2 * linux/net/sunrpc/xprtsock.c
4 * Client-side transport implementation for sockets.
6 * TCP callback races fixes (C) 1998 Red Hat Software <alan@redhat.com>
7 * TCP send fixes (C) 1998 Red Hat Software <alan@redhat.com>
8 * TCP NFS related read + write fixes
9 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
11 * Rewrite of larges part of the code in order to stabilize TCP stuff.
12 * Fix behaviour when socket buffer is full.
13 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
15 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
18 #include <linux/types.h>
19 #include <linux/slab.h>
20 #include <linux/capability.h>
21 #include <linux/pagemap.h>
22 #include <linux/errno.h>
23 #include <linux/socket.h>
25 #include <linux/net.h>
27 #include <linux/udp.h>
28 #include <linux/tcp.h>
29 #include <linux/sunrpc/clnt.h>
30 #include <linux/sunrpc/sched.h>
31 #include <linux/file.h>
34 #include <net/checksum.h>
41 unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
42 unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE;
44 unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
45 unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
48 * We can register our own files under /proc/sys/sunrpc by
49 * calling register_sysctl_table() again. The files in that
50 * directory become the union of all files registered there.
52 * We simply need to make sure that we don't collide with
53 * someone else's file names!
58 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
59 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
60 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
61 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
63 static struct ctl_table_header *sunrpc_table_header;
66 * FIXME: changing the UDP slot table size should also resize the UDP
67 * socket buffers for existing UDP transports
69 static ctl_table xs_tunables_table[] = {
71 .ctl_name = CTL_SLOTTABLE_UDP,
72 .procname = "udp_slot_table_entries",
73 .data = &xprt_udp_slot_table_entries,
74 .maxlen = sizeof(unsigned int),
76 .proc_handler = &proc_dointvec_minmax,
77 .strategy = &sysctl_intvec,
78 .extra1 = &min_slot_table_size,
79 .extra2 = &max_slot_table_size
82 .ctl_name = CTL_SLOTTABLE_TCP,
83 .procname = "tcp_slot_table_entries",
84 .data = &xprt_tcp_slot_table_entries,
85 .maxlen = sizeof(unsigned int),
87 .proc_handler = &proc_dointvec_minmax,
88 .strategy = &sysctl_intvec,
89 .extra1 = &min_slot_table_size,
90 .extra2 = &max_slot_table_size
93 .ctl_name = CTL_MIN_RESVPORT,
94 .procname = "min_resvport",
95 .data = &xprt_min_resvport,
96 .maxlen = sizeof(unsigned int),
98 .proc_handler = &proc_dointvec_minmax,
99 .strategy = &sysctl_intvec,
100 .extra1 = &xprt_min_resvport_limit,
101 .extra2 = &xprt_max_resvport_limit
104 .ctl_name = CTL_MAX_RESVPORT,
105 .procname = "max_resvport",
106 .data = &xprt_max_resvport,
107 .maxlen = sizeof(unsigned int),
109 .proc_handler = &proc_dointvec_minmax,
110 .strategy = &sysctl_intvec,
111 .extra1 = &xprt_min_resvport_limit,
112 .extra2 = &xprt_max_resvport_limit
119 static ctl_table sunrpc_table[] = {
121 .ctl_name = CTL_SUNRPC,
122 .procname = "sunrpc",
124 .child = xs_tunables_table
134 * How many times to try sending a request on a socket before waiting
135 * for the socket buffer to clear.
137 #define XS_SENDMSG_RETRY (10U)
140 * Time out for an RPC UDP socket connect. UDP socket connects are
141 * synchronous, but we set a timeout anyway in case of resource
142 * exhaustion on the local host.
144 #define XS_UDP_CONN_TO (5U * HZ)
147 * Wait duration for an RPC TCP connection to be established. Solaris
148 * NFS over TCP uses 60 seconds, for example, which is in line with how
149 * long a server takes to reboot.
151 #define XS_TCP_CONN_TO (60U * HZ)
154 * Wait duration for a reply from the RPC portmapper.
156 #define XS_BIND_TO (60U * HZ)
159 * Delay if a UDP socket connect error occurs. This is most likely some
160 * kind of resource problem on the local host.
162 #define XS_UDP_REEST_TO (2U * HZ)
165 * The reestablish timeout allows clients to delay for a bit before attempting
166 * to reconnect to a server that just dropped our connection.
168 * We implement an exponential backoff when trying to reestablish a TCP
169 * transport connection with the server. Some servers like to drop a TCP
170 * connection when they are overworked, so we start with a short timeout and
171 * increase over time if the server is down or not responding.
173 #define XS_TCP_INIT_REEST_TO (3U * HZ)
174 #define XS_TCP_MAX_REEST_TO (5U * 60 * HZ)
177 * TCP idle timeout; client drops the transport socket if it is idle
178 * for this long. Note that we also timeout UDP sockets to prevent
179 * holding port numbers when there is no RPC traffic.
181 #define XS_IDLE_DISC_TO (5U * 60 * HZ)
184 # undef RPC_DEBUG_DATA
185 # define RPCDBG_FACILITY RPCDBG_TRANS
188 #ifdef RPC_DEBUG_DATA
189 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
191 u8 *buf = (u8 *) packet;
194 dprintk("RPC: %s\n", msg);
195 for (j = 0; j < count && j < 128; j += 4) {
199 dprintk("0x%04x ", j);
201 dprintk("%02x%02x%02x%02x ",
202 buf[j], buf[j+1], buf[j+2], buf[j+3]);
207 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
214 struct rpc_xprt xprt;
219 struct socket * sock;
223 * State of TCP reply receive
231 unsigned long tcp_copied,
235 * Connection of transports
237 struct delayed_work connect_worker;
238 struct sockaddr_storage addr;
242 * UDP socket buffer size parameters
248 * Saved socket callback addresses
250 void (*old_data_ready)(struct sock *, int);
251 void (*old_state_change)(struct sock *);
252 void (*old_write_space)(struct sock *);
256 * TCP receive state flags
258 #define TCP_RCV_LAST_FRAG (1UL << 0)
259 #define TCP_RCV_COPY_FRAGHDR (1UL << 1)
260 #define TCP_RCV_COPY_XID (1UL << 2)
261 #define TCP_RCV_COPY_DATA (1UL << 3)
263 static void xs_format_ipv4_peer_addresses(struct rpc_xprt *xprt)
265 struct sockaddr_in *addr = (struct sockaddr_in *) &xprt->addr;
268 buf = kzalloc(20, GFP_KERNEL);
270 snprintf(buf, 20, NIPQUAD_FMT,
271 NIPQUAD(addr->sin_addr.s_addr));
273 xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
275 buf = kzalloc(8, GFP_KERNEL);
277 snprintf(buf, 8, "%u",
278 ntohs(addr->sin_port));
280 xprt->address_strings[RPC_DISPLAY_PORT] = buf;
282 buf = kzalloc(8, GFP_KERNEL);
284 if (xprt->prot == IPPROTO_UDP)
285 snprintf(buf, 8, "udp");
287 snprintf(buf, 8, "tcp");
289 xprt->address_strings[RPC_DISPLAY_PROTO] = buf;
291 buf = kzalloc(48, GFP_KERNEL);
293 snprintf(buf, 48, "addr="NIPQUAD_FMT" port=%u proto=%s",
294 NIPQUAD(addr->sin_addr.s_addr),
295 ntohs(addr->sin_port),
296 xprt->prot == IPPROTO_UDP ? "udp" : "tcp");
298 xprt->address_strings[RPC_DISPLAY_ALL] = buf;
300 buf = kzalloc(10, GFP_KERNEL);
302 snprintf(buf, 10, "%02x%02x%02x%02x",
303 NIPQUAD(addr->sin_addr.s_addr));
305 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
307 buf = kzalloc(8, GFP_KERNEL);
309 snprintf(buf, 8, "%4hx",
310 ntohs(addr->sin_port));
312 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
315 static void xs_format_ipv6_peer_addresses(struct rpc_xprt *xprt)
317 struct sockaddr_in6 *addr = (struct sockaddr_in6 *) &xprt->addr;
320 buf = kzalloc(40, GFP_KERNEL);
322 snprintf(buf, 40, NIP6_FMT,
323 NIP6(addr->sin6_addr));
325 xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
327 buf = kzalloc(8, GFP_KERNEL);
329 snprintf(buf, 8, "%u",
330 ntohs(addr->sin6_port));
332 xprt->address_strings[RPC_DISPLAY_PORT] = buf;
334 buf = kzalloc(8, GFP_KERNEL);
336 if (xprt->prot == IPPROTO_UDP)
337 snprintf(buf, 8, "udp");
339 snprintf(buf, 8, "tcp");
341 xprt->address_strings[RPC_DISPLAY_PROTO] = buf;
343 buf = kzalloc(64, GFP_KERNEL);
345 snprintf(buf, 64, "addr="NIP6_FMT" port=%u proto=%s",
346 NIP6(addr->sin6_addr),
347 ntohs(addr->sin6_port),
348 xprt->prot == IPPROTO_UDP ? "udp" : "tcp");
350 xprt->address_strings[RPC_DISPLAY_ALL] = buf;
352 buf = kzalloc(36, GFP_KERNEL);
354 snprintf(buf, 36, NIP6_SEQFMT,
355 NIP6(addr->sin6_addr));
357 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
359 buf = kzalloc(8, GFP_KERNEL);
361 snprintf(buf, 8, "%4hx",
362 ntohs(addr->sin6_port));
364 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
367 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
371 for (i = 0; i < RPC_DISPLAY_MAX; i++)
372 kfree(xprt->address_strings[i]);
375 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
377 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
379 struct msghdr msg = {
381 .msg_namelen = addrlen,
382 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
385 .iov_base = vec->iov_base + base,
386 .iov_len = vec->iov_len - base,
389 if (iov.iov_len != 0)
390 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
391 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
394 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
397 unsigned int remainder;
400 remainder = xdr->page_len - base;
401 base += xdr->page_base;
402 ppage = xdr->pages + (base >> PAGE_SHIFT);
405 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
406 int flags = XS_SENDMSG_FLAGS;
409 if (remainder != 0 || more)
411 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
412 if (remainder == 0 || err != len)
426 * xs_sendpages - write pages directly to a socket
427 * @sock: socket to send on
428 * @addr: UDP only -- address of destination
429 * @addrlen: UDP only -- length of destination address
430 * @xdr: buffer containing this request
431 * @base: starting position in the buffer
434 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
436 unsigned int remainder = xdr->len - base;
442 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
448 if (base < xdr->head[0].iov_len || addr != NULL) {
449 unsigned int len = xdr->head[0].iov_len - base;
451 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
452 if (remainder == 0 || err != len)
457 base -= xdr->head[0].iov_len;
459 if (base < xdr->page_len) {
460 unsigned int len = xdr->page_len - base;
462 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
463 if (remainder == 0 || err != len)
468 base -= xdr->page_len;
470 if (base >= xdr->tail[0].iov_len)
472 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
482 * xs_nospace - place task on wait queue if transmit was incomplete
483 * @task: task to put to sleep
486 static void xs_nospace(struct rpc_task *task)
488 struct rpc_rqst *req = task->tk_rqstp;
489 struct rpc_xprt *xprt = req->rq_xprt;
490 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
492 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
493 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
496 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
497 /* Protect against races with write_space */
498 spin_lock_bh(&xprt->transport_lock);
500 /* Don't race with disconnect */
501 if (!xprt_connected(xprt))
502 task->tk_status = -ENOTCONN;
503 else if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
504 xprt_wait_for_buffer_space(task);
506 spin_unlock_bh(&xprt->transport_lock);
508 /* Keep holding the socket if it is blocked */
509 rpc_delay(task, HZ>>4);
513 * xs_udp_send_request - write an RPC request to a UDP socket
514 * @task: address of RPC task that manages the state of an RPC request
517 * 0: The request has been sent
518 * EAGAIN: The socket was blocked, please call again later to
519 * complete the request
520 * ENOTCONN: Caller needs to invoke connect logic then call again
521 * other: Some other error occured, the request was not sent
523 static int xs_udp_send_request(struct rpc_task *task)
525 struct rpc_rqst *req = task->tk_rqstp;
526 struct rpc_xprt *xprt = req->rq_xprt;
527 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
528 struct xdr_buf *xdr = &req->rq_snd_buf;
531 xs_pktdump("packet data:",
532 req->rq_svec->iov_base,
533 req->rq_svec->iov_len);
535 req->rq_xtime = jiffies;
536 status = xs_sendpages(transport->sock,
537 (struct sockaddr *) &xprt->addr,
541 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
542 xdr->len - req->rq_bytes_sent, status);
544 if (likely(status >= (int) req->rq_slen))
547 /* Still some bytes left; set up for a retry later. */
555 /* When the server has died, an ICMP port unreachable message
556 * prompts ECONNREFUSED. */
562 dprintk("RPC: sendmsg returned unrecognized error %d\n",
570 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
572 u32 reclen = buf->len - sizeof(rpc_fraghdr);
573 rpc_fraghdr *base = buf->head[0].iov_base;
574 *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
578 * xs_tcp_send_request - write an RPC request to a TCP socket
579 * @task: address of RPC task that manages the state of an RPC request
582 * 0: The request has been sent
583 * EAGAIN: The socket was blocked, please call again later to
584 * complete the request
585 * ENOTCONN: Caller needs to invoke connect logic then call again
586 * other: Some other error occured, the request was not sent
588 * XXX: In the case of soft timeouts, should we eventually give up
589 * if sendmsg is not able to make progress?
591 static int xs_tcp_send_request(struct rpc_task *task)
593 struct rpc_rqst *req = task->tk_rqstp;
594 struct rpc_xprt *xprt = req->rq_xprt;
595 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
596 struct xdr_buf *xdr = &req->rq_snd_buf;
598 unsigned int retry = 0;
600 xs_encode_tcp_record_marker(&req->rq_snd_buf);
602 xs_pktdump("packet data:",
603 req->rq_svec->iov_base,
604 req->rq_svec->iov_len);
606 /* Continue transmitting the packet/record. We must be careful
607 * to cope with writespace callbacks arriving _after_ we have
608 * called sendmsg(). */
610 req->rq_xtime = jiffies;
611 status = xs_sendpages(transport->sock,
612 NULL, 0, xdr, req->rq_bytes_sent);
614 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
615 xdr->len - req->rq_bytes_sent, status);
617 if (unlikely(status < 0))
620 /* If we've sent the entire packet, immediately
621 * reset the count of bytes sent. */
622 req->rq_bytes_sent += status;
623 task->tk_bytes_sent += status;
624 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
625 req->rq_bytes_sent = 0;
630 if (retry++ > XS_SENDMSG_RETRY)
645 dprintk("RPC: sendmsg returned unrecognized error %d\n",
647 xprt_disconnect(xprt);
655 * xs_tcp_release_xprt - clean up after a tcp transmission
659 * This cleans up if an error causes us to abort the transmission of a request.
660 * In this case, the socket may need to be reset in order to avoid confusing
663 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
665 struct rpc_rqst *req;
667 if (task != xprt->snd_task)
671 req = task->tk_rqstp;
672 if (req->rq_bytes_sent == 0)
674 if (req->rq_bytes_sent == req->rq_snd_buf.len)
676 set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
678 xprt_release_xprt(xprt, task);
682 * xs_close - close a socket
685 * This is used when all requests are complete; ie, no DRC state remains
686 * on the server we want to save.
688 static void xs_close(struct rpc_xprt *xprt)
690 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
691 struct socket *sock = transport->sock;
692 struct sock *sk = transport->inet;
695 goto clear_close_wait;
697 dprintk("RPC: xs_close xprt %p\n", xprt);
699 write_lock_bh(&sk->sk_callback_lock);
700 transport->inet = NULL;
701 transport->sock = NULL;
703 sk->sk_user_data = NULL;
704 sk->sk_data_ready = transport->old_data_ready;
705 sk->sk_state_change = transport->old_state_change;
706 sk->sk_write_space = transport->old_write_space;
707 write_unlock_bh(&sk->sk_callback_lock);
713 smp_mb__before_clear_bit();
714 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
715 smp_mb__after_clear_bit();
719 * xs_destroy - prepare to shutdown a transport
720 * @xprt: doomed transport
723 static void xs_destroy(struct rpc_xprt *xprt)
725 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
727 dprintk("RPC: xs_destroy xprt %p\n", xprt);
729 cancel_rearming_delayed_work(&transport->connect_worker);
731 xprt_disconnect(xprt);
733 xs_free_peer_addresses(xprt);
738 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
740 return (struct rpc_xprt *) sk->sk_user_data;
744 * xs_udp_data_ready - "data ready" callback for UDP sockets
745 * @sk: socket with data to read
746 * @len: how much data to read
749 static void xs_udp_data_ready(struct sock *sk, int len)
751 struct rpc_task *task;
752 struct rpc_xprt *xprt;
753 struct rpc_rqst *rovr;
755 int err, repsize, copied;
759 read_lock(&sk->sk_callback_lock);
760 dprintk("RPC: xs_udp_data_ready...\n");
761 if (!(xprt = xprt_from_sock(sk)))
764 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
770 repsize = skb->len - sizeof(struct udphdr);
772 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
776 /* Copy the XID from the skb... */
777 xp = skb_header_pointer(skb, sizeof(struct udphdr),
778 sizeof(_xid), &_xid);
782 /* Look up and lock the request corresponding to the given XID */
783 spin_lock(&xprt->transport_lock);
784 rovr = xprt_lookup_rqst(xprt, *xp);
787 task = rovr->rq_task;
789 if ((copied = rovr->rq_private_buf.buflen) > repsize)
792 /* Suck it into the iovec, verify checksum if not done by hw. */
793 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb))
796 /* Something worked... */
797 dst_confirm(skb->dst);
799 xprt_adjust_cwnd(task, copied);
800 xprt_update_rtt(task);
801 xprt_complete_rqst(task, copied);
804 spin_unlock(&xprt->transport_lock);
806 skb_free_datagram(sk, skb);
808 read_unlock(&sk->sk_callback_lock);
811 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
813 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
817 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
818 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
819 used = xdr_skb_read_bits(desc, p, len);
820 transport->tcp_offset += used;
824 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
825 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
826 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
828 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
829 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
831 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
832 transport->tcp_offset = 0;
834 /* Sanity check of the record length */
835 if (unlikely(transport->tcp_reclen < 4)) {
836 dprintk("RPC: invalid TCP record fragment length\n");
837 xprt_disconnect(xprt);
840 dprintk("RPC: reading TCP record fragment of length %d\n",
841 transport->tcp_reclen);
844 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
846 if (transport->tcp_offset == transport->tcp_reclen) {
847 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
848 transport->tcp_offset = 0;
849 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
850 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
851 transport->tcp_flags |= TCP_RCV_COPY_XID;
852 transport->tcp_copied = 0;
857 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
862 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
863 dprintk("RPC: reading XID (%Zu bytes)\n", len);
864 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
865 used = xdr_skb_read_bits(desc, p, len);
866 transport->tcp_offset += used;
869 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
870 transport->tcp_flags |= TCP_RCV_COPY_DATA;
871 transport->tcp_copied = 4;
872 dprintk("RPC: reading reply for XID %08x\n",
873 ntohl(transport->tcp_xid));
874 xs_tcp_check_fraghdr(transport);
877 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
879 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
880 struct rpc_rqst *req;
881 struct xdr_buf *rcvbuf;
885 /* Find and lock the request corresponding to this xid */
886 spin_lock(&xprt->transport_lock);
887 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
889 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
890 dprintk("RPC: XID %08x request not found!\n",
891 ntohl(transport->tcp_xid));
892 spin_unlock(&xprt->transport_lock);
896 rcvbuf = &req->rq_private_buf;
898 if (len > transport->tcp_reclen - transport->tcp_offset) {
899 struct xdr_skb_reader my_desc;
901 len = transport->tcp_reclen - transport->tcp_offset;
902 memcpy(&my_desc, desc, sizeof(my_desc));
904 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
905 &my_desc, xdr_skb_read_bits);
909 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
910 desc, xdr_skb_read_bits);
913 transport->tcp_copied += r;
914 transport->tcp_offset += r;
917 /* Error when copying to the receive buffer,
918 * usually because we weren't able to allocate
919 * additional buffer pages. All we can do now
920 * is turn off TCP_RCV_COPY_DATA, so the request
921 * will not receive any additional updates,
923 * Any remaining data from this record will
926 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
927 dprintk("RPC: XID %08x truncated request\n",
928 ntohl(transport->tcp_xid));
929 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
930 "tcp_offset = %u, tcp_reclen = %u\n",
931 xprt, transport->tcp_copied,
932 transport->tcp_offset, transport->tcp_reclen);
936 dprintk("RPC: XID %08x read %Zd bytes\n",
937 ntohl(transport->tcp_xid), r);
938 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
939 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
940 transport->tcp_offset, transport->tcp_reclen);
942 if (transport->tcp_copied == req->rq_private_buf.buflen)
943 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
944 else if (transport->tcp_offset == transport->tcp_reclen) {
945 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
946 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
950 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
951 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
952 spin_unlock(&xprt->transport_lock);
953 xs_tcp_check_fraghdr(transport);
956 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
960 len = transport->tcp_reclen - transport->tcp_offset;
961 if (len > desc->count)
965 transport->tcp_offset += len;
966 dprintk("RPC: discarded %Zu bytes\n", len);
967 xs_tcp_check_fraghdr(transport);
970 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
972 struct rpc_xprt *xprt = rd_desc->arg.data;
973 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
974 struct xdr_skb_reader desc = {
980 dprintk("RPC: xs_tcp_data_recv started\n");
982 /* Read in a new fragment marker if necessary */
983 /* Can we ever really expect to get completely empty fragments? */
984 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
985 xs_tcp_read_fraghdr(xprt, &desc);
988 /* Read in the xid if necessary */
989 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
990 xs_tcp_read_xid(transport, &desc);
993 /* Read in the request data */
994 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
995 xs_tcp_read_request(xprt, &desc);
998 /* Skip over any trailing bytes on short reads */
999 xs_tcp_read_discard(transport, &desc);
1000 } while (desc.count);
1001 dprintk("RPC: xs_tcp_data_recv done\n");
1002 return len - desc.count;
1006 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1007 * @sk: socket with data to read
1008 * @bytes: how much data to read
1011 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1013 struct rpc_xprt *xprt;
1014 read_descriptor_t rd_desc;
1016 dprintk("RPC: xs_tcp_data_ready...\n");
1018 read_lock(&sk->sk_callback_lock);
1019 if (!(xprt = xprt_from_sock(sk)))
1024 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1025 rd_desc.arg.data = xprt;
1026 rd_desc.count = 65536;
1027 tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1029 read_unlock(&sk->sk_callback_lock);
1033 * xs_tcp_state_change - callback to handle TCP socket state changes
1034 * @sk: socket whose state has changed
1037 static void xs_tcp_state_change(struct sock *sk)
1039 struct rpc_xprt *xprt;
1041 read_lock(&sk->sk_callback_lock);
1042 if (!(xprt = xprt_from_sock(sk)))
1044 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1045 dprintk("RPC: state %x conn %d dead %d zapped %d\n",
1046 sk->sk_state, xprt_connected(xprt),
1047 sock_flag(sk, SOCK_DEAD),
1048 sock_flag(sk, SOCK_ZAPPED));
1050 switch (sk->sk_state) {
1051 case TCP_ESTABLISHED:
1052 spin_lock_bh(&xprt->transport_lock);
1053 if (!xprt_test_and_set_connected(xprt)) {
1054 struct sock_xprt *transport = container_of(xprt,
1055 struct sock_xprt, xprt);
1057 /* Reset TCP record info */
1058 transport->tcp_offset = 0;
1059 transport->tcp_reclen = 0;
1060 transport->tcp_copied = 0;
1061 transport->tcp_flags =
1062 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1064 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1065 xprt_wake_pending_tasks(xprt, 0);
1067 spin_unlock_bh(&xprt->transport_lock);
1072 case TCP_CLOSE_WAIT:
1073 /* Try to schedule an autoclose RPC calls */
1074 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
1075 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
1076 queue_work(rpciod_workqueue, &xprt->task_cleanup);
1078 xprt_disconnect(xprt);
1081 read_unlock(&sk->sk_callback_lock);
1085 * xs_udp_write_space - callback invoked when socket buffer space
1087 * @sk: socket whose state has changed
1089 * Called when more output buffer space is available for this socket.
1090 * We try not to wake our writers until they can make "significant"
1091 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1092 * with a bunch of small requests.
1094 static void xs_udp_write_space(struct sock *sk)
1096 read_lock(&sk->sk_callback_lock);
1098 /* from net/core/sock.c:sock_def_write_space */
1099 if (sock_writeable(sk)) {
1100 struct socket *sock;
1101 struct rpc_xprt *xprt;
1103 if (unlikely(!(sock = sk->sk_socket)))
1105 if (unlikely(!(xprt = xprt_from_sock(sk))))
1107 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
1110 xprt_write_space(xprt);
1114 read_unlock(&sk->sk_callback_lock);
1118 * xs_tcp_write_space - callback invoked when socket buffer space
1120 * @sk: socket whose state has changed
1122 * Called when more output buffer space is available for this socket.
1123 * We try not to wake our writers until they can make "significant"
1124 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1125 * with a bunch of small requests.
1127 static void xs_tcp_write_space(struct sock *sk)
1129 read_lock(&sk->sk_callback_lock);
1131 /* from net/core/stream.c:sk_stream_write_space */
1132 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
1133 struct socket *sock;
1134 struct rpc_xprt *xprt;
1136 if (unlikely(!(sock = sk->sk_socket)))
1138 if (unlikely(!(xprt = xprt_from_sock(sk))))
1140 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
1143 xprt_write_space(xprt);
1147 read_unlock(&sk->sk_callback_lock);
1150 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1152 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1153 struct sock *sk = transport->inet;
1155 if (transport->rcvsize) {
1156 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1157 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1159 if (transport->sndsize) {
1160 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1161 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1162 sk->sk_write_space(sk);
1167 * xs_udp_set_buffer_size - set send and receive limits
1168 * @xprt: generic transport
1169 * @sndsize: requested size of send buffer, in bytes
1170 * @rcvsize: requested size of receive buffer, in bytes
1172 * Set socket send and receive buffer size limits.
1174 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1176 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1178 transport->sndsize = 0;
1180 transport->sndsize = sndsize + 1024;
1181 transport->rcvsize = 0;
1183 transport->rcvsize = rcvsize + 1024;
1185 xs_udp_do_set_buffer_size(xprt);
1189 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1190 * @task: task that timed out
1192 * Adjust the congestion window after a retransmit timeout has occurred.
1194 static void xs_udp_timer(struct rpc_task *task)
1196 xprt_adjust_cwnd(task, -ETIMEDOUT);
1199 static unsigned short xs_get_random_port(void)
1201 unsigned short range = xprt_max_resvport - xprt_min_resvport;
1202 unsigned short rand = (unsigned short) net_random() % range;
1203 return rand + xprt_min_resvport;
1207 * xs_set_port - reset the port number in the remote endpoint address
1208 * @xprt: generic transport
1209 * @port: new port number
1212 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1214 struct sockaddr *addr = (struct sockaddr *) &xprt->addr;
1216 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1218 switch (addr->sa_family) {
1220 ((struct sockaddr_in *)addr)->sin_port = htons(port);
1223 ((struct sockaddr_in6 *)addr)->sin6_port = htons(port);
1230 static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
1232 struct sockaddr_in myaddr = {
1233 .sin_family = AF_INET,
1235 struct sockaddr_in *sa;
1237 unsigned short port = transport->port;
1239 if (!transport->xprt.resvport)
1241 sa = (struct sockaddr_in *)&transport->addr;
1242 myaddr.sin_addr = sa->sin_addr;
1244 myaddr.sin_port = htons(port);
1245 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1247 if (!transport->xprt.resvport)
1250 transport->port = port;
1253 if (port <= xprt_min_resvport)
1254 port = xprt_max_resvport;
1257 } while (err == -EADDRINUSE && port != transport->port);
1258 dprintk("RPC: %s "NIPQUAD_FMT":%u: %s (%d)\n",
1259 __FUNCTION__, NIPQUAD(myaddr.sin_addr),
1260 port, err ? "failed" : "ok", err);
1264 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1265 static struct lock_class_key xs_key[2];
1266 static struct lock_class_key xs_slock_key[2];
1268 static inline void xs_reclassify_socket(struct socket *sock)
1270 struct sock *sk = sock->sk;
1271 BUG_ON(sk->sk_lock.owner != NULL);
1272 switch (sk->sk_family) {
1274 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFS",
1275 &xs_slock_key[0], "sk_lock-AF_INET-NFS", &xs_key[0]);
1279 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFS",
1280 &xs_slock_key[1], "sk_lock-AF_INET6-NFS", &xs_key[1]);
1288 static inline void xs_reclassify_socket(struct socket *sock)
1294 * xs_udp_connect_worker - set up a UDP socket
1295 * @work: RPC transport to connect
1297 * Invoked by a work queue tasklet.
1299 static void xs_udp_connect_worker(struct work_struct *work)
1301 struct sock_xprt *transport =
1302 container_of(work, struct sock_xprt, connect_worker.work);
1303 struct rpc_xprt *xprt = &transport->xprt;
1304 struct socket *sock = transport->sock;
1305 int err, status = -EIO;
1307 if (xprt->shutdown || !xprt_bound(xprt))
1310 /* Start by resetting any existing state */
1313 if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1314 dprintk("RPC: can't create UDP transport socket (%d).\n", -err);
1317 xs_reclassify_socket(sock);
1319 if (xs_bind4(transport, sock)) {
1324 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1325 xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1327 if (!transport->inet) {
1328 struct sock *sk = sock->sk;
1330 write_lock_bh(&sk->sk_callback_lock);
1332 sk->sk_user_data = xprt;
1333 transport->old_data_ready = sk->sk_data_ready;
1334 transport->old_state_change = sk->sk_state_change;
1335 transport->old_write_space = sk->sk_write_space;
1336 sk->sk_data_ready = xs_udp_data_ready;
1337 sk->sk_write_space = xs_udp_write_space;
1338 sk->sk_no_check = UDP_CSUM_NORCV;
1339 sk->sk_allocation = GFP_ATOMIC;
1341 xprt_set_connected(xprt);
1343 /* Reset to new socket */
1344 transport->sock = sock;
1345 transport->inet = sk;
1347 write_unlock_bh(&sk->sk_callback_lock);
1349 xs_udp_do_set_buffer_size(xprt);
1352 xprt_wake_pending_tasks(xprt, status);
1353 xprt_clear_connecting(xprt);
1357 * We need to preserve the port number so the reply cache on the server can
1358 * find our cached RPC replies when we get around to reconnecting.
1360 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1363 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1364 struct sockaddr any;
1366 dprintk("RPC: disconnecting xprt %p to reuse port\n", xprt);
1369 * Disconnect the transport socket by doing a connect operation
1370 * with AF_UNSPEC. This should return immediately...
1372 memset(&any, 0, sizeof(any));
1373 any.sa_family = AF_UNSPEC;
1374 result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1376 dprintk("RPC: AF_UNSPEC connect return code %d\n",
1381 * xs_tcp_connect_worker - connect a TCP socket to a remote endpoint
1382 * @work: RPC transport to connect
1384 * Invoked by a work queue tasklet.
1386 static void xs_tcp_connect_worker(struct work_struct *work)
1388 struct sock_xprt *transport =
1389 container_of(work, struct sock_xprt, connect_worker.work);
1390 struct rpc_xprt *xprt = &transport->xprt;
1391 struct socket *sock = transport->sock;
1392 int err, status = -EIO;
1394 if (xprt->shutdown || !xprt_bound(xprt))
1398 /* start from scratch */
1399 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1400 dprintk("RPC: can't create TCP transport "
1401 "socket (%d).\n", -err);
1404 xs_reclassify_socket(sock);
1406 if (xs_bind4(transport, sock)) {
1411 /* "close" the socket, preserving the local port */
1412 xs_tcp_reuse_connection(xprt);
1414 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1415 xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1417 if (!transport->inet) {
1418 struct sock *sk = sock->sk;
1420 write_lock_bh(&sk->sk_callback_lock);
1422 sk->sk_user_data = xprt;
1423 transport->old_data_ready = sk->sk_data_ready;
1424 transport->old_state_change = sk->sk_state_change;
1425 transport->old_write_space = sk->sk_write_space;
1426 sk->sk_data_ready = xs_tcp_data_ready;
1427 sk->sk_state_change = xs_tcp_state_change;
1428 sk->sk_write_space = xs_tcp_write_space;
1429 sk->sk_allocation = GFP_ATOMIC;
1431 /* socket options */
1432 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1433 sock_reset_flag(sk, SOCK_LINGER);
1434 tcp_sk(sk)->linger2 = 0;
1435 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1437 xprt_clear_connected(xprt);
1439 /* Reset to new socket */
1440 transport->sock = sock;
1441 transport->inet = sk;
1443 write_unlock_bh(&sk->sk_callback_lock);
1446 /* Tell the socket layer to start connecting... */
1447 xprt->stat.connect_count++;
1448 xprt->stat.connect_start = jiffies;
1449 status = kernel_connect(sock, (struct sockaddr *) &xprt->addr,
1450 xprt->addrlen, O_NONBLOCK);
1451 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
1452 xprt, -status, xprt_connected(xprt),
1453 sock->sk->sk_state);
1461 /* retry with existing socket, after a delay */
1464 /* get rid of existing socket, and retry */
1470 xprt_wake_pending_tasks(xprt, status);
1472 xprt_clear_connecting(xprt);
1476 * xs_connect - connect a socket to a remote endpoint
1477 * @task: address of RPC task that manages state of connect request
1479 * TCP: If the remote end dropped the connection, delay reconnecting.
1481 * UDP socket connects are synchronous, but we use a work queue anyway
1482 * to guarantee that even unprivileged user processes can set up a
1483 * socket on a privileged port.
1485 * If a UDP socket connect fails, the delay behavior here prevents
1486 * retry floods (hard mounts).
1488 static void xs_connect(struct rpc_task *task)
1490 struct rpc_xprt *xprt = task->tk_xprt;
1491 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1493 if (xprt_test_and_set_connecting(xprt))
1496 if (transport->sock != NULL) {
1497 dprintk("RPC: xs_connect delayed xprt %p for %lu "
1499 xprt, xprt->reestablish_timeout / HZ);
1500 queue_delayed_work(rpciod_workqueue,
1501 &transport->connect_worker,
1502 xprt->reestablish_timeout);
1503 xprt->reestablish_timeout <<= 1;
1504 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1505 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1507 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
1508 queue_delayed_work(rpciod_workqueue,
1509 &transport->connect_worker, 0);
1514 * xs_udp_print_stats - display UDP socket-specifc stats
1515 * @xprt: rpc_xprt struct containing statistics
1519 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1521 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1523 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1525 xprt->stat.bind_count,
1528 xprt->stat.bad_xids,
1530 xprt->stat.bklog_u);
1534 * xs_tcp_print_stats - display TCP socket-specifc stats
1535 * @xprt: rpc_xprt struct containing statistics
1539 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1541 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1544 if (xprt_connected(xprt))
1545 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1547 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1549 xprt->stat.bind_count,
1550 xprt->stat.connect_count,
1551 xprt->stat.connect_time,
1555 xprt->stat.bad_xids,
1557 xprt->stat.bklog_u);
1560 static struct rpc_xprt_ops xs_udp_ops = {
1561 .set_buffer_size = xs_udp_set_buffer_size,
1562 .reserve_xprt = xprt_reserve_xprt_cong,
1563 .release_xprt = xprt_release_xprt_cong,
1564 .rpcbind = rpcb_getport_async,
1565 .set_port = xs_set_port,
1566 .connect = xs_connect,
1567 .buf_alloc = rpc_malloc,
1568 .buf_free = rpc_free,
1569 .send_request = xs_udp_send_request,
1570 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
1571 .timer = xs_udp_timer,
1572 .release_request = xprt_release_rqst_cong,
1574 .destroy = xs_destroy,
1575 .print_stats = xs_udp_print_stats,
1578 static struct rpc_xprt_ops xs_tcp_ops = {
1579 .reserve_xprt = xprt_reserve_xprt,
1580 .release_xprt = xs_tcp_release_xprt,
1581 .rpcbind = rpcb_getport_async,
1582 .set_port = xs_set_port,
1583 .connect = xs_connect,
1584 .buf_alloc = rpc_malloc,
1585 .buf_free = rpc_free,
1586 .send_request = xs_tcp_send_request,
1587 .set_retrans_timeout = xprt_set_retrans_timeout_def,
1589 .destroy = xs_destroy,
1590 .print_stats = xs_tcp_print_stats,
1593 static struct rpc_xprt *xs_setup_xprt(struct rpc_xprtsock_create *args, unsigned int slot_table_size)
1595 struct rpc_xprt *xprt;
1596 struct sock_xprt *new;
1598 if (args->addrlen > sizeof(xprt->addr)) {
1599 dprintk("RPC: xs_setup_xprt: address too large\n");
1600 return ERR_PTR(-EBADF);
1603 new = kzalloc(sizeof(*new), GFP_KERNEL);
1605 dprintk("RPC: xs_setup_xprt: couldn't allocate "
1607 return ERR_PTR(-ENOMEM);
1611 xprt->max_reqs = slot_table_size;
1612 xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1613 if (xprt->slot == NULL) {
1615 dprintk("RPC: xs_setup_xprt: couldn't allocate slot "
1617 return ERR_PTR(-ENOMEM);
1620 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
1621 xprt->addrlen = args->addrlen;
1623 memcpy(&new->addr, args->srcaddr, args->addrlen);
1624 new->port = xs_get_random_port();
1630 * xs_setup_udp - Set up transport to use a UDP socket
1631 * @args: rpc transport creation arguments
1634 struct rpc_xprt *xs_setup_udp(struct rpc_xprtsock_create *args)
1636 struct rpc_xprt *xprt;
1637 struct sock_xprt *transport;
1639 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
1642 transport = container_of(xprt, struct sock_xprt, xprt);
1644 if (ntohs(((struct sockaddr_in *)args->dstaddr)->sin_port) != 0)
1645 xprt_set_bound(xprt);
1647 xprt->prot = IPPROTO_UDP;
1649 /* XXX: header size can vary due to auth type, IPv6, etc. */
1650 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1652 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_connect_worker);
1653 xprt->bind_timeout = XS_BIND_TO;
1654 xprt->connect_timeout = XS_UDP_CONN_TO;
1655 xprt->reestablish_timeout = XS_UDP_REEST_TO;
1656 xprt->idle_timeout = XS_IDLE_DISC_TO;
1658 xprt->ops = &xs_udp_ops;
1661 xprt->timeout = *args->timeout;
1663 xprt_set_timeout(&xprt->timeout, 5, 5 * HZ);
1665 xs_format_ipv4_peer_addresses(xprt);
1666 dprintk("RPC: set up transport to address %s\n",
1667 xprt->address_strings[RPC_DISPLAY_ALL]);
1673 * xs_setup_tcp - Set up transport to use a TCP socket
1674 * @args: rpc transport creation arguments
1677 struct rpc_xprt *xs_setup_tcp(struct rpc_xprtsock_create *args)
1679 struct rpc_xprt *xprt;
1680 struct sock_xprt *transport;
1682 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
1685 transport = container_of(xprt, struct sock_xprt, xprt);
1687 if (ntohs(((struct sockaddr_in *)args->dstaddr)->sin_port) != 0)
1688 xprt_set_bound(xprt);
1690 xprt->prot = IPPROTO_TCP;
1691 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
1692 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
1694 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker);
1695 xprt->bind_timeout = XS_BIND_TO;
1696 xprt->connect_timeout = XS_TCP_CONN_TO;
1697 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1698 xprt->idle_timeout = XS_IDLE_DISC_TO;
1700 xprt->ops = &xs_tcp_ops;
1703 xprt->timeout = *args->timeout;
1705 xprt_set_timeout(&xprt->timeout, 2, 60 * HZ);
1707 xs_format_ipv4_peer_addresses(xprt);
1708 dprintk("RPC: set up transport to address %s\n",
1709 xprt->address_strings[RPC_DISPLAY_ALL]);
1715 * init_socket_xprt - set up xprtsock's sysctls
1718 int init_socket_xprt(void)
1721 if (!sunrpc_table_header)
1722 sunrpc_table_header = register_sysctl_table(sunrpc_table);
1729 * cleanup_socket_xprt - remove xprtsock's sysctls
1732 void cleanup_socket_xprt(void)
1735 if (sunrpc_table_header) {
1736 unregister_sysctl_table(sunrpc_table_header);
1737 sunrpc_table_header = NULL;