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>
17 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
18 * <gilles.quillard@bull.net>
21 #include <linux/types.h>
22 #include <linux/slab.h>
23 #include <linux/module.h>
24 #include <linux/capability.h>
25 #include <linux/pagemap.h>
26 #include <linux/errno.h>
27 #include <linux/socket.h>
29 #include <linux/net.h>
31 #include <linux/udp.h>
32 #include <linux/tcp.h>
33 #include <linux/sunrpc/clnt.h>
34 #include <linux/sunrpc/sched.h>
35 #include <linux/sunrpc/xprtsock.h>
36 #include <linux/file.h>
39 #include <net/checksum.h>
46 unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
47 unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE;
49 unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
50 unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
53 * We can register our own files under /proc/sys/sunrpc by
54 * calling register_sysctl_table() again. The files in that
55 * directory become the union of all files registered there.
57 * We simply need to make sure that we don't collide with
58 * someone else's file names!
63 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
64 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
65 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
66 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
68 static struct ctl_table_header *sunrpc_table_header;
71 * FIXME: changing the UDP slot table size should also resize the UDP
72 * socket buffers for existing UDP transports
74 static ctl_table xs_tunables_table[] = {
76 .ctl_name = CTL_SLOTTABLE_UDP,
77 .procname = "udp_slot_table_entries",
78 .data = &xprt_udp_slot_table_entries,
79 .maxlen = sizeof(unsigned int),
81 .proc_handler = &proc_dointvec_minmax,
82 .strategy = &sysctl_intvec,
83 .extra1 = &min_slot_table_size,
84 .extra2 = &max_slot_table_size
87 .ctl_name = CTL_SLOTTABLE_TCP,
88 .procname = "tcp_slot_table_entries",
89 .data = &xprt_tcp_slot_table_entries,
90 .maxlen = sizeof(unsigned int),
92 .proc_handler = &proc_dointvec_minmax,
93 .strategy = &sysctl_intvec,
94 .extra1 = &min_slot_table_size,
95 .extra2 = &max_slot_table_size
98 .ctl_name = CTL_MIN_RESVPORT,
99 .procname = "min_resvport",
100 .data = &xprt_min_resvport,
101 .maxlen = sizeof(unsigned int),
103 .proc_handler = &proc_dointvec_minmax,
104 .strategy = &sysctl_intvec,
105 .extra1 = &xprt_min_resvport_limit,
106 .extra2 = &xprt_max_resvport_limit
109 .ctl_name = CTL_MAX_RESVPORT,
110 .procname = "max_resvport",
111 .data = &xprt_max_resvport,
112 .maxlen = sizeof(unsigned int),
114 .proc_handler = &proc_dointvec_minmax,
115 .strategy = &sysctl_intvec,
116 .extra1 = &xprt_min_resvport_limit,
117 .extra2 = &xprt_max_resvport_limit
124 static ctl_table sunrpc_table[] = {
126 .ctl_name = CTL_SUNRPC,
127 .procname = "sunrpc",
129 .child = xs_tunables_table
139 * How many times to try sending a request on a socket before waiting
140 * for the socket buffer to clear.
142 #define XS_SENDMSG_RETRY (10U)
145 * Time out for an RPC UDP socket connect. UDP socket connects are
146 * synchronous, but we set a timeout anyway in case of resource
147 * exhaustion on the local host.
149 #define XS_UDP_CONN_TO (5U * HZ)
152 * Wait duration for an RPC TCP connection to be established. Solaris
153 * NFS over TCP uses 60 seconds, for example, which is in line with how
154 * long a server takes to reboot.
156 #define XS_TCP_CONN_TO (60U * HZ)
159 * Wait duration for a reply from the RPC portmapper.
161 #define XS_BIND_TO (60U * HZ)
164 * Delay if a UDP socket connect error occurs. This is most likely some
165 * kind of resource problem on the local host.
167 #define XS_UDP_REEST_TO (2U * HZ)
170 * The reestablish timeout allows clients to delay for a bit before attempting
171 * to reconnect to a server that just dropped our connection.
173 * We implement an exponential backoff when trying to reestablish a TCP
174 * transport connection with the server. Some servers like to drop a TCP
175 * connection when they are overworked, so we start with a short timeout and
176 * increase over time if the server is down or not responding.
178 #define XS_TCP_INIT_REEST_TO (3U * HZ)
179 #define XS_TCP_MAX_REEST_TO (5U * 60 * HZ)
182 * TCP idle timeout; client drops the transport socket if it is idle
183 * for this long. Note that we also timeout UDP sockets to prevent
184 * holding port numbers when there is no RPC traffic.
186 #define XS_IDLE_DISC_TO (5U * 60 * HZ)
189 # undef RPC_DEBUG_DATA
190 # define RPCDBG_FACILITY RPCDBG_TRANS
193 #ifdef RPC_DEBUG_DATA
194 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
196 u8 *buf = (u8 *) packet;
199 dprintk("RPC: %s\n", msg);
200 for (j = 0; j < count && j < 128; j += 4) {
204 dprintk("0x%04x ", j);
206 dprintk("%02x%02x%02x%02x ",
207 buf[j], buf[j+1], buf[j+2], buf[j+3]);
212 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
219 struct rpc_xprt xprt;
224 struct socket * sock;
228 * State of TCP reply receive
236 unsigned long tcp_copied,
240 * Connection of transports
242 struct delayed_work connect_worker;
243 struct sockaddr_storage addr;
247 * UDP socket buffer size parameters
253 * Saved socket callback addresses
255 void (*old_data_ready)(struct sock *, int);
256 void (*old_state_change)(struct sock *);
257 void (*old_write_space)(struct sock *);
261 * TCP receive state flags
263 #define TCP_RCV_LAST_FRAG (1UL << 0)
264 #define TCP_RCV_COPY_FRAGHDR (1UL << 1)
265 #define TCP_RCV_COPY_XID (1UL << 2)
266 #define TCP_RCV_COPY_DATA (1UL << 3)
268 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
270 return (struct sockaddr *) &xprt->addr;
273 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
275 return (struct sockaddr_in *) &xprt->addr;
278 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
280 return (struct sockaddr_in6 *) &xprt->addr;
283 static void xs_format_ipv4_peer_addresses(struct rpc_xprt *xprt)
285 struct sockaddr_in *addr = xs_addr_in(xprt);
288 buf = kzalloc(20, GFP_KERNEL);
290 snprintf(buf, 20, NIPQUAD_FMT,
291 NIPQUAD(addr->sin_addr.s_addr));
293 xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
295 buf = kzalloc(8, GFP_KERNEL);
297 snprintf(buf, 8, "%u",
298 ntohs(addr->sin_port));
300 xprt->address_strings[RPC_DISPLAY_PORT] = buf;
302 buf = kzalloc(8, GFP_KERNEL);
304 if (xprt->prot == IPPROTO_UDP)
305 snprintf(buf, 8, "udp");
307 snprintf(buf, 8, "tcp");
309 xprt->address_strings[RPC_DISPLAY_PROTO] = buf;
311 buf = kzalloc(48, GFP_KERNEL);
313 snprintf(buf, 48, "addr="NIPQUAD_FMT" port=%u proto=%s",
314 NIPQUAD(addr->sin_addr.s_addr),
315 ntohs(addr->sin_port),
316 xprt->prot == IPPROTO_UDP ? "udp" : "tcp");
318 xprt->address_strings[RPC_DISPLAY_ALL] = buf;
320 buf = kzalloc(10, GFP_KERNEL);
322 snprintf(buf, 10, "%02x%02x%02x%02x",
323 NIPQUAD(addr->sin_addr.s_addr));
325 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
327 buf = kzalloc(8, GFP_KERNEL);
329 snprintf(buf, 8, "%4hx",
330 ntohs(addr->sin_port));
332 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
334 buf = kzalloc(30, GFP_KERNEL);
336 snprintf(buf, 30, NIPQUAD_FMT".%u.%u",
337 NIPQUAD(addr->sin_addr.s_addr),
338 ntohs(addr->sin_port) >> 8,
339 ntohs(addr->sin_port) & 0xff);
341 xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
343 xprt->address_strings[RPC_DISPLAY_NETID] =
344 kstrdup(xprt->prot == IPPROTO_UDP ?
345 RPCBIND_NETID_UDP : RPCBIND_NETID_TCP, GFP_KERNEL);
348 static void xs_format_ipv6_peer_addresses(struct rpc_xprt *xprt)
350 struct sockaddr_in6 *addr = xs_addr_in6(xprt);
353 buf = kzalloc(40, GFP_KERNEL);
355 snprintf(buf, 40, NIP6_FMT,
356 NIP6(addr->sin6_addr));
358 xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
360 buf = kzalloc(8, GFP_KERNEL);
362 snprintf(buf, 8, "%u",
363 ntohs(addr->sin6_port));
365 xprt->address_strings[RPC_DISPLAY_PORT] = buf;
367 buf = kzalloc(8, GFP_KERNEL);
369 if (xprt->prot == IPPROTO_UDP)
370 snprintf(buf, 8, "udp");
372 snprintf(buf, 8, "tcp");
374 xprt->address_strings[RPC_DISPLAY_PROTO] = buf;
376 buf = kzalloc(64, GFP_KERNEL);
378 snprintf(buf, 64, "addr="NIP6_FMT" port=%u proto=%s",
379 NIP6(addr->sin6_addr),
380 ntohs(addr->sin6_port),
381 xprt->prot == IPPROTO_UDP ? "udp" : "tcp");
383 xprt->address_strings[RPC_DISPLAY_ALL] = buf;
385 buf = kzalloc(36, GFP_KERNEL);
387 snprintf(buf, 36, NIP6_SEQFMT,
388 NIP6(addr->sin6_addr));
390 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
392 buf = kzalloc(8, GFP_KERNEL);
394 snprintf(buf, 8, "%4hx",
395 ntohs(addr->sin6_port));
397 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
399 buf = kzalloc(50, GFP_KERNEL);
401 snprintf(buf, 50, NIP6_FMT".%u.%u",
402 NIP6(addr->sin6_addr),
403 ntohs(addr->sin6_port) >> 8,
404 ntohs(addr->sin6_port) & 0xff);
406 xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
408 xprt->address_strings[RPC_DISPLAY_NETID] =
409 kstrdup(xprt->prot == IPPROTO_UDP ?
410 RPCBIND_NETID_UDP6 : RPCBIND_NETID_TCP6, GFP_KERNEL);
413 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
417 for (i = 0; i < RPC_DISPLAY_MAX; i++)
418 kfree(xprt->address_strings[i]);
421 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
423 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
425 struct msghdr msg = {
427 .msg_namelen = addrlen,
428 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
431 .iov_base = vec->iov_base + base,
432 .iov_len = vec->iov_len - base,
435 if (iov.iov_len != 0)
436 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
437 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
440 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
443 unsigned int remainder;
446 remainder = xdr->page_len - base;
447 base += xdr->page_base;
448 ppage = xdr->pages + (base >> PAGE_SHIFT);
451 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
452 int flags = XS_SENDMSG_FLAGS;
455 if (remainder != 0 || more)
457 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
458 if (remainder == 0 || err != len)
472 * xs_sendpages - write pages directly to a socket
473 * @sock: socket to send on
474 * @addr: UDP only -- address of destination
475 * @addrlen: UDP only -- length of destination address
476 * @xdr: buffer containing this request
477 * @base: starting position in the buffer
480 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
482 unsigned int remainder = xdr->len - base;
488 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
494 if (base < xdr->head[0].iov_len || addr != NULL) {
495 unsigned int len = xdr->head[0].iov_len - base;
497 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
498 if (remainder == 0 || err != len)
503 base -= xdr->head[0].iov_len;
505 if (base < xdr->page_len) {
506 unsigned int len = xdr->page_len - base;
508 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
509 if (remainder == 0 || err != len)
514 base -= xdr->page_len;
516 if (base >= xdr->tail[0].iov_len)
518 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
528 * xs_nospace - place task on wait queue if transmit was incomplete
529 * @task: task to put to sleep
532 static void xs_nospace(struct rpc_task *task)
534 struct rpc_rqst *req = task->tk_rqstp;
535 struct rpc_xprt *xprt = req->rq_xprt;
536 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
538 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
539 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
542 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
543 /* Protect against races with write_space */
544 spin_lock_bh(&xprt->transport_lock);
546 /* Don't race with disconnect */
547 if (!xprt_connected(xprt))
548 task->tk_status = -ENOTCONN;
549 else if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
550 xprt_wait_for_buffer_space(task);
552 spin_unlock_bh(&xprt->transport_lock);
554 /* Keep holding the socket if it is blocked */
555 rpc_delay(task, HZ>>4);
559 * xs_udp_send_request - write an RPC request to a UDP socket
560 * @task: address of RPC task that manages the state of an RPC request
563 * 0: The request has been sent
564 * EAGAIN: The socket was blocked, please call again later to
565 * complete the request
566 * ENOTCONN: Caller needs to invoke connect logic then call again
567 * other: Some other error occured, the request was not sent
569 static int xs_udp_send_request(struct rpc_task *task)
571 struct rpc_rqst *req = task->tk_rqstp;
572 struct rpc_xprt *xprt = req->rq_xprt;
573 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
574 struct xdr_buf *xdr = &req->rq_snd_buf;
577 xs_pktdump("packet data:",
578 req->rq_svec->iov_base,
579 req->rq_svec->iov_len);
581 req->rq_xtime = jiffies;
582 status = xs_sendpages(transport->sock,
587 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
588 xdr->len - req->rq_bytes_sent, status);
590 if (likely(status >= (int) req->rq_slen))
593 /* Still some bytes left; set up for a retry later. */
601 /* When the server has died, an ICMP port unreachable message
602 * prompts ECONNREFUSED. */
608 dprintk("RPC: sendmsg returned unrecognized error %d\n",
616 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
618 u32 reclen = buf->len - sizeof(rpc_fraghdr);
619 rpc_fraghdr *base = buf->head[0].iov_base;
620 *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
624 * xs_tcp_send_request - write an RPC request to a TCP socket
625 * @task: address of RPC task that manages the state of an RPC request
628 * 0: The request has been sent
629 * EAGAIN: The socket was blocked, please call again later to
630 * complete the request
631 * ENOTCONN: Caller needs to invoke connect logic then call again
632 * other: Some other error occured, the request was not sent
634 * XXX: In the case of soft timeouts, should we eventually give up
635 * if sendmsg is not able to make progress?
637 static int xs_tcp_send_request(struct rpc_task *task)
639 struct rpc_rqst *req = task->tk_rqstp;
640 struct rpc_xprt *xprt = req->rq_xprt;
641 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
642 struct xdr_buf *xdr = &req->rq_snd_buf;
644 unsigned int retry = 0;
646 xs_encode_tcp_record_marker(&req->rq_snd_buf);
648 xs_pktdump("packet data:",
649 req->rq_svec->iov_base,
650 req->rq_svec->iov_len);
652 /* Continue transmitting the packet/record. We must be careful
653 * to cope with writespace callbacks arriving _after_ we have
654 * called sendmsg(). */
656 req->rq_xtime = jiffies;
657 status = xs_sendpages(transport->sock,
658 NULL, 0, xdr, req->rq_bytes_sent);
660 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
661 xdr->len - req->rq_bytes_sent, status);
663 if (unlikely(status < 0))
666 /* If we've sent the entire packet, immediately
667 * reset the count of bytes sent. */
668 req->rq_bytes_sent += status;
669 task->tk_bytes_sent += status;
670 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
671 req->rq_bytes_sent = 0;
676 if (retry++ > XS_SENDMSG_RETRY)
691 dprintk("RPC: sendmsg returned unrecognized error %d\n",
693 xprt_disconnect(xprt);
701 * xs_tcp_release_xprt - clean up after a tcp transmission
705 * This cleans up if an error causes us to abort the transmission of a request.
706 * In this case, the socket may need to be reset in order to avoid confusing
709 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
711 struct rpc_rqst *req;
713 if (task != xprt->snd_task)
717 req = task->tk_rqstp;
718 if (req->rq_bytes_sent == 0)
720 if (req->rq_bytes_sent == req->rq_snd_buf.len)
722 set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
724 xprt_release_xprt(xprt, task);
728 * xs_close - close a socket
731 * This is used when all requests are complete; ie, no DRC state remains
732 * on the server we want to save.
734 static void xs_close(struct rpc_xprt *xprt)
736 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
737 struct socket *sock = transport->sock;
738 struct sock *sk = transport->inet;
741 goto clear_close_wait;
743 dprintk("RPC: xs_close xprt %p\n", xprt);
745 write_lock_bh(&sk->sk_callback_lock);
746 transport->inet = NULL;
747 transport->sock = NULL;
749 sk->sk_user_data = NULL;
750 sk->sk_data_ready = transport->old_data_ready;
751 sk->sk_state_change = transport->old_state_change;
752 sk->sk_write_space = transport->old_write_space;
753 write_unlock_bh(&sk->sk_callback_lock);
759 smp_mb__before_clear_bit();
760 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
761 smp_mb__after_clear_bit();
765 * xs_destroy - prepare to shutdown a transport
766 * @xprt: doomed transport
769 static void xs_destroy(struct rpc_xprt *xprt)
771 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
773 dprintk("RPC: xs_destroy xprt %p\n", xprt);
775 cancel_rearming_delayed_work(&transport->connect_worker);
777 xprt_disconnect(xprt);
779 xs_free_peer_addresses(xprt);
782 module_put(THIS_MODULE);
785 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
787 return (struct rpc_xprt *) sk->sk_user_data;
791 * xs_udp_data_ready - "data ready" callback for UDP sockets
792 * @sk: socket with data to read
793 * @len: how much data to read
796 static void xs_udp_data_ready(struct sock *sk, int len)
798 struct rpc_task *task;
799 struct rpc_xprt *xprt;
800 struct rpc_rqst *rovr;
802 int err, repsize, copied;
806 read_lock(&sk->sk_callback_lock);
807 dprintk("RPC: xs_udp_data_ready...\n");
808 if (!(xprt = xprt_from_sock(sk)))
811 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
817 repsize = skb->len - sizeof(struct udphdr);
819 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
823 /* Copy the XID from the skb... */
824 xp = skb_header_pointer(skb, sizeof(struct udphdr),
825 sizeof(_xid), &_xid);
829 /* Look up and lock the request corresponding to the given XID */
830 spin_lock(&xprt->transport_lock);
831 rovr = xprt_lookup_rqst(xprt, *xp);
834 task = rovr->rq_task;
836 if ((copied = rovr->rq_private_buf.buflen) > repsize)
839 /* Suck it into the iovec, verify checksum if not done by hw. */
840 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb))
843 /* Something worked... */
844 dst_confirm(skb->dst);
846 xprt_adjust_cwnd(task, copied);
847 xprt_update_rtt(task);
848 xprt_complete_rqst(task, copied);
851 spin_unlock(&xprt->transport_lock);
853 skb_free_datagram(sk, skb);
855 read_unlock(&sk->sk_callback_lock);
858 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
860 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
864 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
865 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
866 used = xdr_skb_read_bits(desc, p, len);
867 transport->tcp_offset += used;
871 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
872 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
873 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
875 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
876 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
878 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
879 transport->tcp_offset = 0;
881 /* Sanity check of the record length */
882 if (unlikely(transport->tcp_reclen < 4)) {
883 dprintk("RPC: invalid TCP record fragment length\n");
884 xprt_disconnect(xprt);
887 dprintk("RPC: reading TCP record fragment of length %d\n",
888 transport->tcp_reclen);
891 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
893 if (transport->tcp_offset == transport->tcp_reclen) {
894 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
895 transport->tcp_offset = 0;
896 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
897 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
898 transport->tcp_flags |= TCP_RCV_COPY_XID;
899 transport->tcp_copied = 0;
904 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
909 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
910 dprintk("RPC: reading XID (%Zu bytes)\n", len);
911 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
912 used = xdr_skb_read_bits(desc, p, len);
913 transport->tcp_offset += used;
916 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
917 transport->tcp_flags |= TCP_RCV_COPY_DATA;
918 transport->tcp_copied = 4;
919 dprintk("RPC: reading reply for XID %08x\n",
920 ntohl(transport->tcp_xid));
921 xs_tcp_check_fraghdr(transport);
924 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
926 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
927 struct rpc_rqst *req;
928 struct xdr_buf *rcvbuf;
932 /* Find and lock the request corresponding to this xid */
933 spin_lock(&xprt->transport_lock);
934 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
936 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
937 dprintk("RPC: XID %08x request not found!\n",
938 ntohl(transport->tcp_xid));
939 spin_unlock(&xprt->transport_lock);
943 rcvbuf = &req->rq_private_buf;
945 if (len > transport->tcp_reclen - transport->tcp_offset) {
946 struct xdr_skb_reader my_desc;
948 len = transport->tcp_reclen - transport->tcp_offset;
949 memcpy(&my_desc, desc, sizeof(my_desc));
951 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
952 &my_desc, xdr_skb_read_bits);
956 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
957 desc, xdr_skb_read_bits);
960 transport->tcp_copied += r;
961 transport->tcp_offset += r;
964 /* Error when copying to the receive buffer,
965 * usually because we weren't able to allocate
966 * additional buffer pages. All we can do now
967 * is turn off TCP_RCV_COPY_DATA, so the request
968 * will not receive any additional updates,
970 * Any remaining data from this record will
973 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
974 dprintk("RPC: XID %08x truncated request\n",
975 ntohl(transport->tcp_xid));
976 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
977 "tcp_offset = %u, tcp_reclen = %u\n",
978 xprt, transport->tcp_copied,
979 transport->tcp_offset, transport->tcp_reclen);
983 dprintk("RPC: XID %08x read %Zd bytes\n",
984 ntohl(transport->tcp_xid), r);
985 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
986 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
987 transport->tcp_offset, transport->tcp_reclen);
989 if (transport->tcp_copied == req->rq_private_buf.buflen)
990 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
991 else if (transport->tcp_offset == transport->tcp_reclen) {
992 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
993 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
997 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
998 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
999 spin_unlock(&xprt->transport_lock);
1000 xs_tcp_check_fraghdr(transport);
1003 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1007 len = transport->tcp_reclen - transport->tcp_offset;
1008 if (len > desc->count)
1011 desc->offset += len;
1012 transport->tcp_offset += len;
1013 dprintk("RPC: discarded %Zu bytes\n", len);
1014 xs_tcp_check_fraghdr(transport);
1017 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1019 struct rpc_xprt *xprt = rd_desc->arg.data;
1020 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1021 struct xdr_skb_reader desc = {
1027 dprintk("RPC: xs_tcp_data_recv started\n");
1029 /* Read in a new fragment marker if necessary */
1030 /* Can we ever really expect to get completely empty fragments? */
1031 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1032 xs_tcp_read_fraghdr(xprt, &desc);
1035 /* Read in the xid if necessary */
1036 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1037 xs_tcp_read_xid(transport, &desc);
1040 /* Read in the request data */
1041 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1042 xs_tcp_read_request(xprt, &desc);
1045 /* Skip over any trailing bytes on short reads */
1046 xs_tcp_read_discard(transport, &desc);
1047 } while (desc.count);
1048 dprintk("RPC: xs_tcp_data_recv done\n");
1049 return len - desc.count;
1053 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1054 * @sk: socket with data to read
1055 * @bytes: how much data to read
1058 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1060 struct rpc_xprt *xprt;
1061 read_descriptor_t rd_desc;
1063 dprintk("RPC: xs_tcp_data_ready...\n");
1065 read_lock(&sk->sk_callback_lock);
1066 if (!(xprt = xprt_from_sock(sk)))
1071 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1072 rd_desc.arg.data = xprt;
1073 rd_desc.count = 65536;
1074 tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1076 read_unlock(&sk->sk_callback_lock);
1080 * xs_tcp_state_change - callback to handle TCP socket state changes
1081 * @sk: socket whose state has changed
1084 static void xs_tcp_state_change(struct sock *sk)
1086 struct rpc_xprt *xprt;
1088 read_lock(&sk->sk_callback_lock);
1089 if (!(xprt = xprt_from_sock(sk)))
1091 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1092 dprintk("RPC: state %x conn %d dead %d zapped %d\n",
1093 sk->sk_state, xprt_connected(xprt),
1094 sock_flag(sk, SOCK_DEAD),
1095 sock_flag(sk, SOCK_ZAPPED));
1097 switch (sk->sk_state) {
1098 case TCP_ESTABLISHED:
1099 spin_lock_bh(&xprt->transport_lock);
1100 if (!xprt_test_and_set_connected(xprt)) {
1101 struct sock_xprt *transport = container_of(xprt,
1102 struct sock_xprt, xprt);
1104 /* Reset TCP record info */
1105 transport->tcp_offset = 0;
1106 transport->tcp_reclen = 0;
1107 transport->tcp_copied = 0;
1108 transport->tcp_flags =
1109 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1111 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1112 xprt_wake_pending_tasks(xprt, 0);
1114 spin_unlock_bh(&xprt->transport_lock);
1119 case TCP_CLOSE_WAIT:
1120 /* Try to schedule an autoclose RPC calls */
1121 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
1122 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
1123 queue_work(rpciod_workqueue, &xprt->task_cleanup);
1125 xprt_disconnect(xprt);
1128 read_unlock(&sk->sk_callback_lock);
1132 * xs_udp_write_space - callback invoked when socket buffer space
1134 * @sk: socket whose state has changed
1136 * Called when more output buffer space is available for this socket.
1137 * We try not to wake our writers until they can make "significant"
1138 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1139 * with a bunch of small requests.
1141 static void xs_udp_write_space(struct sock *sk)
1143 read_lock(&sk->sk_callback_lock);
1145 /* from net/core/sock.c:sock_def_write_space */
1146 if (sock_writeable(sk)) {
1147 struct socket *sock;
1148 struct rpc_xprt *xprt;
1150 if (unlikely(!(sock = sk->sk_socket)))
1152 if (unlikely(!(xprt = xprt_from_sock(sk))))
1154 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
1157 xprt_write_space(xprt);
1161 read_unlock(&sk->sk_callback_lock);
1165 * xs_tcp_write_space - callback invoked when socket buffer space
1167 * @sk: socket whose state has changed
1169 * Called when more output buffer space is available for this socket.
1170 * We try not to wake our writers until they can make "significant"
1171 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1172 * with a bunch of small requests.
1174 static void xs_tcp_write_space(struct sock *sk)
1176 read_lock(&sk->sk_callback_lock);
1178 /* from net/core/stream.c:sk_stream_write_space */
1179 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
1180 struct socket *sock;
1181 struct rpc_xprt *xprt;
1183 if (unlikely(!(sock = sk->sk_socket)))
1185 if (unlikely(!(xprt = xprt_from_sock(sk))))
1187 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
1190 xprt_write_space(xprt);
1194 read_unlock(&sk->sk_callback_lock);
1197 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1199 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1200 struct sock *sk = transport->inet;
1202 if (transport->rcvsize) {
1203 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1204 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1206 if (transport->sndsize) {
1207 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1208 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1209 sk->sk_write_space(sk);
1214 * xs_udp_set_buffer_size - set send and receive limits
1215 * @xprt: generic transport
1216 * @sndsize: requested size of send buffer, in bytes
1217 * @rcvsize: requested size of receive buffer, in bytes
1219 * Set socket send and receive buffer size limits.
1221 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1223 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1225 transport->sndsize = 0;
1227 transport->sndsize = sndsize + 1024;
1228 transport->rcvsize = 0;
1230 transport->rcvsize = rcvsize + 1024;
1232 xs_udp_do_set_buffer_size(xprt);
1236 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1237 * @task: task that timed out
1239 * Adjust the congestion window after a retransmit timeout has occurred.
1241 static void xs_udp_timer(struct rpc_task *task)
1243 xprt_adjust_cwnd(task, -ETIMEDOUT);
1246 static unsigned short xs_get_random_port(void)
1248 unsigned short range = xprt_max_resvport - xprt_min_resvport;
1249 unsigned short rand = (unsigned short) net_random() % range;
1250 return rand + xprt_min_resvport;
1254 * xs_set_port - reset the port number in the remote endpoint address
1255 * @xprt: generic transport
1256 * @port: new port number
1259 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1261 struct sockaddr *addr = xs_addr(xprt);
1263 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1265 switch (addr->sa_family) {
1267 ((struct sockaddr_in *)addr)->sin_port = htons(port);
1270 ((struct sockaddr_in6 *)addr)->sin6_port = htons(port);
1277 static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
1279 struct sockaddr_in myaddr = {
1280 .sin_family = AF_INET,
1282 struct sockaddr_in *sa;
1284 unsigned short port = transport->port;
1286 if (!transport->xprt.resvport)
1288 sa = (struct sockaddr_in *)&transport->addr;
1289 myaddr.sin_addr = sa->sin_addr;
1291 myaddr.sin_port = htons(port);
1292 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1294 if (!transport->xprt.resvport)
1297 transport->port = port;
1300 if (port <= xprt_min_resvport)
1301 port = xprt_max_resvport;
1304 } while (err == -EADDRINUSE && port != transport->port);
1305 dprintk("RPC: %s "NIPQUAD_FMT":%u: %s (%d)\n",
1306 __FUNCTION__, NIPQUAD(myaddr.sin_addr),
1307 port, err ? "failed" : "ok", err);
1311 static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1313 struct sockaddr_in6 myaddr = {
1314 .sin6_family = AF_INET6,
1316 struct sockaddr_in6 *sa;
1318 unsigned short port = transport->port;
1320 if (!transport->xprt.resvport)
1322 sa = (struct sockaddr_in6 *)&transport->addr;
1323 myaddr.sin6_addr = sa->sin6_addr;
1325 myaddr.sin6_port = htons(port);
1326 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1328 if (!transport->xprt.resvport)
1331 transport->port = port;
1334 if (port <= xprt_min_resvport)
1335 port = xprt_max_resvport;
1338 } while (err == -EADDRINUSE && port != transport->port);
1339 dprintk("RPC: xs_bind6 "NIP6_FMT":%u: %s (%d)\n",
1340 NIP6(myaddr.sin6_addr), port, err ? "failed" : "ok", err);
1344 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1345 static struct lock_class_key xs_key[2];
1346 static struct lock_class_key xs_slock_key[2];
1348 static inline void xs_reclassify_socket4(struct socket *sock)
1350 struct sock *sk = sock->sk;
1352 BUG_ON(sk->sk_lock.owner != NULL);
1353 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1354 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1357 static inline void xs_reclassify_socket6(struct socket *sock)
1359 struct sock *sk = sock->sk;
1361 BUG_ON(sk->sk_lock.owner != NULL);
1362 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1363 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1366 static inline void xs_reclassify_socket4(struct socket *sock)
1370 static inline void xs_reclassify_socket6(struct socket *sock)
1375 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1377 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1379 if (!transport->inet) {
1380 struct sock *sk = sock->sk;
1382 write_lock_bh(&sk->sk_callback_lock);
1384 sk->sk_user_data = xprt;
1385 transport->old_data_ready = sk->sk_data_ready;
1386 transport->old_state_change = sk->sk_state_change;
1387 transport->old_write_space = sk->sk_write_space;
1388 sk->sk_data_ready = xs_udp_data_ready;
1389 sk->sk_write_space = xs_udp_write_space;
1390 sk->sk_no_check = UDP_CSUM_NORCV;
1391 sk->sk_allocation = GFP_ATOMIC;
1393 xprt_set_connected(xprt);
1395 /* Reset to new socket */
1396 transport->sock = sock;
1397 transport->inet = sk;
1399 write_unlock_bh(&sk->sk_callback_lock);
1401 xs_udp_do_set_buffer_size(xprt);
1405 * xs_udp_connect_worker4 - set up a UDP socket
1406 * @work: RPC transport to connect
1408 * Invoked by a work queue tasklet.
1410 static void xs_udp_connect_worker4(struct work_struct *work)
1412 struct sock_xprt *transport =
1413 container_of(work, struct sock_xprt, connect_worker.work);
1414 struct rpc_xprt *xprt = &transport->xprt;
1415 struct socket *sock = transport->sock;
1416 int err, status = -EIO;
1418 if (xprt->shutdown || !xprt_bound(xprt))
1421 /* Start by resetting any existing state */
1424 if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1425 dprintk("RPC: can't create UDP transport socket (%d).\n", -err);
1428 xs_reclassify_socket4(sock);
1430 if (xs_bind4(transport, sock)) {
1435 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1436 xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1438 xs_udp_finish_connecting(xprt, sock);
1441 xprt_wake_pending_tasks(xprt, status);
1442 xprt_clear_connecting(xprt);
1446 * xs_udp_connect_worker6 - set up a UDP socket
1447 * @work: RPC transport to connect
1449 * Invoked by a work queue tasklet.
1451 static void xs_udp_connect_worker6(struct work_struct *work)
1453 struct sock_xprt *transport =
1454 container_of(work, struct sock_xprt, connect_worker.work);
1455 struct rpc_xprt *xprt = &transport->xprt;
1456 struct socket *sock = transport->sock;
1457 int err, status = -EIO;
1459 if (xprt->shutdown || !xprt_bound(xprt))
1462 /* Start by resetting any existing state */
1465 if ((err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1466 dprintk("RPC: can't create UDP transport socket (%d).\n", -err);
1469 xs_reclassify_socket6(sock);
1471 if (xs_bind6(transport, sock) < 0) {
1476 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1477 xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1479 xs_udp_finish_connecting(xprt, sock);
1482 xprt_wake_pending_tasks(xprt, status);
1483 xprt_clear_connecting(xprt);
1487 * We need to preserve the port number so the reply cache on the server can
1488 * find our cached RPC replies when we get around to reconnecting.
1490 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1493 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1494 struct sockaddr any;
1496 dprintk("RPC: disconnecting xprt %p to reuse port\n", xprt);
1499 * Disconnect the transport socket by doing a connect operation
1500 * with AF_UNSPEC. This should return immediately...
1502 memset(&any, 0, sizeof(any));
1503 any.sa_family = AF_UNSPEC;
1504 result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1506 dprintk("RPC: AF_UNSPEC connect return code %d\n",
1510 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1512 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1514 if (!transport->inet) {
1515 struct sock *sk = sock->sk;
1517 write_lock_bh(&sk->sk_callback_lock);
1519 sk->sk_user_data = xprt;
1520 transport->old_data_ready = sk->sk_data_ready;
1521 transport->old_state_change = sk->sk_state_change;
1522 transport->old_write_space = sk->sk_write_space;
1523 sk->sk_data_ready = xs_tcp_data_ready;
1524 sk->sk_state_change = xs_tcp_state_change;
1525 sk->sk_write_space = xs_tcp_write_space;
1526 sk->sk_allocation = GFP_ATOMIC;
1528 /* socket options */
1529 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1530 sock_reset_flag(sk, SOCK_LINGER);
1531 tcp_sk(sk)->linger2 = 0;
1532 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1534 xprt_clear_connected(xprt);
1536 /* Reset to new socket */
1537 transport->sock = sock;
1538 transport->inet = sk;
1540 write_unlock_bh(&sk->sk_callback_lock);
1543 /* Tell the socket layer to start connecting... */
1544 xprt->stat.connect_count++;
1545 xprt->stat.connect_start = jiffies;
1546 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
1550 * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1551 * @work: RPC transport to connect
1553 * Invoked by a work queue tasklet.
1555 static void xs_tcp_connect_worker4(struct work_struct *work)
1557 struct sock_xprt *transport =
1558 container_of(work, struct sock_xprt, connect_worker.work);
1559 struct rpc_xprt *xprt = &transport->xprt;
1560 struct socket *sock = transport->sock;
1561 int err, status = -EIO;
1563 if (xprt->shutdown || !xprt_bound(xprt))
1567 /* start from scratch */
1568 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1569 dprintk("RPC: can't create TCP transport socket (%d).\n", -err);
1572 xs_reclassify_socket4(sock);
1574 if (xs_bind4(transport, sock) < 0) {
1579 /* "close" the socket, preserving the local port */
1580 xs_tcp_reuse_connection(xprt);
1582 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1583 xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1585 status = xs_tcp_finish_connecting(xprt, sock);
1586 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
1587 xprt, -status, xprt_connected(xprt),
1588 sock->sk->sk_state);
1596 /* retry with existing socket, after a delay */
1599 /* get rid of existing socket, and retry */
1605 xprt_wake_pending_tasks(xprt, status);
1607 xprt_clear_connecting(xprt);
1611 * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
1612 * @work: RPC transport to connect
1614 * Invoked by a work queue tasklet.
1616 static void xs_tcp_connect_worker6(struct work_struct *work)
1618 struct sock_xprt *transport =
1619 container_of(work, struct sock_xprt, connect_worker.work);
1620 struct rpc_xprt *xprt = &transport->xprt;
1621 struct socket *sock = transport->sock;
1622 int err, status = -EIO;
1624 if (xprt->shutdown || !xprt_bound(xprt))
1628 /* start from scratch */
1629 if ((err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1630 dprintk("RPC: can't create TCP transport socket (%d).\n", -err);
1633 xs_reclassify_socket6(sock);
1635 if (xs_bind6(transport, sock) < 0) {
1640 /* "close" the socket, preserving the local port */
1641 xs_tcp_reuse_connection(xprt);
1643 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1644 xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1646 status = xs_tcp_finish_connecting(xprt, sock);
1647 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
1648 xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1656 /* retry with existing socket, after a delay */
1659 /* get rid of existing socket, and retry */
1665 xprt_wake_pending_tasks(xprt, status);
1667 xprt_clear_connecting(xprt);
1671 * xs_connect - connect a socket to a remote endpoint
1672 * @task: address of RPC task that manages state of connect request
1674 * TCP: If the remote end dropped the connection, delay reconnecting.
1676 * UDP socket connects are synchronous, but we use a work queue anyway
1677 * to guarantee that even unprivileged user processes can set up a
1678 * socket on a privileged port.
1680 * If a UDP socket connect fails, the delay behavior here prevents
1681 * retry floods (hard mounts).
1683 static void xs_connect(struct rpc_task *task)
1685 struct rpc_xprt *xprt = task->tk_xprt;
1686 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1688 if (xprt_test_and_set_connecting(xprt))
1691 if (transport->sock != NULL) {
1692 dprintk("RPC: xs_connect delayed xprt %p for %lu "
1694 xprt, xprt->reestablish_timeout / HZ);
1695 queue_delayed_work(rpciod_workqueue,
1696 &transport->connect_worker,
1697 xprt->reestablish_timeout);
1698 xprt->reestablish_timeout <<= 1;
1699 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1700 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1702 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
1703 queue_delayed_work(rpciod_workqueue,
1704 &transport->connect_worker, 0);
1709 * xs_udp_print_stats - display UDP socket-specifc stats
1710 * @xprt: rpc_xprt struct containing statistics
1714 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1716 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1718 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1720 xprt->stat.bind_count,
1723 xprt->stat.bad_xids,
1725 xprt->stat.bklog_u);
1729 * xs_tcp_print_stats - display TCP socket-specifc stats
1730 * @xprt: rpc_xprt struct containing statistics
1734 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1736 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1739 if (xprt_connected(xprt))
1740 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1742 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1744 xprt->stat.bind_count,
1745 xprt->stat.connect_count,
1746 xprt->stat.connect_time,
1750 xprt->stat.bad_xids,
1752 xprt->stat.bklog_u);
1755 static struct rpc_xprt_ops xs_udp_ops = {
1756 .set_buffer_size = xs_udp_set_buffer_size,
1757 .reserve_xprt = xprt_reserve_xprt_cong,
1758 .release_xprt = xprt_release_xprt_cong,
1759 .rpcbind = rpcb_getport_async,
1760 .set_port = xs_set_port,
1761 .connect = xs_connect,
1762 .buf_alloc = rpc_malloc,
1763 .buf_free = rpc_free,
1764 .send_request = xs_udp_send_request,
1765 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
1766 .timer = xs_udp_timer,
1767 .release_request = xprt_release_rqst_cong,
1769 .destroy = xs_destroy,
1770 .print_stats = xs_udp_print_stats,
1773 static struct rpc_xprt_ops xs_tcp_ops = {
1774 .reserve_xprt = xprt_reserve_xprt,
1775 .release_xprt = xs_tcp_release_xprt,
1776 .rpcbind = rpcb_getport_async,
1777 .set_port = xs_set_port,
1778 .connect = xs_connect,
1779 .buf_alloc = rpc_malloc,
1780 .buf_free = rpc_free,
1781 .send_request = xs_tcp_send_request,
1782 .set_retrans_timeout = xprt_set_retrans_timeout_def,
1784 .destroy = xs_destroy,
1785 .print_stats = xs_tcp_print_stats,
1788 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
1789 unsigned int slot_table_size)
1791 struct rpc_xprt *xprt;
1792 struct sock_xprt *new;
1794 if (args->addrlen > sizeof(xprt->addr)) {
1795 dprintk("RPC: xs_setup_xprt: address too large\n");
1796 return ERR_PTR(-EBADF);
1799 new = kzalloc(sizeof(*new), GFP_KERNEL);
1801 dprintk("RPC: xs_setup_xprt: couldn't allocate "
1803 return ERR_PTR(-ENOMEM);
1807 xprt->max_reqs = slot_table_size;
1808 xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1809 if (xprt->slot == NULL) {
1811 dprintk("RPC: xs_setup_xprt: couldn't allocate slot "
1813 return ERR_PTR(-ENOMEM);
1816 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
1817 xprt->addrlen = args->addrlen;
1819 memcpy(&new->addr, args->srcaddr, args->addrlen);
1820 new->port = xs_get_random_port();
1826 * xs_setup_udp - Set up transport to use a UDP socket
1827 * @args: rpc transport creation arguments
1830 struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
1832 struct sockaddr *addr = args->dstaddr;
1833 struct rpc_xprt *xprt;
1834 struct sock_xprt *transport;
1836 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
1839 transport = container_of(xprt, struct sock_xprt, xprt);
1841 xprt->prot = IPPROTO_UDP;
1843 /* XXX: header size can vary due to auth type, IPv6, etc. */
1844 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1846 xprt->bind_timeout = XS_BIND_TO;
1847 xprt->connect_timeout = XS_UDP_CONN_TO;
1848 xprt->reestablish_timeout = XS_UDP_REEST_TO;
1849 xprt->idle_timeout = XS_IDLE_DISC_TO;
1851 xprt->ops = &xs_udp_ops;
1854 xprt->timeout = *args->timeout;
1856 xprt_set_timeout(&xprt->timeout, 5, 5 * HZ);
1858 switch (addr->sa_family) {
1860 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1861 xprt_set_bound(xprt);
1863 INIT_DELAYED_WORK(&transport->connect_worker,
1864 xs_udp_connect_worker4);
1865 xs_format_ipv4_peer_addresses(xprt);
1868 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
1869 xprt_set_bound(xprt);
1871 INIT_DELAYED_WORK(&transport->connect_worker,
1872 xs_udp_connect_worker6);
1873 xs_format_ipv6_peer_addresses(xprt);
1877 return ERR_PTR(-EAFNOSUPPORT);
1880 dprintk("RPC: set up transport to address %s\n",
1881 xprt->address_strings[RPC_DISPLAY_ALL]);
1883 if (try_module_get(THIS_MODULE))
1888 return ERR_PTR(-EINVAL);
1892 * xs_setup_tcp - Set up transport to use a TCP socket
1893 * @args: rpc transport creation arguments
1896 struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
1898 struct sockaddr *addr = args->dstaddr;
1899 struct rpc_xprt *xprt;
1900 struct sock_xprt *transport;
1902 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
1905 transport = container_of(xprt, struct sock_xprt, xprt);
1907 xprt->prot = IPPROTO_TCP;
1908 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
1909 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
1911 xprt->bind_timeout = XS_BIND_TO;
1912 xprt->connect_timeout = XS_TCP_CONN_TO;
1913 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1914 xprt->idle_timeout = XS_IDLE_DISC_TO;
1916 xprt->ops = &xs_tcp_ops;
1919 xprt->timeout = *args->timeout;
1921 xprt_set_timeout(&xprt->timeout, 2, 60 * HZ);
1923 switch (addr->sa_family) {
1925 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1926 xprt_set_bound(xprt);
1928 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker4);
1929 xs_format_ipv4_peer_addresses(xprt);
1932 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
1933 xprt_set_bound(xprt);
1935 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker6);
1936 xs_format_ipv6_peer_addresses(xprt);
1940 return ERR_PTR(-EAFNOSUPPORT);
1943 dprintk("RPC: set up transport to address %s\n",
1944 xprt->address_strings[RPC_DISPLAY_ALL]);
1946 if (try_module_get(THIS_MODULE))
1951 return ERR_PTR(-EINVAL);
1954 static struct xprt_class xs_udp_transport = {
1955 .list = LIST_HEAD_INIT(xs_udp_transport.list),
1957 .owner = THIS_MODULE,
1958 .ident = IPPROTO_UDP,
1959 .setup = xs_setup_udp,
1962 static struct xprt_class xs_tcp_transport = {
1963 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
1965 .owner = THIS_MODULE,
1966 .ident = IPPROTO_TCP,
1967 .setup = xs_setup_tcp,
1971 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
1974 int init_socket_xprt(void)
1977 if (!sunrpc_table_header)
1978 sunrpc_table_header = register_sysctl_table(sunrpc_table);
1981 xprt_register_transport(&xs_udp_transport);
1982 xprt_register_transport(&xs_tcp_transport);
1988 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
1991 void cleanup_socket_xprt(void)
1994 if (sunrpc_table_header) {
1995 unregister_sysctl_table(sunrpc_table_header);
1996 sunrpc_table_header = NULL;
2000 xprt_unregister_transport(&xs_udp_transport);
2001 xprt_unregister_transport(&xs_tcp_transport);