1 // SPDX-License-Identifier: GPL-2.0
3 * linux/net/sunrpc/xprtsock.c
5 * Client-side transport implementation for sockets.
7 * TCP callback races fixes (C) 1998 Red Hat
8 * TCP send fixes (C) 1998 Red Hat
9 * TCP NFS related read + write fixes
10 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
12 * Rewrite of larges part of the code in order to stabilize TCP stuff.
13 * Fix behaviour when socket buffer is full.
14 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
16 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
18 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
19 * <gilles.quillard@bull.net>
22 #include <linux/types.h>
23 #include <linux/string.h>
24 #include <linux/slab.h>
25 #include <linux/module.h>
26 #include <linux/capability.h>
27 #include <linux/pagemap.h>
28 #include <linux/errno.h>
29 #include <linux/socket.h>
31 #include <linux/net.h>
34 #include <linux/udp.h>
35 #include <linux/tcp.h>
36 #include <linux/sunrpc/clnt.h>
37 #include <linux/sunrpc/addr.h>
38 #include <linux/sunrpc/sched.h>
39 #include <linux/sunrpc/svcsock.h>
40 #include <linux/sunrpc/xprtsock.h>
41 #include <linux/file.h>
42 #ifdef CONFIG_SUNRPC_BACKCHANNEL
43 #include <linux/sunrpc/bc_xprt.h>
47 #include <net/checksum.h>
50 #include <linux/bvec.h>
51 #include <linux/highmem.h>
52 #include <linux/uio.h>
53 #include <linux/sched/mm.h>
55 #include <trace/events/sunrpc.h>
60 static void xs_close(struct rpc_xprt
*xprt
);
61 static void xs_tcp_set_socket_timeouts(struct rpc_xprt
*xprt
,
67 static unsigned int xprt_udp_slot_table_entries
= RPC_DEF_SLOT_TABLE
;
68 static unsigned int xprt_tcp_slot_table_entries
= RPC_MIN_SLOT_TABLE
;
69 static unsigned int xprt_max_tcp_slot_table_entries
= RPC_MAX_SLOT_TABLE
;
71 static unsigned int xprt_min_resvport
= RPC_DEF_MIN_RESVPORT
;
72 static unsigned int xprt_max_resvport
= RPC_DEF_MAX_RESVPORT
;
74 #define XS_TCP_LINGER_TO (15U * HZ)
75 static unsigned int xs_tcp_fin_timeout __read_mostly
= XS_TCP_LINGER_TO
;
78 * We can register our own files under /proc/sys/sunrpc by
79 * calling register_sysctl_table() again. The files in that
80 * directory become the union of all files registered there.
82 * We simply need to make sure that we don't collide with
83 * someone else's file names!
86 static unsigned int min_slot_table_size
= RPC_MIN_SLOT_TABLE
;
87 static unsigned int max_slot_table_size
= RPC_MAX_SLOT_TABLE
;
88 static unsigned int max_tcp_slot_table_limit
= RPC_MAX_SLOT_TABLE_LIMIT
;
89 static unsigned int xprt_min_resvport_limit
= RPC_MIN_RESVPORT
;
90 static unsigned int xprt_max_resvport_limit
= RPC_MAX_RESVPORT
;
92 static struct ctl_table_header
*sunrpc_table_header
;
94 static struct xprt_class xs_local_transport
;
95 static struct xprt_class xs_udp_transport
;
96 static struct xprt_class xs_tcp_transport
;
97 static struct xprt_class xs_bc_tcp_transport
;
100 * FIXME: changing the UDP slot table size should also resize the UDP
101 * socket buffers for existing UDP transports
103 static struct ctl_table xs_tunables_table
[] = {
105 .procname
= "udp_slot_table_entries",
106 .data
= &xprt_udp_slot_table_entries
,
107 .maxlen
= sizeof(unsigned int),
109 .proc_handler
= proc_dointvec_minmax
,
110 .extra1
= &min_slot_table_size
,
111 .extra2
= &max_slot_table_size
114 .procname
= "tcp_slot_table_entries",
115 .data
= &xprt_tcp_slot_table_entries
,
116 .maxlen
= sizeof(unsigned int),
118 .proc_handler
= proc_dointvec_minmax
,
119 .extra1
= &min_slot_table_size
,
120 .extra2
= &max_slot_table_size
123 .procname
= "tcp_max_slot_table_entries",
124 .data
= &xprt_max_tcp_slot_table_entries
,
125 .maxlen
= sizeof(unsigned int),
127 .proc_handler
= proc_dointvec_minmax
,
128 .extra1
= &min_slot_table_size
,
129 .extra2
= &max_tcp_slot_table_limit
132 .procname
= "min_resvport",
133 .data
= &xprt_min_resvport
,
134 .maxlen
= sizeof(unsigned int),
136 .proc_handler
= proc_dointvec_minmax
,
137 .extra1
= &xprt_min_resvport_limit
,
138 .extra2
= &xprt_max_resvport_limit
141 .procname
= "max_resvport",
142 .data
= &xprt_max_resvport
,
143 .maxlen
= sizeof(unsigned int),
145 .proc_handler
= proc_dointvec_minmax
,
146 .extra1
= &xprt_min_resvport_limit
,
147 .extra2
= &xprt_max_resvport_limit
150 .procname
= "tcp_fin_timeout",
151 .data
= &xs_tcp_fin_timeout
,
152 .maxlen
= sizeof(xs_tcp_fin_timeout
),
154 .proc_handler
= proc_dointvec_jiffies
,
159 static struct ctl_table sunrpc_table
[] = {
161 .procname
= "sunrpc",
163 .child
= xs_tunables_table
169 * Wait duration for a reply from the RPC portmapper.
171 #define XS_BIND_TO (60U * HZ)
174 * Delay if a UDP socket connect error occurs. This is most likely some
175 * kind of resource problem on the local host.
177 #define XS_UDP_REEST_TO (2U * HZ)
180 * The reestablish timeout allows clients to delay for a bit before attempting
181 * to reconnect to a server that just dropped our connection.
183 * We implement an exponential backoff when trying to reestablish a TCP
184 * transport connection with the server. Some servers like to drop a TCP
185 * connection when they are overworked, so we start with a short timeout and
186 * increase over time if the server is down or not responding.
188 #define XS_TCP_INIT_REEST_TO (3U * HZ)
191 * TCP idle timeout; client drops the transport socket if it is idle
192 * for this long. Note that we also timeout UDP sockets to prevent
193 * holding port numbers when there is no RPC traffic.
195 #define XS_IDLE_DISC_TO (5U * 60 * HZ)
197 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
198 # undef RPC_DEBUG_DATA
199 # define RPCDBG_FACILITY RPCDBG_TRANS
202 #ifdef RPC_DEBUG_DATA
203 static void xs_pktdump(char *msg
, u32
*packet
, unsigned int count
)
205 u8
*buf
= (u8
*) packet
;
208 dprintk("RPC: %s\n", msg
);
209 for (j
= 0; j
< count
&& j
< 128; j
+= 4) {
213 dprintk("0x%04x ", j
);
215 dprintk("%02x%02x%02x%02x ",
216 buf
[j
], buf
[j
+1], buf
[j
+2], buf
[j
+3]);
221 static inline void xs_pktdump(char *msg
, u32
*packet
, unsigned int count
)
227 static inline struct rpc_xprt
*xprt_from_sock(struct sock
*sk
)
229 return (struct rpc_xprt
*) sk
->sk_user_data
;
232 static inline struct sockaddr
*xs_addr(struct rpc_xprt
*xprt
)
234 return (struct sockaddr
*) &xprt
->addr
;
237 static inline struct sockaddr_un
*xs_addr_un(struct rpc_xprt
*xprt
)
239 return (struct sockaddr_un
*) &xprt
->addr
;
242 static inline struct sockaddr_in
*xs_addr_in(struct rpc_xprt
*xprt
)
244 return (struct sockaddr_in
*) &xprt
->addr
;
247 static inline struct sockaddr_in6
*xs_addr_in6(struct rpc_xprt
*xprt
)
249 return (struct sockaddr_in6
*) &xprt
->addr
;
252 static void xs_format_common_peer_addresses(struct rpc_xprt
*xprt
)
254 struct sockaddr
*sap
= xs_addr(xprt
);
255 struct sockaddr_in6
*sin6
;
256 struct sockaddr_in
*sin
;
257 struct sockaddr_un
*sun
;
260 switch (sap
->sa_family
) {
262 sun
= xs_addr_un(xprt
);
263 strlcpy(buf
, sun
->sun_path
, sizeof(buf
));
264 xprt
->address_strings
[RPC_DISPLAY_ADDR
] =
265 kstrdup(buf
, GFP_KERNEL
);
268 (void)rpc_ntop(sap
, buf
, sizeof(buf
));
269 xprt
->address_strings
[RPC_DISPLAY_ADDR
] =
270 kstrdup(buf
, GFP_KERNEL
);
271 sin
= xs_addr_in(xprt
);
272 snprintf(buf
, sizeof(buf
), "%08x", ntohl(sin
->sin_addr
.s_addr
));
275 (void)rpc_ntop(sap
, buf
, sizeof(buf
));
276 xprt
->address_strings
[RPC_DISPLAY_ADDR
] =
277 kstrdup(buf
, GFP_KERNEL
);
278 sin6
= xs_addr_in6(xprt
);
279 snprintf(buf
, sizeof(buf
), "%pi6", &sin6
->sin6_addr
);
285 xprt
->address_strings
[RPC_DISPLAY_HEX_ADDR
] = kstrdup(buf
, GFP_KERNEL
);
288 static void xs_format_common_peer_ports(struct rpc_xprt
*xprt
)
290 struct sockaddr
*sap
= xs_addr(xprt
);
293 snprintf(buf
, sizeof(buf
), "%u", rpc_get_port(sap
));
294 xprt
->address_strings
[RPC_DISPLAY_PORT
] = kstrdup(buf
, GFP_KERNEL
);
296 snprintf(buf
, sizeof(buf
), "%4hx", rpc_get_port(sap
));
297 xprt
->address_strings
[RPC_DISPLAY_HEX_PORT
] = kstrdup(buf
, GFP_KERNEL
);
300 static void xs_format_peer_addresses(struct rpc_xprt
*xprt
,
301 const char *protocol
,
304 xprt
->address_strings
[RPC_DISPLAY_PROTO
] = protocol
;
305 xprt
->address_strings
[RPC_DISPLAY_NETID
] = netid
;
306 xs_format_common_peer_addresses(xprt
);
307 xs_format_common_peer_ports(xprt
);
310 static void xs_update_peer_port(struct rpc_xprt
*xprt
)
312 kfree(xprt
->address_strings
[RPC_DISPLAY_HEX_PORT
]);
313 kfree(xprt
->address_strings
[RPC_DISPLAY_PORT
]);
315 xs_format_common_peer_ports(xprt
);
318 static void xs_free_peer_addresses(struct rpc_xprt
*xprt
)
322 for (i
= 0; i
< RPC_DISPLAY_MAX
; i
++)
324 case RPC_DISPLAY_PROTO
:
325 case RPC_DISPLAY_NETID
:
328 kfree(xprt
->address_strings
[i
]);
333 xs_alloc_sparse_pages(struct xdr_buf
*buf
, size_t want
, gfp_t gfp
)
337 if (!want
|| !(buf
->flags
& XDRBUF_SPARSE_PAGES
))
339 n
= (buf
->page_base
+ want
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
340 for (i
= 0; i
< n
; i
++) {
343 buf
->bvec
[i
].bv_page
= buf
->pages
[i
] = alloc_page(gfp
);
344 if (!buf
->pages
[i
]) {
346 return i
> buf
->page_base
? i
- buf
->page_base
: 0;
353 xs_sock_recvmsg(struct socket
*sock
, struct msghdr
*msg
, int flags
, size_t seek
)
357 iov_iter_advance(&msg
->msg_iter
, seek
);
358 ret
= sock_recvmsg(sock
, msg
, flags
);
359 return ret
> 0 ? ret
+ seek
: ret
;
363 xs_read_kvec(struct socket
*sock
, struct msghdr
*msg
, int flags
,
364 struct kvec
*kvec
, size_t count
, size_t seek
)
366 iov_iter_kvec(&msg
->msg_iter
, READ
, kvec
, 1, count
);
367 return xs_sock_recvmsg(sock
, msg
, flags
, seek
);
371 xs_read_bvec(struct socket
*sock
, struct msghdr
*msg
, int flags
,
372 struct bio_vec
*bvec
, unsigned long nr
, size_t count
,
375 iov_iter_bvec(&msg
->msg_iter
, READ
, bvec
, nr
, count
);
376 return xs_sock_recvmsg(sock
, msg
, flags
, seek
);
380 xs_read_discard(struct socket
*sock
, struct msghdr
*msg
, int flags
,
383 iov_iter_discard(&msg
->msg_iter
, READ
, count
);
384 return sock_recvmsg(sock
, msg
, flags
);
387 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
389 xs_flush_bvec(const struct bio_vec
*bvec
, size_t count
, size_t seek
)
391 struct bvec_iter bi
= {
396 bvec_iter_advance(bvec
, &bi
, seek
& PAGE_MASK
);
397 for_each_bvec(bv
, bvec
, bi
, bi
)
398 flush_dcache_page(bv
.bv_page
);
402 xs_flush_bvec(const struct bio_vec
*bvec
, size_t count
, size_t seek
)
408 xs_read_xdr_buf(struct socket
*sock
, struct msghdr
*msg
, int flags
,
409 struct xdr_buf
*buf
, size_t count
, size_t seek
, size_t *read
)
411 size_t want
, seek_init
= seek
, offset
= 0;
414 want
= min_t(size_t, count
, buf
->head
[0].iov_len
);
416 ret
= xs_read_kvec(sock
, msg
, flags
, &buf
->head
[0], want
, seek
);
420 if (offset
== count
|| msg
->msg_flags
& (MSG_EOR
|MSG_TRUNC
))
430 want
= xs_alloc_sparse_pages(buf
,
431 min_t(size_t, count
- offset
, buf
->page_len
),
434 ret
= xs_read_bvec(sock
, msg
, flags
, buf
->bvec
,
435 xdr_buf_pagecount(buf
),
436 want
+ buf
->page_base
,
437 seek
+ buf
->page_base
);
440 xs_flush_bvec(buf
->bvec
, ret
, seek
+ buf
->page_base
);
441 ret
-= buf
->page_base
;
443 if (offset
== count
|| msg
->msg_flags
& (MSG_EOR
|MSG_TRUNC
))
453 want
= min_t(size_t, count
- offset
, buf
->tail
[0].iov_len
);
455 ret
= xs_read_kvec(sock
, msg
, flags
, &buf
->tail
[0], want
, seek
);
459 if (offset
== count
|| msg
->msg_flags
& (MSG_EOR
|MSG_TRUNC
))
463 } else if (offset
< seek_init
)
467 *read
= offset
- seek_init
;
475 xs_read_header(struct sock_xprt
*transport
, struct xdr_buf
*buf
)
477 if (!transport
->recv
.copied
) {
478 if (buf
->head
[0].iov_len
>= transport
->recv
.offset
)
479 memcpy(buf
->head
[0].iov_base
,
480 &transport
->recv
.xid
,
481 transport
->recv
.offset
);
482 transport
->recv
.copied
= transport
->recv
.offset
;
487 xs_read_stream_request_done(struct sock_xprt
*transport
)
489 return transport
->recv
.fraghdr
& cpu_to_be32(RPC_LAST_STREAM_FRAGMENT
);
493 xs_read_stream_check_eor(struct sock_xprt
*transport
,
496 if (xs_read_stream_request_done(transport
))
497 msg
->msg_flags
|= MSG_EOR
;
501 xs_read_stream_request(struct sock_xprt
*transport
, struct msghdr
*msg
,
502 int flags
, struct rpc_rqst
*req
)
504 struct xdr_buf
*buf
= &req
->rq_private_buf
;
508 xs_read_header(transport
, buf
);
510 want
= transport
->recv
.len
- transport
->recv
.offset
;
512 ret
= xs_read_xdr_buf(transport
->sock
, msg
, flags
, buf
,
513 transport
->recv
.copied
+ want
,
514 transport
->recv
.copied
,
516 transport
->recv
.offset
+= read
;
517 transport
->recv
.copied
+= read
;
520 if (transport
->recv
.offset
== transport
->recv
.len
)
521 xs_read_stream_check_eor(transport
, msg
);
531 msg
->msg_flags
|= MSG_TRUNC
;
536 return ret
< 0 ? ret
: read
;
540 xs_read_stream_headersize(bool isfrag
)
543 return sizeof(__be32
);
544 return 3 * sizeof(__be32
);
548 xs_read_stream_header(struct sock_xprt
*transport
, struct msghdr
*msg
,
549 int flags
, size_t want
, size_t seek
)
552 .iov_base
= &transport
->recv
.fraghdr
,
555 return xs_read_kvec(transport
->sock
, msg
, flags
, &kvec
, want
, seek
);
558 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
560 xs_read_stream_call(struct sock_xprt
*transport
, struct msghdr
*msg
, int flags
)
562 struct rpc_xprt
*xprt
= &transport
->xprt
;
563 struct rpc_rqst
*req
;
566 /* Is this transport associated with the backchannel? */
570 /* Look up and lock the request corresponding to the given XID */
571 req
= xprt_lookup_bc_request(xprt
, transport
->recv
.xid
);
573 printk(KERN_WARNING
"Callback slot table overflowed\n");
576 if (transport
->recv
.copied
&& !req
->rq_private_buf
.len
)
579 ret
= xs_read_stream_request(transport
, msg
, flags
, req
);
580 if (msg
->msg_flags
& (MSG_EOR
|MSG_TRUNC
))
581 xprt_complete_bc_request(req
, transport
->recv
.copied
);
583 req
->rq_private_buf
.len
= transport
->recv
.copied
;
587 #else /* CONFIG_SUNRPC_BACKCHANNEL */
589 xs_read_stream_call(struct sock_xprt
*transport
, struct msghdr
*msg
, int flags
)
593 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
596 xs_read_stream_reply(struct sock_xprt
*transport
, struct msghdr
*msg
, int flags
)
598 struct rpc_xprt
*xprt
= &transport
->xprt
;
599 struct rpc_rqst
*req
;
602 /* Look up and lock the request corresponding to the given XID */
603 spin_lock(&xprt
->queue_lock
);
604 req
= xprt_lookup_rqst(xprt
, transport
->recv
.xid
);
605 if (!req
|| (transport
->recv
.copied
&& !req
->rq_private_buf
.len
)) {
606 msg
->msg_flags
|= MSG_TRUNC
;
610 spin_unlock(&xprt
->queue_lock
);
612 ret
= xs_read_stream_request(transport
, msg
, flags
, req
);
614 spin_lock(&xprt
->queue_lock
);
615 if (msg
->msg_flags
& (MSG_EOR
|MSG_TRUNC
))
616 xprt_complete_rqst(req
->rq_task
, transport
->recv
.copied
);
618 req
->rq_private_buf
.len
= transport
->recv
.copied
;
619 xprt_unpin_rqst(req
);
621 spin_unlock(&xprt
->queue_lock
);
626 xs_read_stream(struct sock_xprt
*transport
, int flags
)
628 struct msghdr msg
= { 0 };
629 size_t want
, read
= 0;
632 if (transport
->recv
.len
== 0) {
633 want
= xs_read_stream_headersize(transport
->recv
.copied
!= 0);
634 ret
= xs_read_stream_header(transport
, &msg
, flags
, want
,
635 transport
->recv
.offset
);
638 transport
->recv
.offset
= ret
;
639 if (transport
->recv
.offset
!= want
)
640 return transport
->recv
.offset
;
641 transport
->recv
.len
= be32_to_cpu(transport
->recv
.fraghdr
) &
642 RPC_FRAGMENT_SIZE_MASK
;
643 transport
->recv
.offset
-= sizeof(transport
->recv
.fraghdr
);
647 switch (be32_to_cpu(transport
->recv
.calldir
)) {
649 msg
.msg_flags
|= MSG_TRUNC
;
652 ret
= xs_read_stream_call(transport
, &msg
, flags
);
655 ret
= xs_read_stream_reply(transport
, &msg
, flags
);
657 if (msg
.msg_flags
& MSG_TRUNC
) {
658 transport
->recv
.calldir
= cpu_to_be32(-1);
659 transport
->recv
.copied
= -1;
664 if (transport
->recv
.offset
< transport
->recv
.len
) {
665 if (!(msg
.msg_flags
& MSG_TRUNC
))
668 ret
= xs_read_discard(transport
->sock
, &msg
, flags
,
669 transport
->recv
.len
- transport
->recv
.offset
);
672 transport
->recv
.offset
+= ret
;
674 if (transport
->recv
.offset
!= transport
->recv
.len
)
677 if (xs_read_stream_request_done(transport
)) {
678 trace_xs_stream_read_request(transport
);
679 transport
->recv
.copied
= 0;
681 transport
->recv
.offset
= 0;
682 transport
->recv
.len
= 0;
685 return ret
!= 0 ? ret
: -ESHUTDOWN
;
688 static __poll_t
xs_poll_socket(struct sock_xprt
*transport
)
690 return transport
->sock
->ops
->poll(transport
->file
, transport
->sock
,
694 static bool xs_poll_socket_readable(struct sock_xprt
*transport
)
696 __poll_t events
= xs_poll_socket(transport
);
698 return (events
& (EPOLLIN
| EPOLLRDNORM
)) && !(events
& EPOLLRDHUP
);
701 static void xs_poll_check_readable(struct sock_xprt
*transport
)
704 clear_bit(XPRT_SOCK_DATA_READY
, &transport
->sock_state
);
705 if (!xs_poll_socket_readable(transport
))
707 if (!test_and_set_bit(XPRT_SOCK_DATA_READY
, &transport
->sock_state
))
708 queue_work(xprtiod_workqueue
, &transport
->recv_worker
);
711 static void xs_stream_data_receive(struct sock_xprt
*transport
)
716 mutex_lock(&transport
->recv_mutex
);
717 if (transport
->sock
== NULL
)
720 ret
= xs_read_stream(transport
, MSG_DONTWAIT
);
726 if (ret
== -ESHUTDOWN
)
727 kernel_sock_shutdown(transport
->sock
, SHUT_RDWR
);
729 xs_poll_check_readable(transport
);
731 mutex_unlock(&transport
->recv_mutex
);
732 trace_xs_stream_read_data(&transport
->xprt
, ret
, read
);
735 static void xs_stream_data_receive_workfn(struct work_struct
*work
)
737 struct sock_xprt
*transport
=
738 container_of(work
, struct sock_xprt
, recv_worker
);
739 unsigned int pflags
= memalloc_nofs_save();
741 xs_stream_data_receive(transport
);
742 memalloc_nofs_restore(pflags
);
746 xs_stream_reset_connect(struct sock_xprt
*transport
)
748 transport
->recv
.offset
= 0;
749 transport
->recv
.len
= 0;
750 transport
->recv
.copied
= 0;
751 transport
->xmit
.offset
= 0;
755 xs_stream_start_connect(struct sock_xprt
*transport
)
757 transport
->xprt
.stat
.connect_count
++;
758 transport
->xprt
.stat
.connect_start
= jiffies
;
761 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
764 * xs_nospace - handle transmit was incomplete
765 * @req: pointer to RPC request
768 static int xs_nospace(struct rpc_rqst
*req
)
770 struct rpc_xprt
*xprt
= req
->rq_xprt
;
771 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
772 struct sock
*sk
= transport
->inet
;
775 trace_rpc_socket_nospace(req
, transport
);
777 /* Protect against races with write_space */
778 spin_lock(&xprt
->transport_lock
);
780 /* Don't race with disconnect */
781 if (xprt_connected(xprt
)) {
782 /* wait for more buffer space */
783 sk
->sk_write_pending
++;
784 xprt_wait_for_buffer_space(xprt
);
788 spin_unlock(&xprt
->transport_lock
);
790 /* Race breaker in case memory is freed before above code is called */
791 if (ret
== -EAGAIN
) {
792 struct socket_wq
*wq
;
795 wq
= rcu_dereference(sk
->sk_wq
);
796 set_bit(SOCKWQ_ASYNC_NOSPACE
, &wq
->flags
);
799 sk
->sk_write_space(sk
);
805 xs_stream_prepare_request(struct rpc_rqst
*req
)
807 xdr_free_bvec(&req
->rq_rcv_buf
);
808 req
->rq_task
->tk_status
= xdr_alloc_bvec(&req
->rq_rcv_buf
, GFP_KERNEL
);
812 * Determine if the previous message in the stream was aborted before it
813 * could complete transmission.
816 xs_send_request_was_aborted(struct sock_xprt
*transport
, struct rpc_rqst
*req
)
818 return transport
->xmit
.offset
!= 0 && req
->rq_bytes_sent
== 0;
822 * Return the stream record marker field for a record of length < 2^31-1
825 xs_stream_record_marker(struct xdr_buf
*xdr
)
829 return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT
| (u32
)xdr
->len
);
833 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
834 * @req: pointer to RPC request
837 * 0: The request has been sent
838 * EAGAIN: The socket was blocked, please call again later to
839 * complete the request
840 * ENOTCONN: Caller needs to invoke connect logic then call again
841 * other: Some other error occurred, the request was not sent
843 static int xs_local_send_request(struct rpc_rqst
*req
)
845 struct rpc_xprt
*xprt
= req
->rq_xprt
;
846 struct sock_xprt
*transport
=
847 container_of(xprt
, struct sock_xprt
, xprt
);
848 struct xdr_buf
*xdr
= &req
->rq_snd_buf
;
849 rpc_fraghdr rm
= xs_stream_record_marker(xdr
);
850 unsigned int msglen
= rm
? req
->rq_slen
+ sizeof(rm
) : req
->rq_slen
;
851 struct msghdr msg
= {
852 .msg_flags
= XS_SENDMSG_FLAGS
,
857 /* Close the stream if the previous transmission was incomplete */
858 if (xs_send_request_was_aborted(transport
, req
)) {
863 xs_pktdump("packet data:",
864 req
->rq_svec
->iov_base
, req
->rq_svec
->iov_len
);
866 req
->rq_xtime
= ktime_get();
867 status
= xprt_sock_sendmsg(transport
->sock
, &msg
, xdr
,
868 transport
->xmit
.offset
, rm
, &sent
);
869 dprintk("RPC: %s(%u) = %d\n",
870 __func__
, xdr
->len
- transport
->xmit
.offset
, status
);
872 if (status
== -EAGAIN
&& sock_writeable(transport
->inet
))
875 if (likely(sent
> 0) || status
== 0) {
876 transport
->xmit
.offset
+= sent
;
877 req
->rq_bytes_sent
= transport
->xmit
.offset
;
878 if (likely(req
->rq_bytes_sent
>= msglen
)) {
879 req
->rq_xmit_bytes_sent
+= transport
->xmit
.offset
;
880 transport
->xmit
.offset
= 0;
890 status
= xs_nospace(req
);
893 dprintk("RPC: sendmsg returned unrecognized error %d\n",
905 * xs_udp_send_request - write an RPC request to a UDP socket
906 * @req: pointer to RPC request
909 * 0: The request has been sent
910 * EAGAIN: The socket was blocked, please call again later to
911 * complete the request
912 * ENOTCONN: Caller needs to invoke connect logic then call again
913 * other: Some other error occurred, the request was not sent
915 static int xs_udp_send_request(struct rpc_rqst
*req
)
917 struct rpc_xprt
*xprt
= req
->rq_xprt
;
918 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
919 struct xdr_buf
*xdr
= &req
->rq_snd_buf
;
920 struct msghdr msg
= {
921 .msg_name
= xs_addr(xprt
),
922 .msg_namelen
= xprt
->addrlen
,
923 .msg_flags
= XS_SENDMSG_FLAGS
,
928 xs_pktdump("packet data:",
929 req
->rq_svec
->iov_base
,
930 req
->rq_svec
->iov_len
);
932 if (!xprt_bound(xprt
))
935 if (!xprt_request_get_cong(xprt
, req
))
938 req
->rq_xtime
= ktime_get();
939 status
= xprt_sock_sendmsg(transport
->sock
, &msg
, xdr
, 0, 0, &sent
);
941 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
944 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
945 if (status
== -EPERM
)
948 if (status
== -EAGAIN
&& sock_writeable(transport
->inet
))
951 if (sent
> 0 || status
== 0) {
952 req
->rq_xmit_bytes_sent
+= sent
;
953 if (sent
>= req
->rq_slen
)
955 /* Still some bytes left; set up for a retry later. */
963 /* Should we call xs_close() here? */
966 status
= xs_nospace(req
);
973 /* When the server has died, an ICMP port unreachable message
974 * prompts ECONNREFUSED. */
977 dprintk("RPC: sendmsg returned unrecognized error %d\n",
985 * xs_tcp_send_request - write an RPC request to a TCP socket
986 * @req: pointer to RPC request
989 * 0: The request has been sent
990 * EAGAIN: The socket was blocked, please call again later to
991 * complete the request
992 * ENOTCONN: Caller needs to invoke connect logic then call again
993 * other: Some other error occurred, the request was not sent
995 * XXX: In the case of soft timeouts, should we eventually give up
996 * if sendmsg is not able to make progress?
998 static int xs_tcp_send_request(struct rpc_rqst
*req
)
1000 struct rpc_xprt
*xprt
= req
->rq_xprt
;
1001 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1002 struct xdr_buf
*xdr
= &req
->rq_snd_buf
;
1003 rpc_fraghdr rm
= xs_stream_record_marker(xdr
);
1004 unsigned int msglen
= rm
? req
->rq_slen
+ sizeof(rm
) : req
->rq_slen
;
1005 struct msghdr msg
= {
1006 .msg_flags
= XS_SENDMSG_FLAGS
,
1008 bool vm_wait
= false;
1012 /* Close the stream if the previous transmission was incomplete */
1013 if (xs_send_request_was_aborted(transport
, req
)) {
1014 if (transport
->sock
!= NULL
)
1015 kernel_sock_shutdown(transport
->sock
, SHUT_RDWR
);
1018 if (!transport
->inet
)
1021 xs_pktdump("packet data:",
1022 req
->rq_svec
->iov_base
,
1023 req
->rq_svec
->iov_len
);
1025 if (test_bit(XPRT_SOCK_UPD_TIMEOUT
, &transport
->sock_state
))
1026 xs_tcp_set_socket_timeouts(xprt
, transport
->sock
);
1028 /* Continue transmitting the packet/record. We must be careful
1029 * to cope with writespace callbacks arriving _after_ we have
1030 * called sendmsg(). */
1031 req
->rq_xtime
= ktime_get();
1032 tcp_sock_set_cork(transport
->inet
, true);
1034 status
= xprt_sock_sendmsg(transport
->sock
, &msg
, xdr
,
1035 transport
->xmit
.offset
, rm
, &sent
);
1037 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
1038 xdr
->len
- transport
->xmit
.offset
, status
);
1040 /* If we've sent the entire packet, immediately
1041 * reset the count of bytes sent. */
1042 transport
->xmit
.offset
+= sent
;
1043 req
->rq_bytes_sent
= transport
->xmit
.offset
;
1044 if (likely(req
->rq_bytes_sent
>= msglen
)) {
1045 req
->rq_xmit_bytes_sent
+= transport
->xmit
.offset
;
1046 transport
->xmit
.offset
= 0;
1047 if (atomic_long_read(&xprt
->xmit_queuelen
) == 1)
1048 tcp_sock_set_cork(transport
->inet
, false);
1052 WARN_ON_ONCE(sent
== 0 && status
== 0);
1054 if (status
== -EAGAIN
) {
1056 * Return EAGAIN if we're sure we're hitting the
1057 * socket send buffer limits.
1059 if (test_bit(SOCK_NOSPACE
, &transport
->sock
->flags
))
1062 * Did we hit a memory allocation failure?
1068 /* Retry, knowing now that we're below the
1069 * socket send buffer limit
1083 /* Should we call xs_close() here? */
1086 status
= xs_nospace(req
);
1096 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1103 static void xs_save_old_callbacks(struct sock_xprt
*transport
, struct sock
*sk
)
1105 transport
->old_data_ready
= sk
->sk_data_ready
;
1106 transport
->old_state_change
= sk
->sk_state_change
;
1107 transport
->old_write_space
= sk
->sk_write_space
;
1108 transport
->old_error_report
= sk
->sk_error_report
;
1111 static void xs_restore_old_callbacks(struct sock_xprt
*transport
, struct sock
*sk
)
1113 sk
->sk_data_ready
= transport
->old_data_ready
;
1114 sk
->sk_state_change
= transport
->old_state_change
;
1115 sk
->sk_write_space
= transport
->old_write_space
;
1116 sk
->sk_error_report
= transport
->old_error_report
;
1119 static void xs_sock_reset_state_flags(struct rpc_xprt
*xprt
)
1121 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1123 clear_bit(XPRT_SOCK_DATA_READY
, &transport
->sock_state
);
1124 clear_bit(XPRT_SOCK_WAKE_ERROR
, &transport
->sock_state
);
1125 clear_bit(XPRT_SOCK_WAKE_WRITE
, &transport
->sock_state
);
1126 clear_bit(XPRT_SOCK_WAKE_DISCONNECT
, &transport
->sock_state
);
1129 static void xs_run_error_worker(struct sock_xprt
*transport
, unsigned int nr
)
1131 set_bit(nr
, &transport
->sock_state
);
1132 queue_work(xprtiod_workqueue
, &transport
->error_worker
);
1135 static void xs_sock_reset_connection_flags(struct rpc_xprt
*xprt
)
1137 smp_mb__before_atomic();
1138 clear_bit(XPRT_CLOSE_WAIT
, &xprt
->state
);
1139 clear_bit(XPRT_CLOSING
, &xprt
->state
);
1140 xs_sock_reset_state_flags(xprt
);
1141 smp_mb__after_atomic();
1145 * xs_error_report - callback to handle TCP socket state errors
1148 * Note: we don't call sock_error() since there may be a rpc_task
1149 * using the socket, and so we don't want to clear sk->sk_err.
1151 static void xs_error_report(struct sock
*sk
)
1153 struct sock_xprt
*transport
;
1154 struct rpc_xprt
*xprt
;
1156 read_lock_bh(&sk
->sk_callback_lock
);
1157 if (!(xprt
= xprt_from_sock(sk
)))
1160 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1161 transport
->xprt_err
= -sk
->sk_err
;
1162 if (transport
->xprt_err
== 0)
1164 dprintk("RPC: xs_error_report client %p, error=%d...\n",
1165 xprt
, -transport
->xprt_err
);
1166 trace_rpc_socket_error(xprt
, sk
->sk_socket
, transport
->xprt_err
);
1168 /* barrier ensures xprt_err is set before XPRT_SOCK_WAKE_ERROR */
1169 smp_mb__before_atomic();
1170 xs_run_error_worker(transport
, XPRT_SOCK_WAKE_ERROR
);
1172 read_unlock_bh(&sk
->sk_callback_lock
);
1175 static void xs_reset_transport(struct sock_xprt
*transport
)
1177 struct socket
*sock
= transport
->sock
;
1178 struct sock
*sk
= transport
->inet
;
1179 struct rpc_xprt
*xprt
= &transport
->xprt
;
1180 struct file
*filp
= transport
->file
;
1185 if (atomic_read(&transport
->xprt
.swapper
))
1186 sk_clear_memalloc(sk
);
1188 kernel_sock_shutdown(sock
, SHUT_RDWR
);
1190 mutex_lock(&transport
->recv_mutex
);
1191 write_lock_bh(&sk
->sk_callback_lock
);
1192 transport
->inet
= NULL
;
1193 transport
->sock
= NULL
;
1194 transport
->file
= NULL
;
1196 sk
->sk_user_data
= NULL
;
1198 xs_restore_old_callbacks(transport
, sk
);
1199 xprt_clear_connected(xprt
);
1200 write_unlock_bh(&sk
->sk_callback_lock
);
1201 xs_sock_reset_connection_flags(xprt
);
1202 /* Reset stream record info */
1203 xs_stream_reset_connect(transport
);
1204 mutex_unlock(&transport
->recv_mutex
);
1206 trace_rpc_socket_close(xprt
, sock
);
1209 xprt_disconnect_done(xprt
);
1213 * xs_close - close a socket
1216 * This is used when all requests are complete; ie, no DRC state remains
1217 * on the server we want to save.
1219 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1220 * xs_reset_transport() zeroing the socket from underneath a writer.
1222 static void xs_close(struct rpc_xprt
*xprt
)
1224 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1226 dprintk("RPC: xs_close xprt %p\n", xprt
);
1228 xs_reset_transport(transport
);
1229 xprt
->reestablish_timeout
= 0;
1232 static void xs_inject_disconnect(struct rpc_xprt
*xprt
)
1234 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
1236 xprt_disconnect_done(xprt
);
1239 static void xs_xprt_free(struct rpc_xprt
*xprt
)
1241 xs_free_peer_addresses(xprt
);
1246 * xs_destroy - prepare to shutdown a transport
1247 * @xprt: doomed transport
1250 static void xs_destroy(struct rpc_xprt
*xprt
)
1252 struct sock_xprt
*transport
= container_of(xprt
,
1253 struct sock_xprt
, xprt
);
1254 dprintk("RPC: xs_destroy xprt %p\n", xprt
);
1256 cancel_delayed_work_sync(&transport
->connect_worker
);
1258 cancel_work_sync(&transport
->recv_worker
);
1259 cancel_work_sync(&transport
->error_worker
);
1261 module_put(THIS_MODULE
);
1265 * xs_udp_data_read_skb - receive callback for UDP sockets
1271 static void xs_udp_data_read_skb(struct rpc_xprt
*xprt
,
1273 struct sk_buff
*skb
)
1275 struct rpc_task
*task
;
1276 struct rpc_rqst
*rovr
;
1277 int repsize
, copied
;
1283 dprintk("RPC: impossible RPC reply size %d!\n", repsize
);
1287 /* Copy the XID from the skb... */
1288 xp
= skb_header_pointer(skb
, 0, sizeof(_xid
), &_xid
);
1292 /* Look up and lock the request corresponding to the given XID */
1293 spin_lock(&xprt
->queue_lock
);
1294 rovr
= xprt_lookup_rqst(xprt
, *xp
);
1297 xprt_pin_rqst(rovr
);
1298 xprt_update_rtt(rovr
->rq_task
);
1299 spin_unlock(&xprt
->queue_lock
);
1300 task
= rovr
->rq_task
;
1302 if ((copied
= rovr
->rq_private_buf
.buflen
) > repsize
)
1305 /* Suck it into the iovec, verify checksum if not done by hw. */
1306 if (csum_partial_copy_to_xdr(&rovr
->rq_private_buf
, skb
)) {
1307 spin_lock(&xprt
->queue_lock
);
1308 __UDPX_INC_STATS(sk
, UDP_MIB_INERRORS
);
1313 spin_lock(&xprt
->transport_lock
);
1314 xprt_adjust_cwnd(xprt
, task
, copied
);
1315 spin_unlock(&xprt
->transport_lock
);
1316 spin_lock(&xprt
->queue_lock
);
1317 xprt_complete_rqst(task
, copied
);
1318 __UDPX_INC_STATS(sk
, UDP_MIB_INDATAGRAMS
);
1320 xprt_unpin_rqst(rovr
);
1322 spin_unlock(&xprt
->queue_lock
);
1325 static void xs_udp_data_receive(struct sock_xprt
*transport
)
1327 struct sk_buff
*skb
;
1331 mutex_lock(&transport
->recv_mutex
);
1332 sk
= transport
->inet
;
1336 skb
= skb_recv_udp(sk
, 0, 1, &err
);
1339 xs_udp_data_read_skb(&transport
->xprt
, sk
, skb
);
1343 xs_poll_check_readable(transport
);
1345 mutex_unlock(&transport
->recv_mutex
);
1348 static void xs_udp_data_receive_workfn(struct work_struct
*work
)
1350 struct sock_xprt
*transport
=
1351 container_of(work
, struct sock_xprt
, recv_worker
);
1352 unsigned int pflags
= memalloc_nofs_save();
1354 xs_udp_data_receive(transport
);
1355 memalloc_nofs_restore(pflags
);
1359 * xs_data_ready - "data ready" callback for UDP sockets
1360 * @sk: socket with data to read
1363 static void xs_data_ready(struct sock
*sk
)
1365 struct rpc_xprt
*xprt
;
1367 read_lock_bh(&sk
->sk_callback_lock
);
1368 dprintk("RPC: xs_data_ready...\n");
1369 xprt
= xprt_from_sock(sk
);
1371 struct sock_xprt
*transport
= container_of(xprt
,
1372 struct sock_xprt
, xprt
);
1373 transport
->old_data_ready(sk
);
1374 /* Any data means we had a useful conversation, so
1375 * then we don't need to delay the next reconnect
1377 if (xprt
->reestablish_timeout
)
1378 xprt
->reestablish_timeout
= 0;
1379 if (!test_and_set_bit(XPRT_SOCK_DATA_READY
, &transport
->sock_state
))
1380 queue_work(xprtiod_workqueue
, &transport
->recv_worker
);
1382 read_unlock_bh(&sk
->sk_callback_lock
);
1386 * Helper function to force a TCP close if the server is sending
1387 * junk and/or it has put us in CLOSE_WAIT
1389 static void xs_tcp_force_close(struct rpc_xprt
*xprt
)
1391 xprt_force_disconnect(xprt
);
1394 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1395 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt
*xprt
)
1399 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1402 * xs_tcp_state_change - callback to handle TCP socket state changes
1403 * @sk: socket whose state has changed
1406 static void xs_tcp_state_change(struct sock
*sk
)
1408 struct rpc_xprt
*xprt
;
1409 struct sock_xprt
*transport
;
1411 read_lock_bh(&sk
->sk_callback_lock
);
1412 if (!(xprt
= xprt_from_sock(sk
)))
1414 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt
);
1415 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1416 sk
->sk_state
, xprt_connected(xprt
),
1417 sock_flag(sk
, SOCK_DEAD
),
1418 sock_flag(sk
, SOCK_ZAPPED
),
1421 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1422 trace_rpc_socket_state_change(xprt
, sk
->sk_socket
);
1423 switch (sk
->sk_state
) {
1424 case TCP_ESTABLISHED
:
1425 if (!xprt_test_and_set_connected(xprt
)) {
1426 xprt
->connect_cookie
++;
1427 clear_bit(XPRT_SOCK_CONNECTING
, &transport
->sock_state
);
1428 xprt_clear_connecting(xprt
);
1430 xprt
->stat
.connect_count
++;
1431 xprt
->stat
.connect_time
+= (long)jiffies
-
1432 xprt
->stat
.connect_start
;
1433 xs_run_error_worker(transport
, XPRT_SOCK_WAKE_PENDING
);
1437 /* The client initiated a shutdown of the socket */
1438 xprt
->connect_cookie
++;
1439 xprt
->reestablish_timeout
= 0;
1440 set_bit(XPRT_CLOSING
, &xprt
->state
);
1441 smp_mb__before_atomic();
1442 clear_bit(XPRT_CONNECTED
, &xprt
->state
);
1443 clear_bit(XPRT_CLOSE_WAIT
, &xprt
->state
);
1444 smp_mb__after_atomic();
1446 case TCP_CLOSE_WAIT
:
1447 /* The server initiated a shutdown of the socket */
1448 xprt
->connect_cookie
++;
1449 clear_bit(XPRT_CONNECTED
, &xprt
->state
);
1450 xs_run_error_worker(transport
, XPRT_SOCK_WAKE_DISCONNECT
);
1454 * If the server closed down the connection, make sure that
1455 * we back off before reconnecting
1457 if (xprt
->reestablish_timeout
< XS_TCP_INIT_REEST_TO
)
1458 xprt
->reestablish_timeout
= XS_TCP_INIT_REEST_TO
;
1461 set_bit(XPRT_CLOSING
, &xprt
->state
);
1462 smp_mb__before_atomic();
1463 clear_bit(XPRT_CONNECTED
, &xprt
->state
);
1464 smp_mb__after_atomic();
1467 if (test_and_clear_bit(XPRT_SOCK_CONNECTING
,
1468 &transport
->sock_state
))
1469 xprt_clear_connecting(xprt
);
1470 clear_bit(XPRT_CLOSING
, &xprt
->state
);
1471 /* Trigger the socket release */
1472 xs_run_error_worker(transport
, XPRT_SOCK_WAKE_DISCONNECT
);
1475 read_unlock_bh(&sk
->sk_callback_lock
);
1478 static void xs_write_space(struct sock
*sk
)
1480 struct socket_wq
*wq
;
1481 struct sock_xprt
*transport
;
1482 struct rpc_xprt
*xprt
;
1486 clear_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
1488 if (unlikely(!(xprt
= xprt_from_sock(sk
))))
1490 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1492 wq
= rcu_dereference(sk
->sk_wq
);
1493 if (!wq
|| test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE
, &wq
->flags
) == 0)
1496 xs_run_error_worker(transport
, XPRT_SOCK_WAKE_WRITE
);
1497 sk
->sk_write_pending
--;
1503 * xs_udp_write_space - callback invoked when socket buffer space
1505 * @sk: socket whose state has changed
1507 * Called when more output buffer space is available for this socket.
1508 * We try not to wake our writers until they can make "significant"
1509 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1510 * with a bunch of small requests.
1512 static void xs_udp_write_space(struct sock
*sk
)
1514 read_lock_bh(&sk
->sk_callback_lock
);
1516 /* from net/core/sock.c:sock_def_write_space */
1517 if (sock_writeable(sk
))
1520 read_unlock_bh(&sk
->sk_callback_lock
);
1524 * xs_tcp_write_space - callback invoked when socket buffer space
1526 * @sk: socket whose state has changed
1528 * Called when more output buffer space is available for this socket.
1529 * We try not to wake our writers until they can make "significant"
1530 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1531 * with a bunch of small requests.
1533 static void xs_tcp_write_space(struct sock
*sk
)
1535 read_lock_bh(&sk
->sk_callback_lock
);
1537 /* from net/core/stream.c:sk_stream_write_space */
1538 if (sk_stream_is_writeable(sk
))
1541 read_unlock_bh(&sk
->sk_callback_lock
);
1544 static void xs_udp_do_set_buffer_size(struct rpc_xprt
*xprt
)
1546 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1547 struct sock
*sk
= transport
->inet
;
1549 if (transport
->rcvsize
) {
1550 sk
->sk_userlocks
|= SOCK_RCVBUF_LOCK
;
1551 sk
->sk_rcvbuf
= transport
->rcvsize
* xprt
->max_reqs
* 2;
1553 if (transport
->sndsize
) {
1554 sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
;
1555 sk
->sk_sndbuf
= transport
->sndsize
* xprt
->max_reqs
* 2;
1556 sk
->sk_write_space(sk
);
1561 * xs_udp_set_buffer_size - set send and receive limits
1562 * @xprt: generic transport
1563 * @sndsize: requested size of send buffer, in bytes
1564 * @rcvsize: requested size of receive buffer, in bytes
1566 * Set socket send and receive buffer size limits.
1568 static void xs_udp_set_buffer_size(struct rpc_xprt
*xprt
, size_t sndsize
, size_t rcvsize
)
1570 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1572 transport
->sndsize
= 0;
1574 transport
->sndsize
= sndsize
+ 1024;
1575 transport
->rcvsize
= 0;
1577 transport
->rcvsize
= rcvsize
+ 1024;
1579 xs_udp_do_set_buffer_size(xprt
);
1583 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1584 * @xprt: controlling transport
1585 * @task: task that timed out
1587 * Adjust the congestion window after a retransmit timeout has occurred.
1589 static void xs_udp_timer(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
1591 spin_lock(&xprt
->transport_lock
);
1592 xprt_adjust_cwnd(xprt
, task
, -ETIMEDOUT
);
1593 spin_unlock(&xprt
->transport_lock
);
1596 static int xs_get_random_port(void)
1598 unsigned short min
= xprt_min_resvport
, max
= xprt_max_resvport
;
1599 unsigned short range
;
1600 unsigned short rand
;
1604 range
= max
- min
+ 1;
1605 rand
= (unsigned short) prandom_u32() % range
;
1609 static unsigned short xs_sock_getport(struct socket
*sock
)
1611 struct sockaddr_storage buf
;
1612 unsigned short port
= 0;
1614 if (kernel_getsockname(sock
, (struct sockaddr
*)&buf
) < 0)
1616 switch (buf
.ss_family
) {
1618 port
= ntohs(((struct sockaddr_in6
*)&buf
)->sin6_port
);
1621 port
= ntohs(((struct sockaddr_in
*)&buf
)->sin_port
);
1628 * xs_set_port - reset the port number in the remote endpoint address
1629 * @xprt: generic transport
1630 * @port: new port number
1633 static void xs_set_port(struct rpc_xprt
*xprt
, unsigned short port
)
1635 dprintk("RPC: setting port for xprt %p to %u\n", xprt
, port
);
1637 rpc_set_port(xs_addr(xprt
), port
);
1638 xs_update_peer_port(xprt
);
1641 static void xs_set_srcport(struct sock_xprt
*transport
, struct socket
*sock
)
1643 if (transport
->srcport
== 0 && transport
->xprt
.reuseport
)
1644 transport
->srcport
= xs_sock_getport(sock
);
1647 static int xs_get_srcport(struct sock_xprt
*transport
)
1649 int port
= transport
->srcport
;
1651 if (port
== 0 && transport
->xprt
.resvport
)
1652 port
= xs_get_random_port();
1656 unsigned short get_srcport(struct rpc_xprt
*xprt
)
1658 struct sock_xprt
*sock
= container_of(xprt
, struct sock_xprt
, xprt
);
1659 return xs_sock_getport(sock
->sock
);
1661 EXPORT_SYMBOL(get_srcport
);
1663 static unsigned short xs_next_srcport(struct sock_xprt
*transport
, unsigned short port
)
1665 if (transport
->srcport
!= 0)
1666 transport
->srcport
= 0;
1667 if (!transport
->xprt
.resvport
)
1669 if (port
<= xprt_min_resvport
|| port
> xprt_max_resvport
)
1670 return xprt_max_resvport
;
1673 static int xs_bind(struct sock_xprt
*transport
, struct socket
*sock
)
1675 struct sockaddr_storage myaddr
;
1677 int port
= xs_get_srcport(transport
);
1678 unsigned short last
;
1681 * If we are asking for any ephemeral port (i.e. port == 0 &&
1682 * transport->xprt.resvport == 0), don't bind. Let the local
1683 * port selection happen implicitly when the socket is used
1684 * (for example at connect time).
1686 * This ensures that we can continue to establish TCP
1687 * connections even when all local ephemeral ports are already
1688 * a part of some TCP connection. This makes no difference
1689 * for UDP sockets, but also doesn't harm them.
1691 * If we're asking for any reserved port (i.e. port == 0 &&
1692 * transport->xprt.resvport == 1) xs_get_srcport above will
1693 * ensure that port is non-zero and we will bind as needed.
1698 memcpy(&myaddr
, &transport
->srcaddr
, transport
->xprt
.addrlen
);
1700 rpc_set_port((struct sockaddr
*)&myaddr
, port
);
1701 err
= kernel_bind(sock
, (struct sockaddr
*)&myaddr
,
1702 transport
->xprt
.addrlen
);
1704 if (transport
->xprt
.reuseport
)
1705 transport
->srcport
= port
;
1709 port
= xs_next_srcport(transport
, port
);
1712 } while (err
== -EADDRINUSE
&& nloop
!= 2);
1714 if (myaddr
.ss_family
== AF_INET
)
1715 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__
,
1716 &((struct sockaddr_in
*)&myaddr
)->sin_addr
,
1717 port
, err
? "failed" : "ok", err
);
1719 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__
,
1720 &((struct sockaddr_in6
*)&myaddr
)->sin6_addr
,
1721 port
, err
? "failed" : "ok", err
);
1726 * We don't support autobind on AF_LOCAL sockets
1728 static void xs_local_rpcbind(struct rpc_task
*task
)
1730 xprt_set_bound(task
->tk_xprt
);
1733 static void xs_local_set_port(struct rpc_xprt
*xprt
, unsigned short port
)
1737 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1738 static struct lock_class_key xs_key
[2];
1739 static struct lock_class_key xs_slock_key
[2];
1741 static inline void xs_reclassify_socketu(struct socket
*sock
)
1743 struct sock
*sk
= sock
->sk
;
1745 sock_lock_init_class_and_name(sk
, "slock-AF_LOCAL-RPC",
1746 &xs_slock_key
[1], "sk_lock-AF_LOCAL-RPC", &xs_key
[1]);
1749 static inline void xs_reclassify_socket4(struct socket
*sock
)
1751 struct sock
*sk
= sock
->sk
;
1753 sock_lock_init_class_and_name(sk
, "slock-AF_INET-RPC",
1754 &xs_slock_key
[0], "sk_lock-AF_INET-RPC", &xs_key
[0]);
1757 static inline void xs_reclassify_socket6(struct socket
*sock
)
1759 struct sock
*sk
= sock
->sk
;
1761 sock_lock_init_class_and_name(sk
, "slock-AF_INET6-RPC",
1762 &xs_slock_key
[1], "sk_lock-AF_INET6-RPC", &xs_key
[1]);
1765 static inline void xs_reclassify_socket(int family
, struct socket
*sock
)
1767 if (WARN_ON_ONCE(!sock_allow_reclassification(sock
->sk
)))
1772 xs_reclassify_socketu(sock
);
1775 xs_reclassify_socket4(sock
);
1778 xs_reclassify_socket6(sock
);
1783 static inline void xs_reclassify_socket(int family
, struct socket
*sock
)
1788 static void xs_dummy_setup_socket(struct work_struct
*work
)
1792 static struct socket
*xs_create_sock(struct rpc_xprt
*xprt
,
1793 struct sock_xprt
*transport
, int family
, int type
,
1794 int protocol
, bool reuseport
)
1797 struct socket
*sock
;
1800 err
= __sock_create(xprt
->xprt_net
, family
, type
, protocol
, &sock
, 1);
1802 dprintk("RPC: can't create %d transport socket (%d).\n",
1806 xs_reclassify_socket(family
, sock
);
1809 sock_set_reuseport(sock
->sk
);
1811 err
= xs_bind(transport
, sock
);
1817 filp
= sock_alloc_file(sock
, O_NONBLOCK
, NULL
);
1819 return ERR_CAST(filp
);
1820 transport
->file
= filp
;
1824 return ERR_PTR(err
);
1827 static int xs_local_finish_connecting(struct rpc_xprt
*xprt
,
1828 struct socket
*sock
)
1830 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
,
1833 if (!transport
->inet
) {
1834 struct sock
*sk
= sock
->sk
;
1836 write_lock_bh(&sk
->sk_callback_lock
);
1838 xs_save_old_callbacks(transport
, sk
);
1840 sk
->sk_user_data
= xprt
;
1841 sk
->sk_data_ready
= xs_data_ready
;
1842 sk
->sk_write_space
= xs_udp_write_space
;
1843 sock_set_flag(sk
, SOCK_FASYNC
);
1844 sk
->sk_error_report
= xs_error_report
;
1846 xprt_clear_connected(xprt
);
1848 /* Reset to new socket */
1849 transport
->sock
= sock
;
1850 transport
->inet
= sk
;
1852 write_unlock_bh(&sk
->sk_callback_lock
);
1855 xs_stream_start_connect(transport
);
1857 return kernel_connect(sock
, xs_addr(xprt
), xprt
->addrlen
, 0);
1861 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1862 * @transport: socket transport to connect
1864 static int xs_local_setup_socket(struct sock_xprt
*transport
)
1866 struct rpc_xprt
*xprt
= &transport
->xprt
;
1868 struct socket
*sock
;
1871 status
= __sock_create(xprt
->xprt_net
, AF_LOCAL
,
1872 SOCK_STREAM
, 0, &sock
, 1);
1874 dprintk("RPC: can't create AF_LOCAL "
1875 "transport socket (%d).\n", -status
);
1878 xs_reclassify_socket(AF_LOCAL
, sock
);
1880 filp
= sock_alloc_file(sock
, O_NONBLOCK
, NULL
);
1882 status
= PTR_ERR(filp
);
1885 transport
->file
= filp
;
1887 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1888 xprt
, xprt
->address_strings
[RPC_DISPLAY_ADDR
]);
1890 status
= xs_local_finish_connecting(xprt
, sock
);
1891 trace_rpc_socket_connect(xprt
, sock
, status
);
1894 dprintk("RPC: xprt %p connected to %s\n",
1895 xprt
, xprt
->address_strings
[RPC_DISPLAY_ADDR
]);
1896 xprt
->stat
.connect_count
++;
1897 xprt
->stat
.connect_time
+= (long)jiffies
-
1898 xprt
->stat
.connect_start
;
1899 xprt_set_connected(xprt
);
1904 dprintk("RPC: xprt %p: socket %s does not exist\n",
1905 xprt
, xprt
->address_strings
[RPC_DISPLAY_ADDR
]);
1908 dprintk("RPC: xprt %p: connection refused for %s\n",
1909 xprt
, xprt
->address_strings
[RPC_DISPLAY_ADDR
]);
1912 printk(KERN_ERR
"%s: unhandled error (%d) connecting to %s\n",
1914 xprt
->address_strings
[RPC_DISPLAY_ADDR
]);
1918 xprt_clear_connecting(xprt
);
1919 xprt_wake_pending_tasks(xprt
, status
);
1923 static void xs_local_connect(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
1925 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
1928 if (RPC_IS_ASYNC(task
)) {
1930 * We want the AF_LOCAL connect to be resolved in the
1931 * filesystem namespace of the process making the rpc
1932 * call. Thus we connect synchronously.
1934 * If we want to support asynchronous AF_LOCAL calls,
1935 * we'll need to figure out how to pass a namespace to
1938 task
->tk_rpc_status
= -ENOTCONN
;
1939 rpc_exit(task
, -ENOTCONN
);
1942 ret
= xs_local_setup_socket(transport
);
1943 if (ret
&& !RPC_IS_SOFTCONN(task
))
1944 msleep_interruptible(15000);
1947 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
1949 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
1950 * know that we have exclusive access to the socket), to guard against
1951 * races with xs_reset_transport.
1953 static void xs_set_memalloc(struct rpc_xprt
*xprt
)
1955 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
,
1959 * If there's no sock, then we have nothing to set. The
1960 * reconnecting process will get it for us.
1962 if (!transport
->inet
)
1964 if (atomic_read(&xprt
->swapper
))
1965 sk_set_memalloc(transport
->inet
);
1969 * xs_enable_swap - Tag this transport as being used for swap.
1970 * @xprt: transport to tag
1972 * Take a reference to this transport on behalf of the rpc_clnt, and
1973 * optionally mark it for swapping if it wasn't already.
1976 xs_enable_swap(struct rpc_xprt
*xprt
)
1978 struct sock_xprt
*xs
= container_of(xprt
, struct sock_xprt
, xprt
);
1980 if (atomic_inc_return(&xprt
->swapper
) != 1)
1982 if (wait_on_bit_lock(&xprt
->state
, XPRT_LOCKED
, TASK_KILLABLE
))
1983 return -ERESTARTSYS
;
1985 sk_set_memalloc(xs
->inet
);
1986 xprt_release_xprt(xprt
, NULL
);
1991 * xs_disable_swap - Untag this transport as being used for swap.
1992 * @xprt: transport to tag
1994 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
1995 * swapper refcount goes to 0, untag the socket as a memalloc socket.
1998 xs_disable_swap(struct rpc_xprt
*xprt
)
2000 struct sock_xprt
*xs
= container_of(xprt
, struct sock_xprt
, xprt
);
2002 if (!atomic_dec_and_test(&xprt
->swapper
))
2004 if (wait_on_bit_lock(&xprt
->state
, XPRT_LOCKED
, TASK_KILLABLE
))
2007 sk_clear_memalloc(xs
->inet
);
2008 xprt_release_xprt(xprt
, NULL
);
2011 static void xs_set_memalloc(struct rpc_xprt
*xprt
)
2016 xs_enable_swap(struct rpc_xprt
*xprt
)
2022 xs_disable_swap(struct rpc_xprt
*xprt
)
2027 static void xs_udp_finish_connecting(struct rpc_xprt
*xprt
, struct socket
*sock
)
2029 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2031 if (!transport
->inet
) {
2032 struct sock
*sk
= sock
->sk
;
2034 write_lock_bh(&sk
->sk_callback_lock
);
2036 xs_save_old_callbacks(transport
, sk
);
2038 sk
->sk_user_data
= xprt
;
2039 sk
->sk_data_ready
= xs_data_ready
;
2040 sk
->sk_write_space
= xs_udp_write_space
;
2041 sock_set_flag(sk
, SOCK_FASYNC
);
2043 xprt_set_connected(xprt
);
2045 /* Reset to new socket */
2046 transport
->sock
= sock
;
2047 transport
->inet
= sk
;
2049 xs_set_memalloc(xprt
);
2051 write_unlock_bh(&sk
->sk_callback_lock
);
2053 xs_udp_do_set_buffer_size(xprt
);
2055 xprt
->stat
.connect_start
= jiffies
;
2058 static void xs_udp_setup_socket(struct work_struct
*work
)
2060 struct sock_xprt
*transport
=
2061 container_of(work
, struct sock_xprt
, connect_worker
.work
);
2062 struct rpc_xprt
*xprt
= &transport
->xprt
;
2063 struct socket
*sock
;
2066 sock
= xs_create_sock(xprt
, transport
,
2067 xs_addr(xprt
)->sa_family
, SOCK_DGRAM
,
2068 IPPROTO_UDP
, false);
2072 dprintk("RPC: worker connecting xprt %p via %s to "
2073 "%s (port %s)\n", xprt
,
2074 xprt
->address_strings
[RPC_DISPLAY_PROTO
],
2075 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
2076 xprt
->address_strings
[RPC_DISPLAY_PORT
]);
2078 xs_udp_finish_connecting(xprt
, sock
);
2079 trace_rpc_socket_connect(xprt
, sock
, 0);
2082 xprt_clear_connecting(xprt
);
2083 xprt_unlock_connect(xprt
, transport
);
2084 xprt_wake_pending_tasks(xprt
, status
);
2088 * xs_tcp_shutdown - gracefully shut down a TCP socket
2091 * Initiates a graceful shutdown of the TCP socket by calling the
2092 * equivalent of shutdown(SHUT_RDWR);
2094 static void xs_tcp_shutdown(struct rpc_xprt
*xprt
)
2096 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2097 struct socket
*sock
= transport
->sock
;
2098 int skst
= transport
->inet
? transport
->inet
->sk_state
: TCP_CLOSE
;
2102 if (!xprt
->reuseport
) {
2110 case TCP_ESTABLISHED
:
2111 case TCP_CLOSE_WAIT
:
2112 kernel_sock_shutdown(sock
, SHUT_RDWR
);
2113 trace_rpc_socket_shutdown(xprt
, sock
);
2116 xs_reset_transport(transport
);
2120 static void xs_tcp_set_socket_timeouts(struct rpc_xprt
*xprt
,
2121 struct socket
*sock
)
2123 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2124 unsigned int keepidle
;
2125 unsigned int keepcnt
;
2128 spin_lock(&xprt
->transport_lock
);
2129 keepidle
= DIV_ROUND_UP(xprt
->timeout
->to_initval
, HZ
);
2130 keepcnt
= xprt
->timeout
->to_retries
+ 1;
2131 timeo
= jiffies_to_msecs(xprt
->timeout
->to_initval
) *
2132 (xprt
->timeout
->to_retries
+ 1);
2133 clear_bit(XPRT_SOCK_UPD_TIMEOUT
, &transport
->sock_state
);
2134 spin_unlock(&xprt
->transport_lock
);
2136 /* TCP Keepalive options */
2137 sock_set_keepalive(sock
->sk
);
2138 tcp_sock_set_keepidle(sock
->sk
, keepidle
);
2139 tcp_sock_set_keepintvl(sock
->sk
, keepidle
);
2140 tcp_sock_set_keepcnt(sock
->sk
, keepcnt
);
2142 /* TCP user timeout (see RFC5482) */
2143 tcp_sock_set_user_timeout(sock
->sk
, timeo
);
2146 static void xs_tcp_set_connect_timeout(struct rpc_xprt
*xprt
,
2147 unsigned long connect_timeout
,
2148 unsigned long reconnect_timeout
)
2150 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2151 struct rpc_timeout to
;
2152 unsigned long initval
;
2154 spin_lock(&xprt
->transport_lock
);
2155 if (reconnect_timeout
< xprt
->max_reconnect_timeout
)
2156 xprt
->max_reconnect_timeout
= reconnect_timeout
;
2157 if (connect_timeout
< xprt
->connect_timeout
) {
2158 memcpy(&to
, xprt
->timeout
, sizeof(to
));
2159 initval
= DIV_ROUND_UP(connect_timeout
, to
.to_retries
+ 1);
2160 /* Arbitrary lower limit */
2161 if (initval
< XS_TCP_INIT_REEST_TO
<< 1)
2162 initval
= XS_TCP_INIT_REEST_TO
<< 1;
2163 to
.to_initval
= initval
;
2164 to
.to_maxval
= initval
;
2165 memcpy(&transport
->tcp_timeout
, &to
,
2166 sizeof(transport
->tcp_timeout
));
2167 xprt
->timeout
= &transport
->tcp_timeout
;
2168 xprt
->connect_timeout
= connect_timeout
;
2170 set_bit(XPRT_SOCK_UPD_TIMEOUT
, &transport
->sock_state
);
2171 spin_unlock(&xprt
->transport_lock
);
2174 static int xs_tcp_finish_connecting(struct rpc_xprt
*xprt
, struct socket
*sock
)
2176 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2177 int ret
= -ENOTCONN
;
2179 if (!transport
->inet
) {
2180 struct sock
*sk
= sock
->sk
;
2182 /* Avoid temporary address, they are bad for long-lived
2183 * connections such as NFS mounts.
2184 * RFC4941, section 3.6 suggests that:
2185 * Individual applications, which have specific
2186 * knowledge about the normal duration of connections,
2187 * MAY override this as appropriate.
2189 if (xs_addr(xprt
)->sa_family
== PF_INET6
) {
2190 ip6_sock_set_addr_preferences(sk
,
2191 IPV6_PREFER_SRC_PUBLIC
);
2194 xs_tcp_set_socket_timeouts(xprt
, sock
);
2195 tcp_sock_set_nodelay(sk
);
2197 write_lock_bh(&sk
->sk_callback_lock
);
2199 xs_save_old_callbacks(transport
, sk
);
2201 sk
->sk_user_data
= xprt
;
2202 sk
->sk_data_ready
= xs_data_ready
;
2203 sk
->sk_state_change
= xs_tcp_state_change
;
2204 sk
->sk_write_space
= xs_tcp_write_space
;
2205 sock_set_flag(sk
, SOCK_FASYNC
);
2206 sk
->sk_error_report
= xs_error_report
;
2208 /* socket options */
2209 sock_reset_flag(sk
, SOCK_LINGER
);
2211 xprt_clear_connected(xprt
);
2213 /* Reset to new socket */
2214 transport
->sock
= sock
;
2215 transport
->inet
= sk
;
2217 write_unlock_bh(&sk
->sk_callback_lock
);
2220 if (!xprt_bound(xprt
))
2223 xs_set_memalloc(xprt
);
2225 xs_stream_start_connect(transport
);
2227 /* Tell the socket layer to start connecting... */
2228 set_bit(XPRT_SOCK_CONNECTING
, &transport
->sock_state
);
2229 ret
= kernel_connect(sock
, xs_addr(xprt
), xprt
->addrlen
, O_NONBLOCK
);
2232 xs_set_srcport(transport
, sock
);
2236 if (xprt
->reestablish_timeout
< XS_TCP_INIT_REEST_TO
)
2237 xprt
->reestablish_timeout
= XS_TCP_INIT_REEST_TO
;
2239 case -EADDRNOTAVAIL
:
2240 /* Source port number is unavailable. Try a new one! */
2241 transport
->srcport
= 0;
2248 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2249 * @work: queued work item
2251 * Invoked by a work queue tasklet.
2253 static void xs_tcp_setup_socket(struct work_struct
*work
)
2255 struct sock_xprt
*transport
=
2256 container_of(work
, struct sock_xprt
, connect_worker
.work
);
2257 struct socket
*sock
= transport
->sock
;
2258 struct rpc_xprt
*xprt
= &transport
->xprt
;
2262 sock
= xs_create_sock(xprt
, transport
,
2263 xs_addr(xprt
)->sa_family
, SOCK_STREAM
,
2266 status
= PTR_ERR(sock
);
2271 dprintk("RPC: worker connecting xprt %p via %s to "
2272 "%s (port %s)\n", xprt
,
2273 xprt
->address_strings
[RPC_DISPLAY_PROTO
],
2274 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
2275 xprt
->address_strings
[RPC_DISPLAY_PORT
]);
2277 status
= xs_tcp_finish_connecting(xprt
, sock
);
2278 trace_rpc_socket_connect(xprt
, sock
, status
);
2279 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2280 xprt
, -status
, xprt_connected(xprt
),
2281 sock
->sk
->sk_state
);
2284 printk("%s: connect returned unhandled error %d\n",
2287 case -EADDRNOTAVAIL
:
2288 /* We're probably in TIME_WAIT. Get rid of existing socket,
2291 xs_tcp_force_close(xprt
);
2296 xprt_unlock_connect(xprt
, transport
);
2299 /* Happens, for instance, if the user specified a link
2300 * local IPv6 address without a scope-id.
2309 /* xs_tcp_force_close() wakes tasks with a fixed error code.
2310 * We need to wake them first to ensure the correct error code.
2312 xprt_wake_pending_tasks(xprt
, status
);
2313 xs_tcp_force_close(xprt
);
2318 xprt_clear_connecting(xprt
);
2319 xprt_unlock_connect(xprt
, transport
);
2320 xprt_wake_pending_tasks(xprt
, status
);
2324 * xs_connect - connect a socket to a remote endpoint
2325 * @xprt: pointer to transport structure
2326 * @task: address of RPC task that manages state of connect request
2328 * TCP: If the remote end dropped the connection, delay reconnecting.
2330 * UDP socket connects are synchronous, but we use a work queue anyway
2331 * to guarantee that even unprivileged user processes can set up a
2332 * socket on a privileged port.
2334 * If a UDP socket connect fails, the delay behavior here prevents
2335 * retry floods (hard mounts).
2337 static void xs_connect(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
2339 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2340 unsigned long delay
= 0;
2342 WARN_ON_ONCE(!xprt_lock_connect(xprt
, task
, transport
));
2344 if (transport
->sock
!= NULL
) {
2345 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2347 xprt
, xprt
->reestablish_timeout
/ HZ
);
2349 /* Start by resetting any existing state */
2350 xs_reset_transport(transport
);
2352 delay
= xprt_reconnect_delay(xprt
);
2353 xprt_reconnect_backoff(xprt
, XS_TCP_INIT_REEST_TO
);
2356 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt
);
2358 queue_delayed_work(xprtiod_workqueue
,
2359 &transport
->connect_worker
,
2363 static void xs_wake_disconnect(struct sock_xprt
*transport
)
2365 if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT
, &transport
->sock_state
))
2366 xs_tcp_force_close(&transport
->xprt
);
2369 static void xs_wake_write(struct sock_xprt
*transport
)
2371 if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE
, &transport
->sock_state
))
2372 xprt_write_space(&transport
->xprt
);
2375 static void xs_wake_error(struct sock_xprt
*transport
)
2379 if (!test_bit(XPRT_SOCK_WAKE_ERROR
, &transport
->sock_state
))
2381 mutex_lock(&transport
->recv_mutex
);
2382 if (transport
->sock
== NULL
)
2384 if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR
, &transport
->sock_state
))
2386 sockerr
= xchg(&transport
->xprt_err
, 0);
2388 xprt_wake_pending_tasks(&transport
->xprt
, sockerr
);
2390 mutex_unlock(&transport
->recv_mutex
);
2393 static void xs_wake_pending(struct sock_xprt
*transport
)
2395 if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING
, &transport
->sock_state
))
2396 xprt_wake_pending_tasks(&transport
->xprt
, -EAGAIN
);
2399 static void xs_error_handle(struct work_struct
*work
)
2401 struct sock_xprt
*transport
= container_of(work
,
2402 struct sock_xprt
, error_worker
);
2404 xs_wake_disconnect(transport
);
2405 xs_wake_write(transport
);
2406 xs_wake_error(transport
);
2407 xs_wake_pending(transport
);
2411 * xs_local_print_stats - display AF_LOCAL socket-specific stats
2412 * @xprt: rpc_xprt struct containing statistics
2416 static void xs_local_print_stats(struct rpc_xprt
*xprt
, struct seq_file
*seq
)
2420 if (xprt_connected(xprt
))
2421 idle_time
= (long)(jiffies
- xprt
->last_used
) / HZ
;
2423 seq_printf(seq
, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2424 "%llu %llu %lu %llu %llu\n",
2425 xprt
->stat
.bind_count
,
2426 xprt
->stat
.connect_count
,
2427 xprt
->stat
.connect_time
/ HZ
,
2431 xprt
->stat
.bad_xids
,
2434 xprt
->stat
.max_slots
,
2435 xprt
->stat
.sending_u
,
2436 xprt
->stat
.pending_u
);
2440 * xs_udp_print_stats - display UDP socket-specific stats
2441 * @xprt: rpc_xprt struct containing statistics
2445 static void xs_udp_print_stats(struct rpc_xprt
*xprt
, struct seq_file
*seq
)
2447 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2449 seq_printf(seq
, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2452 xprt
->stat
.bind_count
,
2455 xprt
->stat
.bad_xids
,
2458 xprt
->stat
.max_slots
,
2459 xprt
->stat
.sending_u
,
2460 xprt
->stat
.pending_u
);
2464 * xs_tcp_print_stats - display TCP socket-specific stats
2465 * @xprt: rpc_xprt struct containing statistics
2469 static void xs_tcp_print_stats(struct rpc_xprt
*xprt
, struct seq_file
*seq
)
2471 struct sock_xprt
*transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2474 if (xprt_connected(xprt
))
2475 idle_time
= (long)(jiffies
- xprt
->last_used
) / HZ
;
2477 seq_printf(seq
, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2478 "%llu %llu %lu %llu %llu\n",
2480 xprt
->stat
.bind_count
,
2481 xprt
->stat
.connect_count
,
2482 xprt
->stat
.connect_time
/ HZ
,
2486 xprt
->stat
.bad_xids
,
2489 xprt
->stat
.max_slots
,
2490 xprt
->stat
.sending_u
,
2491 xprt
->stat
.pending_u
);
2495 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2496 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2497 * to use the server side send routines.
2499 static int bc_malloc(struct rpc_task
*task
)
2501 struct rpc_rqst
*rqst
= task
->tk_rqstp
;
2502 size_t size
= rqst
->rq_callsize
;
2504 struct rpc_buffer
*buf
;
2506 if (size
> PAGE_SIZE
- sizeof(struct rpc_buffer
)) {
2507 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2512 page
= alloc_page(GFP_KERNEL
);
2516 buf
= page_address(page
);
2517 buf
->len
= PAGE_SIZE
;
2519 rqst
->rq_buffer
= buf
->data
;
2520 rqst
->rq_rbuffer
= (char *)rqst
->rq_buffer
+ rqst
->rq_callsize
;
2525 * Free the space allocated in the bc_alloc routine
2527 static void bc_free(struct rpc_task
*task
)
2529 void *buffer
= task
->tk_rqstp
->rq_buffer
;
2530 struct rpc_buffer
*buf
;
2532 buf
= container_of(buffer
, struct rpc_buffer
, data
);
2533 free_page((unsigned long)buf
);
2536 static int bc_sendto(struct rpc_rqst
*req
)
2538 struct xdr_buf
*xdr
= &req
->rq_snd_buf
;
2539 struct sock_xprt
*transport
=
2540 container_of(req
->rq_xprt
, struct sock_xprt
, xprt
);
2541 struct msghdr msg
= {
2544 rpc_fraghdr marker
= cpu_to_be32(RPC_LAST_STREAM_FRAGMENT
|
2546 unsigned int sent
= 0;
2549 req
->rq_xtime
= ktime_get();
2550 err
= xprt_sock_sendmsg(transport
->sock
, &msg
, xdr
, 0, marker
, &sent
);
2552 if (err
< 0 || sent
!= (xdr
->len
+ sizeof(marker
)))
2558 * bc_send_request - Send a backchannel Call on a TCP socket
2559 * @req: rpc_rqst containing Call message to be sent
2561 * xpt_mutex ensures @rqstp's whole message is written to the socket
2562 * without interruption.
2565 * %0 if the message was sent successfully
2566 * %ENOTCONN if the message was not sent
2568 static int bc_send_request(struct rpc_rqst
*req
)
2570 struct svc_xprt
*xprt
;
2574 * Get the server socket associated with this callback xprt
2576 xprt
= req
->rq_xprt
->bc_xprt
;
2579 * Grab the mutex to serialize data as the connection is shared
2580 * with the fore channel
2582 mutex_lock(&xprt
->xpt_mutex
);
2583 if (test_bit(XPT_DEAD
, &xprt
->xpt_flags
))
2586 len
= bc_sendto(req
);
2587 mutex_unlock(&xprt
->xpt_mutex
);
2596 * The close routine. Since this is client initiated, we do nothing
2599 static void bc_close(struct rpc_xprt
*xprt
)
2601 xprt_disconnect_done(xprt
);
2605 * The xprt destroy routine. Again, because this connection is client
2606 * initiated, we do nothing
2609 static void bc_destroy(struct rpc_xprt
*xprt
)
2611 dprintk("RPC: bc_destroy xprt %p\n", xprt
);
2614 module_put(THIS_MODULE
);
2617 static const struct rpc_xprt_ops xs_local_ops
= {
2618 .reserve_xprt
= xprt_reserve_xprt
,
2619 .release_xprt
= xprt_release_xprt
,
2620 .alloc_slot
= xprt_alloc_slot
,
2621 .free_slot
= xprt_free_slot
,
2622 .rpcbind
= xs_local_rpcbind
,
2623 .set_port
= xs_local_set_port
,
2624 .connect
= xs_local_connect
,
2625 .buf_alloc
= rpc_malloc
,
2626 .buf_free
= rpc_free
,
2627 .prepare_request
= xs_stream_prepare_request
,
2628 .send_request
= xs_local_send_request
,
2629 .wait_for_reply_request
= xprt_wait_for_reply_request_def
,
2631 .destroy
= xs_destroy
,
2632 .print_stats
= xs_local_print_stats
,
2633 .enable_swap
= xs_enable_swap
,
2634 .disable_swap
= xs_disable_swap
,
2637 static const struct rpc_xprt_ops xs_udp_ops
= {
2638 .set_buffer_size
= xs_udp_set_buffer_size
,
2639 .reserve_xprt
= xprt_reserve_xprt_cong
,
2640 .release_xprt
= xprt_release_xprt_cong
,
2641 .alloc_slot
= xprt_alloc_slot
,
2642 .free_slot
= xprt_free_slot
,
2643 .rpcbind
= rpcb_getport_async
,
2644 .set_port
= xs_set_port
,
2645 .connect
= xs_connect
,
2646 .buf_alloc
= rpc_malloc
,
2647 .buf_free
= rpc_free
,
2648 .send_request
= xs_udp_send_request
,
2649 .wait_for_reply_request
= xprt_wait_for_reply_request_rtt
,
2650 .timer
= xs_udp_timer
,
2651 .release_request
= xprt_release_rqst_cong
,
2653 .destroy
= xs_destroy
,
2654 .print_stats
= xs_udp_print_stats
,
2655 .enable_swap
= xs_enable_swap
,
2656 .disable_swap
= xs_disable_swap
,
2657 .inject_disconnect
= xs_inject_disconnect
,
2660 static const struct rpc_xprt_ops xs_tcp_ops
= {
2661 .reserve_xprt
= xprt_reserve_xprt
,
2662 .release_xprt
= xprt_release_xprt
,
2663 .alloc_slot
= xprt_alloc_slot
,
2664 .free_slot
= xprt_free_slot
,
2665 .rpcbind
= rpcb_getport_async
,
2666 .set_port
= xs_set_port
,
2667 .connect
= xs_connect
,
2668 .buf_alloc
= rpc_malloc
,
2669 .buf_free
= rpc_free
,
2670 .prepare_request
= xs_stream_prepare_request
,
2671 .send_request
= xs_tcp_send_request
,
2672 .wait_for_reply_request
= xprt_wait_for_reply_request_def
,
2673 .close
= xs_tcp_shutdown
,
2674 .destroy
= xs_destroy
,
2675 .set_connect_timeout
= xs_tcp_set_connect_timeout
,
2676 .print_stats
= xs_tcp_print_stats
,
2677 .enable_swap
= xs_enable_swap
,
2678 .disable_swap
= xs_disable_swap
,
2679 .inject_disconnect
= xs_inject_disconnect
,
2680 #ifdef CONFIG_SUNRPC_BACKCHANNEL
2681 .bc_setup
= xprt_setup_bc
,
2682 .bc_maxpayload
= xs_tcp_bc_maxpayload
,
2683 .bc_num_slots
= xprt_bc_max_slots
,
2684 .bc_free_rqst
= xprt_free_bc_rqst
,
2685 .bc_destroy
= xprt_destroy_bc
,
2690 * The rpc_xprt_ops for the server backchannel
2693 static const struct rpc_xprt_ops bc_tcp_ops
= {
2694 .reserve_xprt
= xprt_reserve_xprt
,
2695 .release_xprt
= xprt_release_xprt
,
2696 .alloc_slot
= xprt_alloc_slot
,
2697 .free_slot
= xprt_free_slot
,
2698 .buf_alloc
= bc_malloc
,
2699 .buf_free
= bc_free
,
2700 .send_request
= bc_send_request
,
2701 .wait_for_reply_request
= xprt_wait_for_reply_request_def
,
2703 .destroy
= bc_destroy
,
2704 .print_stats
= xs_tcp_print_stats
,
2705 .enable_swap
= xs_enable_swap
,
2706 .disable_swap
= xs_disable_swap
,
2707 .inject_disconnect
= xs_inject_disconnect
,
2710 static int xs_init_anyaddr(const int family
, struct sockaddr
*sap
)
2712 static const struct sockaddr_in sin
= {
2713 .sin_family
= AF_INET
,
2714 .sin_addr
.s_addr
= htonl(INADDR_ANY
),
2716 static const struct sockaddr_in6 sin6
= {
2717 .sin6_family
= AF_INET6
,
2718 .sin6_addr
= IN6ADDR_ANY_INIT
,
2725 memcpy(sap
, &sin
, sizeof(sin
));
2728 memcpy(sap
, &sin6
, sizeof(sin6
));
2731 dprintk("RPC: %s: Bad address family\n", __func__
);
2732 return -EAFNOSUPPORT
;
2737 static struct rpc_xprt
*xs_setup_xprt(struct xprt_create
*args
,
2738 unsigned int slot_table_size
,
2739 unsigned int max_slot_table_size
)
2741 struct rpc_xprt
*xprt
;
2742 struct sock_xprt
*new;
2744 if (args
->addrlen
> sizeof(xprt
->addr
)) {
2745 dprintk("RPC: xs_setup_xprt: address too large\n");
2746 return ERR_PTR(-EBADF
);
2749 xprt
= xprt_alloc(args
->net
, sizeof(*new), slot_table_size
,
2750 max_slot_table_size
);
2752 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2754 return ERR_PTR(-ENOMEM
);
2757 new = container_of(xprt
, struct sock_xprt
, xprt
);
2758 mutex_init(&new->recv_mutex
);
2759 memcpy(&xprt
->addr
, args
->dstaddr
, args
->addrlen
);
2760 xprt
->addrlen
= args
->addrlen
;
2762 memcpy(&new->srcaddr
, args
->srcaddr
, args
->addrlen
);
2765 err
= xs_init_anyaddr(args
->dstaddr
->sa_family
,
2766 (struct sockaddr
*)&new->srcaddr
);
2769 return ERR_PTR(err
);
2776 static const struct rpc_timeout xs_local_default_timeout
= {
2777 .to_initval
= 10 * HZ
,
2778 .to_maxval
= 10 * HZ
,
2783 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2784 * @args: rpc transport creation arguments
2786 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2788 static struct rpc_xprt
*xs_setup_local(struct xprt_create
*args
)
2790 struct sockaddr_un
*sun
= (struct sockaddr_un
*)args
->dstaddr
;
2791 struct sock_xprt
*transport
;
2792 struct rpc_xprt
*xprt
;
2793 struct rpc_xprt
*ret
;
2795 xprt
= xs_setup_xprt(args
, xprt_tcp_slot_table_entries
,
2796 xprt_max_tcp_slot_table_entries
);
2799 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2802 xprt
->xprt_class
= &xs_local_transport
;
2803 xprt
->max_payload
= RPC_MAX_FRAGMENT_SIZE
;
2805 xprt
->bind_timeout
= XS_BIND_TO
;
2806 xprt
->reestablish_timeout
= XS_TCP_INIT_REEST_TO
;
2807 xprt
->idle_timeout
= XS_IDLE_DISC_TO
;
2809 xprt
->ops
= &xs_local_ops
;
2810 xprt
->timeout
= &xs_local_default_timeout
;
2812 INIT_WORK(&transport
->recv_worker
, xs_stream_data_receive_workfn
);
2813 INIT_WORK(&transport
->error_worker
, xs_error_handle
);
2814 INIT_DELAYED_WORK(&transport
->connect_worker
, xs_dummy_setup_socket
);
2816 switch (sun
->sun_family
) {
2818 if (sun
->sun_path
[0] != '/') {
2819 dprintk("RPC: bad AF_LOCAL address: %s\n",
2821 ret
= ERR_PTR(-EINVAL
);
2824 xprt_set_bound(xprt
);
2825 xs_format_peer_addresses(xprt
, "local", RPCBIND_NETID_LOCAL
);
2826 ret
= ERR_PTR(xs_local_setup_socket(transport
));
2831 ret
= ERR_PTR(-EAFNOSUPPORT
);
2835 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2836 xprt
->address_strings
[RPC_DISPLAY_ADDR
]);
2838 if (try_module_get(THIS_MODULE
))
2840 ret
= ERR_PTR(-EINVAL
);
2846 static const struct rpc_timeout xs_udp_default_timeout
= {
2847 .to_initval
= 5 * HZ
,
2848 .to_maxval
= 30 * HZ
,
2849 .to_increment
= 5 * HZ
,
2854 * xs_setup_udp - Set up transport to use a UDP socket
2855 * @args: rpc transport creation arguments
2858 static struct rpc_xprt
*xs_setup_udp(struct xprt_create
*args
)
2860 struct sockaddr
*addr
= args
->dstaddr
;
2861 struct rpc_xprt
*xprt
;
2862 struct sock_xprt
*transport
;
2863 struct rpc_xprt
*ret
;
2865 xprt
= xs_setup_xprt(args
, xprt_udp_slot_table_entries
,
2866 xprt_udp_slot_table_entries
);
2869 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2871 xprt
->prot
= IPPROTO_UDP
;
2872 xprt
->xprt_class
= &xs_udp_transport
;
2873 /* XXX: header size can vary due to auth type, IPv6, etc. */
2874 xprt
->max_payload
= (1U << 16) - (MAX_HEADER
<< 3);
2876 xprt
->bind_timeout
= XS_BIND_TO
;
2877 xprt
->reestablish_timeout
= XS_UDP_REEST_TO
;
2878 xprt
->idle_timeout
= XS_IDLE_DISC_TO
;
2880 xprt
->ops
= &xs_udp_ops
;
2882 xprt
->timeout
= &xs_udp_default_timeout
;
2884 INIT_WORK(&transport
->recv_worker
, xs_udp_data_receive_workfn
);
2885 INIT_WORK(&transport
->error_worker
, xs_error_handle
);
2886 INIT_DELAYED_WORK(&transport
->connect_worker
, xs_udp_setup_socket
);
2888 switch (addr
->sa_family
) {
2890 if (((struct sockaddr_in
*)addr
)->sin_port
!= htons(0))
2891 xprt_set_bound(xprt
);
2893 xs_format_peer_addresses(xprt
, "udp", RPCBIND_NETID_UDP
);
2896 if (((struct sockaddr_in6
*)addr
)->sin6_port
!= htons(0))
2897 xprt_set_bound(xprt
);
2899 xs_format_peer_addresses(xprt
, "udp", RPCBIND_NETID_UDP6
);
2902 ret
= ERR_PTR(-EAFNOSUPPORT
);
2906 if (xprt_bound(xprt
))
2907 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2908 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
2909 xprt
->address_strings
[RPC_DISPLAY_PORT
],
2910 xprt
->address_strings
[RPC_DISPLAY_PROTO
]);
2912 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2913 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
2914 xprt
->address_strings
[RPC_DISPLAY_PROTO
]);
2916 if (try_module_get(THIS_MODULE
))
2918 ret
= ERR_PTR(-EINVAL
);
2924 static const struct rpc_timeout xs_tcp_default_timeout
= {
2925 .to_initval
= 60 * HZ
,
2926 .to_maxval
= 60 * HZ
,
2931 * xs_setup_tcp - Set up transport to use a TCP socket
2932 * @args: rpc transport creation arguments
2935 static struct rpc_xprt
*xs_setup_tcp(struct xprt_create
*args
)
2937 struct sockaddr
*addr
= args
->dstaddr
;
2938 struct rpc_xprt
*xprt
;
2939 struct sock_xprt
*transport
;
2940 struct rpc_xprt
*ret
;
2941 unsigned int max_slot_table_size
= xprt_max_tcp_slot_table_entries
;
2943 if (args
->flags
& XPRT_CREATE_INFINITE_SLOTS
)
2944 max_slot_table_size
= RPC_MAX_SLOT_TABLE_LIMIT
;
2946 xprt
= xs_setup_xprt(args
, xprt_tcp_slot_table_entries
,
2947 max_slot_table_size
);
2950 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
2952 xprt
->prot
= IPPROTO_TCP
;
2953 xprt
->xprt_class
= &xs_tcp_transport
;
2954 xprt
->max_payload
= RPC_MAX_FRAGMENT_SIZE
;
2956 xprt
->bind_timeout
= XS_BIND_TO
;
2957 xprt
->reestablish_timeout
= XS_TCP_INIT_REEST_TO
;
2958 xprt
->idle_timeout
= XS_IDLE_DISC_TO
;
2960 xprt
->ops
= &xs_tcp_ops
;
2961 xprt
->timeout
= &xs_tcp_default_timeout
;
2963 xprt
->max_reconnect_timeout
= xprt
->timeout
->to_maxval
;
2964 xprt
->connect_timeout
= xprt
->timeout
->to_initval
*
2965 (xprt
->timeout
->to_retries
+ 1);
2967 INIT_WORK(&transport
->recv_worker
, xs_stream_data_receive_workfn
);
2968 INIT_WORK(&transport
->error_worker
, xs_error_handle
);
2969 INIT_DELAYED_WORK(&transport
->connect_worker
, xs_tcp_setup_socket
);
2971 switch (addr
->sa_family
) {
2973 if (((struct sockaddr_in
*)addr
)->sin_port
!= htons(0))
2974 xprt_set_bound(xprt
);
2976 xs_format_peer_addresses(xprt
, "tcp", RPCBIND_NETID_TCP
);
2979 if (((struct sockaddr_in6
*)addr
)->sin6_port
!= htons(0))
2980 xprt_set_bound(xprt
);
2982 xs_format_peer_addresses(xprt
, "tcp", RPCBIND_NETID_TCP6
);
2985 ret
= ERR_PTR(-EAFNOSUPPORT
);
2989 if (xprt_bound(xprt
))
2990 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2991 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
2992 xprt
->address_strings
[RPC_DISPLAY_PORT
],
2993 xprt
->address_strings
[RPC_DISPLAY_PROTO
]);
2995 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2996 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
2997 xprt
->address_strings
[RPC_DISPLAY_PROTO
]);
2999 if (try_module_get(THIS_MODULE
))
3001 ret
= ERR_PTR(-EINVAL
);
3008 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3009 * @args: rpc transport creation arguments
3012 static struct rpc_xprt
*xs_setup_bc_tcp(struct xprt_create
*args
)
3014 struct sockaddr
*addr
= args
->dstaddr
;
3015 struct rpc_xprt
*xprt
;
3016 struct sock_xprt
*transport
;
3017 struct svc_sock
*bc_sock
;
3018 struct rpc_xprt
*ret
;
3020 xprt
= xs_setup_xprt(args
, xprt_tcp_slot_table_entries
,
3021 xprt_tcp_slot_table_entries
);
3024 transport
= container_of(xprt
, struct sock_xprt
, xprt
);
3026 xprt
->prot
= IPPROTO_TCP
;
3027 xprt
->xprt_class
= &xs_bc_tcp_transport
;
3028 xprt
->max_payload
= RPC_MAX_FRAGMENT_SIZE
;
3029 xprt
->timeout
= &xs_tcp_default_timeout
;
3032 xprt_set_bound(xprt
);
3033 xprt
->bind_timeout
= 0;
3034 xprt
->reestablish_timeout
= 0;
3035 xprt
->idle_timeout
= 0;
3037 xprt
->ops
= &bc_tcp_ops
;
3039 switch (addr
->sa_family
) {
3041 xs_format_peer_addresses(xprt
, "tcp",
3045 xs_format_peer_addresses(xprt
, "tcp",
3046 RPCBIND_NETID_TCP6
);
3049 ret
= ERR_PTR(-EAFNOSUPPORT
);
3053 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3054 xprt
->address_strings
[RPC_DISPLAY_ADDR
],
3055 xprt
->address_strings
[RPC_DISPLAY_PORT
],
3056 xprt
->address_strings
[RPC_DISPLAY_PROTO
]);
3059 * Once we've associated a backchannel xprt with a connection,
3060 * we want to keep it around as long as the connection lasts,
3061 * in case we need to start using it for a backchannel again;
3062 * this reference won't be dropped until bc_xprt is destroyed.
3065 args
->bc_xprt
->xpt_bc_xprt
= xprt
;
3066 xprt
->bc_xprt
= args
->bc_xprt
;
3067 bc_sock
= container_of(args
->bc_xprt
, struct svc_sock
, sk_xprt
);
3068 transport
->sock
= bc_sock
->sk_sock
;
3069 transport
->inet
= bc_sock
->sk_sk
;
3072 * Since we don't want connections for the backchannel, we set
3073 * the xprt status to connected
3075 xprt_set_connected(xprt
);
3077 if (try_module_get(THIS_MODULE
))
3080 args
->bc_xprt
->xpt_bc_xprt
= NULL
;
3081 args
->bc_xprt
->xpt_bc_xps
= NULL
;
3083 ret
= ERR_PTR(-EINVAL
);
3089 static struct xprt_class xs_local_transport
= {
3090 .list
= LIST_HEAD_INIT(xs_local_transport
.list
),
3091 .name
= "named UNIX socket",
3092 .owner
= THIS_MODULE
,
3093 .ident
= XPRT_TRANSPORT_LOCAL
,
3094 .setup
= xs_setup_local
,
3098 static struct xprt_class xs_udp_transport
= {
3099 .list
= LIST_HEAD_INIT(xs_udp_transport
.list
),
3101 .owner
= THIS_MODULE
,
3102 .ident
= XPRT_TRANSPORT_UDP
,
3103 .setup
= xs_setup_udp
,
3104 .netid
= { "udp", "udp6", "" },
3107 static struct xprt_class xs_tcp_transport
= {
3108 .list
= LIST_HEAD_INIT(xs_tcp_transport
.list
),
3110 .owner
= THIS_MODULE
,
3111 .ident
= XPRT_TRANSPORT_TCP
,
3112 .setup
= xs_setup_tcp
,
3113 .netid
= { "tcp", "tcp6", "" },
3116 static struct xprt_class xs_bc_tcp_transport
= {
3117 .list
= LIST_HEAD_INIT(xs_bc_tcp_transport
.list
),
3118 .name
= "tcp NFSv4.1 backchannel",
3119 .owner
= THIS_MODULE
,
3120 .ident
= XPRT_TRANSPORT_BC_TCP
,
3121 .setup
= xs_setup_bc_tcp
,
3126 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3129 int init_socket_xprt(void)
3131 if (!sunrpc_table_header
)
3132 sunrpc_table_header
= register_sysctl_table(sunrpc_table
);
3134 xprt_register_transport(&xs_local_transport
);
3135 xprt_register_transport(&xs_udp_transport
);
3136 xprt_register_transport(&xs_tcp_transport
);
3137 xprt_register_transport(&xs_bc_tcp_transport
);
3143 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3146 void cleanup_socket_xprt(void)
3148 if (sunrpc_table_header
) {
3149 unregister_sysctl_table(sunrpc_table_header
);
3150 sunrpc_table_header
= NULL
;
3153 xprt_unregister_transport(&xs_local_transport
);
3154 xprt_unregister_transport(&xs_udp_transport
);
3155 xprt_unregister_transport(&xs_tcp_transport
);
3156 xprt_unregister_transport(&xs_bc_tcp_transport
);
3159 static int param_set_portnr(const char *val
, const struct kernel_param
*kp
)
3161 return param_set_uint_minmax(val
, kp
,
3166 static const struct kernel_param_ops param_ops_portnr
= {
3167 .set
= param_set_portnr
,
3168 .get
= param_get_uint
,
3171 #define param_check_portnr(name, p) \
3172 __param_check(name, p, unsigned int);
3174 module_param_named(min_resvport
, xprt_min_resvport
, portnr
, 0644);
3175 module_param_named(max_resvport
, xprt_max_resvport
, portnr
, 0644);
3177 static int param_set_slot_table_size(const char *val
,
3178 const struct kernel_param
*kp
)
3180 return param_set_uint_minmax(val
, kp
,
3182 RPC_MAX_SLOT_TABLE
);
3185 static const struct kernel_param_ops param_ops_slot_table_size
= {
3186 .set
= param_set_slot_table_size
,
3187 .get
= param_get_uint
,
3190 #define param_check_slot_table_size(name, p) \
3191 __param_check(name, p, unsigned int);
3193 static int param_set_max_slot_table_size(const char *val
,
3194 const struct kernel_param
*kp
)
3196 return param_set_uint_minmax(val
, kp
,
3198 RPC_MAX_SLOT_TABLE_LIMIT
);
3201 static const struct kernel_param_ops param_ops_max_slot_table_size
= {
3202 .set
= param_set_max_slot_table_size
,
3203 .get
= param_get_uint
,
3206 #define param_check_max_slot_table_size(name, p) \
3207 __param_check(name, p, unsigned int);
3209 module_param_named(tcp_slot_table_entries
, xprt_tcp_slot_table_entries
,
3210 slot_table_size
, 0644);
3211 module_param_named(tcp_max_slot_table_entries
, xprt_max_tcp_slot_table_entries
,
3212 max_slot_table_size
, 0644);
3213 module_param_named(udp_slot_table_entries
, xprt_udp_slot_table_entries
,
3214 slot_table_size
, 0644);