2 * linux/net/sunrpc/xprtsock.c
4 * Client-side transport implementation for sockets.
6 * TCP callback races fixes (C) 1998 Red Hat Software <alan@redhat.com>
7 * TCP send fixes (C) 1998 Red Hat Software <alan@redhat.com>
8 * TCP NFS related read + write fixes
9 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
11 * Rewrite of larges part of the code in order to stabilize TCP stuff.
12 * Fix behaviour when socket buffer is full.
13 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
15 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
17 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
18 * <gilles.quillard@bull.net>
21 #include <linux/types.h>
22 #include <linux/slab.h>
23 #include <linux/module.h>
24 #include <linux/capability.h>
25 #include <linux/pagemap.h>
26 #include <linux/errno.h>
27 #include <linux/socket.h>
29 #include <linux/net.h>
31 #include <linux/udp.h>
32 #include <linux/tcp.h>
33 #include <linux/sunrpc/clnt.h>
34 #include <linux/sunrpc/sched.h>
35 #include <linux/sunrpc/xprtsock.h>
36 #include <linux/file.h>
39 #include <net/checksum.h>
46 unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
47 unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE;
49 unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
50 unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
53 * We can register our own files under /proc/sys/sunrpc by
54 * calling register_sysctl_table() again. The files in that
55 * directory become the union of all files registered there.
57 * We simply need to make sure that we don't collide with
58 * someone else's file names!
63 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
64 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
65 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
66 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
68 static struct ctl_table_header *sunrpc_table_header;
71 * FIXME: changing the UDP slot table size should also resize the UDP
72 * socket buffers for existing UDP transports
74 static ctl_table xs_tunables_table[] = {
76 .ctl_name = CTL_SLOTTABLE_UDP,
77 .procname = "udp_slot_table_entries",
78 .data = &xprt_udp_slot_table_entries,
79 .maxlen = sizeof(unsigned int),
81 .proc_handler = &proc_dointvec_minmax,
82 .strategy = &sysctl_intvec,
83 .extra1 = &min_slot_table_size,
84 .extra2 = &max_slot_table_size
87 .ctl_name = CTL_SLOTTABLE_TCP,
88 .procname = "tcp_slot_table_entries",
89 .data = &xprt_tcp_slot_table_entries,
90 .maxlen = sizeof(unsigned int),
92 .proc_handler = &proc_dointvec_minmax,
93 .strategy = &sysctl_intvec,
94 .extra1 = &min_slot_table_size,
95 .extra2 = &max_slot_table_size
98 .ctl_name = CTL_MIN_RESVPORT,
99 .procname = "min_resvport",
100 .data = &xprt_min_resvport,
101 .maxlen = sizeof(unsigned int),
103 .proc_handler = &proc_dointvec_minmax,
104 .strategy = &sysctl_intvec,
105 .extra1 = &xprt_min_resvport_limit,
106 .extra2 = &xprt_max_resvport_limit
109 .ctl_name = CTL_MAX_RESVPORT,
110 .procname = "max_resvport",
111 .data = &xprt_max_resvport,
112 .maxlen = sizeof(unsigned int),
114 .proc_handler = &proc_dointvec_minmax,
115 .strategy = &sysctl_intvec,
116 .extra1 = &xprt_min_resvport_limit,
117 .extra2 = &xprt_max_resvport_limit
124 static ctl_table sunrpc_table[] = {
126 .ctl_name = CTL_SUNRPC,
127 .procname = "sunrpc",
129 .child = xs_tunables_table
139 * How many times to try sending a request on a socket before waiting
140 * for the socket buffer to clear.
142 #define XS_SENDMSG_RETRY (10U)
145 * Time out for an RPC UDP socket connect. UDP socket connects are
146 * synchronous, but we set a timeout anyway in case of resource
147 * exhaustion on the local host.
149 #define XS_UDP_CONN_TO (5U * HZ)
152 * Wait duration for an RPC TCP connection to be established. Solaris
153 * NFS over TCP uses 60 seconds, for example, which is in line with how
154 * long a server takes to reboot.
156 #define XS_TCP_CONN_TO (60U * HZ)
159 * Wait duration for a reply from the RPC portmapper.
161 #define XS_BIND_TO (60U * HZ)
164 * Delay if a UDP socket connect error occurs. This is most likely some
165 * kind of resource problem on the local host.
167 #define XS_UDP_REEST_TO (2U * HZ)
170 * The reestablish timeout allows clients to delay for a bit before attempting
171 * to reconnect to a server that just dropped our connection.
173 * We implement an exponential backoff when trying to reestablish a TCP
174 * transport connection with the server. Some servers like to drop a TCP
175 * connection when they are overworked, so we start with a short timeout and
176 * increase over time if the server is down or not responding.
178 #define XS_TCP_INIT_REEST_TO (3U * HZ)
179 #define XS_TCP_MAX_REEST_TO (5U * 60 * HZ)
182 * TCP idle timeout; client drops the transport socket if it is idle
183 * for this long. Note that we also timeout UDP sockets to prevent
184 * holding port numbers when there is no RPC traffic.
186 #define XS_IDLE_DISC_TO (5U * 60 * HZ)
189 # undef RPC_DEBUG_DATA
190 # define RPCDBG_FACILITY RPCDBG_TRANS
193 #ifdef RPC_DEBUG_DATA
194 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
196 u8 *buf = (u8 *) packet;
199 dprintk("RPC: %s\n", msg);
200 for (j = 0; j < count && j < 128; j += 4) {
204 dprintk("0x%04x ", j);
206 dprintk("%02x%02x%02x%02x ",
207 buf[j], buf[j+1], buf[j+2], buf[j+3]);
212 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
219 struct rpc_xprt xprt;
224 struct socket * sock;
228 * State of TCP reply receive
236 unsigned long tcp_copied,
240 * Connection of transports
242 struct delayed_work connect_worker;
243 struct sockaddr_storage addr;
247 * UDP socket buffer size parameters
253 * Saved socket callback addresses
255 void (*old_data_ready)(struct sock *, int);
256 void (*old_state_change)(struct sock *);
257 void (*old_write_space)(struct sock *);
261 * TCP receive state flags
263 #define TCP_RCV_LAST_FRAG (1UL << 0)
264 #define TCP_RCV_COPY_FRAGHDR (1UL << 1)
265 #define TCP_RCV_COPY_XID (1UL << 2)
266 #define TCP_RCV_COPY_DATA (1UL << 3)
268 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
270 return (struct sockaddr *) &xprt->addr;
273 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
275 return (struct sockaddr_in *) &xprt->addr;
278 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
280 return (struct sockaddr_in6 *) &xprt->addr;
283 static void xs_format_ipv4_peer_addresses(struct rpc_xprt *xprt,
284 const char *protocol,
287 struct sockaddr_in *addr = xs_addr_in(xprt);
290 buf = kzalloc(20, GFP_KERNEL);
292 snprintf(buf, 20, NIPQUAD_FMT,
293 NIPQUAD(addr->sin_addr.s_addr));
295 xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
297 buf = kzalloc(8, GFP_KERNEL);
299 snprintf(buf, 8, "%u",
300 ntohs(addr->sin_port));
302 xprt->address_strings[RPC_DISPLAY_PORT] = buf;
304 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
306 buf = kzalloc(48, GFP_KERNEL);
308 snprintf(buf, 48, "addr="NIPQUAD_FMT" port=%u proto=%s",
309 NIPQUAD(addr->sin_addr.s_addr),
310 ntohs(addr->sin_port),
313 xprt->address_strings[RPC_DISPLAY_ALL] = buf;
315 buf = kzalloc(10, GFP_KERNEL);
317 snprintf(buf, 10, "%02x%02x%02x%02x",
318 NIPQUAD(addr->sin_addr.s_addr));
320 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
322 buf = kzalloc(8, GFP_KERNEL);
324 snprintf(buf, 8, "%4hx",
325 ntohs(addr->sin_port));
327 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
329 buf = kzalloc(30, GFP_KERNEL);
331 snprintf(buf, 30, NIPQUAD_FMT".%u.%u",
332 NIPQUAD(addr->sin_addr.s_addr),
333 ntohs(addr->sin_port) >> 8,
334 ntohs(addr->sin_port) & 0xff);
336 xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
338 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
341 static void xs_format_ipv6_peer_addresses(struct rpc_xprt *xprt,
342 const char *protocol,
345 struct sockaddr_in6 *addr = xs_addr_in6(xprt);
348 buf = kzalloc(40, GFP_KERNEL);
350 snprintf(buf, 40, NIP6_FMT,
351 NIP6(addr->sin6_addr));
353 xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
355 buf = kzalloc(8, GFP_KERNEL);
357 snprintf(buf, 8, "%u",
358 ntohs(addr->sin6_port));
360 xprt->address_strings[RPC_DISPLAY_PORT] = buf;
362 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
364 buf = kzalloc(64, GFP_KERNEL);
366 snprintf(buf, 64, "addr="NIP6_FMT" port=%u proto=%s",
367 NIP6(addr->sin6_addr),
368 ntohs(addr->sin6_port),
371 xprt->address_strings[RPC_DISPLAY_ALL] = buf;
373 buf = kzalloc(36, GFP_KERNEL);
375 snprintf(buf, 36, NIP6_SEQFMT,
376 NIP6(addr->sin6_addr));
378 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
380 buf = kzalloc(8, GFP_KERNEL);
382 snprintf(buf, 8, "%4hx",
383 ntohs(addr->sin6_port));
385 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
387 buf = kzalloc(50, GFP_KERNEL);
389 snprintf(buf, 50, NIP6_FMT".%u.%u",
390 NIP6(addr->sin6_addr),
391 ntohs(addr->sin6_port) >> 8,
392 ntohs(addr->sin6_port) & 0xff);
394 xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
396 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
399 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
403 for (i = 0; i < RPC_DISPLAY_MAX; i++)
405 case RPC_DISPLAY_PROTO:
406 case RPC_DISPLAY_NETID:
409 kfree(xprt->address_strings[i]);
413 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
415 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
417 struct msghdr msg = {
419 .msg_namelen = addrlen,
420 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
423 .iov_base = vec->iov_base + base,
424 .iov_len = vec->iov_len - base,
427 if (iov.iov_len != 0)
428 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
429 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
432 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
435 unsigned int remainder;
438 remainder = xdr->page_len - base;
439 base += xdr->page_base;
440 ppage = xdr->pages + (base >> PAGE_SHIFT);
443 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
444 int flags = XS_SENDMSG_FLAGS;
447 if (remainder != 0 || more)
449 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
450 if (remainder == 0 || err != len)
464 * xs_sendpages - write pages directly to a socket
465 * @sock: socket to send on
466 * @addr: UDP only -- address of destination
467 * @addrlen: UDP only -- length of destination address
468 * @xdr: buffer containing this request
469 * @base: starting position in the buffer
472 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
474 unsigned int remainder = xdr->len - base;
480 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
486 if (base < xdr->head[0].iov_len || addr != NULL) {
487 unsigned int len = xdr->head[0].iov_len - base;
489 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
490 if (remainder == 0 || err != len)
495 base -= xdr->head[0].iov_len;
497 if (base < xdr->page_len) {
498 unsigned int len = xdr->page_len - base;
500 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
501 if (remainder == 0 || err != len)
506 base -= xdr->page_len;
508 if (base >= xdr->tail[0].iov_len)
510 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
520 * xs_nospace - place task on wait queue if transmit was incomplete
521 * @task: task to put to sleep
524 static void xs_nospace(struct rpc_task *task)
526 struct rpc_rqst *req = task->tk_rqstp;
527 struct rpc_xprt *xprt = req->rq_xprt;
528 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
530 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
531 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
534 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
535 /* Protect against races with write_space */
536 spin_lock_bh(&xprt->transport_lock);
538 /* Don't race with disconnect */
539 if (!xprt_connected(xprt))
540 task->tk_status = -ENOTCONN;
541 else if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
542 xprt_wait_for_buffer_space(task);
544 spin_unlock_bh(&xprt->transport_lock);
546 /* Keep holding the socket if it is blocked */
547 rpc_delay(task, HZ>>4);
551 * xs_udp_send_request - write an RPC request to a UDP socket
552 * @task: address of RPC task that manages the state of an RPC request
555 * 0: The request has been sent
556 * EAGAIN: The socket was blocked, please call again later to
557 * complete the request
558 * ENOTCONN: Caller needs to invoke connect logic then call again
559 * other: Some other error occured, the request was not sent
561 static int xs_udp_send_request(struct rpc_task *task)
563 struct rpc_rqst *req = task->tk_rqstp;
564 struct rpc_xprt *xprt = req->rq_xprt;
565 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
566 struct xdr_buf *xdr = &req->rq_snd_buf;
569 xs_pktdump("packet data:",
570 req->rq_svec->iov_base,
571 req->rq_svec->iov_len);
573 req->rq_xtime = jiffies;
574 status = xs_sendpages(transport->sock,
579 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
580 xdr->len - req->rq_bytes_sent, status);
583 task->tk_bytes_sent += status;
584 if (status >= req->rq_slen)
586 /* Still some bytes left; set up for a retry later. */
594 /* When the server has died, an ICMP port unreachable message
595 * prompts ECONNREFUSED. */
601 dprintk("RPC: sendmsg returned unrecognized error %d\n",
610 * xs_tcp_shutdown - gracefully shut down a TCP socket
613 * Initiates a graceful shutdown of the TCP socket by calling the
614 * equivalent of shutdown(SHUT_WR);
616 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
618 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
619 struct socket *sock = transport->sock;
622 kernel_sock_shutdown(sock, SHUT_WR);
625 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
627 u32 reclen = buf->len - sizeof(rpc_fraghdr);
628 rpc_fraghdr *base = buf->head[0].iov_base;
629 *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
633 * xs_tcp_send_request - write an RPC request to a TCP socket
634 * @task: address of RPC task that manages the state of an RPC request
637 * 0: The request has been sent
638 * EAGAIN: The socket was blocked, please call again later to
639 * complete the request
640 * ENOTCONN: Caller needs to invoke connect logic then call again
641 * other: Some other error occured, the request was not sent
643 * XXX: In the case of soft timeouts, should we eventually give up
644 * if sendmsg is not able to make progress?
646 static int xs_tcp_send_request(struct rpc_task *task)
648 struct rpc_rqst *req = task->tk_rqstp;
649 struct rpc_xprt *xprt = req->rq_xprt;
650 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
651 struct xdr_buf *xdr = &req->rq_snd_buf;
653 unsigned int retry = 0;
655 xs_encode_tcp_record_marker(&req->rq_snd_buf);
657 xs_pktdump("packet data:",
658 req->rq_svec->iov_base,
659 req->rq_svec->iov_len);
661 /* Continue transmitting the packet/record. We must be careful
662 * to cope with writespace callbacks arriving _after_ we have
663 * called sendmsg(). */
665 req->rq_xtime = jiffies;
666 status = xs_sendpages(transport->sock,
667 NULL, 0, xdr, req->rq_bytes_sent);
669 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
670 xdr->len - req->rq_bytes_sent, status);
672 if (unlikely(status < 0))
675 /* If we've sent the entire packet, immediately
676 * reset the count of bytes sent. */
677 req->rq_bytes_sent += status;
678 task->tk_bytes_sent += status;
679 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
680 req->rq_bytes_sent = 0;
685 if (retry++ > XS_SENDMSG_RETRY)
700 dprintk("RPC: sendmsg returned unrecognized error %d\n",
702 xs_tcp_shutdown(xprt);
710 * xs_tcp_release_xprt - clean up after a tcp transmission
714 * This cleans up if an error causes us to abort the transmission of a request.
715 * In this case, the socket may need to be reset in order to avoid confusing
718 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
720 struct rpc_rqst *req;
722 if (task != xprt->snd_task)
726 req = task->tk_rqstp;
727 if (req->rq_bytes_sent == 0)
729 if (req->rq_bytes_sent == req->rq_snd_buf.len)
731 set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
733 xprt_release_xprt(xprt, task);
737 * xs_close - close a socket
740 * This is used when all requests are complete; ie, no DRC state remains
741 * on the server we want to save.
743 static void xs_close(struct rpc_xprt *xprt)
745 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
746 struct socket *sock = transport->sock;
747 struct sock *sk = transport->inet;
750 goto clear_close_wait;
752 dprintk("RPC: xs_close xprt %p\n", xprt);
754 write_lock_bh(&sk->sk_callback_lock);
755 transport->inet = NULL;
756 transport->sock = NULL;
758 sk->sk_user_data = NULL;
759 sk->sk_data_ready = transport->old_data_ready;
760 sk->sk_state_change = transport->old_state_change;
761 sk->sk_write_space = transport->old_write_space;
762 write_unlock_bh(&sk->sk_callback_lock);
768 smp_mb__before_clear_bit();
769 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
770 clear_bit(XPRT_CLOSING, &xprt->state);
771 smp_mb__after_clear_bit();
772 xprt_disconnect_done(xprt);
776 * xs_destroy - prepare to shutdown a transport
777 * @xprt: doomed transport
780 static void xs_destroy(struct rpc_xprt *xprt)
782 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
784 dprintk("RPC: xs_destroy xprt %p\n", xprt);
786 cancel_rearming_delayed_work(&transport->connect_worker);
789 xs_free_peer_addresses(xprt);
792 module_put(THIS_MODULE);
795 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
797 return (struct rpc_xprt *) sk->sk_user_data;
801 * xs_udp_data_ready - "data ready" callback for UDP sockets
802 * @sk: socket with data to read
803 * @len: how much data to read
806 static void xs_udp_data_ready(struct sock *sk, int len)
808 struct rpc_task *task;
809 struct rpc_xprt *xprt;
810 struct rpc_rqst *rovr;
812 int err, repsize, copied;
816 read_lock(&sk->sk_callback_lock);
817 dprintk("RPC: xs_udp_data_ready...\n");
818 if (!(xprt = xprt_from_sock(sk)))
821 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
827 repsize = skb->len - sizeof(struct udphdr);
829 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
833 /* Copy the XID from the skb... */
834 xp = skb_header_pointer(skb, sizeof(struct udphdr),
835 sizeof(_xid), &_xid);
839 /* Look up and lock the request corresponding to the given XID */
840 spin_lock(&xprt->transport_lock);
841 rovr = xprt_lookup_rqst(xprt, *xp);
844 task = rovr->rq_task;
846 if ((copied = rovr->rq_private_buf.buflen) > repsize)
849 /* Suck it into the iovec, verify checksum if not done by hw. */
850 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
851 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
855 UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
857 /* Something worked... */
858 dst_confirm(skb->dst);
860 xprt_adjust_cwnd(task, copied);
861 xprt_update_rtt(task);
862 xprt_complete_rqst(task, copied);
865 spin_unlock(&xprt->transport_lock);
867 skb_free_datagram(sk, skb);
869 read_unlock(&sk->sk_callback_lock);
872 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
874 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
878 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
879 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
880 used = xdr_skb_read_bits(desc, p, len);
881 transport->tcp_offset += used;
885 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
886 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
887 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
889 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
890 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
892 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
893 transport->tcp_offset = 0;
895 /* Sanity check of the record length */
896 if (unlikely(transport->tcp_reclen < 4)) {
897 dprintk("RPC: invalid TCP record fragment length\n");
898 xprt_force_disconnect(xprt);
901 dprintk("RPC: reading TCP record fragment of length %d\n",
902 transport->tcp_reclen);
905 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
907 if (transport->tcp_offset == transport->tcp_reclen) {
908 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
909 transport->tcp_offset = 0;
910 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
911 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
912 transport->tcp_flags |= TCP_RCV_COPY_XID;
913 transport->tcp_copied = 0;
918 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
923 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
924 dprintk("RPC: reading XID (%Zu bytes)\n", len);
925 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
926 used = xdr_skb_read_bits(desc, p, len);
927 transport->tcp_offset += used;
930 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
931 transport->tcp_flags |= TCP_RCV_COPY_DATA;
932 transport->tcp_copied = 4;
933 dprintk("RPC: reading reply for XID %08x\n",
934 ntohl(transport->tcp_xid));
935 xs_tcp_check_fraghdr(transport);
938 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
940 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
941 struct rpc_rqst *req;
942 struct xdr_buf *rcvbuf;
946 /* Find and lock the request corresponding to this xid */
947 spin_lock(&xprt->transport_lock);
948 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
950 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
951 dprintk("RPC: XID %08x request not found!\n",
952 ntohl(transport->tcp_xid));
953 spin_unlock(&xprt->transport_lock);
957 rcvbuf = &req->rq_private_buf;
959 if (len > transport->tcp_reclen - transport->tcp_offset) {
960 struct xdr_skb_reader my_desc;
962 len = transport->tcp_reclen - transport->tcp_offset;
963 memcpy(&my_desc, desc, sizeof(my_desc));
965 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
966 &my_desc, xdr_skb_read_bits);
970 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
971 desc, xdr_skb_read_bits);
974 transport->tcp_copied += r;
975 transport->tcp_offset += r;
978 /* Error when copying to the receive buffer,
979 * usually because we weren't able to allocate
980 * additional buffer pages. All we can do now
981 * is turn off TCP_RCV_COPY_DATA, so the request
982 * will not receive any additional updates,
984 * Any remaining data from this record will
987 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
988 dprintk("RPC: XID %08x truncated request\n",
989 ntohl(transport->tcp_xid));
990 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
991 "tcp_offset = %u, tcp_reclen = %u\n",
992 xprt, transport->tcp_copied,
993 transport->tcp_offset, transport->tcp_reclen);
997 dprintk("RPC: XID %08x read %Zd bytes\n",
998 ntohl(transport->tcp_xid), r);
999 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1000 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1001 transport->tcp_offset, transport->tcp_reclen);
1003 if (transport->tcp_copied == req->rq_private_buf.buflen)
1004 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1005 else if (transport->tcp_offset == transport->tcp_reclen) {
1006 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1007 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1011 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1012 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1013 spin_unlock(&xprt->transport_lock);
1014 xs_tcp_check_fraghdr(transport);
1017 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1021 len = transport->tcp_reclen - transport->tcp_offset;
1022 if (len > desc->count)
1025 desc->offset += len;
1026 transport->tcp_offset += len;
1027 dprintk("RPC: discarded %Zu bytes\n", len);
1028 xs_tcp_check_fraghdr(transport);
1031 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1033 struct rpc_xprt *xprt = rd_desc->arg.data;
1034 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1035 struct xdr_skb_reader desc = {
1041 dprintk("RPC: xs_tcp_data_recv started\n");
1043 /* Read in a new fragment marker if necessary */
1044 /* Can we ever really expect to get completely empty fragments? */
1045 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1046 xs_tcp_read_fraghdr(xprt, &desc);
1049 /* Read in the xid if necessary */
1050 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1051 xs_tcp_read_xid(transport, &desc);
1054 /* Read in the request data */
1055 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1056 xs_tcp_read_request(xprt, &desc);
1059 /* Skip over any trailing bytes on short reads */
1060 xs_tcp_read_discard(transport, &desc);
1061 } while (desc.count);
1062 dprintk("RPC: xs_tcp_data_recv done\n");
1063 return len - desc.count;
1067 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1068 * @sk: socket with data to read
1069 * @bytes: how much data to read
1072 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1074 struct rpc_xprt *xprt;
1075 read_descriptor_t rd_desc;
1078 dprintk("RPC: xs_tcp_data_ready...\n");
1080 read_lock(&sk->sk_callback_lock);
1081 if (!(xprt = xprt_from_sock(sk)))
1086 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1087 rd_desc.arg.data = xprt;
1089 rd_desc.count = 65536;
1090 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1093 read_unlock(&sk->sk_callback_lock);
1097 * xs_tcp_state_change - callback to handle TCP socket state changes
1098 * @sk: socket whose state has changed
1101 static void xs_tcp_state_change(struct sock *sk)
1103 struct rpc_xprt *xprt;
1105 read_lock(&sk->sk_callback_lock);
1106 if (!(xprt = xprt_from_sock(sk)))
1108 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1109 dprintk("RPC: state %x conn %d dead %d zapped %d\n",
1110 sk->sk_state, xprt_connected(xprt),
1111 sock_flag(sk, SOCK_DEAD),
1112 sock_flag(sk, SOCK_ZAPPED));
1114 switch (sk->sk_state) {
1115 case TCP_ESTABLISHED:
1116 spin_lock_bh(&xprt->transport_lock);
1117 if (!xprt_test_and_set_connected(xprt)) {
1118 struct sock_xprt *transport = container_of(xprt,
1119 struct sock_xprt, xprt);
1121 /* Reset TCP record info */
1122 transport->tcp_offset = 0;
1123 transport->tcp_reclen = 0;
1124 transport->tcp_copied = 0;
1125 transport->tcp_flags =
1126 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1128 xprt_wake_pending_tasks(xprt, 0);
1130 spin_unlock_bh(&xprt->transport_lock);
1133 /* The client initiated a shutdown of the socket */
1134 xprt->reestablish_timeout = 0;
1135 set_bit(XPRT_CLOSING, &xprt->state);
1136 smp_mb__before_clear_bit();
1137 clear_bit(XPRT_CONNECTED, &xprt->state);
1138 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1139 smp_mb__after_clear_bit();
1141 case TCP_CLOSE_WAIT:
1142 /* The server initiated a shutdown of the socket */
1143 set_bit(XPRT_CLOSING, &xprt->state);
1144 xprt_force_disconnect(xprt);
1148 * If the server closed down the connection, make sure that
1149 * we back off before reconnecting
1151 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1152 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1155 smp_mb__before_clear_bit();
1156 clear_bit(XPRT_CONNECTED, &xprt->state);
1157 smp_mb__after_clear_bit();
1160 smp_mb__before_clear_bit();
1161 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1162 clear_bit(XPRT_CLOSING, &xprt->state);
1163 smp_mb__after_clear_bit();
1164 /* Mark transport as closed and wake up all pending tasks */
1165 xprt_disconnect_done(xprt);
1168 read_unlock(&sk->sk_callback_lock);
1172 * xs_udp_write_space - callback invoked when socket buffer space
1174 * @sk: socket whose state has changed
1176 * Called when more output buffer space is available for this socket.
1177 * We try not to wake our writers until they can make "significant"
1178 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1179 * with a bunch of small requests.
1181 static void xs_udp_write_space(struct sock *sk)
1183 read_lock(&sk->sk_callback_lock);
1185 /* from net/core/sock.c:sock_def_write_space */
1186 if (sock_writeable(sk)) {
1187 struct socket *sock;
1188 struct rpc_xprt *xprt;
1190 if (unlikely(!(sock = sk->sk_socket)))
1192 if (unlikely(!(xprt = xprt_from_sock(sk))))
1194 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
1197 xprt_write_space(xprt);
1201 read_unlock(&sk->sk_callback_lock);
1205 * xs_tcp_write_space - callback invoked when socket buffer space
1207 * @sk: socket whose state has changed
1209 * Called when more output buffer space is available for this socket.
1210 * We try not to wake our writers until they can make "significant"
1211 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1212 * with a bunch of small requests.
1214 static void xs_tcp_write_space(struct sock *sk)
1216 read_lock(&sk->sk_callback_lock);
1218 /* from net/core/stream.c:sk_stream_write_space */
1219 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
1220 struct socket *sock;
1221 struct rpc_xprt *xprt;
1223 if (unlikely(!(sock = sk->sk_socket)))
1225 if (unlikely(!(xprt = xprt_from_sock(sk))))
1227 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
1230 xprt_write_space(xprt);
1234 read_unlock(&sk->sk_callback_lock);
1237 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1239 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1240 struct sock *sk = transport->inet;
1242 if (transport->rcvsize) {
1243 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1244 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1246 if (transport->sndsize) {
1247 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1248 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1249 sk->sk_write_space(sk);
1254 * xs_udp_set_buffer_size - set send and receive limits
1255 * @xprt: generic transport
1256 * @sndsize: requested size of send buffer, in bytes
1257 * @rcvsize: requested size of receive buffer, in bytes
1259 * Set socket send and receive buffer size limits.
1261 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1263 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1265 transport->sndsize = 0;
1267 transport->sndsize = sndsize + 1024;
1268 transport->rcvsize = 0;
1270 transport->rcvsize = rcvsize + 1024;
1272 xs_udp_do_set_buffer_size(xprt);
1276 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1277 * @task: task that timed out
1279 * Adjust the congestion window after a retransmit timeout has occurred.
1281 static void xs_udp_timer(struct rpc_task *task)
1283 xprt_adjust_cwnd(task, -ETIMEDOUT);
1286 static unsigned short xs_get_random_port(void)
1288 unsigned short range = xprt_max_resvport - xprt_min_resvport;
1289 unsigned short rand = (unsigned short) net_random() % range;
1290 return rand + xprt_min_resvport;
1294 * xs_set_port - reset the port number in the remote endpoint address
1295 * @xprt: generic transport
1296 * @port: new port number
1299 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1301 struct sockaddr *addr = xs_addr(xprt);
1303 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1305 switch (addr->sa_family) {
1307 ((struct sockaddr_in *)addr)->sin_port = htons(port);
1310 ((struct sockaddr_in6 *)addr)->sin6_port = htons(port);
1317 static unsigned short xs_get_srcport(struct sock_xprt *transport, struct socket *sock)
1319 unsigned short port = transport->port;
1321 if (port == 0 && transport->xprt.resvport)
1322 port = xs_get_random_port();
1326 static unsigned short xs_next_srcport(struct sock_xprt *transport, struct socket *sock, unsigned short port)
1328 if (transport->port != 0)
1329 transport->port = 0;
1330 if (!transport->xprt.resvport)
1332 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1333 return xprt_max_resvport;
1337 static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
1339 struct sockaddr_in myaddr = {
1340 .sin_family = AF_INET,
1342 struct sockaddr_in *sa;
1344 unsigned short port = xs_get_srcport(transport, sock);
1345 unsigned short last;
1347 sa = (struct sockaddr_in *)&transport->addr;
1348 myaddr.sin_addr = sa->sin_addr;
1350 myaddr.sin_port = htons(port);
1351 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1356 transport->port = port;
1360 port = xs_next_srcport(transport, sock, port);
1363 } while (err == -EADDRINUSE && nloop != 2);
1364 dprintk("RPC: %s "NIPQUAD_FMT":%u: %s (%d)\n",
1365 __FUNCTION__, NIPQUAD(myaddr.sin_addr),
1366 port, err ? "failed" : "ok", err);
1370 static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1372 struct sockaddr_in6 myaddr = {
1373 .sin6_family = AF_INET6,
1375 struct sockaddr_in6 *sa;
1377 unsigned short port = xs_get_srcport(transport, sock);
1378 unsigned short last;
1380 sa = (struct sockaddr_in6 *)&transport->addr;
1381 myaddr.sin6_addr = sa->sin6_addr;
1383 myaddr.sin6_port = htons(port);
1384 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1389 transport->port = port;
1393 port = xs_next_srcport(transport, sock, port);
1396 } while (err == -EADDRINUSE && nloop != 2);
1397 dprintk("RPC: xs_bind6 "NIP6_FMT":%u: %s (%d)\n",
1398 NIP6(myaddr.sin6_addr), port, err ? "failed" : "ok", err);
1402 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1403 static struct lock_class_key xs_key[2];
1404 static struct lock_class_key xs_slock_key[2];
1406 static inline void xs_reclassify_socket4(struct socket *sock)
1408 struct sock *sk = sock->sk;
1410 BUG_ON(sock_owned_by_user(sk));
1411 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1412 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1415 static inline void xs_reclassify_socket6(struct socket *sock)
1417 struct sock *sk = sock->sk;
1419 BUG_ON(sock_owned_by_user(sk));
1420 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1421 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1424 static inline void xs_reclassify_socket4(struct socket *sock)
1428 static inline void xs_reclassify_socket6(struct socket *sock)
1433 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1435 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1437 if (!transport->inet) {
1438 struct sock *sk = sock->sk;
1440 write_lock_bh(&sk->sk_callback_lock);
1442 sk->sk_user_data = xprt;
1443 transport->old_data_ready = sk->sk_data_ready;
1444 transport->old_state_change = sk->sk_state_change;
1445 transport->old_write_space = sk->sk_write_space;
1446 sk->sk_data_ready = xs_udp_data_ready;
1447 sk->sk_write_space = xs_udp_write_space;
1448 sk->sk_no_check = UDP_CSUM_NORCV;
1449 sk->sk_allocation = GFP_ATOMIC;
1451 xprt_set_connected(xprt);
1453 /* Reset to new socket */
1454 transport->sock = sock;
1455 transport->inet = sk;
1457 write_unlock_bh(&sk->sk_callback_lock);
1459 xs_udp_do_set_buffer_size(xprt);
1463 * xs_udp_connect_worker4 - set up a UDP socket
1464 * @work: RPC transport to connect
1466 * Invoked by a work queue tasklet.
1468 static void xs_udp_connect_worker4(struct work_struct *work)
1470 struct sock_xprt *transport =
1471 container_of(work, struct sock_xprt, connect_worker.work);
1472 struct rpc_xprt *xprt = &transport->xprt;
1473 struct socket *sock = transport->sock;
1474 int err, status = -EIO;
1476 if (xprt->shutdown || !xprt_bound(xprt))
1479 /* Start by resetting any existing state */
1482 if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1483 dprintk("RPC: can't create UDP transport socket (%d).\n", -err);
1486 xs_reclassify_socket4(sock);
1488 if (xs_bind4(transport, sock)) {
1493 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1494 xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1496 xs_udp_finish_connecting(xprt, sock);
1499 xprt_wake_pending_tasks(xprt, status);
1500 xprt_clear_connecting(xprt);
1504 * xs_udp_connect_worker6 - set up a UDP socket
1505 * @work: RPC transport to connect
1507 * Invoked by a work queue tasklet.
1509 static void xs_udp_connect_worker6(struct work_struct *work)
1511 struct sock_xprt *transport =
1512 container_of(work, struct sock_xprt, connect_worker.work);
1513 struct rpc_xprt *xprt = &transport->xprt;
1514 struct socket *sock = transport->sock;
1515 int err, status = -EIO;
1517 if (xprt->shutdown || !xprt_bound(xprt))
1520 /* Start by resetting any existing state */
1523 if ((err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1524 dprintk("RPC: can't create UDP transport socket (%d).\n", -err);
1527 xs_reclassify_socket6(sock);
1529 if (xs_bind6(transport, sock) < 0) {
1534 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1535 xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1537 xs_udp_finish_connecting(xprt, sock);
1540 xprt_wake_pending_tasks(xprt, status);
1541 xprt_clear_connecting(xprt);
1545 * We need to preserve the port number so the reply cache on the server can
1546 * find our cached RPC replies when we get around to reconnecting.
1548 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1551 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1552 struct sockaddr any;
1554 dprintk("RPC: disconnecting xprt %p to reuse port\n", xprt);
1557 * Disconnect the transport socket by doing a connect operation
1558 * with AF_UNSPEC. This should return immediately...
1560 memset(&any, 0, sizeof(any));
1561 any.sa_family = AF_UNSPEC;
1562 result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1564 dprintk("RPC: AF_UNSPEC connect return code %d\n",
1568 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1570 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1572 if (!transport->inet) {
1573 struct sock *sk = sock->sk;
1575 write_lock_bh(&sk->sk_callback_lock);
1577 sk->sk_user_data = xprt;
1578 transport->old_data_ready = sk->sk_data_ready;
1579 transport->old_state_change = sk->sk_state_change;
1580 transport->old_write_space = sk->sk_write_space;
1581 sk->sk_data_ready = xs_tcp_data_ready;
1582 sk->sk_state_change = xs_tcp_state_change;
1583 sk->sk_write_space = xs_tcp_write_space;
1584 sk->sk_allocation = GFP_ATOMIC;
1586 /* socket options */
1587 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1588 sock_reset_flag(sk, SOCK_LINGER);
1589 tcp_sk(sk)->linger2 = 0;
1590 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1592 xprt_clear_connected(xprt);
1594 /* Reset to new socket */
1595 transport->sock = sock;
1596 transport->inet = sk;
1598 write_unlock_bh(&sk->sk_callback_lock);
1601 /* Tell the socket layer to start connecting... */
1602 xprt->stat.connect_count++;
1603 xprt->stat.connect_start = jiffies;
1604 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
1608 * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1609 * @work: RPC transport to connect
1611 * Invoked by a work queue tasklet.
1613 static void xs_tcp_connect_worker4(struct work_struct *work)
1615 struct sock_xprt *transport =
1616 container_of(work, struct sock_xprt, connect_worker.work);
1617 struct rpc_xprt *xprt = &transport->xprt;
1618 struct socket *sock = transport->sock;
1619 int err, status = -EIO;
1621 if (xprt->shutdown || !xprt_bound(xprt))
1625 /* start from scratch */
1626 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1627 dprintk("RPC: can't create TCP transport socket (%d).\n", -err);
1630 xs_reclassify_socket4(sock);
1632 if (xs_bind4(transport, sock) < 0) {
1637 /* "close" the socket, preserving the local port */
1638 xs_tcp_reuse_connection(xprt);
1640 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1641 xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1643 status = xs_tcp_finish_connecting(xprt, sock);
1644 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
1645 xprt, -status, xprt_connected(xprt),
1646 sock->sk->sk_state);
1654 /* retry with existing socket, after a delay */
1657 /* get rid of existing socket, and retry */
1658 xs_tcp_shutdown(xprt);
1662 xprt_wake_pending_tasks(xprt, status);
1664 xprt_clear_connecting(xprt);
1668 * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
1669 * @work: RPC transport to connect
1671 * Invoked by a work queue tasklet.
1673 static void xs_tcp_connect_worker6(struct work_struct *work)
1675 struct sock_xprt *transport =
1676 container_of(work, struct sock_xprt, connect_worker.work);
1677 struct rpc_xprt *xprt = &transport->xprt;
1678 struct socket *sock = transport->sock;
1679 int err, status = -EIO;
1681 if (xprt->shutdown || !xprt_bound(xprt))
1685 /* start from scratch */
1686 if ((err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1687 dprintk("RPC: can't create TCP transport socket (%d).\n", -err);
1690 xs_reclassify_socket6(sock);
1692 if (xs_bind6(transport, sock) < 0) {
1697 /* "close" the socket, preserving the local port */
1698 xs_tcp_reuse_connection(xprt);
1700 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1701 xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1703 status = xs_tcp_finish_connecting(xprt, sock);
1704 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
1705 xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1713 /* retry with existing socket, after a delay */
1716 /* get rid of existing socket, and retry */
1717 xs_tcp_shutdown(xprt);
1721 xprt_wake_pending_tasks(xprt, status);
1723 xprt_clear_connecting(xprt);
1727 * xs_connect - connect a socket to a remote endpoint
1728 * @task: address of RPC task that manages state of connect request
1730 * TCP: If the remote end dropped the connection, delay reconnecting.
1732 * UDP socket connects are synchronous, but we use a work queue anyway
1733 * to guarantee that even unprivileged user processes can set up a
1734 * socket on a privileged port.
1736 * If a UDP socket connect fails, the delay behavior here prevents
1737 * retry floods (hard mounts).
1739 static void xs_connect(struct rpc_task *task)
1741 struct rpc_xprt *xprt = task->tk_xprt;
1742 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1744 if (xprt_test_and_set_connecting(xprt))
1747 if (transport->sock != NULL) {
1748 dprintk("RPC: xs_connect delayed xprt %p for %lu "
1750 xprt, xprt->reestablish_timeout / HZ);
1751 queue_delayed_work(rpciod_workqueue,
1752 &transport->connect_worker,
1753 xprt->reestablish_timeout);
1754 xprt->reestablish_timeout <<= 1;
1755 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1756 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1758 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
1759 queue_delayed_work(rpciod_workqueue,
1760 &transport->connect_worker, 0);
1764 static void xs_tcp_connect(struct rpc_task *task)
1766 struct rpc_xprt *xprt = task->tk_xprt;
1768 /* Initiate graceful shutdown of the socket if not already done */
1769 if (test_bit(XPRT_CONNECTED, &xprt->state))
1770 xs_tcp_shutdown(xprt);
1771 /* Exit if we need to wait for socket shutdown to complete */
1772 if (test_bit(XPRT_CLOSING, &xprt->state))
1778 * xs_udp_print_stats - display UDP socket-specifc stats
1779 * @xprt: rpc_xprt struct containing statistics
1783 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1785 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1787 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1789 xprt->stat.bind_count,
1792 xprt->stat.bad_xids,
1794 xprt->stat.bklog_u);
1798 * xs_tcp_print_stats - display TCP socket-specifc stats
1799 * @xprt: rpc_xprt struct containing statistics
1803 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1805 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1808 if (xprt_connected(xprt))
1809 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1811 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1813 xprt->stat.bind_count,
1814 xprt->stat.connect_count,
1815 xprt->stat.connect_time,
1819 xprt->stat.bad_xids,
1821 xprt->stat.bklog_u);
1824 static struct rpc_xprt_ops xs_udp_ops = {
1825 .set_buffer_size = xs_udp_set_buffer_size,
1826 .reserve_xprt = xprt_reserve_xprt_cong,
1827 .release_xprt = xprt_release_xprt_cong,
1828 .rpcbind = rpcb_getport_async,
1829 .set_port = xs_set_port,
1830 .connect = xs_connect,
1831 .buf_alloc = rpc_malloc,
1832 .buf_free = rpc_free,
1833 .send_request = xs_udp_send_request,
1834 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
1835 .timer = xs_udp_timer,
1836 .release_request = xprt_release_rqst_cong,
1838 .destroy = xs_destroy,
1839 .print_stats = xs_udp_print_stats,
1842 static struct rpc_xprt_ops xs_tcp_ops = {
1843 .reserve_xprt = xprt_reserve_xprt,
1844 .release_xprt = xs_tcp_release_xprt,
1845 .rpcbind = rpcb_getport_async,
1846 .set_port = xs_set_port,
1847 .connect = xs_tcp_connect,
1848 .buf_alloc = rpc_malloc,
1849 .buf_free = rpc_free,
1850 .send_request = xs_tcp_send_request,
1851 .set_retrans_timeout = xprt_set_retrans_timeout_def,
1852 .close = xs_tcp_shutdown,
1853 .destroy = xs_destroy,
1854 .print_stats = xs_tcp_print_stats,
1857 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
1858 unsigned int slot_table_size)
1860 struct rpc_xprt *xprt;
1861 struct sock_xprt *new;
1863 if (args->addrlen > sizeof(xprt->addr)) {
1864 dprintk("RPC: xs_setup_xprt: address too large\n");
1865 return ERR_PTR(-EBADF);
1868 new = kzalloc(sizeof(*new), GFP_KERNEL);
1870 dprintk("RPC: xs_setup_xprt: couldn't allocate "
1872 return ERR_PTR(-ENOMEM);
1876 xprt->max_reqs = slot_table_size;
1877 xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1878 if (xprt->slot == NULL) {
1880 dprintk("RPC: xs_setup_xprt: couldn't allocate slot "
1882 return ERR_PTR(-ENOMEM);
1885 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
1886 xprt->addrlen = args->addrlen;
1888 memcpy(&new->addr, args->srcaddr, args->addrlen);
1893 static const struct rpc_timeout xs_udp_default_timeout = {
1894 .to_initval = 5 * HZ,
1895 .to_maxval = 30 * HZ,
1896 .to_increment = 5 * HZ,
1901 * xs_setup_udp - Set up transport to use a UDP socket
1902 * @args: rpc transport creation arguments
1905 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
1907 struct sockaddr *addr = args->dstaddr;
1908 struct rpc_xprt *xprt;
1909 struct sock_xprt *transport;
1911 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
1914 transport = container_of(xprt, struct sock_xprt, xprt);
1916 xprt->prot = IPPROTO_UDP;
1918 /* XXX: header size can vary due to auth type, IPv6, etc. */
1919 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1921 xprt->bind_timeout = XS_BIND_TO;
1922 xprt->connect_timeout = XS_UDP_CONN_TO;
1923 xprt->reestablish_timeout = XS_UDP_REEST_TO;
1924 xprt->idle_timeout = XS_IDLE_DISC_TO;
1926 xprt->ops = &xs_udp_ops;
1928 xprt->timeout = &xs_udp_default_timeout;
1930 switch (addr->sa_family) {
1932 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1933 xprt_set_bound(xprt);
1935 INIT_DELAYED_WORK(&transport->connect_worker,
1936 xs_udp_connect_worker4);
1937 xs_format_ipv4_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
1940 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
1941 xprt_set_bound(xprt);
1943 INIT_DELAYED_WORK(&transport->connect_worker,
1944 xs_udp_connect_worker6);
1945 xs_format_ipv6_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
1949 return ERR_PTR(-EAFNOSUPPORT);
1952 dprintk("RPC: set up transport to address %s\n",
1953 xprt->address_strings[RPC_DISPLAY_ALL]);
1955 if (try_module_get(THIS_MODULE))
1960 return ERR_PTR(-EINVAL);
1963 static const struct rpc_timeout xs_tcp_default_timeout = {
1964 .to_initval = 60 * HZ,
1965 .to_maxval = 60 * HZ,
1970 * xs_setup_tcp - Set up transport to use a TCP socket
1971 * @args: rpc transport creation arguments
1974 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
1976 struct sockaddr *addr = args->dstaddr;
1977 struct rpc_xprt *xprt;
1978 struct sock_xprt *transport;
1980 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
1983 transport = container_of(xprt, struct sock_xprt, xprt);
1985 xprt->prot = IPPROTO_TCP;
1986 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
1987 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
1989 xprt->bind_timeout = XS_BIND_TO;
1990 xprt->connect_timeout = XS_TCP_CONN_TO;
1991 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1992 xprt->idle_timeout = XS_IDLE_DISC_TO;
1994 xprt->ops = &xs_tcp_ops;
1995 xprt->timeout = &xs_tcp_default_timeout;
1997 switch (addr->sa_family) {
1999 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2000 xprt_set_bound(xprt);
2002 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker4);
2003 xs_format_ipv4_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2006 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2007 xprt_set_bound(xprt);
2009 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker6);
2010 xs_format_ipv6_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2014 return ERR_PTR(-EAFNOSUPPORT);
2017 dprintk("RPC: set up transport to address %s\n",
2018 xprt->address_strings[RPC_DISPLAY_ALL]);
2020 if (try_module_get(THIS_MODULE))
2025 return ERR_PTR(-EINVAL);
2028 static struct xprt_class xs_udp_transport = {
2029 .list = LIST_HEAD_INIT(xs_udp_transport.list),
2031 .owner = THIS_MODULE,
2032 .ident = IPPROTO_UDP,
2033 .setup = xs_setup_udp,
2036 static struct xprt_class xs_tcp_transport = {
2037 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
2039 .owner = THIS_MODULE,
2040 .ident = IPPROTO_TCP,
2041 .setup = xs_setup_tcp,
2045 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2048 int init_socket_xprt(void)
2051 if (!sunrpc_table_header)
2052 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2055 xprt_register_transport(&xs_udp_transport);
2056 xprt_register_transport(&xs_tcp_transport);
2062 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2065 void cleanup_socket_xprt(void)
2068 if (sunrpc_table_header) {
2069 unregister_sysctl_table(sunrpc_table_header);
2070 sunrpc_table_header = NULL;
2074 xprt_unregister_transport(&xs_udp_transport);
2075 xprt_unregister_transport(&xs_tcp_transport);