2 * linux/net/sunrpc/xprtsock.c
4 * Client-side transport implementation for sockets.
6 * TCP callback races fixes (C) 1998 Red Hat Software <alan@redhat.com>
7 * TCP send fixes (C) 1998 Red Hat Software <alan@redhat.com>
8 * TCP NFS related read + write fixes
9 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
11 * Rewrite of larges part of the code in order to stabilize TCP stuff.
12 * Fix behaviour when socket buffer is full.
13 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
15 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
17 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
18 * <gilles.quillard@bull.net>
21 #include <linux/types.h>
22 #include <linux/slab.h>
23 #include <linux/module.h>
24 #include <linux/capability.h>
25 #include <linux/pagemap.h>
26 #include <linux/errno.h>
27 #include <linux/socket.h>
29 #include <linux/net.h>
31 #include <linux/udp.h>
32 #include <linux/tcp.h>
33 #include <linux/sunrpc/clnt.h>
34 #include <linux/sunrpc/sched.h>
35 #include <linux/sunrpc/xprtsock.h>
36 #include <linux/file.h>
39 #include <net/checksum.h>
46 unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
47 unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE;
49 unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
50 unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
53 * We can register our own files under /proc/sys/sunrpc by
54 * calling register_sysctl_table() again. The files in that
55 * directory become the union of all files registered there.
57 * We simply need to make sure that we don't collide with
58 * someone else's file names!
63 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
64 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
65 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
66 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
68 static struct ctl_table_header *sunrpc_table_header;
71 * FIXME: changing the UDP slot table size should also resize the UDP
72 * socket buffers for existing UDP transports
74 static ctl_table xs_tunables_table[] = {
76 .ctl_name = CTL_SLOTTABLE_UDP,
77 .procname = "udp_slot_table_entries",
78 .data = &xprt_udp_slot_table_entries,
79 .maxlen = sizeof(unsigned int),
81 .proc_handler = &proc_dointvec_minmax,
82 .strategy = &sysctl_intvec,
83 .extra1 = &min_slot_table_size,
84 .extra2 = &max_slot_table_size
87 .ctl_name = CTL_SLOTTABLE_TCP,
88 .procname = "tcp_slot_table_entries",
89 .data = &xprt_tcp_slot_table_entries,
90 .maxlen = sizeof(unsigned int),
92 .proc_handler = &proc_dointvec_minmax,
93 .strategy = &sysctl_intvec,
94 .extra1 = &min_slot_table_size,
95 .extra2 = &max_slot_table_size
98 .ctl_name = CTL_MIN_RESVPORT,
99 .procname = "min_resvport",
100 .data = &xprt_min_resvport,
101 .maxlen = sizeof(unsigned int),
103 .proc_handler = &proc_dointvec_minmax,
104 .strategy = &sysctl_intvec,
105 .extra1 = &xprt_min_resvport_limit,
106 .extra2 = &xprt_max_resvport_limit
109 .ctl_name = CTL_MAX_RESVPORT,
110 .procname = "max_resvport",
111 .data = &xprt_max_resvport,
112 .maxlen = sizeof(unsigned int),
114 .proc_handler = &proc_dointvec_minmax,
115 .strategy = &sysctl_intvec,
116 .extra1 = &xprt_min_resvport_limit,
117 .extra2 = &xprt_max_resvport_limit
124 static ctl_table sunrpc_table[] = {
126 .ctl_name = CTL_SUNRPC,
127 .procname = "sunrpc",
129 .child = xs_tunables_table
139 * How many times to try sending a request on a socket before waiting
140 * for the socket buffer to clear.
142 #define XS_SENDMSG_RETRY (10U)
145 * Time out for an RPC UDP socket connect. UDP socket connects are
146 * synchronous, but we set a timeout anyway in case of resource
147 * exhaustion on the local host.
149 #define XS_UDP_CONN_TO (5U * HZ)
152 * Wait duration for an RPC TCP connection to be established. Solaris
153 * NFS over TCP uses 60 seconds, for example, which is in line with how
154 * long a server takes to reboot.
156 #define XS_TCP_CONN_TO (60U * HZ)
159 * Wait duration for a reply from the RPC portmapper.
161 #define XS_BIND_TO (60U * HZ)
164 * Delay if a UDP socket connect error occurs. This is most likely some
165 * kind of resource problem on the local host.
167 #define XS_UDP_REEST_TO (2U * HZ)
170 * The reestablish timeout allows clients to delay for a bit before attempting
171 * to reconnect to a server that just dropped our connection.
173 * We implement an exponential backoff when trying to reestablish a TCP
174 * transport connection with the server. Some servers like to drop a TCP
175 * connection when they are overworked, so we start with a short timeout and
176 * increase over time if the server is down or not responding.
178 #define XS_TCP_INIT_REEST_TO (3U * HZ)
179 #define XS_TCP_MAX_REEST_TO (5U * 60 * HZ)
182 * TCP idle timeout; client drops the transport socket if it is idle
183 * for this long. Note that we also timeout UDP sockets to prevent
184 * holding port numbers when there is no RPC traffic.
186 #define XS_IDLE_DISC_TO (5U * 60 * HZ)
189 # undef RPC_DEBUG_DATA
190 # define RPCDBG_FACILITY RPCDBG_TRANS
193 #ifdef RPC_DEBUG_DATA
194 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
196 u8 *buf = (u8 *) packet;
199 dprintk("RPC: %s\n", msg);
200 for (j = 0; j < count && j < 128; j += 4) {
204 dprintk("0x%04x ", j);
206 dprintk("%02x%02x%02x%02x ",
207 buf[j], buf[j+1], buf[j+2], buf[j+3]);
212 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
219 struct rpc_xprt xprt;
224 struct socket * sock;
228 * State of TCP reply receive
236 unsigned long tcp_copied,
240 * Connection of transports
242 struct delayed_work connect_worker;
243 struct sockaddr_storage addr;
247 * UDP socket buffer size parameters
253 * Saved socket callback addresses
255 void (*old_data_ready)(struct sock *, int);
256 void (*old_state_change)(struct sock *);
257 void (*old_write_space)(struct sock *);
261 * TCP receive state flags
263 #define TCP_RCV_LAST_FRAG (1UL << 0)
264 #define TCP_RCV_COPY_FRAGHDR (1UL << 1)
265 #define TCP_RCV_COPY_XID (1UL << 2)
266 #define TCP_RCV_COPY_DATA (1UL << 3)
268 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
270 return (struct sockaddr *) &xprt->addr;
273 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
275 return (struct sockaddr_in *) &xprt->addr;
278 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
280 return (struct sockaddr_in6 *) &xprt->addr;
283 static void xs_format_ipv4_peer_addresses(struct rpc_xprt *xprt)
285 struct sockaddr_in *addr = xs_addr_in(xprt);
288 buf = kzalloc(20, GFP_KERNEL);
290 snprintf(buf, 20, NIPQUAD_FMT,
291 NIPQUAD(addr->sin_addr.s_addr));
293 xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
295 buf = kzalloc(8, GFP_KERNEL);
297 snprintf(buf, 8, "%u",
298 ntohs(addr->sin_port));
300 xprt->address_strings[RPC_DISPLAY_PORT] = buf;
302 buf = kzalloc(8, GFP_KERNEL);
304 if (xprt->prot == IPPROTO_UDP)
305 snprintf(buf, 8, "udp");
307 snprintf(buf, 8, "tcp");
309 xprt->address_strings[RPC_DISPLAY_PROTO] = buf;
311 buf = kzalloc(48, GFP_KERNEL);
313 snprintf(buf, 48, "addr="NIPQUAD_FMT" port=%u proto=%s",
314 NIPQUAD(addr->sin_addr.s_addr),
315 ntohs(addr->sin_port),
316 xprt->prot == IPPROTO_UDP ? "udp" : "tcp");
318 xprt->address_strings[RPC_DISPLAY_ALL] = buf;
320 buf = kzalloc(10, GFP_KERNEL);
322 snprintf(buf, 10, "%02x%02x%02x%02x",
323 NIPQUAD(addr->sin_addr.s_addr));
325 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
327 buf = kzalloc(8, GFP_KERNEL);
329 snprintf(buf, 8, "%4hx",
330 ntohs(addr->sin_port));
332 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
334 buf = kzalloc(30, GFP_KERNEL);
336 snprintf(buf, 30, NIPQUAD_FMT".%u.%u",
337 NIPQUAD(addr->sin_addr.s_addr),
338 ntohs(addr->sin_port) >> 8,
339 ntohs(addr->sin_port) & 0xff);
341 xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
343 xprt->address_strings[RPC_DISPLAY_NETID] =
344 kstrdup(xprt->prot == IPPROTO_UDP ?
345 RPCBIND_NETID_UDP : RPCBIND_NETID_TCP, GFP_KERNEL);
348 static void xs_format_ipv6_peer_addresses(struct rpc_xprt *xprt)
350 struct sockaddr_in6 *addr = xs_addr_in6(xprt);
353 buf = kzalloc(40, GFP_KERNEL);
355 snprintf(buf, 40, NIP6_FMT,
356 NIP6(addr->sin6_addr));
358 xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
360 buf = kzalloc(8, GFP_KERNEL);
362 snprintf(buf, 8, "%u",
363 ntohs(addr->sin6_port));
365 xprt->address_strings[RPC_DISPLAY_PORT] = buf;
367 buf = kzalloc(8, GFP_KERNEL);
369 if (xprt->prot == IPPROTO_UDP)
370 snprintf(buf, 8, "udp");
372 snprintf(buf, 8, "tcp");
374 xprt->address_strings[RPC_DISPLAY_PROTO] = buf;
376 buf = kzalloc(64, GFP_KERNEL);
378 snprintf(buf, 64, "addr="NIP6_FMT" port=%u proto=%s",
379 NIP6(addr->sin6_addr),
380 ntohs(addr->sin6_port),
381 xprt->prot == IPPROTO_UDP ? "udp" : "tcp");
383 xprt->address_strings[RPC_DISPLAY_ALL] = buf;
385 buf = kzalloc(36, GFP_KERNEL);
387 snprintf(buf, 36, NIP6_SEQFMT,
388 NIP6(addr->sin6_addr));
390 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
392 buf = kzalloc(8, GFP_KERNEL);
394 snprintf(buf, 8, "%4hx",
395 ntohs(addr->sin6_port));
397 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
399 buf = kzalloc(50, GFP_KERNEL);
401 snprintf(buf, 50, NIP6_FMT".%u.%u",
402 NIP6(addr->sin6_addr),
403 ntohs(addr->sin6_port) >> 8,
404 ntohs(addr->sin6_port) & 0xff);
406 xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
408 xprt->address_strings[RPC_DISPLAY_NETID] =
409 kstrdup(xprt->prot == IPPROTO_UDP ?
410 RPCBIND_NETID_UDP6 : RPCBIND_NETID_TCP6, GFP_KERNEL);
413 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
417 for (i = 0; i < RPC_DISPLAY_MAX; i++)
418 kfree(xprt->address_strings[i]);
421 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
423 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
425 struct msghdr msg = {
427 .msg_namelen = addrlen,
428 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
431 .iov_base = vec->iov_base + base,
432 .iov_len = vec->iov_len - base,
435 if (iov.iov_len != 0)
436 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
437 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
440 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
443 unsigned int remainder;
446 remainder = xdr->page_len - base;
447 base += xdr->page_base;
448 ppage = xdr->pages + (base >> PAGE_SHIFT);
451 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
452 int flags = XS_SENDMSG_FLAGS;
455 if (remainder != 0 || more)
457 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
458 if (remainder == 0 || err != len)
472 * xs_sendpages - write pages directly to a socket
473 * @sock: socket to send on
474 * @addr: UDP only -- address of destination
475 * @addrlen: UDP only -- length of destination address
476 * @xdr: buffer containing this request
477 * @base: starting position in the buffer
480 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
482 unsigned int remainder = xdr->len - base;
488 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
494 if (base < xdr->head[0].iov_len || addr != NULL) {
495 unsigned int len = xdr->head[0].iov_len - base;
497 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
498 if (remainder == 0 || err != len)
503 base -= xdr->head[0].iov_len;
505 if (base < xdr->page_len) {
506 unsigned int len = xdr->page_len - base;
508 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
509 if (remainder == 0 || err != len)
514 base -= xdr->page_len;
516 if (base >= xdr->tail[0].iov_len)
518 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
528 * xs_nospace - place task on wait queue if transmit was incomplete
529 * @task: task to put to sleep
532 static void xs_nospace(struct rpc_task *task)
534 struct rpc_rqst *req = task->tk_rqstp;
535 struct rpc_xprt *xprt = req->rq_xprt;
536 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
538 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
539 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
542 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
543 /* Protect against races with write_space */
544 spin_lock_bh(&xprt->transport_lock);
546 /* Don't race with disconnect */
547 if (!xprt_connected(xprt))
548 task->tk_status = -ENOTCONN;
549 else if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
550 xprt_wait_for_buffer_space(task);
552 spin_unlock_bh(&xprt->transport_lock);
554 /* Keep holding the socket if it is blocked */
555 rpc_delay(task, HZ>>4);
559 * xs_udp_send_request - write an RPC request to a UDP socket
560 * @task: address of RPC task that manages the state of an RPC request
563 * 0: The request has been sent
564 * EAGAIN: The socket was blocked, please call again later to
565 * complete the request
566 * ENOTCONN: Caller needs to invoke connect logic then call again
567 * other: Some other error occured, the request was not sent
569 static int xs_udp_send_request(struct rpc_task *task)
571 struct rpc_rqst *req = task->tk_rqstp;
572 struct rpc_xprt *xprt = req->rq_xprt;
573 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
574 struct xdr_buf *xdr = &req->rq_snd_buf;
577 xs_pktdump("packet data:",
578 req->rq_svec->iov_base,
579 req->rq_svec->iov_len);
581 req->rq_xtime = jiffies;
582 status = xs_sendpages(transport->sock,
587 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
588 xdr->len - req->rq_bytes_sent, status);
591 task->tk_bytes_sent += status;
592 if (status >= req->rq_slen)
594 /* Still some bytes left; set up for a retry later. */
602 /* When the server has died, an ICMP port unreachable message
603 * prompts ECONNREFUSED. */
609 dprintk("RPC: sendmsg returned unrecognized error %d\n",
618 * xs_tcp_shutdown - gracefully shut down a TCP socket
621 * Initiates a graceful shutdown of the TCP socket by calling the
622 * equivalent of shutdown(SHUT_WR);
624 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
626 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
627 struct socket *sock = transport->sock;
630 kernel_sock_shutdown(sock, SHUT_WR);
633 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
635 u32 reclen = buf->len - sizeof(rpc_fraghdr);
636 rpc_fraghdr *base = buf->head[0].iov_base;
637 *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
641 * xs_tcp_send_request - write an RPC request to a TCP socket
642 * @task: address of RPC task that manages the state of an RPC request
645 * 0: The request has been sent
646 * EAGAIN: The socket was blocked, please call again later to
647 * complete the request
648 * ENOTCONN: Caller needs to invoke connect logic then call again
649 * other: Some other error occured, the request was not sent
651 * XXX: In the case of soft timeouts, should we eventually give up
652 * if sendmsg is not able to make progress?
654 static int xs_tcp_send_request(struct rpc_task *task)
656 struct rpc_rqst *req = task->tk_rqstp;
657 struct rpc_xprt *xprt = req->rq_xprt;
658 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
659 struct xdr_buf *xdr = &req->rq_snd_buf;
661 unsigned int retry = 0;
663 xs_encode_tcp_record_marker(&req->rq_snd_buf);
665 xs_pktdump("packet data:",
666 req->rq_svec->iov_base,
667 req->rq_svec->iov_len);
669 /* Continue transmitting the packet/record. We must be careful
670 * to cope with writespace callbacks arriving _after_ we have
671 * called sendmsg(). */
673 req->rq_xtime = jiffies;
674 status = xs_sendpages(transport->sock,
675 NULL, 0, xdr, req->rq_bytes_sent);
677 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
678 xdr->len - req->rq_bytes_sent, status);
680 if (unlikely(status < 0))
683 /* If we've sent the entire packet, immediately
684 * reset the count of bytes sent. */
685 req->rq_bytes_sent += status;
686 task->tk_bytes_sent += status;
687 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
688 req->rq_bytes_sent = 0;
693 if (retry++ > XS_SENDMSG_RETRY)
708 dprintk("RPC: sendmsg returned unrecognized error %d\n",
710 xs_tcp_shutdown(xprt);
718 * xs_tcp_release_xprt - clean up after a tcp transmission
722 * This cleans up if an error causes us to abort the transmission of a request.
723 * In this case, the socket may need to be reset in order to avoid confusing
726 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
728 struct rpc_rqst *req;
730 if (task != xprt->snd_task)
734 req = task->tk_rqstp;
735 if (req->rq_bytes_sent == 0)
737 if (req->rq_bytes_sent == req->rq_snd_buf.len)
739 set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
741 xprt_release_xprt(xprt, task);
745 * xs_close - close a socket
748 * This is used when all requests are complete; ie, no DRC state remains
749 * on the server we want to save.
751 static void xs_close(struct rpc_xprt *xprt)
753 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
754 struct socket *sock = transport->sock;
755 struct sock *sk = transport->inet;
758 goto clear_close_wait;
760 dprintk("RPC: xs_close xprt %p\n", xprt);
762 write_lock_bh(&sk->sk_callback_lock);
763 transport->inet = NULL;
764 transport->sock = NULL;
766 sk->sk_user_data = NULL;
767 sk->sk_data_ready = transport->old_data_ready;
768 sk->sk_state_change = transport->old_state_change;
769 sk->sk_write_space = transport->old_write_space;
770 write_unlock_bh(&sk->sk_callback_lock);
776 smp_mb__before_clear_bit();
777 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
778 clear_bit(XPRT_CLOSING, &xprt->state);
779 smp_mb__after_clear_bit();
780 xprt_disconnect(xprt);
784 * xs_destroy - prepare to shutdown a transport
785 * @xprt: doomed transport
788 static void xs_destroy(struct rpc_xprt *xprt)
790 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
792 dprintk("RPC: xs_destroy xprt %p\n", xprt);
794 cancel_rearming_delayed_work(&transport->connect_worker);
797 xs_free_peer_addresses(xprt);
800 module_put(THIS_MODULE);
803 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
805 return (struct rpc_xprt *) sk->sk_user_data;
809 * xs_udp_data_ready - "data ready" callback for UDP sockets
810 * @sk: socket with data to read
811 * @len: how much data to read
814 static void xs_udp_data_ready(struct sock *sk, int len)
816 struct rpc_task *task;
817 struct rpc_xprt *xprt;
818 struct rpc_rqst *rovr;
820 int err, repsize, copied;
824 read_lock(&sk->sk_callback_lock);
825 dprintk("RPC: xs_udp_data_ready...\n");
826 if (!(xprt = xprt_from_sock(sk)))
829 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
835 repsize = skb->len - sizeof(struct udphdr);
837 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
841 /* Copy the XID from the skb... */
842 xp = skb_header_pointer(skb, sizeof(struct udphdr),
843 sizeof(_xid), &_xid);
847 /* Look up and lock the request corresponding to the given XID */
848 spin_lock(&xprt->transport_lock);
849 rovr = xprt_lookup_rqst(xprt, *xp);
852 task = rovr->rq_task;
854 if ((copied = rovr->rq_private_buf.buflen) > repsize)
857 /* Suck it into the iovec, verify checksum if not done by hw. */
858 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
859 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
863 UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
865 /* Something worked... */
866 dst_confirm(skb->dst);
868 xprt_adjust_cwnd(task, copied);
869 xprt_update_rtt(task);
870 xprt_complete_rqst(task, copied);
873 spin_unlock(&xprt->transport_lock);
875 skb_free_datagram(sk, skb);
877 read_unlock(&sk->sk_callback_lock);
880 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
882 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
886 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
887 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
888 used = xdr_skb_read_bits(desc, p, len);
889 transport->tcp_offset += used;
893 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
894 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
895 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
897 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
898 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
900 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
901 transport->tcp_offset = 0;
903 /* Sanity check of the record length */
904 if (unlikely(transport->tcp_reclen < 4)) {
905 dprintk("RPC: invalid TCP record fragment length\n");
906 xprt_disconnect(xprt);
909 dprintk("RPC: reading TCP record fragment of length %d\n",
910 transport->tcp_reclen);
913 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
915 if (transport->tcp_offset == transport->tcp_reclen) {
916 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
917 transport->tcp_offset = 0;
918 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
919 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
920 transport->tcp_flags |= TCP_RCV_COPY_XID;
921 transport->tcp_copied = 0;
926 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
931 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
932 dprintk("RPC: reading XID (%Zu bytes)\n", len);
933 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
934 used = xdr_skb_read_bits(desc, p, len);
935 transport->tcp_offset += used;
938 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
939 transport->tcp_flags |= TCP_RCV_COPY_DATA;
940 transport->tcp_copied = 4;
941 dprintk("RPC: reading reply for XID %08x\n",
942 ntohl(transport->tcp_xid));
943 xs_tcp_check_fraghdr(transport);
946 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
948 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
949 struct rpc_rqst *req;
950 struct xdr_buf *rcvbuf;
954 /* Find and lock the request corresponding to this xid */
955 spin_lock(&xprt->transport_lock);
956 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
958 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
959 dprintk("RPC: XID %08x request not found!\n",
960 ntohl(transport->tcp_xid));
961 spin_unlock(&xprt->transport_lock);
965 rcvbuf = &req->rq_private_buf;
967 if (len > transport->tcp_reclen - transport->tcp_offset) {
968 struct xdr_skb_reader my_desc;
970 len = transport->tcp_reclen - transport->tcp_offset;
971 memcpy(&my_desc, desc, sizeof(my_desc));
973 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
974 &my_desc, xdr_skb_read_bits);
978 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
979 desc, xdr_skb_read_bits);
982 transport->tcp_copied += r;
983 transport->tcp_offset += r;
986 /* Error when copying to the receive buffer,
987 * usually because we weren't able to allocate
988 * additional buffer pages. All we can do now
989 * is turn off TCP_RCV_COPY_DATA, so the request
990 * will not receive any additional updates,
992 * Any remaining data from this record will
995 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
996 dprintk("RPC: XID %08x truncated request\n",
997 ntohl(transport->tcp_xid));
998 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
999 "tcp_offset = %u, tcp_reclen = %u\n",
1000 xprt, transport->tcp_copied,
1001 transport->tcp_offset, transport->tcp_reclen);
1005 dprintk("RPC: XID %08x read %Zd bytes\n",
1006 ntohl(transport->tcp_xid), r);
1007 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1008 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1009 transport->tcp_offset, transport->tcp_reclen);
1011 if (transport->tcp_copied == req->rq_private_buf.buflen)
1012 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1013 else if (transport->tcp_offset == transport->tcp_reclen) {
1014 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1015 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1019 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1020 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1021 spin_unlock(&xprt->transport_lock);
1022 xs_tcp_check_fraghdr(transport);
1025 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1029 len = transport->tcp_reclen - transport->tcp_offset;
1030 if (len > desc->count)
1033 desc->offset += len;
1034 transport->tcp_offset += len;
1035 dprintk("RPC: discarded %Zu bytes\n", len);
1036 xs_tcp_check_fraghdr(transport);
1039 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1041 struct rpc_xprt *xprt = rd_desc->arg.data;
1042 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1043 struct xdr_skb_reader desc = {
1049 dprintk("RPC: xs_tcp_data_recv started\n");
1051 /* Read in a new fragment marker if necessary */
1052 /* Can we ever really expect to get completely empty fragments? */
1053 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1054 xs_tcp_read_fraghdr(xprt, &desc);
1057 /* Read in the xid if necessary */
1058 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1059 xs_tcp_read_xid(transport, &desc);
1062 /* Read in the request data */
1063 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1064 xs_tcp_read_request(xprt, &desc);
1067 /* Skip over any trailing bytes on short reads */
1068 xs_tcp_read_discard(transport, &desc);
1069 } while (desc.count);
1070 dprintk("RPC: xs_tcp_data_recv done\n");
1071 return len - desc.count;
1075 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1076 * @sk: socket with data to read
1077 * @bytes: how much data to read
1080 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1082 struct rpc_xprt *xprt;
1083 read_descriptor_t rd_desc;
1085 dprintk("RPC: xs_tcp_data_ready...\n");
1087 read_lock(&sk->sk_callback_lock);
1088 if (!(xprt = xprt_from_sock(sk)))
1093 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1094 rd_desc.arg.data = xprt;
1095 rd_desc.count = 65536;
1096 tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1098 read_unlock(&sk->sk_callback_lock);
1102 * xs_tcp_state_change - callback to handle TCP socket state changes
1103 * @sk: socket whose state has changed
1106 static void xs_tcp_state_change(struct sock *sk)
1108 struct rpc_xprt *xprt;
1110 read_lock(&sk->sk_callback_lock);
1111 if (!(xprt = xprt_from_sock(sk)))
1113 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1114 dprintk("RPC: state %x conn %d dead %d zapped %d\n",
1115 sk->sk_state, xprt_connected(xprt),
1116 sock_flag(sk, SOCK_DEAD),
1117 sock_flag(sk, SOCK_ZAPPED));
1119 switch (sk->sk_state) {
1120 case TCP_ESTABLISHED:
1121 spin_lock_bh(&xprt->transport_lock);
1122 if (!xprt_test_and_set_connected(xprt)) {
1123 struct sock_xprt *transport = container_of(xprt,
1124 struct sock_xprt, xprt);
1126 /* Reset TCP record info */
1127 transport->tcp_offset = 0;
1128 transport->tcp_reclen = 0;
1129 transport->tcp_copied = 0;
1130 transport->tcp_flags =
1131 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1133 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1134 xprt_wake_pending_tasks(xprt, 0);
1136 spin_unlock_bh(&xprt->transport_lock);
1139 /* The client initiated a shutdown of the socket */
1140 set_bit(XPRT_CLOSING, &xprt->state);
1141 smp_mb__before_clear_bit();
1142 clear_bit(XPRT_CONNECTED, &xprt->state);
1143 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1144 smp_mb__after_clear_bit();
1146 case TCP_CLOSE_WAIT:
1147 /* The server initiated a shutdown of the socket */
1148 set_bit(XPRT_CLOSING, &xprt->state);
1149 xprt_force_disconnect(xprt);
1152 smp_mb__before_clear_bit();
1153 clear_bit(XPRT_CONNECTED, &xprt->state);
1154 smp_mb__after_clear_bit();
1157 smp_mb__before_clear_bit();
1158 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1159 clear_bit(XPRT_CLOSING, &xprt->state);
1160 smp_mb__after_clear_bit();
1161 /* Mark transport as closed and wake up all pending tasks */
1162 xprt_disconnect(xprt);
1165 read_unlock(&sk->sk_callback_lock);
1169 * xs_udp_write_space - callback invoked when socket buffer space
1171 * @sk: socket whose state has changed
1173 * Called when more output buffer space is available for this socket.
1174 * We try not to wake our writers until they can make "significant"
1175 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1176 * with a bunch of small requests.
1178 static void xs_udp_write_space(struct sock *sk)
1180 read_lock(&sk->sk_callback_lock);
1182 /* from net/core/sock.c:sock_def_write_space */
1183 if (sock_writeable(sk)) {
1184 struct socket *sock;
1185 struct rpc_xprt *xprt;
1187 if (unlikely(!(sock = sk->sk_socket)))
1189 if (unlikely(!(xprt = xprt_from_sock(sk))))
1191 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
1194 xprt_write_space(xprt);
1198 read_unlock(&sk->sk_callback_lock);
1202 * xs_tcp_write_space - callback invoked when socket buffer space
1204 * @sk: socket whose state has changed
1206 * Called when more output buffer space is available for this socket.
1207 * We try not to wake our writers until they can make "significant"
1208 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1209 * with a bunch of small requests.
1211 static void xs_tcp_write_space(struct sock *sk)
1213 read_lock(&sk->sk_callback_lock);
1215 /* from net/core/stream.c:sk_stream_write_space */
1216 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
1217 struct socket *sock;
1218 struct rpc_xprt *xprt;
1220 if (unlikely(!(sock = sk->sk_socket)))
1222 if (unlikely(!(xprt = xprt_from_sock(sk))))
1224 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
1227 xprt_write_space(xprt);
1231 read_unlock(&sk->sk_callback_lock);
1234 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1236 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1237 struct sock *sk = transport->inet;
1239 if (transport->rcvsize) {
1240 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1241 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1243 if (transport->sndsize) {
1244 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1245 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1246 sk->sk_write_space(sk);
1251 * xs_udp_set_buffer_size - set send and receive limits
1252 * @xprt: generic transport
1253 * @sndsize: requested size of send buffer, in bytes
1254 * @rcvsize: requested size of receive buffer, in bytes
1256 * Set socket send and receive buffer size limits.
1258 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1260 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1262 transport->sndsize = 0;
1264 transport->sndsize = sndsize + 1024;
1265 transport->rcvsize = 0;
1267 transport->rcvsize = rcvsize + 1024;
1269 xs_udp_do_set_buffer_size(xprt);
1273 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1274 * @task: task that timed out
1276 * Adjust the congestion window after a retransmit timeout has occurred.
1278 static void xs_udp_timer(struct rpc_task *task)
1280 xprt_adjust_cwnd(task, -ETIMEDOUT);
1283 static unsigned short xs_get_random_port(void)
1285 unsigned short range = xprt_max_resvport - xprt_min_resvport;
1286 unsigned short rand = (unsigned short) net_random() % range;
1287 return rand + xprt_min_resvport;
1291 * xs_set_port - reset the port number in the remote endpoint address
1292 * @xprt: generic transport
1293 * @port: new port number
1296 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1298 struct sockaddr *addr = xs_addr(xprt);
1300 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1302 switch (addr->sa_family) {
1304 ((struct sockaddr_in *)addr)->sin_port = htons(port);
1307 ((struct sockaddr_in6 *)addr)->sin6_port = htons(port);
1314 static unsigned short xs_get_srcport(struct sock_xprt *transport, struct socket *sock)
1316 unsigned short port = transport->port;
1318 if (port == 0 && transport->xprt.resvport)
1319 port = xs_get_random_port();
1323 static unsigned short xs_next_srcport(struct sock_xprt *transport, struct socket *sock, unsigned short port)
1325 if (transport->port != 0)
1326 transport->port = 0;
1327 if (!transport->xprt.resvport)
1329 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1330 return xprt_max_resvport;
1334 static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
1336 struct sockaddr_in myaddr = {
1337 .sin_family = AF_INET,
1339 struct sockaddr_in *sa;
1341 unsigned short port = xs_get_srcport(transport, sock);
1342 unsigned short last;
1344 sa = (struct sockaddr_in *)&transport->addr;
1345 myaddr.sin_addr = sa->sin_addr;
1347 myaddr.sin_port = htons(port);
1348 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1353 transport->port = port;
1357 port = xs_next_srcport(transport, sock, port);
1360 } while (err == -EADDRINUSE && nloop != 2);
1361 dprintk("RPC: %s "NIPQUAD_FMT":%u: %s (%d)\n",
1362 __FUNCTION__, NIPQUAD(myaddr.sin_addr),
1363 port, err ? "failed" : "ok", err);
1367 static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1369 struct sockaddr_in6 myaddr = {
1370 .sin6_family = AF_INET6,
1372 struct sockaddr_in6 *sa;
1374 unsigned short port = xs_get_srcport(transport, sock);
1375 unsigned short last;
1377 sa = (struct sockaddr_in6 *)&transport->addr;
1378 myaddr.sin6_addr = sa->sin6_addr;
1380 myaddr.sin6_port = htons(port);
1381 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1386 transport->port = port;
1390 port = xs_next_srcport(transport, sock, port);
1393 } while (err == -EADDRINUSE && nloop != 2);
1394 dprintk("RPC: xs_bind6 "NIP6_FMT":%u: %s (%d)\n",
1395 NIP6(myaddr.sin6_addr), port, err ? "failed" : "ok", err);
1399 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1400 static struct lock_class_key xs_key[2];
1401 static struct lock_class_key xs_slock_key[2];
1403 static inline void xs_reclassify_socket4(struct socket *sock)
1405 struct sock *sk = sock->sk;
1407 BUG_ON(sock_owned_by_user(sk));
1408 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1409 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1412 static inline void xs_reclassify_socket6(struct socket *sock)
1414 struct sock *sk = sock->sk;
1416 BUG_ON(sock_owned_by_user(sk));
1417 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1418 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1421 static inline void xs_reclassify_socket4(struct socket *sock)
1425 static inline void xs_reclassify_socket6(struct socket *sock)
1430 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1432 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1434 if (!transport->inet) {
1435 struct sock *sk = sock->sk;
1437 write_lock_bh(&sk->sk_callback_lock);
1439 sk->sk_user_data = xprt;
1440 transport->old_data_ready = sk->sk_data_ready;
1441 transport->old_state_change = sk->sk_state_change;
1442 transport->old_write_space = sk->sk_write_space;
1443 sk->sk_data_ready = xs_udp_data_ready;
1444 sk->sk_write_space = xs_udp_write_space;
1445 sk->sk_no_check = UDP_CSUM_NORCV;
1446 sk->sk_allocation = GFP_ATOMIC;
1448 xprt_set_connected(xprt);
1450 /* Reset to new socket */
1451 transport->sock = sock;
1452 transport->inet = sk;
1454 write_unlock_bh(&sk->sk_callback_lock);
1456 xs_udp_do_set_buffer_size(xprt);
1460 * xs_udp_connect_worker4 - set up a UDP socket
1461 * @work: RPC transport to connect
1463 * Invoked by a work queue tasklet.
1465 static void xs_udp_connect_worker4(struct work_struct *work)
1467 struct sock_xprt *transport =
1468 container_of(work, struct sock_xprt, connect_worker.work);
1469 struct rpc_xprt *xprt = &transport->xprt;
1470 struct socket *sock = transport->sock;
1471 int err, status = -EIO;
1473 if (xprt->shutdown || !xprt_bound(xprt))
1476 /* Start by resetting any existing state */
1479 if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1480 dprintk("RPC: can't create UDP transport socket (%d).\n", -err);
1483 xs_reclassify_socket4(sock);
1485 if (xs_bind4(transport, sock)) {
1490 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1491 xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1493 xs_udp_finish_connecting(xprt, sock);
1496 xprt_wake_pending_tasks(xprt, status);
1497 xprt_clear_connecting(xprt);
1501 * xs_udp_connect_worker6 - set up a UDP socket
1502 * @work: RPC transport to connect
1504 * Invoked by a work queue tasklet.
1506 static void xs_udp_connect_worker6(struct work_struct *work)
1508 struct sock_xprt *transport =
1509 container_of(work, struct sock_xprt, connect_worker.work);
1510 struct rpc_xprt *xprt = &transport->xprt;
1511 struct socket *sock = transport->sock;
1512 int err, status = -EIO;
1514 if (xprt->shutdown || !xprt_bound(xprt))
1517 /* Start by resetting any existing state */
1520 if ((err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1521 dprintk("RPC: can't create UDP transport socket (%d).\n", -err);
1524 xs_reclassify_socket6(sock);
1526 if (xs_bind6(transport, sock) < 0) {
1531 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1532 xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1534 xs_udp_finish_connecting(xprt, sock);
1537 xprt_wake_pending_tasks(xprt, status);
1538 xprt_clear_connecting(xprt);
1542 * We need to preserve the port number so the reply cache on the server can
1543 * find our cached RPC replies when we get around to reconnecting.
1545 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1548 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1549 struct sockaddr any;
1551 dprintk("RPC: disconnecting xprt %p to reuse port\n", xprt);
1554 * Disconnect the transport socket by doing a connect operation
1555 * with AF_UNSPEC. This should return immediately...
1557 memset(&any, 0, sizeof(any));
1558 any.sa_family = AF_UNSPEC;
1559 result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1561 dprintk("RPC: AF_UNSPEC connect return code %d\n",
1565 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1567 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1569 if (!transport->inet) {
1570 struct sock *sk = sock->sk;
1572 write_lock_bh(&sk->sk_callback_lock);
1574 sk->sk_user_data = xprt;
1575 transport->old_data_ready = sk->sk_data_ready;
1576 transport->old_state_change = sk->sk_state_change;
1577 transport->old_write_space = sk->sk_write_space;
1578 sk->sk_data_ready = xs_tcp_data_ready;
1579 sk->sk_state_change = xs_tcp_state_change;
1580 sk->sk_write_space = xs_tcp_write_space;
1581 sk->sk_allocation = GFP_ATOMIC;
1583 /* socket options */
1584 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1585 sock_reset_flag(sk, SOCK_LINGER);
1586 tcp_sk(sk)->linger2 = 0;
1587 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1589 xprt_clear_connected(xprt);
1591 /* Reset to new socket */
1592 transport->sock = sock;
1593 transport->inet = sk;
1595 write_unlock_bh(&sk->sk_callback_lock);
1598 /* Tell the socket layer to start connecting... */
1599 xprt->stat.connect_count++;
1600 xprt->stat.connect_start = jiffies;
1601 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
1605 * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1606 * @work: RPC transport to connect
1608 * Invoked by a work queue tasklet.
1610 static void xs_tcp_connect_worker4(struct work_struct *work)
1612 struct sock_xprt *transport =
1613 container_of(work, struct sock_xprt, connect_worker.work);
1614 struct rpc_xprt *xprt = &transport->xprt;
1615 struct socket *sock = transport->sock;
1616 int err, status = -EIO;
1618 if (xprt->shutdown || !xprt_bound(xprt))
1622 /* start from scratch */
1623 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1624 dprintk("RPC: can't create TCP transport socket (%d).\n", -err);
1627 xs_reclassify_socket4(sock);
1629 if (xs_bind4(transport, sock) < 0) {
1634 /* "close" the socket, preserving the local port */
1635 xs_tcp_reuse_connection(xprt);
1637 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1638 xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1640 status = xs_tcp_finish_connecting(xprt, sock);
1641 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
1642 xprt, -status, xprt_connected(xprt),
1643 sock->sk->sk_state);
1651 /* retry with existing socket, after a delay */
1654 /* get rid of existing socket, and retry */
1655 xs_tcp_shutdown(xprt);
1659 xprt_wake_pending_tasks(xprt, status);
1661 xprt_clear_connecting(xprt);
1665 * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
1666 * @work: RPC transport to connect
1668 * Invoked by a work queue tasklet.
1670 static void xs_tcp_connect_worker6(struct work_struct *work)
1672 struct sock_xprt *transport =
1673 container_of(work, struct sock_xprt, connect_worker.work);
1674 struct rpc_xprt *xprt = &transport->xprt;
1675 struct socket *sock = transport->sock;
1676 int err, status = -EIO;
1678 if (xprt->shutdown || !xprt_bound(xprt))
1682 /* start from scratch */
1683 if ((err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1684 dprintk("RPC: can't create TCP transport socket (%d).\n", -err);
1687 xs_reclassify_socket6(sock);
1689 if (xs_bind6(transport, sock) < 0) {
1694 /* "close" the socket, preserving the local port */
1695 xs_tcp_reuse_connection(xprt);
1697 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1698 xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1700 status = xs_tcp_finish_connecting(xprt, sock);
1701 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
1702 xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1710 /* retry with existing socket, after a delay */
1713 /* get rid of existing socket, and retry */
1714 xs_tcp_shutdown(xprt);
1718 xprt_wake_pending_tasks(xprt, status);
1720 xprt_clear_connecting(xprt);
1724 * xs_connect - connect a socket to a remote endpoint
1725 * @task: address of RPC task that manages state of connect request
1727 * TCP: If the remote end dropped the connection, delay reconnecting.
1729 * UDP socket connects are synchronous, but we use a work queue anyway
1730 * to guarantee that even unprivileged user processes can set up a
1731 * socket on a privileged port.
1733 * If a UDP socket connect fails, the delay behavior here prevents
1734 * retry floods (hard mounts).
1736 static void xs_connect(struct rpc_task *task)
1738 struct rpc_xprt *xprt = task->tk_xprt;
1739 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1741 if (xprt_test_and_set_connecting(xprt))
1744 if (transport->sock != NULL) {
1745 dprintk("RPC: xs_connect delayed xprt %p for %lu "
1747 xprt, xprt->reestablish_timeout / HZ);
1748 queue_delayed_work(rpciod_workqueue,
1749 &transport->connect_worker,
1750 xprt->reestablish_timeout);
1751 xprt->reestablish_timeout <<= 1;
1752 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1753 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1755 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
1756 queue_delayed_work(rpciod_workqueue,
1757 &transport->connect_worker, 0);
1761 static void xs_tcp_connect(struct rpc_task *task)
1763 struct rpc_xprt *xprt = task->tk_xprt;
1765 /* Initiate graceful shutdown of the socket if not already done */
1766 if (test_bit(XPRT_CONNECTED, &xprt->state))
1767 xs_tcp_shutdown(xprt);
1768 /* Exit if we need to wait for socket shutdown to complete */
1769 if (test_bit(XPRT_CLOSING, &xprt->state))
1775 * xs_udp_print_stats - display UDP socket-specifc stats
1776 * @xprt: rpc_xprt struct containing statistics
1780 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1782 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1784 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1786 xprt->stat.bind_count,
1789 xprt->stat.bad_xids,
1791 xprt->stat.bklog_u);
1795 * xs_tcp_print_stats - display TCP socket-specifc stats
1796 * @xprt: rpc_xprt struct containing statistics
1800 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1802 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1805 if (xprt_connected(xprt))
1806 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1808 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1810 xprt->stat.bind_count,
1811 xprt->stat.connect_count,
1812 xprt->stat.connect_time,
1816 xprt->stat.bad_xids,
1818 xprt->stat.bklog_u);
1821 static struct rpc_xprt_ops xs_udp_ops = {
1822 .set_buffer_size = xs_udp_set_buffer_size,
1823 .reserve_xprt = xprt_reserve_xprt_cong,
1824 .release_xprt = xprt_release_xprt_cong,
1825 .rpcbind = rpcb_getport_async,
1826 .set_port = xs_set_port,
1827 .connect = xs_connect,
1828 .buf_alloc = rpc_malloc,
1829 .buf_free = rpc_free,
1830 .send_request = xs_udp_send_request,
1831 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
1832 .timer = xs_udp_timer,
1833 .release_request = xprt_release_rqst_cong,
1835 .destroy = xs_destroy,
1836 .print_stats = xs_udp_print_stats,
1839 static struct rpc_xprt_ops xs_tcp_ops = {
1840 .reserve_xprt = xprt_reserve_xprt,
1841 .release_xprt = xs_tcp_release_xprt,
1842 .rpcbind = rpcb_getport_async,
1843 .set_port = xs_set_port,
1844 .connect = xs_tcp_connect,
1845 .buf_alloc = rpc_malloc,
1846 .buf_free = rpc_free,
1847 .send_request = xs_tcp_send_request,
1848 .set_retrans_timeout = xprt_set_retrans_timeout_def,
1849 .close = xs_tcp_shutdown,
1850 .destroy = xs_destroy,
1851 .print_stats = xs_tcp_print_stats,
1854 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
1855 unsigned int slot_table_size)
1857 struct rpc_xprt *xprt;
1858 struct sock_xprt *new;
1860 if (args->addrlen > sizeof(xprt->addr)) {
1861 dprintk("RPC: xs_setup_xprt: address too large\n");
1862 return ERR_PTR(-EBADF);
1865 new = kzalloc(sizeof(*new), GFP_KERNEL);
1867 dprintk("RPC: xs_setup_xprt: couldn't allocate "
1869 return ERR_PTR(-ENOMEM);
1873 xprt->max_reqs = slot_table_size;
1874 xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1875 if (xprt->slot == NULL) {
1877 dprintk("RPC: xs_setup_xprt: couldn't allocate slot "
1879 return ERR_PTR(-ENOMEM);
1882 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
1883 xprt->addrlen = args->addrlen;
1885 memcpy(&new->addr, args->srcaddr, args->addrlen);
1891 * xs_setup_udp - Set up transport to use a UDP socket
1892 * @args: rpc transport creation arguments
1895 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
1897 struct sockaddr *addr = args->dstaddr;
1898 struct rpc_xprt *xprt;
1899 struct sock_xprt *transport;
1901 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
1904 transport = container_of(xprt, struct sock_xprt, xprt);
1906 xprt->prot = IPPROTO_UDP;
1908 /* XXX: header size can vary due to auth type, IPv6, etc. */
1909 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1911 xprt->bind_timeout = XS_BIND_TO;
1912 xprt->connect_timeout = XS_UDP_CONN_TO;
1913 xprt->reestablish_timeout = XS_UDP_REEST_TO;
1914 xprt->idle_timeout = XS_IDLE_DISC_TO;
1916 xprt->ops = &xs_udp_ops;
1919 xprt->timeout = *args->timeout;
1921 xprt_set_timeout(&xprt->timeout, 5, 5 * HZ);
1923 switch (addr->sa_family) {
1925 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1926 xprt_set_bound(xprt);
1928 INIT_DELAYED_WORK(&transport->connect_worker,
1929 xs_udp_connect_worker4);
1930 xs_format_ipv4_peer_addresses(xprt);
1933 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
1934 xprt_set_bound(xprt);
1936 INIT_DELAYED_WORK(&transport->connect_worker,
1937 xs_udp_connect_worker6);
1938 xs_format_ipv6_peer_addresses(xprt);
1942 return ERR_PTR(-EAFNOSUPPORT);
1945 dprintk("RPC: set up transport to address %s\n",
1946 xprt->address_strings[RPC_DISPLAY_ALL]);
1948 if (try_module_get(THIS_MODULE))
1953 return ERR_PTR(-EINVAL);
1957 * xs_setup_tcp - Set up transport to use a TCP socket
1958 * @args: rpc transport creation arguments
1961 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
1963 struct sockaddr *addr = args->dstaddr;
1964 struct rpc_xprt *xprt;
1965 struct sock_xprt *transport;
1967 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
1970 transport = container_of(xprt, struct sock_xprt, xprt);
1972 xprt->prot = IPPROTO_TCP;
1973 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
1974 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
1976 xprt->bind_timeout = XS_BIND_TO;
1977 xprt->connect_timeout = XS_TCP_CONN_TO;
1978 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1979 xprt->idle_timeout = XS_IDLE_DISC_TO;
1981 xprt->ops = &xs_tcp_ops;
1984 xprt->timeout = *args->timeout;
1986 xprt_set_timeout(&xprt->timeout, 2, 60 * HZ);
1988 switch (addr->sa_family) {
1990 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1991 xprt_set_bound(xprt);
1993 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker4);
1994 xs_format_ipv4_peer_addresses(xprt);
1997 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
1998 xprt_set_bound(xprt);
2000 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker6);
2001 xs_format_ipv6_peer_addresses(xprt);
2005 return ERR_PTR(-EAFNOSUPPORT);
2008 dprintk("RPC: set up transport to address %s\n",
2009 xprt->address_strings[RPC_DISPLAY_ALL]);
2011 if (try_module_get(THIS_MODULE))
2016 return ERR_PTR(-EINVAL);
2019 static struct xprt_class xs_udp_transport = {
2020 .list = LIST_HEAD_INIT(xs_udp_transport.list),
2022 .owner = THIS_MODULE,
2023 .ident = IPPROTO_UDP,
2024 .setup = xs_setup_udp,
2027 static struct xprt_class xs_tcp_transport = {
2028 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
2030 .owner = THIS_MODULE,
2031 .ident = IPPROTO_TCP,
2032 .setup = xs_setup_tcp,
2036 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2039 int init_socket_xprt(void)
2042 if (!sunrpc_table_header)
2043 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2046 xprt_register_transport(&xs_udp_transport);
2047 xprt_register_transport(&xs_tcp_transport);
2053 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2056 void cleanup_socket_xprt(void)
2059 if (sunrpc_table_header) {
2060 unregister_sysctl_table(sunrpc_table_header);
2061 sunrpc_table_header = NULL;
2065 xprt_unregister_transport(&xs_udp_transport);
2066 xprt_unregister_transport(&xs_tcp_transport);