595 * to cope with writespace callbacks arriving _after_ we have
596 * called sendmsg(). */
598 status
= xs_sendpages(transport
->sock
,
599 NULL
, 0, xdr
, req
->rq_bytes_sent
);
601 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
602 xdr
->len
- req
->rq_bytes_sent
, status
);
604 if (unlikely(status
< 0))
607 /* If we've sent the entire packet, immediately
608 * reset the count of bytes sent. */
609 req
->rq_bytes_sent
+= status
;
610 req
->rq_xmit_bytes_sent
+= status
;
611 if (likely(req
->rq_bytes_sent
>= req
->rq_slen
)) {
612 req
->rq_bytes_sent
= 0;
625 /* Should we call xs_close() here? */
628 status
= xs_nospace(task
);
631 dprintk("RPC: sendmsg returned unrecognized error %d\n",
635 xs_tcp_shutdown(xprt
);
638 clear_bit(SOCK_ASYNC_NOSPACE
, &transport
->sock
->flags
);
645 * xs_tcp_release_xprt - clean up after a tcp transmission
649 * This cleans up if an error causes us to abort the transmission of a request.
650 * In this case, the socket may need to be reset in order to avoid confusing
653 static void xs_tcp_release_xprt(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
655 struct rpc_rqst
*req
;
657 if (task
!= xprt
->snd_task
)
661 req
= task
->tk_rqstp
;
662 if (req
->rq_bytes_sent
== 0)
664 if (req
->rq_bytes_sent
== req
->rq_snd_buf
.len
)
666 set_bit(XPRT_CLOSE_WAIT
, &task
->tk_xprt
->state
);
668 xprt_release_xprt(xprt
, task
);
671 static void xs_save_old_callbacks(struct sock_xprt
*transport
, struct sock
*sk
)
673 transport
->old_data_ready
= sk
->sk_data_ready
;
674 transport
->old_state_change
= sk
->sk_state_change
;
675 transport
->old_write_space
= sk
->sk_write_space
;
676 transport
->old_error_report
= sk
->sk_error_report
;
679 static void xs_restore_old_callbacks(struct sock_xprt
*transport
, struct sock
*sk
)
681 sk
->sk_data_ready
= transport
->old_data_ready
;
682 sk
->sk_state_change
= transport
->old_state_change
;
683 sk
->sk_write_space
= transport
->old_write_space
;
684 sk
->sk_error_report
= transport
->old_error_report
;
687 static void xs_reset_transport(struct sock_xprt
*transport
)
689 struct socket
*sock
= transport
->sock
;
690 struct sock
*sk
= transport
->inet
;
695 write_lock_bh(&sk
->sk_callback_lock
);
696 transport
->inet
= NULL
;
697 transport
->sock
= NULL
;
699 sk
->sk_user_data
= NULL
;
701 xs_restore_old_callbacks(transport
, sk
);
702 write_unlock_bh(&sk
->sk_callback_lock
);
710 * xs_close - close a socket
713 * This is used when all requests are complete; ie, no DRC state remains
714 * on the server we want to save.
716 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
717 * xs_reset_transport() zeroing the socket from underneath a writer.
719 static void xs_close(struct rpc_xprt
*xprt
)
721 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
723 dprintk("RPC: xs_close xprt %p\n", xprt
);
725 xs_reset_transport(transport
);
726 xprt
->reestablish_timeout
= 0;
728 smp_mb__before_clear_bit();
729 clear_bit(XPRT_CONNECTION_ABORT
, &xprt
->state
);
730 clear_bit(XPRT_CLOSE_WAIT
, &xprt
->state
);
731 clear_bit(XPRT_CLOSING
, &xprt
->state
);
732 smp_mb__after_clear_bit();
733 xprt_disconnect_done(xprt
);
736 static void xs_tcp_close(struct rpc_xprt
*xprt
)
738 if (test_and_clear_bit(XPRT_CONNECTION_CLOSE
, &xprt
->state
))
741 xs_tcp_shutdown(xprt
);
745 * xs_destroy - prepare to shutdown a transport
746 * @xprt: doomed transport
749 static void xs_destroy(struct rpc_xprt
*xprt
)
751 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
753 dprintk("RPC: xs_destroy xprt %p\n", xprt
);
755 cancel_rearming_delayed_work(&transport
->connect_worker
);
758 xs_free_peer_addresses(xprt
);
761 module_put(THIS_MODULE
);
764 static inline struct rpc_xprt
*xprt_from_sock(struct sock
*sk
)
766 return (struct rpc_xprt
*) sk
->sk_user_data
;
770 * xs_udp_data_ready - "data ready" callback for UDP sockets
771 * @sk: socket with data to read
772 * @len: how much data to read
775 static void xs_udp_data_ready(struct sock
*sk
, int len
)
777 struct rpc_task
*task
;
778 struct rpc_xprt
*xprt
;
779 struct rpc_rqst
*rovr
;
781 int err
, repsize
, copied
;
785 read_lock_bh(&sk
->sk_callback_lock
);
786 dprintk("RPC: xs_udp_data_ready...\n");
787 if (!(xprt
= xprt_from_sock(sk
)))
790 if ((skb
= skb_recv_datagram(sk
, 0, 1, &err
)) == NULL
)
796 repsize
= skb
->len
- sizeof(struct udphdr
);
798 dprintk("RPC: impossible RPC reply size %d!\n", repsize
);
802 /* Copy the XID from the skb... */
803 xp
= skb_header_pointer(skb
, sizeof(struct udphdr
),
804 sizeof(_xid
), &_xid
);
808 /* Look up and lock the request corresponding to the given XID */
809 spin_lock(&xprt
->transport_lock
);
810 rovr
= xprt_lookup_rqst(xprt
, *xp
);
813 task
= rovr
->rq_task
;
815 if ((copied
= rovr
->rq_private_buf
.buflen
) > repsize
)
818 /* Suck it into the iovec, verify checksum if not done by hw. */
819 if (csum_partial_copy_to_xdr(&rovr
->rq_private_buf
, skb
)) {
820 UDPX_INC_STATS_BH(sk
, UDP_MIB_INERRORS
);
824 UDPX_INC_STATS_BH(sk
, UDP_MIB_INDATAGRAMS
);
826 /* Something worked... */
827 dst_confirm(skb_dst(skb
));
829 xprt_adjust_cwnd(task
, copied
);
830 xprt_complete_rqst(task
, copied
);
833 spin_unlock(&xprt
->transport_lock
);
835 skb_free_datagram(sk
, skb
);
837 read_unlock_bh(&sk
->sk_callback_lock
);
840 static inline void xs_tcp_read_fraghdr(struct rpc_xprt
*xprt
, struct xdr_skb_reader
*desc
)
842 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
846 p
= ((char *) &transport
->tcp_fraghdr
) + transport
->tcp_offset
;
847 len
= sizeof(transport
->tcp_fraghdr
) - transport
->tcp_offset
;
848 used
= xdr_skb_read_bits(desc
, p
, len
);
849 transport
->tcp_offset
+= used
;
853 transport
->tcp_reclen
= ntohl(transport
->tcp_fraghdr
);
854 if (transport
->tcp_reclen
& RPC_LAST_STREAM_FRAGMENT
)
855 transport
->tcp_flags
|= TCP_RCV_LAST_FRAG
;
857 transport
->tcp_flags
&= ~TCP_RCV_LAST_FRAG
;
858 transport
->tcp_reclen
&= RPC_FRAGMENT_SIZE_MASK
;
860 transport
->tcp_flags
&= ~TCP_RCV_COPY_FRAGHDR
;
861 transport
->tcp_offset
= 0;
863 /* Sanity check of the record length */
864 if (unlikely(transport
->tcp_reclen
< 8)) {
865 dprintk("RPC: invalid TCP record fragment length\n");
866 xprt_force_disconnect(xprt
);
869 dprintk("RPC: reading TCP record fragment of length %d\n",
870 transport
->tcp_reclen
);
873 static void xs_tcp_check_fraghdr(struct sock_xprt
*transport
)
875 if (transport
->tcp_offset
== transport
->tcp_reclen
) {
876 transport
->tcp_flags
|= TCP_RCV_COPY_FRAGHDR
;
877 transport
->tcp_offset
= 0;
878 if (transport
->tcp_flags
& TCP_RCV_LAST_FRAG
) {
879 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
880 transport
->tcp_flags
|= TCP_RCV_COPY_XID
;
881 transport
->tcp_copied
= 0;
886 static inline void xs_tcp_read_xid(struct sock_xprt
*transport
, struct xdr_skb_reader
*desc
)
891 len
= sizeof(transport
->tcp_xid
) - transport
->tcp_offset
;
892 dprintk("RPC: reading XID (%Zu bytes)\n", len
);
893 p
= ((char *) &transport
->tcp_xid
) + transport
->tcp_offset
;
894 used
= xdr_skb_read_bits(desc
, p
, len
);
895 transport
->tcp_offset
+= used
;
898 transport
->tcp_flags
&= ~TCP_RCV_COPY_XID
;
899 transport
->tcp_flags
|= TCP_RCV_READ_CALLDIR
;
900 transport
->tcp_copied
= 4;
901 dprintk("RPC: reading %s XID %08x\n",
902 (transport
->tcp_flags
& TCP_RPC_REPLY
) ? "reply for"
904 ntohl(transport
->tcp_xid
));
905 xs_tcp_check_fraghdr(transport
);
908 static inline void xs_tcp_read_calldir(struct sock_xprt
*transport
,
909 struct xdr_skb_reader
*desc
)
916 * We want transport->tcp_offset to be 8 at the end of this routine
917 * (4 bytes for the xid and 4 bytes for the call/reply flag).
918 * When this function is called for the first time,
919 * transport->tcp_offset is 4 (after having already read the xid).
921 offset
= transport
->tcp_offset
- sizeof(transport
->tcp_xid
);
922 len
= sizeof(transport
->tcp_calldir
) - offset
;
923 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len
);
924 p
= ((char *) &transport
->tcp_calldir
) + offset
;
925 used
= xdr_skb_read_bits(desc
, p
, len
);
926 transport
->tcp_offset
+= used
;
929 transport
->tcp_flags
&= ~TCP_RCV_READ_CALLDIR
;
931 * We don't yet have the XDR buffer, so we will write the calldir
932 * out after we get the buffer from the 'struct rpc_rqst'
934 switch (ntohl(transport
->tcp_calldir
)) {
936 transport
->tcp_flags
|= TCP_RCV_COPY_CALLDIR
;
937 transport
->tcp_flags
|= TCP_RCV_COPY_DATA
;
938 transport
->tcp_flags
|= TCP_RPC_REPLY
;
941 transport
->tcp_flags
|= TCP_RCV_COPY_CALLDIR
;
942 transport
->tcp_flags
|= TCP_RCV_COPY_DATA
;
943 transport
->tcp_flags
&= ~TCP_RPC_REPLY
;
946 dprintk("RPC: invalid request message type\n");
947 xprt_force_disconnect(&transport
->xprt
);
949 xs_tcp_check_fraghdr(transport
);
952 static inline void xs_tcp_read_common(struct rpc_xprt
*xprt
,
953 struct xdr_skb_reader
*desc
,
954 struct rpc_rqst
*req
)
956 struct sock_xprt
*transport
=
957 container_of(xprt
, struct sock_xprt
, xprt
);
958 struct xdr_buf
*rcvbuf
;
962 rcvbuf
= &req
->rq_private_buf
;
964 if (transport
->tcp_flags
& TCP_RCV_COPY_CALLDIR
) {
966 * Save the RPC direction in the XDR buffer
968 memcpy(rcvbuf
->head
[0].iov_base
+ transport
->tcp_copied
,
969 &transport
->tcp_calldir
,
970 sizeof(transport
->tcp_calldir
));
971 transport
->tcp_copied
+= sizeof(transport
->tcp_calldir
);
972 transport
->tcp_flags
&= ~TCP_RCV_COPY_CALLDIR
;
976 if (len
> transport
->tcp_reclen
- transport
->tcp_offset
) {
977 struct xdr_skb_reader my_desc
;
979 len
= transport
->tcp_reclen
- transport
->tcp_offset
;
980 memcpy(&my_desc
, desc
, sizeof(my_desc
));
982 r
= xdr_partial_copy_from_skb(rcvbuf
, transport
->tcp_copied
,
983 &my_desc
, xdr_skb_read_bits
);
987 r
= xdr_partial_copy_from_skb(rcvbuf
, transport
->tcp_copied
,
988 desc
, xdr_skb_read_bits
);
991 transport
->tcp_copied
+= r
;
992 transport
->tcp_offset
+= r
;
995 /* Error when copying to the receive buffer,
996 * usually because we weren't able to allocate
997 * additional buffer pages. All we can do now
998 * is turn off TCP_RCV_COPY_DATA, so the request
999 * will not receive any additional updates,
1001 * Any remaining data from this record will
1004 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
1005 dprintk("RPC: XID %08x truncated request\n",
1006 ntohl(transport
->tcp_xid
));
1007 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1008 "tcp_offset = %u, tcp_reclen = %u\n",
1009 xprt
, transport
->tcp_copied
,
1010 transport
->tcp_offset
, transport
->tcp_reclen
);
1014 dprintk("RPC: XID %08x read %Zd bytes\n",
1015 ntohl(transport
->tcp_xid
), r
);
1016 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1017 "tcp_reclen = %u\n", xprt
, transport
->tcp_copied
,
1018 transport
->tcp_offset
, transport
->tcp_reclen
);
1020 if (transport
->tcp_copied
== req
->rq_private_buf
.buflen
)
1021 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
1022 else if (transport
->tcp_offset
== transport
->tcp_reclen
) {
1023 if (transport
->tcp_flags
& TCP_RCV_LAST_FRAG
)
1024 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
1029 * Finds the request corresponding to the RPC xid and invokes the common
1030 * tcp read code to read the data.
1032 static inline int xs_tcp_read_reply(struct rpc_xprt
*xprt
,
1033 struct xdr_skb_reader
*desc
)
1035 struct sock_xprt
*transport
=
1036 container_of(xprt
, struct sock_xprt
, xprt
);
1037 struct rpc_rqst
*req
;
1039 dprintk("RPC: read reply XID %08x\n", ntohl(transport
->tcp_xid
));
1041 /* Find and lock the request corresponding to this xid */
1042 spin_lock(&xprt
->transport_lock
);
1043 req
= xprt_lookup_rqst(xprt
, transport
->tcp_xid
);
1045 dprintk("RPC: XID %08x request not found!\n",
1046 ntohl(transport
->tcp_xid
));
1047 spin_unlock(&xprt
->transport_lock
);
1051 xs_tcp_read_common(xprt
, desc
, req
);
1053 if (!(transport
->tcp_flags
& TCP_RCV_COPY_DATA
))
1054 xprt_complete_rqst(req
->rq_task
, transport
->tcp_copied
);
1056 spin_unlock(&xprt
->transport_lock
);
1060 #if defined(CONFIG_NFS_V4_1)
1062 * Obtains an rpc_rqst previously allocated and invokes the common
1063 * tcp read code to read the data. The result is placed in the callback
1065 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1066 * connection and return -1.
1068 static inline int xs_tcp_read_callback(struct rpc_xprt
*xprt
,
1069 struct xdr_skb_reader
*desc
)
1071 struct sock_xprt
*transport
=
1072 container_of(xprt
, struct sock_xprt
, xprt
);
1073 struct rpc_rqst
*req
;
1075 req
= xprt_alloc_bc_request(xprt
);
1077 printk(KERN_WARNING
"Callback slot table overflowed\n");
1078 xprt_force_disconnect(xprt
);
1082 req
->rq_xid
= transport
->tcp_xid
;
1083 dprintk("RPC: read callback XID %08x\n", ntohl(req
->rq_xid
));
1084 xs_tcp_read_common(xprt
, desc
, req
);
1086 if (!(transport
->tcp_flags
& TCP_RCV_COPY_DATA
)) {
1087 struct svc_serv
*bc_serv
= xprt
->bc_serv
;
1090 * Add callback request to callback list. The callback
1091 * service sleeps on the sv_cb_waitq waiting for new
1092 * requests. Wake it up after adding enqueing the
1095 dprintk("RPC: add callback request to list\n");
1096 spin_lock(&bc_serv
->sv_cb_lock
);
1097 list_add(&req
->rq_bc_list
, &bc_serv
->sv_cb_list
);
1098 spin_unlock(&bc_serv
->sv_cb_lock
);
1099 wake_up(&bc_serv
->sv_cb_waitq
);
1102 req
->rq_private_buf
.len
= transport
->tcp_copied
;
1107 static inline int _xs_tcp_read_data(struct rpc_xprt
*xprt
,
1108 struct xdr_skb_reader
*desc
)
1110 struct sock_xprt
*transport
=
1111 container_of(xprt
, struct sock_xprt
, xprt
);
1113 return (transport
->tcp_flags
& TCP_RPC_REPLY
) ?
1114 xs_tcp_read_reply(xprt
, desc
) :
1115 xs_tcp_read_callback(xprt
, desc
);
1118 static inline int _xs_tcp_read_data(struct rpc_xprt
*xprt
,
1119 struct xdr_skb_reader
*desc
)
1121 return xs_tcp_read_reply(xprt
, desc
);
1123 #endif /* CONFIG_NFS_V4_1 */
1126 * Read data off the transport. This can be either an RPC_CALL or an
1127 * RPC_REPLY. Relay the processing to helper functions.
1129 static void xs_tcp_read_data(struct rpc_xprt
*xprt
,
1130 struct xdr_skb_reader
*desc
)
1132 struct sock_xprt
*transport
=
1133 container_of(xprt
, struct sock_xprt
, xprt
);
1135 if (_xs_tcp_read_data(xprt
, desc
) == 0)
1136 xs_tcp_check_fraghdr(transport
);
1139 * The transport_lock protects the request handling.
1140 * There's no need to hold it to update the tcp_flags.
1142 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
1146 static inline void xs_tcp_read_discard(struct sock_xprt
*transport
, struct xdr_skb_reader
*desc
)
1150 len
= transport
->tcp_reclen
- transport
->tcp_offset
;
1151 if (len
> desc
->count
)
1154 desc
->offset
+= len
;
1155 transport
->tcp_offset
+= len
;
1156 dprintk("RPC: discarded %Zu bytes\n", len
);
1157 xs_tcp_check_fraghdr(transport
);
1160 static int xs_tcp_data_recv(read_descriptor_t
*rd_desc
, struct sk_buff
*skb
, unsigned int offset
, size_t len
)
1162 struct rpc_xprt
*xprt
= rd_desc
->arg
.data
;
1163 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1164 struct xdr_skb_reader desc
= {
1170 dprintk("RPC: xs_tcp_data_recv started\n");
1172 /* Read in a new fragment marker if necessary */
1173 /* Can we ever really expect to get completely empty fragments? */
1174 if (transport
->tcp_flags
& TCP_RCV_COPY_FRAGHDR
) {
1175 xs_tcp_read_fraghdr(xprt
, &desc
);
1178 /* Read in the xid if necessary */
1179 if (transport
->tcp_flags
& TCP_RCV_COPY_XID
) {
1180 xs_tcp_read_xid(transport
, &desc
);
1183 /* Read in the call/reply flag */
1184 if (transport
->tcp_flags
& TCP_RCV_READ_CALLDIR
) {
1185 xs_tcp_read_calldir(transport
, &desc
);
1188 /* Read in the request data */
1189 if (transport
->tcp_flags
& TCP_RCV_COPY_DATA
) {
1190 xs_tcp_read_data(xprt
, &desc
);
1193 /* Skip over any trailing bytes on short reads */
1194 xs_tcp_read_discard(transport
, &desc
);
1195 } while (desc
.count
);
1196 dprintk("RPC: xs_tcp_data_recv done\n");
1197 return len
- desc
.count
;
1201 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1202 * @sk: socket with data to read
1203 * @bytes: how much data to read
1206 static void xs_tcp_data_ready(struct sock
*sk
, int bytes
)
1208 struct rpc_xprt
*xprt
;
1209 read_descriptor_t rd_desc
;
1212 dprintk("RPC: xs_tcp_data_ready...\n");
1214 read_lock_bh(&sk
->sk_callback_lock
);
1215 if (!(xprt
= xprt_from_sock(sk
)))
1220 /* Any data means we had a useful conversation, so
1221 * the we don't need to delay the next reconnect
1223 if (xprt
->reestablish_timeout
)
1224 xprt
->reestablish_timeout
= 0;
1226 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1227 rd_desc
.arg
.data
= xprt
;
1229 rd_desc
.count
= 65536;
1230 read
= tcp_read_sock(sk
, &rd_desc
, xs_tcp_data_recv
);
1233 read_unlock_bh(&sk
->sk_callback_lock
);
1237 * Do the equivalent of linger/linger2 handling for dealing with
1238 * broken servers that don't close the socket in a timely
1241 static void xs_tcp_schedule_linger_timeout(struct rpc_xprt
*xprt
,
1242 unsigned long timeout
)
1244 struct sock_xprt
*transport
;
1246 if (xprt_test_and_set_connecting(xprt
))
1248 set_bit(XPRT_CONNECTION_ABORT
, &xprt
->state
);
1249 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1250 queue_delayed_work(rpciod_workqueue
, &transport
->connect_worker
,
1254 static void xs_tcp_cancel_linger_timeout(struct rpc_xprt
*xprt
)
1256 struct sock_xprt
*transport
;
1258 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1260 if (!test_bit(XPRT_CONNECTION_ABORT
, &xprt
->state
) ||
1261 !cancel_delayed_work(&transport
->connect_worker
))
1263 clear_bit(XPRT_CONNECTION_ABORT
, &xprt
->state
);
1264 xprt_clear_connecting(xprt
);
1267 static void xs_sock_mark_closed(struct rpc_xprt
*xprt
)
1269 smp_mb__before_clear_bit();
1270 clear_bit(XPRT_CLOSE_WAIT
, &xprt
->state
);
1271 clear_bit(XPRT_CLOSING
, &xprt
->state
);
1272 smp_mb__after_clear_bit();
1273 /* Mark transport as closed and wake up all pending tasks */
1274 xprt_disconnect_done(xprt
);
1278 * xs_tcp_state_change - callback to handle TCP socket state changes
1279 * @sk: socket whose state has changed
1282 static void xs_tcp_state_change(struct sock
*sk
)
1284 struct rpc_xprt
*xprt
;
1286 read_lock_bh(&sk
->sk_callback_lock
);
1287 if (!(xprt
= xprt_from_sock(sk
)))
1289 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt
);
1290 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1291 sk
->sk_state
, xprt_connected(xprt
),
1292 sock_flag(sk
, SOCK_DEAD
),
1293 sock_flag(sk
, SOCK_ZAPPED
),
1296 switch (sk
->sk_state
) {
1297 case TCP_ESTABLISHED
:
1298 spin_lock(&xprt
->transport_lock
);
1299 if (!xprt_test_and_set_connected(xprt
)) {
1300 struct sock_xprt
*transport
= container_of(xprt
,
1301 struct sock_xprt
, xprt
);
1303 /* Reset TCP record info */
1304 transport
->tcp_offset
= 0;
1305 transport
->tcp_reclen
= 0;
1306 transport
->tcp_copied
= 0;
1307 transport
->tcp_flags
=
1308 TCP_RCV_COPY_FRAGHDR
| TCP_RCV_COPY_XID
;
1310 xprt_wake_pending_tasks(xprt
, -EAGAIN
);
1312 spin_unlock(&xprt
->transport_lock
);
1315 /* The client initiated a shutdown of the socket */
1316 xprt
->connect_cookie
++;
1317 xprt
->reestablish_timeout
= 0;
1318 set_bit(XPRT_CLOSING
, &xprt
->state
);
1319 smp_mb__before_clear_bit();
1320 clear_bit(XPRT_CONNECTED
, &xprt
->state
);
1321 clear_bit(XPRT_CLOSE_WAIT
, &xprt
->state
);
1322 smp_mb__after_clear_bit();
1323 xs_tcp_schedule_linger_timeout(xprt
, xs_tcp_fin_timeout
);
1325 case TCP_CLOSE_WAIT
:
1326 /* The server initiated a shutdown of the socket */
1327 xprt_force_disconnect(xprt
);
1329 xprt
->connect_cookie
++;
1332 * If the server closed down the connection, make sure that
1333 * we back off before reconnecting
1335 if (xprt
->reestablish_timeout
< XS_TCP_INIT_REEST_TO
)
1336 xprt
->reestablish_timeout
= XS_TCP_INIT_REEST_TO
;
1339 set_bit(XPRT_CLOSING
, &xprt
->state
);
1340 xs_tcp_schedule_linger_timeout(xprt
, xs_tcp_fin_timeout
);
1341 smp_mb__before_clear_bit();
1342 clear_bit(XPRT_CONNECTED
, &xprt
->state
);
1343 smp_mb__after_clear_bit();
1346 xs_tcp_cancel_linger_timeout(xprt
);
1347 xs_sock_mark_closed(xprt
);
1350 read_unlock_bh(&sk
->sk_callback_lock
);
1354 * xs_error_report - callback mainly for catching socket errors
1357 static void xs_error_report(struct sock
*sk
)
1359 struct rpc_xprt
*xprt
;
1361 read_lock_bh(&sk
->sk_callback_lock
);
1362 if (!(xprt
= xprt_from_sock(sk
)))
1364 dprintk("RPC: %s client %p...\n"
1366 __func__
, xprt
, sk
->sk_err
);
1367 xprt_wake_pending_tasks(xprt
, -EAGAIN
);
1369 read_unlock_bh(&sk
->sk_callback_lock
);
1372 static void xs_write_space(struct sock
*sk
)
1374 struct socket
*sock
;
1375 struct rpc_xprt
*xprt
;
1377 if (unlikely(!(sock
= sk
->sk_socket
)))
1379 clear_bit(SOCK_NOSPACE
, &sock
->flags
);
1381 if (unlikely(!(xprt
= xprt_from_sock(sk
))))
1383 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE
, &sock
->flags
) == 0)
1386 xprt_write_space(xprt
);
1390 * xs_udp_write_space - callback invoked when socket buffer space
1392 * @sk: socket whose state has changed
1394 * Called when more output buffer space is available for this socket.
1395 * We try not to wake our writers until they can make "significant"
1396 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1397 * with a bunch of small requests.
1399 static void xs_udp_write_space(struct sock
*sk
)
1401 read_lock_bh(&sk
->sk_callback_lock
);
1403 /* from net/core/sock.c:sock_def_write_space */
1404 if (sock_writeable(sk
))
1407 read_unlock_bh(&sk
->sk_callback_lock
);
1411 * xs_tcp_write_space - callback invoked when socket buffer space
1413 * @sk: socket whose state has changed
1415 * Called when more output buffer space is available for this socket.
1416 * We try not to wake our writers until they can make "significant"
1417 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1418 * with a bunch of small requests.
1420 static void xs_tcp_write_space(struct sock
*sk
)
1422 read_lock_bh(&sk
->sk_callback_lock
);
1424 /* from net/core/stream.c:sk_stream_write_space */
1425 if (sk_stream_wspace(sk
) >= sk_stream_min_wspace(sk
))
1428 read_unlock_bh(&sk
->sk_callback_lock
);
1431 static void xs_udp_do_set_buffer_size(struct rpc_xprt
*xprt
)
1433 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1434 struct sock
*sk
= transport
->inet
;
1436 if (transport
->rcvsize
) {
1437 sk
->sk_userlocks
|= SOCK_RCVBUF_LOCK
;
1438 sk
->sk_rcvbuf
= transport
->rcvsize
* xprt
->max_reqs
* 2;
1440 if (transport
->sndsize
) {
1441 sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
;
1442 sk
->sk_sndbuf
= transport
->sndsize
* xprt
->max_reqs
* 2;
1443 sk
->sk_write_space(sk
);
1448 * xs_udp_set_buffer_size - set send and receive limits
1449 * @xprt: generic transport
1450 * @sndsize: requested size of send buffer, in bytes
1451 * @rcvsize: requested size of receive buffer, in bytes
1453 * Set socket send and receive buffer size limits.
1455 static void xs_udp_set_buffer_size(struct rpc_xprt
*xprt
, size_t sndsize
, size_t rcvsize
)
1457 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1459 transport
->sndsize
= 0;
1461 transport
->sndsize
= sndsize
+ 1024;
1462 transport
->rcvsize
= 0;
1464 transport
->rcvsize
= rcvsize
+ 1024;
1466 xs_udp_do_set_buffer_size(xprt
);
1470 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1471 * @task: task that timed out
1473 * Adjust the congestion window after a retransmit timeout has occurred.
1475 static void xs_udp_timer(struct rpc_task
*task
)
1477 xprt_adjust_cwnd(task
, -ETIMEDOUT
);
1480 static unsigned short xs_get_random_port(void)
1482 unsigned short range
= xprt_max_resvport
- xprt_min_resvport
;
1483 unsigned short rand
= (unsigned short) net_random() % range
;
1484 return rand
+ xprt_min_resvport
;
1488 * xs_set_port - reset the port number in the remote endpoint address
1489 * @xprt: generic transport
1490 * @port: new port number
1493 static void xs_set_port(struct rpc_xprt
*xprt
, unsigned short port
)
1495 dprintk("RPC: setting port for xprt %p to %u\n", xprt
, port
);
1497 rpc_set_port(xs_addr(xprt
), port
);
1498 xs_update_peer_port(xprt
);
1501 static unsigned short xs_get_srcport(struct sock_xprt
*transport
, struct socket
*sock
)
1503 unsigned short port
= transport
->srcport
;
1505 if (port
== 0 && transport
->xprt
.resvport
)
1506 port
= xs_get_random_port();
1510 static unsigned short xs_next_srcport(struct sock_xprt
*transport
, struct socket
*sock
, unsigned short port
)
1512 if (transport
->srcport
!= 0)
1513 transport
->srcport
= 0;
1514 if (!transport
->xprt
.resvport
)
1516 if (port
<= xprt_min_resvport
|| port
> xprt_max_resvport
)
1517 return xprt_max_resvport
;
1521 static int xs_bind4(struct sock_xprt
*transport
, struct socket
*sock
)
1523 struct sockaddr_in myaddr
= {
1524 .sin_family
= AF_INET
,
1526 struct sockaddr_in
*sa
;
1528 unsigned short port
= xs_get_srcport(transport
, sock
);
1529 unsigned short last
;
1531 sa
= (struct sockaddr_in
*)&transport
->srcaddr
;
1532 myaddr
.sin_addr
= sa
->sin_addr
;
1534 myaddr
.sin_port
= htons(port
);
1535 err
= kernel_bind(sock
, (struct sockaddr
*) &myaddr
,
1540 transport
->srcport
= port
;
1544 port
= xs_next_srcport(transport
, sock
, port
);
1547 } while (err
== -EADDRINUSE
&& nloop
!= 2);
1548 dprintk("RPC: %s %pI4:%u: %s (%d)\n",
1549 __func__
, &myaddr
.sin_addr
,
1550 port
, err
? "failed" : "ok", err
);
1554 static int xs_bind6(struct sock_xprt
*transport
, struct socket
*sock
)
1556 struct sockaddr_in6 myaddr
= {
1557 .sin6_family
= AF_INET6
,
1559 struct sockaddr_in6
*sa
;
1561 unsigned short port
= xs_get_srcport(transport
, sock
);
1562 unsigned short last
;
1564 sa
= (struct sockaddr_in6
*)&transport
->srcaddr
;
1565 myaddr
.sin6_addr
= sa
->sin6_addr
;
1567 myaddr
.sin6_port
= htons(port
);
1568 err
= kernel_bind(sock
, (struct sockaddr
*) &myaddr
,
1573 transport
->srcport
= port
;
1577 port
= xs_next_srcport(transport
, sock
, port
);
1580 } while (err
== -EADDRINUSE
&& nloop
!= 2);
1581 dprintk("RPC: xs_bind6 %pI6:%u: %s (%d)\n",
1582 &myaddr
.sin6_addr
, port
, err
? "failed" : "ok", err
);
1586 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1587 static struct lock_class_key xs_key
[2];
1588 static struct lock_class_key xs_slock_key
[2];
1590 static inline void xs_reclassify_socket4(struct socket
*sock
)
1592 struct sock
*sk
= sock
->sk
;
1594 BUG_ON(sock_owned_by_user(sk
));
1595 sock_lock_init_class_and_name(sk
, "slock-AF_INET-RPC",
1596 &xs_slock_key
[0], "sk_lock-AF_INET-RPC", &xs_key
[0]);
1599 static inline void xs_reclassify_socket6(struct socket
*sock
)
1601 struct sock
*sk
= sock
->sk
;
1603 BUG_ON(sock_owned_by_user(sk
));
1604 sock_lock_init_class_and_name(sk
, "slock-AF_INET6-RPC",
1605 &xs_slock_key
[1], "sk_lock-AF_INET6-RPC", &xs_key
[1]);
1608 static inline void xs_reclassify_socket4(struct socket
*sock
)
1612 static inline void xs_reclassify_socket6(struct socket
*sock
)
1617 static void xs_udp_finish_connecting(struct rpc_xprt
*xprt
, struct socket
*sock
)
1619 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1621 if (!transport
->inet
) {
1622 struct sock
*sk
= sock
->sk
;
1624 write_lock_bh(&sk
->sk_callback_lock
);
1626 xs_save_old_callbacks(transport
, sk
);
1628 sk
->sk_user_data
= xprt
;
1629 sk
->sk_data_ready
= xs_udp_data_ready
;
1630 sk
->sk_write_space
= xs_udp_write_space
;
1631 sk
->sk_error_report
= xs_error_report
;
1632 sk
->sk_no_check
= UDP_CSUM_NORCV
;
1633 sk
->sk_allocation
= GFP_ATOMIC
;
1635 xprt_set_connected(xprt
);
1637 /* Reset to new socket */
1638 transport
->sock
= sock
;
1639 transport
->inet
= sk
;
1641 write_unlock_bh(&sk
->sk_callback_lock
);
1643 xs_udp_do_set_buffer_size(xprt
);
1647 * xs_udp_connect_worker4 - set up a UDP socket
1648 * @work: RPC transport to connect
1650 * Invoked by a work queue tasklet.
1652 static void xs_udp_connect_worker4(struct work_struct
*work
)
1654 struct sock_xprt
*transport
=
1655 container_of(work
, struct sock_xprt
, connect_worker
.work
);
1656 struct rpc_xprt
*xprt
= &transport
->xprt
;
1657 struct socket
*sock
= transport
->sock
;
1658 int err
, status
= -EIO
;
1663 /* Start by resetting any existing state */
1664 xs_reset_transport(transport
);
1666 err
= sock_create_kern(PF_INET
, SOCK_DGRAM
, IPPROTO_UDP
, &sock
);
1668 dprintk("RPC: can't create UDP transport socket (%d).\n", -err
);
1671 xs_reclassify_socket4(sock
);
1673 if (xs_bind4(transport
, sock
)) {
1678 dprintk("RPC: worker connecting xprt %p via %s to "
1679 "%s (port %s)\n", xprt
,
1680 xprt
->address_strings
[RPC_DISPLAY_PROTO
],
1681 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
1682 xprt
->address_strings
[RPC_DISPLAY_PORT
]);
1684 xs_udp_finish_connecting(xprt
, sock
);
1687 xprt_clear_connecting(xprt
);
1688 xprt_wake_pending_tasks(xprt
, status
);
1692 * xs_udp_connect_worker6 - set up a UDP socket
1693 * @work: RPC transport to connect
1695 * Invoked by a work queue tasklet.
1697 static void xs_udp_connect_worker6(struct work_struct
*work
)
1699 struct sock_xprt
*transport
=
1700 container_of(work
, struct sock_xprt
, connect_worker
.work
);
1701 struct rpc_xprt
*xprt
= &transport
->xprt
;
1702 struct socket
*sock
= transport
->sock
;
1703 int err
, status
= -EIO
;
1708 /* Start by resetting any existing state */
1709 xs_reset_transport(transport
);
1711 err
= sock_create_kern(PF_INET6
, SOCK_DGRAM
, IPPROTO_UDP
, &sock
);
1713 dprintk("RPC: can't create UDP transport socket (%d).\n", -err
);
1716 xs_reclassify_socket6(sock
);
1718 if (xs_bind6(transport
, sock
) < 0) {
1723 dprintk("RPC: worker connecting xprt %p via %s to "
1724 "%s (port %s)\n", xprt
,
1725 xprt
->address_strings
[RPC_DISPLAY_PROTO
],
1726 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
1727 xprt
->address_strings
[RPC_DISPLAY_PORT
]);
1729 xs_udp_finish_connecting(xprt
, sock
);
1732 xprt_clear_connecting(xprt
);
1733 xprt_wake_pending_tasks(xprt
, status
);
1737 * We need to preserve the port number so the reply cache on the server can
1738 * find our cached RPC replies when we get around to reconnecting.
1740 static void xs_abort_connection(struct rpc_xprt
*xprt
, struct sock_xprt
*transport
)
1743 struct sockaddr any
;
1745 dprintk("RPC: disconnecting xprt %p to reuse port\n", xprt
);
1748 * Disconnect the transport socket by doing a connect operation
1749 * with AF_UNSPEC. This should return immediately...
1751 memset(&any
, 0, sizeof(any
));
1752 any
.sa_family
= AF_UNSPEC
;
1753 result
= kernel_connect(transport
->sock
, &any
, sizeof(any
), 0);
1755 xs_sock_mark_closed(xprt
);
1757 dprintk("RPC: AF_UNSPEC connect return code %d\n",
1761 static void xs_tcp_reuse_connection(struct rpc_xprt
*xprt
, struct sock_xprt
*transport
)
1763 unsigned int state
= transport
->inet
->sk_state
;
1765 if (state
== TCP_CLOSE
&& transport
->sock
->state
== SS_UNCONNECTED
) {
1766 /* we don't need to abort the connection if the socket
1767 * hasn't undergone a shutdown
1769 if (transport
->inet
->sk_shutdown
== 0)
1771 dprintk("RPC: %s: TCP_CLOSEd and sk_shutdown set to %d\n",
1772 __func__
, transport
->inet
->sk_shutdown
);
1774 if ((1 << state
) & (TCPF_ESTABLISHED
|TCPF_SYN_SENT
)) {
1775 /* we don't need to abort the connection if the socket
1776 * hasn't undergone a shutdown
1778 if (transport
->inet
->sk_shutdown
== 0)
1780 dprintk("RPC: %s: ESTABLISHED/SYN_SENT "
1781 "sk_shutdown set to %d\n",
1782 __func__
, transport
->inet
->sk_shutdown
);
1784 xs_abort_connection(xprt
, transport
);
1787 static int xs_tcp_finish_connecting(struct rpc_xprt
*xprt
, struct socket
*sock
)
1789 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1791 if (!transport
->inet
) {
1792 struct sock
*sk
= sock
->sk
;
1794 write_lock_bh(&sk
->sk_callback_lock
);
1796 xs_save_old_callbacks(transport
, sk
);
1798 sk
->sk_user_data
= xprt
;
1799 sk
->sk_data_ready
= xs_tcp_data_ready
;
1800 sk
->sk_state_change
= xs_tcp_state_change
;
1801 sk
->sk_write_space
= xs_tcp_write_space
;
1802 sk
->sk_error_report
= xs_error_report
;
1803 sk
->sk_allocation
= GFP_ATOMIC
;
1805 /* socket options */
1806 sk
->sk_userlocks
|= SOCK_BINDPORT_LOCK
;
1807 sock_reset_flag(sk
, SOCK_LINGER
);
1808 tcp_sk(sk
)->linger2
= 0;
1809 tcp_sk(sk
)->nonagle
|= TCP_NAGLE_OFF
;
1811 xprt_clear_connected(xprt
);
1813 /* Reset to new socket */
1814 transport
->sock
= sock
;
1815 transport
->inet
= sk
;
1817 write_unlock_bh(&sk
->sk_callback_lock
);
1820 if (!xprt_bound(xprt
))
1823 /* Tell the socket layer to start connecting... */
1824 xprt
->stat
.connect_count
++;
1825 xprt
->stat
.connect_start
= jiffies
;
1826 return kernel_connect(sock
, xs_addr(xprt
), xprt
->addrlen
, O_NONBLOCK
);
1830 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
1831 * @xprt: RPC transport to connect
1832 * @transport: socket transport to connect
1833 * @create_sock: function to create a socket of the correct type
1835 * Invoked by a work queue tasklet.
1837 static void xs_tcp_setup_socket(struct rpc_xprt
*xprt
,
1838 struct sock_xprt
*transport
,
1839 struct socket
*(*create_sock
)(struct rpc_xprt
*,
1840 struct sock_xprt
*))
1842 struct socket
*sock
= transport
->sock
;
1849 clear_bit(XPRT_CONNECTION_ABORT
, &xprt
->state
);
1850 sock
= create_sock(xprt
, transport
);
1852 status
= PTR_ERR(sock
);
1858 abort_and_exit
= test_and_clear_bit(XPRT_CONNECTION_ABORT
,
1860 /* "close" the socket, preserving the local port */
1861 xs_tcp_reuse_connection(xprt
, transport
);
1867 dprintk("RPC: worker connecting xprt %p via %s to "
1868 "%s (port %s)\n", xprt
,
1869 xprt
->address_strings
[RPC_DISPLAY_PROTO
],
1870 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
1871 xprt
->address_strings
[RPC_DISPLAY_PORT
]);
1873 status
= xs_tcp_finish_connecting(xprt
, sock
);
1874 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
1875 xprt
, -status
, xprt_connected(xprt
),
1876 sock
->sk
->sk_state
);
1879 printk("%s: connect returned unhandled error %d\n",
1881 case -EADDRNOTAVAIL
:
1882 /* We're probably in TIME_WAIT. Get rid of existing socket,
1885 set_bit(XPRT_CONNECTION_CLOSE
, &xprt
->state
);
1886 xprt_force_disconnect(xprt
);
1891 /* retry with existing socket, after a delay */
1895 xprt_clear_connecting(xprt
);
1898 /* Happens, for instance, if the user specified a link
1899 * local IPv6 address without a scope-id.
1906 xprt_clear_connecting(xprt
);
1907 xprt_wake_pending_tasks(xprt
, status
);
1910 static struct socket
*xs_create_tcp_sock4(struct rpc_xprt
*xprt
,
1911 struct sock_xprt
*transport
)
1913 struct socket
*sock
;
1916 /* start from scratch */
1917 err
= sock_create_kern(PF_INET
, SOCK_STREAM
, IPPROTO_TCP
, &sock
);
1919 dprintk("RPC: can't create TCP transport socket (%d).\n",
1923 xs_reclassify_socket4(sock
);
1925 if (xs_bind4(transport
, sock
) < 0) {
1931 return ERR_PTR(-EIO
);
1935 * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1936 * @work: RPC transport to connect
1938 * Invoked by a work queue tasklet.
1940 static void xs_tcp_connect_worker4(struct work_struct
*work
)
1942 struct sock_xprt
*transport
=
1943 container_of(work
, struct sock_xprt
, connect_worker
.work
);
1944 struct rpc_xprt
*xprt
= &transport
->xprt
;
1946 xs_tcp_setup_socket(xprt
, transport
, xs_create_tcp_sock4
);
1949 static struct socket
*xs_create_tcp_sock6(struct rpc_xprt
*xprt
,
1950 struct sock_xprt
*transport
)
1952 struct socket
*sock
;
1955 /* start from scratch */
1956 err
= sock_create_kern(PF_INET6
, SOCK_STREAM
, IPPROTO_TCP
, &sock
);
1958 dprintk("RPC: can't create TCP transport socket (%d).\n",
1962 xs_reclassify_socket6(sock
);
1964 if (xs_bind6(transport
, sock
) < 0) {
1970 return ERR_PTR(-EIO
);
1974 * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
1975 * @work: RPC transport to connect
1977 * Invoked by a work queue tasklet.
1979 static void xs_tcp_connect_worker6(struct work_struct
*work
)
1981 struct sock_xprt
*transport
=
1982 container_of(work
, struct sock_xprt
, connect_worker
.work
);
1983 struct rpc_xprt
*xprt
= &transport
->xprt
;
1985 xs_tcp_setup_socket(xprt
, transport
, xs_create_tcp_sock6
);
1989 * xs_connect - connect a socket to a remote endpoint
1990 * @task: address of RPC task that manages state of connect request
1992 * TCP: If the remote end dropped the connection, delay reconnecting.
1994 * UDP socket connects are synchronous, but we use a work queue anyway
1995 * to guarantee that even unprivileged user processes can set up a
1996 * socket on a privileged port.
1998 * If a UDP socket connect fails, the delay behavior here prevents
1999 * retry floods (hard mounts).
2001 static void xs_connect(struct rpc_task
*task
)
2003 struct rpc_xprt
*xprt
= task
->tk_xprt
;
2004 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2006 if (transport
->sock
!= NULL
&& !RPC_IS_SOFTCONN(task
)) {
2007 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2009 xprt
, xprt
->reestablish_timeout
/ HZ
);
2010 queue_delayed_work(rpciod_workqueue
,
2011 &transport
->connect_worker
,
2012 xprt
->reestablish_timeout
);
2013 xprt
->reestablish_timeout
<<= 1;
2014 if (xprt
->reestablish_timeout
< XS_TCP_INIT_REEST_TO
)
2015 xprt
->reestablish_timeout
= XS_TCP_INIT_REEST_TO
;
2016 if (xprt
->reestablish_timeout
> XS_TCP_MAX_REEST_TO
)
2017 xprt
->reestablish_timeout
= XS_TCP_MAX_REEST_TO
;
2019 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt
);
2020 queue_delayed_work(rpciod_workqueue
,
2021 &transport
->connect_worker
, 0);
2026 * xs_udp_print_stats - display UDP socket-specifc stats
2027 * @xprt: rpc_xprt struct containing statistics
2031 static void xs_udp_print_stats(struct rpc_xprt
*xprt
, struct seq_file
*seq
)
2033 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2035 seq_printf(seq
, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
2037 xprt
->stat
.bind_count
,
2040 xprt
->stat
.bad_xids
,
2042 xprt
->stat
.bklog_u
);
2046 * xs_tcp_print_stats - display TCP socket-specifc stats
2047 * @xprt: rpc_xprt struct containing statistics
2051 static void xs_tcp_print_stats(struct rpc_xprt
*xprt
, struct seq_file
*seq
)
2053 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2056 if (xprt_connected(xprt
))
2057 idle_time
= (long)(jiffies
- xprt
->last_used
) / HZ
;
2059 seq_printf(seq
, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
2061 xprt
->stat
.bind_count
,
2062 xprt
->stat
.connect_count
,
2063 xprt
->stat
.connect_time
,
2067 xprt
->stat
.bad_xids
,
2069 xprt
->stat
.bklog_u
);
2073 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2074 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2075 * to use the server side send routines.
2077 static void *bc_malloc(struct rpc_task
*task
, size_t size
)
2080 struct rpc_buffer
*buf
;
2082 BUG_ON(size
> PAGE_SIZE
- sizeof(struct rpc_buffer
));
2083 page
= alloc_page(GFP_KERNEL
);
2088 buf
= page_address(page
);
2089 buf
->len
= PAGE_SIZE
;
2095 * Free the space allocated in the bc_alloc routine
2097 static void bc_free(void *buffer
)
2099 struct rpc_buffer
*buf
;
2104 buf
= container_of(buffer
, struct rpc_buffer
, data
);
2105 free_page((unsigned long)buf
);
2109 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2110 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2112 static int bc_sendto(struct rpc_rqst
*req
)
2115 struct xdr_buf
*xbufp
= &req
->rq_snd_buf
;
2116 struct rpc_xprt
*xprt
= req
->rq_xprt
;
2117 struct sock_xprt
*transport
=
2118 container_of(xprt
, struct sock_xprt
, xprt
);
2119 struct socket
*sock
= transport
->sock
;
2120 unsigned long headoff
;
2121 unsigned long tailoff
;
2124 * Set up the rpc header and record marker stuff
2126 xs_encode_tcp_record_marker(xbufp
);
2128 tailoff
= (unsigned long)xbufp
->tail
[0].iov_base
& ~PAGE_MASK
;
2129 headoff
= (unsigned long)xbufp
->head
[0].iov_base
& ~PAGE_MASK
;
2130 len
= svc_send_common(sock
, xbufp
,
2131 virt_to_page(xbufp
->head
[0].iov_base
), headoff
,
2132 xbufp
->tail
[0].iov_base
, tailoff
);
2134 if (len
!= xbufp
->len
) {
2135 printk(KERN_NOTICE
"Error sending entire callback!\n");
2143 * The send routine. Borrows from svc_send
2145 static int bc_send_request(struct rpc_task
*task
)
2147 struct rpc_rqst
*req
= task
->tk_rqstp
;
2148 struct svc_xprt
*xprt
;
2149 struct svc_sock
*svsk
;
2152 dprintk("sending request with xid: %08x\n", ntohl(req
->rq_xid
));
2154 * Get the server socket associated with this callback xprt
2156 xprt
= req
->rq_xprt
->bc_xprt
;
2157 svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
2160 * Grab the mutex to serialize data as the connection is shared
2161 * with the fore channel
2163 if (!mutex_trylock(&xprt
->xpt_mutex
)) {
2164 rpc_sleep_on(&xprt
->xpt_bc_pending
, task
, NULL
);
2165 if (!mutex_trylock(&xprt
->xpt_mutex
))
2167 rpc_wake_up_queued_task(&xprt
->xpt_bc_pending
, task
);
2169 if (test_bit(XPT_DEAD
, &xprt
->xpt_flags
))
2172 len
= bc_sendto(req
);
2173 mutex_unlock(&xprt
->xpt_mutex
);
2182 * The close routine. Since this is client initiated, we do nothing
2185 static void bc_close(struct rpc_xprt
*xprt
)
2190 * The xprt destroy routine. Again, because this connection is client
2191 * initiated, we do nothing
2194 static void bc_destroy(struct rpc_xprt
*xprt
)
2198 static struct rpc_xprt_ops xs_udp_ops
= {
2199 .set_buffer_size
= xs_udp_set_buffer_size
,
2200 .reserve_xprt
= xprt_reserve_xprt_cong
,
2201 .release_xprt
= xprt_release_xprt_cong
,
2202 .rpcbind
= rpcb_getport_async
,
2203 .set_port
= xs_set_port
,
2204 .connect
= xs_connect
,
2205 .buf_alloc
= rpc_malloc
,
2206 .buf_free
= rpc_free
,
2207 .send_request
= xs_udp_send_request
,
2208 .set_retrans_timeout
= xprt_set_retrans_timeout_rtt
,
2209 .timer
= xs_udp_timer
,
2210 .release_request
= xprt_release_rqst_cong
,
2212 .destroy
= xs_destroy
,
2213 .print_stats
= xs_udp_print_stats
,
2216 static struct rpc_xprt_ops xs_tcp_ops
= {
2217 .reserve_xprt
= xprt_reserve_xprt
,
2218 .release_xprt
= xs_tcp_release_xprt
,
2219 .rpcbind
= rpcb_getport_async
,
2220 .set_port
= xs_set_port
,
2221 .connect
= xs_connect
,
2222 .buf_alloc
= rpc_malloc
,
2223 .buf_free
= rpc_free
,
2224 .send_request
= xs_tcp_send_request
,
2225 .set_retrans_timeout
= xprt_set_retrans_timeout_def
,
2226 .close
= xs_tcp_close
,
2227 .destroy
= xs_destroy
,
2228 .print_stats
= xs_tcp_print_stats
,
2232 * The rpc_xprt_ops for the server backchannel
2235 static struct rpc_xprt_ops bc_tcp_ops
= {
2236 .reserve_xprt
= xprt_reserve_xprt
,
2237 .release_xprt
= xprt_release_xprt
,
2238 .buf_alloc
= bc_malloc
,
2239 .buf_free
= bc_free
,
2240 .send_request
= bc_send_request
,
2241 .set_retrans_timeout
= xprt_set_retrans_timeout_def
,
2243 .destroy
= bc_destroy
,
2244 .print_stats
= xs_tcp_print_stats
,
2247 static struct rpc_xprt
*xs_setup_xprt(struct xprt_create
*args
,
2248 unsigned int slot_table_size
)
2250 struct rpc_xprt
*xprt
;
2251 struct sock_xprt
*new;
2253 if (args
->addrlen
> sizeof(xprt
->addr
)) {
2254 dprintk("RPC: xs_setup_xprt: address too large\n");
2255 return ERR_PTR(-EBADF
);
2258 new = kzalloc(sizeof(*new), GFP_KERNEL
);
2260 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2262 return ERR_PTR(-ENOMEM
);
2266 xprt
->max_reqs
= slot_table_size
;
2267 xprt
->slot
= kcalloc(xprt
->max_reqs
, sizeof(struct rpc_rqst
), GFP_KERNEL
);
2268 if (xprt
->slot
== NULL
) {
2270 dprintk("RPC: xs_setup_xprt: couldn't allocate slot "
2272 return ERR_PTR(-ENOMEM
);
2275 memcpy(&xprt
->addr
, args
->dstaddr
, args
->addrlen
);
2276 xprt
->addrlen
= args
->addrlen
;
2278 memcpy(&new->srcaddr
, args
->srcaddr
, args
->addrlen
);
2283 static const struct rpc_timeout xs_udp_default_timeout
= {
2284 .to_initval
= 5 * HZ
,
2285 .to_maxval
= 30 * HZ
,
2286 .to_increment
= 5 * HZ
,
2291 * xs_setup_udp - Set up transport to use a UDP socket
2292 * @args: rpc transport creation arguments
2295 static struct rpc_xprt
*xs_setup_udp(struct xprt_create
*args
)
2297 struct sockaddr
*addr
= args
->dstaddr
;
2298 struct rpc_xprt
*xprt
;
2299 struct sock_xprt
*transport
;
2300 struct rpc_xprt
*ret
;
2302 xprt
= xs_setup_xprt(args
, xprt_udp_slot_table_entries
);
2305 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2307 xprt
->prot
= IPPROTO_UDP
;
2309 xprt
->max_payload
= (1U << 16) - (MAX_HEADER
<< 3);
2311 xprt
->bind_timeout
= XS_BIND_TO
;
2312 xprt
->reestablish_timeout
= XS_UDP_REEST_TO
;
2313 xprt
->idle_timeout
= XS_IDLE_DISC_TO
;
2315 xprt
->ops
= &xs_udp_ops
;
2317 xprt
->timeout
= &xs_udp_default_timeout
;
2319 switch (addr
->sa_family
) {
2321 if (((struct sockaddr_in
*)addr
)->sin_port
!= htons(0))
2322 xprt_set_bound(xprt
);
2324 INIT_DELAYED_WORK(&transport
->connect_worker
,
2325 xs_udp_connect_worker4
);
2326 xs_format_peer_addresses(xprt
, "udp", RPCBIND_NETID_UDP
);
2329 if (((struct sockaddr_in6
*)addr
)->sin6_port
!= htons(0))
2330 xprt_set_bound(xprt
);
2332 INIT_DELAYED_WORK(&transport
->connect_worker
,
2333 xs_udp_connect_worker6
);
2334 xs_format_peer_addresses(xprt
, "udp", RPCBIND_NETID_UDP6
);
2337 ret
= ERR_PTR(-EAFNOSUPPORT
);
2341 if (xprt_bound(xprt
))
2342 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2343 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
2344 xprt
->address_strings
[RPC_DISPLAY_PORT
],
2345 xprt
->address_strings
[RPC_DISPLAY_PROTO
]);
2347 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2348 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
2349 xprt
->address_strings
[RPC_DISPLAY_PROTO
]);
2351 if (try_module_get(THIS_MODULE
))
2353 ret
= ERR_PTR(-EINVAL
);
2360 static const struct rpc_timeout xs_tcp_default_timeout
= {
2361 .to_initval
= 60 * HZ
,
2362 .to_maxval
= 60 * HZ
,
2367 * xs_setup_tcp - Set up transport to use a TCP socket
2368 * @args: rpc transport creation arguments
2371 static struct rpc_xprt
*xs_setup_tcp(struct xprt_create
*args
)
2373 struct sockaddr
*addr
= args
->dstaddr
;
2374 struct rpc_xprt
*xprt
;
2375 struct sock_xprt
*transport
;
2376 struct rpc_xprt
*ret
;
2378 xprt
= xs_setup_xprt(args
, xprt_tcp_slot_table_entries
);
2381 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2383 xprt
->prot
= IPPROTO_TCP
;
2384 xprt
->tsh_size
= sizeof(rpc_fraghdr
) / sizeof(u32
);
2385 xprt
->max_payload
= RPC_MAX_FRAGMENT_SIZE
;
2387 xprt
->bind_timeout
= XS_BIND_TO
;
2388 xprt
->reestablish_timeout
= XS_TCP_INIT_REEST_TO
;
2389 xprt
->idle_timeout
= XS_IDLE_DISC_TO
;
2391 xprt
->ops
= &xs_tcp_ops
;
2392 xprt
->timeout
= &xs_tcp_default_timeout
;
2394 switch (addr
->sa_family
) {
2396 if (((struct sockaddr_in
*)addr
)->sin_port
!= htons(0))
2397 xprt_set_bound(xprt
);
2399 INIT_DELAYED_WORK(&transport
->connect_worker
,
2400 xs_tcp_connect_worker4
);
2401 xs_format_peer_addresses(xprt
, "tcp", RPCBIND_NETID_TCP
);
2404 if (((struct sockaddr_in6
*)addr
)->sin6_port
!= htons(0))
2405 xprt_set_bound(xprt
);
2407 INIT_DELAYED_WORK(&transport
->connect_worker
,
2408 xs_tcp_connect_worker6
);
2409 xs_format_peer_addresses(xprt
, "tcp", RPCBIND_NETID_TCP6
);
2412 ret
= ERR_PTR(-EAFNOSUPPORT
);
2416 if (xprt_bound(xprt
))
2417 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2418 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
2419 xprt
->address_strings
[RPC_DISPLAY_PORT
],
2420 xprt
->address_strings
[RPC_DISPLAY_PROTO
]);
2422 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2423 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
2424 xprt
->address_strings
[RPC_DISPLAY_PROTO
]);
2427 if (try_module_get(THIS_MODULE
))
2429 ret
= ERR_PTR(-EINVAL
);
2437 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2438 * @args: rpc transport creation arguments
2441 static struct rpc_xprt
*xs_setup_bc_tcp(struct xprt_create
*args
)
2443 struct sockaddr
*addr
= args
->dstaddr
;
2444 struct rpc_xprt
*xprt
;
2445 struct sock_xprt
*transport
;
2446 struct svc_sock
*bc_sock
;
2447 struct rpc_xprt
*ret
;
2449 xprt
= xs_setup_xprt(args
, xprt_tcp_slot_table_entries
);
2452 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2454 xprt
->prot
= IPPROTO_TCP
;
2455 xprt
->tsh_size
= sizeof(rpc_fraghdr
) / sizeof(u32
);
2456 xprt
->max_payload
= RPC_MAX_FRAGMENT_SIZE
;
2457 xprt
->timeout
= &xs_tcp_default_timeout
;
2460 xprt_set_bound(xprt
);
2461 xprt
->bind_timeout
= 0;
2462 xprt
->reestablish_timeout
= 0;
2463 xprt
->idle_timeout
= 0;
2466 * The backchannel uses the same socket connection as the
2469 xprt
->bc_xprt
= args
->bc_xprt
;
2470 bc_sock
= container_of(args
->bc_xprt
, struct svc_sock
, sk_xprt
);
2471 bc_sock
->sk_bc_xprt
= xprt
;
2472 transport
->sock
= bc_sock
->sk_sock
;
2473 transport
->inet
= bc_sock
->sk_sk
;
2475 xprt
->ops
= &bc_tcp_ops
;
2477 switch (addr
->sa_family
) {
2479 xs_format_peer_addresses(xprt
, "tcp",
2483 xs_format_peer_addresses(xprt
, "tcp",
2484 RPCBIND_NETID_TCP6
);
2487 ret
= ERR_PTR(-EAFNOSUPPORT
);
2491 if (xprt_bound(xprt
))
2492 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2493 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
2494 xprt
->address_strings
[RPC_DISPLAY_PORT
],
2495 xprt
->address_strings
[RPC_DISPLAY_PROTO
]);
2497 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2498 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
2499 xprt
->address_strings
[RPC_DISPLAY_PROTO
]);
2502 * Since we don't want connections for the backchannel, we set
2503 * the xprt status to connected
2505 xprt_set_connected(xprt
);
2508 if (try_module_get(THIS_MODULE
))
2510 ret
= ERR_PTR(-EINVAL
);
2517 static struct xprt_class xs_udp_transport
= {
2518 .list
= LIST_HEAD_INIT(xs_udp_transport
.list
),
2520 .owner
= THIS_MODULE
,
2521 .ident
= XPRT_TRANSPORT_UDP
,
2522 .setup
= xs_setup_udp
,
2525 static struct xprt_class xs_tcp_transport
= {
2526 .list
= LIST_HEAD_INIT(xs_tcp_transport
.list
),
2528 .owner
= THIS_MODULE
,
2529 .ident
= XPRT_TRANSPORT_TCP
,
2530 .setup
= xs_setup_tcp
,
2533 static struct xprt_class xs_bc_tcp_transport
= {
2534 .list
= LIST_HEAD_INIT(xs_bc_tcp_transport
.list
),
2535 .name
= "tcp NFSv4.1 backchannel",
2536 .owner
= THIS_MODULE
,
2537 .ident
= XPRT_TRANSPORT_BC_TCP
,
2538 .setup
= xs_setup_bc_tcp
,
2542 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2545 int init_socket_xprt(void)
2548 if (!sunrpc_table_header
)
2549 sunrpc_table_header
= register_sysctl_table(sunrpc_table
);
2552 xprt_register_transport(&xs_udp_transport
);
2553 xprt_register_transport(&xs_tcp_transport
);
2554 xprt_register_transport(&xs_bc_tcp_transport
);
2560 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2563 void cleanup_socket_xprt(void)
2566 if (sunrpc_table_header
) {
2567 unregister_sysctl_table(sunrpc_table_header
);
2568 sunrpc_table_header
= NULL
;
2572 xprt_unregister_transport(&xs_udp_transport
);
2573 xprt_unregister_transport(&xs_tcp_transport
);
2574 xprt_unregister_transport(&xs_bc_tcp_transport
);
2577 static int param_set_uint_minmax(const char *val
,
2578 const struct kernel_param
*kp
,
2579 unsigned int min
, unsigned int max
)
2586 ret
= strict_strtoul(val
, 0, &num
);
2587 if (ret
== -EINVAL
|| num
< min
|| num
> max
)
2589 *((unsigned int *)kp
->arg
) = num
;
2593 static int param_set_portnr(const char *val
, const struct kernel_param
*kp
)
2595 return param_set_uint_minmax(val
, kp
,
2600 static struct kernel_param_ops param_ops_portnr
= {
2601 .set
= param_set_portnr
,
2602 .get
= param_get_uint
,
2605 #define param_check_portnr(name, p) \
2606 __param_check(name, p, unsigned int);
2608 module_param_named(min_resvport
, xprt_min_resvport
, portnr
, 0644);
2609 module_param_named(max_resvport
, xprt_max_resvport
, portnr
, 0644);
2611 static int param_set_slot_table_size(const char *val
,
2612 const struct kernel_param
*kp
)
2614 return param_set_uint_minmax(val
, kp
,
2616 RPC_MAX_SLOT_TABLE
);
2619 static struct kernel_param_ops param_ops_slot_table_size
= {
2620 .set
= param_set_slot_table_size
,
2621 .get
= param_get_uint
,
2624 #define param_check_slot_table_size(name, p) \
2625 __param_check(name, p, unsigned int);
2627 module_param_named(tcp_slot_table_entries
, xprt_tcp_slot_table_entries
,
2628 slot_table_size
, 0644);
2629 module_param_named(udp_slot_table_entries
, xprt_udp_slot_table_entries
,
2630 slot_table_size
, 0644);