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_limit
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_limit
,
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)
180 #define XS_TCP_MAX_REEST_TO (5U * 60 * HZ)
183 * TCP idle timeout; client drops the transport socket if it is idle
184 * for this long. Note that we also timeout UDP sockets to prevent
185 * holding port numbers when there is no RPC traffic.
187 #define XS_IDLE_DISC_TO (5U * 60 * HZ)
189 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
190 # undef RPC_DEBUG_DATA
191 # define RPCDBG_FACILITY RPCDBG_TRANS
194 #ifdef RPC_DEBUG_DATA
195 static void xs_pktdump(char *msg
, u32
*packet
, unsigned int count
)
197 u8
*buf
= (u8
*) packet
;
200 dprintk("RPC: %s\n", msg
);
201 for (j
= 0; j
< count
&& j
< 128; j
+= 4) {
205 dprintk("0x%04x ", j
);
207 dprintk("%02x%02x%02x%02x ",
208 buf
[j
], buf
[j
+1], buf
[j
+2], buf
[j
+3]);
213 static inline void xs_pktdump(char *msg
, u32
*packet
, unsigned int count
)
219 static inline struct rpc_xprt
*xprt_from_sock(struct sock
*sk
)
221 return (struct rpc_xprt
*) sk
->sk_user_data
;
224 static inline struct sockaddr
*xs_addr(struct rpc_xprt
*xprt
)
226 return (struct sockaddr
*) &xprt
->addr
;
229 static inline struct sockaddr_un
*xs_addr_un(struct rpc_xprt
*xprt
)
231 return (struct sockaddr_un
*) &xprt
->addr
;
234 static inline struct sockaddr_in
*xs_addr_in(struct rpc_xprt
*xprt
)
236 return (struct sockaddr_in
*) &xprt
->addr
;
239 static inline struct sockaddr_in6
*xs_addr_in6(struct rpc_xprt
*xprt
)
241 return (struct sockaddr_in6
*) &xprt
->addr
;
244 static void xs_format_common_peer_addresses(struct rpc_xprt
*xprt
)
246 struct sockaddr
*sap
= xs_addr(xprt
);
247 struct sockaddr_in6
*sin6
;
248 struct sockaddr_in
*sin
;
249 struct sockaddr_un
*sun
;
252 switch (sap
->sa_family
) {
254 sun
= xs_addr_un(xprt
);
255 strlcpy(buf
, sun
->sun_path
, sizeof(buf
));
256 xprt
->address_strings
[RPC_DISPLAY_ADDR
] =
257 kstrdup(buf
, GFP_KERNEL
);
260 (void)rpc_ntop(sap
, buf
, sizeof(buf
));
261 xprt
->address_strings
[RPC_DISPLAY_ADDR
] =
262 kstrdup(buf
, GFP_KERNEL
);
263 sin
= xs_addr_in(xprt
);
264 snprintf(buf
, sizeof(buf
), "%08x", ntohl(sin
->sin_addr
.s_addr
));
267 (void)rpc_ntop(sap
, buf
, sizeof(buf
));
268 xprt
->address_strings
[RPC_DISPLAY_ADDR
] =
269 kstrdup(buf
, GFP_KERNEL
);
270 sin6
= xs_addr_in6(xprt
);
271 snprintf(buf
, sizeof(buf
), "%pi6", &sin6
->sin6_addr
);
277 xprt
->address_strings
[RPC_DISPLAY_HEX_ADDR
] = kstrdup(buf
, GFP_KERNEL
);
280 static void xs_format_common_peer_ports(struct rpc_xprt
*xprt
)
282 struct sockaddr
*sap
= xs_addr(xprt
);
285 snprintf(buf
, sizeof(buf
), "%u", rpc_get_port(sap
));
286 xprt
->address_strings
[RPC_DISPLAY_PORT
] = kstrdup(buf
, GFP_KERNEL
);
288 snprintf(buf
, sizeof(buf
), "%4hx", rpc_get_port(sap
));
289 xprt
->address_strings
[RPC_DISPLAY_HEX_PORT
] = kstrdup(buf
, GFP_KERNEL
);
292 static void xs_format_peer_addresses(struct rpc_xprt
*xprt
,
293 const char *protocol
,
296 xprt
->address_strings
[RPC_DISPLAY_PROTO
] = protocol
;
297 xprt
->address_strings
[RPC_DISPLAY_NETID
] = netid
;
298 xs_format_common_peer_addresses(xprt
);
299 xs_format_common_peer_ports(xprt
);
302 static void xs_update_peer_port(struct rpc_xprt
*xprt
)
304 kfree(xprt
->address_strings
[RPC_DISPLAY_HEX_PORT
]);
305 kfree(xprt
->address_strings
[RPC_DISPLAY_PORT
]);
307 xs_format_common_peer_ports(xprt
);
310 static void xs_free_peer_addresses(struct rpc_xprt
*xprt
)
314 for (i
= 0; i
< RPC_DISPLAY_MAX
; i
++)
316 case RPC_DISPLAY_PROTO
:
317 case RPC_DISPLAY_NETID
:
320 kfree(xprt
->address_strings
[i
]);
324 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
326 static int xs_send_kvec(struct socket
*sock
, struct sockaddr
*addr
, int addrlen
, struct kvec
*vec
, unsigned int base
, int more
)
328 struct msghdr msg
= {
330 .msg_namelen
= addrlen
,
331 .msg_flags
= XS_SENDMSG_FLAGS
| (more
? MSG_MORE
: 0),
334 .iov_base
= vec
->iov_base
+ base
,
335 .iov_len
= vec
->iov_len
- base
,
338 if (iov
.iov_len
!= 0)
339 return kernel_sendmsg(sock
, &msg
, &iov
, 1, iov
.iov_len
);
340 return kernel_sendmsg(sock
, &msg
, NULL
, 0, 0);
343 static int xs_send_pagedata(struct socket
*sock
, struct xdr_buf
*xdr
, unsigned int base
, int more
, bool zerocopy
, int *sent_p
)
345 ssize_t (*do_sendpage
)(struct socket
*sock
, struct page
*page
,
346 int offset
, size_t size
, int flags
);
348 unsigned int remainder
;
351 remainder
= xdr
->page_len
- base
;
352 base
+= xdr
->page_base
;
353 ppage
= xdr
->pages
+ (base
>> PAGE_SHIFT
);
355 do_sendpage
= sock
->ops
->sendpage
;
357 do_sendpage
= sock_no_sendpage
;
359 unsigned int len
= min_t(unsigned int, PAGE_SIZE
- base
, remainder
);
360 int flags
= XS_SENDMSG_FLAGS
;
366 flags
|= MSG_SENDPAGE_NOTLAST
| MSG_MORE
;
367 err
= do_sendpage(sock
, *ppage
, base
, len
, flags
);
368 if (remainder
== 0 || err
!= len
)
382 * xs_sendpages - write pages directly to a socket
383 * @sock: socket to send on
384 * @addr: UDP only -- address of destination
385 * @addrlen: UDP only -- length of destination address
386 * @xdr: buffer containing this request
387 * @base: starting position in the buffer
388 * @zerocopy: true if it is safe to use sendpage()
389 * @sent_p: return the total number of bytes successfully queued for sending
392 static int xs_sendpages(struct socket
*sock
, struct sockaddr
*addr
, int addrlen
, struct xdr_buf
*xdr
, unsigned int base
, bool zerocopy
, int *sent_p
)
394 unsigned int remainder
= xdr
->len
- base
;
406 if (base
< xdr
->head
[0].iov_len
|| addr
!= NULL
) {
407 unsigned int len
= xdr
->head
[0].iov_len
- base
;
409 err
= xs_send_kvec(sock
, addr
, addrlen
, &xdr
->head
[0], base
, remainder
!= 0);
410 if (remainder
== 0 || err
!= len
)
415 base
-= xdr
->head
[0].iov_len
;
417 if (base
< xdr
->page_len
) {
418 unsigned int len
= xdr
->page_len
- base
;
420 err
= xs_send_pagedata(sock
, xdr
, base
, remainder
!= 0, zerocopy
, &sent
);
422 if (remainder
== 0 || sent
!= len
)
426 base
-= xdr
->page_len
;
428 if (base
>= xdr
->tail
[0].iov_len
)
430 err
= xs_send_kvec(sock
, NULL
, 0, &xdr
->tail
[0], base
, 0);
439 static void xs_nospace_callback(struct rpc_task
*task
)
441 struct sock_xprt
*transport
= container_of(task
->tk_rqstp
->rq_xprt
, struct sock_xprt
, xprt
);
443 transport
->inet
->sk_write_pending
--;
447 * xs_nospace - place task on wait queue if transmit was incomplete
448 * @task: task to put to sleep
451 static int xs_nospace(struct rpc_task
*task
)
453 struct rpc_rqst
*req
= task
->tk_rqstp
;
454 struct rpc_xprt
*xprt
= req
->rq_xprt
;
455 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
456 struct sock
*sk
= transport
->inet
;
459 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
460 task
->tk_pid
, req
->rq_slen
- req
->rq_bytes_sent
,
463 /* Protect against races with write_space */
464 spin_lock_bh(&xprt
->transport_lock
);
466 /* Don't race with disconnect */
467 if (xprt_connected(xprt
)) {
468 /* wait for more buffer space */
469 sk
->sk_write_pending
++;
470 xprt_wait_for_buffer_space(task
, xs_nospace_callback
);
474 spin_unlock_bh(&xprt
->transport_lock
);
476 /* Race breaker in case memory is freed before above code is called */
477 sk
->sk_write_space(sk
);
482 * Construct a stream transport record marker in @buf.
484 static inline void xs_encode_stream_record_marker(struct xdr_buf
*buf
)
486 u32 reclen
= buf
->len
- sizeof(rpc_fraghdr
);
487 rpc_fraghdr
*base
= buf
->head
[0].iov_base
;
488 *base
= cpu_to_be32(RPC_LAST_STREAM_FRAGMENT
| reclen
);
492 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
493 * @task: RPC task that manages the state of an RPC request
496 * 0: The request has been sent
497 * EAGAIN: The socket was blocked, please call again later to
498 * complete the request
499 * ENOTCONN: Caller needs to invoke connect logic then call again
500 * other: Some other error occured, the request was not sent
502 static int xs_local_send_request(struct rpc_task
*task
)
504 struct rpc_rqst
*req
= task
->tk_rqstp
;
505 struct rpc_xprt
*xprt
= req
->rq_xprt
;
506 struct sock_xprt
*transport
=
507 container_of(xprt
, struct sock_xprt
, xprt
);
508 struct xdr_buf
*xdr
= &req
->rq_snd_buf
;
512 xs_encode_stream_record_marker(&req
->rq_snd_buf
);
514 xs_pktdump("packet data:",
515 req
->rq_svec
->iov_base
, req
->rq_svec
->iov_len
);
517 status
= xs_sendpages(transport
->sock
, NULL
, 0, xdr
, req
->rq_bytes_sent
,
519 dprintk("RPC: %s(%u) = %d\n",
520 __func__
, xdr
->len
- req
->rq_bytes_sent
, status
);
522 if (status
== -EAGAIN
&& sock_writeable(transport
->inet
))
525 if (likely(sent
> 0) || status
== 0) {
526 req
->rq_bytes_sent
+= sent
;
527 req
->rq_xmit_bytes_sent
+= sent
;
528 if (likely(req
->rq_bytes_sent
>= req
->rq_slen
)) {
529 req
->rq_bytes_sent
= 0;
539 status
= xs_nospace(task
);
542 dprintk("RPC: sendmsg returned unrecognized error %d\n",
553 * xs_udp_send_request - write an RPC request to a UDP socket
554 * @task: address of RPC task that manages the state of an RPC request
557 * 0: The request has been sent
558 * EAGAIN: The socket was blocked, please call again later to
559 * complete the request
560 * ENOTCONN: Caller needs to invoke connect logic then call again
561 * other: Some other error occurred, the request was not sent
563 static int xs_udp_send_request(struct rpc_task
*task
)
565 struct rpc_rqst
*req
= task
->tk_rqstp
;
566 struct rpc_xprt
*xprt
= req
->rq_xprt
;
567 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
568 struct xdr_buf
*xdr
= &req
->rq_snd_buf
;
572 xs_pktdump("packet data:",
573 req
->rq_svec
->iov_base
,
574 req
->rq_svec
->iov_len
);
576 if (!xprt_bound(xprt
))
578 status
= xs_sendpages(transport
->sock
, xs_addr(xprt
), xprt
->addrlen
,
579 xdr
, req
->rq_bytes_sent
, true, &sent
);
581 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
582 xdr
->len
- req
->rq_bytes_sent
, status
);
584 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
585 if (status
== -EPERM
)
588 if (status
== -EAGAIN
&& sock_writeable(transport
->inet
))
591 if (sent
> 0 || status
== 0) {
592 req
->rq_xmit_bytes_sent
+= sent
;
593 if (sent
>= req
->rq_slen
)
595 /* Still some bytes left; set up for a retry later. */
603 /* Should we call xs_close() here? */
606 status
= xs_nospace(task
);
613 /* When the server has died, an ICMP port unreachable message
614 * prompts ECONNREFUSED. */
617 dprintk("RPC: sendmsg returned unrecognized error %d\n",
625 * xs_tcp_send_request - write an RPC request to a TCP socket
626 * @task: address of RPC task that manages the state of an RPC request
629 * 0: The request has been sent
630 * EAGAIN: The socket was blocked, please call again later to
631 * complete the request
632 * ENOTCONN: Caller needs to invoke connect logic then call again
633 * other: Some other error occurred, the request was not sent
635 * XXX: In the case of soft timeouts, should we eventually give up
636 * if sendmsg is not able to make progress?
638 static int xs_tcp_send_request(struct rpc_task
*task
)
640 struct rpc_rqst
*req
= task
->tk_rqstp
;
641 struct rpc_xprt
*xprt
= req
->rq_xprt
;
642 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
643 struct xdr_buf
*xdr
= &req
->rq_snd_buf
;
644 bool zerocopy
= true;
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;
687 if (status
== -EAGAIN
&& sk_stream_is_writeable(transport
->inet
))
693 /* Should we call xs_close() here? */
696 status
= xs_nospace(task
);
706 dprintk("RPC: sendmsg returned unrecognized error %d\n",
714 * xs_tcp_release_xprt - clean up after a tcp transmission
718 * This cleans up if an error causes us to abort the transmission of a request.
719 * In this case, the socket may need to be reset in order to avoid confusing
722 static void xs_tcp_release_xprt(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
724 struct rpc_rqst
*req
;
726 if (task
!= xprt
->snd_task
)
730 req
= task
->tk_rqstp
;
733 if (req
->rq_bytes_sent
== 0)
735 if (req
->rq_bytes_sent
== req
->rq_snd_buf
.len
)
737 set_bit(XPRT_CLOSE_WAIT
, &xprt
->state
);
739 xprt_release_xprt(xprt
, task
);
742 static void xs_save_old_callbacks(struct sock_xprt
*transport
, struct sock
*sk
)
744 transport
->old_data_ready
= sk
->sk_data_ready
;
745 transport
->old_state_change
= sk
->sk_state_change
;
746 transport
->old_write_space
= sk
->sk_write_space
;
747 transport
->old_error_report
= sk
->sk_error_report
;
750 static void xs_restore_old_callbacks(struct sock_xprt
*transport
, struct sock
*sk
)
752 sk
->sk_data_ready
= transport
->old_data_ready
;
753 sk
->sk_state_change
= transport
->old_state_change
;
754 sk
->sk_write_space
= transport
->old_write_space
;
755 sk
->sk_error_report
= transport
->old_error_report
;
758 static void xs_sock_reset_connection_flags(struct rpc_xprt
*xprt
)
760 smp_mb__before_atomic();
761 clear_bit(XPRT_CLOSE_WAIT
, &xprt
->state
);
762 clear_bit(XPRT_CLOSING
, &xprt
->state
);
763 smp_mb__after_atomic();
766 static void xs_sock_mark_closed(struct rpc_xprt
*xprt
)
768 xs_sock_reset_connection_flags(xprt
);
769 /* Mark transport as closed and wake up all pending tasks */
770 xprt_disconnect_done(xprt
);
774 * xs_error_report - callback to handle TCP socket state errors
777 * Note: we don't call sock_error() since there may be a rpc_task
778 * using the socket, and so we don't want to clear sk->sk_err.
780 static void xs_error_report(struct sock
*sk
)
782 struct rpc_xprt
*xprt
;
785 read_lock_bh(&sk
->sk_callback_lock
);
786 if (!(xprt
= xprt_from_sock(sk
)))
792 /* Is this a reset event? */
793 if (sk
->sk_state
== TCP_CLOSE
)
794 xs_sock_mark_closed(xprt
);
795 dprintk("RPC: xs_error_report client %p, error=%d...\n",
797 trace_rpc_socket_error(xprt
, sk
->sk_socket
, err
);
798 xprt_wake_pending_tasks(xprt
, err
);
800 read_unlock_bh(&sk
->sk_callback_lock
);
803 static void xs_reset_transport(struct sock_xprt
*transport
)
805 struct socket
*sock
= transport
->sock
;
806 struct sock
*sk
= transport
->inet
;
807 struct rpc_xprt
*xprt
= &transport
->xprt
;
812 if (atomic_read(&transport
->xprt
.swapper
))
813 sk_clear_memalloc(sk
);
815 kernel_sock_shutdown(sock
, SHUT_RDWR
);
817 mutex_lock(&transport
->recv_mutex
);
818 write_lock_bh(&sk
->sk_callback_lock
);
819 transport
->inet
= NULL
;
820 transport
->sock
= NULL
;
822 sk
->sk_user_data
= NULL
;
824 xs_restore_old_callbacks(transport
, sk
);
825 xprt_clear_connected(xprt
);
826 write_unlock_bh(&sk
->sk_callback_lock
);
827 xs_sock_reset_connection_flags(xprt
);
828 mutex_unlock(&transport
->recv_mutex
);
830 trace_rpc_socket_close(xprt
, sock
);
835 * xs_close - close a socket
838 * This is used when all requests are complete; ie, no DRC state remains
839 * on the server we want to save.
841 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
842 * xs_reset_transport() zeroing the socket from underneath a writer.
844 static void xs_close(struct rpc_xprt
*xprt
)
846 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
848 dprintk("RPC: xs_close xprt %p\n", xprt
);
850 xs_reset_transport(transport
);
851 xprt
->reestablish_timeout
= 0;
853 xprt_disconnect_done(xprt
);
856 static void xs_inject_disconnect(struct rpc_xprt
*xprt
)
858 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
860 xprt_disconnect_done(xprt
);
863 static void xs_xprt_free(struct rpc_xprt
*xprt
)
865 xs_free_peer_addresses(xprt
);
870 * xs_destroy - prepare to shutdown a transport
871 * @xprt: doomed transport
874 static void xs_destroy(struct rpc_xprt
*xprt
)
876 struct sock_xprt
*transport
= container_of(xprt
,
877 struct sock_xprt
, xprt
);
878 dprintk("RPC: xs_destroy xprt %p\n", xprt
);
880 cancel_delayed_work_sync(&transport
->connect_worker
);
882 cancel_work_sync(&transport
->recv_worker
);
884 module_put(THIS_MODULE
);
887 static int xs_local_copy_to_xdr(struct xdr_buf
*xdr
, struct sk_buff
*skb
)
889 struct xdr_skb_reader desc
= {
891 .offset
= sizeof(rpc_fraghdr
),
892 .count
= skb
->len
- sizeof(rpc_fraghdr
),
895 if (xdr_partial_copy_from_skb(xdr
, 0, &desc
, xdr_skb_read_bits
) < 0)
903 * xs_local_data_read_skb
908 * Currently this assumes we can read the whole reply in a single gulp.
910 static void xs_local_data_read_skb(struct rpc_xprt
*xprt
,
914 struct rpc_task
*task
;
915 struct rpc_rqst
*rovr
;
920 repsize
= skb
->len
- sizeof(rpc_fraghdr
);
922 dprintk("RPC: impossible RPC reply size %d\n", repsize
);
926 /* Copy the XID from the skb... */
927 xp
= skb_header_pointer(skb
, sizeof(rpc_fraghdr
), sizeof(_xid
), &_xid
);
931 /* Look up and lock the request corresponding to the given XID */
932 spin_lock_bh(&xprt
->transport_lock
);
933 rovr
= xprt_lookup_rqst(xprt
, *xp
);
936 task
= rovr
->rq_task
;
938 copied
= rovr
->rq_private_buf
.buflen
;
939 if (copied
> repsize
)
942 if (xs_local_copy_to_xdr(&rovr
->rq_private_buf
, skb
)) {
943 dprintk("RPC: sk_buff copy failed\n");
947 xprt_complete_rqst(task
, copied
);
950 spin_unlock_bh(&xprt
->transport_lock
);
953 static void xs_local_data_receive(struct sock_xprt
*transport
)
959 mutex_lock(&transport
->recv_mutex
);
960 sk
= transport
->inet
;
964 skb
= skb_recv_datagram(sk
, 0, 1, &err
);
967 xs_local_data_read_skb(&transport
->xprt
, sk
, skb
);
968 skb_free_datagram(sk
, skb
);
971 mutex_unlock(&transport
->recv_mutex
);
974 static void xs_local_data_receive_workfn(struct work_struct
*work
)
976 struct sock_xprt
*transport
=
977 container_of(work
, struct sock_xprt
, recv_worker
);
978 xs_local_data_receive(transport
);
982 * xs_udp_data_read_skb - receive callback for UDP sockets
988 static void xs_udp_data_read_skb(struct rpc_xprt
*xprt
,
992 struct rpc_task
*task
;
993 struct rpc_rqst
*rovr
;
1000 dprintk("RPC: impossible RPC reply size %d!\n", repsize
);
1004 /* Copy the XID from the skb... */
1005 xp
= skb_header_pointer(skb
, 0, sizeof(_xid
), &_xid
);
1009 /* Look up and lock the request corresponding to the given XID */
1010 spin_lock_bh(&xprt
->transport_lock
);
1011 rovr
= xprt_lookup_rqst(xprt
, *xp
);
1014 task
= rovr
->rq_task
;
1016 if ((copied
= rovr
->rq_private_buf
.buflen
) > repsize
)
1019 /* Suck it into the iovec, verify checksum if not done by hw. */
1020 if (csum_partial_copy_to_xdr(&rovr
->rq_private_buf
, skb
)) {
1021 __UDPX_INC_STATS(sk
, UDP_MIB_INERRORS
);
1025 __UDPX_INC_STATS(sk
, UDP_MIB_INDATAGRAMS
);
1027 xprt_adjust_cwnd(xprt
, task
, copied
);
1028 xprt_complete_rqst(task
, copied
);
1031 spin_unlock_bh(&xprt
->transport_lock
);
1034 static void xs_udp_data_receive(struct sock_xprt
*transport
)
1036 struct sk_buff
*skb
;
1040 mutex_lock(&transport
->recv_mutex
);
1041 sk
= transport
->inet
;
1045 skb
= skb_recv_datagram(sk
, 0, 1, &err
);
1048 xs_udp_data_read_skb(&transport
->xprt
, sk
, skb
);
1049 skb_free_datagram(sk
, skb
);
1052 mutex_unlock(&transport
->recv_mutex
);
1055 static void xs_udp_data_receive_workfn(struct work_struct
*work
)
1057 struct sock_xprt
*transport
=
1058 container_of(work
, struct sock_xprt
, recv_worker
);
1059 xs_udp_data_receive(transport
);
1063 * xs_data_ready - "data ready" callback for UDP sockets
1064 * @sk: socket with data to read
1067 static void xs_data_ready(struct sock
*sk
)
1069 struct rpc_xprt
*xprt
;
1071 read_lock_bh(&sk
->sk_callback_lock
);
1072 dprintk("RPC: xs_data_ready...\n");
1073 xprt
= xprt_from_sock(sk
);
1075 struct sock_xprt
*transport
= container_of(xprt
,
1076 struct sock_xprt
, xprt
);
1077 queue_work(rpciod_workqueue
, &transport
->recv_worker
);
1079 read_unlock_bh(&sk
->sk_callback_lock
);
1083 * Helper function to force a TCP close if the server is sending
1084 * junk and/or it has put us in CLOSE_WAIT
1086 static void xs_tcp_force_close(struct rpc_xprt
*xprt
)
1088 xprt_force_disconnect(xprt
);
1091 static inline void xs_tcp_read_fraghdr(struct rpc_xprt
*xprt
, struct xdr_skb_reader
*desc
)
1093 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1097 p
= ((char *) &transport
->tcp_fraghdr
) + transport
->tcp_offset
;
1098 len
= sizeof(transport
->tcp_fraghdr
) - transport
->tcp_offset
;
1099 used
= xdr_skb_read_bits(desc
, p
, len
);
1100 transport
->tcp_offset
+= used
;
1104 transport
->tcp_reclen
= ntohl(transport
->tcp_fraghdr
);
1105 if (transport
->tcp_reclen
& RPC_LAST_STREAM_FRAGMENT
)
1106 transport
->tcp_flags
|= TCP_RCV_LAST_FRAG
;
1108 transport
->tcp_flags
&= ~TCP_RCV_LAST_FRAG
;
1109 transport
->tcp_reclen
&= RPC_FRAGMENT_SIZE_MASK
;
1111 transport
->tcp_flags
&= ~TCP_RCV_COPY_FRAGHDR
;
1112 transport
->tcp_offset
= 0;
1114 /* Sanity check of the record length */
1115 if (unlikely(transport
->tcp_reclen
< 8)) {
1116 dprintk("RPC: invalid TCP record fragment length\n");
1117 xs_tcp_force_close(xprt
);
1120 dprintk("RPC: reading TCP record fragment of length %d\n",
1121 transport
->tcp_reclen
);
1124 static void xs_tcp_check_fraghdr(struct sock_xprt
*transport
)
1126 if (transport
->tcp_offset
== transport
->tcp_reclen
) {
1127 transport
->tcp_flags
|= TCP_RCV_COPY_FRAGHDR
;
1128 transport
->tcp_offset
= 0;
1129 if (transport
->tcp_flags
& TCP_RCV_LAST_FRAG
) {
1130 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
1131 transport
->tcp_flags
|= TCP_RCV_COPY_XID
;
1132 transport
->tcp_copied
= 0;
1137 static inline void xs_tcp_read_xid(struct sock_xprt
*transport
, struct xdr_skb_reader
*desc
)
1142 len
= sizeof(transport
->tcp_xid
) - transport
->tcp_offset
;
1143 dprintk("RPC: reading XID (%Zu bytes)\n", len
);
1144 p
= ((char *) &transport
->tcp_xid
) + transport
->tcp_offset
;
1145 used
= xdr_skb_read_bits(desc
, p
, len
);
1146 transport
->tcp_offset
+= used
;
1149 transport
->tcp_flags
&= ~TCP_RCV_COPY_XID
;
1150 transport
->tcp_flags
|= TCP_RCV_READ_CALLDIR
;
1151 transport
->tcp_copied
= 4;
1152 dprintk("RPC: reading %s XID %08x\n",
1153 (transport
->tcp_flags
& TCP_RPC_REPLY
) ? "reply for"
1155 ntohl(transport
->tcp_xid
));
1156 xs_tcp_check_fraghdr(transport
);
1159 static inline void xs_tcp_read_calldir(struct sock_xprt
*transport
,
1160 struct xdr_skb_reader
*desc
)
1167 * We want transport->tcp_offset to be 8 at the end of this routine
1168 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1169 * When this function is called for the first time,
1170 * transport->tcp_offset is 4 (after having already read the xid).
1172 offset
= transport
->tcp_offset
- sizeof(transport
->tcp_xid
);
1173 len
= sizeof(transport
->tcp_calldir
) - offset
;
1174 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len
);
1175 p
= ((char *) &transport
->tcp_calldir
) + offset
;
1176 used
= xdr_skb_read_bits(desc
, p
, len
);
1177 transport
->tcp_offset
+= used
;
1180 transport
->tcp_flags
&= ~TCP_RCV_READ_CALLDIR
;
1182 * We don't yet have the XDR buffer, so we will write the calldir
1183 * out after we get the buffer from the 'struct rpc_rqst'
1185 switch (ntohl(transport
->tcp_calldir
)) {
1187 transport
->tcp_flags
|= TCP_RCV_COPY_CALLDIR
;
1188 transport
->tcp_flags
|= TCP_RCV_COPY_DATA
;
1189 transport
->tcp_flags
|= TCP_RPC_REPLY
;
1192 transport
->tcp_flags
|= TCP_RCV_COPY_CALLDIR
;
1193 transport
->tcp_flags
|= TCP_RCV_COPY_DATA
;
1194 transport
->tcp_flags
&= ~TCP_RPC_REPLY
;
1197 dprintk("RPC: invalid request message type\n");
1198 xs_tcp_force_close(&transport
->xprt
);
1200 xs_tcp_check_fraghdr(transport
);
1203 static inline void xs_tcp_read_common(struct rpc_xprt
*xprt
,
1204 struct xdr_skb_reader
*desc
,
1205 struct rpc_rqst
*req
)
1207 struct sock_xprt
*transport
=
1208 container_of(xprt
, struct sock_xprt
, xprt
);
1209 struct xdr_buf
*rcvbuf
;
1213 rcvbuf
= &req
->rq_private_buf
;
1215 if (transport
->tcp_flags
& TCP_RCV_COPY_CALLDIR
) {
1217 * Save the RPC direction in the XDR buffer
1219 memcpy(rcvbuf
->head
[0].iov_base
+ transport
->tcp_copied
,
1220 &transport
->tcp_calldir
,
1221 sizeof(transport
->tcp_calldir
));
1222 transport
->tcp_copied
+= sizeof(transport
->tcp_calldir
);
1223 transport
->tcp_flags
&= ~TCP_RCV_COPY_CALLDIR
;
1227 if (len
> transport
->tcp_reclen
- transport
->tcp_offset
) {
1228 struct xdr_skb_reader my_desc
;
1230 len
= transport
->tcp_reclen
- transport
->tcp_offset
;
1231 memcpy(&my_desc
, desc
, sizeof(my_desc
));
1232 my_desc
.count
= len
;
1233 r
= xdr_partial_copy_from_skb(rcvbuf
, transport
->tcp_copied
,
1234 &my_desc
, xdr_skb_read_bits
);
1238 r
= xdr_partial_copy_from_skb(rcvbuf
, transport
->tcp_copied
,
1239 desc
, xdr_skb_read_bits
);
1242 transport
->tcp_copied
+= r
;
1243 transport
->tcp_offset
+= r
;
1246 /* Error when copying to the receive buffer,
1247 * usually because we weren't able to allocate
1248 * additional buffer pages. All we can do now
1249 * is turn off TCP_RCV_COPY_DATA, so the request
1250 * will not receive any additional updates,
1252 * Any remaining data from this record will
1255 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
1256 dprintk("RPC: XID %08x truncated request\n",
1257 ntohl(transport
->tcp_xid
));
1258 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1259 "tcp_offset = %u, tcp_reclen = %u\n",
1260 xprt
, transport
->tcp_copied
,
1261 transport
->tcp_offset
, transport
->tcp_reclen
);
1265 dprintk("RPC: XID %08x read %Zd bytes\n",
1266 ntohl(transport
->tcp_xid
), r
);
1267 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1268 "tcp_reclen = %u\n", xprt
, transport
->tcp_copied
,
1269 transport
->tcp_offset
, transport
->tcp_reclen
);
1271 if (transport
->tcp_copied
== req
->rq_private_buf
.buflen
)
1272 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
1273 else if (transport
->tcp_offset
== transport
->tcp_reclen
) {
1274 if (transport
->tcp_flags
& TCP_RCV_LAST_FRAG
)
1275 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
1280 * Finds the request corresponding to the RPC xid and invokes the common
1281 * tcp read code to read the data.
1283 static inline int xs_tcp_read_reply(struct rpc_xprt
*xprt
,
1284 struct xdr_skb_reader
*desc
)
1286 struct sock_xprt
*transport
=
1287 container_of(xprt
, struct sock_xprt
, xprt
);
1288 struct rpc_rqst
*req
;
1290 dprintk("RPC: read reply XID %08x\n", ntohl(transport
->tcp_xid
));
1292 /* Find and lock the request corresponding to this xid */
1293 spin_lock_bh(&xprt
->transport_lock
);
1294 req
= xprt_lookup_rqst(xprt
, transport
->tcp_xid
);
1296 dprintk("RPC: XID %08x request not found!\n",
1297 ntohl(transport
->tcp_xid
));
1298 spin_unlock_bh(&xprt
->transport_lock
);
1302 xs_tcp_read_common(xprt
, desc
, req
);
1304 if (!(transport
->tcp_flags
& TCP_RCV_COPY_DATA
))
1305 xprt_complete_rqst(req
->rq_task
, transport
->tcp_copied
);
1307 spin_unlock_bh(&xprt
->transport_lock
);
1311 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1313 * Obtains an rpc_rqst previously allocated and invokes the common
1314 * tcp read code to read the data. The result is placed in the callback
1316 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1317 * connection and return -1.
1319 static int xs_tcp_read_callback(struct rpc_xprt
*xprt
,
1320 struct xdr_skb_reader
*desc
)
1322 struct sock_xprt
*transport
=
1323 container_of(xprt
, struct sock_xprt
, xprt
);
1324 struct rpc_rqst
*req
;
1326 /* Look up and lock the request corresponding to the given XID */
1327 spin_lock_bh(&xprt
->transport_lock
);
1328 req
= xprt_lookup_bc_request(xprt
, transport
->tcp_xid
);
1330 spin_unlock_bh(&xprt
->transport_lock
);
1331 printk(KERN_WARNING
"Callback slot table overflowed\n");
1332 xprt_force_disconnect(xprt
);
1336 dprintk("RPC: read callback XID %08x\n", ntohl(req
->rq_xid
));
1337 xs_tcp_read_common(xprt
, desc
, req
);
1339 if (!(transport
->tcp_flags
& TCP_RCV_COPY_DATA
))
1340 xprt_complete_bc_request(req
, transport
->tcp_copied
);
1341 spin_unlock_bh(&xprt
->transport_lock
);
1346 static inline int _xs_tcp_read_data(struct rpc_xprt
*xprt
,
1347 struct xdr_skb_reader
*desc
)
1349 struct sock_xprt
*transport
=
1350 container_of(xprt
, struct sock_xprt
, xprt
);
1352 return (transport
->tcp_flags
& TCP_RPC_REPLY
) ?
1353 xs_tcp_read_reply(xprt
, desc
) :
1354 xs_tcp_read_callback(xprt
, desc
);
1357 static int xs_tcp_bc_up(struct svc_serv
*serv
, struct net
*net
)
1361 ret
= svc_create_xprt(serv
, "tcp-bc", net
, PF_INET
, 0,
1362 SVC_SOCK_ANONYMOUS
);
1368 static inline int _xs_tcp_read_data(struct rpc_xprt
*xprt
,
1369 struct xdr_skb_reader
*desc
)
1371 return xs_tcp_read_reply(xprt
, desc
);
1373 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1376 * Read data off the transport. This can be either an RPC_CALL or an
1377 * RPC_REPLY. Relay the processing to helper functions.
1379 static void xs_tcp_read_data(struct rpc_xprt
*xprt
,
1380 struct xdr_skb_reader
*desc
)
1382 struct sock_xprt
*transport
=
1383 container_of(xprt
, struct sock_xprt
, xprt
);
1385 if (_xs_tcp_read_data(xprt
, desc
) == 0)
1386 xs_tcp_check_fraghdr(transport
);
1389 * The transport_lock protects the request handling.
1390 * There's no need to hold it to update the tcp_flags.
1392 transport
->tcp_flags
&= ~TCP_RCV_COPY_DATA
;
1396 static inline void xs_tcp_read_discard(struct sock_xprt
*transport
, struct xdr_skb_reader
*desc
)
1400 len
= transport
->tcp_reclen
- transport
->tcp_offset
;
1401 if (len
> desc
->count
)
1404 desc
->offset
+= len
;
1405 transport
->tcp_offset
+= len
;
1406 dprintk("RPC: discarded %Zu bytes\n", len
);
1407 xs_tcp_check_fraghdr(transport
);
1410 static int xs_tcp_data_recv(read_descriptor_t
*rd_desc
, struct sk_buff
*skb
, unsigned int offset
, size_t len
)
1412 struct rpc_xprt
*xprt
= rd_desc
->arg
.data
;
1413 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1414 struct xdr_skb_reader desc
= {
1420 dprintk("RPC: xs_tcp_data_recv started\n");
1422 trace_xs_tcp_data_recv(transport
);
1423 /* Read in a new fragment marker if necessary */
1424 /* Can we ever really expect to get completely empty fragments? */
1425 if (transport
->tcp_flags
& TCP_RCV_COPY_FRAGHDR
) {
1426 xs_tcp_read_fraghdr(xprt
, &desc
);
1429 /* Read in the xid if necessary */
1430 if (transport
->tcp_flags
& TCP_RCV_COPY_XID
) {
1431 xs_tcp_read_xid(transport
, &desc
);
1434 /* Read in the call/reply flag */
1435 if (transport
->tcp_flags
& TCP_RCV_READ_CALLDIR
) {
1436 xs_tcp_read_calldir(transport
, &desc
);
1439 /* Read in the request data */
1440 if (transport
->tcp_flags
& TCP_RCV_COPY_DATA
) {
1441 xs_tcp_read_data(xprt
, &desc
);
1444 /* Skip over any trailing bytes on short reads */
1445 xs_tcp_read_discard(transport
, &desc
);
1446 } while (desc
.count
);
1447 trace_xs_tcp_data_recv(transport
);
1448 dprintk("RPC: xs_tcp_data_recv done\n");
1449 return len
- desc
.count
;
1452 static void xs_tcp_data_receive(struct sock_xprt
*transport
)
1454 struct rpc_xprt
*xprt
= &transport
->xprt
;
1456 read_descriptor_t rd_desc
= {
1457 .count
= 2*1024*1024,
1460 unsigned long total
= 0;
1463 mutex_lock(&transport
->recv_mutex
);
1464 sk
= transport
->inet
;
1468 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1471 read
= tcp_read_sock(sk
, &rd_desc
, xs_tcp_data_recv
);
1476 rd_desc
.count
= 65536;
1479 mutex_unlock(&transport
->recv_mutex
);
1480 trace_xs_tcp_data_ready(xprt
, read
, total
);
1483 static void xs_tcp_data_receive_workfn(struct work_struct
*work
)
1485 struct sock_xprt
*transport
=
1486 container_of(work
, struct sock_xprt
, recv_worker
);
1487 xs_tcp_data_receive(transport
);
1491 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1492 * @sk: socket with data to read
1495 static void xs_tcp_data_ready(struct sock
*sk
)
1497 struct sock_xprt
*transport
;
1498 struct rpc_xprt
*xprt
;
1500 dprintk("RPC: xs_tcp_data_ready...\n");
1502 read_lock_bh(&sk
->sk_callback_lock
);
1503 if (!(xprt
= xprt_from_sock(sk
)))
1505 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1507 /* Any data means we had a useful conversation, so
1508 * the we don't need to delay the next reconnect
1510 if (xprt
->reestablish_timeout
)
1511 xprt
->reestablish_timeout
= 0;
1512 queue_work(rpciod_workqueue
, &transport
->recv_worker
);
1515 read_unlock_bh(&sk
->sk_callback_lock
);
1519 * xs_tcp_state_change - callback to handle TCP socket state changes
1520 * @sk: socket whose state has changed
1523 static void xs_tcp_state_change(struct sock
*sk
)
1525 struct rpc_xprt
*xprt
;
1526 struct sock_xprt
*transport
;
1528 read_lock_bh(&sk
->sk_callback_lock
);
1529 if (!(xprt
= xprt_from_sock(sk
)))
1531 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt
);
1532 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1533 sk
->sk_state
, xprt_connected(xprt
),
1534 sock_flag(sk
, SOCK_DEAD
),
1535 sock_flag(sk
, SOCK_ZAPPED
),
1538 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1539 trace_rpc_socket_state_change(xprt
, sk
->sk_socket
);
1540 switch (sk
->sk_state
) {
1541 case TCP_ESTABLISHED
:
1542 spin_lock(&xprt
->transport_lock
);
1543 if (!xprt_test_and_set_connected(xprt
)) {
1545 /* Reset TCP record info */
1546 transport
->tcp_offset
= 0;
1547 transport
->tcp_reclen
= 0;
1548 transport
->tcp_copied
= 0;
1549 transport
->tcp_flags
=
1550 TCP_RCV_COPY_FRAGHDR
| TCP_RCV_COPY_XID
;
1551 xprt
->connect_cookie
++;
1552 clear_bit(XPRT_SOCK_CONNECTING
, &transport
->sock_state
);
1553 xprt_clear_connecting(xprt
);
1555 xprt_wake_pending_tasks(xprt
, -EAGAIN
);
1557 spin_unlock(&xprt
->transport_lock
);
1560 /* The client initiated a shutdown of the socket */
1561 xprt
->connect_cookie
++;
1562 xprt
->reestablish_timeout
= 0;
1563 set_bit(XPRT_CLOSING
, &xprt
->state
);
1564 smp_mb__before_atomic();
1565 clear_bit(XPRT_CONNECTED
, &xprt
->state
);
1566 clear_bit(XPRT_CLOSE_WAIT
, &xprt
->state
);
1567 smp_mb__after_atomic();
1569 case TCP_CLOSE_WAIT
:
1570 /* The server initiated a shutdown of the socket */
1571 xprt
->connect_cookie
++;
1572 clear_bit(XPRT_CONNECTED
, &xprt
->state
);
1573 xs_tcp_force_close(xprt
);
1576 * If the server closed down the connection, make sure that
1577 * we back off before reconnecting
1579 if (xprt
->reestablish_timeout
< XS_TCP_INIT_REEST_TO
)
1580 xprt
->reestablish_timeout
= XS_TCP_INIT_REEST_TO
;
1583 set_bit(XPRT_CLOSING
, &xprt
->state
);
1584 smp_mb__before_atomic();
1585 clear_bit(XPRT_CONNECTED
, &xprt
->state
);
1586 smp_mb__after_atomic();
1589 if (test_and_clear_bit(XPRT_SOCK_CONNECTING
,
1590 &transport
->sock_state
))
1591 xprt_clear_connecting(xprt
);
1592 xs_sock_mark_closed(xprt
);
1595 read_unlock_bh(&sk
->sk_callback_lock
);
1598 static void xs_write_space(struct sock
*sk
)
1600 struct socket_wq
*wq
;
1601 struct rpc_xprt
*xprt
;
1605 clear_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1607 if (unlikely(!(xprt
= xprt_from_sock(sk
))))
1610 wq
= rcu_dereference(sk
->sk_wq
);
1611 if (!wq
|| test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE
, &wq
->flags
) == 0)
1614 xprt_write_space(xprt
);
1620 * xs_udp_write_space - callback invoked when socket buffer space
1622 * @sk: socket whose state has changed
1624 * Called when more output buffer space is available for this socket.
1625 * We try not to wake our writers until they can make "significant"
1626 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1627 * with a bunch of small requests.
1629 static void xs_udp_write_space(struct sock
*sk
)
1631 read_lock_bh(&sk
->sk_callback_lock
);
1633 /* from net/core/sock.c:sock_def_write_space */
1634 if (sock_writeable(sk
))
1637 read_unlock_bh(&sk
->sk_callback_lock
);
1641 * xs_tcp_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_tcp_write_space(struct sock
*sk
)
1652 read_lock_bh(&sk
->sk_callback_lock
);
1654 /* from net/core/stream.c:sk_stream_write_space */
1655 if (sk_stream_is_writeable(sk
))
1658 read_unlock_bh(&sk
->sk_callback_lock
);
1661 static void xs_udp_do_set_buffer_size(struct rpc_xprt
*xprt
)
1663 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1664 struct sock
*sk
= transport
->inet
;
1666 if (transport
->rcvsize
) {
1667 sk
->sk_userlocks
|= SOCK_RCVBUF_LOCK
;
1668 sk
->sk_rcvbuf
= transport
->rcvsize
* xprt
->max_reqs
* 2;
1670 if (transport
->sndsize
) {
1671 sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
;
1672 sk
->sk_sndbuf
= transport
->sndsize
* xprt
->max_reqs
* 2;
1673 sk
->sk_write_space(sk
);
1678 * xs_udp_set_buffer_size - set send and receive limits
1679 * @xprt: generic transport
1680 * @sndsize: requested size of send buffer, in bytes
1681 * @rcvsize: requested size of receive buffer, in bytes
1683 * Set socket send and receive buffer size limits.
1685 static void xs_udp_set_buffer_size(struct rpc_xprt
*xprt
, size_t sndsize
, size_t rcvsize
)
1687 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1689 transport
->sndsize
= 0;
1691 transport
->sndsize
= sndsize
+ 1024;
1692 transport
->rcvsize
= 0;
1694 transport
->rcvsize
= rcvsize
+ 1024;
1696 xs_udp_do_set_buffer_size(xprt
);
1700 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1701 * @task: task that timed out
1703 * Adjust the congestion window after a retransmit timeout has occurred.
1705 static void xs_udp_timer(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
1707 xprt_adjust_cwnd(xprt
, task
, -ETIMEDOUT
);
1710 static unsigned short xs_get_random_port(void)
1712 unsigned short range
= xprt_max_resvport
- xprt_min_resvport
;
1713 unsigned short rand
= (unsigned short) prandom_u32() % range
;
1714 return rand
+ xprt_min_resvport
;
1718 * xs_set_reuseaddr_port - set the socket's port and address reuse options
1721 * Note that this function has to be called on all sockets that share the
1722 * same port, and it must be called before binding.
1724 static void xs_sock_set_reuseport(struct socket
*sock
)
1728 kernel_setsockopt(sock
, SOL_SOCKET
, SO_REUSEPORT
,
1729 (char *)&opt
, sizeof(opt
));
1732 static unsigned short xs_sock_getport(struct socket
*sock
)
1734 struct sockaddr_storage buf
;
1736 unsigned short port
= 0;
1738 if (kernel_getsockname(sock
, (struct sockaddr
*)&buf
, &buflen
) < 0)
1740 switch (buf
.ss_family
) {
1742 port
= ntohs(((struct sockaddr_in6
*)&buf
)->sin6_port
);
1745 port
= ntohs(((struct sockaddr_in
*)&buf
)->sin_port
);
1752 * xs_set_port - reset the port number in the remote endpoint address
1753 * @xprt: generic transport
1754 * @port: new port number
1757 static void xs_set_port(struct rpc_xprt
*xprt
, unsigned short port
)
1759 dprintk("RPC: setting port for xprt %p to %u\n", xprt
, port
);
1761 rpc_set_port(xs_addr(xprt
), port
);
1762 xs_update_peer_port(xprt
);
1765 static void xs_set_srcport(struct sock_xprt
*transport
, struct socket
*sock
)
1767 if (transport
->srcport
== 0)
1768 transport
->srcport
= xs_sock_getport(sock
);
1771 static unsigned short xs_get_srcport(struct sock_xprt
*transport
)
1773 unsigned short port
= transport
->srcport
;
1775 if (port
== 0 && transport
->xprt
.resvport
)
1776 port
= xs_get_random_port();
1780 static unsigned short xs_next_srcport(struct sock_xprt
*transport
, unsigned short port
)
1782 if (transport
->srcport
!= 0)
1783 transport
->srcport
= 0;
1784 if (!transport
->xprt
.resvport
)
1786 if (port
<= xprt_min_resvport
|| port
> xprt_max_resvport
)
1787 return xprt_max_resvport
;
1790 static int xs_bind(struct sock_xprt
*transport
, struct socket
*sock
)
1792 struct sockaddr_storage myaddr
;
1794 unsigned short port
= xs_get_srcport(transport
);
1795 unsigned short last
;
1798 * If we are asking for any ephemeral port (i.e. port == 0 &&
1799 * transport->xprt.resvport == 0), don't bind. Let the local
1800 * port selection happen implicitly when the socket is used
1801 * (for example at connect time).
1803 * This ensures that we can continue to establish TCP
1804 * connections even when all local ephemeral ports are already
1805 * a part of some TCP connection. This makes no difference
1806 * for UDP sockets, but also doens't harm them.
1808 * If we're asking for any reserved port (i.e. port == 0 &&
1809 * transport->xprt.resvport == 1) xs_get_srcport above will
1810 * ensure that port is non-zero and we will bind as needed.
1815 memcpy(&myaddr
, &transport
->srcaddr
, transport
->xprt
.addrlen
);
1817 rpc_set_port((struct sockaddr
*)&myaddr
, port
);
1818 err
= kernel_bind(sock
, (struct sockaddr
*)&myaddr
,
1819 transport
->xprt
.addrlen
);
1821 transport
->srcport
= port
;
1825 port
= xs_next_srcport(transport
, port
);
1828 } while (err
== -EADDRINUSE
&& nloop
!= 2);
1830 if (myaddr
.ss_family
== AF_INET
)
1831 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__
,
1832 &((struct sockaddr_in
*)&myaddr
)->sin_addr
,
1833 port
, err
? "failed" : "ok", err
);
1835 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__
,
1836 &((struct sockaddr_in6
*)&myaddr
)->sin6_addr
,
1837 port
, err
? "failed" : "ok", err
);
1842 * We don't support autobind on AF_LOCAL sockets
1844 static void xs_local_rpcbind(struct rpc_task
*task
)
1846 xprt_set_bound(task
->tk_xprt
);
1849 static void xs_local_set_port(struct rpc_xprt
*xprt
, unsigned short port
)
1853 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1854 static struct lock_class_key xs_key
[2];
1855 static struct lock_class_key xs_slock_key
[2];
1857 static inline void xs_reclassify_socketu(struct socket
*sock
)
1859 struct sock
*sk
= sock
->sk
;
1861 sock_lock_init_class_and_name(sk
, "slock-AF_LOCAL-RPC",
1862 &xs_slock_key
[1], "sk_lock-AF_LOCAL-RPC", &xs_key
[1]);
1865 static inline void xs_reclassify_socket4(struct socket
*sock
)
1867 struct sock
*sk
= sock
->sk
;
1869 sock_lock_init_class_and_name(sk
, "slock-AF_INET-RPC",
1870 &xs_slock_key
[0], "sk_lock-AF_INET-RPC", &xs_key
[0]);
1873 static inline void xs_reclassify_socket6(struct socket
*sock
)
1875 struct sock
*sk
= sock
->sk
;
1877 sock_lock_init_class_and_name(sk
, "slock-AF_INET6-RPC",
1878 &xs_slock_key
[1], "sk_lock-AF_INET6-RPC", &xs_key
[1]);
1881 static inline void xs_reclassify_socket(int family
, struct socket
*sock
)
1883 if (WARN_ON_ONCE(!sock_allow_reclassification(sock
->sk
)))
1888 xs_reclassify_socketu(sock
);
1891 xs_reclassify_socket4(sock
);
1894 xs_reclassify_socket6(sock
);
1899 static inline void xs_reclassify_socket(int family
, struct socket
*sock
)
1904 static void xs_dummy_setup_socket(struct work_struct
*work
)
1908 static struct socket
*xs_create_sock(struct rpc_xprt
*xprt
,
1909 struct sock_xprt
*transport
, int family
, int type
,
1910 int protocol
, bool reuseport
)
1912 struct socket
*sock
;
1915 err
= __sock_create(xprt
->xprt_net
, family
, type
, protocol
, &sock
, 1);
1917 dprintk("RPC: can't create %d transport socket (%d).\n",
1921 xs_reclassify_socket(family
, sock
);
1924 xs_sock_set_reuseport(sock
);
1926 err
= xs_bind(transport
, sock
);
1934 return ERR_PTR(err
);
1937 static int xs_local_finish_connecting(struct rpc_xprt
*xprt
,
1938 struct socket
*sock
)
1940 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
,
1943 if (!transport
->inet
) {
1944 struct sock
*sk
= sock
->sk
;
1946 write_lock_bh(&sk
->sk_callback_lock
);
1948 xs_save_old_callbacks(transport
, sk
);
1950 sk
->sk_user_data
= xprt
;
1951 sk
->sk_data_ready
= xs_data_ready
;
1952 sk
->sk_write_space
= xs_udp_write_space
;
1953 sock_set_flag(sk
, SOCK_FASYNC
);
1954 sk
->sk_error_report
= xs_error_report
;
1955 sk
->sk_allocation
= GFP_NOIO
;
1957 xprt_clear_connected(xprt
);
1959 /* Reset to new socket */
1960 transport
->sock
= sock
;
1961 transport
->inet
= sk
;
1963 write_unlock_bh(&sk
->sk_callback_lock
);
1966 /* Tell the socket layer to start connecting... */
1967 xprt
->stat
.connect_count
++;
1968 xprt
->stat
.connect_start
= jiffies
;
1969 return kernel_connect(sock
, xs_addr(xprt
), xprt
->addrlen
, 0);
1973 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1974 * @transport: socket transport to connect
1976 static int xs_local_setup_socket(struct sock_xprt
*transport
)
1978 struct rpc_xprt
*xprt
= &transport
->xprt
;
1979 struct socket
*sock
;
1982 status
= __sock_create(xprt
->xprt_net
, AF_LOCAL
,
1983 SOCK_STREAM
, 0, &sock
, 1);
1985 dprintk("RPC: can't create AF_LOCAL "
1986 "transport socket (%d).\n", -status
);
1989 xs_reclassify_socket(AF_LOCAL
, sock
);
1991 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1992 xprt
, xprt
->address_strings
[RPC_DISPLAY_ADDR
]);
1994 status
= xs_local_finish_connecting(xprt
, sock
);
1995 trace_rpc_socket_connect(xprt
, sock
, status
);
1998 dprintk("RPC: xprt %p connected to %s\n",
1999 xprt
, xprt
->address_strings
[RPC_DISPLAY_ADDR
]);
2000 xprt_set_connected(xprt
);
2004 dprintk("RPC: xprt %p: socket %s does not exist\n",
2005 xprt
, xprt
->address_strings
[RPC_DISPLAY_ADDR
]);
2008 dprintk("RPC: xprt %p: connection refused for %s\n",
2009 xprt
, xprt
->address_strings
[RPC_DISPLAY_ADDR
]);
2012 printk(KERN_ERR
"%s: unhandled error (%d) connecting to %s\n",
2014 xprt
->address_strings
[RPC_DISPLAY_ADDR
]);
2018 xprt_clear_connecting(xprt
);
2019 xprt_wake_pending_tasks(xprt
, status
);
2023 static void xs_local_connect(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
2025 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2028 if (RPC_IS_ASYNC(task
)) {
2030 * We want the AF_LOCAL connect to be resolved in the
2031 * filesystem namespace of the process making the rpc
2032 * call. Thus we connect synchronously.
2034 * If we want to support asynchronous AF_LOCAL calls,
2035 * we'll need to figure out how to pass a namespace to
2038 rpc_exit(task
, -ENOTCONN
);
2041 ret
= xs_local_setup_socket(transport
);
2042 if (ret
&& !RPC_IS_SOFTCONN(task
))
2043 msleep_interruptible(15000);
2046 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2048 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2049 * know that we have exclusive access to the socket), to guard against
2050 * races with xs_reset_transport.
2052 static void xs_set_memalloc(struct rpc_xprt
*xprt
)
2054 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
,
2058 * If there's no sock, then we have nothing to set. The
2059 * reconnecting process will get it for us.
2061 if (!transport
->inet
)
2063 if (atomic_read(&xprt
->swapper
))
2064 sk_set_memalloc(transport
->inet
);
2068 * xs_enable_swap - Tag this transport as being used for swap.
2069 * @xprt: transport to tag
2071 * Take a reference to this transport on behalf of the rpc_clnt, and
2072 * optionally mark it for swapping if it wasn't already.
2075 xs_enable_swap(struct rpc_xprt
*xprt
)
2077 struct sock_xprt
*xs
= container_of(xprt
, struct sock_xprt
, xprt
);
2079 if (atomic_inc_return(&xprt
->swapper
) != 1)
2081 if (wait_on_bit_lock(&xprt
->state
, XPRT_LOCKED
, TASK_KILLABLE
))
2082 return -ERESTARTSYS
;
2084 sk_set_memalloc(xs
->inet
);
2085 xprt_release_xprt(xprt
, NULL
);
2090 * xs_disable_swap - Untag this transport as being used for swap.
2091 * @xprt: transport to tag
2093 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2094 * swapper refcount goes to 0, untag the socket as a memalloc socket.
2097 xs_disable_swap(struct rpc_xprt
*xprt
)
2099 struct sock_xprt
*xs
= container_of(xprt
, struct sock_xprt
, xprt
);
2101 if (!atomic_dec_and_test(&xprt
->swapper
))
2103 if (wait_on_bit_lock(&xprt
->state
, XPRT_LOCKED
, TASK_KILLABLE
))
2106 sk_clear_memalloc(xs
->inet
);
2107 xprt_release_xprt(xprt
, NULL
);
2110 static void xs_set_memalloc(struct rpc_xprt
*xprt
)
2115 xs_enable_swap(struct rpc_xprt
*xprt
)
2121 xs_disable_swap(struct rpc_xprt
*xprt
)
2126 static void xs_udp_finish_connecting(struct rpc_xprt
*xprt
, struct socket
*sock
)
2128 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2130 if (!transport
->inet
) {
2131 struct sock
*sk
= sock
->sk
;
2133 write_lock_bh(&sk
->sk_callback_lock
);
2135 xs_save_old_callbacks(transport
, sk
);
2137 sk
->sk_user_data
= xprt
;
2138 sk
->sk_data_ready
= xs_data_ready
;
2139 sk
->sk_write_space
= xs_udp_write_space
;
2140 sock_set_flag(sk
, SOCK_FASYNC
);
2141 sk
->sk_allocation
= GFP_NOIO
;
2143 xprt_set_connected(xprt
);
2145 /* Reset to new socket */
2146 transport
->sock
= sock
;
2147 transport
->inet
= sk
;
2149 xs_set_memalloc(xprt
);
2151 write_unlock_bh(&sk
->sk_callback_lock
);
2153 xs_udp_do_set_buffer_size(xprt
);
2156 static void xs_udp_setup_socket(struct work_struct
*work
)
2158 struct sock_xprt
*transport
=
2159 container_of(work
, struct sock_xprt
, connect_worker
.work
);
2160 struct rpc_xprt
*xprt
= &transport
->xprt
;
2161 struct socket
*sock
= transport
->sock
;
2164 sock
= xs_create_sock(xprt
, transport
,
2165 xs_addr(xprt
)->sa_family
, SOCK_DGRAM
,
2166 IPPROTO_UDP
, false);
2170 dprintk("RPC: worker connecting xprt %p via %s to "
2171 "%s (port %s)\n", xprt
,
2172 xprt
->address_strings
[RPC_DISPLAY_PROTO
],
2173 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
2174 xprt
->address_strings
[RPC_DISPLAY_PORT
]);
2176 xs_udp_finish_connecting(xprt
, sock
);
2177 trace_rpc_socket_connect(xprt
, sock
, 0);
2180 xprt_unlock_connect(xprt
, transport
);
2181 xprt_clear_connecting(xprt
);
2182 xprt_wake_pending_tasks(xprt
, status
);
2186 * xs_tcp_shutdown - gracefully shut down a TCP socket
2189 * Initiates a graceful shutdown of the TCP socket by calling the
2190 * equivalent of shutdown(SHUT_RDWR);
2192 static void xs_tcp_shutdown(struct rpc_xprt
*xprt
)
2194 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2195 struct socket
*sock
= transport
->sock
;
2199 if (xprt_connected(xprt
)) {
2200 kernel_sock_shutdown(sock
, SHUT_RDWR
);
2201 trace_rpc_socket_shutdown(xprt
, sock
);
2203 xs_reset_transport(transport
);
2206 static int xs_tcp_finish_connecting(struct rpc_xprt
*xprt
, struct socket
*sock
)
2208 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2209 int ret
= -ENOTCONN
;
2211 if (!transport
->inet
) {
2212 struct sock
*sk
= sock
->sk
;
2213 unsigned int keepidle
= xprt
->timeout
->to_initval
/ HZ
;
2214 unsigned int keepcnt
= xprt
->timeout
->to_retries
+ 1;
2215 unsigned int opt_on
= 1;
2218 /* TCP Keepalive options */
2219 kernel_setsockopt(sock
, SOL_SOCKET
, SO_KEEPALIVE
,
2220 (char *)&opt_on
, sizeof(opt_on
));
2221 kernel_setsockopt(sock
, SOL_TCP
, TCP_KEEPIDLE
,
2222 (char *)&keepidle
, sizeof(keepidle
));
2223 kernel_setsockopt(sock
, SOL_TCP
, TCP_KEEPINTVL
,
2224 (char *)&keepidle
, sizeof(keepidle
));
2225 kernel_setsockopt(sock
, SOL_TCP
, TCP_KEEPCNT
,
2226 (char *)&keepcnt
, sizeof(keepcnt
));
2228 /* TCP user timeout (see RFC5482) */
2229 timeo
= jiffies_to_msecs(xprt
->timeout
->to_initval
) *
2230 (xprt
->timeout
->to_retries
+ 1);
2231 kernel_setsockopt(sock
, SOL_TCP
, TCP_USER_TIMEOUT
,
2232 (char *)&timeo
, sizeof(timeo
));
2234 write_lock_bh(&sk
->sk_callback_lock
);
2236 xs_save_old_callbacks(transport
, sk
);
2238 sk
->sk_user_data
= xprt
;
2239 sk
->sk_data_ready
= xs_tcp_data_ready
;
2240 sk
->sk_state_change
= xs_tcp_state_change
;
2241 sk
->sk_write_space
= xs_tcp_write_space
;
2242 sock_set_flag(sk
, SOCK_FASYNC
);
2243 sk
->sk_error_report
= xs_error_report
;
2244 sk
->sk_allocation
= GFP_NOIO
;
2246 /* socket options */
2247 sock_reset_flag(sk
, SOCK_LINGER
);
2248 tcp_sk(sk
)->nonagle
|= TCP_NAGLE_OFF
;
2250 xprt_clear_connected(xprt
);
2252 /* Reset to new socket */
2253 transport
->sock
= sock
;
2254 transport
->inet
= sk
;
2256 write_unlock_bh(&sk
->sk_callback_lock
);
2259 if (!xprt_bound(xprt
))
2262 xs_set_memalloc(xprt
);
2264 /* Tell the socket layer to start connecting... */
2265 xprt
->stat
.connect_count
++;
2266 xprt
->stat
.connect_start
= jiffies
;
2267 set_bit(XPRT_SOCK_CONNECTING
, &transport
->sock_state
);
2268 ret
= kernel_connect(sock
, xs_addr(xprt
), xprt
->addrlen
, O_NONBLOCK
);
2271 xs_set_srcport(transport
, sock
);
2274 if (xprt
->reestablish_timeout
< XS_TCP_INIT_REEST_TO
)
2275 xprt
->reestablish_timeout
= XS_TCP_INIT_REEST_TO
;
2282 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2284 * Invoked by a work queue tasklet.
2286 static void xs_tcp_setup_socket(struct work_struct
*work
)
2288 struct sock_xprt
*transport
=
2289 container_of(work
, struct sock_xprt
, connect_worker
.work
);
2290 struct socket
*sock
= transport
->sock
;
2291 struct rpc_xprt
*xprt
= &transport
->xprt
;
2295 sock
= xs_create_sock(xprt
, transport
,
2296 xs_addr(xprt
)->sa_family
, SOCK_STREAM
,
2299 status
= PTR_ERR(sock
);
2304 dprintk("RPC: worker connecting xprt %p via %s to "
2305 "%s (port %s)\n", xprt
,
2306 xprt
->address_strings
[RPC_DISPLAY_PROTO
],
2307 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
2308 xprt
->address_strings
[RPC_DISPLAY_PORT
]);
2310 status
= xs_tcp_finish_connecting(xprt
, sock
);
2311 trace_rpc_socket_connect(xprt
, sock
, status
);
2312 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2313 xprt
, -status
, xprt_connected(xprt
),
2314 sock
->sk
->sk_state
);
2317 printk("%s: connect returned unhandled error %d\n",
2319 case -EADDRNOTAVAIL
:
2320 /* We're probably in TIME_WAIT. Get rid of existing socket,
2323 xs_tcp_force_close(xprt
);
2328 xprt_unlock_connect(xprt
, transport
);
2331 /* Happens, for instance, if the user specified a link
2332 * local IPv6 address without a scope-id.
2339 /* retry with existing socket, after a delay */
2340 xs_tcp_force_close(xprt
);
2345 xprt_unlock_connect(xprt
, transport
);
2346 xprt_clear_connecting(xprt
);
2347 xprt_wake_pending_tasks(xprt
, status
);
2351 * xs_connect - connect a socket to a remote endpoint
2352 * @xprt: pointer to transport structure
2353 * @task: address of RPC task that manages state of connect request
2355 * TCP: If the remote end dropped the connection, delay reconnecting.
2357 * UDP socket connects are synchronous, but we use a work queue anyway
2358 * to guarantee that even unprivileged user processes can set up a
2359 * socket on a privileged port.
2361 * If a UDP socket connect fails, the delay behavior here prevents
2362 * retry floods (hard mounts).
2364 static void xs_connect(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
2366 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2368 WARN_ON_ONCE(!xprt_lock_connect(xprt
, task
, transport
));
2370 if (transport
->sock
!= NULL
) {
2371 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2373 xprt
, xprt
->reestablish_timeout
/ HZ
);
2375 /* Start by resetting any existing state */
2376 xs_reset_transport(transport
);
2378 queue_delayed_work(rpciod_workqueue
,
2379 &transport
->connect_worker
,
2380 xprt
->reestablish_timeout
);
2381 xprt
->reestablish_timeout
<<= 1;
2382 if (xprt
->reestablish_timeout
< XS_TCP_INIT_REEST_TO
)
2383 xprt
->reestablish_timeout
= XS_TCP_INIT_REEST_TO
;
2384 if (xprt
->reestablish_timeout
> XS_TCP_MAX_REEST_TO
)
2385 xprt
->reestablish_timeout
= XS_TCP_MAX_REEST_TO
;
2387 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt
);
2388 queue_delayed_work(rpciod_workqueue
,
2389 &transport
->connect_worker
, 0);
2394 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2395 * @xprt: rpc_xprt struct containing statistics
2399 static void xs_local_print_stats(struct rpc_xprt
*xprt
, struct seq_file
*seq
)
2403 if (xprt_connected(xprt
))
2404 idle_time
= (long)(jiffies
- xprt
->last_used
) / HZ
;
2406 seq_printf(seq
, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2407 "%llu %llu %lu %llu %llu\n",
2408 xprt
->stat
.bind_count
,
2409 xprt
->stat
.connect_count
,
2410 xprt
->stat
.connect_time
,
2414 xprt
->stat
.bad_xids
,
2417 xprt
->stat
.max_slots
,
2418 xprt
->stat
.sending_u
,
2419 xprt
->stat
.pending_u
);
2423 * xs_udp_print_stats - display UDP socket-specifc stats
2424 * @xprt: rpc_xprt struct containing statistics
2428 static void xs_udp_print_stats(struct rpc_xprt
*xprt
, struct seq_file
*seq
)
2430 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2432 seq_printf(seq
, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2435 xprt
->stat
.bind_count
,
2438 xprt
->stat
.bad_xids
,
2441 xprt
->stat
.max_slots
,
2442 xprt
->stat
.sending_u
,
2443 xprt
->stat
.pending_u
);
2447 * xs_tcp_print_stats - display TCP socket-specifc stats
2448 * @xprt: rpc_xprt struct containing statistics
2452 static void xs_tcp_print_stats(struct rpc_xprt
*xprt
, struct seq_file
*seq
)
2454 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2457 if (xprt_connected(xprt
))
2458 idle_time
= (long)(jiffies
- xprt
->last_used
) / HZ
;
2460 seq_printf(seq
, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2461 "%llu %llu %lu %llu %llu\n",
2463 xprt
->stat
.bind_count
,
2464 xprt
->stat
.connect_count
,
2465 xprt
->stat
.connect_time
,
2469 xprt
->stat
.bad_xids
,
2472 xprt
->stat
.max_slots
,
2473 xprt
->stat
.sending_u
,
2474 xprt
->stat
.pending_u
);
2478 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2479 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2480 * to use the server side send routines.
2482 static void *bc_malloc(struct rpc_task
*task
, size_t size
)
2485 struct rpc_buffer
*buf
;
2487 WARN_ON_ONCE(size
> PAGE_SIZE
- sizeof(struct rpc_buffer
));
2488 if (size
> PAGE_SIZE
- sizeof(struct rpc_buffer
))
2491 page
= alloc_page(GFP_KERNEL
);
2495 buf
= page_address(page
);
2496 buf
->len
= PAGE_SIZE
;
2502 * Free the space allocated in the bc_alloc routine
2504 static void bc_free(void *buffer
)
2506 struct rpc_buffer
*buf
;
2511 buf
= container_of(buffer
, struct rpc_buffer
, data
);
2512 free_page((unsigned long)buf
);
2516 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2517 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2519 static int bc_sendto(struct rpc_rqst
*req
)
2522 struct xdr_buf
*xbufp
= &req
->rq_snd_buf
;
2523 struct rpc_xprt
*xprt
= req
->rq_xprt
;
2524 struct sock_xprt
*transport
=
2525 container_of(xprt
, struct sock_xprt
, xprt
);
2526 struct socket
*sock
= transport
->sock
;
2527 unsigned long headoff
;
2528 unsigned long tailoff
;
2530 xs_encode_stream_record_marker(xbufp
);
2532 tailoff
= (unsigned long)xbufp
->tail
[0].iov_base
& ~PAGE_MASK
;
2533 headoff
= (unsigned long)xbufp
->head
[0].iov_base
& ~PAGE_MASK
;
2534 len
= svc_send_common(sock
, xbufp
,
2535 virt_to_page(xbufp
->head
[0].iov_base
), headoff
,
2536 xbufp
->tail
[0].iov_base
, tailoff
);
2538 if (len
!= xbufp
->len
) {
2539 printk(KERN_NOTICE
"Error sending entire callback!\n");
2547 * The send routine. Borrows from svc_send
2549 static int bc_send_request(struct rpc_task
*task
)
2551 struct rpc_rqst
*req
= task
->tk_rqstp
;
2552 struct svc_xprt
*xprt
;
2555 dprintk("sending request with xid: %08x\n", ntohl(req
->rq_xid
));
2557 * Get the server socket associated with this callback xprt
2559 xprt
= req
->rq_xprt
->bc_xprt
;
2562 * Grab the mutex to serialize data as the connection is shared
2563 * with the fore channel
2565 if (!mutex_trylock(&xprt
->xpt_mutex
)) {
2566 rpc_sleep_on(&xprt
->xpt_bc_pending
, task
, NULL
);
2567 if (!mutex_trylock(&xprt
->xpt_mutex
))
2569 rpc_wake_up_queued_task(&xprt
->xpt_bc_pending
, task
);
2571 if (test_bit(XPT_DEAD
, &xprt
->xpt_flags
))
2574 len
= bc_sendto(req
);
2575 mutex_unlock(&xprt
->xpt_mutex
);
2584 * The close routine. Since this is client initiated, we do nothing
2587 static void bc_close(struct rpc_xprt
*xprt
)
2592 * The xprt destroy routine. Again, because this connection is client
2593 * initiated, we do nothing
2596 static void bc_destroy(struct rpc_xprt
*xprt
)
2598 dprintk("RPC: bc_destroy xprt %p\n", xprt
);
2601 module_put(THIS_MODULE
);
2604 static struct rpc_xprt_ops xs_local_ops
= {
2605 .reserve_xprt
= xprt_reserve_xprt
,
2606 .release_xprt
= xs_tcp_release_xprt
,
2607 .alloc_slot
= xprt_alloc_slot
,
2608 .rpcbind
= xs_local_rpcbind
,
2609 .set_port
= xs_local_set_port
,
2610 .connect
= xs_local_connect
,
2611 .buf_alloc
= rpc_malloc
,
2612 .buf_free
= rpc_free
,
2613 .send_request
= xs_local_send_request
,
2614 .set_retrans_timeout
= xprt_set_retrans_timeout_def
,
2616 .destroy
= xs_destroy
,
2617 .print_stats
= xs_local_print_stats
,
2618 .enable_swap
= xs_enable_swap
,
2619 .disable_swap
= xs_disable_swap
,
2622 static struct rpc_xprt_ops xs_udp_ops
= {
2623 .set_buffer_size
= xs_udp_set_buffer_size
,
2624 .reserve_xprt
= xprt_reserve_xprt_cong
,
2625 .release_xprt
= xprt_release_xprt_cong
,
2626 .alloc_slot
= xprt_alloc_slot
,
2627 .rpcbind
= rpcb_getport_async
,
2628 .set_port
= xs_set_port
,
2629 .connect
= xs_connect
,
2630 .buf_alloc
= rpc_malloc
,
2631 .buf_free
= rpc_free
,
2632 .send_request
= xs_udp_send_request
,
2633 .set_retrans_timeout
= xprt_set_retrans_timeout_rtt
,
2634 .timer
= xs_udp_timer
,
2635 .release_request
= xprt_release_rqst_cong
,
2637 .destroy
= xs_destroy
,
2638 .print_stats
= xs_udp_print_stats
,
2639 .enable_swap
= xs_enable_swap
,
2640 .disable_swap
= xs_disable_swap
,
2641 .inject_disconnect
= xs_inject_disconnect
,
2644 static struct rpc_xprt_ops xs_tcp_ops
= {
2645 .reserve_xprt
= xprt_reserve_xprt
,
2646 .release_xprt
= xs_tcp_release_xprt
,
2647 .alloc_slot
= xprt_lock_and_alloc_slot
,
2648 .rpcbind
= rpcb_getport_async
,
2649 .set_port
= xs_set_port
,
2650 .connect
= xs_connect
,
2651 .buf_alloc
= rpc_malloc
,
2652 .buf_free
= rpc_free
,
2653 .send_request
= xs_tcp_send_request
,
2654 .set_retrans_timeout
= xprt_set_retrans_timeout_def
,
2655 .close
= xs_tcp_shutdown
,
2656 .destroy
= xs_destroy
,
2657 .print_stats
= xs_tcp_print_stats
,
2658 .enable_swap
= xs_enable_swap
,
2659 .disable_swap
= xs_disable_swap
,
2660 .inject_disconnect
= xs_inject_disconnect
,
2661 #ifdef CONFIG_SUNRPC_BACKCHANNEL
2662 .bc_setup
= xprt_setup_bc
,
2663 .bc_up
= xs_tcp_bc_up
,
2664 .bc_free_rqst
= xprt_free_bc_rqst
,
2665 .bc_destroy
= xprt_destroy_bc
,
2670 * The rpc_xprt_ops for the server backchannel
2673 static struct rpc_xprt_ops bc_tcp_ops
= {
2674 .reserve_xprt
= xprt_reserve_xprt
,
2675 .release_xprt
= xprt_release_xprt
,
2676 .alloc_slot
= xprt_alloc_slot
,
2677 .buf_alloc
= bc_malloc
,
2678 .buf_free
= bc_free
,
2679 .send_request
= bc_send_request
,
2680 .set_retrans_timeout
= xprt_set_retrans_timeout_def
,
2682 .destroy
= bc_destroy
,
2683 .print_stats
= xs_tcp_print_stats
,
2684 .enable_swap
= xs_enable_swap
,
2685 .disable_swap
= xs_disable_swap
,
2686 .inject_disconnect
= xs_inject_disconnect
,
2689 static int xs_init_anyaddr(const int family
, struct sockaddr
*sap
)
2691 static const struct sockaddr_in sin
= {
2692 .sin_family
= AF_INET
,
2693 .sin_addr
.s_addr
= htonl(INADDR_ANY
),
2695 static const struct sockaddr_in6 sin6
= {
2696 .sin6_family
= AF_INET6
,
2697 .sin6_addr
= IN6ADDR_ANY_INIT
,
2704 memcpy(sap
, &sin
, sizeof(sin
));
2707 memcpy(sap
, &sin6
, sizeof(sin6
));
2710 dprintk("RPC: %s: Bad address family\n", __func__
);
2711 return -EAFNOSUPPORT
;
2716 static struct rpc_xprt
*xs_setup_xprt(struct xprt_create
*args
,
2717 unsigned int slot_table_size
,
2718 unsigned int max_slot_table_size
)
2720 struct rpc_xprt
*xprt
;
2721 struct sock_xprt
*new;
2723 if (args
->addrlen
> sizeof(xprt
->addr
)) {
2724 dprintk("RPC: xs_setup_xprt: address too large\n");
2725 return ERR_PTR(-EBADF
);
2728 xprt
= xprt_alloc(args
->net
, sizeof(*new), slot_table_size
,
2729 max_slot_table_size
);
2731 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2733 return ERR_PTR(-ENOMEM
);
2736 new = container_of(xprt
, struct sock_xprt
, xprt
);
2737 mutex_init(&new->recv_mutex
);
2738 memcpy(&xprt
->addr
, args
->dstaddr
, args
->addrlen
);
2739 xprt
->addrlen
= args
->addrlen
;
2741 memcpy(&new->srcaddr
, args
->srcaddr
, args
->addrlen
);
2744 err
= xs_init_anyaddr(args
->dstaddr
->sa_family
,
2745 (struct sockaddr
*)&new->srcaddr
);
2748 return ERR_PTR(err
);
2755 static const struct rpc_timeout xs_local_default_timeout
= {
2756 .to_initval
= 10 * HZ
,
2757 .to_maxval
= 10 * HZ
,
2762 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2763 * @args: rpc transport creation arguments
2765 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2767 static struct rpc_xprt
*xs_setup_local(struct xprt_create
*args
)
2769 struct sockaddr_un
*sun
= (struct sockaddr_un
*)args
->dstaddr
;
2770 struct sock_xprt
*transport
;
2771 struct rpc_xprt
*xprt
;
2772 struct rpc_xprt
*ret
;
2774 xprt
= xs_setup_xprt(args
, xprt_tcp_slot_table_entries
,
2775 xprt_max_tcp_slot_table_entries
);
2778 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2781 xprt
->tsh_size
= sizeof(rpc_fraghdr
) / sizeof(u32
);
2782 xprt
->max_payload
= RPC_MAX_FRAGMENT_SIZE
;
2784 xprt
->bind_timeout
= XS_BIND_TO
;
2785 xprt
->reestablish_timeout
= XS_TCP_INIT_REEST_TO
;
2786 xprt
->idle_timeout
= XS_IDLE_DISC_TO
;
2788 xprt
->ops
= &xs_local_ops
;
2789 xprt
->timeout
= &xs_local_default_timeout
;
2791 INIT_WORK(&transport
->recv_worker
, xs_local_data_receive_workfn
);
2792 INIT_DELAYED_WORK(&transport
->connect_worker
,
2793 xs_dummy_setup_socket
);
2795 switch (sun
->sun_family
) {
2797 if (sun
->sun_path
[0] != '/') {
2798 dprintk("RPC: bad AF_LOCAL address: %s\n",
2800 ret
= ERR_PTR(-EINVAL
);
2803 xprt_set_bound(xprt
);
2804 xs_format_peer_addresses(xprt
, "local", RPCBIND_NETID_LOCAL
);
2805 ret
= ERR_PTR(xs_local_setup_socket(transport
));
2810 ret
= ERR_PTR(-EAFNOSUPPORT
);
2814 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2815 xprt
->address_strings
[RPC_DISPLAY_ADDR
]);
2817 if (try_module_get(THIS_MODULE
))
2819 ret
= ERR_PTR(-EINVAL
);
2825 static const struct rpc_timeout xs_udp_default_timeout
= {
2826 .to_initval
= 5 * HZ
,
2827 .to_maxval
= 30 * HZ
,
2828 .to_increment
= 5 * HZ
,
2833 * xs_setup_udp - Set up transport to use a UDP socket
2834 * @args: rpc transport creation arguments
2837 static struct rpc_xprt
*xs_setup_udp(struct xprt_create
*args
)
2839 struct sockaddr
*addr
= args
->dstaddr
;
2840 struct rpc_xprt
*xprt
;
2841 struct sock_xprt
*transport
;
2842 struct rpc_xprt
*ret
;
2844 xprt
= xs_setup_xprt(args
, xprt_udp_slot_table_entries
,
2845 xprt_udp_slot_table_entries
);
2848 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2850 xprt
->prot
= IPPROTO_UDP
;
2852 /* XXX: header size can vary due to auth type, IPv6, etc. */
2853 xprt
->max_payload
= (1U << 16) - (MAX_HEADER
<< 3);
2855 xprt
->bind_timeout
= XS_BIND_TO
;
2856 xprt
->reestablish_timeout
= XS_UDP_REEST_TO
;
2857 xprt
->idle_timeout
= XS_IDLE_DISC_TO
;
2859 xprt
->ops
= &xs_udp_ops
;
2861 xprt
->timeout
= &xs_udp_default_timeout
;
2863 INIT_WORK(&transport
->recv_worker
, xs_udp_data_receive_workfn
);
2864 INIT_DELAYED_WORK(&transport
->connect_worker
, xs_udp_setup_socket
);
2866 switch (addr
->sa_family
) {
2868 if (((struct sockaddr_in
*)addr
)->sin_port
!= htons(0))
2869 xprt_set_bound(xprt
);
2871 xs_format_peer_addresses(xprt
, "udp", RPCBIND_NETID_UDP
);
2874 if (((struct sockaddr_in6
*)addr
)->sin6_port
!= htons(0))
2875 xprt_set_bound(xprt
);
2877 xs_format_peer_addresses(xprt
, "udp", RPCBIND_NETID_UDP6
);
2880 ret
= ERR_PTR(-EAFNOSUPPORT
);
2884 if (xprt_bound(xprt
))
2885 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2886 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
2887 xprt
->address_strings
[RPC_DISPLAY_PORT
],
2888 xprt
->address_strings
[RPC_DISPLAY_PROTO
]);
2890 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2891 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
2892 xprt
->address_strings
[RPC_DISPLAY_PROTO
]);
2894 if (try_module_get(THIS_MODULE
))
2896 ret
= ERR_PTR(-EINVAL
);
2902 static const struct rpc_timeout xs_tcp_default_timeout
= {
2903 .to_initval
= 60 * HZ
,
2904 .to_maxval
= 60 * HZ
,
2909 * xs_setup_tcp - Set up transport to use a TCP socket
2910 * @args: rpc transport creation arguments
2913 static struct rpc_xprt
*xs_setup_tcp(struct xprt_create
*args
)
2915 struct sockaddr
*addr
= args
->dstaddr
;
2916 struct rpc_xprt
*xprt
;
2917 struct sock_xprt
*transport
;
2918 struct rpc_xprt
*ret
;
2919 unsigned int max_slot_table_size
= xprt_max_tcp_slot_table_entries
;
2921 if (args
->flags
& XPRT_CREATE_INFINITE_SLOTS
)
2922 max_slot_table_size
= RPC_MAX_SLOT_TABLE_LIMIT
;
2924 xprt
= xs_setup_xprt(args
, xprt_tcp_slot_table_entries
,
2925 max_slot_table_size
);
2928 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2930 xprt
->prot
= IPPROTO_TCP
;
2931 xprt
->tsh_size
= sizeof(rpc_fraghdr
) / sizeof(u32
);
2932 xprt
->max_payload
= RPC_MAX_FRAGMENT_SIZE
;
2934 xprt
->bind_timeout
= XS_BIND_TO
;
2935 xprt
->reestablish_timeout
= XS_TCP_INIT_REEST_TO
;
2936 xprt
->idle_timeout
= XS_IDLE_DISC_TO
;
2938 xprt
->ops
= &xs_tcp_ops
;
2939 xprt
->timeout
= &xs_tcp_default_timeout
;
2941 INIT_WORK(&transport
->recv_worker
, xs_tcp_data_receive_workfn
);
2942 INIT_DELAYED_WORK(&transport
->connect_worker
, xs_tcp_setup_socket
);
2944 switch (addr
->sa_family
) {
2946 if (((struct sockaddr_in
*)addr
)->sin_port
!= htons(0))
2947 xprt_set_bound(xprt
);
2949 xs_format_peer_addresses(xprt
, "tcp", RPCBIND_NETID_TCP
);
2952 if (((struct sockaddr_in6
*)addr
)->sin6_port
!= htons(0))
2953 xprt_set_bound(xprt
);
2955 xs_format_peer_addresses(xprt
, "tcp", RPCBIND_NETID_TCP6
);
2958 ret
= ERR_PTR(-EAFNOSUPPORT
);
2962 if (xprt_bound(xprt
))
2963 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2964 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
2965 xprt
->address_strings
[RPC_DISPLAY_PORT
],
2966 xprt
->address_strings
[RPC_DISPLAY_PROTO
]);
2968 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2969 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
2970 xprt
->address_strings
[RPC_DISPLAY_PROTO
]);
2972 if (try_module_get(THIS_MODULE
))
2974 ret
= ERR_PTR(-EINVAL
);
2981 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2982 * @args: rpc transport creation arguments
2985 static struct rpc_xprt
*xs_setup_bc_tcp(struct xprt_create
*args
)
2987 struct sockaddr
*addr
= args
->dstaddr
;
2988 struct rpc_xprt
*xprt
;
2989 struct sock_xprt
*transport
;
2990 struct svc_sock
*bc_sock
;
2991 struct rpc_xprt
*ret
;
2993 xprt
= xs_setup_xprt(args
, xprt_tcp_slot_table_entries
,
2994 xprt_tcp_slot_table_entries
);
2997 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2999 xprt
->prot
= IPPROTO_TCP
;
3000 xprt
->tsh_size
= sizeof(rpc_fraghdr
) / sizeof(u32
);
3001 xprt
->max_payload
= RPC_MAX_FRAGMENT_SIZE
;
3002 xprt
->timeout
= &xs_tcp_default_timeout
;
3005 xprt_set_bound(xprt
);
3006 xprt
->bind_timeout
= 0;
3007 xprt
->reestablish_timeout
= 0;
3008 xprt
->idle_timeout
= 0;
3010 xprt
->ops
= &bc_tcp_ops
;
3012 switch (addr
->sa_family
) {
3014 xs_format_peer_addresses(xprt
, "tcp",
3018 xs_format_peer_addresses(xprt
, "tcp",
3019 RPCBIND_NETID_TCP6
);
3022 ret
= ERR_PTR(-EAFNOSUPPORT
);
3026 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3027 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
3028 xprt
->address_strings
[RPC_DISPLAY_PORT
],
3029 xprt
->address_strings
[RPC_DISPLAY_PROTO
]);
3032 * Once we've associated a backchannel xprt with a connection,
3033 * we want to keep it around as long as the connection lasts,
3034 * in case we need to start using it for a backchannel again;
3035 * this reference won't be dropped until bc_xprt is destroyed.
3038 args
->bc_xprt
->xpt_bc_xprt
= xprt
;
3039 xprt
->bc_xprt
= args
->bc_xprt
;
3040 bc_sock
= container_of(args
->bc_xprt
, struct svc_sock
, sk_xprt
);
3041 transport
->sock
= bc_sock
->sk_sock
;
3042 transport
->inet
= bc_sock
->sk_sk
;
3045 * Since we don't want connections for the backchannel, we set
3046 * the xprt status to connected
3048 xprt_set_connected(xprt
);
3050 if (try_module_get(THIS_MODULE
))
3053 args
->bc_xprt
->xpt_bc_xprt
= NULL
;
3055 ret
= ERR_PTR(-EINVAL
);
3061 static struct xprt_class xs_local_transport
= {
3062 .list
= LIST_HEAD_INIT(xs_local_transport
.list
),
3063 .name
= "named UNIX socket",
3064 .owner
= THIS_MODULE
,
3065 .ident
= XPRT_TRANSPORT_LOCAL
,
3066 .setup
= xs_setup_local
,
3069 static struct xprt_class xs_udp_transport
= {
3070 .list
= LIST_HEAD_INIT(xs_udp_transport
.list
),
3072 .owner
= THIS_MODULE
,
3073 .ident
= XPRT_TRANSPORT_UDP
,
3074 .setup
= xs_setup_udp
,
3077 static struct xprt_class xs_tcp_transport
= {
3078 .list
= LIST_HEAD_INIT(xs_tcp_transport
.list
),
3080 .owner
= THIS_MODULE
,
3081 .ident
= XPRT_TRANSPORT_TCP
,
3082 .setup
= xs_setup_tcp
,
3085 static struct xprt_class xs_bc_tcp_transport
= {
3086 .list
= LIST_HEAD_INIT(xs_bc_tcp_transport
.list
),
3087 .name
= "tcp NFSv4.1 backchannel",
3088 .owner
= THIS_MODULE
,
3089 .ident
= XPRT_TRANSPORT_BC_TCP
,
3090 .setup
= xs_setup_bc_tcp
,
3094 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3097 int init_socket_xprt(void)
3099 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3100 if (!sunrpc_table_header
)
3101 sunrpc_table_header
= register_sysctl_table(sunrpc_table
);
3104 xprt_register_transport(&xs_local_transport
);
3105 xprt_register_transport(&xs_udp_transport
);
3106 xprt_register_transport(&xs_tcp_transport
);
3107 xprt_register_transport(&xs_bc_tcp_transport
);
3113 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3116 void cleanup_socket_xprt(void)
3118 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3119 if (sunrpc_table_header
) {
3120 unregister_sysctl_table(sunrpc_table_header
);
3121 sunrpc_table_header
= NULL
;
3125 xprt_unregister_transport(&xs_local_transport
);
3126 xprt_unregister_transport(&xs_udp_transport
);
3127 xprt_unregister_transport(&xs_tcp_transport
);
3128 xprt_unregister_transport(&xs_bc_tcp_transport
);
3131 static int param_set_uint_minmax(const char *val
,
3132 const struct kernel_param
*kp
,
3133 unsigned int min
, unsigned int max
)
3140 ret
= kstrtouint(val
, 0, &num
);
3141 if (ret
== -EINVAL
|| num
< min
|| num
> max
)
3143 *((unsigned int *)kp
->arg
) = num
;
3147 static int param_set_portnr(const char *val
, const struct kernel_param
*kp
)
3149 return param_set_uint_minmax(val
, kp
,
3154 static const struct kernel_param_ops param_ops_portnr
= {
3155 .set
= param_set_portnr
,
3156 .get
= param_get_uint
,
3159 #define param_check_portnr(name, p) \
3160 __param_check(name, p, unsigned int);
3162 module_param_named(min_resvport
, xprt_min_resvport
, portnr
, 0644);
3163 module_param_named(max_resvport
, xprt_max_resvport
, portnr
, 0644);
3165 static int param_set_slot_table_size(const char *val
,
3166 const struct kernel_param
*kp
)
3168 return param_set_uint_minmax(val
, kp
,
3170 RPC_MAX_SLOT_TABLE
);
3173 static const struct kernel_param_ops param_ops_slot_table_size
= {
3174 .set
= param_set_slot_table_size
,
3175 .get
= param_get_uint
,
3178 #define param_check_slot_table_size(name, p) \
3179 __param_check(name, p, unsigned int);
3181 static int param_set_max_slot_table_size(const char *val
,
3182 const struct kernel_param
*kp
)
3184 return param_set_uint_minmax(val
, kp
,
3186 RPC_MAX_SLOT_TABLE_LIMIT
);
3189 static const struct kernel_param_ops param_ops_max_slot_table_size
= {
3190 .set
= param_set_max_slot_table_size
,
3191 .get
= param_get_uint
,
3194 #define param_check_max_slot_table_size(name, p) \
3195 __param_check(name, p, unsigned int);
3197 module_param_named(tcp_slot_table_entries
, xprt_tcp_slot_table_entries
,
3198 slot_table_size
, 0644);
3199 module_param_named(tcp_max_slot_table_entries
, xprt_max_tcp_slot_table_entries
,
3200 max_slot_table_size
, 0644);
3201 module_param_named(udp_slot_table_entries
, xprt_udp_slot_table_entries
,
3202 slot_table_size
, 0644);