2 * linux/net/sunrpc/xprtsock.c
4 * Client-side transport implementation for sockets.
6 * TCP callback races fixes (C) 1998 Red Hat Software <alan@redhat.com>
7 * TCP send fixes (C) 1998 Red Hat Software <alan@redhat.com>
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>
18 #include <linux/types.h>
19 #include <linux/slab.h>
20 #include <linux/capability.h>
21 #include <linux/pagemap.h>
22 #include <linux/errno.h>
23 #include <linux/socket.h>
25 #include <linux/net.h>
27 #include <linux/udp.h>
28 #include <linux/tcp.h>
29 #include <linux/sunrpc/clnt.h>
30 #include <linux/sunrpc/sched.h>
31 #include <linux/file.h>
34 #include <net/checksum.h>
41 unsigned int xprt_udp_slot_table_entries
= RPC_DEF_SLOT_TABLE
;
42 unsigned int xprt_tcp_slot_table_entries
= RPC_DEF_SLOT_TABLE
;
44 unsigned int xprt_min_resvport
= RPC_DEF_MIN_RESVPORT
;
45 unsigned int xprt_max_resvport
= RPC_DEF_MAX_RESVPORT
;
48 * We can register our own files under /proc/sys/sunrpc by
49 * calling register_sysctl_table() again. The files in that
50 * directory become the union of all files registered there.
52 * We simply need to make sure that we don't collide with
53 * someone else's file names!
58 static unsigned int min_slot_table_size
= RPC_MIN_SLOT_TABLE
;
59 static unsigned int max_slot_table_size
= RPC_MAX_SLOT_TABLE
;
60 static unsigned int xprt_min_resvport_limit
= RPC_MIN_RESVPORT
;
61 static unsigned int xprt_max_resvport_limit
= RPC_MAX_RESVPORT
;
63 static struct ctl_table_header
*sunrpc_table_header
;
66 * FIXME: changing the UDP slot table size should also resize the UDP
67 * socket buffers for existing UDP transports
69 static ctl_table xs_tunables_table
[] = {
71 .ctl_name
= CTL_SLOTTABLE_UDP
,
72 .procname
= "udp_slot_table_entries",
73 .data
= &xprt_udp_slot_table_entries
,
74 .maxlen
= sizeof(unsigned int),
76 .proc_handler
= &proc_dointvec_minmax
,
77 .strategy
= &sysctl_intvec
,
78 .extra1
= &min_slot_table_size
,
79 .extra2
= &max_slot_table_size
82 .ctl_name
= CTL_SLOTTABLE_TCP
,
83 .procname
= "tcp_slot_table_entries",
84 .data
= &xprt_tcp_slot_table_entries
,
85 .maxlen
= sizeof(unsigned int),
87 .proc_handler
= &proc_dointvec_minmax
,
88 .strategy
= &sysctl_intvec
,
89 .extra1
= &min_slot_table_size
,
90 .extra2
= &max_slot_table_size
93 .ctl_name
= CTL_MIN_RESVPORT
,
94 .procname
= "min_resvport",
95 .data
= &xprt_min_resvport
,
96 .maxlen
= sizeof(unsigned int),
98 .proc_handler
= &proc_dointvec_minmax
,
99 .strategy
= &sysctl_intvec
,
100 .extra1
= &xprt_min_resvport_limit
,
101 .extra2
= &xprt_max_resvport_limit
104 .ctl_name
= CTL_MAX_RESVPORT
,
105 .procname
= "max_resvport",
106 .data
= &xprt_max_resvport
,
107 .maxlen
= sizeof(unsigned int),
109 .proc_handler
= &proc_dointvec_minmax
,
110 .strategy
= &sysctl_intvec
,
111 .extra1
= &xprt_min_resvport_limit
,
112 .extra2
= &xprt_max_resvport_limit
119 static ctl_table sunrpc_table
[] = {
121 .ctl_name
= CTL_SUNRPC
,
122 .procname
= "sunrpc",
124 .child
= xs_tunables_table
134 * How many times to try sending a request on a socket before waiting
135 * for the socket buffer to clear.
137 #define XS_SENDMSG_RETRY (10U)
140 * Time out for an RPC UDP socket connect. UDP socket connects are
141 * synchronous, but we set a timeout anyway in case of resource
142 * exhaustion on the local host.
144 #define XS_UDP_CONN_TO (5U * HZ)
147 * Wait duration for an RPC TCP connection to be established. Solaris
148 * NFS over TCP uses 60 seconds, for example, which is in line with how
149 * long a server takes to reboot.
151 #define XS_TCP_CONN_TO (60U * HZ)
154 * Wait duration for a reply from the RPC portmapper.
156 #define XS_BIND_TO (60U * HZ)
159 * Delay if a UDP socket connect error occurs. This is most likely some
160 * kind of resource problem on the local host.
162 #define XS_UDP_REEST_TO (2U * HZ)
165 * The reestablish timeout allows clients to delay for a bit before attempting
166 * to reconnect to a server that just dropped our connection.
168 * We implement an exponential backoff when trying to reestablish a TCP
169 * transport connection with the server. Some servers like to drop a TCP
170 * connection when they are overworked, so we start with a short timeout and
171 * increase over time if the server is down or not responding.
173 #define XS_TCP_INIT_REEST_TO (3U * HZ)
174 #define XS_TCP_MAX_REEST_TO (5U * 60 * HZ)
177 * TCP idle timeout; client drops the transport socket if it is idle
178 * for this long. Note that we also timeout UDP sockets to prevent
179 * holding port numbers when there is no RPC traffic.
181 #define XS_IDLE_DISC_TO (5U * 60 * HZ)
184 # undef RPC_DEBUG_DATA
185 # define RPCDBG_FACILITY RPCDBG_TRANS
188 #ifdef RPC_DEBUG_DATA
189 static void xs_pktdump(char *msg
, u32
*packet
, unsigned int count
)
191 u8
*buf
= (u8
*) packet
;
194 dprintk("RPC: %s\n", msg
);
195 for (j
= 0; j
< count
&& j
< 128; j
+= 4) {
199 dprintk("0x%04x ", j
);
201 dprintk("%02x%02x%02x%02x ",
202 buf
[j
], buf
[j
+1], buf
[j
+2], buf
[j
+3]);
207 static inline void xs_pktdump(char *msg
, u32
*packet
, unsigned int count
)
214 struct rpc_xprt xprt
;
219 struct socket
* sock
;
223 * State of TCP reply receive
231 unsigned long tcp_copied
,
235 * Connection of transports
237 struct delayed_work connect_worker
;
238 struct sockaddr_storage addr
;
242 * UDP socket buffer size parameters
248 * Saved socket callback addresses
250 void (*old_data_ready
)(struct sock
*, int);
251 void (*old_state_change
)(struct sock
*);
252 void (*old_write_space
)(struct sock
*);
256 * TCP receive state flags
258 #define TCP_RCV_LAST_FRAG (1UL << 0)
259 #define TCP_RCV_COPY_FRAGHDR (1UL << 1)
260 #define TCP_RCV_COPY_XID (1UL << 2)
261 #define TCP_RCV_COPY_DATA (1UL << 3)
263 static void xs_format_peer_addresses(struct rpc_xprt
*xprt
)
265 struct sockaddr_in
*addr
= (struct sockaddr_in
*) &xprt
->addr
;
268 buf
= kzalloc(20, GFP_KERNEL
);
270 snprintf(buf
, 20, "%u.%u.%u.%u",
271 NIPQUAD(addr
->sin_addr
.s_addr
));
273 xprt
->address_strings
[RPC_DISPLAY_ADDR
] = buf
;
275 buf
= kzalloc(8, GFP_KERNEL
);
277 snprintf(buf
, 8, "%u",
278 ntohs(addr
->sin_port
));
280 xprt
->address_strings
[RPC_DISPLAY_PORT
] = buf
;
282 if (xprt
->prot
== IPPROTO_UDP
)
283 xprt
->address_strings
[RPC_DISPLAY_PROTO
] = "udp";
285 xprt
->address_strings
[RPC_DISPLAY_PROTO
] = "tcp";
287 buf
= kzalloc(48, GFP_KERNEL
);
289 snprintf(buf
, 48, "addr=%u.%u.%u.%u port=%u proto=%s",
290 NIPQUAD(addr
->sin_addr
.s_addr
),
291 ntohs(addr
->sin_port
),
292 xprt
->prot
== IPPROTO_UDP
? "udp" : "tcp");
294 xprt
->address_strings
[RPC_DISPLAY_ALL
] = buf
;
297 static void xs_free_peer_addresses(struct rpc_xprt
*xprt
)
299 kfree(xprt
->address_strings
[RPC_DISPLAY_ADDR
]);
300 kfree(xprt
->address_strings
[RPC_DISPLAY_PORT
]);
301 kfree(xprt
->address_strings
[RPC_DISPLAY_ALL
]);
304 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
306 static int xs_send_kvec(struct socket
*sock
, struct sockaddr
*addr
, int addrlen
, struct kvec
*vec
, unsigned int base
, int more
)
308 struct msghdr msg
= {
310 .msg_namelen
= addrlen
,
311 .msg_flags
= XS_SENDMSG_FLAGS
| (more
? MSG_MORE
: 0),
314 .iov_base
= vec
->iov_base
+ base
,
315 .iov_len
= vec
->iov_len
- base
,
318 if (iov
.iov_len
!= 0)
319 return kernel_sendmsg(sock
, &msg
, &iov
, 1, iov
.iov_len
);
320 return kernel_sendmsg(sock
, &msg
, NULL
, 0, 0);
323 static int xs_send_pagedata(struct socket
*sock
, struct xdr_buf
*xdr
, unsigned int base
, int more
)
326 unsigned int remainder
;
329 remainder
= xdr
->page_len
- base
;
330 base
+= xdr
->page_base
;
331 ppage
= xdr
->pages
+ (base
>> PAGE_SHIFT
);
334 unsigned int len
= min_t(unsigned int, PAGE_SIZE
- base
, remainder
);
335 int flags
= XS_SENDMSG_FLAGS
;
338 if (remainder
!= 0 || more
)
340 err
= sock
->ops
->sendpage(sock
, *ppage
, base
, len
, flags
);
341 if (remainder
== 0 || err
!= len
)
355 * xs_sendpages - write pages directly to a socket
356 * @sock: socket to send on
357 * @addr: UDP only -- address of destination
358 * @addrlen: UDP only -- length of destination address
359 * @xdr: buffer containing this request
360 * @base: starting position in the buffer
363 static int xs_sendpages(struct socket
*sock
, struct sockaddr
*addr
, int addrlen
, struct xdr_buf
*xdr
, unsigned int base
)
365 unsigned int remainder
= xdr
->len
- base
;
371 clear_bit(SOCK_ASYNC_NOSPACE
, &sock
->flags
);
377 if (base
< xdr
->head
[0].iov_len
|| addr
!= NULL
) {
378 unsigned int len
= xdr
->head
[0].iov_len
- base
;
380 err
= xs_send_kvec(sock
, addr
, addrlen
, &xdr
->head
[0], base
, remainder
!= 0);
381 if (remainder
== 0 || err
!= len
)
386 base
-= xdr
->head
[0].iov_len
;
388 if (base
< xdr
->page_len
) {
389 unsigned int len
= xdr
->page_len
- base
;
391 err
= xs_send_pagedata(sock
, xdr
, base
, remainder
!= 0);
392 if (remainder
== 0 || err
!= len
)
397 base
-= xdr
->page_len
;
399 if (base
>= xdr
->tail
[0].iov_len
)
401 err
= xs_send_kvec(sock
, NULL
, 0, &xdr
->tail
[0], base
, 0);
411 * xs_nospace - place task on wait queue if transmit was incomplete
412 * @task: task to put to sleep
415 static void xs_nospace(struct rpc_task
*task
)
417 struct rpc_rqst
*req
= task
->tk_rqstp
;
418 struct rpc_xprt
*xprt
= req
->rq_xprt
;
419 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
421 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
422 task
->tk_pid
, req
->rq_slen
- req
->rq_bytes_sent
,
425 if (test_bit(SOCK_ASYNC_NOSPACE
, &transport
->sock
->flags
)) {
426 /* Protect against races with write_space */
427 spin_lock_bh(&xprt
->transport_lock
);
429 /* Don't race with disconnect */
430 if (!xprt_connected(xprt
))
431 task
->tk_status
= -ENOTCONN
;
432 else if (test_bit(SOCK_NOSPACE
, &transport
->sock
->flags
))
433 xprt_wait_for_buffer_space(task
);
435 spin_unlock_bh(&xprt
->transport_lock
);
437 /* Keep holding the socket if it is blocked */
438 rpc_delay(task
, HZ
>>4);
442 * xs_udp_send_request - write an RPC request to a UDP socket
443 * @task: address of RPC task that manages the state of an RPC request
446 * 0: The request has been sent
447 * EAGAIN: The socket was blocked, please call again later to
448 * complete the request
449 * ENOTCONN: Caller needs to invoke connect logic then call again
450 * other: Some other error occured, the request was not sent
452 static int xs_udp_send_request(struct rpc_task
*task
)
454 struct rpc_rqst
*req
= task
->tk_rqstp
;
455 struct rpc_xprt
*xprt
= req
->rq_xprt
;
456 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
457 struct xdr_buf
*xdr
= &req
->rq_snd_buf
;
460 xs_pktdump("packet data:",
461 req
->rq_svec
->iov_base
,
462 req
->rq_svec
->iov_len
);
464 req
->rq_xtime
= jiffies
;
465 status
= xs_sendpages(transport
->sock
,
466 (struct sockaddr
*) &xprt
->addr
,
470 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
471 xdr
->len
- req
->rq_bytes_sent
, status
);
473 if (likely(status
>= (int) req
->rq_slen
))
476 /* Still some bytes left; set up for a retry later. */
484 /* When the server has died, an ICMP port unreachable message
485 * prompts ECONNREFUSED. */
491 dprintk("RPC: sendmsg returned unrecognized error %d\n",
499 static inline void xs_encode_tcp_record_marker(struct xdr_buf
*buf
)
501 u32 reclen
= buf
->len
- sizeof(rpc_fraghdr
);
502 rpc_fraghdr
*base
= buf
->head
[0].iov_base
;
503 *base
= htonl(RPC_LAST_STREAM_FRAGMENT
| reclen
);
507 * xs_tcp_send_request - write an RPC request to a TCP socket
508 * @task: address of RPC task that manages the state of an RPC request
511 * 0: The request has been sent
512 * EAGAIN: The socket was blocked, please call again later to
513 * complete the request
514 * ENOTCONN: Caller needs to invoke connect logic then call again
515 * other: Some other error occured, the request was not sent
517 * XXX: In the case of soft timeouts, should we eventually give up
518 * if sendmsg is not able to make progress?
520 static int xs_tcp_send_request(struct rpc_task
*task
)
522 struct rpc_rqst
*req
= task
->tk_rqstp
;
523 struct rpc_xprt
*xprt
= req
->rq_xprt
;
524 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
525 struct xdr_buf
*xdr
= &req
->rq_snd_buf
;
527 unsigned int retry
= 0;
529 xs_encode_tcp_record_marker(&req
->rq_snd_buf
);
531 xs_pktdump("packet data:",
532 req
->rq_svec
->iov_base
,
533 req
->rq_svec
->iov_len
);
535 /* Continue transmitting the packet/record. We must be careful
536 * to cope with writespace callbacks arriving _after_ we have
537 * called sendmsg(). */
539 req
->rq_xtime
= jiffies
;
540 status
= xs_sendpages(transport
->sock
,
541 NULL
, 0, xdr
, req
->rq_bytes_sent
);
543 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
544 xdr
->len
- req
->rq_bytes_sent
, status
);
546 if (unlikely(status
< 0))
549 /* If we've sent the entire packet, immediately
550 * reset the count of bytes sent. */
551 req
->rq_bytes_sent
+= status
;
552 task
->tk_bytes_sent
+= status
;
553 if (likely(req
->rq_bytes_sent
>= req
->rq_slen
)) {
554 req
->rq_bytes_sent
= 0;
559 if (retry
++ > XS_SENDMSG_RETRY
)
574 dprintk("RPC: sendmsg returned unrecognized error %d\n",
576 xprt_disconnect(xprt
);
584 * xs_tcp_release_xprt - clean up after a tcp transmission
588 * This cleans up if an error causes us to abort the transmission of a request.
589 * In this case, the socket may need to be reset in order to avoid confusing
592 static void xs_tcp_release_xprt(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
594 struct rpc_rqst
*req
;
596 if (task
!= xprt
->snd_task
)
600 req
= task
->tk_rqstp
;
601 if (req
->rq_bytes_sent
== 0)
603 if (req
->rq_bytes_sent
== req
->rq_snd_buf
.len
)
605 set_bit(XPRT_CLOSE_WAIT
, &task
->tk_xprt
->state
);
607 xprt_release_xprt(xprt
, task
);
611 * xs_close - close a socket
614 * This is used when all requests are complete; ie, no DRC state remains
615 * on the server we want to save.
617 static void xs_close(struct rpc_xprt
*xprt
)
619 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
620 struct socket
*sock
= transport
->sock
;
621 struct sock
*sk
= transport
->inet
;
624 goto clear_close_wait
;
626 dprintk("RPC: xs_close xprt %p\n", xprt
);
628 write_lock_bh(&sk
->sk_callback_lock
);
629 transport
->inet
= NULL
;
630 transport
->sock
= NULL
;
632 sk
->sk_user_data
= NULL
;
633 sk
->sk_data_ready
= transport
->old_data_ready
;
634 sk
->sk_state_change
= transport
->old_state_change
;
635 sk
->sk_write_space
= transport
->old_write_space
;
636 write_unlock_bh(&sk
->sk_callback_lock
);
642 smp_mb__before_clear_bit();
643 clear_bit(XPRT_CLOSE_WAIT
, &xprt
->state
);
644 smp_mb__after_clear_bit();
648 * xs_destroy - prepare to shutdown a transport
649 * @xprt: doomed transport
652 static void xs_destroy(struct rpc_xprt
*xprt
)
654 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
656 dprintk("RPC: xs_destroy xprt %p\n", xprt
);
658 cancel_rearming_delayed_work(&transport
->connect_worker
);
660 xprt_disconnect(xprt
);
662 xs_free_peer_addresses(xprt
);
667 static inline struct rpc_xprt
*xprt_from_sock(struct sock
*sk
)
669 return (struct rpc_xprt
*) sk
->sk_user_data
;
673 * xs_udp_data_ready - "data ready" callback for UDP sockets
674 * @sk: socket with data to read
675 * @len: how much data to read
678 static void xs_udp_data_ready(struct sock
*sk
, int len
)
680 struct rpc_task
*task
;
681 struct rpc_xprt
*xprt
;
682 struct rpc_rqst
*rovr
;
684 int err
, repsize
, copied
;
688 read_lock(&sk
->sk_callback_lock
);
689 dprintk("RPC: xs_udp_data_ready...\n");
690 if (!(xprt
= xprt_from_sock(sk
)))
693 if ((skb
= skb_recv_datagram(sk
, 0, 1, &err
)) == NULL
)
699 repsize
= skb
->len
- sizeof(struct udphdr
);
701 dprintk("RPC: impossible RPC reply size %d!\n", repsize
);
705 /* Copy the XID from the skb... */
706 xp
= skb_header_pointer(skb
, sizeof(struct udphdr
),
707 sizeof(_xid
), &_xid
);
711 /* Look up and lock the request corresponding to the given XID */
712 spin_lock(&xprt
->transport_lock
);
713 rovr
= xprt_lookup_rqst(xprt
, *xp
);
716 task
= rovr
->rq_task
;
718 if ((copied
= rovr
->rq_private_buf
.buflen
) > repsize
)
721 /* Suck it into the iovec, verify checksum if not done by hw. */
722 if (csum_partial_copy_to_xdr(&rovr
->rq_private_buf
, skb
))
725 /* Something worked... */
726 dst_confirm(skb
->dst
);
728 xprt_adjust_cwnd(task
, copied
);
729 xprt_update_rtt(task
);
730 xprt_complete_rqst(task
, copied
);
733 spin_unlock(&xprt
->transport_lock
);
735 skb_free_datagram(sk
, skb
);
737 read_unlock(&sk
->sk_callback_lock
);
740 static inline void xs_tcp_read_fraghdr(struct rpc_xprt
*xprt
, struct xdr_skb_reader
*desc
)
742 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
746 p
= ((char *) &transport
->tcp_fraghdr
) + transport
->tcp_offset
;
747 len
= sizeof(transport
->tcp_fraghdr
) - transport
->tcp_offset
;
748 used
= xdr_skb_read_bits(desc
, p
, len
);
749 transport
->tcp_offset
+= used
;
753 transport
->tcp_reclen
= ntohl(transport
->tcp_fraghdr
);
754 if (transport
->tcp_reclen
& RPC_LAST_STREAM_FRAGMENT
)
755 transport
->tcp_flags
|= TCP_RCV_LAST_FRAG
;
757 transport
->tcp_flags
&= ~TCP_RCV_LAST_FRAG
;
758 transport
->tcp_reclen
&= RPC_FRAGMENT_SIZE_MASK
;
760 transport
->tcp_flags
&= ~TCP_RCV_COPY_FRAGHDR
;
761 transport
->tcp_offset
= 0;
763 /* Sanity check of the record length */
764 if (unlikely(transport
->tcp_reclen
< 4)) {
765 dprintk("RPC: invalid TCP record fragment length\n");
766 xprt_disconnect(xprt
);
769 dprintk("RPC: reading TCP record fragment of length %d\n",
770 transport
->tcp_reclen
);
773 static void xs_tcp_check_fraghdr(struct sock_xprt
*transport
)
775 if (transport
->tcp_offset
== transport
->tcp_reclen
) {
776 transport
->tcp_flags
|= TCP_RCV_COPY_FRAGHDR
;
777 transport
->tcp_offset
= 0;
778 if (transport
->tcp_flags
& TCP_RCV_LAST_FRAG
) {
779 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
780 transport
->tcp_flags
|= TCP_RCV_COPY_XID
;
781 transport
->tcp_copied
= 0;
786 static inline void xs_tcp_read_xid(struct sock_xprt
*transport
, struct xdr_skb_reader
*desc
)
791 len
= sizeof(transport
->tcp_xid
) - transport
->tcp_offset
;
792 dprintk("RPC: reading XID (%Zu bytes)\n", len
);
793 p
= ((char *) &transport
->tcp_xid
) + transport
->tcp_offset
;
794 used
= xdr_skb_read_bits(desc
, p
, len
);
795 transport
->tcp_offset
+= used
;
798 transport
->tcp_flags
&= ~TCP_RCV_COPY_XID
;
799 transport
->tcp_flags
|= TCP_RCV_COPY_DATA
;
800 transport
->tcp_copied
= 4;
801 dprintk("RPC: reading reply for XID %08x\n",
802 ntohl(transport
->tcp_xid
));
803 xs_tcp_check_fraghdr(transport
);
806 static inline void xs_tcp_read_request(struct rpc_xprt
*xprt
, struct xdr_skb_reader
*desc
)
808 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
809 struct rpc_rqst
*req
;
810 struct xdr_buf
*rcvbuf
;
814 /* Find and lock the request corresponding to this xid */
815 spin_lock(&xprt
->transport_lock
);
816 req
= xprt_lookup_rqst(xprt
, transport
->tcp_xid
);
818 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
819 dprintk("RPC: XID %08x request not found!\n",
820 ntohl(transport
->tcp_xid
));
821 spin_unlock(&xprt
->transport_lock
);
825 rcvbuf
= &req
->rq_private_buf
;
827 if (len
> transport
->tcp_reclen
- transport
->tcp_offset
) {
828 struct xdr_skb_reader my_desc
;
830 len
= transport
->tcp_reclen
- transport
->tcp_offset
;
831 memcpy(&my_desc
, desc
, sizeof(my_desc
));
833 r
= xdr_partial_copy_from_skb(rcvbuf
, transport
->tcp_copied
,
834 &my_desc
, xdr_skb_read_bits
);
838 r
= xdr_partial_copy_from_skb(rcvbuf
, transport
->tcp_copied
,
839 desc
, xdr_skb_read_bits
);
842 transport
->tcp_copied
+= r
;
843 transport
->tcp_offset
+= r
;
846 /* Error when copying to the receive buffer,
847 * usually because we weren't able to allocate
848 * additional buffer pages. All we can do now
849 * is turn off TCP_RCV_COPY_DATA, so the request
850 * will not receive any additional updates,
852 * Any remaining data from this record will
855 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
856 dprintk("RPC: XID %08x truncated request\n",
857 ntohl(transport
->tcp_xid
));
858 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
859 "tcp_offset = %u, tcp_reclen = %u\n",
860 xprt
, transport
->tcp_copied
,
861 transport
->tcp_offset
, transport
->tcp_reclen
);
865 dprintk("RPC: XID %08x read %Zd bytes\n",
866 ntohl(transport
->tcp_xid
), r
);
867 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
868 "tcp_reclen = %u\n", xprt
, transport
->tcp_copied
,
869 transport
->tcp_offset
, transport
->tcp_reclen
);
871 if (transport
->tcp_copied
== req
->rq_private_buf
.buflen
)
872 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
873 else if (transport
->tcp_offset
== transport
->tcp_reclen
) {
874 if (transport
->tcp_flags
& TCP_RCV_LAST_FRAG
)
875 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
879 if (!(transport
->tcp_flags
& TCP_RCV_COPY_DATA
))
880 xprt_complete_rqst(req
->rq_task
, transport
->tcp_copied
);
881 spin_unlock(&xprt
->transport_lock
);
882 xs_tcp_check_fraghdr(transport
);
885 static inline void xs_tcp_read_discard(struct sock_xprt
*transport
, struct xdr_skb_reader
*desc
)
889 len
= transport
->tcp_reclen
- transport
->tcp_offset
;
890 if (len
> desc
->count
)
894 transport
->tcp_offset
+= len
;
895 dprintk("RPC: discarded %Zu bytes\n", len
);
896 xs_tcp_check_fraghdr(transport
);
899 static int xs_tcp_data_recv(read_descriptor_t
*rd_desc
, struct sk_buff
*skb
, unsigned int offset
, size_t len
)
901 struct rpc_xprt
*xprt
= rd_desc
->arg
.data
;
902 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
903 struct xdr_skb_reader desc
= {
909 dprintk("RPC: xs_tcp_data_recv started\n");
911 /* Read in a new fragment marker if necessary */
912 /* Can we ever really expect to get completely empty fragments? */
913 if (transport
->tcp_flags
& TCP_RCV_COPY_FRAGHDR
) {
914 xs_tcp_read_fraghdr(xprt
, &desc
);
917 /* Read in the xid if necessary */
918 if (transport
->tcp_flags
& TCP_RCV_COPY_XID
) {
919 xs_tcp_read_xid(transport
, &desc
);
922 /* Read in the request data */
923 if (transport
->tcp_flags
& TCP_RCV_COPY_DATA
) {
924 xs_tcp_read_request(xprt
, &desc
);
927 /* Skip over any trailing bytes on short reads */
928 xs_tcp_read_discard(transport
, &desc
);
929 } while (desc
.count
);
930 dprintk("RPC: xs_tcp_data_recv done\n");
931 return len
- desc
.count
;
935 * xs_tcp_data_ready - "data ready" callback for TCP sockets
936 * @sk: socket with data to read
937 * @bytes: how much data to read
940 static void xs_tcp_data_ready(struct sock
*sk
, int bytes
)
942 struct rpc_xprt
*xprt
;
943 read_descriptor_t rd_desc
;
945 dprintk("RPC: xs_tcp_data_ready...\n");
947 read_lock(&sk
->sk_callback_lock
);
948 if (!(xprt
= xprt_from_sock(sk
)))
953 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
954 rd_desc
.arg
.data
= xprt
;
955 rd_desc
.count
= 65536;
956 tcp_read_sock(sk
, &rd_desc
, xs_tcp_data_recv
);
958 read_unlock(&sk
->sk_callback_lock
);
962 * xs_tcp_state_change - callback to handle TCP socket state changes
963 * @sk: socket whose state has changed
966 static void xs_tcp_state_change(struct sock
*sk
)
968 struct rpc_xprt
*xprt
;
970 read_lock(&sk
->sk_callback_lock
);
971 if (!(xprt
= xprt_from_sock(sk
)))
973 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt
);
974 dprintk("RPC: state %x conn %d dead %d zapped %d\n",
975 sk
->sk_state
, xprt_connected(xprt
),
976 sock_flag(sk
, SOCK_DEAD
),
977 sock_flag(sk
, SOCK_ZAPPED
));
979 switch (sk
->sk_state
) {
980 case TCP_ESTABLISHED
:
981 spin_lock_bh(&xprt
->transport_lock
);
982 if (!xprt_test_and_set_connected(xprt
)) {
983 struct sock_xprt
*transport
= container_of(xprt
,
984 struct sock_xprt
, xprt
);
986 /* Reset TCP record info */
987 transport
->tcp_offset
= 0;
988 transport
->tcp_reclen
= 0;
989 transport
->tcp_copied
= 0;
990 transport
->tcp_flags
=
991 TCP_RCV_COPY_FRAGHDR
| TCP_RCV_COPY_XID
;
993 xprt
->reestablish_timeout
= XS_TCP_INIT_REEST_TO
;
994 xprt_wake_pending_tasks(xprt
, 0);
996 spin_unlock_bh(&xprt
->transport_lock
);
1001 case TCP_CLOSE_WAIT
:
1002 /* Try to schedule an autoclose RPC calls */
1003 set_bit(XPRT_CLOSE_WAIT
, &xprt
->state
);
1004 if (test_and_set_bit(XPRT_LOCKED
, &xprt
->state
) == 0)
1005 queue_work(rpciod_workqueue
, &xprt
->task_cleanup
);
1007 xprt_disconnect(xprt
);
1010 read_unlock(&sk
->sk_callback_lock
);
1014 * xs_udp_write_space - callback invoked when socket buffer space
1016 * @sk: socket whose state has changed
1018 * Called when more output buffer space is available for this socket.
1019 * We try not to wake our writers until they can make "significant"
1020 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1021 * with a bunch of small requests.
1023 static void xs_udp_write_space(struct sock
*sk
)
1025 read_lock(&sk
->sk_callback_lock
);
1027 /* from net/core/sock.c:sock_def_write_space */
1028 if (sock_writeable(sk
)) {
1029 struct socket
*sock
;
1030 struct rpc_xprt
*xprt
;
1032 if (unlikely(!(sock
= sk
->sk_socket
)))
1034 if (unlikely(!(xprt
= xprt_from_sock(sk
))))
1036 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE
, &sock
->flags
)))
1039 xprt_write_space(xprt
);
1043 read_unlock(&sk
->sk_callback_lock
);
1047 * xs_tcp_write_space - callback invoked when socket buffer space
1049 * @sk: socket whose state has changed
1051 * Called when more output buffer space is available for this socket.
1052 * We try not to wake our writers until they can make "significant"
1053 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1054 * with a bunch of small requests.
1056 static void xs_tcp_write_space(struct sock
*sk
)
1058 read_lock(&sk
->sk_callback_lock
);
1060 /* from net/core/stream.c:sk_stream_write_space */
1061 if (sk_stream_wspace(sk
) >= sk_stream_min_wspace(sk
)) {
1062 struct socket
*sock
;
1063 struct rpc_xprt
*xprt
;
1065 if (unlikely(!(sock
= sk
->sk_socket
)))
1067 if (unlikely(!(xprt
= xprt_from_sock(sk
))))
1069 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE
, &sock
->flags
)))
1072 xprt_write_space(xprt
);
1076 read_unlock(&sk
->sk_callback_lock
);
1079 static void xs_udp_do_set_buffer_size(struct rpc_xprt
*xprt
)
1081 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1082 struct sock
*sk
= transport
->inet
;
1084 if (transport
->rcvsize
) {
1085 sk
->sk_userlocks
|= SOCK_RCVBUF_LOCK
;
1086 sk
->sk_rcvbuf
= transport
->rcvsize
* xprt
->max_reqs
* 2;
1088 if (transport
->sndsize
) {
1089 sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
;
1090 sk
->sk_sndbuf
= transport
->sndsize
* xprt
->max_reqs
* 2;
1091 sk
->sk_write_space(sk
);
1096 * xs_udp_set_buffer_size - set send and receive limits
1097 * @xprt: generic transport
1098 * @sndsize: requested size of send buffer, in bytes
1099 * @rcvsize: requested size of receive buffer, in bytes
1101 * Set socket send and receive buffer size limits.
1103 static void xs_udp_set_buffer_size(struct rpc_xprt
*xprt
, size_t sndsize
, size_t rcvsize
)
1105 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1107 transport
->sndsize
= 0;
1109 transport
->sndsize
= sndsize
+ 1024;
1110 transport
->rcvsize
= 0;
1112 transport
->rcvsize
= rcvsize
+ 1024;
1114 xs_udp_do_set_buffer_size(xprt
);
1118 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1119 * @task: task that timed out
1121 * Adjust the congestion window after a retransmit timeout has occurred.
1123 static void xs_udp_timer(struct rpc_task
*task
)
1125 xprt_adjust_cwnd(task
, -ETIMEDOUT
);
1128 static unsigned short xs_get_random_port(void)
1130 unsigned short range
= xprt_max_resvport
- xprt_min_resvport
;
1131 unsigned short rand
= (unsigned short) net_random() % range
;
1132 return rand
+ xprt_min_resvport
;
1136 * xs_set_port - reset the port number in the remote endpoint address
1137 * @xprt: generic transport
1138 * @port: new port number
1141 static void xs_set_port(struct rpc_xprt
*xprt
, unsigned short port
)
1143 struct sockaddr_in
*sap
= (struct sockaddr_in
*) &xprt
->addr
;
1145 dprintk("RPC: setting port for xprt %p to %u\n", xprt
, port
);
1147 sap
->sin_port
= htons(port
);
1150 static int xs_bind(struct sock_xprt
*transport
, struct socket
*sock
)
1152 struct sockaddr_in myaddr
= {
1153 .sin_family
= AF_INET
,
1155 struct sockaddr_in
*sa
;
1157 unsigned short port
= transport
->port
;
1159 if (!transport
->xprt
.resvport
)
1161 sa
= (struct sockaddr_in
*)&transport
->addr
;
1162 myaddr
.sin_addr
= sa
->sin_addr
;
1164 myaddr
.sin_port
= htons(port
);
1165 err
= kernel_bind(sock
, (struct sockaddr
*) &myaddr
,
1167 if (!transport
->xprt
.resvport
)
1170 transport
->port
= port
;
1173 if (port
<= xprt_min_resvport
)
1174 port
= xprt_max_resvport
;
1177 } while (err
== -EADDRINUSE
&& port
!= transport
->port
);
1178 dprintk("RPC: xs_bind "NIPQUAD_FMT
":%u: %s (%d)\n",
1179 NIPQUAD(myaddr
.sin_addr
), port
, err
? "failed" : "ok", err
);
1183 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1184 static struct lock_class_key xs_key
[2];
1185 static struct lock_class_key xs_slock_key
[2];
1187 static inline void xs_reclassify_socket(struct socket
*sock
)
1189 struct sock
*sk
= sock
->sk
;
1190 BUG_ON(sk
->sk_lock
.owner
!= NULL
);
1191 switch (sk
->sk_family
) {
1193 sock_lock_init_class_and_name(sk
, "slock-AF_INET-NFS",
1194 &xs_slock_key
[0], "sk_lock-AF_INET-NFS", &xs_key
[0]);
1198 sock_lock_init_class_and_name(sk
, "slock-AF_INET6-NFS",
1199 &xs_slock_key
[1], "sk_lock-AF_INET6-NFS", &xs_key
[1]);
1207 static inline void xs_reclassify_socket(struct socket
*sock
)
1213 * xs_udp_connect_worker - set up a UDP socket
1214 * @work: RPC transport to connect
1216 * Invoked by a work queue tasklet.
1218 static void xs_udp_connect_worker(struct work_struct
*work
)
1220 struct sock_xprt
*transport
=
1221 container_of(work
, struct sock_xprt
, connect_worker
.work
);
1222 struct rpc_xprt
*xprt
= &transport
->xprt
;
1223 struct socket
*sock
= transport
->sock
;
1224 int err
, status
= -EIO
;
1226 if (xprt
->shutdown
|| !xprt_bound(xprt
))
1229 /* Start by resetting any existing state */
1232 if ((err
= sock_create_kern(PF_INET
, SOCK_DGRAM
, IPPROTO_UDP
, &sock
)) < 0) {
1233 dprintk("RPC: can't create UDP transport socket (%d).\n", -err
);
1236 xs_reclassify_socket(sock
);
1238 if (xs_bind(transport
, sock
)) {
1243 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1244 xprt
, xprt
->address_strings
[RPC_DISPLAY_ALL
]);
1246 if (!transport
->inet
) {
1247 struct sock
*sk
= sock
->sk
;
1249 write_lock_bh(&sk
->sk_callback_lock
);
1251 sk
->sk_user_data
= xprt
;
1252 transport
->old_data_ready
= sk
->sk_data_ready
;
1253 transport
->old_state_change
= sk
->sk_state_change
;
1254 transport
->old_write_space
= sk
->sk_write_space
;
1255 sk
->sk_data_ready
= xs_udp_data_ready
;
1256 sk
->sk_write_space
= xs_udp_write_space
;
1257 sk
->sk_no_check
= UDP_CSUM_NORCV
;
1258 sk
->sk_allocation
= GFP_ATOMIC
;
1260 xprt_set_connected(xprt
);
1262 /* Reset to new socket */
1263 transport
->sock
= sock
;
1264 transport
->inet
= sk
;
1266 write_unlock_bh(&sk
->sk_callback_lock
);
1268 xs_udp_do_set_buffer_size(xprt
);
1271 xprt_wake_pending_tasks(xprt
, status
);
1272 xprt_clear_connecting(xprt
);
1276 * We need to preserve the port number so the reply cache on the server can
1277 * find our cached RPC replies when we get around to reconnecting.
1279 static void xs_tcp_reuse_connection(struct rpc_xprt
*xprt
)
1282 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1283 struct sockaddr any
;
1285 dprintk("RPC: disconnecting xprt %p to reuse port\n", xprt
);
1288 * Disconnect the transport socket by doing a connect operation
1289 * with AF_UNSPEC. This should return immediately...
1291 memset(&any
, 0, sizeof(any
));
1292 any
.sa_family
= AF_UNSPEC
;
1293 result
= kernel_connect(transport
->sock
, &any
, sizeof(any
), 0);
1295 dprintk("RPC: AF_UNSPEC connect return code %d\n",
1300 * xs_tcp_connect_worker - connect a TCP socket to a remote endpoint
1301 * @work: RPC transport to connect
1303 * Invoked by a work queue tasklet.
1305 static void xs_tcp_connect_worker(struct work_struct
*work
)
1307 struct sock_xprt
*transport
=
1308 container_of(work
, struct sock_xprt
, connect_worker
.work
);
1309 struct rpc_xprt
*xprt
= &transport
->xprt
;
1310 struct socket
*sock
= transport
->sock
;
1311 int err
, status
= -EIO
;
1313 if (xprt
->shutdown
|| !xprt_bound(xprt
))
1317 /* start from scratch */
1318 if ((err
= sock_create_kern(PF_INET
, SOCK_STREAM
, IPPROTO_TCP
, &sock
)) < 0) {
1319 dprintk("RPC: can't create TCP transport "
1320 "socket (%d).\n", -err
);
1323 xs_reclassify_socket(sock
);
1325 if (xs_bind(transport
, sock
)) {
1330 /* "close" the socket, preserving the local port */
1331 xs_tcp_reuse_connection(xprt
);
1333 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1334 xprt
, xprt
->address_strings
[RPC_DISPLAY_ALL
]);
1336 if (!transport
->inet
) {
1337 struct sock
*sk
= sock
->sk
;
1339 write_lock_bh(&sk
->sk_callback_lock
);
1341 sk
->sk_user_data
= xprt
;
1342 transport
->old_data_ready
= sk
->sk_data_ready
;
1343 transport
->old_state_change
= sk
->sk_state_change
;
1344 transport
->old_write_space
= sk
->sk_write_space
;
1345 sk
->sk_data_ready
= xs_tcp_data_ready
;
1346 sk
->sk_state_change
= xs_tcp_state_change
;
1347 sk
->sk_write_space
= xs_tcp_write_space
;
1348 sk
->sk_allocation
= GFP_ATOMIC
;
1350 /* socket options */
1351 sk
->sk_userlocks
|= SOCK_BINDPORT_LOCK
;
1352 sock_reset_flag(sk
, SOCK_LINGER
);
1353 tcp_sk(sk
)->linger2
= 0;
1354 tcp_sk(sk
)->nonagle
|= TCP_NAGLE_OFF
;
1356 xprt_clear_connected(xprt
);
1358 /* Reset to new socket */
1359 transport
->sock
= sock
;
1360 transport
->inet
= sk
;
1362 write_unlock_bh(&sk
->sk_callback_lock
);
1365 /* Tell the socket layer to start connecting... */
1366 xprt
->stat
.connect_count
++;
1367 xprt
->stat
.connect_start
= jiffies
;
1368 status
= kernel_connect(sock
, (struct sockaddr
*) &xprt
->addr
,
1369 xprt
->addrlen
, O_NONBLOCK
);
1370 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
1371 xprt
, -status
, xprt_connected(xprt
),
1372 sock
->sk
->sk_state
);
1380 /* retry with existing socket, after a delay */
1383 /* get rid of existing socket, and retry */
1389 xprt_wake_pending_tasks(xprt
, status
);
1391 xprt_clear_connecting(xprt
);
1395 * xs_connect - connect a socket to a remote endpoint
1396 * @task: address of RPC task that manages state of connect request
1398 * TCP: If the remote end dropped the connection, delay reconnecting.
1400 * UDP socket connects are synchronous, but we use a work queue anyway
1401 * to guarantee that even unprivileged user processes can set up a
1402 * socket on a privileged port.
1404 * If a UDP socket connect fails, the delay behavior here prevents
1405 * retry floods (hard mounts).
1407 static void xs_connect(struct rpc_task
*task
)
1409 struct rpc_xprt
*xprt
= task
->tk_xprt
;
1410 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1412 if (xprt_test_and_set_connecting(xprt
))
1415 if (transport
->sock
!= NULL
) {
1416 dprintk("RPC: xs_connect delayed xprt %p for %lu "
1418 xprt
, xprt
->reestablish_timeout
/ HZ
);
1419 queue_delayed_work(rpciod_workqueue
,
1420 &transport
->connect_worker
,
1421 xprt
->reestablish_timeout
);
1422 xprt
->reestablish_timeout
<<= 1;
1423 if (xprt
->reestablish_timeout
> XS_TCP_MAX_REEST_TO
)
1424 xprt
->reestablish_timeout
= XS_TCP_MAX_REEST_TO
;
1426 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt
);
1427 queue_delayed_work(rpciod_workqueue
,
1428 &transport
->connect_worker
, 0);
1433 * xs_udp_print_stats - display UDP socket-specifc stats
1434 * @xprt: rpc_xprt struct containing statistics
1438 static void xs_udp_print_stats(struct rpc_xprt
*xprt
, struct seq_file
*seq
)
1440 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1442 seq_printf(seq
, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1444 xprt
->stat
.bind_count
,
1447 xprt
->stat
.bad_xids
,
1449 xprt
->stat
.bklog_u
);
1453 * xs_tcp_print_stats - display TCP socket-specifc stats
1454 * @xprt: rpc_xprt struct containing statistics
1458 static void xs_tcp_print_stats(struct rpc_xprt
*xprt
, struct seq_file
*seq
)
1460 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1463 if (xprt_connected(xprt
))
1464 idle_time
= (long)(jiffies
- xprt
->last_used
) / HZ
;
1466 seq_printf(seq
, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1468 xprt
->stat
.bind_count
,
1469 xprt
->stat
.connect_count
,
1470 xprt
->stat
.connect_time
,
1474 xprt
->stat
.bad_xids
,
1476 xprt
->stat
.bklog_u
);
1479 static struct rpc_xprt_ops xs_udp_ops
= {
1480 .set_buffer_size
= xs_udp_set_buffer_size
,
1481 .reserve_xprt
= xprt_reserve_xprt_cong
,
1482 .release_xprt
= xprt_release_xprt_cong
,
1483 .rpcbind
= rpcb_getport_async
,
1484 .set_port
= xs_set_port
,
1485 .connect
= xs_connect
,
1486 .buf_alloc
= rpc_malloc
,
1487 .buf_free
= rpc_free
,
1488 .send_request
= xs_udp_send_request
,
1489 .set_retrans_timeout
= xprt_set_retrans_timeout_rtt
,
1490 .timer
= xs_udp_timer
,
1491 .release_request
= xprt_release_rqst_cong
,
1493 .destroy
= xs_destroy
,
1494 .print_stats
= xs_udp_print_stats
,
1497 static struct rpc_xprt_ops xs_tcp_ops
= {
1498 .reserve_xprt
= xprt_reserve_xprt
,
1499 .release_xprt
= xs_tcp_release_xprt
,
1500 .rpcbind
= rpcb_getport_async
,
1501 .set_port
= xs_set_port
,
1502 .connect
= xs_connect
,
1503 .buf_alloc
= rpc_malloc
,
1504 .buf_free
= rpc_free
,
1505 .send_request
= xs_tcp_send_request
,
1506 .set_retrans_timeout
= xprt_set_retrans_timeout_def
,
1508 .destroy
= xs_destroy
,
1509 .print_stats
= xs_tcp_print_stats
,
1512 static struct rpc_xprt
*xs_setup_xprt(struct rpc_xprtsock_create
*args
, unsigned int slot_table_size
)
1514 struct rpc_xprt
*xprt
;
1515 struct sock_xprt
*new;
1517 if (args
->addrlen
> sizeof(xprt
->addr
)) {
1518 dprintk("RPC: xs_setup_xprt: address too large\n");
1519 return ERR_PTR(-EBADF
);
1522 new = kzalloc(sizeof(*new), GFP_KERNEL
);
1524 dprintk("RPC: xs_setup_xprt: couldn't allocate "
1526 return ERR_PTR(-ENOMEM
);
1530 xprt
->max_reqs
= slot_table_size
;
1531 xprt
->slot
= kcalloc(xprt
->max_reqs
, sizeof(struct rpc_rqst
), GFP_KERNEL
);
1532 if (xprt
->slot
== NULL
) {
1534 dprintk("RPC: xs_setup_xprt: couldn't allocate slot "
1536 return ERR_PTR(-ENOMEM
);
1539 memcpy(&xprt
->addr
, args
->dstaddr
, args
->addrlen
);
1540 xprt
->addrlen
= args
->addrlen
;
1542 memcpy(&new->addr
, args
->srcaddr
, args
->addrlen
);
1543 new->port
= xs_get_random_port();
1549 * xs_setup_udp - Set up transport to use a UDP socket
1550 * @args: rpc transport creation arguments
1553 struct rpc_xprt
*xs_setup_udp(struct rpc_xprtsock_create
*args
)
1555 struct rpc_xprt
*xprt
;
1556 struct sock_xprt
*transport
;
1558 xprt
= xs_setup_xprt(args
, xprt_udp_slot_table_entries
);
1561 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1563 if (ntohs(((struct sockaddr_in
*)args
->dstaddr
)->sin_port
) != 0)
1564 xprt_set_bound(xprt
);
1566 xprt
->prot
= IPPROTO_UDP
;
1568 /* XXX: header size can vary due to auth type, IPv6, etc. */
1569 xprt
->max_payload
= (1U << 16) - (MAX_HEADER
<< 3);
1571 INIT_DELAYED_WORK(&transport
->connect_worker
, xs_udp_connect_worker
);
1572 xprt
->bind_timeout
= XS_BIND_TO
;
1573 xprt
->connect_timeout
= XS_UDP_CONN_TO
;
1574 xprt
->reestablish_timeout
= XS_UDP_REEST_TO
;
1575 xprt
->idle_timeout
= XS_IDLE_DISC_TO
;
1577 xprt
->ops
= &xs_udp_ops
;
1580 xprt
->timeout
= *args
->timeout
;
1582 xprt_set_timeout(&xprt
->timeout
, 5, 5 * HZ
);
1584 xs_format_peer_addresses(xprt
);
1585 dprintk("RPC: set up transport to address %s\n",
1586 xprt
->address_strings
[RPC_DISPLAY_ALL
]);
1592 * xs_setup_tcp - Set up transport to use a TCP socket
1593 * @args: rpc transport creation arguments
1596 struct rpc_xprt
*xs_setup_tcp(struct rpc_xprtsock_create
*args
)
1598 struct rpc_xprt
*xprt
;
1599 struct sock_xprt
*transport
;
1601 xprt
= xs_setup_xprt(args
, xprt_tcp_slot_table_entries
);
1604 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1606 if (ntohs(((struct sockaddr_in
*)args
->dstaddr
)->sin_port
) != 0)
1607 xprt_set_bound(xprt
);
1609 xprt
->prot
= IPPROTO_TCP
;
1610 xprt
->tsh_size
= sizeof(rpc_fraghdr
) / sizeof(u32
);
1611 xprt
->max_payload
= RPC_MAX_FRAGMENT_SIZE
;
1613 INIT_DELAYED_WORK(&transport
->connect_worker
, xs_tcp_connect_worker
);
1614 xprt
->bind_timeout
= XS_BIND_TO
;
1615 xprt
->connect_timeout
= XS_TCP_CONN_TO
;
1616 xprt
->reestablish_timeout
= XS_TCP_INIT_REEST_TO
;
1617 xprt
->idle_timeout
= XS_IDLE_DISC_TO
;
1619 xprt
->ops
= &xs_tcp_ops
;
1622 xprt
->timeout
= *args
->timeout
;
1624 xprt_set_timeout(&xprt
->timeout
, 2, 60 * HZ
);
1626 xs_format_peer_addresses(xprt
);
1627 dprintk("RPC: set up transport to address %s\n",
1628 xprt
->address_strings
[RPC_DISPLAY_ALL
]);
1634 * init_socket_xprt - set up xprtsock's sysctls
1637 int init_socket_xprt(void)
1640 if (!sunrpc_table_header
)
1641 sunrpc_table_header
= register_sysctl_table(sunrpc_table
);
1648 * cleanup_socket_xprt - remove xprtsock's sysctls
1651 void cleanup_socket_xprt(void)
1654 if (sunrpc_table_header
) {
1655 unregister_sysctl_table(sunrpc_table_header
);
1656 sunrpc_table_header
= NULL
;