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/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 * Time out for an RPC UDP socket connect. UDP socket connects are
140 * synchronous, but we set a timeout anyway in case of resource
141 * exhaustion on the local host.
143 #define XS_UDP_CONN_TO (5U * HZ)
146 * Wait duration for an RPC TCP connection to be established. Solaris
147 * NFS over TCP uses 60 seconds, for example, which is in line with how
148 * long a server takes to reboot.
150 #define XS_TCP_CONN_TO (60U * HZ)
153 * Wait duration for a reply from the RPC portmapper.
155 #define XS_BIND_TO (60U * HZ)
158 * Delay if a UDP socket connect error occurs. This is most likely some
159 * kind of resource problem on the local host.
161 #define XS_UDP_REEST_TO (2U * HZ)
164 * The reestablish timeout allows clients to delay for a bit before attempting
165 * to reconnect to a server that just dropped our connection.
167 * We implement an exponential backoff when trying to reestablish a TCP
168 * transport connection with the server. Some servers like to drop a TCP
169 * connection when they are overworked, so we start with a short timeout and
170 * increase over time if the server is down or not responding.
172 #define XS_TCP_INIT_REEST_TO (3U * HZ)
173 #define XS_TCP_MAX_REEST_TO (5U * 60 * HZ)
176 * TCP idle timeout; client drops the transport socket if it is idle
177 * for this long. Note that we also timeout UDP sockets to prevent
178 * holding port numbers when there is no RPC traffic.
180 #define XS_IDLE_DISC_TO (5U * 60 * HZ)
183 # undef RPC_DEBUG_DATA
184 # define RPCDBG_FACILITY RPCDBG_TRANS
187 #ifdef RPC_DEBUG_DATA
188 static void xs_pktdump(char *msg
, u32
*packet
, unsigned int count
)
190 u8
*buf
= (u8
*) packet
;
193 dprintk("RPC: %s\n", msg
);
194 for (j
= 0; j
< count
&& j
< 128; j
+= 4) {
198 dprintk("0x%04x ", j
);
200 dprintk("%02x%02x%02x%02x ",
201 buf
[j
], buf
[j
+1], buf
[j
+2], buf
[j
+3]);
206 static inline void xs_pktdump(char *msg
, u32
*packet
, unsigned int count
)
213 struct rpc_xprt xprt
;
218 struct socket
* sock
;
222 * State of TCP reply receive
230 unsigned long tcp_copied
,
234 * Connection of transports
236 struct delayed_work connect_worker
;
237 struct sockaddr_storage addr
;
241 * UDP socket buffer size parameters
247 * Saved socket callback addresses
249 void (*old_data_ready
)(struct sock
*, int);
250 void (*old_state_change
)(struct sock
*);
251 void (*old_write_space
)(struct sock
*);
252 void (*old_error_report
)(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 inline struct sockaddr
*xs_addr(struct rpc_xprt
*xprt
)
265 return (struct sockaddr
*) &xprt
->addr
;
268 static inline struct sockaddr_in
*xs_addr_in(struct rpc_xprt
*xprt
)
270 return (struct sockaddr_in
*) &xprt
->addr
;
273 static inline struct sockaddr_in6
*xs_addr_in6(struct rpc_xprt
*xprt
)
275 return (struct sockaddr_in6
*) &xprt
->addr
;
278 static void xs_format_ipv4_peer_addresses(struct rpc_xprt
*xprt
,
279 const char *protocol
,
282 struct sockaddr_in
*addr
= xs_addr_in(xprt
);
285 buf
= kzalloc(20, GFP_KERNEL
);
287 snprintf(buf
, 20, NIPQUAD_FMT
,
288 NIPQUAD(addr
->sin_addr
.s_addr
));
290 xprt
->address_strings
[RPC_DISPLAY_ADDR
] = buf
;
292 buf
= kzalloc(8, GFP_KERNEL
);
294 snprintf(buf
, 8, "%u",
295 ntohs(addr
->sin_port
));
297 xprt
->address_strings
[RPC_DISPLAY_PORT
] = buf
;
299 xprt
->address_strings
[RPC_DISPLAY_PROTO
] = protocol
;
301 buf
= kzalloc(48, GFP_KERNEL
);
303 snprintf(buf
, 48, "addr="NIPQUAD_FMT
" port=%u proto=%s",
304 NIPQUAD(addr
->sin_addr
.s_addr
),
305 ntohs(addr
->sin_port
),
308 xprt
->address_strings
[RPC_DISPLAY_ALL
] = buf
;
310 buf
= kzalloc(10, GFP_KERNEL
);
312 snprintf(buf
, 10, "%02x%02x%02x%02x",
313 NIPQUAD(addr
->sin_addr
.s_addr
));
315 xprt
->address_strings
[RPC_DISPLAY_HEX_ADDR
] = buf
;
317 buf
= kzalloc(8, GFP_KERNEL
);
319 snprintf(buf
, 8, "%4hx",
320 ntohs(addr
->sin_port
));
322 xprt
->address_strings
[RPC_DISPLAY_HEX_PORT
] = buf
;
324 buf
= kzalloc(30, GFP_KERNEL
);
326 snprintf(buf
, 30, NIPQUAD_FMT
".%u.%u",
327 NIPQUAD(addr
->sin_addr
.s_addr
),
328 ntohs(addr
->sin_port
) >> 8,
329 ntohs(addr
->sin_port
) & 0xff);
331 xprt
->address_strings
[RPC_DISPLAY_UNIVERSAL_ADDR
] = buf
;
333 xprt
->address_strings
[RPC_DISPLAY_NETID
] = netid
;
336 static void xs_format_ipv6_peer_addresses(struct rpc_xprt
*xprt
,
337 const char *protocol
,
340 struct sockaddr_in6
*addr
= xs_addr_in6(xprt
);
343 buf
= kzalloc(40, GFP_KERNEL
);
345 snprintf(buf
, 40, NIP6_FMT
,
346 NIP6(addr
->sin6_addr
));
348 xprt
->address_strings
[RPC_DISPLAY_ADDR
] = buf
;
350 buf
= kzalloc(8, GFP_KERNEL
);
352 snprintf(buf
, 8, "%u",
353 ntohs(addr
->sin6_port
));
355 xprt
->address_strings
[RPC_DISPLAY_PORT
] = buf
;
357 xprt
->address_strings
[RPC_DISPLAY_PROTO
] = protocol
;
359 buf
= kzalloc(64, GFP_KERNEL
);
361 snprintf(buf
, 64, "addr="NIP6_FMT
" port=%u proto=%s",
362 NIP6(addr
->sin6_addr
),
363 ntohs(addr
->sin6_port
),
366 xprt
->address_strings
[RPC_DISPLAY_ALL
] = buf
;
368 buf
= kzalloc(36, GFP_KERNEL
);
370 snprintf(buf
, 36, NIP6_SEQFMT
,
371 NIP6(addr
->sin6_addr
));
373 xprt
->address_strings
[RPC_DISPLAY_HEX_ADDR
] = buf
;
375 buf
= kzalloc(8, GFP_KERNEL
);
377 snprintf(buf
, 8, "%4hx",
378 ntohs(addr
->sin6_port
));
380 xprt
->address_strings
[RPC_DISPLAY_HEX_PORT
] = buf
;
382 buf
= kzalloc(50, GFP_KERNEL
);
384 snprintf(buf
, 50, NIP6_FMT
".%u.%u",
385 NIP6(addr
->sin6_addr
),
386 ntohs(addr
->sin6_port
) >> 8,
387 ntohs(addr
->sin6_port
) & 0xff);
389 xprt
->address_strings
[RPC_DISPLAY_UNIVERSAL_ADDR
] = buf
;
391 xprt
->address_strings
[RPC_DISPLAY_NETID
] = netid
;
394 static void xs_free_peer_addresses(struct rpc_xprt
*xprt
)
398 for (i
= 0; i
< RPC_DISPLAY_MAX
; i
++)
400 case RPC_DISPLAY_PROTO
:
401 case RPC_DISPLAY_NETID
:
404 kfree(xprt
->address_strings
[i
]);
408 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
410 static int xs_send_kvec(struct socket
*sock
, struct sockaddr
*addr
, int addrlen
, struct kvec
*vec
, unsigned int base
, int more
)
412 struct msghdr msg
= {
414 .msg_namelen
= addrlen
,
415 .msg_flags
= XS_SENDMSG_FLAGS
| (more
? MSG_MORE
: 0),
418 .iov_base
= vec
->iov_base
+ base
,
419 .iov_len
= vec
->iov_len
- base
,
422 if (iov
.iov_len
!= 0)
423 return kernel_sendmsg(sock
, &msg
, &iov
, 1, iov
.iov_len
);
424 return kernel_sendmsg(sock
, &msg
, NULL
, 0, 0);
427 static int xs_send_pagedata(struct socket
*sock
, struct xdr_buf
*xdr
, unsigned int base
, int more
)
430 unsigned int remainder
;
433 remainder
= xdr
->page_len
- base
;
434 base
+= xdr
->page_base
;
435 ppage
= xdr
->pages
+ (base
>> PAGE_SHIFT
);
438 unsigned int len
= min_t(unsigned int, PAGE_SIZE
- base
, remainder
);
439 int flags
= XS_SENDMSG_FLAGS
;
442 if (remainder
!= 0 || more
)
444 err
= sock
->ops
->sendpage(sock
, *ppage
, base
, len
, flags
);
445 if (remainder
== 0 || err
!= len
)
459 * xs_sendpages - write pages directly to a socket
460 * @sock: socket to send on
461 * @addr: UDP only -- address of destination
462 * @addrlen: UDP only -- length of destination address
463 * @xdr: buffer containing this request
464 * @base: starting position in the buffer
467 static int xs_sendpages(struct socket
*sock
, struct sockaddr
*addr
, int addrlen
, struct xdr_buf
*xdr
, unsigned int base
)
469 unsigned int remainder
= xdr
->len
- base
;
475 clear_bit(SOCK_ASYNC_NOSPACE
, &sock
->flags
);
481 if (base
< xdr
->head
[0].iov_len
|| addr
!= NULL
) {
482 unsigned int len
= xdr
->head
[0].iov_len
- base
;
484 err
= xs_send_kvec(sock
, addr
, addrlen
, &xdr
->head
[0], base
, remainder
!= 0);
485 if (remainder
== 0 || err
!= len
)
490 base
-= xdr
->head
[0].iov_len
;
492 if (base
< xdr
->page_len
) {
493 unsigned int len
= xdr
->page_len
- base
;
495 err
= xs_send_pagedata(sock
, xdr
, base
, remainder
!= 0);
496 if (remainder
== 0 || err
!= len
)
501 base
-= xdr
->page_len
;
503 if (base
>= xdr
->tail
[0].iov_len
)
505 err
= xs_send_kvec(sock
, NULL
, 0, &xdr
->tail
[0], base
, 0);
514 static void xs_nospace_callback(struct rpc_task
*task
)
516 struct sock_xprt
*transport
= container_of(task
->tk_rqstp
->rq_xprt
, struct sock_xprt
, xprt
);
518 transport
->inet
->sk_write_pending
--;
519 clear_bit(SOCK_ASYNC_NOSPACE
, &transport
->sock
->flags
);
523 * xs_nospace - place task on wait queue if transmit was incomplete
524 * @task: task to put to sleep
527 static void xs_nospace(struct rpc_task
*task
)
529 struct rpc_rqst
*req
= task
->tk_rqstp
;
530 struct rpc_xprt
*xprt
= req
->rq_xprt
;
531 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
533 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
534 task
->tk_pid
, req
->rq_slen
- req
->rq_bytes_sent
,
537 /* Protect against races with write_space */
538 spin_lock_bh(&xprt
->transport_lock
);
540 /* Don't race with disconnect */
541 if (xprt_connected(xprt
)) {
542 if (test_bit(SOCK_ASYNC_NOSPACE
, &transport
->sock
->flags
)) {
544 * Notify TCP that we're limited by the application
547 set_bit(SOCK_NOSPACE
, &transport
->sock
->flags
);
548 transport
->inet
->sk_write_pending
++;
549 /* ...and wait for more buffer space */
550 xprt_wait_for_buffer_space(task
, xs_nospace_callback
);
553 clear_bit(SOCK_ASYNC_NOSPACE
, &transport
->sock
->flags
);
554 task
->tk_status
= -ENOTCONN
;
557 spin_unlock_bh(&xprt
->transport_lock
);
561 * xs_udp_send_request - write an RPC request to a UDP socket
562 * @task: address of RPC task that manages the state of an RPC request
565 * 0: The request has been sent
566 * EAGAIN: The socket was blocked, please call again later to
567 * complete the request
568 * ENOTCONN: Caller needs to invoke connect logic then call again
569 * other: Some other error occured, the request was not sent
571 static int xs_udp_send_request(struct rpc_task
*task
)
573 struct rpc_rqst
*req
= task
->tk_rqstp
;
574 struct rpc_xprt
*xprt
= req
->rq_xprt
;
575 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
576 struct xdr_buf
*xdr
= &req
->rq_snd_buf
;
579 xs_pktdump("packet data:",
580 req
->rq_svec
->iov_base
,
581 req
->rq_svec
->iov_len
);
583 status
= xs_sendpages(transport
->sock
,
588 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
589 xdr
->len
- req
->rq_bytes_sent
, status
);
592 task
->tk_bytes_sent
+= status
;
593 if (status
>= req
->rq_slen
)
595 /* Still some bytes left; set up for a retry later. */
606 /* When the server has died, an ICMP port unreachable message
607 * prompts ECONNREFUSED. */
608 clear_bit(SOCK_ASYNC_NOSPACE
, &transport
->sock
->flags
);
611 clear_bit(SOCK_ASYNC_NOSPACE
, &transport
->sock
->flags
);
612 dprintk("RPC: sendmsg returned unrecognized error %d\n",
620 * xs_tcp_shutdown - gracefully shut down a TCP socket
623 * Initiates a graceful shutdown of the TCP socket by calling the
624 * equivalent of shutdown(SHUT_WR);
626 static void xs_tcp_shutdown(struct rpc_xprt
*xprt
)
628 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
629 struct socket
*sock
= transport
->sock
;
632 kernel_sock_shutdown(sock
, SHUT_WR
);
635 static inline void xs_encode_tcp_record_marker(struct xdr_buf
*buf
)
637 u32 reclen
= buf
->len
- sizeof(rpc_fraghdr
);
638 rpc_fraghdr
*base
= buf
->head
[0].iov_base
;
639 *base
= htonl(RPC_LAST_STREAM_FRAGMENT
| reclen
);
643 * xs_tcp_send_request - write an RPC request to a TCP socket
644 * @task: address of RPC task that manages the state of an RPC request
647 * 0: The request has been sent
648 * EAGAIN: The socket was blocked, please call again later to
649 * complete the request
650 * ENOTCONN: Caller needs to invoke connect logic then call again
651 * other: Some other error occured, the request was not sent
653 * XXX: In the case of soft timeouts, should we eventually give up
654 * if sendmsg is not able to make progress?
656 static int xs_tcp_send_request(struct rpc_task
*task
)
658 struct rpc_rqst
*req
= task
->tk_rqstp
;
659 struct rpc_xprt
*xprt
= req
->rq_xprt
;
660 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
661 struct xdr_buf
*xdr
= &req
->rq_snd_buf
;
664 xs_encode_tcp_record_marker(&req
->rq_snd_buf
);
666 xs_pktdump("packet data:",
667 req
->rq_svec
->iov_base
,
668 req
->rq_svec
->iov_len
);
670 /* Continue transmitting the packet/record. We must be careful
671 * to cope with writespace callbacks arriving _after_ we have
672 * called sendmsg(). */
674 status
= xs_sendpages(transport
->sock
,
675 NULL
, 0, xdr
, req
->rq_bytes_sent
);
677 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
678 xdr
->len
- req
->rq_bytes_sent
, status
);
680 if (unlikely(status
< 0))
683 /* If we've sent the entire packet, immediately
684 * reset the count of bytes sent. */
685 req
->rq_bytes_sent
+= status
;
686 task
->tk_bytes_sent
+= status
;
687 if (likely(req
->rq_bytes_sent
>= req
->rq_slen
)) {
688 req
->rq_bytes_sent
= 0;
703 xs_tcp_shutdown(xprt
);
708 clear_bit(SOCK_ASYNC_NOSPACE
, &transport
->sock
->flags
);
711 dprintk("RPC: sendmsg returned unrecognized error %d\n",
713 clear_bit(SOCK_ASYNC_NOSPACE
, &transport
->sock
->flags
);
714 xs_tcp_shutdown(xprt
);
721 * xs_tcp_release_xprt - clean up after a tcp transmission
725 * This cleans up if an error causes us to abort the transmission of a request.
726 * In this case, the socket may need to be reset in order to avoid confusing
729 static void xs_tcp_release_xprt(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
731 struct rpc_rqst
*req
;
733 if (task
!= xprt
->snd_task
)
737 req
= task
->tk_rqstp
;
738 if (req
->rq_bytes_sent
== 0)
740 if (req
->rq_bytes_sent
== req
->rq_snd_buf
.len
)
742 set_bit(XPRT_CLOSE_WAIT
, &task
->tk_xprt
->state
);
744 xprt_release_xprt(xprt
, task
);
747 static void xs_save_old_callbacks(struct sock_xprt
*transport
, struct sock
*sk
)
749 transport
->old_data_ready
= sk
->sk_data_ready
;
750 transport
->old_state_change
= sk
->sk_state_change
;
751 transport
->old_write_space
= sk
->sk_write_space
;
752 transport
->old_error_report
= sk
->sk_error_report
;
755 static void xs_restore_old_callbacks(struct sock_xprt
*transport
, struct sock
*sk
)
757 sk
->sk_data_ready
= transport
->old_data_ready
;
758 sk
->sk_state_change
= transport
->old_state_change
;
759 sk
->sk_write_space
= transport
->old_write_space
;
760 sk
->sk_error_report
= transport
->old_error_report
;
764 * xs_close - close a socket
767 * This is used when all requests are complete; ie, no DRC state remains
768 * on the server we want to save.
770 static void xs_close(struct rpc_xprt
*xprt
)
772 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
773 struct socket
*sock
= transport
->sock
;
774 struct sock
*sk
= transport
->inet
;
777 goto clear_close_wait
;
779 dprintk("RPC: xs_close xprt %p\n", xprt
);
781 write_lock_bh(&sk
->sk_callback_lock
);
782 transport
->inet
= NULL
;
783 transport
->sock
= NULL
;
785 sk
->sk_user_data
= NULL
;
787 xs_restore_old_callbacks(transport
, sk
);
788 write_unlock_bh(&sk
->sk_callback_lock
);
794 smp_mb__before_clear_bit();
795 clear_bit(XPRT_CLOSE_WAIT
, &xprt
->state
);
796 clear_bit(XPRT_CLOSING
, &xprt
->state
);
797 smp_mb__after_clear_bit();
798 xprt_disconnect_done(xprt
);
802 * xs_destroy - prepare to shutdown a transport
803 * @xprt: doomed transport
806 static void xs_destroy(struct rpc_xprt
*xprt
)
808 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
810 dprintk("RPC: xs_destroy xprt %p\n", xprt
);
812 cancel_rearming_delayed_work(&transport
->connect_worker
);
815 xs_free_peer_addresses(xprt
);
818 module_put(THIS_MODULE
);
821 static inline struct rpc_xprt
*xprt_from_sock(struct sock
*sk
)
823 return (struct rpc_xprt
*) sk
->sk_user_data
;
827 * xs_udp_data_ready - "data ready" callback for UDP sockets
828 * @sk: socket with data to read
829 * @len: how much data to read
832 static void xs_udp_data_ready(struct sock
*sk
, int len
)
834 struct rpc_task
*task
;
835 struct rpc_xprt
*xprt
;
836 struct rpc_rqst
*rovr
;
838 int err
, repsize
, copied
;
842 read_lock(&sk
->sk_callback_lock
);
843 dprintk("RPC: xs_udp_data_ready...\n");
844 if (!(xprt
= xprt_from_sock(sk
)))
847 if ((skb
= skb_recv_datagram(sk
, 0, 1, &err
)) == NULL
)
853 repsize
= skb
->len
- sizeof(struct udphdr
);
855 dprintk("RPC: impossible RPC reply size %d!\n", repsize
);
859 /* Copy the XID from the skb... */
860 xp
= skb_header_pointer(skb
, sizeof(struct udphdr
),
861 sizeof(_xid
), &_xid
);
865 /* Look up and lock the request corresponding to the given XID */
866 spin_lock(&xprt
->transport_lock
);
867 rovr
= xprt_lookup_rqst(xprt
, *xp
);
870 task
= rovr
->rq_task
;
872 if ((copied
= rovr
->rq_private_buf
.buflen
) > repsize
)
875 /* Suck it into the iovec, verify checksum if not done by hw. */
876 if (csum_partial_copy_to_xdr(&rovr
->rq_private_buf
, skb
)) {
877 UDPX_INC_STATS_BH(sk
, UDP_MIB_INERRORS
);
881 UDPX_INC_STATS_BH(sk
, UDP_MIB_INDATAGRAMS
);
883 /* Something worked... */
884 dst_confirm(skb
->dst
);
886 xprt_adjust_cwnd(task
, copied
);
887 xprt_update_rtt(task
);
888 xprt_complete_rqst(task
, copied
);
891 spin_unlock(&xprt
->transport_lock
);
893 skb_free_datagram(sk
, skb
);
895 read_unlock(&sk
->sk_callback_lock
);
898 static inline void xs_tcp_read_fraghdr(struct rpc_xprt
*xprt
, struct xdr_skb_reader
*desc
)
900 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
904 p
= ((char *) &transport
->tcp_fraghdr
) + transport
->tcp_offset
;
905 len
= sizeof(transport
->tcp_fraghdr
) - transport
->tcp_offset
;
906 used
= xdr_skb_read_bits(desc
, p
, len
);
907 transport
->tcp_offset
+= used
;
911 transport
->tcp_reclen
= ntohl(transport
->tcp_fraghdr
);
912 if (transport
->tcp_reclen
& RPC_LAST_STREAM_FRAGMENT
)
913 transport
->tcp_flags
|= TCP_RCV_LAST_FRAG
;
915 transport
->tcp_flags
&= ~TCP_RCV_LAST_FRAG
;
916 transport
->tcp_reclen
&= RPC_FRAGMENT_SIZE_MASK
;
918 transport
->tcp_flags
&= ~TCP_RCV_COPY_FRAGHDR
;
919 transport
->tcp_offset
= 0;
921 /* Sanity check of the record length */
922 if (unlikely(transport
->tcp_reclen
< 4)) {
923 dprintk("RPC: invalid TCP record fragment length\n");
924 xprt_force_disconnect(xprt
);
927 dprintk("RPC: reading TCP record fragment of length %d\n",
928 transport
->tcp_reclen
);
931 static void xs_tcp_check_fraghdr(struct sock_xprt
*transport
)
933 if (transport
->tcp_offset
== transport
->tcp_reclen
) {
934 transport
->tcp_flags
|= TCP_RCV_COPY_FRAGHDR
;
935 transport
->tcp_offset
= 0;
936 if (transport
->tcp_flags
& TCP_RCV_LAST_FRAG
) {
937 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
938 transport
->tcp_flags
|= TCP_RCV_COPY_XID
;
939 transport
->tcp_copied
= 0;
944 static inline void xs_tcp_read_xid(struct sock_xprt
*transport
, struct xdr_skb_reader
*desc
)
949 len
= sizeof(transport
->tcp_xid
) - transport
->tcp_offset
;
950 dprintk("RPC: reading XID (%Zu bytes)\n", len
);
951 p
= ((char *) &transport
->tcp_xid
) + transport
->tcp_offset
;
952 used
= xdr_skb_read_bits(desc
, p
, len
);
953 transport
->tcp_offset
+= used
;
956 transport
->tcp_flags
&= ~TCP_RCV_COPY_XID
;
957 transport
->tcp_flags
|= TCP_RCV_COPY_DATA
;
958 transport
->tcp_copied
= 4;
959 dprintk("RPC: reading reply for XID %08x\n",
960 ntohl(transport
->tcp_xid
));
961 xs_tcp_check_fraghdr(transport
);
964 static inline void xs_tcp_read_request(struct rpc_xprt
*xprt
, struct xdr_skb_reader
*desc
)
966 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
967 struct rpc_rqst
*req
;
968 struct xdr_buf
*rcvbuf
;
972 /* Find and lock the request corresponding to this xid */
973 spin_lock(&xprt
->transport_lock
);
974 req
= xprt_lookup_rqst(xprt
, transport
->tcp_xid
);
976 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
977 dprintk("RPC: XID %08x request not found!\n",
978 ntohl(transport
->tcp_xid
));
979 spin_unlock(&xprt
->transport_lock
);
983 rcvbuf
= &req
->rq_private_buf
;
985 if (len
> transport
->tcp_reclen
- transport
->tcp_offset
) {
986 struct xdr_skb_reader my_desc
;
988 len
= transport
->tcp_reclen
- transport
->tcp_offset
;
989 memcpy(&my_desc
, desc
, sizeof(my_desc
));
991 r
= xdr_partial_copy_from_skb(rcvbuf
, transport
->tcp_copied
,
992 &my_desc
, xdr_skb_read_bits
);
996 r
= xdr_partial_copy_from_skb(rcvbuf
, transport
->tcp_copied
,
997 desc
, xdr_skb_read_bits
);
1000 transport
->tcp_copied
+= r
;
1001 transport
->tcp_offset
+= r
;
1004 /* Error when copying to the receive buffer,
1005 * usually because we weren't able to allocate
1006 * additional buffer pages. All we can do now
1007 * is turn off TCP_RCV_COPY_DATA, so the request
1008 * will not receive any additional updates,
1010 * Any remaining data from this record will
1013 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
1014 dprintk("RPC: XID %08x truncated request\n",
1015 ntohl(transport
->tcp_xid
));
1016 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1017 "tcp_offset = %u, tcp_reclen = %u\n",
1018 xprt
, transport
->tcp_copied
,
1019 transport
->tcp_offset
, transport
->tcp_reclen
);
1023 dprintk("RPC: XID %08x read %Zd bytes\n",
1024 ntohl(transport
->tcp_xid
), r
);
1025 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1026 "tcp_reclen = %u\n", xprt
, transport
->tcp_copied
,
1027 transport
->tcp_offset
, transport
->tcp_reclen
);
1029 if (transport
->tcp_copied
== req
->rq_private_buf
.buflen
)
1030 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
1031 else if (transport
->tcp_offset
== transport
->tcp_reclen
) {
1032 if (transport
->tcp_flags
& TCP_RCV_LAST_FRAG
)
1033 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
1037 if (!(transport
->tcp_flags
& TCP_RCV_COPY_DATA
))
1038 xprt_complete_rqst(req
->rq_task
, transport
->tcp_copied
);
1039 spin_unlock(&xprt
->transport_lock
);
1040 xs_tcp_check_fraghdr(transport
);
1043 static inline void xs_tcp_read_discard(struct sock_xprt
*transport
, struct xdr_skb_reader
*desc
)
1047 len
= transport
->tcp_reclen
- transport
->tcp_offset
;
1048 if (len
> desc
->count
)
1051 desc
->offset
+= len
;
1052 transport
->tcp_offset
+= len
;
1053 dprintk("RPC: discarded %Zu bytes\n", len
);
1054 xs_tcp_check_fraghdr(transport
);
1057 static int xs_tcp_data_recv(read_descriptor_t
*rd_desc
, struct sk_buff
*skb
, unsigned int offset
, size_t len
)
1059 struct rpc_xprt
*xprt
= rd_desc
->arg
.data
;
1060 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1061 struct xdr_skb_reader desc
= {
1067 dprintk("RPC: xs_tcp_data_recv started\n");
1069 /* Read in a new fragment marker if necessary */
1070 /* Can we ever really expect to get completely empty fragments? */
1071 if (transport
->tcp_flags
& TCP_RCV_COPY_FRAGHDR
) {
1072 xs_tcp_read_fraghdr(xprt
, &desc
);
1075 /* Read in the xid if necessary */
1076 if (transport
->tcp_flags
& TCP_RCV_COPY_XID
) {
1077 xs_tcp_read_xid(transport
, &desc
);
1080 /* Read in the request data */
1081 if (transport
->tcp_flags
& TCP_RCV_COPY_DATA
) {
1082 xs_tcp_read_request(xprt
, &desc
);
1085 /* Skip over any trailing bytes on short reads */
1086 xs_tcp_read_discard(transport
, &desc
);
1087 } while (desc
.count
);
1088 dprintk("RPC: xs_tcp_data_recv done\n");
1089 return len
- desc
.count
;
1093 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1094 * @sk: socket with data to read
1095 * @bytes: how much data to read
1098 static void xs_tcp_data_ready(struct sock
*sk
, int bytes
)
1100 struct rpc_xprt
*xprt
;
1101 read_descriptor_t rd_desc
;
1104 dprintk("RPC: xs_tcp_data_ready...\n");
1106 read_lock(&sk
->sk_callback_lock
);
1107 if (!(xprt
= xprt_from_sock(sk
)))
1112 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1113 rd_desc
.arg
.data
= xprt
;
1115 rd_desc
.count
= 65536;
1116 read
= tcp_read_sock(sk
, &rd_desc
, xs_tcp_data_recv
);
1119 read_unlock(&sk
->sk_callback_lock
);
1123 * xs_tcp_state_change - callback to handle TCP socket state changes
1124 * @sk: socket whose state has changed
1127 static void xs_tcp_state_change(struct sock
*sk
)
1129 struct rpc_xprt
*xprt
;
1131 read_lock(&sk
->sk_callback_lock
);
1132 if (!(xprt
= xprt_from_sock(sk
)))
1134 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt
);
1135 dprintk("RPC: state %x conn %d dead %d zapped %d\n",
1136 sk
->sk_state
, xprt_connected(xprt
),
1137 sock_flag(sk
, SOCK_DEAD
),
1138 sock_flag(sk
, SOCK_ZAPPED
));
1140 switch (sk
->sk_state
) {
1141 case TCP_ESTABLISHED
:
1142 spin_lock_bh(&xprt
->transport_lock
);
1143 if (!xprt_test_and_set_connected(xprt
)) {
1144 struct sock_xprt
*transport
= container_of(xprt
,
1145 struct sock_xprt
, xprt
);
1147 /* Reset TCP record info */
1148 transport
->tcp_offset
= 0;
1149 transport
->tcp_reclen
= 0;
1150 transport
->tcp_copied
= 0;
1151 transport
->tcp_flags
=
1152 TCP_RCV_COPY_FRAGHDR
| TCP_RCV_COPY_XID
;
1154 xprt_wake_pending_tasks(xprt
, 0);
1156 spin_unlock_bh(&xprt
->transport_lock
);
1159 /* The client initiated a shutdown of the socket */
1160 xprt
->connect_cookie
++;
1161 xprt
->reestablish_timeout
= 0;
1162 set_bit(XPRT_CLOSING
, &xprt
->state
);
1163 smp_mb__before_clear_bit();
1164 clear_bit(XPRT_CONNECTED
, &xprt
->state
);
1165 clear_bit(XPRT_CLOSE_WAIT
, &xprt
->state
);
1166 smp_mb__after_clear_bit();
1168 case TCP_CLOSE_WAIT
:
1169 /* The server initiated a shutdown of the socket */
1170 set_bit(XPRT_CLOSING
, &xprt
->state
);
1171 xprt_force_disconnect(xprt
);
1173 xprt
->connect_cookie
++;
1176 * If the server closed down the connection, make sure that
1177 * we back off before reconnecting
1179 if (xprt
->reestablish_timeout
< XS_TCP_INIT_REEST_TO
)
1180 xprt
->reestablish_timeout
= XS_TCP_INIT_REEST_TO
;
1183 smp_mb__before_clear_bit();
1184 clear_bit(XPRT_CONNECTED
, &xprt
->state
);
1185 smp_mb__after_clear_bit();
1188 smp_mb__before_clear_bit();
1189 clear_bit(XPRT_CLOSE_WAIT
, &xprt
->state
);
1190 clear_bit(XPRT_CLOSING
, &xprt
->state
);
1191 smp_mb__after_clear_bit();
1192 /* Mark transport as closed and wake up all pending tasks */
1193 xprt_disconnect_done(xprt
);
1196 read_unlock(&sk
->sk_callback_lock
);
1200 * xs_tcp_error_report - callback mainly for catching RST events
1203 static void xs_tcp_error_report(struct sock
*sk
)
1205 struct rpc_xprt
*xprt
;
1207 read_lock(&sk
->sk_callback_lock
);
1208 if (sk
->sk_err
!= ECONNRESET
|| sk
->sk_state
!= TCP_ESTABLISHED
)
1210 if (!(xprt
= xprt_from_sock(sk
)))
1212 dprintk("RPC: %s client %p...\n"
1214 __func__
, xprt
, sk
->sk_err
);
1216 xprt_force_disconnect(xprt
);
1218 read_unlock(&sk
->sk_callback_lock
);
1222 * xs_udp_write_space - callback invoked when socket buffer space
1224 * @sk: socket whose state has changed
1226 * Called when more output buffer space is available for this socket.
1227 * We try not to wake our writers until they can make "significant"
1228 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1229 * with a bunch of small requests.
1231 static void xs_udp_write_space(struct sock
*sk
)
1233 read_lock(&sk
->sk_callback_lock
);
1235 /* from net/core/sock.c:sock_def_write_space */
1236 if (sock_writeable(sk
)) {
1237 struct socket
*sock
;
1238 struct rpc_xprt
*xprt
;
1240 if (unlikely(!(sock
= sk
->sk_socket
)))
1242 clear_bit(SOCK_NOSPACE
, &sock
->flags
);
1244 if (unlikely(!(xprt
= xprt_from_sock(sk
))))
1246 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE
, &sock
->flags
) == 0)
1249 xprt_write_space(xprt
);
1253 read_unlock(&sk
->sk_callback_lock
);
1257 * xs_tcp_write_space - callback invoked when socket buffer space
1259 * @sk: socket whose state has changed
1261 * Called when more output buffer space is available for this socket.
1262 * We try not to wake our writers until they can make "significant"
1263 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1264 * with a bunch of small requests.
1266 static void xs_tcp_write_space(struct sock
*sk
)
1268 read_lock(&sk
->sk_callback_lock
);
1270 /* from net/core/stream.c:sk_stream_write_space */
1271 if (sk_stream_wspace(sk
) >= sk_stream_min_wspace(sk
)) {
1272 struct socket
*sock
;
1273 struct rpc_xprt
*xprt
;
1275 if (unlikely(!(sock
= sk
->sk_socket
)))
1277 clear_bit(SOCK_NOSPACE
, &sock
->flags
);
1279 if (unlikely(!(xprt
= xprt_from_sock(sk
))))
1281 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE
, &sock
->flags
) == 0)
1284 xprt_write_space(xprt
);
1288 read_unlock(&sk
->sk_callback_lock
);
1291 static void xs_udp_do_set_buffer_size(struct rpc_xprt
*xprt
)
1293 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1294 struct sock
*sk
= transport
->inet
;
1296 if (transport
->rcvsize
) {
1297 sk
->sk_userlocks
|= SOCK_RCVBUF_LOCK
;
1298 sk
->sk_rcvbuf
= transport
->rcvsize
* xprt
->max_reqs
* 2;
1300 if (transport
->sndsize
) {
1301 sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
;
1302 sk
->sk_sndbuf
= transport
->sndsize
* xprt
->max_reqs
* 2;
1303 sk
->sk_write_space(sk
);
1308 * xs_udp_set_buffer_size - set send and receive limits
1309 * @xprt: generic transport
1310 * @sndsize: requested size of send buffer, in bytes
1311 * @rcvsize: requested size of receive buffer, in bytes
1313 * Set socket send and receive buffer size limits.
1315 static void xs_udp_set_buffer_size(struct rpc_xprt
*xprt
, size_t sndsize
, size_t rcvsize
)
1317 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1319 transport
->sndsize
= 0;
1321 transport
->sndsize
= sndsize
+ 1024;
1322 transport
->rcvsize
= 0;
1324 transport
->rcvsize
= rcvsize
+ 1024;
1326 xs_udp_do_set_buffer_size(xprt
);
1330 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1331 * @task: task that timed out
1333 * Adjust the congestion window after a retransmit timeout has occurred.
1335 static void xs_udp_timer(struct rpc_task
*task
)
1337 xprt_adjust_cwnd(task
, -ETIMEDOUT
);
1340 static unsigned short xs_get_random_port(void)
1342 unsigned short range
= xprt_max_resvport
- xprt_min_resvport
;
1343 unsigned short rand
= (unsigned short) net_random() % range
;
1344 return rand
+ xprt_min_resvport
;
1348 * xs_set_port - reset the port number in the remote endpoint address
1349 * @xprt: generic transport
1350 * @port: new port number
1353 static void xs_set_port(struct rpc_xprt
*xprt
, unsigned short port
)
1355 struct sockaddr
*addr
= xs_addr(xprt
);
1357 dprintk("RPC: setting port for xprt %p to %u\n", xprt
, port
);
1359 switch (addr
->sa_family
) {
1361 ((struct sockaddr_in
*)addr
)->sin_port
= htons(port
);
1364 ((struct sockaddr_in6
*)addr
)->sin6_port
= htons(port
);
1371 static unsigned short xs_get_srcport(struct sock_xprt
*transport
, struct socket
*sock
)
1373 unsigned short port
= transport
->port
;
1375 if (port
== 0 && transport
->xprt
.resvport
)
1376 port
= xs_get_random_port();
1380 static unsigned short xs_next_srcport(struct sock_xprt
*transport
, struct socket
*sock
, unsigned short port
)
1382 if (transport
->port
!= 0)
1383 transport
->port
= 0;
1384 if (!transport
->xprt
.resvport
)
1386 if (port
<= xprt_min_resvport
|| port
> xprt_max_resvport
)
1387 return xprt_max_resvport
;
1391 static int xs_bind4(struct sock_xprt
*transport
, struct socket
*sock
)
1393 struct sockaddr_in myaddr
= {
1394 .sin_family
= AF_INET
,
1396 struct sockaddr_in
*sa
;
1398 unsigned short port
= xs_get_srcport(transport
, sock
);
1399 unsigned short last
;
1401 sa
= (struct sockaddr_in
*)&transport
->addr
;
1402 myaddr
.sin_addr
= sa
->sin_addr
;
1404 myaddr
.sin_port
= htons(port
);
1405 err
= kernel_bind(sock
, (struct sockaddr
*) &myaddr
,
1410 transport
->port
= port
;
1414 port
= xs_next_srcport(transport
, sock
, port
);
1417 } while (err
== -EADDRINUSE
&& nloop
!= 2);
1418 dprintk("RPC: %s "NIPQUAD_FMT
":%u: %s (%d)\n",
1419 __func__
, NIPQUAD(myaddr
.sin_addr
),
1420 port
, err
? "failed" : "ok", err
);
1424 static int xs_bind6(struct sock_xprt
*transport
, struct socket
*sock
)
1426 struct sockaddr_in6 myaddr
= {
1427 .sin6_family
= AF_INET6
,
1429 struct sockaddr_in6
*sa
;
1431 unsigned short port
= xs_get_srcport(transport
, sock
);
1432 unsigned short last
;
1434 sa
= (struct sockaddr_in6
*)&transport
->addr
;
1435 myaddr
.sin6_addr
= sa
->sin6_addr
;
1437 myaddr
.sin6_port
= htons(port
);
1438 err
= kernel_bind(sock
, (struct sockaddr
*) &myaddr
,
1443 transport
->port
= port
;
1447 port
= xs_next_srcport(transport
, sock
, port
);
1450 } while (err
== -EADDRINUSE
&& nloop
!= 2);
1451 dprintk("RPC: xs_bind6 "NIP6_FMT
":%u: %s (%d)\n",
1452 NIP6(myaddr
.sin6_addr
), port
, err
? "failed" : "ok", err
);
1456 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1457 static struct lock_class_key xs_key
[2];
1458 static struct lock_class_key xs_slock_key
[2];
1460 static inline void xs_reclassify_socket4(struct socket
*sock
)
1462 struct sock
*sk
= sock
->sk
;
1464 BUG_ON(sock_owned_by_user(sk
));
1465 sock_lock_init_class_and_name(sk
, "slock-AF_INET-RPC",
1466 &xs_slock_key
[0], "sk_lock-AF_INET-RPC", &xs_key
[0]);
1469 static inline void xs_reclassify_socket6(struct socket
*sock
)
1471 struct sock
*sk
= sock
->sk
;
1473 BUG_ON(sock_owned_by_user(sk
));
1474 sock_lock_init_class_and_name(sk
, "slock-AF_INET6-RPC",
1475 &xs_slock_key
[1], "sk_lock-AF_INET6-RPC", &xs_key
[1]);
1478 static inline void xs_reclassify_socket4(struct socket
*sock
)
1482 static inline void xs_reclassify_socket6(struct socket
*sock
)
1487 static void xs_udp_finish_connecting(struct rpc_xprt
*xprt
, struct socket
*sock
)
1489 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1491 if (!transport
->inet
) {
1492 struct sock
*sk
= sock
->sk
;
1494 write_lock_bh(&sk
->sk_callback_lock
);
1496 xs_save_old_callbacks(transport
, sk
);
1498 sk
->sk_user_data
= xprt
;
1499 sk
->sk_data_ready
= xs_udp_data_ready
;
1500 sk
->sk_write_space
= xs_udp_write_space
;
1501 sk
->sk_no_check
= UDP_CSUM_NORCV
;
1502 sk
->sk_allocation
= GFP_ATOMIC
;
1504 xprt_set_connected(xprt
);
1506 /* Reset to new socket */
1507 transport
->sock
= sock
;
1508 transport
->inet
= sk
;
1510 write_unlock_bh(&sk
->sk_callback_lock
);
1512 xs_udp_do_set_buffer_size(xprt
);
1516 * xs_udp_connect_worker4 - set up a UDP socket
1517 * @work: RPC transport to connect
1519 * Invoked by a work queue tasklet.
1521 static void xs_udp_connect_worker4(struct work_struct
*work
)
1523 struct sock_xprt
*transport
=
1524 container_of(work
, struct sock_xprt
, connect_worker
.work
);
1525 struct rpc_xprt
*xprt
= &transport
->xprt
;
1526 struct socket
*sock
= transport
->sock
;
1527 int err
, status
= -EIO
;
1529 if (xprt
->shutdown
|| !xprt_bound(xprt
))
1532 /* Start by resetting any existing state */
1535 if ((err
= sock_create_kern(PF_INET
, SOCK_DGRAM
, IPPROTO_UDP
, &sock
)) < 0) {
1536 dprintk("RPC: can't create UDP transport socket (%d).\n", -err
);
1539 xs_reclassify_socket4(sock
);
1541 if (xs_bind4(transport
, sock
)) {
1546 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1547 xprt
, xprt
->address_strings
[RPC_DISPLAY_ALL
]);
1549 xs_udp_finish_connecting(xprt
, sock
);
1552 xprt_wake_pending_tasks(xprt
, status
);
1553 xprt_clear_connecting(xprt
);
1557 * xs_udp_connect_worker6 - set up a UDP socket
1558 * @work: RPC transport to connect
1560 * Invoked by a work queue tasklet.
1562 static void xs_udp_connect_worker6(struct work_struct
*work
)
1564 struct sock_xprt
*transport
=
1565 container_of(work
, struct sock_xprt
, connect_worker
.work
);
1566 struct rpc_xprt
*xprt
= &transport
->xprt
;
1567 struct socket
*sock
= transport
->sock
;
1568 int err
, status
= -EIO
;
1570 if (xprt
->shutdown
|| !xprt_bound(xprt
))
1573 /* Start by resetting any existing state */
1576 if ((err
= sock_create_kern(PF_INET6
, SOCK_DGRAM
, IPPROTO_UDP
, &sock
)) < 0) {
1577 dprintk("RPC: can't create UDP transport socket (%d).\n", -err
);
1580 xs_reclassify_socket6(sock
);
1582 if (xs_bind6(transport
, sock
) < 0) {
1587 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1588 xprt
, xprt
->address_strings
[RPC_DISPLAY_ALL
]);
1590 xs_udp_finish_connecting(xprt
, sock
);
1593 xprt_wake_pending_tasks(xprt
, status
);
1594 xprt_clear_connecting(xprt
);
1598 * We need to preserve the port number so the reply cache on the server can
1599 * find our cached RPC replies when we get around to reconnecting.
1601 static void xs_tcp_reuse_connection(struct rpc_xprt
*xprt
)
1604 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1605 struct sockaddr any
;
1607 dprintk("RPC: disconnecting xprt %p to reuse port\n", xprt
);
1610 * Disconnect the transport socket by doing a connect operation
1611 * with AF_UNSPEC. This should return immediately...
1613 memset(&any
, 0, sizeof(any
));
1614 any
.sa_family
= AF_UNSPEC
;
1615 result
= kernel_connect(transport
->sock
, &any
, sizeof(any
), 0);
1617 dprintk("RPC: AF_UNSPEC connect return code %d\n",
1621 static int xs_tcp_finish_connecting(struct rpc_xprt
*xprt
, struct socket
*sock
)
1623 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1625 if (!transport
->inet
) {
1626 struct sock
*sk
= sock
->sk
;
1628 write_lock_bh(&sk
->sk_callback_lock
);
1630 xs_save_old_callbacks(transport
, sk
);
1632 sk
->sk_user_data
= xprt
;
1633 sk
->sk_data_ready
= xs_tcp_data_ready
;
1634 sk
->sk_state_change
= xs_tcp_state_change
;
1635 sk
->sk_write_space
= xs_tcp_write_space
;
1636 sk
->sk_error_report
= xs_tcp_error_report
;
1637 sk
->sk_allocation
= GFP_ATOMIC
;
1639 /* socket options */
1640 sk
->sk_userlocks
|= SOCK_BINDPORT_LOCK
;
1641 sock_reset_flag(sk
, SOCK_LINGER
);
1642 tcp_sk(sk
)->linger2
= 0;
1643 tcp_sk(sk
)->nonagle
|= TCP_NAGLE_OFF
;
1645 xprt_clear_connected(xprt
);
1647 /* Reset to new socket */
1648 transport
->sock
= sock
;
1649 transport
->inet
= sk
;
1651 write_unlock_bh(&sk
->sk_callback_lock
);
1654 /* Tell the socket layer to start connecting... */
1655 xprt
->stat
.connect_count
++;
1656 xprt
->stat
.connect_start
= jiffies
;
1657 return kernel_connect(sock
, xs_addr(xprt
), xprt
->addrlen
, O_NONBLOCK
);
1661 * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1662 * @work: RPC transport to connect
1664 * Invoked by a work queue tasklet.
1666 static void xs_tcp_connect_worker4(struct work_struct
*work
)
1668 struct sock_xprt
*transport
=
1669 container_of(work
, struct sock_xprt
, connect_worker
.work
);
1670 struct rpc_xprt
*xprt
= &transport
->xprt
;
1671 struct socket
*sock
= transport
->sock
;
1672 int err
, status
= -EIO
;
1674 if (xprt
->shutdown
|| !xprt_bound(xprt
))
1678 /* start from scratch */
1679 if ((err
= sock_create_kern(PF_INET
, SOCK_STREAM
, IPPROTO_TCP
, &sock
)) < 0) {
1680 dprintk("RPC: can't create TCP transport socket (%d).\n", -err
);
1683 xs_reclassify_socket4(sock
);
1685 if (xs_bind4(transport
, sock
) < 0) {
1690 /* "close" the socket, preserving the local port */
1691 xs_tcp_reuse_connection(xprt
);
1693 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1694 xprt
, xprt
->address_strings
[RPC_DISPLAY_ALL
]);
1696 status
= xs_tcp_finish_connecting(xprt
, sock
);
1697 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
1698 xprt
, -status
, xprt_connected(xprt
),
1699 sock
->sk
->sk_state
);
1707 /* retry with existing socket, after a delay */
1710 /* get rid of existing socket, and retry */
1711 xs_tcp_shutdown(xprt
);
1715 xprt_wake_pending_tasks(xprt
, status
);
1717 xprt_clear_connecting(xprt
);
1721 * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
1722 * @work: RPC transport to connect
1724 * Invoked by a work queue tasklet.
1726 static void xs_tcp_connect_worker6(struct work_struct
*work
)
1728 struct sock_xprt
*transport
=
1729 container_of(work
, struct sock_xprt
, connect_worker
.work
);
1730 struct rpc_xprt
*xprt
= &transport
->xprt
;
1731 struct socket
*sock
= transport
->sock
;
1732 int err
, status
= -EIO
;
1734 if (xprt
->shutdown
|| !xprt_bound(xprt
))
1738 /* start from scratch */
1739 if ((err
= sock_create_kern(PF_INET6
, SOCK_STREAM
, IPPROTO_TCP
, &sock
)) < 0) {
1740 dprintk("RPC: can't create TCP transport socket (%d).\n", -err
);
1743 xs_reclassify_socket6(sock
);
1745 if (xs_bind6(transport
, sock
) < 0) {
1750 /* "close" the socket, preserving the local port */
1751 xs_tcp_reuse_connection(xprt
);
1753 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1754 xprt
, xprt
->address_strings
[RPC_DISPLAY_ALL
]);
1756 status
= xs_tcp_finish_connecting(xprt
, sock
);
1757 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
1758 xprt
, -status
, xprt_connected(xprt
), sock
->sk
->sk_state
);
1766 /* retry with existing socket, after a delay */
1769 /* get rid of existing socket, and retry */
1770 xs_tcp_shutdown(xprt
);
1774 xprt_wake_pending_tasks(xprt
, status
);
1776 xprt_clear_connecting(xprt
);
1780 * xs_connect - connect a socket to a remote endpoint
1781 * @task: address of RPC task that manages state of connect request
1783 * TCP: If the remote end dropped the connection, delay reconnecting.
1785 * UDP socket connects are synchronous, but we use a work queue anyway
1786 * to guarantee that even unprivileged user processes can set up a
1787 * socket on a privileged port.
1789 * If a UDP socket connect fails, the delay behavior here prevents
1790 * retry floods (hard mounts).
1792 static void xs_connect(struct rpc_task
*task
)
1794 struct rpc_xprt
*xprt
= task
->tk_xprt
;
1795 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1797 if (xprt_test_and_set_connecting(xprt
))
1800 if (transport
->sock
!= NULL
) {
1801 dprintk("RPC: xs_connect delayed xprt %p for %lu "
1803 xprt
, xprt
->reestablish_timeout
/ HZ
);
1804 queue_delayed_work(rpciod_workqueue
,
1805 &transport
->connect_worker
,
1806 xprt
->reestablish_timeout
);
1807 xprt
->reestablish_timeout
<<= 1;
1808 if (xprt
->reestablish_timeout
> XS_TCP_MAX_REEST_TO
)
1809 xprt
->reestablish_timeout
= XS_TCP_MAX_REEST_TO
;
1811 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt
);
1812 queue_delayed_work(rpciod_workqueue
,
1813 &transport
->connect_worker
, 0);
1817 static void xs_tcp_connect(struct rpc_task
*task
)
1819 struct rpc_xprt
*xprt
= task
->tk_xprt
;
1821 /* Initiate graceful shutdown of the socket if not already done */
1822 if (test_bit(XPRT_CONNECTED
, &xprt
->state
))
1823 xs_tcp_shutdown(xprt
);
1824 /* Exit if we need to wait for socket shutdown to complete */
1825 if (test_bit(XPRT_CLOSING
, &xprt
->state
))
1831 * xs_udp_print_stats - display UDP socket-specifc stats
1832 * @xprt: rpc_xprt struct containing statistics
1836 static void xs_udp_print_stats(struct rpc_xprt
*xprt
, struct seq_file
*seq
)
1838 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1840 seq_printf(seq
, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1842 xprt
->stat
.bind_count
,
1845 xprt
->stat
.bad_xids
,
1847 xprt
->stat
.bklog_u
);
1851 * xs_tcp_print_stats - display TCP socket-specifc stats
1852 * @xprt: rpc_xprt struct containing statistics
1856 static void xs_tcp_print_stats(struct rpc_xprt
*xprt
, struct seq_file
*seq
)
1858 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1861 if (xprt_connected(xprt
))
1862 idle_time
= (long)(jiffies
- xprt
->last_used
) / HZ
;
1864 seq_printf(seq
, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1866 xprt
->stat
.bind_count
,
1867 xprt
->stat
.connect_count
,
1868 xprt
->stat
.connect_time
,
1872 xprt
->stat
.bad_xids
,
1874 xprt
->stat
.bklog_u
);
1877 static struct rpc_xprt_ops xs_udp_ops
= {
1878 .set_buffer_size
= xs_udp_set_buffer_size
,
1879 .reserve_xprt
= xprt_reserve_xprt_cong
,
1880 .release_xprt
= xprt_release_xprt_cong
,
1881 .rpcbind
= rpcb_getport_async
,
1882 .set_port
= xs_set_port
,
1883 .connect
= xs_connect
,
1884 .buf_alloc
= rpc_malloc
,
1885 .buf_free
= rpc_free
,
1886 .send_request
= xs_udp_send_request
,
1887 .set_retrans_timeout
= xprt_set_retrans_timeout_rtt
,
1888 .timer
= xs_udp_timer
,
1889 .release_request
= xprt_release_rqst_cong
,
1891 .destroy
= xs_destroy
,
1892 .print_stats
= xs_udp_print_stats
,
1895 static struct rpc_xprt_ops xs_tcp_ops
= {
1896 .reserve_xprt
= xprt_reserve_xprt
,
1897 .release_xprt
= xs_tcp_release_xprt
,
1898 .rpcbind
= rpcb_getport_async
,
1899 .set_port
= xs_set_port
,
1900 .connect
= xs_tcp_connect
,
1901 .buf_alloc
= rpc_malloc
,
1902 .buf_free
= rpc_free
,
1903 .send_request
= xs_tcp_send_request
,
1904 .set_retrans_timeout
= xprt_set_retrans_timeout_def
,
1905 .close
= xs_tcp_shutdown
,
1906 .destroy
= xs_destroy
,
1907 .print_stats
= xs_tcp_print_stats
,
1910 static struct rpc_xprt
*xs_setup_xprt(struct xprt_create
*args
,
1911 unsigned int slot_table_size
)
1913 struct rpc_xprt
*xprt
;
1914 struct sock_xprt
*new;
1916 if (args
->addrlen
> sizeof(xprt
->addr
)) {
1917 dprintk("RPC: xs_setup_xprt: address too large\n");
1918 return ERR_PTR(-EBADF
);
1921 new = kzalloc(sizeof(*new), GFP_KERNEL
);
1923 dprintk("RPC: xs_setup_xprt: couldn't allocate "
1925 return ERR_PTR(-ENOMEM
);
1929 xprt
->max_reqs
= slot_table_size
;
1930 xprt
->slot
= kcalloc(xprt
->max_reqs
, sizeof(struct rpc_rqst
), GFP_KERNEL
);
1931 if (xprt
->slot
== NULL
) {
1933 dprintk("RPC: xs_setup_xprt: couldn't allocate slot "
1935 return ERR_PTR(-ENOMEM
);
1938 memcpy(&xprt
->addr
, args
->dstaddr
, args
->addrlen
);
1939 xprt
->addrlen
= args
->addrlen
;
1941 memcpy(&new->addr
, args
->srcaddr
, args
->addrlen
);
1946 static const struct rpc_timeout xs_udp_default_timeout
= {
1947 .to_initval
= 5 * HZ
,
1948 .to_maxval
= 30 * HZ
,
1949 .to_increment
= 5 * HZ
,
1954 * xs_setup_udp - Set up transport to use a UDP socket
1955 * @args: rpc transport creation arguments
1958 static struct rpc_xprt
*xs_setup_udp(struct xprt_create
*args
)
1960 struct sockaddr
*addr
= args
->dstaddr
;
1961 struct rpc_xprt
*xprt
;
1962 struct sock_xprt
*transport
;
1964 xprt
= xs_setup_xprt(args
, xprt_udp_slot_table_entries
);
1967 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1969 xprt
->prot
= IPPROTO_UDP
;
1971 /* XXX: header size can vary due to auth type, IPv6, etc. */
1972 xprt
->max_payload
= (1U << 16) - (MAX_HEADER
<< 3);
1974 xprt
->bind_timeout
= XS_BIND_TO
;
1975 xprt
->connect_timeout
= XS_UDP_CONN_TO
;
1976 xprt
->reestablish_timeout
= XS_UDP_REEST_TO
;
1977 xprt
->idle_timeout
= XS_IDLE_DISC_TO
;
1979 xprt
->ops
= &xs_udp_ops
;
1981 xprt
->timeout
= &xs_udp_default_timeout
;
1983 switch (addr
->sa_family
) {
1985 if (((struct sockaddr_in
*)addr
)->sin_port
!= htons(0))
1986 xprt_set_bound(xprt
);
1988 INIT_DELAYED_WORK(&transport
->connect_worker
,
1989 xs_udp_connect_worker4
);
1990 xs_format_ipv4_peer_addresses(xprt
, "udp", RPCBIND_NETID_UDP
);
1993 if (((struct sockaddr_in6
*)addr
)->sin6_port
!= htons(0))
1994 xprt_set_bound(xprt
);
1996 INIT_DELAYED_WORK(&transport
->connect_worker
,
1997 xs_udp_connect_worker6
);
1998 xs_format_ipv6_peer_addresses(xprt
, "udp", RPCBIND_NETID_UDP6
);
2002 return ERR_PTR(-EAFNOSUPPORT
);
2005 dprintk("RPC: set up transport to address %s\n",
2006 xprt
->address_strings
[RPC_DISPLAY_ALL
]);
2008 if (try_module_get(THIS_MODULE
))
2013 return ERR_PTR(-EINVAL
);
2016 static const struct rpc_timeout xs_tcp_default_timeout
= {
2017 .to_initval
= 60 * HZ
,
2018 .to_maxval
= 60 * HZ
,
2023 * xs_setup_tcp - Set up transport to use a TCP socket
2024 * @args: rpc transport creation arguments
2027 static struct rpc_xprt
*xs_setup_tcp(struct xprt_create
*args
)
2029 struct sockaddr
*addr
= args
->dstaddr
;
2030 struct rpc_xprt
*xprt
;
2031 struct sock_xprt
*transport
;
2033 xprt
= xs_setup_xprt(args
, xprt_tcp_slot_table_entries
);
2036 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2038 xprt
->prot
= IPPROTO_TCP
;
2039 xprt
->tsh_size
= sizeof(rpc_fraghdr
) / sizeof(u32
);
2040 xprt
->max_payload
= RPC_MAX_FRAGMENT_SIZE
;
2042 xprt
->bind_timeout
= XS_BIND_TO
;
2043 xprt
->connect_timeout
= XS_TCP_CONN_TO
;
2044 xprt
->reestablish_timeout
= XS_TCP_INIT_REEST_TO
;
2045 xprt
->idle_timeout
= XS_IDLE_DISC_TO
;
2047 xprt
->ops
= &xs_tcp_ops
;
2048 xprt
->timeout
= &xs_tcp_default_timeout
;
2050 switch (addr
->sa_family
) {
2052 if (((struct sockaddr_in
*)addr
)->sin_port
!= htons(0))
2053 xprt_set_bound(xprt
);
2055 INIT_DELAYED_WORK(&transport
->connect_worker
, xs_tcp_connect_worker4
);
2056 xs_format_ipv4_peer_addresses(xprt
, "tcp", RPCBIND_NETID_TCP
);
2059 if (((struct sockaddr_in6
*)addr
)->sin6_port
!= htons(0))
2060 xprt_set_bound(xprt
);
2062 INIT_DELAYED_WORK(&transport
->connect_worker
, xs_tcp_connect_worker6
);
2063 xs_format_ipv6_peer_addresses(xprt
, "tcp", RPCBIND_NETID_TCP6
);
2067 return ERR_PTR(-EAFNOSUPPORT
);
2070 dprintk("RPC: set up transport to address %s\n",
2071 xprt
->address_strings
[RPC_DISPLAY_ALL
]);
2073 if (try_module_get(THIS_MODULE
))
2078 return ERR_PTR(-EINVAL
);
2081 static struct xprt_class xs_udp_transport
= {
2082 .list
= LIST_HEAD_INIT(xs_udp_transport
.list
),
2084 .owner
= THIS_MODULE
,
2085 .ident
= IPPROTO_UDP
,
2086 .setup
= xs_setup_udp
,
2089 static struct xprt_class xs_tcp_transport
= {
2090 .list
= LIST_HEAD_INIT(xs_tcp_transport
.list
),
2092 .owner
= THIS_MODULE
,
2093 .ident
= IPPROTO_TCP
,
2094 .setup
= xs_setup_tcp
,
2098 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2101 int init_socket_xprt(void)
2104 if (!sunrpc_table_header
)
2105 sunrpc_table_header
= register_sysctl_table(sunrpc_table
);
2108 xprt_register_transport(&xs_udp_transport
);
2109 xprt_register_transport(&xs_tcp_transport
);
2115 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2118 void cleanup_socket_xprt(void)
2121 if (sunrpc_table_header
) {
2122 unregister_sysctl_table(sunrpc_table_header
);
2123 sunrpc_table_header
= NULL
;
2127 xprt_unregister_transport(&xs_udp_transport
);
2128 xprt_unregister_transport(&xs_tcp_transport
);