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 inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
268 return (struct sockaddr *) &xprt->addr;
271 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
273 return (struct sockaddr_in *) &xprt->addr;
276 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
278 return (struct sockaddr_in6 *) &xprt->addr;
281 static void xs_format_ipv4_peer_addresses(struct rpc_xprt *xprt)
283 struct sockaddr_in *addr = xs_addr_in(xprt);
286 buf = kzalloc(20, GFP_KERNEL);
288 snprintf(buf, 20, NIPQUAD_FMT,
289 NIPQUAD(addr->sin_addr.s_addr));
291 xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
293 buf = kzalloc(8, GFP_KERNEL);
295 snprintf(buf, 8, "%u",
296 ntohs(addr->sin_port));
298 xprt->address_strings[RPC_DISPLAY_PORT] = buf;
300 buf = kzalloc(8, GFP_KERNEL);
302 if (xprt->prot == IPPROTO_UDP)
303 snprintf(buf, 8, "udp");
305 snprintf(buf, 8, "tcp");
307 xprt->address_strings[RPC_DISPLAY_PROTO] = buf;
309 buf = kzalloc(48, GFP_KERNEL);
311 snprintf(buf, 48, "addr="NIPQUAD_FMT" port=%u proto=%s",
312 NIPQUAD(addr->sin_addr.s_addr),
313 ntohs(addr->sin_port),
314 xprt->prot == IPPROTO_UDP ? "udp" : "tcp");
316 xprt->address_strings[RPC_DISPLAY_ALL] = buf;
318 buf = kzalloc(10, GFP_KERNEL);
320 snprintf(buf, 10, "%02x%02x%02x%02x",
321 NIPQUAD(addr->sin_addr.s_addr));
323 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
325 buf = kzalloc(8, GFP_KERNEL);
327 snprintf(buf, 8, "%4hx",
328 ntohs(addr->sin_port));
330 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
332 buf = kzalloc(30, GFP_KERNEL);
334 snprintf(buf, 30, NIPQUAD_FMT".%u.%u",
335 NIPQUAD(addr->sin_addr.s_addr),
336 ntohs(addr->sin_port) >> 8,
337 ntohs(addr->sin_port) & 0xff);
339 xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
342 static void xs_format_ipv6_peer_addresses(struct rpc_xprt *xprt)
344 struct sockaddr_in6 *addr = xs_addr_in6(xprt);
347 buf = kzalloc(40, GFP_KERNEL);
349 snprintf(buf, 40, NIP6_FMT,
350 NIP6(addr->sin6_addr));
352 xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
354 buf = kzalloc(8, GFP_KERNEL);
356 snprintf(buf, 8, "%u",
357 ntohs(addr->sin6_port));
359 xprt->address_strings[RPC_DISPLAY_PORT] = buf;
361 buf = kzalloc(8, GFP_KERNEL);
363 if (xprt->prot == IPPROTO_UDP)
364 snprintf(buf, 8, "udp");
366 snprintf(buf, 8, "tcp");
368 xprt->address_strings[RPC_DISPLAY_PROTO] = buf;
370 buf = kzalloc(64, GFP_KERNEL);
372 snprintf(buf, 64, "addr="NIP6_FMT" port=%u proto=%s",
373 NIP6(addr->sin6_addr),
374 ntohs(addr->sin6_port),
375 xprt->prot == IPPROTO_UDP ? "udp" : "tcp");
377 xprt->address_strings[RPC_DISPLAY_ALL] = buf;
379 buf = kzalloc(36, GFP_KERNEL);
381 snprintf(buf, 36, NIP6_SEQFMT,
382 NIP6(addr->sin6_addr));
384 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
386 buf = kzalloc(8, GFP_KERNEL);
388 snprintf(buf, 8, "%4hx",
389 ntohs(addr->sin6_port));
391 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
393 buf = kzalloc(50, GFP_KERNEL);
395 snprintf(buf, 50, NIP6_FMT".%u.%u",
396 NIP6(addr->sin6_addr),
397 ntohs(addr->sin6_port) >> 8,
398 ntohs(addr->sin6_port) & 0xff);
400 xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
403 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
407 for (i = 0; i < RPC_DISPLAY_MAX; i++)
408 kfree(xprt->address_strings[i]);
411 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
413 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
415 struct msghdr msg = {
417 .msg_namelen = addrlen,
418 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
421 .iov_base = vec->iov_base + base,
422 .iov_len = vec->iov_len - base,
425 if (iov.iov_len != 0)
426 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
427 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
430 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
433 unsigned int remainder;
436 remainder = xdr->page_len - base;
437 base += xdr->page_base;
438 ppage = xdr->pages + (base >> PAGE_SHIFT);
441 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
442 int flags = XS_SENDMSG_FLAGS;
445 if (remainder != 0 || more)
447 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
448 if (remainder == 0 || err != len)
462 * xs_sendpages - write pages directly to a socket
463 * @sock: socket to send on
464 * @addr: UDP only -- address of destination
465 * @addrlen: UDP only -- length of destination address
466 * @xdr: buffer containing this request
467 * @base: starting position in the buffer
470 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
472 unsigned int remainder = xdr->len - base;
478 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
484 if (base < xdr->head[0].iov_len || addr != NULL) {
485 unsigned int len = xdr->head[0].iov_len - base;
487 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
488 if (remainder == 0 || err != len)
493 base -= xdr->head[0].iov_len;
495 if (base < xdr->page_len) {
496 unsigned int len = xdr->page_len - base;
498 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
499 if (remainder == 0 || err != len)
504 base -= xdr->page_len;
506 if (base >= xdr->tail[0].iov_len)
508 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
518 * xs_nospace - place task on wait queue if transmit was incomplete
519 * @task: task to put to sleep
522 static void xs_nospace(struct rpc_task *task)
524 struct rpc_rqst *req = task->tk_rqstp;
525 struct rpc_xprt *xprt = req->rq_xprt;
526 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
528 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
529 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
532 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
533 /* Protect against races with write_space */
534 spin_lock_bh(&xprt->transport_lock);
536 /* Don't race with disconnect */
537 if (!xprt_connected(xprt))
538 task->tk_status = -ENOTCONN;
539 else if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
540 xprt_wait_for_buffer_space(task);
542 spin_unlock_bh(&xprt->transport_lock);
544 /* Keep holding the socket if it is blocked */
545 rpc_delay(task, HZ>>4);
549 * xs_udp_send_request - write an RPC request to a UDP socket
550 * @task: address of RPC task that manages the state of an RPC request
553 * 0: The request has been sent
554 * EAGAIN: The socket was blocked, please call again later to
555 * complete the request
556 * ENOTCONN: Caller needs to invoke connect logic then call again
557 * other: Some other error occured, the request was not sent
559 static int xs_udp_send_request(struct rpc_task *task)
561 struct rpc_rqst *req = task->tk_rqstp;
562 struct rpc_xprt *xprt = req->rq_xprt;
563 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
564 struct xdr_buf *xdr = &req->rq_snd_buf;
567 xs_pktdump("packet data:",
568 req->rq_svec->iov_base,
569 req->rq_svec->iov_len);
571 req->rq_xtime = jiffies;
572 status = xs_sendpages(transport->sock,
577 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
578 xdr->len - req->rq_bytes_sent, status);
580 if (likely(status >= (int) req->rq_slen))
583 /* Still some bytes left; set up for a retry later. */
591 /* When the server has died, an ICMP port unreachable message
592 * prompts ECONNREFUSED. */
598 dprintk("RPC: sendmsg returned unrecognized error %d\n",
606 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
608 u32 reclen = buf->len - sizeof(rpc_fraghdr);
609 rpc_fraghdr *base = buf->head[0].iov_base;
610 *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
614 * xs_tcp_send_request - write an RPC request to a TCP socket
615 * @task: address of RPC task that manages the state of an RPC request
618 * 0: The request has been sent
619 * EAGAIN: The socket was blocked, please call again later to
620 * complete the request
621 * ENOTCONN: Caller needs to invoke connect logic then call again
622 * other: Some other error occured, the request was not sent
624 * XXX: In the case of soft timeouts, should we eventually give up
625 * if sendmsg is not able to make progress?
627 static int xs_tcp_send_request(struct rpc_task *task)
629 struct rpc_rqst *req = task->tk_rqstp;
630 struct rpc_xprt *xprt = req->rq_xprt;
631 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
632 struct xdr_buf *xdr = &req->rq_snd_buf;
634 unsigned int retry = 0;
636 xs_encode_tcp_record_marker(&req->rq_snd_buf);
638 xs_pktdump("packet data:",
639 req->rq_svec->iov_base,
640 req->rq_svec->iov_len);
642 /* Continue transmitting the packet/record. We must be careful
643 * to cope with writespace callbacks arriving _after_ we have
644 * called sendmsg(). */
646 req->rq_xtime = jiffies;
647 status = xs_sendpages(transport->sock,
648 NULL, 0, xdr, req->rq_bytes_sent);
650 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
651 xdr->len - req->rq_bytes_sent, status);
653 if (unlikely(status < 0))
656 /* If we've sent the entire packet, immediately
657 * reset the count of bytes sent. */
658 req->rq_bytes_sent += status;
659 task->tk_bytes_sent += status;
660 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
661 req->rq_bytes_sent = 0;
666 if (retry++ > XS_SENDMSG_RETRY)
681 dprintk("RPC: sendmsg returned unrecognized error %d\n",
683 xprt_disconnect(xprt);
691 * xs_tcp_release_xprt - clean up after a tcp transmission
695 * This cleans up if an error causes us to abort the transmission of a request.
696 * In this case, the socket may need to be reset in order to avoid confusing
699 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
701 struct rpc_rqst *req;
703 if (task != xprt->snd_task)
707 req = task->tk_rqstp;
708 if (req->rq_bytes_sent == 0)
710 if (req->rq_bytes_sent == req->rq_snd_buf.len)
712 set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
714 xprt_release_xprt(xprt, task);
718 * xs_close - close a socket
721 * This is used when all requests are complete; ie, no DRC state remains
722 * on the server we want to save.
724 static void xs_close(struct rpc_xprt *xprt)
726 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
727 struct socket *sock = transport->sock;
728 struct sock *sk = transport->inet;
731 goto clear_close_wait;
733 dprintk("RPC: xs_close xprt %p\n", xprt);
735 write_lock_bh(&sk->sk_callback_lock);
736 transport->inet = NULL;
737 transport->sock = NULL;
739 sk->sk_user_data = NULL;
740 sk->sk_data_ready = transport->old_data_ready;
741 sk->sk_state_change = transport->old_state_change;
742 sk->sk_write_space = transport->old_write_space;
743 write_unlock_bh(&sk->sk_callback_lock);
749 smp_mb__before_clear_bit();
750 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
751 smp_mb__after_clear_bit();
755 * xs_destroy - prepare to shutdown a transport
756 * @xprt: doomed transport
759 static void xs_destroy(struct rpc_xprt *xprt)
761 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
763 dprintk("RPC: xs_destroy xprt %p\n", xprt);
765 cancel_rearming_delayed_work(&transport->connect_worker);
767 xprt_disconnect(xprt);
769 xs_free_peer_addresses(xprt);
774 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
776 return (struct rpc_xprt *) sk->sk_user_data;
780 * xs_udp_data_ready - "data ready" callback for UDP sockets
781 * @sk: socket with data to read
782 * @len: how much data to read
785 static void xs_udp_data_ready(struct sock *sk, int len)
787 struct rpc_task *task;
788 struct rpc_xprt *xprt;
789 struct rpc_rqst *rovr;
791 int err, repsize, copied;
795 read_lock(&sk->sk_callback_lock);
796 dprintk("RPC: xs_udp_data_ready...\n");
797 if (!(xprt = xprt_from_sock(sk)))
800 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
806 repsize = skb->len - sizeof(struct udphdr);
808 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
812 /* Copy the XID from the skb... */
813 xp = skb_header_pointer(skb, sizeof(struct udphdr),
814 sizeof(_xid), &_xid);
818 /* Look up and lock the request corresponding to the given XID */
819 spin_lock(&xprt->transport_lock);
820 rovr = xprt_lookup_rqst(xprt, *xp);
823 task = rovr->rq_task;
825 if ((copied = rovr->rq_private_buf.buflen) > repsize)
828 /* Suck it into the iovec, verify checksum if not done by hw. */
829 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb))
832 /* Something worked... */
833 dst_confirm(skb->dst);
835 xprt_adjust_cwnd(task, copied);
836 xprt_update_rtt(task);
837 xprt_complete_rqst(task, copied);
840 spin_unlock(&xprt->transport_lock);
842 skb_free_datagram(sk, skb);
844 read_unlock(&sk->sk_callback_lock);
847 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
849 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
853 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
854 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
855 used = xdr_skb_read_bits(desc, p, len);
856 transport->tcp_offset += used;
860 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
861 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
862 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
864 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
865 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
867 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
868 transport->tcp_offset = 0;
870 /* Sanity check of the record length */
871 if (unlikely(transport->tcp_reclen < 4)) {
872 dprintk("RPC: invalid TCP record fragment length\n");
873 xprt_disconnect(xprt);
876 dprintk("RPC: reading TCP record fragment of length %d\n",
877 transport->tcp_reclen);
880 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
882 if (transport->tcp_offset == transport->tcp_reclen) {
883 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
884 transport->tcp_offset = 0;
885 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
886 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
887 transport->tcp_flags |= TCP_RCV_COPY_XID;
888 transport->tcp_copied = 0;
893 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
898 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
899 dprintk("RPC: reading XID (%Zu bytes)\n", len);
900 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
901 used = xdr_skb_read_bits(desc, p, len);
902 transport->tcp_offset += used;
905 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
906 transport->tcp_flags |= TCP_RCV_COPY_DATA;
907 transport->tcp_copied = 4;
908 dprintk("RPC: reading reply for XID %08x\n",
909 ntohl(transport->tcp_xid));
910 xs_tcp_check_fraghdr(transport);
913 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
915 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
916 struct rpc_rqst *req;
917 struct xdr_buf *rcvbuf;
921 /* Find and lock the request corresponding to this xid */
922 spin_lock(&xprt->transport_lock);
923 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
925 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
926 dprintk("RPC: XID %08x request not found!\n",
927 ntohl(transport->tcp_xid));
928 spin_unlock(&xprt->transport_lock);
932 rcvbuf = &req->rq_private_buf;
934 if (len > transport->tcp_reclen - transport->tcp_offset) {
935 struct xdr_skb_reader my_desc;
937 len = transport->tcp_reclen - transport->tcp_offset;
938 memcpy(&my_desc, desc, sizeof(my_desc));
940 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
941 &my_desc, xdr_skb_read_bits);
945 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
946 desc, xdr_skb_read_bits);
949 transport->tcp_copied += r;
950 transport->tcp_offset += r;
953 /* Error when copying to the receive buffer,
954 * usually because we weren't able to allocate
955 * additional buffer pages. All we can do now
956 * is turn off TCP_RCV_COPY_DATA, so the request
957 * will not receive any additional updates,
959 * Any remaining data from this record will
962 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
963 dprintk("RPC: XID %08x truncated request\n",
964 ntohl(transport->tcp_xid));
965 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
966 "tcp_offset = %u, tcp_reclen = %u\n",
967 xprt, transport->tcp_copied,
968 transport->tcp_offset, transport->tcp_reclen);
972 dprintk("RPC: XID %08x read %Zd bytes\n",
973 ntohl(transport->tcp_xid), r);
974 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
975 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
976 transport->tcp_offset, transport->tcp_reclen);
978 if (transport->tcp_copied == req->rq_private_buf.buflen)
979 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
980 else if (transport->tcp_offset == transport->tcp_reclen) {
981 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
982 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
986 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
987 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
988 spin_unlock(&xprt->transport_lock);
989 xs_tcp_check_fraghdr(transport);
992 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
996 len = transport->tcp_reclen - transport->tcp_offset;
997 if (len > desc->count)
1000 desc->offset += len;
1001 transport->tcp_offset += len;
1002 dprintk("RPC: discarded %Zu bytes\n", len);
1003 xs_tcp_check_fraghdr(transport);
1006 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1008 struct rpc_xprt *xprt = rd_desc->arg.data;
1009 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1010 struct xdr_skb_reader desc = {
1016 dprintk("RPC: xs_tcp_data_recv started\n");
1018 /* Read in a new fragment marker if necessary */
1019 /* Can we ever really expect to get completely empty fragments? */
1020 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1021 xs_tcp_read_fraghdr(xprt, &desc);
1024 /* Read in the xid if necessary */
1025 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1026 xs_tcp_read_xid(transport, &desc);
1029 /* Read in the request data */
1030 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1031 xs_tcp_read_request(xprt, &desc);
1034 /* Skip over any trailing bytes on short reads */
1035 xs_tcp_read_discard(transport, &desc);
1036 } while (desc.count);
1037 dprintk("RPC: xs_tcp_data_recv done\n");
1038 return len - desc.count;
1042 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1043 * @sk: socket with data to read
1044 * @bytes: how much data to read
1047 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1049 struct rpc_xprt *xprt;
1050 read_descriptor_t rd_desc;
1052 dprintk("RPC: xs_tcp_data_ready...\n");
1054 read_lock(&sk->sk_callback_lock);
1055 if (!(xprt = xprt_from_sock(sk)))
1060 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1061 rd_desc.arg.data = xprt;
1062 rd_desc.count = 65536;
1063 tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1065 read_unlock(&sk->sk_callback_lock);
1069 * xs_tcp_state_change - callback to handle TCP socket state changes
1070 * @sk: socket whose state has changed
1073 static void xs_tcp_state_change(struct sock *sk)
1075 struct rpc_xprt *xprt;
1077 read_lock(&sk->sk_callback_lock);
1078 if (!(xprt = xprt_from_sock(sk)))
1080 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1081 dprintk("RPC: state %x conn %d dead %d zapped %d\n",
1082 sk->sk_state, xprt_connected(xprt),
1083 sock_flag(sk, SOCK_DEAD),
1084 sock_flag(sk, SOCK_ZAPPED));
1086 switch (sk->sk_state) {
1087 case TCP_ESTABLISHED:
1088 spin_lock_bh(&xprt->transport_lock);
1089 if (!xprt_test_and_set_connected(xprt)) {
1090 struct sock_xprt *transport = container_of(xprt,
1091 struct sock_xprt, xprt);
1093 /* Reset TCP record info */
1094 transport->tcp_offset = 0;
1095 transport->tcp_reclen = 0;
1096 transport->tcp_copied = 0;
1097 transport->tcp_flags =
1098 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1100 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1101 xprt_wake_pending_tasks(xprt, 0);
1103 spin_unlock_bh(&xprt->transport_lock);
1108 case TCP_CLOSE_WAIT:
1109 /* Try to schedule an autoclose RPC calls */
1110 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
1111 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
1112 queue_work(rpciod_workqueue, &xprt->task_cleanup);
1114 xprt_disconnect(xprt);
1117 read_unlock(&sk->sk_callback_lock);
1121 * xs_udp_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_udp_write_space(struct sock *sk)
1132 read_lock(&sk->sk_callback_lock);
1134 /* from net/core/sock.c:sock_def_write_space */
1135 if (sock_writeable(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);
1154 * xs_tcp_write_space - callback invoked when socket buffer space
1156 * @sk: socket whose state has changed
1158 * Called when more output buffer space is available for this socket.
1159 * We try not to wake our writers until they can make "significant"
1160 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1161 * with a bunch of small requests.
1163 static void xs_tcp_write_space(struct sock *sk)
1165 read_lock(&sk->sk_callback_lock);
1167 /* from net/core/stream.c:sk_stream_write_space */
1168 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
1169 struct socket *sock;
1170 struct rpc_xprt *xprt;
1172 if (unlikely(!(sock = sk->sk_socket)))
1174 if (unlikely(!(xprt = xprt_from_sock(sk))))
1176 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
1179 xprt_write_space(xprt);
1183 read_unlock(&sk->sk_callback_lock);
1186 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1188 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1189 struct sock *sk = transport->inet;
1191 if (transport->rcvsize) {
1192 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1193 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1195 if (transport->sndsize) {
1196 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1197 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1198 sk->sk_write_space(sk);
1203 * xs_udp_set_buffer_size - set send and receive limits
1204 * @xprt: generic transport
1205 * @sndsize: requested size of send buffer, in bytes
1206 * @rcvsize: requested size of receive buffer, in bytes
1208 * Set socket send and receive buffer size limits.
1210 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1212 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1214 transport->sndsize = 0;
1216 transport->sndsize = sndsize + 1024;
1217 transport->rcvsize = 0;
1219 transport->rcvsize = rcvsize + 1024;
1221 xs_udp_do_set_buffer_size(xprt);
1225 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1226 * @task: task that timed out
1228 * Adjust the congestion window after a retransmit timeout has occurred.
1230 static void xs_udp_timer(struct rpc_task *task)
1232 xprt_adjust_cwnd(task, -ETIMEDOUT);
1235 static unsigned short xs_get_random_port(void)
1237 unsigned short range = xprt_max_resvport - xprt_min_resvport;
1238 unsigned short rand = (unsigned short) net_random() % range;
1239 return rand + xprt_min_resvport;
1243 * xs_set_port - reset the port number in the remote endpoint address
1244 * @xprt: generic transport
1245 * @port: new port number
1248 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1250 struct sockaddr *addr = xs_addr(xprt);
1252 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1254 switch (addr->sa_family) {
1256 ((struct sockaddr_in *)addr)->sin_port = htons(port);
1259 ((struct sockaddr_in6 *)addr)->sin6_port = htons(port);
1266 static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
1268 struct sockaddr_in myaddr = {
1269 .sin_family = AF_INET,
1271 struct sockaddr_in *sa;
1273 unsigned short port = transport->port;
1275 if (!transport->xprt.resvport)
1277 sa = (struct sockaddr_in *)&transport->addr;
1278 myaddr.sin_addr = sa->sin_addr;
1280 myaddr.sin_port = htons(port);
1281 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1283 if (!transport->xprt.resvport)
1286 transport->port = port;
1289 if (port <= xprt_min_resvport)
1290 port = xprt_max_resvport;
1293 } while (err == -EADDRINUSE && port != transport->port);
1294 dprintk("RPC: %s "NIPQUAD_FMT":%u: %s (%d)\n",
1295 __FUNCTION__, NIPQUAD(myaddr.sin_addr),
1296 port, err ? "failed" : "ok", err);
1300 static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1302 struct sockaddr_in6 myaddr = {
1303 .sin6_family = AF_INET6,
1305 struct sockaddr_in6 *sa;
1307 unsigned short port = transport->port;
1309 if (!transport->xprt.resvport)
1311 sa = (struct sockaddr_in6 *)&transport->addr;
1312 myaddr.sin6_addr = sa->sin6_addr;
1314 myaddr.sin6_port = htons(port);
1315 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1317 if (!transport->xprt.resvport)
1320 transport->port = port;
1323 if (port <= xprt_min_resvport)
1324 port = xprt_max_resvport;
1327 } while (err == -EADDRINUSE && port != transport->port);
1328 dprintk("RPC: xs_bind6 "NIP6_FMT":%u: %s (%d)\n",
1329 NIP6(myaddr.sin6_addr), port, err ? "failed" : "ok", err);
1333 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1334 static struct lock_class_key xs_key[2];
1335 static struct lock_class_key xs_slock_key[2];
1337 static inline void xs_reclassify_socket4(struct socket *sock)
1339 struct sock *sk = sock->sk;
1341 BUG_ON(sk->sk_lock.owner != NULL);
1342 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1343 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1346 static inline void xs_reclassify_socket6(struct socket *sock)
1348 struct sock *sk = sock->sk;
1350 BUG_ON(sk->sk_lock.owner != NULL);
1351 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1352 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1355 static inline void xs_reclassify_socket4(struct socket *sock)
1359 static inline void xs_reclassify_socket6(struct socket *sock)
1364 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1366 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1368 if (!transport->inet) {
1369 struct sock *sk = sock->sk;
1371 write_lock_bh(&sk->sk_callback_lock);
1373 sk->sk_user_data = xprt;
1374 transport->old_data_ready = sk->sk_data_ready;
1375 transport->old_state_change = sk->sk_state_change;
1376 transport->old_write_space = sk->sk_write_space;
1377 sk->sk_data_ready = xs_udp_data_ready;
1378 sk->sk_write_space = xs_udp_write_space;
1379 sk->sk_no_check = UDP_CSUM_NORCV;
1380 sk->sk_allocation = GFP_ATOMIC;
1382 xprt_set_connected(xprt);
1384 /* Reset to new socket */
1385 transport->sock = sock;
1386 transport->inet = sk;
1388 write_unlock_bh(&sk->sk_callback_lock);
1390 xs_udp_do_set_buffer_size(xprt);
1394 * xs_udp_connect_worker4 - set up a UDP socket
1395 * @work: RPC transport to connect
1397 * Invoked by a work queue tasklet.
1399 static void xs_udp_connect_worker4(struct work_struct *work)
1401 struct sock_xprt *transport =
1402 container_of(work, struct sock_xprt, connect_worker.work);
1403 struct rpc_xprt *xprt = &transport->xprt;
1404 struct socket *sock = transport->sock;
1405 int err, status = -EIO;
1407 if (xprt->shutdown || !xprt_bound(xprt))
1410 /* Start by resetting any existing state */
1413 if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1414 dprintk("RPC: can't create UDP transport socket (%d).\n", -err);
1417 xs_reclassify_socket4(sock);
1419 if (xs_bind4(transport, sock)) {
1424 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1425 xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1427 xs_udp_finish_connecting(xprt, sock);
1430 xprt_wake_pending_tasks(xprt, status);
1431 xprt_clear_connecting(xprt);
1435 * xs_udp_connect_worker6 - set up a UDP socket
1436 * @work: RPC transport to connect
1438 * Invoked by a work queue tasklet.
1440 static void xs_udp_connect_worker6(struct work_struct *work)
1442 struct sock_xprt *transport =
1443 container_of(work, struct sock_xprt, connect_worker.work);
1444 struct rpc_xprt *xprt = &transport->xprt;
1445 struct socket *sock = transport->sock;
1446 int err, status = -EIO;
1448 if (xprt->shutdown || !xprt_bound(xprt))
1451 /* Start by resetting any existing state */
1454 if ((err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1455 dprintk("RPC: can't create UDP transport socket (%d).\n", -err);
1458 xs_reclassify_socket6(sock);
1460 if (xs_bind6(transport, sock) < 0) {
1465 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1466 xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1468 xs_udp_finish_connecting(xprt, sock);
1471 xprt_wake_pending_tasks(xprt, status);
1472 xprt_clear_connecting(xprt);
1476 * We need to preserve the port number so the reply cache on the server can
1477 * find our cached RPC replies when we get around to reconnecting.
1479 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1482 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1483 struct sockaddr any;
1485 dprintk("RPC: disconnecting xprt %p to reuse port\n", xprt);
1488 * Disconnect the transport socket by doing a connect operation
1489 * with AF_UNSPEC. This should return immediately...
1491 memset(&any, 0, sizeof(any));
1492 any.sa_family = AF_UNSPEC;
1493 result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1495 dprintk("RPC: AF_UNSPEC connect return code %d\n",
1499 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1501 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1503 if (!transport->inet) {
1504 struct sock *sk = sock->sk;
1506 write_lock_bh(&sk->sk_callback_lock);
1508 sk->sk_user_data = xprt;
1509 transport->old_data_ready = sk->sk_data_ready;
1510 transport->old_state_change = sk->sk_state_change;
1511 transport->old_write_space = sk->sk_write_space;
1512 sk->sk_data_ready = xs_tcp_data_ready;
1513 sk->sk_state_change = xs_tcp_state_change;
1514 sk->sk_write_space = xs_tcp_write_space;
1515 sk->sk_allocation = GFP_ATOMIC;
1517 /* socket options */
1518 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1519 sock_reset_flag(sk, SOCK_LINGER);
1520 tcp_sk(sk)->linger2 = 0;
1521 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1523 xprt_clear_connected(xprt);
1525 /* Reset to new socket */
1526 transport->sock = sock;
1527 transport->inet = sk;
1529 write_unlock_bh(&sk->sk_callback_lock);
1532 /* Tell the socket layer to start connecting... */
1533 xprt->stat.connect_count++;
1534 xprt->stat.connect_start = jiffies;
1535 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
1539 * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1540 * @work: RPC transport to connect
1542 * Invoked by a work queue tasklet.
1544 static void xs_tcp_connect_worker4(struct work_struct *work)
1546 struct sock_xprt *transport =
1547 container_of(work, struct sock_xprt, connect_worker.work);
1548 struct rpc_xprt *xprt = &transport->xprt;
1549 struct socket *sock = transport->sock;
1550 int err, status = -EIO;
1552 if (xprt->shutdown || !xprt_bound(xprt))
1556 /* start from scratch */
1557 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1558 dprintk("RPC: can't create TCP transport socket (%d).\n", -err);
1561 xs_reclassify_socket4(sock);
1563 if (xs_bind4(transport, sock) < 0) {
1568 /* "close" the socket, preserving the local port */
1569 xs_tcp_reuse_connection(xprt);
1571 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1572 xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1574 status = xs_tcp_finish_connecting(xprt, sock);
1575 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
1576 xprt, -status, xprt_connected(xprt),
1577 sock->sk->sk_state);
1585 /* retry with existing socket, after a delay */
1588 /* get rid of existing socket, and retry */
1594 xprt_wake_pending_tasks(xprt, status);
1596 xprt_clear_connecting(xprt);
1600 * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
1601 * @work: RPC transport to connect
1603 * Invoked by a work queue tasklet.
1605 static void xs_tcp_connect_worker6(struct work_struct *work)
1607 struct sock_xprt *transport =
1608 container_of(work, struct sock_xprt, connect_worker.work);
1609 struct rpc_xprt *xprt = &transport->xprt;
1610 struct socket *sock = transport->sock;
1611 int err, status = -EIO;
1613 if (xprt->shutdown || !xprt_bound(xprt))
1617 /* start from scratch */
1618 if ((err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1619 dprintk("RPC: can't create TCP transport socket (%d).\n", -err);
1622 xs_reclassify_socket6(sock);
1624 if (xs_bind6(transport, sock) < 0) {
1629 /* "close" the socket, preserving the local port */
1630 xs_tcp_reuse_connection(xprt);
1632 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1633 xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1635 status = xs_tcp_finish_connecting(xprt, sock);
1636 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
1637 xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1645 /* retry with existing socket, after a delay */
1648 /* get rid of existing socket, and retry */
1654 xprt_wake_pending_tasks(xprt, status);
1656 xprt_clear_connecting(xprt);
1660 * xs_connect - connect a socket to a remote endpoint
1661 * @task: address of RPC task that manages state of connect request
1663 * TCP: If the remote end dropped the connection, delay reconnecting.
1665 * UDP socket connects are synchronous, but we use a work queue anyway
1666 * to guarantee that even unprivileged user processes can set up a
1667 * socket on a privileged port.
1669 * If a UDP socket connect fails, the delay behavior here prevents
1670 * retry floods (hard mounts).
1672 static void xs_connect(struct rpc_task *task)
1674 struct rpc_xprt *xprt = task->tk_xprt;
1675 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1677 if (xprt_test_and_set_connecting(xprt))
1680 if (transport->sock != NULL) {
1681 dprintk("RPC: xs_connect delayed xprt %p for %lu "
1683 xprt, xprt->reestablish_timeout / HZ);
1684 queue_delayed_work(rpciod_workqueue,
1685 &transport->connect_worker,
1686 xprt->reestablish_timeout);
1687 xprt->reestablish_timeout <<= 1;
1688 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1689 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1691 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
1692 queue_delayed_work(rpciod_workqueue,
1693 &transport->connect_worker, 0);
1698 * xs_udp_print_stats - display UDP socket-specifc stats
1699 * @xprt: rpc_xprt struct containing statistics
1703 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1705 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1707 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1709 xprt->stat.bind_count,
1712 xprt->stat.bad_xids,
1714 xprt->stat.bklog_u);
1718 * xs_tcp_print_stats - display TCP socket-specifc stats
1719 * @xprt: rpc_xprt struct containing statistics
1723 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1725 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1728 if (xprt_connected(xprt))
1729 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1731 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1733 xprt->stat.bind_count,
1734 xprt->stat.connect_count,
1735 xprt->stat.connect_time,
1739 xprt->stat.bad_xids,
1741 xprt->stat.bklog_u);
1744 static struct rpc_xprt_ops xs_udp_ops = {
1745 .set_buffer_size = xs_udp_set_buffer_size,
1746 .reserve_xprt = xprt_reserve_xprt_cong,
1747 .release_xprt = xprt_release_xprt_cong,
1748 .rpcbind = rpcb_getport_async,
1749 .set_port = xs_set_port,
1750 .connect = xs_connect,
1751 .buf_alloc = rpc_malloc,
1752 .buf_free = rpc_free,
1753 .send_request = xs_udp_send_request,
1754 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
1755 .timer = xs_udp_timer,
1756 .release_request = xprt_release_rqst_cong,
1758 .destroy = xs_destroy,
1759 .print_stats = xs_udp_print_stats,
1762 static struct rpc_xprt_ops xs_tcp_ops = {
1763 .reserve_xprt = xprt_reserve_xprt,
1764 .release_xprt = xs_tcp_release_xprt,
1765 .rpcbind = rpcb_getport_async,
1766 .set_port = xs_set_port,
1767 .connect = xs_connect,
1768 .buf_alloc = rpc_malloc,
1769 .buf_free = rpc_free,
1770 .send_request = xs_tcp_send_request,
1771 .set_retrans_timeout = xprt_set_retrans_timeout_def,
1773 .destroy = xs_destroy,
1774 .print_stats = xs_tcp_print_stats,
1777 static struct rpc_xprt *xs_setup_xprt(struct rpc_xprtsock_create *args, unsigned int slot_table_size)
1779 struct rpc_xprt *xprt;
1780 struct sock_xprt *new;
1782 if (args->addrlen > sizeof(xprt->addr)) {
1783 dprintk("RPC: xs_setup_xprt: address too large\n");
1784 return ERR_PTR(-EBADF);
1787 new = kzalloc(sizeof(*new), GFP_KERNEL);
1789 dprintk("RPC: xs_setup_xprt: couldn't allocate "
1791 return ERR_PTR(-ENOMEM);
1795 xprt->max_reqs = slot_table_size;
1796 xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1797 if (xprt->slot == NULL) {
1799 dprintk("RPC: xs_setup_xprt: couldn't allocate slot "
1801 return ERR_PTR(-ENOMEM);
1804 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
1805 xprt->addrlen = args->addrlen;
1807 memcpy(&new->addr, args->srcaddr, args->addrlen);
1808 new->port = xs_get_random_port();
1814 * xs_setup_udp - Set up transport to use a UDP socket
1815 * @args: rpc transport creation arguments
1818 struct rpc_xprt *xs_setup_udp(struct rpc_xprtsock_create *args)
1820 struct sockaddr *addr = args->dstaddr;
1821 struct rpc_xprt *xprt;
1822 struct sock_xprt *transport;
1824 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
1827 transport = container_of(xprt, struct sock_xprt, xprt);
1829 xprt->prot = IPPROTO_UDP;
1831 /* XXX: header size can vary due to auth type, IPv6, etc. */
1832 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1834 xprt->bind_timeout = XS_BIND_TO;
1835 xprt->connect_timeout = XS_UDP_CONN_TO;
1836 xprt->reestablish_timeout = XS_UDP_REEST_TO;
1837 xprt->idle_timeout = XS_IDLE_DISC_TO;
1839 xprt->ops = &xs_udp_ops;
1842 xprt->timeout = *args->timeout;
1844 xprt_set_timeout(&xprt->timeout, 5, 5 * HZ);
1846 switch (addr->sa_family) {
1848 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1849 xprt_set_bound(xprt);
1851 INIT_DELAYED_WORK(&transport->connect_worker,
1852 xs_udp_connect_worker4);
1853 xs_format_ipv4_peer_addresses(xprt);
1856 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
1857 xprt_set_bound(xprt);
1859 INIT_DELAYED_WORK(&transport->connect_worker,
1860 xs_udp_connect_worker6);
1861 xs_format_ipv6_peer_addresses(xprt);
1865 return ERR_PTR(-EAFNOSUPPORT);
1868 dprintk("RPC: set up transport to address %s\n",
1869 xprt->address_strings[RPC_DISPLAY_ALL]);
1875 * xs_setup_tcp - Set up transport to use a TCP socket
1876 * @args: rpc transport creation arguments
1879 struct rpc_xprt *xs_setup_tcp(struct rpc_xprtsock_create *args)
1881 struct sockaddr *addr = args->dstaddr;
1882 struct rpc_xprt *xprt;
1883 struct sock_xprt *transport;
1885 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
1888 transport = container_of(xprt, struct sock_xprt, xprt);
1890 xprt->prot = IPPROTO_TCP;
1891 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
1892 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
1894 xprt->bind_timeout = XS_BIND_TO;
1895 xprt->connect_timeout = XS_TCP_CONN_TO;
1896 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1897 xprt->idle_timeout = XS_IDLE_DISC_TO;
1899 xprt->ops = &xs_tcp_ops;
1902 xprt->timeout = *args->timeout;
1904 xprt_set_timeout(&xprt->timeout, 2, 60 * HZ);
1906 switch (addr->sa_family) {
1908 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1909 xprt_set_bound(xprt);
1911 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker4);
1912 xs_format_ipv4_peer_addresses(xprt);
1915 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
1916 xprt_set_bound(xprt);
1918 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker6);
1919 xs_format_ipv6_peer_addresses(xprt);
1923 return ERR_PTR(-EAFNOSUPPORT);
1926 dprintk("RPC: set up transport to address %s\n",
1927 xprt->address_strings[RPC_DISPLAY_ALL]);
1933 * init_socket_xprt - set up xprtsock's sysctls
1936 int init_socket_xprt(void)
1939 if (!sunrpc_table_header)
1940 sunrpc_table_header = register_sysctl_table(sunrpc_table);
1947 * cleanup_socket_xprt - remove xprtsock's sysctls
1950 void cleanup_socket_xprt(void)
1953 if (sunrpc_table_header) {
1954 unregister_sysctl_table(sunrpc_table_header);
1955 sunrpc_table_header = NULL;