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/capability.h>
24 #include <linux/pagemap.h>
25 #include <linux/errno.h>
26 #include <linux/socket.h>
28 #include <linux/net.h>
30 #include <linux/udp.h>
31 #include <linux/tcp.h>
32 #include <linux/sunrpc/clnt.h>
33 #include <linux/sunrpc/sched.h>
34 #include <linux/file.h>
37 #include <net/checksum.h>
44 unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
45 unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE;
47 unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
48 unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
51 * We can register our own files under /proc/sys/sunrpc by
52 * calling register_sysctl_table() again. The files in that
53 * directory become the union of all files registered there.
55 * We simply need to make sure that we don't collide with
56 * someone else's file names!
61 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
62 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
63 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
64 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
66 static struct ctl_table_header *sunrpc_table_header;
69 * FIXME: changing the UDP slot table size should also resize the UDP
70 * socket buffers for existing UDP transports
72 static ctl_table xs_tunables_table[] = {
74 .ctl_name = CTL_SLOTTABLE_UDP,
75 .procname = "udp_slot_table_entries",
76 .data = &xprt_udp_slot_table_entries,
77 .maxlen = sizeof(unsigned int),
79 .proc_handler = &proc_dointvec_minmax,
80 .strategy = &sysctl_intvec,
81 .extra1 = &min_slot_table_size,
82 .extra2 = &max_slot_table_size
85 .ctl_name = CTL_SLOTTABLE_TCP,
86 .procname = "tcp_slot_table_entries",
87 .data = &xprt_tcp_slot_table_entries,
88 .maxlen = sizeof(unsigned int),
90 .proc_handler = &proc_dointvec_minmax,
91 .strategy = &sysctl_intvec,
92 .extra1 = &min_slot_table_size,
93 .extra2 = &max_slot_table_size
96 .ctl_name = CTL_MIN_RESVPORT,
97 .procname = "min_resvport",
98 .data = &xprt_min_resvport,
99 .maxlen = sizeof(unsigned int),
101 .proc_handler = &proc_dointvec_minmax,
102 .strategy = &sysctl_intvec,
103 .extra1 = &xprt_min_resvport_limit,
104 .extra2 = &xprt_max_resvport_limit
107 .ctl_name = CTL_MAX_RESVPORT,
108 .procname = "max_resvport",
109 .data = &xprt_max_resvport,
110 .maxlen = sizeof(unsigned int),
112 .proc_handler = &proc_dointvec_minmax,
113 .strategy = &sysctl_intvec,
114 .extra1 = &xprt_min_resvport_limit,
115 .extra2 = &xprt_max_resvport_limit
122 static ctl_table sunrpc_table[] = {
124 .ctl_name = CTL_SUNRPC,
125 .procname = "sunrpc",
127 .child = xs_tunables_table
137 * How many times to try sending a request on a socket before waiting
138 * for the socket buffer to clear.
140 #define XS_SENDMSG_RETRY (10U)
143 * Time out for an RPC UDP socket connect. UDP socket connects are
144 * synchronous, but we set a timeout anyway in case of resource
145 * exhaustion on the local host.
147 #define XS_UDP_CONN_TO (5U * HZ)
150 * Wait duration for an RPC TCP connection to be established. Solaris
151 * NFS over TCP uses 60 seconds, for example, which is in line with how
152 * long a server takes to reboot.
154 #define XS_TCP_CONN_TO (60U * HZ)
157 * Wait duration for a reply from the RPC portmapper.
159 #define XS_BIND_TO (60U * HZ)
162 * Delay if a UDP socket connect error occurs. This is most likely some
163 * kind of resource problem on the local host.
165 #define XS_UDP_REEST_TO (2U * HZ)
168 * The reestablish timeout allows clients to delay for a bit before attempting
169 * to reconnect to a server that just dropped our connection.
171 * We implement an exponential backoff when trying to reestablish a TCP
172 * transport connection with the server. Some servers like to drop a TCP
173 * connection when they are overworked, so we start with a short timeout and
174 * increase over time if the server is down or not responding.
176 #define XS_TCP_INIT_REEST_TO (3U * HZ)
177 #define XS_TCP_MAX_REEST_TO (5U * 60 * HZ)
180 * TCP idle timeout; client drops the transport socket if it is idle
181 * for this long. Note that we also timeout UDP sockets to prevent
182 * holding port numbers when there is no RPC traffic.
184 #define XS_IDLE_DISC_TO (5U * 60 * HZ)
187 # undef RPC_DEBUG_DATA
188 # define RPCDBG_FACILITY RPCDBG_TRANS
191 #ifdef RPC_DEBUG_DATA
192 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
194 u8 *buf = (u8 *) packet;
197 dprintk("RPC: %s\n", msg);
198 for (j = 0; j < count && j < 128; j += 4) {
202 dprintk("0x%04x ", j);
204 dprintk("%02x%02x%02x%02x ",
205 buf[j], buf[j+1], buf[j+2], buf[j+3]);
210 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
217 struct rpc_xprt xprt;
222 struct socket * sock;
226 * State of TCP reply receive
234 unsigned long tcp_copied,
238 * Connection of transports
240 struct delayed_work connect_worker;
241 struct sockaddr_storage addr;
245 * UDP socket buffer size parameters
251 * Saved socket callback addresses
253 void (*old_data_ready)(struct sock *, int);
254 void (*old_state_change)(struct sock *);
255 void (*old_write_space)(struct sock *);
259 * TCP receive state flags
261 #define TCP_RCV_LAST_FRAG (1UL << 0)
262 #define TCP_RCV_COPY_FRAGHDR (1UL << 1)
263 #define TCP_RCV_COPY_XID (1UL << 2)
264 #define TCP_RCV_COPY_DATA (1UL << 3)
266 static void xs_format_ipv4_peer_addresses(struct rpc_xprt *xprt)
268 struct sockaddr_in *addr = (struct sockaddr_in *) &xprt->addr;
271 buf = kzalloc(20, GFP_KERNEL);
273 snprintf(buf, 20, NIPQUAD_FMT,
274 NIPQUAD(addr->sin_addr.s_addr));
276 xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
278 buf = kzalloc(8, GFP_KERNEL);
280 snprintf(buf, 8, "%u",
281 ntohs(addr->sin_port));
283 xprt->address_strings[RPC_DISPLAY_PORT] = buf;
285 buf = kzalloc(8, GFP_KERNEL);
287 if (xprt->prot == IPPROTO_UDP)
288 snprintf(buf, 8, "udp");
290 snprintf(buf, 8, "tcp");
292 xprt->address_strings[RPC_DISPLAY_PROTO] = buf;
294 buf = kzalloc(48, GFP_KERNEL);
296 snprintf(buf, 48, "addr="NIPQUAD_FMT" port=%u proto=%s",
297 NIPQUAD(addr->sin_addr.s_addr),
298 ntohs(addr->sin_port),
299 xprt->prot == IPPROTO_UDP ? "udp" : "tcp");
301 xprt->address_strings[RPC_DISPLAY_ALL] = buf;
303 buf = kzalloc(10, GFP_KERNEL);
305 snprintf(buf, 10, "%02x%02x%02x%02x",
306 NIPQUAD(addr->sin_addr.s_addr));
308 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
310 buf = kzalloc(8, GFP_KERNEL);
312 snprintf(buf, 8, "%4hx",
313 ntohs(addr->sin_port));
315 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
318 static void xs_format_ipv6_peer_addresses(struct rpc_xprt *xprt)
320 struct sockaddr_in6 *addr = (struct sockaddr_in6 *) &xprt->addr;
323 buf = kzalloc(40, GFP_KERNEL);
325 snprintf(buf, 40, NIP6_FMT,
326 NIP6(addr->sin6_addr));
328 xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
330 buf = kzalloc(8, GFP_KERNEL);
332 snprintf(buf, 8, "%u",
333 ntohs(addr->sin6_port));
335 xprt->address_strings[RPC_DISPLAY_PORT] = buf;
337 buf = kzalloc(8, GFP_KERNEL);
339 if (xprt->prot == IPPROTO_UDP)
340 snprintf(buf, 8, "udp");
342 snprintf(buf, 8, "tcp");
344 xprt->address_strings[RPC_DISPLAY_PROTO] = buf;
346 buf = kzalloc(64, GFP_KERNEL);
348 snprintf(buf, 64, "addr="NIP6_FMT" port=%u proto=%s",
349 NIP6(addr->sin6_addr),
350 ntohs(addr->sin6_port),
351 xprt->prot == IPPROTO_UDP ? "udp" : "tcp");
353 xprt->address_strings[RPC_DISPLAY_ALL] = buf;
355 buf = kzalloc(36, GFP_KERNEL);
357 snprintf(buf, 36, NIP6_SEQFMT,
358 NIP6(addr->sin6_addr));
360 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
362 buf = kzalloc(8, GFP_KERNEL);
364 snprintf(buf, 8, "%4hx",
365 ntohs(addr->sin6_port));
367 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
370 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
374 for (i = 0; i < RPC_DISPLAY_MAX; i++)
375 kfree(xprt->address_strings[i]);
378 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
380 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
382 struct msghdr msg = {
384 .msg_namelen = addrlen,
385 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
388 .iov_base = vec->iov_base + base,
389 .iov_len = vec->iov_len - base,
392 if (iov.iov_len != 0)
393 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
394 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
397 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
400 unsigned int remainder;
403 remainder = xdr->page_len - base;
404 base += xdr->page_base;
405 ppage = xdr->pages + (base >> PAGE_SHIFT);
408 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
409 int flags = XS_SENDMSG_FLAGS;
412 if (remainder != 0 || more)
414 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
415 if (remainder == 0 || err != len)
429 * xs_sendpages - write pages directly to a socket
430 * @sock: socket to send on
431 * @addr: UDP only -- address of destination
432 * @addrlen: UDP only -- length of destination address
433 * @xdr: buffer containing this request
434 * @base: starting position in the buffer
437 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
439 unsigned int remainder = xdr->len - base;
445 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
451 if (base < xdr->head[0].iov_len || addr != NULL) {
452 unsigned int len = xdr->head[0].iov_len - base;
454 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
455 if (remainder == 0 || err != len)
460 base -= xdr->head[0].iov_len;
462 if (base < xdr->page_len) {
463 unsigned int len = xdr->page_len - base;
465 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
466 if (remainder == 0 || err != len)
471 base -= xdr->page_len;
473 if (base >= xdr->tail[0].iov_len)
475 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
485 * xs_nospace - place task on wait queue if transmit was incomplete
486 * @task: task to put to sleep
489 static void xs_nospace(struct rpc_task *task)
491 struct rpc_rqst *req = task->tk_rqstp;
492 struct rpc_xprt *xprt = req->rq_xprt;
493 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
495 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
496 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
499 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
500 /* Protect against races with write_space */
501 spin_lock_bh(&xprt->transport_lock);
503 /* Don't race with disconnect */
504 if (!xprt_connected(xprt))
505 task->tk_status = -ENOTCONN;
506 else if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
507 xprt_wait_for_buffer_space(task);
509 spin_unlock_bh(&xprt->transport_lock);
511 /* Keep holding the socket if it is blocked */
512 rpc_delay(task, HZ>>4);
516 * xs_udp_send_request - write an RPC request to a UDP socket
517 * @task: address of RPC task that manages the state of an RPC request
520 * 0: The request has been sent
521 * EAGAIN: The socket was blocked, please call again later to
522 * complete the request
523 * ENOTCONN: Caller needs to invoke connect logic then call again
524 * other: Some other error occured, the request was not sent
526 static int xs_udp_send_request(struct rpc_task *task)
528 struct rpc_rqst *req = task->tk_rqstp;
529 struct rpc_xprt *xprt = req->rq_xprt;
530 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
531 struct xdr_buf *xdr = &req->rq_snd_buf;
534 xs_pktdump("packet data:",
535 req->rq_svec->iov_base,
536 req->rq_svec->iov_len);
538 req->rq_xtime = jiffies;
539 status = xs_sendpages(transport->sock,
540 (struct sockaddr *) &xprt->addr,
544 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
545 xdr->len - req->rq_bytes_sent, status);
547 if (likely(status >= (int) req->rq_slen))
550 /* Still some bytes left; set up for a retry later. */
558 /* When the server has died, an ICMP port unreachable message
559 * prompts ECONNREFUSED. */
565 dprintk("RPC: sendmsg returned unrecognized error %d\n",
573 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
575 u32 reclen = buf->len - sizeof(rpc_fraghdr);
576 rpc_fraghdr *base = buf->head[0].iov_base;
577 *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
581 * xs_tcp_send_request - write an RPC request to a TCP socket
582 * @task: address of RPC task that manages the state of an RPC request
585 * 0: The request has been sent
586 * EAGAIN: The socket was blocked, please call again later to
587 * complete the request
588 * ENOTCONN: Caller needs to invoke connect logic then call again
589 * other: Some other error occured, the request was not sent
591 * XXX: In the case of soft timeouts, should we eventually give up
592 * if sendmsg is not able to make progress?
594 static int xs_tcp_send_request(struct rpc_task *task)
596 struct rpc_rqst *req = task->tk_rqstp;
597 struct rpc_xprt *xprt = req->rq_xprt;
598 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
599 struct xdr_buf *xdr = &req->rq_snd_buf;
601 unsigned int retry = 0;
603 xs_encode_tcp_record_marker(&req->rq_snd_buf);
605 xs_pktdump("packet data:",
606 req->rq_svec->iov_base,
607 req->rq_svec->iov_len);
609 /* Continue transmitting the packet/record. We must be careful
610 * to cope with writespace callbacks arriving _after_ we have
611 * called sendmsg(). */
613 req->rq_xtime = jiffies;
614 status = xs_sendpages(transport->sock,
615 NULL, 0, xdr, req->rq_bytes_sent);
617 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
618 xdr->len - req->rq_bytes_sent, status);
620 if (unlikely(status < 0))
623 /* If we've sent the entire packet, immediately
624 * reset the count of bytes sent. */
625 req->rq_bytes_sent += status;
626 task->tk_bytes_sent += status;
627 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
628 req->rq_bytes_sent = 0;
633 if (retry++ > XS_SENDMSG_RETRY)
648 dprintk("RPC: sendmsg returned unrecognized error %d\n",
650 xprt_disconnect(xprt);
658 * xs_tcp_release_xprt - clean up after a tcp transmission
662 * This cleans up if an error causes us to abort the transmission of a request.
663 * In this case, the socket may need to be reset in order to avoid confusing
666 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
668 struct rpc_rqst *req;
670 if (task != xprt->snd_task)
674 req = task->tk_rqstp;
675 if (req->rq_bytes_sent == 0)
677 if (req->rq_bytes_sent == req->rq_snd_buf.len)
679 set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
681 xprt_release_xprt(xprt, task);
685 * xs_close - close a socket
688 * This is used when all requests are complete; ie, no DRC state remains
689 * on the server we want to save.
691 static void xs_close(struct rpc_xprt *xprt)
693 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
694 struct socket *sock = transport->sock;
695 struct sock *sk = transport->inet;
698 goto clear_close_wait;
700 dprintk("RPC: xs_close xprt %p\n", xprt);
702 write_lock_bh(&sk->sk_callback_lock);
703 transport->inet = NULL;
704 transport->sock = NULL;
706 sk->sk_user_data = NULL;
707 sk->sk_data_ready = transport->old_data_ready;
708 sk->sk_state_change = transport->old_state_change;
709 sk->sk_write_space = transport->old_write_space;
710 write_unlock_bh(&sk->sk_callback_lock);
716 smp_mb__before_clear_bit();
717 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
718 smp_mb__after_clear_bit();
722 * xs_destroy - prepare to shutdown a transport
723 * @xprt: doomed transport
726 static void xs_destroy(struct rpc_xprt *xprt)
728 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
730 dprintk("RPC: xs_destroy xprt %p\n", xprt);
732 cancel_rearming_delayed_work(&transport->connect_worker);
734 xprt_disconnect(xprt);
736 xs_free_peer_addresses(xprt);
741 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
743 return (struct rpc_xprt *) sk->sk_user_data;
747 * xs_udp_data_ready - "data ready" callback for UDP sockets
748 * @sk: socket with data to read
749 * @len: how much data to read
752 static void xs_udp_data_ready(struct sock *sk, int len)
754 struct rpc_task *task;
755 struct rpc_xprt *xprt;
756 struct rpc_rqst *rovr;
758 int err, repsize, copied;
762 read_lock(&sk->sk_callback_lock);
763 dprintk("RPC: xs_udp_data_ready...\n");
764 if (!(xprt = xprt_from_sock(sk)))
767 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
773 repsize = skb->len - sizeof(struct udphdr);
775 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
779 /* Copy the XID from the skb... */
780 xp = skb_header_pointer(skb, sizeof(struct udphdr),
781 sizeof(_xid), &_xid);
785 /* Look up and lock the request corresponding to the given XID */
786 spin_lock(&xprt->transport_lock);
787 rovr = xprt_lookup_rqst(xprt, *xp);
790 task = rovr->rq_task;
792 if ((copied = rovr->rq_private_buf.buflen) > repsize)
795 /* Suck it into the iovec, verify checksum if not done by hw. */
796 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb))
799 /* Something worked... */
800 dst_confirm(skb->dst);
802 xprt_adjust_cwnd(task, copied);
803 xprt_update_rtt(task);
804 xprt_complete_rqst(task, copied);
807 spin_unlock(&xprt->transport_lock);
809 skb_free_datagram(sk, skb);
811 read_unlock(&sk->sk_callback_lock);
814 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
816 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
820 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
821 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
822 used = xdr_skb_read_bits(desc, p, len);
823 transport->tcp_offset += used;
827 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
828 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
829 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
831 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
832 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
834 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
835 transport->tcp_offset = 0;
837 /* Sanity check of the record length */
838 if (unlikely(transport->tcp_reclen < 4)) {
839 dprintk("RPC: invalid TCP record fragment length\n");
840 xprt_disconnect(xprt);
843 dprintk("RPC: reading TCP record fragment of length %d\n",
844 transport->tcp_reclen);
847 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
849 if (transport->tcp_offset == transport->tcp_reclen) {
850 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
851 transport->tcp_offset = 0;
852 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
853 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
854 transport->tcp_flags |= TCP_RCV_COPY_XID;
855 transport->tcp_copied = 0;
860 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
865 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
866 dprintk("RPC: reading XID (%Zu bytes)\n", len);
867 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
868 used = xdr_skb_read_bits(desc, p, len);
869 transport->tcp_offset += used;
872 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
873 transport->tcp_flags |= TCP_RCV_COPY_DATA;
874 transport->tcp_copied = 4;
875 dprintk("RPC: reading reply for XID %08x\n",
876 ntohl(transport->tcp_xid));
877 xs_tcp_check_fraghdr(transport);
880 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
882 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
883 struct rpc_rqst *req;
884 struct xdr_buf *rcvbuf;
888 /* Find and lock the request corresponding to this xid */
889 spin_lock(&xprt->transport_lock);
890 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
892 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
893 dprintk("RPC: XID %08x request not found!\n",
894 ntohl(transport->tcp_xid));
895 spin_unlock(&xprt->transport_lock);
899 rcvbuf = &req->rq_private_buf;
901 if (len > transport->tcp_reclen - transport->tcp_offset) {
902 struct xdr_skb_reader my_desc;
904 len = transport->tcp_reclen - transport->tcp_offset;
905 memcpy(&my_desc, desc, sizeof(my_desc));
907 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
908 &my_desc, xdr_skb_read_bits);
912 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
913 desc, xdr_skb_read_bits);
916 transport->tcp_copied += r;
917 transport->tcp_offset += r;
920 /* Error when copying to the receive buffer,
921 * usually because we weren't able to allocate
922 * additional buffer pages. All we can do now
923 * is turn off TCP_RCV_COPY_DATA, so the request
924 * will not receive any additional updates,
926 * Any remaining data from this record will
929 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
930 dprintk("RPC: XID %08x truncated request\n",
931 ntohl(transport->tcp_xid));
932 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
933 "tcp_offset = %u, tcp_reclen = %u\n",
934 xprt, transport->tcp_copied,
935 transport->tcp_offset, transport->tcp_reclen);
939 dprintk("RPC: XID %08x read %Zd bytes\n",
940 ntohl(transport->tcp_xid), r);
941 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
942 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
943 transport->tcp_offset, transport->tcp_reclen);
945 if (transport->tcp_copied == req->rq_private_buf.buflen)
946 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
947 else if (transport->tcp_offset == transport->tcp_reclen) {
948 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
949 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
953 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
954 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
955 spin_unlock(&xprt->transport_lock);
956 xs_tcp_check_fraghdr(transport);
959 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
963 len = transport->tcp_reclen - transport->tcp_offset;
964 if (len > desc->count)
968 transport->tcp_offset += len;
969 dprintk("RPC: discarded %Zu bytes\n", len);
970 xs_tcp_check_fraghdr(transport);
973 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
975 struct rpc_xprt *xprt = rd_desc->arg.data;
976 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
977 struct xdr_skb_reader desc = {
983 dprintk("RPC: xs_tcp_data_recv started\n");
985 /* Read in a new fragment marker if necessary */
986 /* Can we ever really expect to get completely empty fragments? */
987 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
988 xs_tcp_read_fraghdr(xprt, &desc);
991 /* Read in the xid if necessary */
992 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
993 xs_tcp_read_xid(transport, &desc);
996 /* Read in the request data */
997 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
998 xs_tcp_read_request(xprt, &desc);
1001 /* Skip over any trailing bytes on short reads */
1002 xs_tcp_read_discard(transport, &desc);
1003 } while (desc.count);
1004 dprintk("RPC: xs_tcp_data_recv done\n");
1005 return len - desc.count;
1009 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1010 * @sk: socket with data to read
1011 * @bytes: how much data to read
1014 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1016 struct rpc_xprt *xprt;
1017 read_descriptor_t rd_desc;
1019 dprintk("RPC: xs_tcp_data_ready...\n");
1021 read_lock(&sk->sk_callback_lock);
1022 if (!(xprt = xprt_from_sock(sk)))
1027 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1028 rd_desc.arg.data = xprt;
1029 rd_desc.count = 65536;
1030 tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1032 read_unlock(&sk->sk_callback_lock);
1036 * xs_tcp_state_change - callback to handle TCP socket state changes
1037 * @sk: socket whose state has changed
1040 static void xs_tcp_state_change(struct sock *sk)
1042 struct rpc_xprt *xprt;
1044 read_lock(&sk->sk_callback_lock);
1045 if (!(xprt = xprt_from_sock(sk)))
1047 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1048 dprintk("RPC: state %x conn %d dead %d zapped %d\n",
1049 sk->sk_state, xprt_connected(xprt),
1050 sock_flag(sk, SOCK_DEAD),
1051 sock_flag(sk, SOCK_ZAPPED));
1053 switch (sk->sk_state) {
1054 case TCP_ESTABLISHED:
1055 spin_lock_bh(&xprt->transport_lock);
1056 if (!xprt_test_and_set_connected(xprt)) {
1057 struct sock_xprt *transport = container_of(xprt,
1058 struct sock_xprt, xprt);
1060 /* Reset TCP record info */
1061 transport->tcp_offset = 0;
1062 transport->tcp_reclen = 0;
1063 transport->tcp_copied = 0;
1064 transport->tcp_flags =
1065 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1067 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1068 xprt_wake_pending_tasks(xprt, 0);
1070 spin_unlock_bh(&xprt->transport_lock);
1075 case TCP_CLOSE_WAIT:
1076 /* Try to schedule an autoclose RPC calls */
1077 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
1078 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
1079 queue_work(rpciod_workqueue, &xprt->task_cleanup);
1081 xprt_disconnect(xprt);
1084 read_unlock(&sk->sk_callback_lock);
1088 * xs_udp_write_space - callback invoked when socket buffer space
1090 * @sk: socket whose state has changed
1092 * Called when more output buffer space is available for this socket.
1093 * We try not to wake our writers until they can make "significant"
1094 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1095 * with a bunch of small requests.
1097 static void xs_udp_write_space(struct sock *sk)
1099 read_lock(&sk->sk_callback_lock);
1101 /* from net/core/sock.c:sock_def_write_space */
1102 if (sock_writeable(sk)) {
1103 struct socket *sock;
1104 struct rpc_xprt *xprt;
1106 if (unlikely(!(sock = sk->sk_socket)))
1108 if (unlikely(!(xprt = xprt_from_sock(sk))))
1110 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
1113 xprt_write_space(xprt);
1117 read_unlock(&sk->sk_callback_lock);
1121 * xs_tcp_write_space - callback invoked when socket buffer space
1123 * @sk: socket whose state has changed
1125 * Called when more output buffer space is available for this socket.
1126 * We try not to wake our writers until they can make "significant"
1127 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1128 * with a bunch of small requests.
1130 static void xs_tcp_write_space(struct sock *sk)
1132 read_lock(&sk->sk_callback_lock);
1134 /* from net/core/stream.c:sk_stream_write_space */
1135 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
1136 struct socket *sock;
1137 struct rpc_xprt *xprt;
1139 if (unlikely(!(sock = sk->sk_socket)))
1141 if (unlikely(!(xprt = xprt_from_sock(sk))))
1143 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
1146 xprt_write_space(xprt);
1150 read_unlock(&sk->sk_callback_lock);
1153 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1155 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1156 struct sock *sk = transport->inet;
1158 if (transport->rcvsize) {
1159 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1160 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1162 if (transport->sndsize) {
1163 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1164 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1165 sk->sk_write_space(sk);
1170 * xs_udp_set_buffer_size - set send and receive limits
1171 * @xprt: generic transport
1172 * @sndsize: requested size of send buffer, in bytes
1173 * @rcvsize: requested size of receive buffer, in bytes
1175 * Set socket send and receive buffer size limits.
1177 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1179 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1181 transport->sndsize = 0;
1183 transport->sndsize = sndsize + 1024;
1184 transport->rcvsize = 0;
1186 transport->rcvsize = rcvsize + 1024;
1188 xs_udp_do_set_buffer_size(xprt);
1192 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1193 * @task: task that timed out
1195 * Adjust the congestion window after a retransmit timeout has occurred.
1197 static void xs_udp_timer(struct rpc_task *task)
1199 xprt_adjust_cwnd(task, -ETIMEDOUT);
1202 static unsigned short xs_get_random_port(void)
1204 unsigned short range = xprt_max_resvport - xprt_min_resvport;
1205 unsigned short rand = (unsigned short) net_random() % range;
1206 return rand + xprt_min_resvport;
1210 * xs_set_port - reset the port number in the remote endpoint address
1211 * @xprt: generic transport
1212 * @port: new port number
1215 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1217 struct sockaddr *addr = (struct sockaddr *) &xprt->addr;
1219 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1221 switch (addr->sa_family) {
1223 ((struct sockaddr_in *)addr)->sin_port = htons(port);
1226 ((struct sockaddr_in6 *)addr)->sin6_port = htons(port);
1233 static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
1235 struct sockaddr_in myaddr = {
1236 .sin_family = AF_INET,
1238 struct sockaddr_in *sa;
1240 unsigned short port = transport->port;
1242 if (!transport->xprt.resvport)
1244 sa = (struct sockaddr_in *)&transport->addr;
1245 myaddr.sin_addr = sa->sin_addr;
1247 myaddr.sin_port = htons(port);
1248 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1250 if (!transport->xprt.resvport)
1253 transport->port = port;
1256 if (port <= xprt_min_resvport)
1257 port = xprt_max_resvport;
1260 } while (err == -EADDRINUSE && port != transport->port);
1261 dprintk("RPC: %s "NIPQUAD_FMT":%u: %s (%d)\n",
1262 __FUNCTION__, NIPQUAD(myaddr.sin_addr),
1263 port, err ? "failed" : "ok", err);
1267 static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1269 struct sockaddr_in6 myaddr = {
1270 .sin6_family = AF_INET6,
1272 struct sockaddr_in6 *sa;
1274 unsigned short port = transport->port;
1276 if (!transport->xprt.resvport)
1278 sa = (struct sockaddr_in6 *)&transport->addr;
1279 myaddr.sin6_addr = sa->sin6_addr;
1281 myaddr.sin6_port = htons(port);
1282 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1284 if (!transport->xprt.resvport)
1287 transport->port = port;
1290 if (port <= xprt_min_resvport)
1291 port = xprt_max_resvport;
1294 } while (err == -EADDRINUSE && port != transport->port);
1295 dprintk("RPC: xs_bind6 "NIP6_FMT":%u: %s (%d)\n",
1296 NIP6(myaddr.sin6_addr), port, err ? "failed" : "ok", err);
1300 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1301 static struct lock_class_key xs_key[2];
1302 static struct lock_class_key xs_slock_key[2];
1304 static inline void xs_reclassify_socket(struct socket *sock)
1306 struct sock *sk = sock->sk;
1307 BUG_ON(sk->sk_lock.owner != NULL);
1308 switch (sk->sk_family) {
1310 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFS",
1311 &xs_slock_key[0], "sk_lock-AF_INET-NFS", &xs_key[0]);
1315 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFS",
1316 &xs_slock_key[1], "sk_lock-AF_INET6-NFS", &xs_key[1]);
1324 static inline void xs_reclassify_socket(struct socket *sock)
1329 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1331 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1333 if (!transport->inet) {
1334 struct sock *sk = sock->sk;
1336 write_lock_bh(&sk->sk_callback_lock);
1338 sk->sk_user_data = xprt;
1339 transport->old_data_ready = sk->sk_data_ready;
1340 transport->old_state_change = sk->sk_state_change;
1341 transport->old_write_space = sk->sk_write_space;
1342 sk->sk_data_ready = xs_udp_data_ready;
1343 sk->sk_write_space = xs_udp_write_space;
1344 sk->sk_no_check = UDP_CSUM_NORCV;
1345 sk->sk_allocation = GFP_ATOMIC;
1347 xprt_set_connected(xprt);
1349 /* Reset to new socket */
1350 transport->sock = sock;
1351 transport->inet = sk;
1353 write_unlock_bh(&sk->sk_callback_lock);
1355 xs_udp_do_set_buffer_size(xprt);
1359 * xs_udp_connect_worker4 - set up a UDP socket
1360 * @work: RPC transport to connect
1362 * Invoked by a work queue tasklet.
1364 static void xs_udp_connect_worker4(struct work_struct *work)
1366 struct sock_xprt *transport =
1367 container_of(work, struct sock_xprt, connect_worker.work);
1368 struct rpc_xprt *xprt = &transport->xprt;
1369 struct socket *sock = transport->sock;
1370 int err, status = -EIO;
1372 if (xprt->shutdown || !xprt_bound(xprt))
1375 /* Start by resetting any existing state */
1378 if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1379 dprintk("RPC: can't create UDP transport socket (%d).\n", -err);
1382 xs_reclassify_socket(sock);
1384 if (xs_bind4(transport, sock)) {
1389 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1390 xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1392 xs_udp_finish_connecting(xprt, sock);
1395 xprt_wake_pending_tasks(xprt, status);
1396 xprt_clear_connecting(xprt);
1400 * xs_udp_connect_worker6 - set up a UDP socket
1401 * @work: RPC transport to connect
1403 * Invoked by a work queue tasklet.
1405 static void xs_udp_connect_worker6(struct work_struct *work)
1407 struct sock_xprt *transport =
1408 container_of(work, struct sock_xprt, connect_worker.work);
1409 struct rpc_xprt *xprt = &transport->xprt;
1410 struct socket *sock = transport->sock;
1411 int err, status = -EIO;
1413 if (xprt->shutdown || !xprt_bound(xprt))
1416 /* Start by resetting any existing state */
1419 if ((err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1420 dprintk("RPC: can't create UDP transport socket (%d).\n", -err);
1423 xs_reclassify_socket(sock);
1425 if (xs_bind6(transport, sock) < 0) {
1430 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1431 xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1433 xs_udp_finish_connecting(xprt, sock);
1436 xprt_wake_pending_tasks(xprt, status);
1437 xprt_clear_connecting(xprt);
1441 * We need to preserve the port number so the reply cache on the server can
1442 * find our cached RPC replies when we get around to reconnecting.
1444 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1447 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1448 struct sockaddr any;
1450 dprintk("RPC: disconnecting xprt %p to reuse port\n", xprt);
1453 * Disconnect the transport socket by doing a connect operation
1454 * with AF_UNSPEC. This should return immediately...
1456 memset(&any, 0, sizeof(any));
1457 any.sa_family = AF_UNSPEC;
1458 result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1460 dprintk("RPC: AF_UNSPEC connect return code %d\n",
1464 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1466 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1468 if (!transport->inet) {
1469 struct sock *sk = sock->sk;
1471 write_lock_bh(&sk->sk_callback_lock);
1473 sk->sk_user_data = xprt;
1474 transport->old_data_ready = sk->sk_data_ready;
1475 transport->old_state_change = sk->sk_state_change;
1476 transport->old_write_space = sk->sk_write_space;
1477 sk->sk_data_ready = xs_tcp_data_ready;
1478 sk->sk_state_change = xs_tcp_state_change;
1479 sk->sk_write_space = xs_tcp_write_space;
1480 sk->sk_allocation = GFP_ATOMIC;
1482 /* socket options */
1483 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1484 sock_reset_flag(sk, SOCK_LINGER);
1485 tcp_sk(sk)->linger2 = 0;
1486 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1488 xprt_clear_connected(xprt);
1490 /* Reset to new socket */
1491 transport->sock = sock;
1492 transport->inet = sk;
1494 write_unlock_bh(&sk->sk_callback_lock);
1497 /* Tell the socket layer to start connecting... */
1498 xprt->stat.connect_count++;
1499 xprt->stat.connect_start = jiffies;
1500 return kernel_connect(sock, (struct sockaddr *) &xprt->addr,
1501 xprt->addrlen, O_NONBLOCK);
1505 * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1506 * @work: RPC transport to connect
1508 * Invoked by a work queue tasklet.
1510 static void xs_tcp_connect_worker4(struct work_struct *work)
1512 struct sock_xprt *transport =
1513 container_of(work, struct sock_xprt, connect_worker.work);
1514 struct rpc_xprt *xprt = &transport->xprt;
1515 struct socket *sock = transport->sock;
1516 int err, status = -EIO;
1518 if (xprt->shutdown || !xprt_bound(xprt))
1522 /* start from scratch */
1523 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1524 dprintk("RPC: can't create TCP transport socket (%d).\n", -err);
1527 xs_reclassify_socket(sock);
1529 if (xs_bind4(transport, sock) < 0) {
1534 /* "close" the socket, preserving the local port */
1535 xs_tcp_reuse_connection(xprt);
1537 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1538 xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1540 status = xs_tcp_finish_connecting(xprt, sock);
1541 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
1542 xprt, -status, xprt_connected(xprt),
1543 sock->sk->sk_state);
1551 /* retry with existing socket, after a delay */
1554 /* get rid of existing socket, and retry */
1560 xprt_wake_pending_tasks(xprt, status);
1562 xprt_clear_connecting(xprt);
1566 * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
1567 * @work: RPC transport to connect
1569 * Invoked by a work queue tasklet.
1571 static void xs_tcp_connect_worker6(struct work_struct *work)
1573 struct sock_xprt *transport =
1574 container_of(work, struct sock_xprt, connect_worker.work);
1575 struct rpc_xprt *xprt = &transport->xprt;
1576 struct socket *sock = transport->sock;
1577 int err, status = -EIO;
1579 if (xprt->shutdown || !xprt_bound(xprt))
1583 /* start from scratch */
1584 if ((err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1585 dprintk("RPC: can't create TCP transport socket (%d).\n", -err);
1588 xs_reclassify_socket(sock);
1590 if (xs_bind6(transport, sock) < 0) {
1595 /* "close" the socket, preserving the local port */
1596 xs_tcp_reuse_connection(xprt);
1598 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1599 xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1601 status = xs_tcp_finish_connecting(xprt, sock);
1602 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
1603 xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1611 /* retry with existing socket, after a delay */
1614 /* get rid of existing socket, and retry */
1620 xprt_wake_pending_tasks(xprt, status);
1622 xprt_clear_connecting(xprt);
1626 * xs_connect - connect a socket to a remote endpoint
1627 * @task: address of RPC task that manages state of connect request
1629 * TCP: If the remote end dropped the connection, delay reconnecting.
1631 * UDP socket connects are synchronous, but we use a work queue anyway
1632 * to guarantee that even unprivileged user processes can set up a
1633 * socket on a privileged port.
1635 * If a UDP socket connect fails, the delay behavior here prevents
1636 * retry floods (hard mounts).
1638 static void xs_connect(struct rpc_task *task)
1640 struct rpc_xprt *xprt = task->tk_xprt;
1641 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1643 if (xprt_test_and_set_connecting(xprt))
1646 if (transport->sock != NULL) {
1647 dprintk("RPC: xs_connect delayed xprt %p for %lu "
1649 xprt, xprt->reestablish_timeout / HZ);
1650 queue_delayed_work(rpciod_workqueue,
1651 &transport->connect_worker,
1652 xprt->reestablish_timeout);
1653 xprt->reestablish_timeout <<= 1;
1654 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1655 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1657 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
1658 queue_delayed_work(rpciod_workqueue,
1659 &transport->connect_worker, 0);
1664 * xs_udp_print_stats - display UDP socket-specifc stats
1665 * @xprt: rpc_xprt struct containing statistics
1669 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1671 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1673 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1675 xprt->stat.bind_count,
1678 xprt->stat.bad_xids,
1680 xprt->stat.bklog_u);
1684 * xs_tcp_print_stats - display TCP socket-specifc stats
1685 * @xprt: rpc_xprt struct containing statistics
1689 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1691 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1694 if (xprt_connected(xprt))
1695 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1697 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1699 xprt->stat.bind_count,
1700 xprt->stat.connect_count,
1701 xprt->stat.connect_time,
1705 xprt->stat.bad_xids,
1707 xprt->stat.bklog_u);
1710 static struct rpc_xprt_ops xs_udp_ops = {
1711 .set_buffer_size = xs_udp_set_buffer_size,
1712 .reserve_xprt = xprt_reserve_xprt_cong,
1713 .release_xprt = xprt_release_xprt_cong,
1714 .rpcbind = rpcb_getport_async,
1715 .set_port = xs_set_port,
1716 .connect = xs_connect,
1717 .buf_alloc = rpc_malloc,
1718 .buf_free = rpc_free,
1719 .send_request = xs_udp_send_request,
1720 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
1721 .timer = xs_udp_timer,
1722 .release_request = xprt_release_rqst_cong,
1724 .destroy = xs_destroy,
1725 .print_stats = xs_udp_print_stats,
1728 static struct rpc_xprt_ops xs_tcp_ops = {
1729 .reserve_xprt = xprt_reserve_xprt,
1730 .release_xprt = xs_tcp_release_xprt,
1731 .rpcbind = rpcb_getport_async,
1732 .set_port = xs_set_port,
1733 .connect = xs_connect,
1734 .buf_alloc = rpc_malloc,
1735 .buf_free = rpc_free,
1736 .send_request = xs_tcp_send_request,
1737 .set_retrans_timeout = xprt_set_retrans_timeout_def,
1739 .destroy = xs_destroy,
1740 .print_stats = xs_tcp_print_stats,
1743 static struct rpc_xprt *xs_setup_xprt(struct rpc_xprtsock_create *args, unsigned int slot_table_size)
1745 struct rpc_xprt *xprt;
1746 struct sock_xprt *new;
1748 if (args->addrlen > sizeof(xprt->addr)) {
1749 dprintk("RPC: xs_setup_xprt: address too large\n");
1750 return ERR_PTR(-EBADF);
1753 new = kzalloc(sizeof(*new), GFP_KERNEL);
1755 dprintk("RPC: xs_setup_xprt: couldn't allocate "
1757 return ERR_PTR(-ENOMEM);
1761 xprt->max_reqs = slot_table_size;
1762 xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1763 if (xprt->slot == NULL) {
1765 dprintk("RPC: xs_setup_xprt: couldn't allocate slot "
1767 return ERR_PTR(-ENOMEM);
1770 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
1771 xprt->addrlen = args->addrlen;
1773 memcpy(&new->addr, args->srcaddr, args->addrlen);
1774 new->port = xs_get_random_port();
1780 * xs_setup_udp - Set up transport to use a UDP socket
1781 * @args: rpc transport creation arguments
1784 struct rpc_xprt *xs_setup_udp(struct rpc_xprtsock_create *args)
1786 struct sockaddr *addr = args->dstaddr;
1787 struct rpc_xprt *xprt;
1788 struct sock_xprt *transport;
1790 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
1793 transport = container_of(xprt, struct sock_xprt, xprt);
1795 xprt->prot = IPPROTO_UDP;
1797 /* XXX: header size can vary due to auth type, IPv6, etc. */
1798 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1800 xprt->bind_timeout = XS_BIND_TO;
1801 xprt->connect_timeout = XS_UDP_CONN_TO;
1802 xprt->reestablish_timeout = XS_UDP_REEST_TO;
1803 xprt->idle_timeout = XS_IDLE_DISC_TO;
1805 xprt->ops = &xs_udp_ops;
1808 xprt->timeout = *args->timeout;
1810 xprt_set_timeout(&xprt->timeout, 5, 5 * HZ);
1812 switch (addr->sa_family) {
1814 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1815 xprt_set_bound(xprt);
1817 INIT_DELAYED_WORK(&transport->connect_worker,
1818 xs_udp_connect_worker4);
1819 xs_format_ipv4_peer_addresses(xprt);
1822 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
1823 xprt_set_bound(xprt);
1825 INIT_DELAYED_WORK(&transport->connect_worker,
1826 xs_udp_connect_worker6);
1827 xs_format_ipv6_peer_addresses(xprt);
1831 return ERR_PTR(-EAFNOSUPPORT);
1834 dprintk("RPC: set up transport to address %s\n",
1835 xprt->address_strings[RPC_DISPLAY_ALL]);
1841 * xs_setup_tcp - Set up transport to use a TCP socket
1842 * @args: rpc transport creation arguments
1845 struct rpc_xprt *xs_setup_tcp(struct rpc_xprtsock_create *args)
1847 struct sockaddr *addr = args->dstaddr;
1848 struct rpc_xprt *xprt;
1849 struct sock_xprt *transport;
1851 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
1854 transport = container_of(xprt, struct sock_xprt, xprt);
1856 xprt->prot = IPPROTO_TCP;
1857 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
1858 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
1860 xprt->bind_timeout = XS_BIND_TO;
1861 xprt->connect_timeout = XS_TCP_CONN_TO;
1862 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1863 xprt->idle_timeout = XS_IDLE_DISC_TO;
1865 xprt->ops = &xs_tcp_ops;
1868 xprt->timeout = *args->timeout;
1870 xprt_set_timeout(&xprt->timeout, 2, 60 * HZ);
1872 switch (addr->sa_family) {
1874 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1875 xprt_set_bound(xprt);
1877 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker4);
1878 xs_format_ipv4_peer_addresses(xprt);
1881 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
1882 xprt_set_bound(xprt);
1884 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker6);
1885 xs_format_ipv6_peer_addresses(xprt);
1889 return ERR_PTR(-EAFNOSUPPORT);
1892 dprintk("RPC: set up transport to address %s\n",
1893 xprt->address_strings[RPC_DISPLAY_ALL]);
1899 * init_socket_xprt - set up xprtsock's sysctls
1902 int init_socket_xprt(void)
1905 if (!sunrpc_table_header)
1906 sunrpc_table_header = register_sysctl_table(sunrpc_table);
1913 * cleanup_socket_xprt - remove xprtsock's sysctls
1916 void cleanup_socket_xprt(void)
1919 if (sunrpc_table_header) {
1920 unregister_sysctl_table(sunrpc_table_header);
1921 sunrpc_table_header = NULL;