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/string.h>
23 #include <linux/slab.h>
24 #include <linux/module.h>
25 #include <linux/capability.h>
26 #include <linux/pagemap.h>
27 #include <linux/errno.h>
28 #include <linux/socket.h>
30 #include <linux/net.h>
33 #include <linux/udp.h>
34 #include <linux/tcp.h>
35 #include <linux/sunrpc/clnt.h>
36 #include <linux/sunrpc/addr.h>
37 #include <linux/sunrpc/sched.h>
38 #include <linux/sunrpc/svcsock.h>
39 #include <linux/sunrpc/xprtsock.h>
40 #include <linux/file.h>
41 #ifdef CONFIG_SUNRPC_BACKCHANNEL
42 #include <linux/sunrpc/bc_xprt.h>
46 #include <net/checksum.h>
50 #include <trace/events/sunrpc.h>
54 static void xs_close(struct rpc_xprt
*xprt
);
59 static unsigned int xprt_udp_slot_table_entries
= RPC_DEF_SLOT_TABLE
;
60 static unsigned int xprt_tcp_slot_table_entries
= RPC_MIN_SLOT_TABLE
;
61 static unsigned int xprt_max_tcp_slot_table_entries
= RPC_MAX_SLOT_TABLE
;
63 static unsigned int xprt_min_resvport
= RPC_DEF_MIN_RESVPORT
;
64 static unsigned int xprt_max_resvport
= RPC_DEF_MAX_RESVPORT
;
66 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
68 #define XS_TCP_LINGER_TO (15U * HZ)
69 static unsigned int xs_tcp_fin_timeout __read_mostly
= XS_TCP_LINGER_TO
;
72 * We can register our own files under /proc/sys/sunrpc by
73 * calling register_sysctl_table() again. The files in that
74 * directory become the union of all files registered there.
76 * We simply need to make sure that we don't collide with
77 * someone else's file names!
80 static unsigned int min_slot_table_size
= RPC_MIN_SLOT_TABLE
;
81 static unsigned int max_slot_table_size
= RPC_MAX_SLOT_TABLE
;
82 static unsigned int max_tcp_slot_table_limit
= RPC_MAX_SLOT_TABLE_LIMIT
;
83 static unsigned int xprt_min_resvport_limit
= RPC_MIN_RESVPORT
;
84 static unsigned int xprt_max_resvport_limit
= RPC_MAX_RESVPORT
;
86 static struct ctl_table_header
*sunrpc_table_header
;
89 * FIXME: changing the UDP slot table size should also resize the UDP
90 * socket buffers for existing UDP transports
92 static struct ctl_table xs_tunables_table
[] = {
94 .procname
= "udp_slot_table_entries",
95 .data
= &xprt_udp_slot_table_entries
,
96 .maxlen
= sizeof(unsigned int),
98 .proc_handler
= proc_dointvec_minmax
,
99 .extra1
= &min_slot_table_size
,
100 .extra2
= &max_slot_table_size
103 .procname
= "tcp_slot_table_entries",
104 .data
= &xprt_tcp_slot_table_entries
,
105 .maxlen
= sizeof(unsigned int),
107 .proc_handler
= proc_dointvec_minmax
,
108 .extra1
= &min_slot_table_size
,
109 .extra2
= &max_slot_table_size
112 .procname
= "tcp_max_slot_table_entries",
113 .data
= &xprt_max_tcp_slot_table_entries
,
114 .maxlen
= sizeof(unsigned int),
116 .proc_handler
= proc_dointvec_minmax
,
117 .extra1
= &min_slot_table_size
,
118 .extra2
= &max_tcp_slot_table_limit
121 .procname
= "min_resvport",
122 .data
= &xprt_min_resvport
,
123 .maxlen
= sizeof(unsigned int),
125 .proc_handler
= proc_dointvec_minmax
,
126 .extra1
= &xprt_min_resvport_limit
,
127 .extra2
= &xprt_max_resvport
130 .procname
= "max_resvport",
131 .data
= &xprt_max_resvport
,
132 .maxlen
= sizeof(unsigned int),
134 .proc_handler
= proc_dointvec_minmax
,
135 .extra1
= &xprt_min_resvport
,
136 .extra2
= &xprt_max_resvport_limit
139 .procname
= "tcp_fin_timeout",
140 .data
= &xs_tcp_fin_timeout
,
141 .maxlen
= sizeof(xs_tcp_fin_timeout
),
143 .proc_handler
= proc_dointvec_jiffies
,
148 static struct ctl_table sunrpc_table
[] = {
150 .procname
= "sunrpc",
152 .child
= xs_tunables_table
160 * Wait duration for a reply from the RPC portmapper.
162 #define XS_BIND_TO (60U * HZ)
165 * Delay if a UDP socket connect error occurs. This is most likely some
166 * kind of resource problem on the local host.
168 #define XS_UDP_REEST_TO (2U * HZ)
171 * The reestablish timeout allows clients to delay for a bit before attempting
172 * to reconnect to a server that just dropped our connection.
174 * We implement an exponential backoff when trying to reestablish a TCP
175 * transport connection with the server. Some servers like to drop a TCP
176 * connection when they are overworked, so we start with a short timeout and
177 * increase over time if the server is down or not responding.
179 #define XS_TCP_INIT_REEST_TO (3U * 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)
188 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
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
)
218 static inline struct rpc_xprt
*xprt_from_sock(struct sock
*sk
)
220 return (struct rpc_xprt
*) sk
->sk_user_data
;
223 static inline struct sockaddr
*xs_addr(struct rpc_xprt
*xprt
)
225 return (struct sockaddr
*) &xprt
->addr
;
228 static inline struct sockaddr_un
*xs_addr_un(struct rpc_xprt
*xprt
)
230 return (struct sockaddr_un
*) &xprt
->addr
;
233 static inline struct sockaddr_in
*xs_addr_in(struct rpc_xprt
*xprt
)
235 return (struct sockaddr_in
*) &xprt
->addr
;
238 static inline struct sockaddr_in6
*xs_addr_in6(struct rpc_xprt
*xprt
)
240 return (struct sockaddr_in6
*) &xprt
->addr
;
243 static void xs_format_common_peer_addresses(struct rpc_xprt
*xprt
)
245 struct sockaddr
*sap
= xs_addr(xprt
);
246 struct sockaddr_in6
*sin6
;
247 struct sockaddr_in
*sin
;
248 struct sockaddr_un
*sun
;
251 switch (sap
->sa_family
) {
253 sun
= xs_addr_un(xprt
);
254 strlcpy(buf
, sun
->sun_path
, sizeof(buf
));
255 xprt
->address_strings
[RPC_DISPLAY_ADDR
] =
256 kstrdup(buf
, GFP_KERNEL
);
259 (void)rpc_ntop(sap
, buf
, sizeof(buf
));
260 xprt
->address_strings
[RPC_DISPLAY_ADDR
] =
261 kstrdup(buf
, GFP_KERNEL
);
262 sin
= xs_addr_in(xprt
);
263 snprintf(buf
, sizeof(buf
), "%08x", ntohl(sin
->sin_addr
.s_addr
));
266 (void)rpc_ntop(sap
, buf
, sizeof(buf
));
267 xprt
->address_strings
[RPC_DISPLAY_ADDR
] =
268 kstrdup(buf
, GFP_KERNEL
);
269 sin6
= xs_addr_in6(xprt
);
270 snprintf(buf
, sizeof(buf
), "%pi6", &sin6
->sin6_addr
);
276 xprt
->address_strings
[RPC_DISPLAY_HEX_ADDR
] = kstrdup(buf
, GFP_KERNEL
);
279 static void xs_format_common_peer_ports(struct rpc_xprt
*xprt
)
281 struct sockaddr
*sap
= xs_addr(xprt
);
284 snprintf(buf
, sizeof(buf
), "%u", rpc_get_port(sap
));
285 xprt
->address_strings
[RPC_DISPLAY_PORT
] = kstrdup(buf
, GFP_KERNEL
);
287 snprintf(buf
, sizeof(buf
), "%4hx", rpc_get_port(sap
));
288 xprt
->address_strings
[RPC_DISPLAY_HEX_PORT
] = kstrdup(buf
, GFP_KERNEL
);
291 static void xs_format_peer_addresses(struct rpc_xprt
*xprt
,
292 const char *protocol
,
295 xprt
->address_strings
[RPC_DISPLAY_PROTO
] = protocol
;
296 xprt
->address_strings
[RPC_DISPLAY_NETID
] = netid
;
297 xs_format_common_peer_addresses(xprt
);
298 xs_format_common_peer_ports(xprt
);
301 static void xs_update_peer_port(struct rpc_xprt
*xprt
)
303 kfree(xprt
->address_strings
[RPC_DISPLAY_HEX_PORT
]);
304 kfree(xprt
->address_strings
[RPC_DISPLAY_PORT
]);
306 xs_format_common_peer_ports(xprt
);
309 static void xs_free_peer_addresses(struct rpc_xprt
*xprt
)
313 for (i
= 0; i
< RPC_DISPLAY_MAX
; i
++)
315 case RPC_DISPLAY_PROTO
:
316 case RPC_DISPLAY_NETID
:
319 kfree(xprt
->address_strings
[i
]);
323 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
325 static int xs_send_kvec(struct socket
*sock
, struct sockaddr
*addr
, int addrlen
, struct kvec
*vec
, unsigned int base
, int more
)
327 struct msghdr msg
= {
329 .msg_namelen
= addrlen
,
330 .msg_flags
= XS_SENDMSG_FLAGS
| (more
? MSG_MORE
: 0),
333 .iov_base
= vec
->iov_base
+ base
,
334 .iov_len
= vec
->iov_len
- base
,
337 if (iov
.iov_len
!= 0)
338 return kernel_sendmsg(sock
, &msg
, &iov
, 1, iov
.iov_len
);
339 return kernel_sendmsg(sock
, &msg
, NULL
, 0, 0);
342 static int xs_send_pagedata(struct socket
*sock
, struct xdr_buf
*xdr
, unsigned int base
, int more
, bool zerocopy
, int *sent_p
)
344 ssize_t (*do_sendpage
)(struct socket
*sock
, struct page
*page
,
345 int offset
, size_t size
, int flags
);
347 unsigned int remainder
;
350 remainder
= xdr
->page_len
- base
;
351 base
+= xdr
->page_base
;
352 ppage
= xdr
->pages
+ (base
>> PAGE_SHIFT
);
354 do_sendpage
= sock
->ops
->sendpage
;
356 do_sendpage
= sock_no_sendpage
;
358 unsigned int len
= min_t(unsigned int, PAGE_SIZE
- base
, remainder
);
359 int flags
= XS_SENDMSG_FLAGS
;
365 flags
|= MSG_SENDPAGE_NOTLAST
| MSG_MORE
;
366 err
= do_sendpage(sock
, *ppage
, base
, len
, flags
);
367 if (remainder
== 0 || err
!= len
)
381 * xs_sendpages - write pages directly to a socket
382 * @sock: socket to send on
383 * @addr: UDP only -- address of destination
384 * @addrlen: UDP only -- length of destination address
385 * @xdr: buffer containing this request
386 * @base: starting position in the buffer
387 * @zerocopy: true if it is safe to use sendpage()
388 * @sent_p: return the total number of bytes successfully queued for sending
391 static int xs_sendpages(struct socket
*sock
, struct sockaddr
*addr
, int addrlen
, struct xdr_buf
*xdr
, unsigned int base
, bool zerocopy
, int *sent_p
)
393 unsigned int remainder
= xdr
->len
- base
;
405 if (base
< xdr
->head
[0].iov_len
|| addr
!= NULL
) {
406 unsigned int len
= xdr
->head
[0].iov_len
- base
;
408 err
= xs_send_kvec(sock
, addr
, addrlen
, &xdr
->head
[0], base
, remainder
!= 0);
409 if (remainder
== 0 || err
!= len
)
414 base
-= xdr
->head
[0].iov_len
;
416 if (base
< xdr
->page_len
) {
417 unsigned int len
= xdr
->page_len
- base
;
419 err
= xs_send_pagedata(sock
, xdr
, base
, remainder
!= 0, zerocopy
, &sent
);
421 if (remainder
== 0 || sent
!= len
)
425 base
-= xdr
->page_len
;
427 if (base
>= xdr
->tail
[0].iov_len
)
429 err
= xs_send_kvec(sock
, NULL
, 0, &xdr
->tail
[0], base
, 0);
438 static void xs_nospace_callback(struct rpc_task
*task
)
440 struct sock_xprt
*transport
= container_of(task
->tk_rqstp
->rq_xprt
, struct sock_xprt
, xprt
);
442 transport
->inet
->sk_write_pending
--;
446 * xs_nospace - place task on wait queue if transmit was incomplete
447 * @task: task to put to sleep
450 static int xs_nospace(struct rpc_task
*task
)
452 struct rpc_rqst
*req
= task
->tk_rqstp
;
453 struct rpc_xprt
*xprt
= req
->rq_xprt
;
454 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
455 struct sock
*sk
= transport
->inet
;
458 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
459 task
->tk_pid
, req
->rq_slen
- req
->rq_bytes_sent
,
462 /* Protect against races with write_space */
463 spin_lock_bh(&xprt
->transport_lock
);
465 /* Don't race with disconnect */
466 if (xprt_connected(xprt
)) {
467 /* wait for more buffer space */
468 sk
->sk_write_pending
++;
469 xprt_wait_for_buffer_space(task
, xs_nospace_callback
);
473 spin_unlock_bh(&xprt
->transport_lock
);
475 /* Race breaker in case memory is freed before above code is called */
476 sk
->sk_write_space(sk
);
481 * Construct a stream transport record marker in @buf.
483 static inline void xs_encode_stream_record_marker(struct xdr_buf
*buf
)
485 u32 reclen
= buf
->len
- sizeof(rpc_fraghdr
);
486 rpc_fraghdr
*base
= buf
->head
[0].iov_base
;
487 *base
= cpu_to_be32(RPC_LAST_STREAM_FRAGMENT
| reclen
);
491 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
492 * @task: RPC task that manages the state of an RPC request
495 * 0: The request has been sent
496 * EAGAIN: The socket was blocked, please call again later to
497 * complete the request
498 * ENOTCONN: Caller needs to invoke connect logic then call again
499 * other: Some other error occured, the request was not sent
501 static int xs_local_send_request(struct rpc_task
*task
)
503 struct rpc_rqst
*req
= task
->tk_rqstp
;
504 struct rpc_xprt
*xprt
= req
->rq_xprt
;
505 struct sock_xprt
*transport
=
506 container_of(xprt
, struct sock_xprt
, xprt
);
507 struct xdr_buf
*xdr
= &req
->rq_snd_buf
;
511 xs_encode_stream_record_marker(&req
->rq_snd_buf
);
513 xs_pktdump("packet data:",
514 req
->rq_svec
->iov_base
, req
->rq_svec
->iov_len
);
516 status
= xs_sendpages(transport
->sock
, NULL
, 0, xdr
, req
->rq_bytes_sent
,
518 dprintk("RPC: %s(%u) = %d\n",
519 __func__
, xdr
->len
- req
->rq_bytes_sent
, status
);
521 if (status
== -EAGAIN
&& sock_writeable(transport
->inet
))
524 if (likely(sent
> 0) || status
== 0) {
525 req
->rq_bytes_sent
+= sent
;
526 req
->rq_xmit_bytes_sent
+= sent
;
527 if (likely(req
->rq_bytes_sent
>= req
->rq_slen
)) {
528 req
->rq_bytes_sent
= 0;
538 status
= xs_nospace(task
);
541 dprintk("RPC: sendmsg returned unrecognized error %d\n",
552 * xs_udp_send_request - write an RPC request to a UDP socket
553 * @task: address of RPC task that manages the state of an RPC request
556 * 0: The request has been sent
557 * EAGAIN: The socket was blocked, please call again later to
558 * complete the request
559 * ENOTCONN: Caller needs to invoke connect logic then call again
560 * other: Some other error occurred, the request was not sent
562 static int xs_udp_send_request(struct rpc_task
*task
)
564 struct rpc_rqst
*req
= task
->tk_rqstp
;
565 struct rpc_xprt
*xprt
= req
->rq_xprt
;
566 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
567 struct xdr_buf
*xdr
= &req
->rq_snd_buf
;
571 xs_pktdump("packet data:",
572 req
->rq_svec
->iov_base
,
573 req
->rq_svec
->iov_len
);
575 if (!xprt_bound(xprt
))
577 status
= xs_sendpages(transport
->sock
, xs_addr(xprt
), xprt
->addrlen
,
578 xdr
, req
->rq_bytes_sent
, true, &sent
);
580 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
581 xdr
->len
- req
->rq_bytes_sent
, status
);
583 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
584 if (status
== -EPERM
)
587 if (status
== -EAGAIN
&& sock_writeable(transport
->inet
))
590 if (sent
> 0 || status
== 0) {
591 req
->rq_xmit_bytes_sent
+= sent
;
592 if (sent
>= req
->rq_slen
)
594 /* Still some bytes left; set up for a retry later. */
602 /* Should we call xs_close() here? */
605 status
= xs_nospace(task
);
612 /* When the server has died, an ICMP port unreachable message
613 * prompts ECONNREFUSED. */
616 dprintk("RPC: sendmsg returned unrecognized error %d\n",
624 * xs_tcp_send_request - write an RPC request to a TCP socket
625 * @task: address of RPC task that manages the state of an RPC request
628 * 0: The request has been sent
629 * EAGAIN: The socket was blocked, please call again later to
630 * complete the request
631 * ENOTCONN: Caller needs to invoke connect logic then call again
632 * other: Some other error occurred, the request was not sent
634 * XXX: In the case of soft timeouts, should we eventually give up
635 * if sendmsg is not able to make progress?
637 static int xs_tcp_send_request(struct rpc_task
*task
)
639 struct rpc_rqst
*req
= task
->tk_rqstp
;
640 struct rpc_xprt
*xprt
= req
->rq_xprt
;
641 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
642 struct xdr_buf
*xdr
= &req
->rq_snd_buf
;
643 bool zerocopy
= true;
644 bool vm_wait
= false;
648 xs_encode_stream_record_marker(&req
->rq_snd_buf
);
650 xs_pktdump("packet data:",
651 req
->rq_svec
->iov_base
,
652 req
->rq_svec
->iov_len
);
653 /* Don't use zero copy if this is a resend. If the RPC call
654 * completes while the socket holds a reference to the pages,
655 * then we may end up resending corrupted data.
657 if (task
->tk_flags
& RPC_TASK_SENT
)
660 /* Continue transmitting the packet/record. We must be careful
661 * to cope with writespace callbacks arriving _after_ we have
662 * called sendmsg(). */
665 status
= xs_sendpages(transport
->sock
, NULL
, 0, xdr
,
666 req
->rq_bytes_sent
, zerocopy
, &sent
);
668 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
669 xdr
->len
- req
->rq_bytes_sent
, status
);
671 /* If we've sent the entire packet, immediately
672 * reset the count of bytes sent. */
673 req
->rq_bytes_sent
+= sent
;
674 req
->rq_xmit_bytes_sent
+= sent
;
675 if (likely(req
->rq_bytes_sent
>= req
->rq_slen
)) {
676 req
->rq_bytes_sent
= 0;
680 WARN_ON_ONCE(sent
== 0 && status
== 0);
682 if (status
== -EAGAIN
) {
684 * Return EAGAIN if we're sure we're hitting the
685 * socket send buffer limits.
687 if (test_bit(SOCK_NOSPACE
, &transport
->sock
->flags
))
690 * Did we hit a memory allocation failure?
696 /* Retry, knowing now that we're below the
697 * socket send buffer limit
711 /* Should we call xs_close() here? */
714 status
= xs_nospace(task
);
724 dprintk("RPC: sendmsg returned unrecognized error %d\n",
732 * xs_tcp_release_xprt - clean up after a tcp transmission
736 * This cleans up if an error causes us to abort the transmission of a request.
737 * In this case, the socket may need to be reset in order to avoid confusing
740 static void xs_tcp_release_xprt(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
742 struct rpc_rqst
*req
;
744 if (task
!= xprt
->snd_task
)
748 req
= task
->tk_rqstp
;
751 if (req
->rq_bytes_sent
== 0)
753 if (req
->rq_bytes_sent
== req
->rq_snd_buf
.len
)
755 set_bit(XPRT_CLOSE_WAIT
, &xprt
->state
);
757 xprt_release_xprt(xprt
, task
);
760 static void xs_save_old_callbacks(struct sock_xprt
*transport
, struct sock
*sk
)
762 transport
->old_data_ready
= sk
->sk_data_ready
;
763 transport
->old_state_change
= sk
->sk_state_change
;
764 transport
->old_write_space
= sk
->sk_write_space
;
765 transport
->old_error_report
= sk
->sk_error_report
;
768 static void xs_restore_old_callbacks(struct sock_xprt
*transport
, struct sock
*sk
)
770 sk
->sk_data_ready
= transport
->old_data_ready
;
771 sk
->sk_state_change
= transport
->old_state_change
;
772 sk
->sk_write_space
= transport
->old_write_space
;
773 sk
->sk_error_report
= transport
->old_error_report
;
776 static void xs_sock_reset_state_flags(struct rpc_xprt
*xprt
)
778 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
780 clear_bit(XPRT_SOCK_DATA_READY
, &transport
->sock_state
);
783 static void xs_sock_reset_connection_flags(struct rpc_xprt
*xprt
)
785 smp_mb__before_atomic();
786 clear_bit(XPRT_CLOSE_WAIT
, &xprt
->state
);
787 clear_bit(XPRT_CLOSING
, &xprt
->state
);
788 xs_sock_reset_state_flags(xprt
);
789 smp_mb__after_atomic();
792 static void xs_sock_mark_closed(struct rpc_xprt
*xprt
)
794 xs_sock_reset_connection_flags(xprt
);
795 /* Mark transport as closed and wake up all pending tasks */
796 xprt_disconnect_done(xprt
);
800 * xs_error_report - callback to handle TCP socket state errors
803 * Note: we don't call sock_error() since there may be a rpc_task
804 * using the socket, and so we don't want to clear sk->sk_err.
806 static void xs_error_report(struct sock
*sk
)
808 struct rpc_xprt
*xprt
;
811 read_lock_bh(&sk
->sk_callback_lock
);
812 if (!(xprt
= xprt_from_sock(sk
)))
818 /* Is this a reset event? */
819 if (sk
->sk_state
== TCP_CLOSE
)
820 xs_sock_mark_closed(xprt
);
821 dprintk("RPC: xs_error_report client %p, error=%d...\n",
823 trace_rpc_socket_error(xprt
, sk
->sk_socket
, err
);
824 xprt_wake_pending_tasks(xprt
, err
);
826 read_unlock_bh(&sk
->sk_callback_lock
);
829 static void xs_reset_transport(struct sock_xprt
*transport
)
831 struct socket
*sock
= transport
->sock
;
832 struct sock
*sk
= transport
->inet
;
833 struct rpc_xprt
*xprt
= &transport
->xprt
;
838 if (atomic_read(&transport
->xprt
.swapper
))
839 sk_clear_memalloc(sk
);
841 kernel_sock_shutdown(sock
, SHUT_RDWR
);
843 mutex_lock(&transport
->recv_mutex
);
844 write_lock_bh(&sk
->sk_callback_lock
);
845 transport
->inet
= NULL
;
846 transport
->sock
= NULL
;
848 sk
->sk_user_data
= NULL
;
850 xs_restore_old_callbacks(transport
, sk
);
851 xprt_clear_connected(xprt
);
852 write_unlock_bh(&sk
->sk_callback_lock
);
853 xs_sock_reset_connection_flags(xprt
);
854 mutex_unlock(&transport
->recv_mutex
);
856 trace_rpc_socket_close(xprt
, sock
);
861 * xs_close - close a socket
864 * This is used when all requests are complete; ie, no DRC state remains
865 * on the server we want to save.
867 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
868 * xs_reset_transport() zeroing the socket from underneath a writer.
870 static void xs_close(struct rpc_xprt
*xprt
)
872 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
874 dprintk("RPC: xs_close xprt %p\n", xprt
);
876 xs_reset_transport(transport
);
877 xprt
->reestablish_timeout
= 0;
879 xprt_disconnect_done(xprt
);
882 static void xs_inject_disconnect(struct rpc_xprt
*xprt
)
884 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
886 xprt_disconnect_done(xprt
);
889 static void xs_xprt_free(struct rpc_xprt
*xprt
)
891 xs_free_peer_addresses(xprt
);
896 * xs_destroy - prepare to shutdown a transport
897 * @xprt: doomed transport
900 static void xs_destroy(struct rpc_xprt
*xprt
)
902 struct sock_xprt
*transport
= container_of(xprt
,
903 struct sock_xprt
, xprt
);
904 dprintk("RPC: xs_destroy xprt %p\n", xprt
);
906 cancel_delayed_work_sync(&transport
->connect_worker
);
908 cancel_work_sync(&transport
->recv_worker
);
910 module_put(THIS_MODULE
);
913 static int xs_local_copy_to_xdr(struct xdr_buf
*xdr
, struct sk_buff
*skb
)
915 struct xdr_skb_reader desc
= {
917 .offset
= sizeof(rpc_fraghdr
),
918 .count
= skb
->len
- sizeof(rpc_fraghdr
),
921 if (xdr_partial_copy_from_skb(xdr
, 0, &desc
, xdr_skb_read_bits
) < 0)
929 * xs_local_data_read_skb
934 * Currently this assumes we can read the whole reply in a single gulp.
936 static void xs_local_data_read_skb(struct rpc_xprt
*xprt
,
940 struct rpc_task
*task
;
941 struct rpc_rqst
*rovr
;
946 repsize
= skb
->len
- sizeof(rpc_fraghdr
);
948 dprintk("RPC: impossible RPC reply size %d\n", repsize
);
952 /* Copy the XID from the skb... */
953 xp
= skb_header_pointer(skb
, sizeof(rpc_fraghdr
), sizeof(_xid
), &_xid
);
957 /* Look up and lock the request corresponding to the given XID */
958 spin_lock_bh(&xprt
->transport_lock
);
959 rovr
= xprt_lookup_rqst(xprt
, *xp
);
962 task
= rovr
->rq_task
;
964 copied
= rovr
->rq_private_buf
.buflen
;
965 if (copied
> repsize
)
968 if (xs_local_copy_to_xdr(&rovr
->rq_private_buf
, skb
)) {
969 dprintk("RPC: sk_buff copy failed\n");
973 xprt_complete_rqst(task
, copied
);
976 spin_unlock_bh(&xprt
->transport_lock
);
979 static void xs_local_data_receive(struct sock_xprt
*transport
)
985 mutex_lock(&transport
->recv_mutex
);
986 sk
= transport
->inet
;
990 skb
= skb_recv_datagram(sk
, 0, 1, &err
);
992 xs_local_data_read_skb(&transport
->xprt
, sk
, skb
);
993 skb_free_datagram(sk
, skb
);
996 if (!test_and_clear_bit(XPRT_SOCK_DATA_READY
, &transport
->sock_state
))
1000 mutex_unlock(&transport
->recv_mutex
);
1003 static void xs_local_data_receive_workfn(struct work_struct
*work
)
1005 struct sock_xprt
*transport
=
1006 container_of(work
, struct sock_xprt
, recv_worker
);
1007 xs_local_data_receive(transport
);
1011 * xs_udp_data_read_skb - receive callback for UDP sockets
1017 static void xs_udp_data_read_skb(struct rpc_xprt
*xprt
,
1019 struct sk_buff
*skb
)
1021 struct rpc_task
*task
;
1022 struct rpc_rqst
*rovr
;
1023 int repsize
, copied
;
1029 dprintk("RPC: impossible RPC reply size %d!\n", repsize
);
1033 /* Copy the XID from the skb... */
1034 xp
= skb_header_pointer(skb
, 0, sizeof(_xid
), &_xid
);
1038 /* Look up and lock the request corresponding to the given XID */
1039 spin_lock_bh(&xprt
->transport_lock
);
1040 rovr
= xprt_lookup_rqst(xprt
, *xp
);
1043 task
= rovr
->rq_task
;
1045 if ((copied
= rovr
->rq_private_buf
.buflen
) > repsize
)
1048 /* Suck it into the iovec, verify checksum if not done by hw. */
1049 if (csum_partial_copy_to_xdr(&rovr
->rq_private_buf
, skb
)) {
1050 __UDPX_INC_STATS(sk
, UDP_MIB_INERRORS
);
1054 __UDPX_INC_STATS(sk
, UDP_MIB_INDATAGRAMS
);
1056 xprt_adjust_cwnd(xprt
, task
, copied
);
1057 xprt_complete_rqst(task
, copied
);
1060 spin_unlock_bh(&xprt
->transport_lock
);
1063 static void xs_udp_data_receive(struct sock_xprt
*transport
)
1065 struct sk_buff
*skb
;
1069 mutex_lock(&transport
->recv_mutex
);
1070 sk
= transport
->inet
;
1074 skb
= skb_recv_datagram(sk
, 0, 1, &err
);
1076 xs_udp_data_read_skb(&transport
->xprt
, sk
, skb
);
1077 skb_free_datagram(sk
, skb
);
1080 if (!test_and_clear_bit(XPRT_SOCK_DATA_READY
, &transport
->sock_state
))
1084 mutex_unlock(&transport
->recv_mutex
);
1087 static void xs_udp_data_receive_workfn(struct work_struct
*work
)
1089 struct sock_xprt
*transport
=
1090 container_of(work
, struct sock_xprt
, recv_worker
);
1091 xs_udp_data_receive(transport
);
1095 * xs_data_ready - "data ready" callback for UDP sockets
1096 * @sk: socket with data to read
1099 static void xs_data_ready(struct sock
*sk
)
1101 struct rpc_xprt
*xprt
;
1103 read_lock_bh(&sk
->sk_callback_lock
);
1104 dprintk("RPC: xs_data_ready...\n");
1105 xprt
= xprt_from_sock(sk
);
1107 struct sock_xprt
*transport
= container_of(xprt
,
1108 struct sock_xprt
, xprt
);
1109 transport
->old_data_ready(sk
);
1110 /* Any data means we had a useful conversation, so
1111 * then we don't need to delay the next reconnect
1113 if (xprt
->reestablish_timeout
)
1114 xprt
->reestablish_timeout
= 0;
1115 if (!test_and_set_bit(XPRT_SOCK_DATA_READY
, &transport
->sock_state
))
1116 queue_work(xprtiod_workqueue
, &transport
->recv_worker
);
1118 read_unlock_bh(&sk
->sk_callback_lock
);
1122 * Helper function to force a TCP close if the server is sending
1123 * junk and/or it has put us in CLOSE_WAIT
1125 static void xs_tcp_force_close(struct rpc_xprt
*xprt
)
1127 xprt_force_disconnect(xprt
);
1130 static inline void xs_tcp_read_fraghdr(struct rpc_xprt
*xprt
, struct xdr_skb_reader
*desc
)
1132 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1136 p
= ((char *) &transport
->tcp_fraghdr
) + transport
->tcp_offset
;
1137 len
= sizeof(transport
->tcp_fraghdr
) - transport
->tcp_offset
;
1138 used
= xdr_skb_read_bits(desc
, p
, len
);
1139 transport
->tcp_offset
+= used
;
1143 transport
->tcp_reclen
= ntohl(transport
->tcp_fraghdr
);
1144 if (transport
->tcp_reclen
& RPC_LAST_STREAM_FRAGMENT
)
1145 transport
->tcp_flags
|= TCP_RCV_LAST_FRAG
;
1147 transport
->tcp_flags
&= ~TCP_RCV_LAST_FRAG
;
1148 transport
->tcp_reclen
&= RPC_FRAGMENT_SIZE_MASK
;
1150 transport
->tcp_flags
&= ~TCP_RCV_COPY_FRAGHDR
;
1151 transport
->tcp_offset
= 0;
1153 /* Sanity check of the record length */
1154 if (unlikely(transport
->tcp_reclen
< 8)) {
1155 dprintk("RPC: invalid TCP record fragment length\n");
1156 xs_tcp_force_close(xprt
);
1159 dprintk("RPC: reading TCP record fragment of length %d\n",
1160 transport
->tcp_reclen
);
1163 static void xs_tcp_check_fraghdr(struct sock_xprt
*transport
)
1165 if (transport
->tcp_offset
== transport
->tcp_reclen
) {
1166 transport
->tcp_flags
|= TCP_RCV_COPY_FRAGHDR
;
1167 transport
->tcp_offset
= 0;
1168 if (transport
->tcp_flags
& TCP_RCV_LAST_FRAG
) {
1169 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
1170 transport
->tcp_flags
|= TCP_RCV_COPY_XID
;
1171 transport
->tcp_copied
= 0;
1176 static inline void xs_tcp_read_xid(struct sock_xprt
*transport
, struct xdr_skb_reader
*desc
)
1181 len
= sizeof(transport
->tcp_xid
) - transport
->tcp_offset
;
1182 dprintk("RPC: reading XID (%Zu bytes)\n", len
);
1183 p
= ((char *) &transport
->tcp_xid
) + transport
->tcp_offset
;
1184 used
= xdr_skb_read_bits(desc
, p
, len
);
1185 transport
->tcp_offset
+= used
;
1188 transport
->tcp_flags
&= ~TCP_RCV_COPY_XID
;
1189 transport
->tcp_flags
|= TCP_RCV_READ_CALLDIR
;
1190 transport
->tcp_copied
= 4;
1191 dprintk("RPC: reading %s XID %08x\n",
1192 (transport
->tcp_flags
& TCP_RPC_REPLY
) ? "reply for"
1194 ntohl(transport
->tcp_xid
));
1195 xs_tcp_check_fraghdr(transport
);
1198 static inline void xs_tcp_read_calldir(struct sock_xprt
*transport
,
1199 struct xdr_skb_reader
*desc
)
1206 * We want transport->tcp_offset to be 8 at the end of this routine
1207 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1208 * When this function is called for the first time,
1209 * transport->tcp_offset is 4 (after having already read the xid).
1211 offset
= transport
->tcp_offset
- sizeof(transport
->tcp_xid
);
1212 len
= sizeof(transport
->tcp_calldir
) - offset
;
1213 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len
);
1214 p
= ((char *) &transport
->tcp_calldir
) + offset
;
1215 used
= xdr_skb_read_bits(desc
, p
, len
);
1216 transport
->tcp_offset
+= used
;
1219 transport
->tcp_flags
&= ~TCP_RCV_READ_CALLDIR
;
1221 * We don't yet have the XDR buffer, so we will write the calldir
1222 * out after we get the buffer from the 'struct rpc_rqst'
1224 switch (ntohl(transport
->tcp_calldir
)) {
1226 transport
->tcp_flags
|= TCP_RCV_COPY_CALLDIR
;
1227 transport
->tcp_flags
|= TCP_RCV_COPY_DATA
;
1228 transport
->tcp_flags
|= TCP_RPC_REPLY
;
1231 transport
->tcp_flags
|= TCP_RCV_COPY_CALLDIR
;
1232 transport
->tcp_flags
|= TCP_RCV_COPY_DATA
;
1233 transport
->tcp_flags
&= ~TCP_RPC_REPLY
;
1236 dprintk("RPC: invalid request message type\n");
1237 xs_tcp_force_close(&transport
->xprt
);
1239 xs_tcp_check_fraghdr(transport
);
1242 static inline void xs_tcp_read_common(struct rpc_xprt
*xprt
,
1243 struct xdr_skb_reader
*desc
,
1244 struct rpc_rqst
*req
)
1246 struct sock_xprt
*transport
=
1247 container_of(xprt
, struct sock_xprt
, xprt
);
1248 struct xdr_buf
*rcvbuf
;
1252 rcvbuf
= &req
->rq_private_buf
;
1254 if (transport
->tcp_flags
& TCP_RCV_COPY_CALLDIR
) {
1256 * Save the RPC direction in the XDR buffer
1258 memcpy(rcvbuf
->head
[0].iov_base
+ transport
->tcp_copied
,
1259 &transport
->tcp_calldir
,
1260 sizeof(transport
->tcp_calldir
));
1261 transport
->tcp_copied
+= sizeof(transport
->tcp_calldir
);
1262 transport
->tcp_flags
&= ~TCP_RCV_COPY_CALLDIR
;
1266 if (len
> transport
->tcp_reclen
- transport
->tcp_offset
) {
1267 struct xdr_skb_reader my_desc
;
1269 len
= transport
->tcp_reclen
- transport
->tcp_offset
;
1270 memcpy(&my_desc
, desc
, sizeof(my_desc
));
1271 my_desc
.count
= len
;
1272 r
= xdr_partial_copy_from_skb(rcvbuf
, transport
->tcp_copied
,
1273 &my_desc
, xdr_skb_read_bits
);
1277 r
= xdr_partial_copy_from_skb(rcvbuf
, transport
->tcp_copied
,
1278 desc
, xdr_skb_read_bits
);
1281 transport
->tcp_copied
+= r
;
1282 transport
->tcp_offset
+= r
;
1285 /* Error when copying to the receive buffer,
1286 * usually because we weren't able to allocate
1287 * additional buffer pages. All we can do now
1288 * is turn off TCP_RCV_COPY_DATA, so the request
1289 * will not receive any additional updates,
1291 * Any remaining data from this record will
1294 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
1295 dprintk("RPC: XID %08x truncated request\n",
1296 ntohl(transport
->tcp_xid
));
1297 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1298 "tcp_offset = %u, tcp_reclen = %u\n",
1299 xprt
, transport
->tcp_copied
,
1300 transport
->tcp_offset
, transport
->tcp_reclen
);
1304 dprintk("RPC: XID %08x read %Zd bytes\n",
1305 ntohl(transport
->tcp_xid
), r
);
1306 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1307 "tcp_reclen = %u\n", xprt
, transport
->tcp_copied
,
1308 transport
->tcp_offset
, transport
->tcp_reclen
);
1310 if (transport
->tcp_copied
== req
->rq_private_buf
.buflen
)
1311 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
1312 else if (transport
->tcp_offset
== transport
->tcp_reclen
) {
1313 if (transport
->tcp_flags
& TCP_RCV_LAST_FRAG
)
1314 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
1319 * Finds the request corresponding to the RPC xid and invokes the common
1320 * tcp read code to read the data.
1322 static inline int xs_tcp_read_reply(struct rpc_xprt
*xprt
,
1323 struct xdr_skb_reader
*desc
)
1325 struct sock_xprt
*transport
=
1326 container_of(xprt
, struct sock_xprt
, xprt
);
1327 struct rpc_rqst
*req
;
1329 dprintk("RPC: read reply XID %08x\n", ntohl(transport
->tcp_xid
));
1331 /* Find and lock the request corresponding to this xid */
1332 spin_lock_bh(&xprt
->transport_lock
);
1333 req
= xprt_lookup_rqst(xprt
, transport
->tcp_xid
);
1335 dprintk("RPC: XID %08x request not found!\n",
1336 ntohl(transport
->tcp_xid
));
1337 spin_unlock_bh(&xprt
->transport_lock
);
1341 xs_tcp_read_common(xprt
, desc
, req
);
1343 if (!(transport
->tcp_flags
& TCP_RCV_COPY_DATA
))
1344 xprt_complete_rqst(req
->rq_task
, transport
->tcp_copied
);
1346 spin_unlock_bh(&xprt
->transport_lock
);
1350 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1352 * Obtains an rpc_rqst previously allocated and invokes the common
1353 * tcp read code to read the data. The result is placed in the callback
1355 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1356 * connection and return -1.
1358 static int xs_tcp_read_callback(struct rpc_xprt
*xprt
,
1359 struct xdr_skb_reader
*desc
)
1361 struct sock_xprt
*transport
=
1362 container_of(xprt
, struct sock_xprt
, xprt
);
1363 struct rpc_rqst
*req
;
1365 /* Look up and lock the request corresponding to the given XID */
1366 spin_lock_bh(&xprt
->transport_lock
);
1367 req
= xprt_lookup_bc_request(xprt
, transport
->tcp_xid
);
1369 spin_unlock_bh(&xprt
->transport_lock
);
1370 printk(KERN_WARNING
"Callback slot table overflowed\n");
1371 xprt_force_disconnect(xprt
);
1375 dprintk("RPC: read callback XID %08x\n", ntohl(req
->rq_xid
));
1376 xs_tcp_read_common(xprt
, desc
, req
);
1378 if (!(transport
->tcp_flags
& TCP_RCV_COPY_DATA
))
1379 xprt_complete_bc_request(req
, transport
->tcp_copied
);
1380 spin_unlock_bh(&xprt
->transport_lock
);
1385 static inline int _xs_tcp_read_data(struct rpc_xprt
*xprt
,
1386 struct xdr_skb_reader
*desc
)
1388 struct sock_xprt
*transport
=
1389 container_of(xprt
, struct sock_xprt
, xprt
);
1391 return (transport
->tcp_flags
& TCP_RPC_REPLY
) ?
1392 xs_tcp_read_reply(xprt
, desc
) :
1393 xs_tcp_read_callback(xprt
, desc
);
1396 static int xs_tcp_bc_up(struct svc_serv
*serv
, struct net
*net
)
1400 ret
= svc_create_xprt(serv
, "tcp-bc", net
, PF_INET
, 0,
1401 SVC_SOCK_ANONYMOUS
);
1407 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt
*xprt
)
1412 static inline int _xs_tcp_read_data(struct rpc_xprt
*xprt
,
1413 struct xdr_skb_reader
*desc
)
1415 return xs_tcp_read_reply(xprt
, desc
);
1417 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1420 * Read data off the transport. This can be either an RPC_CALL or an
1421 * RPC_REPLY. Relay the processing to helper functions.
1423 static void xs_tcp_read_data(struct rpc_xprt
*xprt
,
1424 struct xdr_skb_reader
*desc
)
1426 struct sock_xprt
*transport
=
1427 container_of(xprt
, struct sock_xprt
, xprt
);
1429 if (_xs_tcp_read_data(xprt
, desc
) == 0)
1430 xs_tcp_check_fraghdr(transport
);
1433 * The transport_lock protects the request handling.
1434 * There's no need to hold it to update the tcp_flags.
1436 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
1440 static inline void xs_tcp_read_discard(struct sock_xprt
*transport
, struct xdr_skb_reader
*desc
)
1444 len
= transport
->tcp_reclen
- transport
->tcp_offset
;
1445 if (len
> desc
->count
)
1448 desc
->offset
+= len
;
1449 transport
->tcp_offset
+= len
;
1450 dprintk("RPC: discarded %Zu bytes\n", len
);
1451 xs_tcp_check_fraghdr(transport
);
1454 static int xs_tcp_data_recv(read_descriptor_t
*rd_desc
, struct sk_buff
*skb
, unsigned int offset
, size_t len
)
1456 struct rpc_xprt
*xprt
= rd_desc
->arg
.data
;
1457 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1458 struct xdr_skb_reader desc
= {
1464 dprintk("RPC: xs_tcp_data_recv started\n");
1466 trace_xs_tcp_data_recv(transport
);
1467 /* Read in a new fragment marker if necessary */
1468 /* Can we ever really expect to get completely empty fragments? */
1469 if (transport
->tcp_flags
& TCP_RCV_COPY_FRAGHDR
) {
1470 xs_tcp_read_fraghdr(xprt
, &desc
);
1473 /* Read in the xid if necessary */
1474 if (transport
->tcp_flags
& TCP_RCV_COPY_XID
) {
1475 xs_tcp_read_xid(transport
, &desc
);
1478 /* Read in the call/reply flag */
1479 if (transport
->tcp_flags
& TCP_RCV_READ_CALLDIR
) {
1480 xs_tcp_read_calldir(transport
, &desc
);
1483 /* Read in the request data */
1484 if (transport
->tcp_flags
& TCP_RCV_COPY_DATA
) {
1485 xs_tcp_read_data(xprt
, &desc
);
1488 /* Skip over any trailing bytes on short reads */
1489 xs_tcp_read_discard(transport
, &desc
);
1490 } while (desc
.count
);
1491 trace_xs_tcp_data_recv(transport
);
1492 dprintk("RPC: xs_tcp_data_recv done\n");
1493 return len
- desc
.count
;
1496 static void xs_tcp_data_receive(struct sock_xprt
*transport
)
1498 struct rpc_xprt
*xprt
= &transport
->xprt
;
1500 read_descriptor_t rd_desc
= {
1501 .count
= 2*1024*1024,
1504 unsigned long total
= 0;
1507 mutex_lock(&transport
->recv_mutex
);
1508 sk
= transport
->inet
;
1512 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1515 read
= tcp_read_sock(sk
, &rd_desc
, xs_tcp_data_recv
);
1517 clear_bit(XPRT_SOCK_DATA_READY
, &transport
->sock_state
);
1519 if (!test_bit(XPRT_SOCK_DATA_READY
, &transport
->sock_state
))
1525 rd_desc
.count
= 65536;
1528 mutex_unlock(&transport
->recv_mutex
);
1529 trace_xs_tcp_data_ready(xprt
, read
, total
);
1532 static void xs_tcp_data_receive_workfn(struct work_struct
*work
)
1534 struct sock_xprt
*transport
=
1535 container_of(work
, struct sock_xprt
, recv_worker
);
1536 xs_tcp_data_receive(transport
);
1540 * xs_tcp_state_change - callback to handle TCP socket state changes
1541 * @sk: socket whose state has changed
1544 static void xs_tcp_state_change(struct sock
*sk
)
1546 struct rpc_xprt
*xprt
;
1547 struct sock_xprt
*transport
;
1549 read_lock_bh(&sk
->sk_callback_lock
);
1550 if (!(xprt
= xprt_from_sock(sk
)))
1552 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt
);
1553 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1554 sk
->sk_state
, xprt_connected(xprt
),
1555 sock_flag(sk
, SOCK_DEAD
),
1556 sock_flag(sk
, SOCK_ZAPPED
),
1559 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1560 trace_rpc_socket_state_change(xprt
, sk
->sk_socket
);
1561 switch (sk
->sk_state
) {
1562 case TCP_ESTABLISHED
:
1563 spin_lock(&xprt
->transport_lock
);
1564 if (!xprt_test_and_set_connected(xprt
)) {
1566 /* Reset TCP record info */
1567 transport
->tcp_offset
= 0;
1568 transport
->tcp_reclen
= 0;
1569 transport
->tcp_copied
= 0;
1570 transport
->tcp_flags
=
1571 TCP_RCV_COPY_FRAGHDR
| TCP_RCV_COPY_XID
;
1572 xprt
->connect_cookie
++;
1573 clear_bit(XPRT_SOCK_CONNECTING
, &transport
->sock_state
);
1574 xprt_clear_connecting(xprt
);
1576 xprt_wake_pending_tasks(xprt
, -EAGAIN
);
1578 spin_unlock(&xprt
->transport_lock
);
1581 /* The client initiated a shutdown of the socket */
1582 xprt
->connect_cookie
++;
1583 xprt
->reestablish_timeout
= 0;
1584 set_bit(XPRT_CLOSING
, &xprt
->state
);
1585 smp_mb__before_atomic();
1586 clear_bit(XPRT_CONNECTED
, &xprt
->state
);
1587 clear_bit(XPRT_CLOSE_WAIT
, &xprt
->state
);
1588 smp_mb__after_atomic();
1590 case TCP_CLOSE_WAIT
:
1591 /* The server initiated a shutdown of the socket */
1592 xprt
->connect_cookie
++;
1593 clear_bit(XPRT_CONNECTED
, &xprt
->state
);
1594 xs_tcp_force_close(xprt
);
1597 * If the server closed down the connection, make sure that
1598 * we back off before reconnecting
1600 if (xprt
->reestablish_timeout
< XS_TCP_INIT_REEST_TO
)
1601 xprt
->reestablish_timeout
= XS_TCP_INIT_REEST_TO
;
1604 set_bit(XPRT_CLOSING
, &xprt
->state
);
1605 smp_mb__before_atomic();
1606 clear_bit(XPRT_CONNECTED
, &xprt
->state
);
1607 smp_mb__after_atomic();
1610 if (test_and_clear_bit(XPRT_SOCK_CONNECTING
,
1611 &transport
->sock_state
))
1612 xprt_clear_connecting(xprt
);
1613 xs_sock_mark_closed(xprt
);
1616 read_unlock_bh(&sk
->sk_callback_lock
);
1619 static void xs_write_space(struct sock
*sk
)
1621 struct socket_wq
*wq
;
1622 struct rpc_xprt
*xprt
;
1626 clear_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1628 if (unlikely(!(xprt
= xprt_from_sock(sk
))))
1631 wq
= rcu_dereference(sk
->sk_wq
);
1632 if (!wq
|| test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE
, &wq
->flags
) == 0)
1635 xprt_write_space(xprt
);
1641 * xs_udp_write_space - callback invoked when socket buffer space
1643 * @sk: socket whose state has changed
1645 * Called when more output buffer space is available for this socket.
1646 * We try not to wake our writers until they can make "significant"
1647 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1648 * with a bunch of small requests.
1650 static void xs_udp_write_space(struct sock
*sk
)
1652 read_lock_bh(&sk
->sk_callback_lock
);
1654 /* from net/core/sock.c:sock_def_write_space */
1655 if (sock_writeable(sk
))
1658 read_unlock_bh(&sk
->sk_callback_lock
);
1662 * xs_tcp_write_space - callback invoked when socket buffer space
1664 * @sk: socket whose state has changed
1666 * Called when more output buffer space is available for this socket.
1667 * We try not to wake our writers until they can make "significant"
1668 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1669 * with a bunch of small requests.
1671 static void xs_tcp_write_space(struct sock
*sk
)
1673 read_lock_bh(&sk
->sk_callback_lock
);
1675 /* from net/core/stream.c:sk_stream_write_space */
1676 if (sk_stream_is_writeable(sk
))
1679 read_unlock_bh(&sk
->sk_callback_lock
);
1682 static void xs_udp_do_set_buffer_size(struct rpc_xprt
*xprt
)
1684 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1685 struct sock
*sk
= transport
->inet
;
1687 if (transport
->rcvsize
) {
1688 sk
->sk_userlocks
|= SOCK_RCVBUF_LOCK
;
1689 sk
->sk_rcvbuf
= transport
->rcvsize
* xprt
->max_reqs
* 2;
1691 if (transport
->sndsize
) {
1692 sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
;
1693 sk
->sk_sndbuf
= transport
->sndsize
* xprt
->max_reqs
* 2;
1694 sk
->sk_write_space(sk
);
1699 * xs_udp_set_buffer_size - set send and receive limits
1700 * @xprt: generic transport
1701 * @sndsize: requested size of send buffer, in bytes
1702 * @rcvsize: requested size of receive buffer, in bytes
1704 * Set socket send and receive buffer size limits.
1706 static void xs_udp_set_buffer_size(struct rpc_xprt
*xprt
, size_t sndsize
, size_t rcvsize
)
1708 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1710 transport
->sndsize
= 0;
1712 transport
->sndsize
= sndsize
+ 1024;
1713 transport
->rcvsize
= 0;
1715 transport
->rcvsize
= rcvsize
+ 1024;
1717 xs_udp_do_set_buffer_size(xprt
);
1721 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1722 * @task: task that timed out
1724 * Adjust the congestion window after a retransmit timeout has occurred.
1726 static void xs_udp_timer(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
1728 xprt_adjust_cwnd(xprt
, task
, -ETIMEDOUT
);
1731 static unsigned short xs_get_random_port(void)
1733 unsigned short range
= xprt_max_resvport
- xprt_min_resvport
+ 1;
1734 unsigned short rand
= (unsigned short) prandom_u32() % range
;
1735 return rand
+ xprt_min_resvport
;
1739 * xs_set_reuseaddr_port - set the socket's port and address reuse options
1742 * Note that this function has to be called on all sockets that share the
1743 * same port, and it must be called before binding.
1745 static void xs_sock_set_reuseport(struct socket
*sock
)
1749 kernel_setsockopt(sock
, SOL_SOCKET
, SO_REUSEPORT
,
1750 (char *)&opt
, sizeof(opt
));
1753 static unsigned short xs_sock_getport(struct socket
*sock
)
1755 struct sockaddr_storage buf
;
1757 unsigned short port
= 0;
1759 if (kernel_getsockname(sock
, (struct sockaddr
*)&buf
, &buflen
) < 0)
1761 switch (buf
.ss_family
) {
1763 port
= ntohs(((struct sockaddr_in6
*)&buf
)->sin6_port
);
1766 port
= ntohs(((struct sockaddr_in
*)&buf
)->sin_port
);
1773 * xs_set_port - reset the port number in the remote endpoint address
1774 * @xprt: generic transport
1775 * @port: new port number
1778 static void xs_set_port(struct rpc_xprt
*xprt
, unsigned short port
)
1780 dprintk("RPC: setting port for xprt %p to %u\n", xprt
, port
);
1782 rpc_set_port(xs_addr(xprt
), port
);
1783 xs_update_peer_port(xprt
);
1786 static void xs_set_srcport(struct sock_xprt
*transport
, struct socket
*sock
)
1788 if (transport
->srcport
== 0)
1789 transport
->srcport
= xs_sock_getport(sock
);
1792 static unsigned short xs_get_srcport(struct sock_xprt
*transport
)
1794 unsigned short port
= transport
->srcport
;
1796 if (port
== 0 && transport
->xprt
.resvport
)
1797 port
= xs_get_random_port();
1801 static unsigned short xs_next_srcport(struct sock_xprt
*transport
, unsigned short port
)
1803 if (transport
->srcport
!= 0)
1804 transport
->srcport
= 0;
1805 if (!transport
->xprt
.resvport
)
1807 if (port
<= xprt_min_resvport
|| port
> xprt_max_resvport
)
1808 return xprt_max_resvport
;
1811 static int xs_bind(struct sock_xprt
*transport
, struct socket
*sock
)
1813 struct sockaddr_storage myaddr
;
1815 unsigned short port
= xs_get_srcport(transport
);
1816 unsigned short last
;
1819 * If we are asking for any ephemeral port (i.e. port == 0 &&
1820 * transport->xprt.resvport == 0), don't bind. Let the local
1821 * port selection happen implicitly when the socket is used
1822 * (for example at connect time).
1824 * This ensures that we can continue to establish TCP
1825 * connections even when all local ephemeral ports are already
1826 * a part of some TCP connection. This makes no difference
1827 * for UDP sockets, but also doens't harm them.
1829 * If we're asking for any reserved port (i.e. port == 0 &&
1830 * transport->xprt.resvport == 1) xs_get_srcport above will
1831 * ensure that port is non-zero and we will bind as needed.
1836 memcpy(&myaddr
, &transport
->srcaddr
, transport
->xprt
.addrlen
);
1838 rpc_set_port((struct sockaddr
*)&myaddr
, port
);
1839 err
= kernel_bind(sock
, (struct sockaddr
*)&myaddr
,
1840 transport
->xprt
.addrlen
);
1842 transport
->srcport
= port
;
1846 port
= xs_next_srcport(transport
, port
);
1849 } while (err
== -EADDRINUSE
&& nloop
!= 2);
1851 if (myaddr
.ss_family
== AF_INET
)
1852 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__
,
1853 &((struct sockaddr_in
*)&myaddr
)->sin_addr
,
1854 port
, err
? "failed" : "ok", err
);
1856 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__
,
1857 &((struct sockaddr_in6
*)&myaddr
)->sin6_addr
,
1858 port
, err
? "failed" : "ok", err
);
1863 * We don't support autobind on AF_LOCAL sockets
1865 static void xs_local_rpcbind(struct rpc_task
*task
)
1867 xprt_set_bound(task
->tk_xprt
);
1870 static void xs_local_set_port(struct rpc_xprt
*xprt
, unsigned short port
)
1874 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1875 static struct lock_class_key xs_key
[2];
1876 static struct lock_class_key xs_slock_key
[2];
1878 static inline void xs_reclassify_socketu(struct socket
*sock
)
1880 struct sock
*sk
= sock
->sk
;
1882 sock_lock_init_class_and_name(sk
, "slock-AF_LOCAL-RPC",
1883 &xs_slock_key
[1], "sk_lock-AF_LOCAL-RPC", &xs_key
[1]);
1886 static inline void xs_reclassify_socket4(struct socket
*sock
)
1888 struct sock
*sk
= sock
->sk
;
1890 sock_lock_init_class_and_name(sk
, "slock-AF_INET-RPC",
1891 &xs_slock_key
[0], "sk_lock-AF_INET-RPC", &xs_key
[0]);
1894 static inline void xs_reclassify_socket6(struct socket
*sock
)
1896 struct sock
*sk
= sock
->sk
;
1898 sock_lock_init_class_and_name(sk
, "slock-AF_INET6-RPC",
1899 &xs_slock_key
[1], "sk_lock-AF_INET6-RPC", &xs_key
[1]);
1902 static inline void xs_reclassify_socket(int family
, struct socket
*sock
)
1904 if (WARN_ON_ONCE(!sock_allow_reclassification(sock
->sk
)))
1909 xs_reclassify_socketu(sock
);
1912 xs_reclassify_socket4(sock
);
1915 xs_reclassify_socket6(sock
);
1920 static inline void xs_reclassify_socket(int family
, struct socket
*sock
)
1925 static void xs_dummy_setup_socket(struct work_struct
*work
)
1929 static struct socket
*xs_create_sock(struct rpc_xprt
*xprt
,
1930 struct sock_xprt
*transport
, int family
, int type
,
1931 int protocol
, bool reuseport
)
1933 struct socket
*sock
;
1936 err
= __sock_create(xprt
->xprt_net
, family
, type
, protocol
, &sock
, 1);
1938 dprintk("RPC: can't create %d transport socket (%d).\n",
1942 xs_reclassify_socket(family
, sock
);
1945 xs_sock_set_reuseport(sock
);
1947 err
= xs_bind(transport
, sock
);
1955 return ERR_PTR(err
);
1958 static int xs_local_finish_connecting(struct rpc_xprt
*xprt
,
1959 struct socket
*sock
)
1961 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
,
1964 if (!transport
->inet
) {
1965 struct sock
*sk
= sock
->sk
;
1967 write_lock_bh(&sk
->sk_callback_lock
);
1969 xs_save_old_callbacks(transport
, sk
);
1971 sk
->sk_user_data
= xprt
;
1972 sk
->sk_data_ready
= xs_data_ready
;
1973 sk
->sk_write_space
= xs_udp_write_space
;
1974 sock_set_flag(sk
, SOCK_FASYNC
);
1975 sk
->sk_error_report
= xs_error_report
;
1976 sk
->sk_allocation
= GFP_NOIO
;
1978 xprt_clear_connected(xprt
);
1980 /* Reset to new socket */
1981 transport
->sock
= sock
;
1982 transport
->inet
= sk
;
1984 write_unlock_bh(&sk
->sk_callback_lock
);
1987 /* Tell the socket layer to start connecting... */
1988 xprt
->stat
.connect_count
++;
1989 xprt
->stat
.connect_start
= jiffies
;
1990 return kernel_connect(sock
, xs_addr(xprt
), xprt
->addrlen
, 0);
1994 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1995 * @transport: socket transport to connect
1997 static int xs_local_setup_socket(struct sock_xprt
*transport
)
1999 struct rpc_xprt
*xprt
= &transport
->xprt
;
2000 struct socket
*sock
;
2003 status
= __sock_create(xprt
->xprt_net
, AF_LOCAL
,
2004 SOCK_STREAM
, 0, &sock
, 1);
2006 dprintk("RPC: can't create AF_LOCAL "
2007 "transport socket (%d).\n", -status
);
2010 xs_reclassify_socket(AF_LOCAL
, sock
);
2012 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
2013 xprt
, xprt
->address_strings
[RPC_DISPLAY_ADDR
]);
2015 status
= xs_local_finish_connecting(xprt
, sock
);
2016 trace_rpc_socket_connect(xprt
, sock
, status
);
2019 dprintk("RPC: xprt %p connected to %s\n",
2020 xprt
, xprt
->address_strings
[RPC_DISPLAY_ADDR
]);
2021 xprt_set_connected(xprt
);
2025 dprintk("RPC: xprt %p: socket %s does not exist\n",
2026 xprt
, xprt
->address_strings
[RPC_DISPLAY_ADDR
]);
2029 dprintk("RPC: xprt %p: connection refused for %s\n",
2030 xprt
, xprt
->address_strings
[RPC_DISPLAY_ADDR
]);
2033 printk(KERN_ERR
"%s: unhandled error (%d) connecting to %s\n",
2035 xprt
->address_strings
[RPC_DISPLAY_ADDR
]);
2039 xprt_clear_connecting(xprt
);
2040 xprt_wake_pending_tasks(xprt
, status
);
2044 static void xs_local_connect(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
2046 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2049 if (RPC_IS_ASYNC(task
)) {
2051 * We want the AF_LOCAL connect to be resolved in the
2052 * filesystem namespace of the process making the rpc
2053 * call. Thus we connect synchronously.
2055 * If we want to support asynchronous AF_LOCAL calls,
2056 * we'll need to figure out how to pass a namespace to
2059 rpc_exit(task
, -ENOTCONN
);
2062 ret
= xs_local_setup_socket(transport
);
2063 if (ret
&& !RPC_IS_SOFTCONN(task
))
2064 msleep_interruptible(15000);
2067 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2069 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2070 * know that we have exclusive access to the socket), to guard against
2071 * races with xs_reset_transport.
2073 static void xs_set_memalloc(struct rpc_xprt
*xprt
)
2075 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
,
2079 * If there's no sock, then we have nothing to set. The
2080 * reconnecting process will get it for us.
2082 if (!transport
->inet
)
2084 if (atomic_read(&xprt
->swapper
))
2085 sk_set_memalloc(transport
->inet
);
2089 * xs_enable_swap - Tag this transport as being used for swap.
2090 * @xprt: transport to tag
2092 * Take a reference to this transport on behalf of the rpc_clnt, and
2093 * optionally mark it for swapping if it wasn't already.
2096 xs_enable_swap(struct rpc_xprt
*xprt
)
2098 struct sock_xprt
*xs
= container_of(xprt
, struct sock_xprt
, xprt
);
2100 if (atomic_inc_return(&xprt
->swapper
) != 1)
2102 if (wait_on_bit_lock(&xprt
->state
, XPRT_LOCKED
, TASK_KILLABLE
))
2103 return -ERESTARTSYS
;
2105 sk_set_memalloc(xs
->inet
);
2106 xprt_release_xprt(xprt
, NULL
);
2111 * xs_disable_swap - Untag this transport as being used for swap.
2112 * @xprt: transport to tag
2114 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2115 * swapper refcount goes to 0, untag the socket as a memalloc socket.
2118 xs_disable_swap(struct rpc_xprt
*xprt
)
2120 struct sock_xprt
*xs
= container_of(xprt
, struct sock_xprt
, xprt
);
2122 if (!atomic_dec_and_test(&xprt
->swapper
))
2124 if (wait_on_bit_lock(&xprt
->state
, XPRT_LOCKED
, TASK_KILLABLE
))
2127 sk_clear_memalloc(xs
->inet
);
2128 xprt_release_xprt(xprt
, NULL
);
2131 static void xs_set_memalloc(struct rpc_xprt
*xprt
)
2136 xs_enable_swap(struct rpc_xprt
*xprt
)
2142 xs_disable_swap(struct rpc_xprt
*xprt
)
2147 static void xs_udp_finish_connecting(struct rpc_xprt
*xprt
, struct socket
*sock
)
2149 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2151 if (!transport
->inet
) {
2152 struct sock
*sk
= sock
->sk
;
2154 write_lock_bh(&sk
->sk_callback_lock
);
2156 xs_save_old_callbacks(transport
, sk
);
2158 sk
->sk_user_data
= xprt
;
2159 sk
->sk_data_ready
= xs_data_ready
;
2160 sk
->sk_write_space
= xs_udp_write_space
;
2161 sock_set_flag(sk
, SOCK_FASYNC
);
2162 sk
->sk_allocation
= GFP_NOIO
;
2164 xprt_set_connected(xprt
);
2166 /* Reset to new socket */
2167 transport
->sock
= sock
;
2168 transport
->inet
= sk
;
2170 xs_set_memalloc(xprt
);
2172 write_unlock_bh(&sk
->sk_callback_lock
);
2174 xs_udp_do_set_buffer_size(xprt
);
2176 xprt
->stat
.connect_start
= jiffies
;
2179 static void xs_udp_setup_socket(struct work_struct
*work
)
2181 struct sock_xprt
*transport
=
2182 container_of(work
, struct sock_xprt
, connect_worker
.work
);
2183 struct rpc_xprt
*xprt
= &transport
->xprt
;
2184 struct socket
*sock
= transport
->sock
;
2187 sock
= xs_create_sock(xprt
, transport
,
2188 xs_addr(xprt
)->sa_family
, SOCK_DGRAM
,
2189 IPPROTO_UDP
, false);
2193 dprintk("RPC: worker connecting xprt %p via %s to "
2194 "%s (port %s)\n", xprt
,
2195 xprt
->address_strings
[RPC_DISPLAY_PROTO
],
2196 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
2197 xprt
->address_strings
[RPC_DISPLAY_PORT
]);
2199 xs_udp_finish_connecting(xprt
, sock
);
2200 trace_rpc_socket_connect(xprt
, sock
, 0);
2203 xprt_unlock_connect(xprt
, transport
);
2204 xprt_clear_connecting(xprt
);
2205 xprt_wake_pending_tasks(xprt
, status
);
2209 * xs_tcp_shutdown - gracefully shut down a TCP socket
2212 * Initiates a graceful shutdown of the TCP socket by calling the
2213 * equivalent of shutdown(SHUT_RDWR);
2215 static void xs_tcp_shutdown(struct rpc_xprt
*xprt
)
2217 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2218 struct socket
*sock
= transport
->sock
;
2222 if (xprt_connected(xprt
)) {
2223 kernel_sock_shutdown(sock
, SHUT_RDWR
);
2224 trace_rpc_socket_shutdown(xprt
, sock
);
2226 xs_reset_transport(transport
);
2229 static int xs_tcp_finish_connecting(struct rpc_xprt
*xprt
, struct socket
*sock
)
2231 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2232 int ret
= -ENOTCONN
;
2234 if (!transport
->inet
) {
2235 struct sock
*sk
= sock
->sk
;
2236 unsigned int keepidle
= xprt
->timeout
->to_initval
/ HZ
;
2237 unsigned int keepcnt
= xprt
->timeout
->to_retries
+ 1;
2238 unsigned int opt_on
= 1;
2240 unsigned int addr_pref
= IPV6_PREFER_SRC_PUBLIC
;
2242 /* TCP Keepalive options */
2243 kernel_setsockopt(sock
, SOL_SOCKET
, SO_KEEPALIVE
,
2244 (char *)&opt_on
, sizeof(opt_on
));
2245 kernel_setsockopt(sock
, SOL_TCP
, TCP_KEEPIDLE
,
2246 (char *)&keepidle
, sizeof(keepidle
));
2247 kernel_setsockopt(sock
, SOL_TCP
, TCP_KEEPINTVL
,
2248 (char *)&keepidle
, sizeof(keepidle
));
2249 kernel_setsockopt(sock
, SOL_TCP
, TCP_KEEPCNT
,
2250 (char *)&keepcnt
, sizeof(keepcnt
));
2252 /* Avoid temporary address, they are bad for long-lived
2253 * connections such as NFS mounts.
2254 * RFC4941, section 3.6 suggests that:
2255 * Individual applications, which have specific
2256 * knowledge about the normal duration of connections,
2257 * MAY override this as appropriate.
2259 kernel_setsockopt(sock
, SOL_IPV6
, IPV6_ADDR_PREFERENCES
,
2260 (char *)&addr_pref
, sizeof(addr_pref
));
2262 /* TCP user timeout (see RFC5482) */
2263 timeo
= jiffies_to_msecs(xprt
->timeout
->to_initval
) *
2264 (xprt
->timeout
->to_retries
+ 1);
2265 kernel_setsockopt(sock
, SOL_TCP
, TCP_USER_TIMEOUT
,
2266 (char *)&timeo
, sizeof(timeo
));
2268 write_lock_bh(&sk
->sk_callback_lock
);
2270 xs_save_old_callbacks(transport
, sk
);
2272 sk
->sk_user_data
= xprt
;
2273 sk
->sk_data_ready
= xs_data_ready
;
2274 sk
->sk_state_change
= xs_tcp_state_change
;
2275 sk
->sk_write_space
= xs_tcp_write_space
;
2276 sock_set_flag(sk
, SOCK_FASYNC
);
2277 sk
->sk_error_report
= xs_error_report
;
2278 sk
->sk_allocation
= GFP_NOIO
;
2280 /* socket options */
2281 sock_reset_flag(sk
, SOCK_LINGER
);
2282 tcp_sk(sk
)->nonagle
|= TCP_NAGLE_OFF
;
2284 xprt_clear_connected(xprt
);
2286 /* Reset to new socket */
2287 transport
->sock
= sock
;
2288 transport
->inet
= sk
;
2290 write_unlock_bh(&sk
->sk_callback_lock
);
2293 if (!xprt_bound(xprt
))
2296 xs_set_memalloc(xprt
);
2298 /* Tell the socket layer to start connecting... */
2299 xprt
->stat
.connect_count
++;
2300 xprt
->stat
.connect_start
= jiffies
;
2301 set_bit(XPRT_SOCK_CONNECTING
, &transport
->sock_state
);
2302 ret
= kernel_connect(sock
, xs_addr(xprt
), xprt
->addrlen
, O_NONBLOCK
);
2305 xs_set_srcport(transport
, sock
);
2308 if (xprt
->reestablish_timeout
< XS_TCP_INIT_REEST_TO
)
2309 xprt
->reestablish_timeout
= XS_TCP_INIT_REEST_TO
;
2311 case -EADDRNOTAVAIL
:
2312 /* Source port number is unavailable. Try a new one! */
2313 transport
->srcport
= 0;
2320 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2322 * Invoked by a work queue tasklet.
2324 static void xs_tcp_setup_socket(struct work_struct
*work
)
2326 struct sock_xprt
*transport
=
2327 container_of(work
, struct sock_xprt
, connect_worker
.work
);
2328 struct socket
*sock
= transport
->sock
;
2329 struct rpc_xprt
*xprt
= &transport
->xprt
;
2333 sock
= xs_create_sock(xprt
, transport
,
2334 xs_addr(xprt
)->sa_family
, SOCK_STREAM
,
2337 status
= PTR_ERR(sock
);
2342 dprintk("RPC: worker connecting xprt %p via %s to "
2343 "%s (port %s)\n", xprt
,
2344 xprt
->address_strings
[RPC_DISPLAY_PROTO
],
2345 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
2346 xprt
->address_strings
[RPC_DISPLAY_PORT
]);
2348 status
= xs_tcp_finish_connecting(xprt
, sock
);
2349 trace_rpc_socket_connect(xprt
, sock
, status
);
2350 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2351 xprt
, -status
, xprt_connected(xprt
),
2352 sock
->sk
->sk_state
);
2355 printk("%s: connect returned unhandled error %d\n",
2357 case -EADDRNOTAVAIL
:
2358 /* We're probably in TIME_WAIT. Get rid of existing socket,
2361 xs_tcp_force_close(xprt
);
2366 xprt_unlock_connect(xprt
, transport
);
2369 /* Happens, for instance, if the user specified a link
2370 * local IPv6 address without a scope-id.
2377 /* retry with existing socket, after a delay */
2378 xs_tcp_force_close(xprt
);
2383 xprt_unlock_connect(xprt
, transport
);
2384 xprt_clear_connecting(xprt
);
2385 xprt_wake_pending_tasks(xprt
, status
);
2388 static unsigned long xs_reconnect_delay(const struct rpc_xprt
*xprt
)
2390 unsigned long start
, now
= jiffies
;
2392 start
= xprt
->stat
.connect_start
+ xprt
->reestablish_timeout
;
2393 if (time_after(start
, now
))
2398 static void xs_reconnect_backoff(struct rpc_xprt
*xprt
)
2400 xprt
->reestablish_timeout
<<= 1;
2401 if (xprt
->reestablish_timeout
> xprt
->max_reconnect_timeout
)
2402 xprt
->reestablish_timeout
= xprt
->max_reconnect_timeout
;
2403 if (xprt
->reestablish_timeout
< XS_TCP_INIT_REEST_TO
)
2404 xprt
->reestablish_timeout
= XS_TCP_INIT_REEST_TO
;
2408 * xs_connect - connect a socket to a remote endpoint
2409 * @xprt: pointer to transport structure
2410 * @task: address of RPC task that manages state of connect request
2412 * TCP: If the remote end dropped the connection, delay reconnecting.
2414 * UDP socket connects are synchronous, but we use a work queue anyway
2415 * to guarantee that even unprivileged user processes can set up a
2416 * socket on a privileged port.
2418 * If a UDP socket connect fails, the delay behavior here prevents
2419 * retry floods (hard mounts).
2421 static void xs_connect(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
2423 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2424 unsigned long delay
= 0;
2426 WARN_ON_ONCE(!xprt_lock_connect(xprt
, task
, transport
));
2428 if (transport
->sock
!= NULL
) {
2429 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2431 xprt
, xprt
->reestablish_timeout
/ HZ
);
2433 /* Start by resetting any existing state */
2434 xs_reset_transport(transport
);
2436 delay
= xs_reconnect_delay(xprt
);
2437 xs_reconnect_backoff(xprt
);
2440 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt
);
2442 queue_delayed_work(xprtiod_workqueue
,
2443 &transport
->connect_worker
,
2448 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2449 * @xprt: rpc_xprt struct containing statistics
2453 static void xs_local_print_stats(struct rpc_xprt
*xprt
, struct seq_file
*seq
)
2457 if (xprt_connected(xprt
))
2458 idle_time
= (long)(jiffies
- xprt
->last_used
) / HZ
;
2460 seq_printf(seq
, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2461 "%llu %llu %lu %llu %llu\n",
2462 xprt
->stat
.bind_count
,
2463 xprt
->stat
.connect_count
,
2464 xprt
->stat
.connect_time
,
2468 xprt
->stat
.bad_xids
,
2471 xprt
->stat
.max_slots
,
2472 xprt
->stat
.sending_u
,
2473 xprt
->stat
.pending_u
);
2477 * xs_udp_print_stats - display UDP socket-specifc stats
2478 * @xprt: rpc_xprt struct containing statistics
2482 static void xs_udp_print_stats(struct rpc_xprt
*xprt
, struct seq_file
*seq
)
2484 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2486 seq_printf(seq
, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2489 xprt
->stat
.bind_count
,
2492 xprt
->stat
.bad_xids
,
2495 xprt
->stat
.max_slots
,
2496 xprt
->stat
.sending_u
,
2497 xprt
->stat
.pending_u
);
2501 * xs_tcp_print_stats - display TCP socket-specifc stats
2502 * @xprt: rpc_xprt struct containing statistics
2506 static void xs_tcp_print_stats(struct rpc_xprt
*xprt
, struct seq_file
*seq
)
2508 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2511 if (xprt_connected(xprt
))
2512 idle_time
= (long)(jiffies
- xprt
->last_used
) / HZ
;
2514 seq_printf(seq
, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2515 "%llu %llu %lu %llu %llu\n",
2517 xprt
->stat
.bind_count
,
2518 xprt
->stat
.connect_count
,
2519 xprt
->stat
.connect_time
,
2523 xprt
->stat
.bad_xids
,
2526 xprt
->stat
.max_slots
,
2527 xprt
->stat
.sending_u
,
2528 xprt
->stat
.pending_u
);
2532 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2533 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2534 * to use the server side send routines.
2536 static void *bc_malloc(struct rpc_task
*task
, size_t size
)
2539 struct rpc_buffer
*buf
;
2541 WARN_ON_ONCE(size
> PAGE_SIZE
- sizeof(struct rpc_buffer
));
2542 if (size
> PAGE_SIZE
- sizeof(struct rpc_buffer
))
2545 page
= alloc_page(GFP_KERNEL
);
2549 buf
= page_address(page
);
2550 buf
->len
= PAGE_SIZE
;
2556 * Free the space allocated in the bc_alloc routine
2558 static void bc_free(void *buffer
)
2560 struct rpc_buffer
*buf
;
2565 buf
= container_of(buffer
, struct rpc_buffer
, data
);
2566 free_page((unsigned long)buf
);
2570 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2571 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2573 static int bc_sendto(struct rpc_rqst
*req
)
2576 struct xdr_buf
*xbufp
= &req
->rq_snd_buf
;
2577 struct rpc_xprt
*xprt
= req
->rq_xprt
;
2578 struct sock_xprt
*transport
=
2579 container_of(xprt
, struct sock_xprt
, xprt
);
2580 struct socket
*sock
= transport
->sock
;
2581 unsigned long headoff
;
2582 unsigned long tailoff
;
2584 xs_encode_stream_record_marker(xbufp
);
2586 tailoff
= (unsigned long)xbufp
->tail
[0].iov_base
& ~PAGE_MASK
;
2587 headoff
= (unsigned long)xbufp
->head
[0].iov_base
& ~PAGE_MASK
;
2588 len
= svc_send_common(sock
, xbufp
,
2589 virt_to_page(xbufp
->head
[0].iov_base
), headoff
,
2590 xbufp
->tail
[0].iov_base
, tailoff
);
2592 if (len
!= xbufp
->len
) {
2593 printk(KERN_NOTICE
"Error sending entire callback!\n");
2601 * The send routine. Borrows from svc_send
2603 static int bc_send_request(struct rpc_task
*task
)
2605 struct rpc_rqst
*req
= task
->tk_rqstp
;
2606 struct svc_xprt
*xprt
;
2609 dprintk("sending request with xid: %08x\n", ntohl(req
->rq_xid
));
2611 * Get the server socket associated with this callback xprt
2613 xprt
= req
->rq_xprt
->bc_xprt
;
2616 * Grab the mutex to serialize data as the connection is shared
2617 * with the fore channel
2619 if (!mutex_trylock(&xprt
->xpt_mutex
)) {
2620 rpc_sleep_on(&xprt
->xpt_bc_pending
, task
, NULL
);
2621 if (!mutex_trylock(&xprt
->xpt_mutex
))
2623 rpc_wake_up_queued_task(&xprt
->xpt_bc_pending
, task
);
2625 if (test_bit(XPT_DEAD
, &xprt
->xpt_flags
))
2628 len
= bc_sendto(req
);
2629 mutex_unlock(&xprt
->xpt_mutex
);
2638 * The close routine. Since this is client initiated, we do nothing
2641 static void bc_close(struct rpc_xprt
*xprt
)
2646 * The xprt destroy routine. Again, because this connection is client
2647 * initiated, we do nothing
2650 static void bc_destroy(struct rpc_xprt
*xprt
)
2652 dprintk("RPC: bc_destroy xprt %p\n", xprt
);
2655 module_put(THIS_MODULE
);
2658 static struct rpc_xprt_ops xs_local_ops
= {
2659 .reserve_xprt
= xprt_reserve_xprt
,
2660 .release_xprt
= xs_tcp_release_xprt
,
2661 .alloc_slot
= xprt_alloc_slot
,
2662 .rpcbind
= xs_local_rpcbind
,
2663 .set_port
= xs_local_set_port
,
2664 .connect
= xs_local_connect
,
2665 .buf_alloc
= rpc_malloc
,
2666 .buf_free
= rpc_free
,
2667 .send_request
= xs_local_send_request
,
2668 .set_retrans_timeout
= xprt_set_retrans_timeout_def
,
2670 .destroy
= xs_destroy
,
2671 .print_stats
= xs_local_print_stats
,
2672 .enable_swap
= xs_enable_swap
,
2673 .disable_swap
= xs_disable_swap
,
2676 static struct rpc_xprt_ops xs_udp_ops
= {
2677 .set_buffer_size
= xs_udp_set_buffer_size
,
2678 .reserve_xprt
= xprt_reserve_xprt_cong
,
2679 .release_xprt
= xprt_release_xprt_cong
,
2680 .alloc_slot
= xprt_alloc_slot
,
2681 .rpcbind
= rpcb_getport_async
,
2682 .set_port
= xs_set_port
,
2683 .connect
= xs_connect
,
2684 .buf_alloc
= rpc_malloc
,
2685 .buf_free
= rpc_free
,
2686 .send_request
= xs_udp_send_request
,
2687 .set_retrans_timeout
= xprt_set_retrans_timeout_rtt
,
2688 .timer
= xs_udp_timer
,
2689 .release_request
= xprt_release_rqst_cong
,
2691 .destroy
= xs_destroy
,
2692 .print_stats
= xs_udp_print_stats
,
2693 .enable_swap
= xs_enable_swap
,
2694 .disable_swap
= xs_disable_swap
,
2695 .inject_disconnect
= xs_inject_disconnect
,
2698 static struct rpc_xprt_ops xs_tcp_ops
= {
2699 .reserve_xprt
= xprt_reserve_xprt
,
2700 .release_xprt
= xs_tcp_release_xprt
,
2701 .alloc_slot
= xprt_lock_and_alloc_slot
,
2702 .rpcbind
= rpcb_getport_async
,
2703 .set_port
= xs_set_port
,
2704 .connect
= xs_connect
,
2705 .buf_alloc
= rpc_malloc
,
2706 .buf_free
= rpc_free
,
2707 .send_request
= xs_tcp_send_request
,
2708 .set_retrans_timeout
= xprt_set_retrans_timeout_def
,
2709 .close
= xs_tcp_shutdown
,
2710 .destroy
= xs_destroy
,
2711 .print_stats
= xs_tcp_print_stats
,
2712 .enable_swap
= xs_enable_swap
,
2713 .disable_swap
= xs_disable_swap
,
2714 .inject_disconnect
= xs_inject_disconnect
,
2715 #ifdef CONFIG_SUNRPC_BACKCHANNEL
2716 .bc_setup
= xprt_setup_bc
,
2717 .bc_up
= xs_tcp_bc_up
,
2718 .bc_maxpayload
= xs_tcp_bc_maxpayload
,
2719 .bc_free_rqst
= xprt_free_bc_rqst
,
2720 .bc_destroy
= xprt_destroy_bc
,
2725 * The rpc_xprt_ops for the server backchannel
2728 static struct rpc_xprt_ops bc_tcp_ops
= {
2729 .reserve_xprt
= xprt_reserve_xprt
,
2730 .release_xprt
= xprt_release_xprt
,
2731 .alloc_slot
= xprt_alloc_slot
,
2732 .buf_alloc
= bc_malloc
,
2733 .buf_free
= bc_free
,
2734 .send_request
= bc_send_request
,
2735 .set_retrans_timeout
= xprt_set_retrans_timeout_def
,
2737 .destroy
= bc_destroy
,
2738 .print_stats
= xs_tcp_print_stats
,
2739 .enable_swap
= xs_enable_swap
,
2740 .disable_swap
= xs_disable_swap
,
2741 .inject_disconnect
= xs_inject_disconnect
,
2744 static int xs_init_anyaddr(const int family
, struct sockaddr
*sap
)
2746 static const struct sockaddr_in sin
= {
2747 .sin_family
= AF_INET
,
2748 .sin_addr
.s_addr
= htonl(INADDR_ANY
),
2750 static const struct sockaddr_in6 sin6
= {
2751 .sin6_family
= AF_INET6
,
2752 .sin6_addr
= IN6ADDR_ANY_INIT
,
2759 memcpy(sap
, &sin
, sizeof(sin
));
2762 memcpy(sap
, &sin6
, sizeof(sin6
));
2765 dprintk("RPC: %s: Bad address family\n", __func__
);
2766 return -EAFNOSUPPORT
;
2771 static struct rpc_xprt
*xs_setup_xprt(struct xprt_create
*args
,
2772 unsigned int slot_table_size
,
2773 unsigned int max_slot_table_size
)
2775 struct rpc_xprt
*xprt
;
2776 struct sock_xprt
*new;
2778 if (args
->addrlen
> sizeof(xprt
->addr
)) {
2779 dprintk("RPC: xs_setup_xprt: address too large\n");
2780 return ERR_PTR(-EBADF
);
2783 xprt
= xprt_alloc(args
->net
, sizeof(*new), slot_table_size
,
2784 max_slot_table_size
);
2786 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2788 return ERR_PTR(-ENOMEM
);
2791 new = container_of(xprt
, struct sock_xprt
, xprt
);
2792 mutex_init(&new->recv_mutex
);
2793 memcpy(&xprt
->addr
, args
->dstaddr
, args
->addrlen
);
2794 xprt
->addrlen
= args
->addrlen
;
2796 memcpy(&new->srcaddr
, args
->srcaddr
, args
->addrlen
);
2799 err
= xs_init_anyaddr(args
->dstaddr
->sa_family
,
2800 (struct sockaddr
*)&new->srcaddr
);
2803 return ERR_PTR(err
);
2810 static const struct rpc_timeout xs_local_default_timeout
= {
2811 .to_initval
= 10 * HZ
,
2812 .to_maxval
= 10 * HZ
,
2817 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2818 * @args: rpc transport creation arguments
2820 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2822 static struct rpc_xprt
*xs_setup_local(struct xprt_create
*args
)
2824 struct sockaddr_un
*sun
= (struct sockaddr_un
*)args
->dstaddr
;
2825 struct sock_xprt
*transport
;
2826 struct rpc_xprt
*xprt
;
2827 struct rpc_xprt
*ret
;
2829 xprt
= xs_setup_xprt(args
, xprt_tcp_slot_table_entries
,
2830 xprt_max_tcp_slot_table_entries
);
2833 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2836 xprt
->tsh_size
= sizeof(rpc_fraghdr
) / sizeof(u32
);
2837 xprt
->max_payload
= RPC_MAX_FRAGMENT_SIZE
;
2839 xprt
->bind_timeout
= XS_BIND_TO
;
2840 xprt
->reestablish_timeout
= XS_TCP_INIT_REEST_TO
;
2841 xprt
->idle_timeout
= XS_IDLE_DISC_TO
;
2843 xprt
->ops
= &xs_local_ops
;
2844 xprt
->timeout
= &xs_local_default_timeout
;
2846 INIT_WORK(&transport
->recv_worker
, xs_local_data_receive_workfn
);
2847 INIT_DELAYED_WORK(&transport
->connect_worker
,
2848 xs_dummy_setup_socket
);
2850 switch (sun
->sun_family
) {
2852 if (sun
->sun_path
[0] != '/') {
2853 dprintk("RPC: bad AF_LOCAL address: %s\n",
2855 ret
= ERR_PTR(-EINVAL
);
2858 xprt_set_bound(xprt
);
2859 xs_format_peer_addresses(xprt
, "local", RPCBIND_NETID_LOCAL
);
2860 ret
= ERR_PTR(xs_local_setup_socket(transport
));
2865 ret
= ERR_PTR(-EAFNOSUPPORT
);
2869 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2870 xprt
->address_strings
[RPC_DISPLAY_ADDR
]);
2872 if (try_module_get(THIS_MODULE
))
2874 ret
= ERR_PTR(-EINVAL
);
2880 static const struct rpc_timeout xs_udp_default_timeout
= {
2881 .to_initval
= 5 * HZ
,
2882 .to_maxval
= 30 * HZ
,
2883 .to_increment
= 5 * HZ
,
2888 * xs_setup_udp - Set up transport to use a UDP socket
2889 * @args: rpc transport creation arguments
2892 static struct rpc_xprt
*xs_setup_udp(struct xprt_create
*args
)
2894 struct sockaddr
*addr
= args
->dstaddr
;
2895 struct rpc_xprt
*xprt
;
2896 struct sock_xprt
*transport
;
2897 struct rpc_xprt
*ret
;
2899 xprt
= xs_setup_xprt(args
, xprt_udp_slot_table_entries
,
2900 xprt_udp_slot_table_entries
);
2903 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2905 xprt
->prot
= IPPROTO_UDP
;
2907 /* XXX: header size can vary due to auth type, IPv6, etc. */
2908 xprt
->max_payload
= (1U << 16) - (MAX_HEADER
<< 3);
2910 xprt
->bind_timeout
= XS_BIND_TO
;
2911 xprt
->reestablish_timeout
= XS_UDP_REEST_TO
;
2912 xprt
->idle_timeout
= XS_IDLE_DISC_TO
;
2914 xprt
->ops
= &xs_udp_ops
;
2916 xprt
->timeout
= &xs_udp_default_timeout
;
2918 INIT_WORK(&transport
->recv_worker
, xs_udp_data_receive_workfn
);
2919 INIT_DELAYED_WORK(&transport
->connect_worker
, xs_udp_setup_socket
);
2921 switch (addr
->sa_family
) {
2923 if (((struct sockaddr_in
*)addr
)->sin_port
!= htons(0))
2924 xprt_set_bound(xprt
);
2926 xs_format_peer_addresses(xprt
, "udp", RPCBIND_NETID_UDP
);
2929 if (((struct sockaddr_in6
*)addr
)->sin6_port
!= htons(0))
2930 xprt_set_bound(xprt
);
2932 xs_format_peer_addresses(xprt
, "udp", RPCBIND_NETID_UDP6
);
2935 ret
= ERR_PTR(-EAFNOSUPPORT
);
2939 if (xprt_bound(xprt
))
2940 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2941 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
2942 xprt
->address_strings
[RPC_DISPLAY_PORT
],
2943 xprt
->address_strings
[RPC_DISPLAY_PROTO
]);
2945 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2946 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
2947 xprt
->address_strings
[RPC_DISPLAY_PROTO
]);
2949 if (try_module_get(THIS_MODULE
))
2951 ret
= ERR_PTR(-EINVAL
);
2957 static const struct rpc_timeout xs_tcp_default_timeout
= {
2958 .to_initval
= 60 * HZ
,
2959 .to_maxval
= 60 * HZ
,
2964 * xs_setup_tcp - Set up transport to use a TCP socket
2965 * @args: rpc transport creation arguments
2968 static struct rpc_xprt
*xs_setup_tcp(struct xprt_create
*args
)
2970 struct sockaddr
*addr
= args
->dstaddr
;
2971 struct rpc_xprt
*xprt
;
2972 struct sock_xprt
*transport
;
2973 struct rpc_xprt
*ret
;
2974 unsigned int max_slot_table_size
= xprt_max_tcp_slot_table_entries
;
2976 if (args
->flags
& XPRT_CREATE_INFINITE_SLOTS
)
2977 max_slot_table_size
= RPC_MAX_SLOT_TABLE_LIMIT
;
2979 xprt
= xs_setup_xprt(args
, xprt_tcp_slot_table_entries
,
2980 max_slot_table_size
);
2983 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2985 xprt
->prot
= IPPROTO_TCP
;
2986 xprt
->tsh_size
= sizeof(rpc_fraghdr
) / sizeof(u32
);
2987 xprt
->max_payload
= RPC_MAX_FRAGMENT_SIZE
;
2989 xprt
->bind_timeout
= XS_BIND_TO
;
2990 xprt
->reestablish_timeout
= XS_TCP_INIT_REEST_TO
;
2991 xprt
->idle_timeout
= XS_IDLE_DISC_TO
;
2993 xprt
->ops
= &xs_tcp_ops
;
2994 xprt
->timeout
= &xs_tcp_default_timeout
;
2996 xprt
->max_reconnect_timeout
= xprt
->timeout
->to_maxval
;
2998 INIT_WORK(&transport
->recv_worker
, xs_tcp_data_receive_workfn
);
2999 INIT_DELAYED_WORK(&transport
->connect_worker
, xs_tcp_setup_socket
);
3001 switch (addr
->sa_family
) {
3003 if (((struct sockaddr_in
*)addr
)->sin_port
!= htons(0))
3004 xprt_set_bound(xprt
);
3006 xs_format_peer_addresses(xprt
, "tcp", RPCBIND_NETID_TCP
);
3009 if (((struct sockaddr_in6
*)addr
)->sin6_port
!= htons(0))
3010 xprt_set_bound(xprt
);
3012 xs_format_peer_addresses(xprt
, "tcp", RPCBIND_NETID_TCP6
);
3015 ret
= ERR_PTR(-EAFNOSUPPORT
);
3019 if (xprt_bound(xprt
))
3020 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3021 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
3022 xprt
->address_strings
[RPC_DISPLAY_PORT
],
3023 xprt
->address_strings
[RPC_DISPLAY_PROTO
]);
3025 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
3026 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
3027 xprt
->address_strings
[RPC_DISPLAY_PROTO
]);
3029 if (try_module_get(THIS_MODULE
))
3031 ret
= ERR_PTR(-EINVAL
);
3038 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3039 * @args: rpc transport creation arguments
3042 static struct rpc_xprt
*xs_setup_bc_tcp(struct xprt_create
*args
)
3044 struct sockaddr
*addr
= args
->dstaddr
;
3045 struct rpc_xprt
*xprt
;
3046 struct sock_xprt
*transport
;
3047 struct svc_sock
*bc_sock
;
3048 struct rpc_xprt
*ret
;
3050 xprt
= xs_setup_xprt(args
, xprt_tcp_slot_table_entries
,
3051 xprt_tcp_slot_table_entries
);
3054 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
3056 xprt
->prot
= IPPROTO_TCP
;
3057 xprt
->tsh_size
= sizeof(rpc_fraghdr
) / sizeof(u32
);
3058 xprt
->max_payload
= RPC_MAX_FRAGMENT_SIZE
;
3059 xprt
->timeout
= &xs_tcp_default_timeout
;
3062 xprt_set_bound(xprt
);
3063 xprt
->bind_timeout
= 0;
3064 xprt
->reestablish_timeout
= 0;
3065 xprt
->idle_timeout
= 0;
3067 xprt
->ops
= &bc_tcp_ops
;
3069 switch (addr
->sa_family
) {
3071 xs_format_peer_addresses(xprt
, "tcp",
3075 xs_format_peer_addresses(xprt
, "tcp",
3076 RPCBIND_NETID_TCP6
);
3079 ret
= ERR_PTR(-EAFNOSUPPORT
);
3083 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3084 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
3085 xprt
->address_strings
[RPC_DISPLAY_PORT
],
3086 xprt
->address_strings
[RPC_DISPLAY_PROTO
]);
3089 * Once we've associated a backchannel xprt with a connection,
3090 * we want to keep it around as long as the connection lasts,
3091 * in case we need to start using it for a backchannel again;
3092 * this reference won't be dropped until bc_xprt is destroyed.
3095 args
->bc_xprt
->xpt_bc_xprt
= xprt
;
3096 xprt
->bc_xprt
= args
->bc_xprt
;
3097 bc_sock
= container_of(args
->bc_xprt
, struct svc_sock
, sk_xprt
);
3098 transport
->sock
= bc_sock
->sk_sock
;
3099 transport
->inet
= bc_sock
->sk_sk
;
3102 * Since we don't want connections for the backchannel, we set
3103 * the xprt status to connected
3105 xprt_set_connected(xprt
);
3107 if (try_module_get(THIS_MODULE
))
3110 args
->bc_xprt
->xpt_bc_xprt
= NULL
;
3111 args
->bc_xprt
->xpt_bc_xps
= NULL
;
3113 ret
= ERR_PTR(-EINVAL
);
3119 static struct xprt_class xs_local_transport
= {
3120 .list
= LIST_HEAD_INIT(xs_local_transport
.list
),
3121 .name
= "named UNIX socket",
3122 .owner
= THIS_MODULE
,
3123 .ident
= XPRT_TRANSPORT_LOCAL
,
3124 .setup
= xs_setup_local
,
3127 static struct xprt_class xs_udp_transport
= {
3128 .list
= LIST_HEAD_INIT(xs_udp_transport
.list
),
3130 .owner
= THIS_MODULE
,
3131 .ident
= XPRT_TRANSPORT_UDP
,
3132 .setup
= xs_setup_udp
,
3135 static struct xprt_class xs_tcp_transport
= {
3136 .list
= LIST_HEAD_INIT(xs_tcp_transport
.list
),
3138 .owner
= THIS_MODULE
,
3139 .ident
= XPRT_TRANSPORT_TCP
,
3140 .setup
= xs_setup_tcp
,
3143 static struct xprt_class xs_bc_tcp_transport
= {
3144 .list
= LIST_HEAD_INIT(xs_bc_tcp_transport
.list
),
3145 .name
= "tcp NFSv4.1 backchannel",
3146 .owner
= THIS_MODULE
,
3147 .ident
= XPRT_TRANSPORT_BC_TCP
,
3148 .setup
= xs_setup_bc_tcp
,
3152 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3155 int init_socket_xprt(void)
3157 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3158 if (!sunrpc_table_header
)
3159 sunrpc_table_header
= register_sysctl_table(sunrpc_table
);
3162 xprt_register_transport(&xs_local_transport
);
3163 xprt_register_transport(&xs_udp_transport
);
3164 xprt_register_transport(&xs_tcp_transport
);
3165 xprt_register_transport(&xs_bc_tcp_transport
);
3171 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3174 void cleanup_socket_xprt(void)
3176 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3177 if (sunrpc_table_header
) {
3178 unregister_sysctl_table(sunrpc_table_header
);
3179 sunrpc_table_header
= NULL
;
3183 xprt_unregister_transport(&xs_local_transport
);
3184 xprt_unregister_transport(&xs_udp_transport
);
3185 xprt_unregister_transport(&xs_tcp_transport
);
3186 xprt_unregister_transport(&xs_bc_tcp_transport
);
3189 static int param_set_uint_minmax(const char *val
,
3190 const struct kernel_param
*kp
,
3191 unsigned int min
, unsigned int max
)
3198 ret
= kstrtouint(val
, 0, &num
);
3199 if (ret
== -EINVAL
|| num
< min
|| num
> max
)
3201 *((unsigned int *)kp
->arg
) = num
;
3205 static int param_set_portnr(const char *val
, const struct kernel_param
*kp
)
3207 if (kp
->arg
== &xprt_min_resvport
)
3208 return param_set_uint_minmax(val
, kp
,
3211 return param_set_uint_minmax(val
, kp
,
3216 static const struct kernel_param_ops param_ops_portnr
= {
3217 .set
= param_set_portnr
,
3218 .get
= param_get_uint
,
3221 #define param_check_portnr(name, p) \
3222 __param_check(name, p, unsigned int);
3224 module_param_named(min_resvport
, xprt_min_resvport
, portnr
, 0644);
3225 module_param_named(max_resvport
, xprt_max_resvport
, portnr
, 0644);
3227 static int param_set_slot_table_size(const char *val
,
3228 const struct kernel_param
*kp
)
3230 return param_set_uint_minmax(val
, kp
,
3232 RPC_MAX_SLOT_TABLE
);
3235 static const struct kernel_param_ops param_ops_slot_table_size
= {
3236 .set
= param_set_slot_table_size
,
3237 .get
= param_get_uint
,
3240 #define param_check_slot_table_size(name, p) \
3241 __param_check(name, p, unsigned int);
3243 static int param_set_max_slot_table_size(const char *val
,
3244 const struct kernel_param
*kp
)
3246 return param_set_uint_minmax(val
, kp
,
3248 RPC_MAX_SLOT_TABLE_LIMIT
);
3251 static const struct kernel_param_ops param_ops_max_slot_table_size
= {
3252 .set
= param_set_max_slot_table_size
,
3253 .get
= param_get_uint
,
3256 #define param_check_max_slot_table_size(name, p) \
3257 __param_check(name, p, unsigned int);
3259 module_param_named(tcp_slot_table_entries
, xprt_tcp_slot_table_entries
,
3260 slot_table_size
, 0644);
3261 module_param_named(tcp_max_slot_table_entries
, xprt_max_tcp_slot_table_entries
,
3262 max_slot_table_size
, 0644);
3263 module_param_named(udp_slot_table_entries
, xprt_udp_slot_table_entries
,
3264 slot_table_size
, 0644);