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>
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/xprtsock.h>
36 #include <linux/file.h>
39 #include <net/checksum.h>
46 unsigned int xprt_udp_slot_table_entries
= RPC_DEF_SLOT_TABLE
;
47 unsigned int xprt_tcp_slot_table_entries
= RPC_DEF_SLOT_TABLE
;
49 unsigned int xprt_min_resvport
= RPC_DEF_MIN_RESVPORT
;
50 unsigned int xprt_max_resvport
= RPC_DEF_MAX_RESVPORT
;
53 * We can register our own files under /proc/sys/sunrpc by
54 * calling register_sysctl_table() again. The files in that
55 * directory become the union of all files registered there.
57 * We simply need to make sure that we don't collide with
58 * someone else's file names!
63 static unsigned int min_slot_table_size
= RPC_MIN_SLOT_TABLE
;
64 static unsigned int max_slot_table_size
= RPC_MAX_SLOT_TABLE
;
65 static unsigned int xprt_min_resvport_limit
= RPC_MIN_RESVPORT
;
66 static unsigned int xprt_max_resvport_limit
= RPC_MAX_RESVPORT
;
68 static struct ctl_table_header
*sunrpc_table_header
;
71 * FIXME: changing the UDP slot table size should also resize the UDP
72 * socket buffers for existing UDP transports
74 static ctl_table xs_tunables_table
[] = {
76 .ctl_name
= CTL_SLOTTABLE_UDP
,
77 .procname
= "udp_slot_table_entries",
78 .data
= &xprt_udp_slot_table_entries
,
79 .maxlen
= sizeof(unsigned int),
81 .proc_handler
= &proc_dointvec_minmax
,
82 .strategy
= &sysctl_intvec
,
83 .extra1
= &min_slot_table_size
,
84 .extra2
= &max_slot_table_size
87 .ctl_name
= CTL_SLOTTABLE_TCP
,
88 .procname
= "tcp_slot_table_entries",
89 .data
= &xprt_tcp_slot_table_entries
,
90 .maxlen
= sizeof(unsigned int),
92 .proc_handler
= &proc_dointvec_minmax
,
93 .strategy
= &sysctl_intvec
,
94 .extra1
= &min_slot_table_size
,
95 .extra2
= &max_slot_table_size
98 .ctl_name
= CTL_MIN_RESVPORT
,
99 .procname
= "min_resvport",
100 .data
= &xprt_min_resvport
,
101 .maxlen
= sizeof(unsigned int),
103 .proc_handler
= &proc_dointvec_minmax
,
104 .strategy
= &sysctl_intvec
,
105 .extra1
= &xprt_min_resvport_limit
,
106 .extra2
= &xprt_max_resvport_limit
109 .ctl_name
= CTL_MAX_RESVPORT
,
110 .procname
= "max_resvport",
111 .data
= &xprt_max_resvport
,
112 .maxlen
= sizeof(unsigned int),
114 .proc_handler
= &proc_dointvec_minmax
,
115 .strategy
= &sysctl_intvec
,
116 .extra1
= &xprt_min_resvport_limit
,
117 .extra2
= &xprt_max_resvport_limit
124 static ctl_table sunrpc_table
[] = {
126 .ctl_name
= CTL_SUNRPC
,
127 .procname
= "sunrpc",
129 .child
= xs_tunables_table
139 * How many times to try sending a request on a socket before waiting
140 * for the socket buffer to clear.
142 #define XS_SENDMSG_RETRY (10U)
145 * Time out for an RPC UDP socket connect. UDP socket connects are
146 * synchronous, but we set a timeout anyway in case of resource
147 * exhaustion on the local host.
149 #define XS_UDP_CONN_TO (5U * HZ)
152 * Wait duration for an RPC TCP connection to be established. Solaris
153 * NFS over TCP uses 60 seconds, for example, which is in line with how
154 * long a server takes to reboot.
156 #define XS_TCP_CONN_TO (60U * HZ)
159 * Wait duration for a reply from the RPC portmapper.
161 #define XS_BIND_TO (60U * HZ)
164 * Delay if a UDP socket connect error occurs. This is most likely some
165 * kind of resource problem on the local host.
167 #define XS_UDP_REEST_TO (2U * HZ)
170 * The reestablish timeout allows clients to delay for a bit before attempting
171 * to reconnect to a server that just dropped our connection.
173 * We implement an exponential backoff when trying to reestablish a TCP
174 * transport connection with the server. Some servers like to drop a TCP
175 * connection when they are overworked, so we start with a short timeout and
176 * increase over time if the server is down or not responding.
178 #define XS_TCP_INIT_REEST_TO (3U * HZ)
179 #define XS_TCP_MAX_REEST_TO (5U * 60 * HZ)
182 * TCP idle timeout; client drops the transport socket if it is idle
183 * for this long. Note that we also timeout UDP sockets to prevent
184 * holding port numbers when there is no RPC traffic.
186 #define XS_IDLE_DISC_TO (5U * 60 * HZ)
189 # undef RPC_DEBUG_DATA
190 # define RPCDBG_FACILITY RPCDBG_TRANS
193 #ifdef RPC_DEBUG_DATA
194 static void xs_pktdump(char *msg
, u32
*packet
, unsigned int count
)
196 u8
*buf
= (u8
*) packet
;
199 dprintk("RPC: %s\n", msg
);
200 for (j
= 0; j
< count
&& j
< 128; j
+= 4) {
204 dprintk("0x%04x ", j
);
206 dprintk("%02x%02x%02x%02x ",
207 buf
[j
], buf
[j
+1], buf
[j
+2], buf
[j
+3]);
212 static inline void xs_pktdump(char *msg
, u32
*packet
, unsigned int count
)
219 struct rpc_xprt xprt
;
224 struct socket
* sock
;
228 * State of TCP reply receive
236 unsigned long tcp_copied
,
240 * Connection of transports
242 struct delayed_work connect_worker
;
243 struct sockaddr_storage addr
;
247 * UDP socket buffer size parameters
253 * Saved socket callback addresses
255 void (*old_data_ready
)(struct sock
*, int);
256 void (*old_state_change
)(struct sock
*);
257 void (*old_write_space
)(struct sock
*);
261 * TCP receive state flags
263 #define TCP_RCV_LAST_FRAG (1UL << 0)
264 #define TCP_RCV_COPY_FRAGHDR (1UL << 1)
265 #define TCP_RCV_COPY_XID (1UL << 2)
266 #define TCP_RCV_COPY_DATA (1UL << 3)
268 static inline struct sockaddr
*xs_addr(struct rpc_xprt
*xprt
)
270 return (struct sockaddr
*) &xprt
->addr
;
273 static inline struct sockaddr_in
*xs_addr_in(struct rpc_xprt
*xprt
)
275 return (struct sockaddr_in
*) &xprt
->addr
;
278 static inline struct sockaddr_in6
*xs_addr_in6(struct rpc_xprt
*xprt
)
280 return (struct sockaddr_in6
*) &xprt
->addr
;
283 static void xs_format_ipv4_peer_addresses(struct rpc_xprt
*xprt
)
285 struct sockaddr_in
*addr
= xs_addr_in(xprt
);
288 buf
= kzalloc(20, GFP_KERNEL
);
290 snprintf(buf
, 20, NIPQUAD_FMT
,
291 NIPQUAD(addr
->sin_addr
.s_addr
));
293 xprt
->address_strings
[RPC_DISPLAY_ADDR
] = buf
;
295 buf
= kzalloc(8, GFP_KERNEL
);
297 snprintf(buf
, 8, "%u",
298 ntohs(addr
->sin_port
));
300 xprt
->address_strings
[RPC_DISPLAY_PORT
] = buf
;
302 buf
= kzalloc(8, GFP_KERNEL
);
304 if (xprt
->prot
== IPPROTO_UDP
)
305 snprintf(buf
, 8, "udp");
307 snprintf(buf
, 8, "tcp");
309 xprt
->address_strings
[RPC_DISPLAY_PROTO
] = buf
;
311 buf
= kzalloc(48, GFP_KERNEL
);
313 snprintf(buf
, 48, "addr="NIPQUAD_FMT
" port=%u proto=%s",
314 NIPQUAD(addr
->sin_addr
.s_addr
),
315 ntohs(addr
->sin_port
),
316 xprt
->prot
== IPPROTO_UDP
? "udp" : "tcp");
318 xprt
->address_strings
[RPC_DISPLAY_ALL
] = buf
;
320 buf
= kzalloc(10, GFP_KERNEL
);
322 snprintf(buf
, 10, "%02x%02x%02x%02x",
323 NIPQUAD(addr
->sin_addr
.s_addr
));
325 xprt
->address_strings
[RPC_DISPLAY_HEX_ADDR
] = buf
;
327 buf
= kzalloc(8, GFP_KERNEL
);
329 snprintf(buf
, 8, "%4hx",
330 ntohs(addr
->sin_port
));
332 xprt
->address_strings
[RPC_DISPLAY_HEX_PORT
] = buf
;
334 buf
= kzalloc(30, GFP_KERNEL
);
336 snprintf(buf
, 30, NIPQUAD_FMT
".%u.%u",
337 NIPQUAD(addr
->sin_addr
.s_addr
),
338 ntohs(addr
->sin_port
) >> 8,
339 ntohs(addr
->sin_port
) & 0xff);
341 xprt
->address_strings
[RPC_DISPLAY_UNIVERSAL_ADDR
] = buf
;
343 xprt
->address_strings
[RPC_DISPLAY_NETID
] =
344 kstrdup(xprt
->prot
== IPPROTO_UDP
?
345 RPCBIND_NETID_UDP
: RPCBIND_NETID_TCP
, GFP_KERNEL
);
348 static void xs_format_ipv6_peer_addresses(struct rpc_xprt
*xprt
)
350 struct sockaddr_in6
*addr
= xs_addr_in6(xprt
);
353 buf
= kzalloc(40, GFP_KERNEL
);
355 snprintf(buf
, 40, NIP6_FMT
,
356 NIP6(addr
->sin6_addr
));
358 xprt
->address_strings
[RPC_DISPLAY_ADDR
] = buf
;
360 buf
= kzalloc(8, GFP_KERNEL
);
362 snprintf(buf
, 8, "%u",
363 ntohs(addr
->sin6_port
));
365 xprt
->address_strings
[RPC_DISPLAY_PORT
] = buf
;
367 buf
= kzalloc(8, GFP_KERNEL
);
369 if (xprt
->prot
== IPPROTO_UDP
)
370 snprintf(buf
, 8, "udp");
372 snprintf(buf
, 8, "tcp");
374 xprt
->address_strings
[RPC_DISPLAY_PROTO
] = buf
;
376 buf
= kzalloc(64, GFP_KERNEL
);
378 snprintf(buf
, 64, "addr="NIP6_FMT
" port=%u proto=%s",
379 NIP6(addr
->sin6_addr
),
380 ntohs(addr
->sin6_port
),
381 xprt
->prot
== IPPROTO_UDP
? "udp" : "tcp");
383 xprt
->address_strings
[RPC_DISPLAY_ALL
] = buf
;
385 buf
= kzalloc(36, GFP_KERNEL
);
387 snprintf(buf
, 36, NIP6_SEQFMT
,
388 NIP6(addr
->sin6_addr
));
390 xprt
->address_strings
[RPC_DISPLAY_HEX_ADDR
] = buf
;
392 buf
= kzalloc(8, GFP_KERNEL
);
394 snprintf(buf
, 8, "%4hx",
395 ntohs(addr
->sin6_port
));
397 xprt
->address_strings
[RPC_DISPLAY_HEX_PORT
] = buf
;
399 buf
= kzalloc(50, GFP_KERNEL
);
401 snprintf(buf
, 50, NIP6_FMT
".%u.%u",
402 NIP6(addr
->sin6_addr
),
403 ntohs(addr
->sin6_port
) >> 8,
404 ntohs(addr
->sin6_port
) & 0xff);
406 xprt
->address_strings
[RPC_DISPLAY_UNIVERSAL_ADDR
] = buf
;
408 xprt
->address_strings
[RPC_DISPLAY_NETID
] =
409 kstrdup(xprt
->prot
== IPPROTO_UDP
?
410 RPCBIND_NETID_UDP6
: RPCBIND_NETID_TCP6
, GFP_KERNEL
);
413 static void xs_free_peer_addresses(struct rpc_xprt
*xprt
)
417 for (i
= 0; i
< RPC_DISPLAY_MAX
; i
++)
418 kfree(xprt
->address_strings
[i
]);
421 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
423 static int xs_send_kvec(struct socket
*sock
, struct sockaddr
*addr
, int addrlen
, struct kvec
*vec
, unsigned int base
, int more
)
425 struct msghdr msg
= {
427 .msg_namelen
= addrlen
,
428 .msg_flags
= XS_SENDMSG_FLAGS
| (more
? MSG_MORE
: 0),
431 .iov_base
= vec
->iov_base
+ base
,
432 .iov_len
= vec
->iov_len
- base
,
435 if (iov
.iov_len
!= 0)
436 return kernel_sendmsg(sock
, &msg
, &iov
, 1, iov
.iov_len
);
437 return kernel_sendmsg(sock
, &msg
, NULL
, 0, 0);
440 static int xs_send_pagedata(struct socket
*sock
, struct xdr_buf
*xdr
, unsigned int base
, int more
)
443 unsigned int remainder
;
446 remainder
= xdr
->page_len
- base
;
447 base
+= xdr
->page_base
;
448 ppage
= xdr
->pages
+ (base
>> PAGE_SHIFT
);
451 unsigned int len
= min_t(unsigned int, PAGE_SIZE
- base
, remainder
);
452 int flags
= XS_SENDMSG_FLAGS
;
455 if (remainder
!= 0 || more
)
457 err
= sock
->ops
->sendpage(sock
, *ppage
, base
, len
, flags
);
458 if (remainder
== 0 || err
!= len
)
472 * xs_sendpages - write pages directly to a socket
473 * @sock: socket to send on
474 * @addr: UDP only -- address of destination
475 * @addrlen: UDP only -- length of destination address
476 * @xdr: buffer containing this request
477 * @base: starting position in the buffer
480 static int xs_sendpages(struct socket
*sock
, struct sockaddr
*addr
, int addrlen
, struct xdr_buf
*xdr
, unsigned int base
)
482 unsigned int remainder
= xdr
->len
- base
;
488 clear_bit(SOCK_ASYNC_NOSPACE
, &sock
->flags
);
494 if (base
< xdr
->head
[0].iov_len
|| addr
!= NULL
) {
495 unsigned int len
= xdr
->head
[0].iov_len
- base
;
497 err
= xs_send_kvec(sock
, addr
, addrlen
, &xdr
->head
[0], base
, remainder
!= 0);
498 if (remainder
== 0 || err
!= len
)
503 base
-= xdr
->head
[0].iov_len
;
505 if (base
< xdr
->page_len
) {
506 unsigned int len
= xdr
->page_len
- base
;
508 err
= xs_send_pagedata(sock
, xdr
, base
, remainder
!= 0);
509 if (remainder
== 0 || err
!= len
)
514 base
-= xdr
->page_len
;
516 if (base
>= xdr
->tail
[0].iov_len
)
518 err
= xs_send_kvec(sock
, NULL
, 0, &xdr
->tail
[0], base
, 0);
528 * xs_nospace - place task on wait queue if transmit was incomplete
529 * @task: task to put to sleep
532 static void xs_nospace(struct rpc_task
*task
)
534 struct rpc_rqst
*req
= task
->tk_rqstp
;
535 struct rpc_xprt
*xprt
= req
->rq_xprt
;
536 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
538 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
539 task
->tk_pid
, req
->rq_slen
- req
->rq_bytes_sent
,
542 if (test_bit(SOCK_ASYNC_NOSPACE
, &transport
->sock
->flags
)) {
543 /* Protect against races with write_space */
544 spin_lock_bh(&xprt
->transport_lock
);
546 /* Don't race with disconnect */
547 if (!xprt_connected(xprt
))
548 task
->tk_status
= -ENOTCONN
;
549 else if (test_bit(SOCK_NOSPACE
, &transport
->sock
->flags
))
550 xprt_wait_for_buffer_space(task
);
552 spin_unlock_bh(&xprt
->transport_lock
);
554 /* Keep holding the socket if it is blocked */
555 rpc_delay(task
, HZ
>>4);
559 * xs_udp_send_request - write an RPC request to a UDP socket
560 * @task: address of RPC task that manages the state of an RPC request
563 * 0: The request has been sent
564 * EAGAIN: The socket was blocked, please call again later to
565 * complete the request
566 * ENOTCONN: Caller needs to invoke connect logic then call again
567 * other: Some other error occured, the request was not sent
569 static int xs_udp_send_request(struct rpc_task
*task
)
571 struct rpc_rqst
*req
= task
->tk_rqstp
;
572 struct rpc_xprt
*xprt
= req
->rq_xprt
;
573 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
574 struct xdr_buf
*xdr
= &req
->rq_snd_buf
;
577 xs_pktdump("packet data:",
578 req
->rq_svec
->iov_base
,
579 req
->rq_svec
->iov_len
);
581 req
->rq_xtime
= jiffies
;
582 status
= xs_sendpages(transport
->sock
,
587 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
588 xdr
->len
- req
->rq_bytes_sent
, status
);
591 task
->tk_bytes_sent
+= status
;
592 if (status
>= req
->rq_slen
)
594 /* Still some bytes left; set up for a retry later. */
602 /* When the server has died, an ICMP port unreachable message
603 * prompts ECONNREFUSED. */
609 dprintk("RPC: sendmsg returned unrecognized error %d\n",
617 static inline void xs_encode_tcp_record_marker(struct xdr_buf
*buf
)
619 u32 reclen
= buf
->len
- sizeof(rpc_fraghdr
);
620 rpc_fraghdr
*base
= buf
->head
[0].iov_base
;
621 *base
= htonl(RPC_LAST_STREAM_FRAGMENT
| reclen
);
625 * xs_tcp_send_request - write an RPC request to a TCP socket
626 * @task: address of RPC task that manages the state of an RPC request
629 * 0: The request has been sent
630 * EAGAIN: The socket was blocked, please call again later to
631 * complete the request
632 * ENOTCONN: Caller needs to invoke connect logic then call again
633 * other: Some other error occured, the request was not sent
635 * XXX: In the case of soft timeouts, should we eventually give up
636 * if sendmsg is not able to make progress?
638 static int xs_tcp_send_request(struct rpc_task
*task
)
640 struct rpc_rqst
*req
= task
->tk_rqstp
;
641 struct rpc_xprt
*xprt
= req
->rq_xprt
;
642 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
643 struct xdr_buf
*xdr
= &req
->rq_snd_buf
;
645 unsigned int retry
= 0;
647 xs_encode_tcp_record_marker(&req
->rq_snd_buf
);
649 xs_pktdump("packet data:",
650 req
->rq_svec
->iov_base
,
651 req
->rq_svec
->iov_len
);
653 /* Continue transmitting the packet/record. We must be careful
654 * to cope with writespace callbacks arriving _after_ we have
655 * called sendmsg(). */
657 req
->rq_xtime
= jiffies
;
658 status
= xs_sendpages(transport
->sock
,
659 NULL
, 0, xdr
, req
->rq_bytes_sent
);
661 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
662 xdr
->len
- req
->rq_bytes_sent
, status
);
664 if (unlikely(status
< 0))
667 /* If we've sent the entire packet, immediately
668 * reset the count of bytes sent. */
669 req
->rq_bytes_sent
+= status
;
670 task
->tk_bytes_sent
+= status
;
671 if (likely(req
->rq_bytes_sent
>= req
->rq_slen
)) {
672 req
->rq_bytes_sent
= 0;
677 if (retry
++ > XS_SENDMSG_RETRY
)
692 dprintk("RPC: sendmsg returned unrecognized error %d\n",
694 xprt_disconnect(xprt
);
702 * xs_tcp_release_xprt - clean up after a tcp transmission
706 * This cleans up if an error causes us to abort the transmission of a request.
707 * In this case, the socket may need to be reset in order to avoid confusing
710 static void xs_tcp_release_xprt(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
712 struct rpc_rqst
*req
;
714 if (task
!= xprt
->snd_task
)
718 req
= task
->tk_rqstp
;
719 if (req
->rq_bytes_sent
== 0)
721 if (req
->rq_bytes_sent
== req
->rq_snd_buf
.len
)
723 set_bit(XPRT_CLOSE_WAIT
, &task
->tk_xprt
->state
);
725 xprt_release_xprt(xprt
, task
);
729 * xs_close - close a socket
732 * This is used when all requests are complete; ie, no DRC state remains
733 * on the server we want to save.
735 static void xs_close(struct rpc_xprt
*xprt
)
737 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
738 struct socket
*sock
= transport
->sock
;
739 struct sock
*sk
= transport
->inet
;
742 goto clear_close_wait
;
744 dprintk("RPC: xs_close xprt %p\n", xprt
);
746 write_lock_bh(&sk
->sk_callback_lock
);
747 transport
->inet
= NULL
;
748 transport
->sock
= NULL
;
750 sk
->sk_user_data
= NULL
;
751 sk
->sk_data_ready
= transport
->old_data_ready
;
752 sk
->sk_state_change
= transport
->old_state_change
;
753 sk
->sk_write_space
= transport
->old_write_space
;
754 write_unlock_bh(&sk
->sk_callback_lock
);
760 smp_mb__before_clear_bit();
761 clear_bit(XPRT_CLOSE_WAIT
, &xprt
->state
);
762 smp_mb__after_clear_bit();
766 * xs_destroy - prepare to shutdown a transport
767 * @xprt: doomed transport
770 static void xs_destroy(struct rpc_xprt
*xprt
)
772 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
774 dprintk("RPC: xs_destroy xprt %p\n", xprt
);
776 cancel_rearming_delayed_work(&transport
->connect_worker
);
778 xprt_disconnect(xprt
);
780 xs_free_peer_addresses(xprt
);
783 module_put(THIS_MODULE
);
786 static inline struct rpc_xprt
*xprt_from_sock(struct sock
*sk
)
788 return (struct rpc_xprt
*) sk
->sk_user_data
;
792 * xs_udp_data_ready - "data ready" callback for UDP sockets
793 * @sk: socket with data to read
794 * @len: how much data to read
797 static void xs_udp_data_ready(struct sock
*sk
, int len
)
799 struct rpc_task
*task
;
800 struct rpc_xprt
*xprt
;
801 struct rpc_rqst
*rovr
;
803 int err
, repsize
, copied
;
807 read_lock(&sk
->sk_callback_lock
);
808 dprintk("RPC: xs_udp_data_ready...\n");
809 if (!(xprt
= xprt_from_sock(sk
)))
812 if ((skb
= skb_recv_datagram(sk
, 0, 1, &err
)) == NULL
)
818 repsize
= skb
->len
- sizeof(struct udphdr
);
820 dprintk("RPC: impossible RPC reply size %d!\n", repsize
);
824 /* Copy the XID from the skb... */
825 xp
= skb_header_pointer(skb
, sizeof(struct udphdr
),
826 sizeof(_xid
), &_xid
);
830 /* Look up and lock the request corresponding to the given XID */
831 spin_lock(&xprt
->transport_lock
);
832 rovr
= xprt_lookup_rqst(xprt
, *xp
);
835 task
= rovr
->rq_task
;
837 if ((copied
= rovr
->rq_private_buf
.buflen
) > repsize
)
840 /* Suck it into the iovec, verify checksum if not done by hw. */
841 if (csum_partial_copy_to_xdr(&rovr
->rq_private_buf
, skb
)) {
842 UDPX_INC_STATS_BH(sk
, UDP_MIB_INERRORS
);
846 UDPX_INC_STATS_BH(sk
, UDP_MIB_INDATAGRAMS
);
848 /* Something worked... */
849 dst_confirm(skb
->dst
);
851 xprt_adjust_cwnd(task
, copied
);
852 xprt_update_rtt(task
);
853 xprt_complete_rqst(task
, copied
);
856 spin_unlock(&xprt
->transport_lock
);
858 skb_free_datagram(sk
, skb
);
860 read_unlock(&sk
->sk_callback_lock
);
863 static inline void xs_tcp_read_fraghdr(struct rpc_xprt
*xprt
, struct xdr_skb_reader
*desc
)
865 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
869 p
= ((char *) &transport
->tcp_fraghdr
) + transport
->tcp_offset
;
870 len
= sizeof(transport
->tcp_fraghdr
) - transport
->tcp_offset
;
871 used
= xdr_skb_read_bits(desc
, p
, len
);
872 transport
->tcp_offset
+= used
;
876 transport
->tcp_reclen
= ntohl(transport
->tcp_fraghdr
);
877 if (transport
->tcp_reclen
& RPC_LAST_STREAM_FRAGMENT
)
878 transport
->tcp_flags
|= TCP_RCV_LAST_FRAG
;
880 transport
->tcp_flags
&= ~TCP_RCV_LAST_FRAG
;
881 transport
->tcp_reclen
&= RPC_FRAGMENT_SIZE_MASK
;
883 transport
->tcp_flags
&= ~TCP_RCV_COPY_FRAGHDR
;
884 transport
->tcp_offset
= 0;
886 /* Sanity check of the record length */
887 if (unlikely(transport
->tcp_reclen
< 4)) {
888 dprintk("RPC: invalid TCP record fragment length\n");
889 xprt_disconnect(xprt
);
892 dprintk("RPC: reading TCP record fragment of length %d\n",
893 transport
->tcp_reclen
);
896 static void xs_tcp_check_fraghdr(struct sock_xprt
*transport
)
898 if (transport
->tcp_offset
== transport
->tcp_reclen
) {
899 transport
->tcp_flags
|= TCP_RCV_COPY_FRAGHDR
;
900 transport
->tcp_offset
= 0;
901 if (transport
->tcp_flags
& TCP_RCV_LAST_FRAG
) {
902 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
903 transport
->tcp_flags
|= TCP_RCV_COPY_XID
;
904 transport
->tcp_copied
= 0;
909 static inline void xs_tcp_read_xid(struct sock_xprt
*transport
, struct xdr_skb_reader
*desc
)
914 len
= sizeof(transport
->tcp_xid
) - transport
->tcp_offset
;
915 dprintk("RPC: reading XID (%Zu bytes)\n", len
);
916 p
= ((char *) &transport
->tcp_xid
) + transport
->tcp_offset
;
917 used
= xdr_skb_read_bits(desc
, p
, len
);
918 transport
->tcp_offset
+= used
;
921 transport
->tcp_flags
&= ~TCP_RCV_COPY_XID
;
922 transport
->tcp_flags
|= TCP_RCV_COPY_DATA
;
923 transport
->tcp_copied
= 4;
924 dprintk("RPC: reading reply for XID %08x\n",
925 ntohl(transport
->tcp_xid
));
926 xs_tcp_check_fraghdr(transport
);
929 static inline void xs_tcp_read_request(struct rpc_xprt
*xprt
, struct xdr_skb_reader
*desc
)
931 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
932 struct rpc_rqst
*req
;
933 struct xdr_buf
*rcvbuf
;
937 /* Find and lock the request corresponding to this xid */
938 spin_lock(&xprt
->transport_lock
);
939 req
= xprt_lookup_rqst(xprt
, transport
->tcp_xid
);
941 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
942 dprintk("RPC: XID %08x request not found!\n",
943 ntohl(transport
->tcp_xid
));
944 spin_unlock(&xprt
->transport_lock
);
948 rcvbuf
= &req
->rq_private_buf
;
950 if (len
> transport
->tcp_reclen
- transport
->tcp_offset
) {
951 struct xdr_skb_reader my_desc
;
953 len
= transport
->tcp_reclen
- transport
->tcp_offset
;
954 memcpy(&my_desc
, desc
, sizeof(my_desc
));
956 r
= xdr_partial_copy_from_skb(rcvbuf
, transport
->tcp_copied
,
957 &my_desc
, xdr_skb_read_bits
);
961 r
= xdr_partial_copy_from_skb(rcvbuf
, transport
->tcp_copied
,
962 desc
, xdr_skb_read_bits
);
965 transport
->tcp_copied
+= r
;
966 transport
->tcp_offset
+= r
;
969 /* Error when copying to the receive buffer,
970 * usually because we weren't able to allocate
971 * additional buffer pages. All we can do now
972 * is turn off TCP_RCV_COPY_DATA, so the request
973 * will not receive any additional updates,
975 * Any remaining data from this record will
978 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
979 dprintk("RPC: XID %08x truncated request\n",
980 ntohl(transport
->tcp_xid
));
981 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
982 "tcp_offset = %u, tcp_reclen = %u\n",
983 xprt
, transport
->tcp_copied
,
984 transport
->tcp_offset
, transport
->tcp_reclen
);
988 dprintk("RPC: XID %08x read %Zd bytes\n",
989 ntohl(transport
->tcp_xid
), r
);
990 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
991 "tcp_reclen = %u\n", xprt
, transport
->tcp_copied
,
992 transport
->tcp_offset
, transport
->tcp_reclen
);
994 if (transport
->tcp_copied
== req
->rq_private_buf
.buflen
)
995 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
996 else if (transport
->tcp_offset
== transport
->tcp_reclen
) {
997 if (transport
->tcp_flags
& TCP_RCV_LAST_FRAG
)
998 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
1002 if (!(transport
->tcp_flags
& TCP_RCV_COPY_DATA
))
1003 xprt_complete_rqst(req
->rq_task
, transport
->tcp_copied
);
1004 spin_unlock(&xprt
->transport_lock
);
1005 xs_tcp_check_fraghdr(transport
);
1008 static inline void xs_tcp_read_discard(struct sock_xprt
*transport
, struct xdr_skb_reader
*desc
)
1012 len
= transport
->tcp_reclen
- transport
->tcp_offset
;
1013 if (len
> desc
->count
)
1016 desc
->offset
+= len
;
1017 transport
->tcp_offset
+= len
;
1018 dprintk("RPC: discarded %Zu bytes\n", len
);
1019 xs_tcp_check_fraghdr(transport
);
1022 static int xs_tcp_data_recv(read_descriptor_t
*rd_desc
, struct sk_buff
*skb
, unsigned int offset
, size_t len
)
1024 struct rpc_xprt
*xprt
= rd_desc
->arg
.data
;
1025 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1026 struct xdr_skb_reader desc
= {
1032 dprintk("RPC: xs_tcp_data_recv started\n");
1034 /* Read in a new fragment marker if necessary */
1035 /* Can we ever really expect to get completely empty fragments? */
1036 if (transport
->tcp_flags
& TCP_RCV_COPY_FRAGHDR
) {
1037 xs_tcp_read_fraghdr(xprt
, &desc
);
1040 /* Read in the xid if necessary */
1041 if (transport
->tcp_flags
& TCP_RCV_COPY_XID
) {
1042 xs_tcp_read_xid(transport
, &desc
);
1045 /* Read in the request data */
1046 if (transport
->tcp_flags
& TCP_RCV_COPY_DATA
) {
1047 xs_tcp_read_request(xprt
, &desc
);
1050 /* Skip over any trailing bytes on short reads */
1051 xs_tcp_read_discard(transport
, &desc
);
1052 } while (desc
.count
);
1053 dprintk("RPC: xs_tcp_data_recv done\n");
1054 return len
- desc
.count
;
1058 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1059 * @sk: socket with data to read
1060 * @bytes: how much data to read
1063 static void xs_tcp_data_ready(struct sock
*sk
, int bytes
)
1065 struct rpc_xprt
*xprt
;
1066 read_descriptor_t rd_desc
;
1068 dprintk("RPC: xs_tcp_data_ready...\n");
1070 read_lock(&sk
->sk_callback_lock
);
1071 if (!(xprt
= xprt_from_sock(sk
)))
1076 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1077 rd_desc
.arg
.data
= xprt
;
1078 rd_desc
.count
= 65536;
1079 tcp_read_sock(sk
, &rd_desc
, xs_tcp_data_recv
);
1081 read_unlock(&sk
->sk_callback_lock
);
1085 * xs_tcp_state_change - callback to handle TCP socket state changes
1086 * @sk: socket whose state has changed
1089 static void xs_tcp_state_change(struct sock
*sk
)
1091 struct rpc_xprt
*xprt
;
1093 read_lock(&sk
->sk_callback_lock
);
1094 if (!(xprt
= xprt_from_sock(sk
)))
1096 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt
);
1097 dprintk("RPC: state %x conn %d dead %d zapped %d\n",
1098 sk
->sk_state
, xprt_connected(xprt
),
1099 sock_flag(sk
, SOCK_DEAD
),
1100 sock_flag(sk
, SOCK_ZAPPED
));
1102 switch (sk
->sk_state
) {
1103 case TCP_ESTABLISHED
:
1104 spin_lock_bh(&xprt
->transport_lock
);
1105 if (!xprt_test_and_set_connected(xprt
)) {
1106 struct sock_xprt
*transport
= container_of(xprt
,
1107 struct sock_xprt
, xprt
);
1109 /* Reset TCP record info */
1110 transport
->tcp_offset
= 0;
1111 transport
->tcp_reclen
= 0;
1112 transport
->tcp_copied
= 0;
1113 transport
->tcp_flags
=
1114 TCP_RCV_COPY_FRAGHDR
| TCP_RCV_COPY_XID
;
1116 xprt
->reestablish_timeout
= XS_TCP_INIT_REEST_TO
;
1117 xprt_wake_pending_tasks(xprt
, 0);
1119 spin_unlock_bh(&xprt
->transport_lock
);
1124 case TCP_CLOSE_WAIT
:
1125 /* Try to schedule an autoclose RPC calls */
1126 set_bit(XPRT_CLOSE_WAIT
, &xprt
->state
);
1127 if (test_and_set_bit(XPRT_LOCKED
, &xprt
->state
) == 0)
1128 queue_work(rpciod_workqueue
, &xprt
->task_cleanup
);
1130 xprt_disconnect(xprt
);
1133 read_unlock(&sk
->sk_callback_lock
);
1137 * xs_udp_write_space - callback invoked when socket buffer space
1139 * @sk: socket whose state has changed
1141 * Called when more output buffer space is available for this socket.
1142 * We try not to wake our writers until they can make "significant"
1143 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1144 * with a bunch of small requests.
1146 static void xs_udp_write_space(struct sock
*sk
)
1148 read_lock(&sk
->sk_callback_lock
);
1150 /* from net/core/sock.c:sock_def_write_space */
1151 if (sock_writeable(sk
)) {
1152 struct socket
*sock
;
1153 struct rpc_xprt
*xprt
;
1155 if (unlikely(!(sock
= sk
->sk_socket
)))
1157 if (unlikely(!(xprt
= xprt_from_sock(sk
))))
1159 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE
, &sock
->flags
)))
1162 xprt_write_space(xprt
);
1166 read_unlock(&sk
->sk_callback_lock
);
1170 * xs_tcp_write_space - callback invoked when socket buffer space
1172 * @sk: socket whose state has changed
1174 * Called when more output buffer space is available for this socket.
1175 * We try not to wake our writers until they can make "significant"
1176 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1177 * with a bunch of small requests.
1179 static void xs_tcp_write_space(struct sock
*sk
)
1181 read_lock(&sk
->sk_callback_lock
);
1183 /* from net/core/stream.c:sk_stream_write_space */
1184 if (sk_stream_wspace(sk
) >= sk_stream_min_wspace(sk
)) {
1185 struct socket
*sock
;
1186 struct rpc_xprt
*xprt
;
1188 if (unlikely(!(sock
= sk
->sk_socket
)))
1190 if (unlikely(!(xprt
= xprt_from_sock(sk
))))
1192 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE
, &sock
->flags
)))
1195 xprt_write_space(xprt
);
1199 read_unlock(&sk
->sk_callback_lock
);
1202 static void xs_udp_do_set_buffer_size(struct rpc_xprt
*xprt
)
1204 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1205 struct sock
*sk
= transport
->inet
;
1207 if (transport
->rcvsize
) {
1208 sk
->sk_userlocks
|= SOCK_RCVBUF_LOCK
;
1209 sk
->sk_rcvbuf
= transport
->rcvsize
* xprt
->max_reqs
* 2;
1211 if (transport
->sndsize
) {
1212 sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
;
1213 sk
->sk_sndbuf
= transport
->sndsize
* xprt
->max_reqs
* 2;
1214 sk
->sk_write_space(sk
);
1219 * xs_udp_set_buffer_size - set send and receive limits
1220 * @xprt: generic transport
1221 * @sndsize: requested size of send buffer, in bytes
1222 * @rcvsize: requested size of receive buffer, in bytes
1224 * Set socket send and receive buffer size limits.
1226 static void xs_udp_set_buffer_size(struct rpc_xprt
*xprt
, size_t sndsize
, size_t rcvsize
)
1228 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1230 transport
->sndsize
= 0;
1232 transport
->sndsize
= sndsize
+ 1024;
1233 transport
->rcvsize
= 0;
1235 transport
->rcvsize
= rcvsize
+ 1024;
1237 xs_udp_do_set_buffer_size(xprt
);
1241 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1242 * @task: task that timed out
1244 * Adjust the congestion window after a retransmit timeout has occurred.
1246 static void xs_udp_timer(struct rpc_task
*task
)
1248 xprt_adjust_cwnd(task
, -ETIMEDOUT
);
1251 static unsigned short xs_get_random_port(void)
1253 unsigned short range
= xprt_max_resvport
- xprt_min_resvport
;
1254 unsigned short rand
= (unsigned short) net_random() % range
;
1255 return rand
+ xprt_min_resvport
;
1259 * xs_set_port - reset the port number in the remote endpoint address
1260 * @xprt: generic transport
1261 * @port: new port number
1264 static void xs_set_port(struct rpc_xprt
*xprt
, unsigned short port
)
1266 struct sockaddr
*addr
= xs_addr(xprt
);
1268 dprintk("RPC: setting port for xprt %p to %u\n", xprt
, port
);
1270 switch (addr
->sa_family
) {
1272 ((struct sockaddr_in
*)addr
)->sin_port
= htons(port
);
1275 ((struct sockaddr_in6
*)addr
)->sin6_port
= htons(port
);
1282 static int xs_bind4(struct sock_xprt
*transport
, struct socket
*sock
)
1284 struct sockaddr_in myaddr
= {
1285 .sin_family
= AF_INET
,
1287 struct sockaddr_in
*sa
;
1289 unsigned short port
= transport
->port
;
1291 if (!transport
->xprt
.resvport
)
1293 sa
= (struct sockaddr_in
*)&transport
->addr
;
1294 myaddr
.sin_addr
= sa
->sin_addr
;
1296 myaddr
.sin_port
= htons(port
);
1297 err
= kernel_bind(sock
, (struct sockaddr
*) &myaddr
,
1299 if (!transport
->xprt
.resvport
)
1302 transport
->port
= port
;
1305 if (port
<= xprt_min_resvport
)
1306 port
= xprt_max_resvport
;
1309 } while (err
== -EADDRINUSE
&& port
!= transport
->port
);
1310 dprintk("RPC: %s "NIPQUAD_FMT
":%u: %s (%d)\n",
1311 __FUNCTION__
, NIPQUAD(myaddr
.sin_addr
),
1312 port
, err
? "failed" : "ok", err
);
1316 static int xs_bind6(struct sock_xprt
*transport
, struct socket
*sock
)
1318 struct sockaddr_in6 myaddr
= {
1319 .sin6_family
= AF_INET6
,
1321 struct sockaddr_in6
*sa
;
1323 unsigned short port
= transport
->port
;
1325 if (!transport
->xprt
.resvport
)
1327 sa
= (struct sockaddr_in6
*)&transport
->addr
;
1328 myaddr
.sin6_addr
= sa
->sin6_addr
;
1330 myaddr
.sin6_port
= htons(port
);
1331 err
= kernel_bind(sock
, (struct sockaddr
*) &myaddr
,
1333 if (!transport
->xprt
.resvport
)
1336 transport
->port
= port
;
1339 if (port
<= xprt_min_resvport
)
1340 port
= xprt_max_resvport
;
1343 } while (err
== -EADDRINUSE
&& port
!= transport
->port
);
1344 dprintk("RPC: xs_bind6 "NIP6_FMT
":%u: %s (%d)\n",
1345 NIP6(myaddr
.sin6_addr
), port
, err
? "failed" : "ok", err
);
1349 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1350 static struct lock_class_key xs_key
[2];
1351 static struct lock_class_key xs_slock_key
[2];
1353 static inline void xs_reclassify_socket4(struct socket
*sock
)
1355 struct sock
*sk
= sock
->sk
;
1357 BUG_ON(sock_owned_by_user(sk
));
1358 sock_lock_init_class_and_name(sk
, "slock-AF_INET-RPC",
1359 &xs_slock_key
[0], "sk_lock-AF_INET-RPC", &xs_key
[0]);
1362 static inline void xs_reclassify_socket6(struct socket
*sock
)
1364 struct sock
*sk
= sock
->sk
;
1366 BUG_ON(sock_owned_by_user(sk
));
1367 sock_lock_init_class_and_name(sk
, "slock-AF_INET6-RPC",
1368 &xs_slock_key
[1], "sk_lock-AF_INET6-RPC", &xs_key
[1]);
1371 static inline void xs_reclassify_socket4(struct socket
*sock
)
1375 static inline void xs_reclassify_socket6(struct socket
*sock
)
1380 static void xs_udp_finish_connecting(struct rpc_xprt
*xprt
, struct socket
*sock
)
1382 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1384 if (!transport
->inet
) {
1385 struct sock
*sk
= sock
->sk
;
1387 write_lock_bh(&sk
->sk_callback_lock
);
1389 sk
->sk_user_data
= xprt
;
1390 transport
->old_data_ready
= sk
->sk_data_ready
;
1391 transport
->old_state_change
= sk
->sk_state_change
;
1392 transport
->old_write_space
= sk
->sk_write_space
;
1393 sk
->sk_data_ready
= xs_udp_data_ready
;
1394 sk
->sk_write_space
= xs_udp_write_space
;
1395 sk
->sk_no_check
= UDP_CSUM_NORCV
;
1396 sk
->sk_allocation
= GFP_ATOMIC
;
1398 xprt_set_connected(xprt
);
1400 /* Reset to new socket */
1401 transport
->sock
= sock
;
1402 transport
->inet
= sk
;
1404 write_unlock_bh(&sk
->sk_callback_lock
);
1406 xs_udp_do_set_buffer_size(xprt
);
1410 * xs_udp_connect_worker4 - set up a UDP socket
1411 * @work: RPC transport to connect
1413 * Invoked by a work queue tasklet.
1415 static void xs_udp_connect_worker4(struct work_struct
*work
)
1417 struct sock_xprt
*transport
=
1418 container_of(work
, struct sock_xprt
, connect_worker
.work
);
1419 struct rpc_xprt
*xprt
= &transport
->xprt
;
1420 struct socket
*sock
= transport
->sock
;
1421 int err
, status
= -EIO
;
1423 if (xprt
->shutdown
|| !xprt_bound(xprt
))
1426 /* Start by resetting any existing state */
1429 if ((err
= sock_create_kern(PF_INET
, SOCK_DGRAM
, IPPROTO_UDP
, &sock
)) < 0) {
1430 dprintk("RPC: can't create UDP transport socket (%d).\n", -err
);
1433 xs_reclassify_socket4(sock
);
1435 if (xs_bind4(transport
, sock
)) {
1440 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1441 xprt
, xprt
->address_strings
[RPC_DISPLAY_ALL
]);
1443 xs_udp_finish_connecting(xprt
, sock
);
1446 xprt_wake_pending_tasks(xprt
, status
);
1447 xprt_clear_connecting(xprt
);
1451 * xs_udp_connect_worker6 - set up a UDP socket
1452 * @work: RPC transport to connect
1454 * Invoked by a work queue tasklet.
1456 static void xs_udp_connect_worker6(struct work_struct
*work
)
1458 struct sock_xprt
*transport
=
1459 container_of(work
, struct sock_xprt
, connect_worker
.work
);
1460 struct rpc_xprt
*xprt
= &transport
->xprt
;
1461 struct socket
*sock
= transport
->sock
;
1462 int err
, status
= -EIO
;
1464 if (xprt
->shutdown
|| !xprt_bound(xprt
))
1467 /* Start by resetting any existing state */
1470 if ((err
= sock_create_kern(PF_INET6
, SOCK_DGRAM
, IPPROTO_UDP
, &sock
)) < 0) {
1471 dprintk("RPC: can't create UDP transport socket (%d).\n", -err
);
1474 xs_reclassify_socket6(sock
);
1476 if (xs_bind6(transport
, sock
) < 0) {
1481 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1482 xprt
, xprt
->address_strings
[RPC_DISPLAY_ALL
]);
1484 xs_udp_finish_connecting(xprt
, sock
);
1487 xprt_wake_pending_tasks(xprt
, status
);
1488 xprt_clear_connecting(xprt
);
1492 * We need to preserve the port number so the reply cache on the server can
1493 * find our cached RPC replies when we get around to reconnecting.
1495 static void xs_tcp_reuse_connection(struct rpc_xprt
*xprt
)
1498 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1499 struct sockaddr any
;
1501 dprintk("RPC: disconnecting xprt %p to reuse port\n", xprt
);
1504 * Disconnect the transport socket by doing a connect operation
1505 * with AF_UNSPEC. This should return immediately...
1507 memset(&any
, 0, sizeof(any
));
1508 any
.sa_family
= AF_UNSPEC
;
1509 result
= kernel_connect(transport
->sock
, &any
, sizeof(any
), 0);
1511 dprintk("RPC: AF_UNSPEC connect return code %d\n",
1515 static int xs_tcp_finish_connecting(struct rpc_xprt
*xprt
, struct socket
*sock
)
1517 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1519 if (!transport
->inet
) {
1520 struct sock
*sk
= sock
->sk
;
1522 write_lock_bh(&sk
->sk_callback_lock
);
1524 sk
->sk_user_data
= xprt
;
1525 transport
->old_data_ready
= sk
->sk_data_ready
;
1526 transport
->old_state_change
= sk
->sk_state_change
;
1527 transport
->old_write_space
= sk
->sk_write_space
;
1528 sk
->sk_data_ready
= xs_tcp_data_ready
;
1529 sk
->sk_state_change
= xs_tcp_state_change
;
1530 sk
->sk_write_space
= xs_tcp_write_space
;
1531 sk
->sk_allocation
= GFP_ATOMIC
;
1533 /* socket options */
1534 sk
->sk_userlocks
|= SOCK_BINDPORT_LOCK
;
1535 sock_reset_flag(sk
, SOCK_LINGER
);
1536 tcp_sk(sk
)->linger2
= 0;
1537 tcp_sk(sk
)->nonagle
|= TCP_NAGLE_OFF
;
1539 xprt_clear_connected(xprt
);
1541 /* Reset to new socket */
1542 transport
->sock
= sock
;
1543 transport
->inet
= sk
;
1545 write_unlock_bh(&sk
->sk_callback_lock
);
1548 /* Tell the socket layer to start connecting... */
1549 xprt
->stat
.connect_count
++;
1550 xprt
->stat
.connect_start
= jiffies
;
1551 return kernel_connect(sock
, xs_addr(xprt
), xprt
->addrlen
, O_NONBLOCK
);
1555 * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1556 * @work: RPC transport to connect
1558 * Invoked by a work queue tasklet.
1560 static void xs_tcp_connect_worker4(struct work_struct
*work
)
1562 struct sock_xprt
*transport
=
1563 container_of(work
, struct sock_xprt
, connect_worker
.work
);
1564 struct rpc_xprt
*xprt
= &transport
->xprt
;
1565 struct socket
*sock
= transport
->sock
;
1566 int err
, status
= -EIO
;
1568 if (xprt
->shutdown
|| !xprt_bound(xprt
))
1572 /* start from scratch */
1573 if ((err
= sock_create_kern(PF_INET
, SOCK_STREAM
, IPPROTO_TCP
, &sock
)) < 0) {
1574 dprintk("RPC: can't create TCP transport socket (%d).\n", -err
);
1577 xs_reclassify_socket4(sock
);
1579 if (xs_bind4(transport
, sock
) < 0) {
1584 /* "close" the socket, preserving the local port */
1585 xs_tcp_reuse_connection(xprt
);
1587 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1588 xprt
, xprt
->address_strings
[RPC_DISPLAY_ALL
]);
1590 status
= xs_tcp_finish_connecting(xprt
, sock
);
1591 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
1592 xprt
, -status
, xprt_connected(xprt
),
1593 sock
->sk
->sk_state
);
1601 /* retry with existing socket, after a delay */
1604 /* get rid of existing socket, and retry */
1610 xprt_wake_pending_tasks(xprt
, status
);
1612 xprt_clear_connecting(xprt
);
1616 * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
1617 * @work: RPC transport to connect
1619 * Invoked by a work queue tasklet.
1621 static void xs_tcp_connect_worker6(struct work_struct
*work
)
1623 struct sock_xprt
*transport
=
1624 container_of(work
, struct sock_xprt
, connect_worker
.work
);
1625 struct rpc_xprt
*xprt
= &transport
->xprt
;
1626 struct socket
*sock
= transport
->sock
;
1627 int err
, status
= -EIO
;
1629 if (xprt
->shutdown
|| !xprt_bound(xprt
))
1633 /* start from scratch */
1634 if ((err
= sock_create_kern(PF_INET6
, SOCK_STREAM
, IPPROTO_TCP
, &sock
)) < 0) {
1635 dprintk("RPC: can't create TCP transport socket (%d).\n", -err
);
1638 xs_reclassify_socket6(sock
);
1640 if (xs_bind6(transport
, sock
) < 0) {
1645 /* "close" the socket, preserving the local port */
1646 xs_tcp_reuse_connection(xprt
);
1648 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1649 xprt
, xprt
->address_strings
[RPC_DISPLAY_ALL
]);
1651 status
= xs_tcp_finish_connecting(xprt
, sock
);
1652 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
1653 xprt
, -status
, xprt_connected(xprt
), sock
->sk
->sk_state
);
1661 /* retry with existing socket, after a delay */
1664 /* get rid of existing socket, and retry */
1670 xprt_wake_pending_tasks(xprt
, status
);
1672 xprt_clear_connecting(xprt
);
1676 * xs_connect - connect a socket to a remote endpoint
1677 * @task: address of RPC task that manages state of connect request
1679 * TCP: If the remote end dropped the connection, delay reconnecting.
1681 * UDP socket connects are synchronous, but we use a work queue anyway
1682 * to guarantee that even unprivileged user processes can set up a
1683 * socket on a privileged port.
1685 * If a UDP socket connect fails, the delay behavior here prevents
1686 * retry floods (hard mounts).
1688 static void xs_connect(struct rpc_task
*task
)
1690 struct rpc_xprt
*xprt
= task
->tk_xprt
;
1691 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1693 if (xprt_test_and_set_connecting(xprt
))
1696 if (transport
->sock
!= NULL
) {
1697 dprintk("RPC: xs_connect delayed xprt %p for %lu "
1699 xprt
, xprt
->reestablish_timeout
/ HZ
);
1700 queue_delayed_work(rpciod_workqueue
,
1701 &transport
->connect_worker
,
1702 xprt
->reestablish_timeout
);
1703 xprt
->reestablish_timeout
<<= 1;
1704 if (xprt
->reestablish_timeout
> XS_TCP_MAX_REEST_TO
)
1705 xprt
->reestablish_timeout
= XS_TCP_MAX_REEST_TO
;
1707 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt
);
1708 queue_delayed_work(rpciod_workqueue
,
1709 &transport
->connect_worker
, 0);
1714 * xs_udp_print_stats - display UDP socket-specifc stats
1715 * @xprt: rpc_xprt struct containing statistics
1719 static void xs_udp_print_stats(struct rpc_xprt
*xprt
, struct seq_file
*seq
)
1721 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1723 seq_printf(seq
, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1725 xprt
->stat
.bind_count
,
1728 xprt
->stat
.bad_xids
,
1730 xprt
->stat
.bklog_u
);
1734 * xs_tcp_print_stats - display TCP socket-specifc stats
1735 * @xprt: rpc_xprt struct containing statistics
1739 static void xs_tcp_print_stats(struct rpc_xprt
*xprt
, struct seq_file
*seq
)
1741 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1744 if (xprt_connected(xprt
))
1745 idle_time
= (long)(jiffies
- xprt
->last_used
) / HZ
;
1747 seq_printf(seq
, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1749 xprt
->stat
.bind_count
,
1750 xprt
->stat
.connect_count
,
1751 xprt
->stat
.connect_time
,
1755 xprt
->stat
.bad_xids
,
1757 xprt
->stat
.bklog_u
);
1760 static struct rpc_xprt_ops xs_udp_ops
= {
1761 .set_buffer_size
= xs_udp_set_buffer_size
,
1762 .reserve_xprt
= xprt_reserve_xprt_cong
,
1763 .release_xprt
= xprt_release_xprt_cong
,
1764 .rpcbind
= rpcb_getport_async
,
1765 .set_port
= xs_set_port
,
1766 .connect
= xs_connect
,
1767 .buf_alloc
= rpc_malloc
,
1768 .buf_free
= rpc_free
,
1769 .send_request
= xs_udp_send_request
,
1770 .set_retrans_timeout
= xprt_set_retrans_timeout_rtt
,
1771 .timer
= xs_udp_timer
,
1772 .release_request
= xprt_release_rqst_cong
,
1774 .destroy
= xs_destroy
,
1775 .print_stats
= xs_udp_print_stats
,
1778 static struct rpc_xprt_ops xs_tcp_ops
= {
1779 .reserve_xprt
= xprt_reserve_xprt
,
1780 .release_xprt
= xs_tcp_release_xprt
,
1781 .rpcbind
= rpcb_getport_async
,
1782 .set_port
= xs_set_port
,
1783 .connect
= xs_connect
,
1784 .buf_alloc
= rpc_malloc
,
1785 .buf_free
= rpc_free
,
1786 .send_request
= xs_tcp_send_request
,
1787 .set_retrans_timeout
= xprt_set_retrans_timeout_def
,
1789 .destroy
= xs_destroy
,
1790 .print_stats
= xs_tcp_print_stats
,
1793 static struct rpc_xprt
*xs_setup_xprt(struct xprt_create
*args
,
1794 unsigned int slot_table_size
)
1796 struct rpc_xprt
*xprt
;
1797 struct sock_xprt
*new;
1799 if (args
->addrlen
> sizeof(xprt
->addr
)) {
1800 dprintk("RPC: xs_setup_xprt: address too large\n");
1801 return ERR_PTR(-EBADF
);
1804 new = kzalloc(sizeof(*new), GFP_KERNEL
);
1806 dprintk("RPC: xs_setup_xprt: couldn't allocate "
1808 return ERR_PTR(-ENOMEM
);
1812 xprt
->max_reqs
= slot_table_size
;
1813 xprt
->slot
= kcalloc(xprt
->max_reqs
, sizeof(struct rpc_rqst
), GFP_KERNEL
);
1814 if (xprt
->slot
== NULL
) {
1816 dprintk("RPC: xs_setup_xprt: couldn't allocate slot "
1818 return ERR_PTR(-ENOMEM
);
1821 memcpy(&xprt
->addr
, args
->dstaddr
, args
->addrlen
);
1822 xprt
->addrlen
= args
->addrlen
;
1824 memcpy(&new->addr
, args
->srcaddr
, args
->addrlen
);
1825 new->port
= xs_get_random_port();
1831 * xs_setup_udp - Set up transport to use a UDP socket
1832 * @args: rpc transport creation arguments
1835 static struct rpc_xprt
*xs_setup_udp(struct xprt_create
*args
)
1837 struct sockaddr
*addr
= args
->dstaddr
;
1838 struct rpc_xprt
*xprt
;
1839 struct sock_xprt
*transport
;
1841 xprt
= xs_setup_xprt(args
, xprt_udp_slot_table_entries
);
1844 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1846 xprt
->prot
= IPPROTO_UDP
;
1848 /* XXX: header size can vary due to auth type, IPv6, etc. */
1849 xprt
->max_payload
= (1U << 16) - (MAX_HEADER
<< 3);
1851 xprt
->bind_timeout
= XS_BIND_TO
;
1852 xprt
->connect_timeout
= XS_UDP_CONN_TO
;
1853 xprt
->reestablish_timeout
= XS_UDP_REEST_TO
;
1854 xprt
->idle_timeout
= XS_IDLE_DISC_TO
;
1856 xprt
->ops
= &xs_udp_ops
;
1859 xprt
->timeout
= *args
->timeout
;
1861 xprt_set_timeout(&xprt
->timeout
, 5, 5 * HZ
);
1863 switch (addr
->sa_family
) {
1865 if (((struct sockaddr_in
*)addr
)->sin_port
!= htons(0))
1866 xprt_set_bound(xprt
);
1868 INIT_DELAYED_WORK(&transport
->connect_worker
,
1869 xs_udp_connect_worker4
);
1870 xs_format_ipv4_peer_addresses(xprt
);
1873 if (((struct sockaddr_in6
*)addr
)->sin6_port
!= htons(0))
1874 xprt_set_bound(xprt
);
1876 INIT_DELAYED_WORK(&transport
->connect_worker
,
1877 xs_udp_connect_worker6
);
1878 xs_format_ipv6_peer_addresses(xprt
);
1882 return ERR_PTR(-EAFNOSUPPORT
);
1885 dprintk("RPC: set up transport to address %s\n",
1886 xprt
->address_strings
[RPC_DISPLAY_ALL
]);
1888 if (try_module_get(THIS_MODULE
))
1893 return ERR_PTR(-EINVAL
);
1897 * xs_setup_tcp - Set up transport to use a TCP socket
1898 * @args: rpc transport creation arguments
1901 static struct rpc_xprt
*xs_setup_tcp(struct xprt_create
*args
)
1903 struct sockaddr
*addr
= args
->dstaddr
;
1904 struct rpc_xprt
*xprt
;
1905 struct sock_xprt
*transport
;
1907 xprt
= xs_setup_xprt(args
, xprt_tcp_slot_table_entries
);
1910 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1912 xprt
->prot
= IPPROTO_TCP
;
1913 xprt
->tsh_size
= sizeof(rpc_fraghdr
) / sizeof(u32
);
1914 xprt
->max_payload
= RPC_MAX_FRAGMENT_SIZE
;
1916 xprt
->bind_timeout
= XS_BIND_TO
;
1917 xprt
->connect_timeout
= XS_TCP_CONN_TO
;
1918 xprt
->reestablish_timeout
= XS_TCP_INIT_REEST_TO
;
1919 xprt
->idle_timeout
= XS_IDLE_DISC_TO
;
1921 xprt
->ops
= &xs_tcp_ops
;
1924 xprt
->timeout
= *args
->timeout
;
1926 xprt_set_timeout(&xprt
->timeout
, 2, 60 * HZ
);
1928 switch (addr
->sa_family
) {
1930 if (((struct sockaddr_in
*)addr
)->sin_port
!= htons(0))
1931 xprt_set_bound(xprt
);
1933 INIT_DELAYED_WORK(&transport
->connect_worker
, xs_tcp_connect_worker4
);
1934 xs_format_ipv4_peer_addresses(xprt
);
1937 if (((struct sockaddr_in6
*)addr
)->sin6_port
!= htons(0))
1938 xprt_set_bound(xprt
);
1940 INIT_DELAYED_WORK(&transport
->connect_worker
, xs_tcp_connect_worker6
);
1941 xs_format_ipv6_peer_addresses(xprt
);
1945 return ERR_PTR(-EAFNOSUPPORT
);
1948 dprintk("RPC: set up transport to address %s\n",
1949 xprt
->address_strings
[RPC_DISPLAY_ALL
]);
1951 if (try_module_get(THIS_MODULE
))
1956 return ERR_PTR(-EINVAL
);
1959 static struct xprt_class xs_udp_transport
= {
1960 .list
= LIST_HEAD_INIT(xs_udp_transport
.list
),
1962 .owner
= THIS_MODULE
,
1963 .ident
= IPPROTO_UDP
,
1964 .setup
= xs_setup_udp
,
1967 static struct xprt_class xs_tcp_transport
= {
1968 .list
= LIST_HEAD_INIT(xs_tcp_transport
.list
),
1970 .owner
= THIS_MODULE
,
1971 .ident
= IPPROTO_TCP
,
1972 .setup
= xs_setup_tcp
,
1976 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
1979 int init_socket_xprt(void)
1982 if (!sunrpc_table_header
)
1983 sunrpc_table_header
= register_sysctl_table(sunrpc_table
);
1986 xprt_register_transport(&xs_udp_transport
);
1987 xprt_register_transport(&xs_tcp_transport
);
1993 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
1996 void cleanup_socket_xprt(void)
1999 if (sunrpc_table_header
) {
2000 unregister_sysctl_table(sunrpc_table_header
);
2001 sunrpc_table_header
= NULL
;
2005 xprt_unregister_transport(&xs_udp_transport
);
2006 xprt_unregister_transport(&xs_tcp_transport
);