2 * linux/net/sunrpc/xprtsock.c
4 * Client-side transport implementation for sockets.
6 * TCP callback races fixes (C) 1998 Red Hat Software <alan@redhat.com>
7 * TCP send fixes (C) 1998 Red Hat Software <alan@redhat.com>
8 * TCP NFS related read + write fixes
9 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
11 * Rewrite of larges part of the code in order to stabilize TCP stuff.
12 * Fix behaviour when socket buffer is full.
13 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
15 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
17 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
18 * <gilles.quillard@bull.net>
21 #include <linux/types.h>
22 #include <linux/slab.h>
23 #include <linux/capability.h>
24 #include <linux/pagemap.h>
25 #include <linux/errno.h>
26 #include <linux/socket.h>
28 #include <linux/net.h>
30 #include <linux/udp.h>
31 #include <linux/tcp.h>
32 #include <linux/sunrpc/clnt.h>
33 #include <linux/sunrpc/sched.h>
34 #include <linux/file.h>
37 #include <net/checksum.h>
44 unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
45 unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE;
47 unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
48 unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
51 * We can register our own files under /proc/sys/sunrpc by
52 * calling register_sysctl_table() again. The files in that
53 * directory become the union of all files registered there.
55 * We simply need to make sure that we don't collide with
56 * someone else's file names!
61 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
62 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
63 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
64 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
66 static struct ctl_table_header *sunrpc_table_header;
69 * FIXME: changing the UDP slot table size should also resize the UDP
70 * socket buffers for existing UDP transports
72 static ctl_table xs_tunables_table[] = {
74 .ctl_name = CTL_SLOTTABLE_UDP,
75 .procname = "udp_slot_table_entries",
76 .data = &xprt_udp_slot_table_entries,
77 .maxlen = sizeof(unsigned int),
79 .proc_handler = &proc_dointvec_minmax,
80 .strategy = &sysctl_intvec,
81 .extra1 = &min_slot_table_size,
82 .extra2 = &max_slot_table_size
85 .ctl_name = CTL_SLOTTABLE_TCP,
86 .procname = "tcp_slot_table_entries",
87 .data = &xprt_tcp_slot_table_entries,
88 .maxlen = sizeof(unsigned int),
90 .proc_handler = &proc_dointvec_minmax,
91 .strategy = &sysctl_intvec,
92 .extra1 = &min_slot_table_size,
93 .extra2 = &max_slot_table_size
96 .ctl_name = CTL_MIN_RESVPORT,
97 .procname = "min_resvport",
98 .data = &xprt_min_resvport,
99 .maxlen = sizeof(unsigned int),
101 .proc_handler = &proc_dointvec_minmax,
102 .strategy = &sysctl_intvec,
103 .extra1 = &xprt_min_resvport_limit,
104 .extra2 = &xprt_max_resvport_limit
107 .ctl_name = CTL_MAX_RESVPORT,
108 .procname = "max_resvport",
109 .data = &xprt_max_resvport,
110 .maxlen = sizeof(unsigned int),
112 .proc_handler = &proc_dointvec_minmax,
113 .strategy = &sysctl_intvec,
114 .extra1 = &xprt_min_resvport_limit,
115 .extra2 = &xprt_max_resvport_limit
122 static ctl_table sunrpc_table[] = {
124 .ctl_name = CTL_SUNRPC,
125 .procname = "sunrpc",
127 .child = xs_tunables_table
137 * How many times to try sending a request on a socket before waiting
138 * for the socket buffer to clear.
140 #define XS_SENDMSG_RETRY (10U)
143 * Time out for an RPC UDP socket connect. UDP socket connects are
144 * synchronous, but we set a timeout anyway in case of resource
145 * exhaustion on the local host.
147 #define XS_UDP_CONN_TO (5U * HZ)
150 * Wait duration for an RPC TCP connection to be established. Solaris
151 * NFS over TCP uses 60 seconds, for example, which is in line with how
152 * long a server takes to reboot.
154 #define XS_TCP_CONN_TO (60U * HZ)
157 * Wait duration for a reply from the RPC portmapper.
159 #define XS_BIND_TO (60U * HZ)
162 * Delay if a UDP socket connect error occurs. This is most likely some
163 * kind of resource problem on the local host.
165 #define XS_UDP_REEST_TO (2U * HZ)
168 * The reestablish timeout allows clients to delay for a bit before attempting
169 * to reconnect to a server that just dropped our connection.
171 * We implement an exponential backoff when trying to reestablish a TCP
172 * transport connection with the server. Some servers like to drop a TCP
173 * connection when they are overworked, so we start with a short timeout and
174 * increase over time if the server is down or not responding.
176 #define XS_TCP_INIT_REEST_TO (3U * HZ)
177 #define XS_TCP_MAX_REEST_TO (5U * 60 * HZ)
180 * TCP idle timeout; client drops the transport socket if it is idle
181 * for this long. Note that we also timeout UDP sockets to prevent
182 * holding port numbers when there is no RPC traffic.
184 #define XS_IDLE_DISC_TO (5U * 60 * HZ)
187 # undef RPC_DEBUG_DATA
188 # define RPCDBG_FACILITY RPCDBG_TRANS
191 #ifdef RPC_DEBUG_DATA
192 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
194 u8 *buf = (u8 *) packet;
197 dprintk("RPC: %s\n", msg);
198 for (j = 0; j < count && j < 128; j += 4) {
202 dprintk("0x%04x ", j);
204 dprintk("%02x%02x%02x%02x ",
205 buf[j], buf[j+1], buf[j+2], buf[j+3]);
210 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
217 struct rpc_xprt xprt;
222 struct socket * sock;
226 * State of TCP reply receive
234 unsigned long tcp_copied,
238 * Connection of transports
240 struct delayed_work connect_worker;
241 struct sockaddr_storage addr;
245 * UDP socket buffer size parameters
251 * Saved socket callback addresses
253 void (*old_data_ready)(struct sock *, int);
254 void (*old_state_change)(struct sock *);
255 void (*old_write_space)(struct sock *);
259 * TCP receive state flags
261 #define TCP_RCV_LAST_FRAG (1UL << 0)
262 #define TCP_RCV_COPY_FRAGHDR (1UL << 1)
263 #define TCP_RCV_COPY_XID (1UL << 2)
264 #define TCP_RCV_COPY_DATA (1UL << 3)
266 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
268 return (struct sockaddr *) &xprt->addr;
271 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
273 return (struct sockaddr_in *) &xprt->addr;
276 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
278 return (struct sockaddr_in6 *) &xprt->addr;
281 static void xs_format_ipv4_peer_addresses(struct rpc_xprt *xprt)
283 struct sockaddr_in *addr = xs_addr_in(xprt);
286 buf = kzalloc(20, GFP_KERNEL);
288 snprintf(buf, 20, NIPQUAD_FMT,
289 NIPQUAD(addr->sin_addr.s_addr));
291 xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
293 buf = kzalloc(8, GFP_KERNEL);
295 snprintf(buf, 8, "%u",
296 ntohs(addr->sin_port));
298 xprt->address_strings[RPC_DISPLAY_PORT] = buf;
300 buf = kzalloc(8, GFP_KERNEL);
302 if (xprt->prot == IPPROTO_UDP)
303 snprintf(buf, 8, "udp");
305 snprintf(buf, 8, "tcp");
307 xprt->address_strings[RPC_DISPLAY_PROTO] = buf;
309 buf = kzalloc(48, GFP_KERNEL);
311 snprintf(buf, 48, "addr="NIPQUAD_FMT" port=%u proto=%s",
312 NIPQUAD(addr->sin_addr.s_addr),
313 ntohs(addr->sin_port),
314 xprt->prot == IPPROTO_UDP ? "udp" : "tcp");
316 xprt->address_strings[RPC_DISPLAY_ALL] = buf;
318 buf = kzalloc(10, GFP_KERNEL);
320 snprintf(buf, 10, "%02x%02x%02x%02x",
321 NIPQUAD(addr->sin_addr.s_addr));
323 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
325 buf = kzalloc(8, GFP_KERNEL);
327 snprintf(buf, 8, "%4hx",
328 ntohs(addr->sin_port));
330 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
332 buf = kzalloc(30, GFP_KERNEL);
334 snprintf(buf, 30, NIPQUAD_FMT".%u.%u",
335 NIPQUAD(addr->sin_addr.s_addr),
336 ntohs(addr->sin_port) >> 8,
337 ntohs(addr->sin_port) & 0xff);
339 xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
341 xprt->address_strings[RPC_DISPLAY_NETID] =
342 kstrdup(xprt->prot == IPPROTO_UDP ?
343 RPCBIND_NETID_UDP : RPCBIND_NETID_TCP, GFP_KERNEL);
346 static void xs_format_ipv6_peer_addresses(struct rpc_xprt *xprt)
348 struct sockaddr_in6 *addr = xs_addr_in6(xprt);
351 buf = kzalloc(40, GFP_KERNEL);
353 snprintf(buf, 40, NIP6_FMT,
354 NIP6(addr->sin6_addr));
356 xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
358 buf = kzalloc(8, GFP_KERNEL);
360 snprintf(buf, 8, "%u",
361 ntohs(addr->sin6_port));
363 xprt->address_strings[RPC_DISPLAY_PORT] = buf;
365 buf = kzalloc(8, GFP_KERNEL);
367 if (xprt->prot == IPPROTO_UDP)
368 snprintf(buf, 8, "udp");
370 snprintf(buf, 8, "tcp");
372 xprt->address_strings[RPC_DISPLAY_PROTO] = buf;
374 buf = kzalloc(64, GFP_KERNEL);
376 snprintf(buf, 64, "addr="NIP6_FMT" port=%u proto=%s",
377 NIP6(addr->sin6_addr),
378 ntohs(addr->sin6_port),
379 xprt->prot == IPPROTO_UDP ? "udp" : "tcp");
381 xprt->address_strings[RPC_DISPLAY_ALL] = buf;
383 buf = kzalloc(36, GFP_KERNEL);
385 snprintf(buf, 36, NIP6_SEQFMT,
386 NIP6(addr->sin6_addr));
388 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
390 buf = kzalloc(8, GFP_KERNEL);
392 snprintf(buf, 8, "%4hx",
393 ntohs(addr->sin6_port));
395 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
397 buf = kzalloc(50, GFP_KERNEL);
399 snprintf(buf, 50, NIP6_FMT".%u.%u",
400 NIP6(addr->sin6_addr),
401 ntohs(addr->sin6_port) >> 8,
402 ntohs(addr->sin6_port) & 0xff);
404 xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
406 xprt->address_strings[RPC_DISPLAY_NETID] =
407 kstrdup(xprt->prot == IPPROTO_UDP ?
408 RPCBIND_NETID_UDP6 : RPCBIND_NETID_TCP6, GFP_KERNEL);
411 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
415 for (i = 0; i < RPC_DISPLAY_MAX; i++)
416 kfree(xprt->address_strings[i]);
419 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
421 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
423 struct msghdr msg = {
425 .msg_namelen = addrlen,
426 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
429 .iov_base = vec->iov_base + base,
430 .iov_len = vec->iov_len - base,
433 if (iov.iov_len != 0)
434 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
435 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
438 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
441 unsigned int remainder;
444 remainder = xdr->page_len - base;
445 base += xdr->page_base;
446 ppage = xdr->pages + (base >> PAGE_SHIFT);
449 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
450 int flags = XS_SENDMSG_FLAGS;
453 if (remainder != 0 || more)
455 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
456 if (remainder == 0 || err != len)
470 * xs_sendpages - write pages directly to a socket
471 * @sock: socket to send on
472 * @addr: UDP only -- address of destination
473 * @addrlen: UDP only -- length of destination address
474 * @xdr: buffer containing this request
475 * @base: starting position in the buffer
478 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
480 unsigned int remainder = xdr->len - base;
486 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
492 if (base < xdr->head[0].iov_len || addr != NULL) {
493 unsigned int len = xdr->head[0].iov_len - base;
495 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
496 if (remainder == 0 || err != len)
501 base -= xdr->head[0].iov_len;
503 if (base < xdr->page_len) {
504 unsigned int len = xdr->page_len - base;
506 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
507 if (remainder == 0 || err != len)
512 base -= xdr->page_len;
514 if (base >= xdr->tail[0].iov_len)
516 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
526 * xs_nospace - place task on wait queue if transmit was incomplete
527 * @task: task to put to sleep
530 static void xs_nospace(struct rpc_task *task)
532 struct rpc_rqst *req = task->tk_rqstp;
533 struct rpc_xprt *xprt = req->rq_xprt;
534 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
536 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
537 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
540 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
541 /* Protect against races with write_space */
542 spin_lock_bh(&xprt->transport_lock);
544 /* Don't race with disconnect */
545 if (!xprt_connected(xprt))
546 task->tk_status = -ENOTCONN;
547 else if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
548 xprt_wait_for_buffer_space(task);
550 spin_unlock_bh(&xprt->transport_lock);
552 /* Keep holding the socket if it is blocked */
553 rpc_delay(task, HZ>>4);
557 * xs_udp_send_request - write an RPC request to a UDP socket
558 * @task: address of RPC task that manages the state of an RPC request
561 * 0: The request has been sent
562 * EAGAIN: The socket was blocked, please call again later to
563 * complete the request
564 * ENOTCONN: Caller needs to invoke connect logic then call again
565 * other: Some other error occured, the request was not sent
567 static int xs_udp_send_request(struct rpc_task *task)
569 struct rpc_rqst *req = task->tk_rqstp;
570 struct rpc_xprt *xprt = req->rq_xprt;
571 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
572 struct xdr_buf *xdr = &req->rq_snd_buf;
575 xs_pktdump("packet data:",
576 req->rq_svec->iov_base,
577 req->rq_svec->iov_len);
579 req->rq_xtime = jiffies;
580 status = xs_sendpages(transport->sock,
585 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
586 xdr->len - req->rq_bytes_sent, status);
588 if (likely(status >= (int) req->rq_slen))
591 /* Still some bytes left; set up for a retry later. */
599 /* When the server has died, an ICMP port unreachable message
600 * prompts ECONNREFUSED. */
606 dprintk("RPC: sendmsg returned unrecognized error %d\n",
614 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
616 u32 reclen = buf->len - sizeof(rpc_fraghdr);
617 rpc_fraghdr *base = buf->head[0].iov_base;
618 *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
622 * xs_tcp_send_request - write an RPC request to a TCP socket
623 * @task: address of RPC task that manages the state of an RPC request
626 * 0: The request has been sent
627 * EAGAIN: The socket was blocked, please call again later to
628 * complete the request
629 * ENOTCONN: Caller needs to invoke connect logic then call again
630 * other: Some other error occured, the request was not sent
632 * XXX: In the case of soft timeouts, should we eventually give up
633 * if sendmsg is not able to make progress?
635 static int xs_tcp_send_request(struct rpc_task *task)
637 struct rpc_rqst *req = task->tk_rqstp;
638 struct rpc_xprt *xprt = req->rq_xprt;
639 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
640 struct xdr_buf *xdr = &req->rq_snd_buf;
642 unsigned int retry = 0;
644 xs_encode_tcp_record_marker(&req->rq_snd_buf);
646 xs_pktdump("packet data:",
647 req->rq_svec->iov_base,
648 req->rq_svec->iov_len);
650 /* Continue transmitting the packet/record. We must be careful
651 * to cope with writespace callbacks arriving _after_ we have
652 * called sendmsg(). */
654 req->rq_xtime = jiffies;
655 status = xs_sendpages(transport->sock,
656 NULL, 0, xdr, req->rq_bytes_sent);
658 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
659 xdr->len - req->rq_bytes_sent, status);
661 if (unlikely(status < 0))
664 /* If we've sent the entire packet, immediately
665 * reset the count of bytes sent. */
666 req->rq_bytes_sent += status;
667 task->tk_bytes_sent += status;
668 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
669 req->rq_bytes_sent = 0;
674 if (retry++ > XS_SENDMSG_RETRY)
689 dprintk("RPC: sendmsg returned unrecognized error %d\n",
691 xprt_disconnect(xprt);
699 * xs_tcp_release_xprt - clean up after a tcp transmission
703 * This cleans up if an error causes us to abort the transmission of a request.
704 * In this case, the socket may need to be reset in order to avoid confusing
707 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
709 struct rpc_rqst *req;
711 if (task != xprt->snd_task)
715 req = task->tk_rqstp;
716 if (req->rq_bytes_sent == 0)
718 if (req->rq_bytes_sent == req->rq_snd_buf.len)
720 set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
722 xprt_release_xprt(xprt, task);
726 * xs_close - close a socket
729 * This is used when all requests are complete; ie, no DRC state remains
730 * on the server we want to save.
732 static void xs_close(struct rpc_xprt *xprt)
734 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
735 struct socket *sock = transport->sock;
736 struct sock *sk = transport->inet;
739 goto clear_close_wait;
741 dprintk("RPC: xs_close xprt %p\n", xprt);
743 write_lock_bh(&sk->sk_callback_lock);
744 transport->inet = NULL;
745 transport->sock = NULL;
747 sk->sk_user_data = NULL;
748 sk->sk_data_ready = transport->old_data_ready;
749 sk->sk_state_change = transport->old_state_change;
750 sk->sk_write_space = transport->old_write_space;
751 write_unlock_bh(&sk->sk_callback_lock);
757 smp_mb__before_clear_bit();
758 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
759 smp_mb__after_clear_bit();
763 * xs_destroy - prepare to shutdown a transport
764 * @xprt: doomed transport
767 static void xs_destroy(struct rpc_xprt *xprt)
769 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
771 dprintk("RPC: xs_destroy xprt %p\n", xprt);
773 cancel_rearming_delayed_work(&transport->connect_worker);
775 xprt_disconnect(xprt);
777 xs_free_peer_addresses(xprt);
782 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
784 return (struct rpc_xprt *) sk->sk_user_data;
788 * xs_udp_data_ready - "data ready" callback for UDP sockets
789 * @sk: socket with data to read
790 * @len: how much data to read
793 static void xs_udp_data_ready(struct sock *sk, int len)
795 struct rpc_task *task;
796 struct rpc_xprt *xprt;
797 struct rpc_rqst *rovr;
799 int err, repsize, copied;
803 read_lock(&sk->sk_callback_lock);
804 dprintk("RPC: xs_udp_data_ready...\n");
805 if (!(xprt = xprt_from_sock(sk)))
808 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
814 repsize = skb->len - sizeof(struct udphdr);
816 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
820 /* Copy the XID from the skb... */
821 xp = skb_header_pointer(skb, sizeof(struct udphdr),
822 sizeof(_xid), &_xid);
826 /* Look up and lock the request corresponding to the given XID */
827 spin_lock(&xprt->transport_lock);
828 rovr = xprt_lookup_rqst(xprt, *xp);
831 task = rovr->rq_task;
833 if ((copied = rovr->rq_private_buf.buflen) > repsize)
836 /* Suck it into the iovec, verify checksum if not done by hw. */
837 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb))
840 /* Something worked... */
841 dst_confirm(skb->dst);
843 xprt_adjust_cwnd(task, copied);
844 xprt_update_rtt(task);
845 xprt_complete_rqst(task, copied);
848 spin_unlock(&xprt->transport_lock);
850 skb_free_datagram(sk, skb);
852 read_unlock(&sk->sk_callback_lock);
855 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
857 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
861 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
862 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
863 used = xdr_skb_read_bits(desc, p, len);
864 transport->tcp_offset += used;
868 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
869 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
870 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
872 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
873 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
875 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
876 transport->tcp_offset = 0;
878 /* Sanity check of the record length */
879 if (unlikely(transport->tcp_reclen < 4)) {
880 dprintk("RPC: invalid TCP record fragment length\n");
881 xprt_disconnect(xprt);
884 dprintk("RPC: reading TCP record fragment of length %d\n",
885 transport->tcp_reclen);
888 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
890 if (transport->tcp_offset == transport->tcp_reclen) {
891 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
892 transport->tcp_offset = 0;
893 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
894 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
895 transport->tcp_flags |= TCP_RCV_COPY_XID;
896 transport->tcp_copied = 0;
901 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
906 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
907 dprintk("RPC: reading XID (%Zu bytes)\n", len);
908 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
909 used = xdr_skb_read_bits(desc, p, len);
910 transport->tcp_offset += used;
913 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
914 transport->tcp_flags |= TCP_RCV_COPY_DATA;
915 transport->tcp_copied = 4;
916 dprintk("RPC: reading reply for XID %08x\n",
917 ntohl(transport->tcp_xid));
918 xs_tcp_check_fraghdr(transport);
921 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
923 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
924 struct rpc_rqst *req;
925 struct xdr_buf *rcvbuf;
929 /* Find and lock the request corresponding to this xid */
930 spin_lock(&xprt->transport_lock);
931 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
933 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
934 dprintk("RPC: XID %08x request not found!\n",
935 ntohl(transport->tcp_xid));
936 spin_unlock(&xprt->transport_lock);
940 rcvbuf = &req->rq_private_buf;
942 if (len > transport->tcp_reclen - transport->tcp_offset) {
943 struct xdr_skb_reader my_desc;
945 len = transport->tcp_reclen - transport->tcp_offset;
946 memcpy(&my_desc, desc, sizeof(my_desc));
948 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
949 &my_desc, xdr_skb_read_bits);
953 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
954 desc, xdr_skb_read_bits);
957 transport->tcp_copied += r;
958 transport->tcp_offset += r;
961 /* Error when copying to the receive buffer,
962 * usually because we weren't able to allocate
963 * additional buffer pages. All we can do now
964 * is turn off TCP_RCV_COPY_DATA, so the request
965 * will not receive any additional updates,
967 * Any remaining data from this record will
970 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
971 dprintk("RPC: XID %08x truncated request\n",
972 ntohl(transport->tcp_xid));
973 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
974 "tcp_offset = %u, tcp_reclen = %u\n",
975 xprt, transport->tcp_copied,
976 transport->tcp_offset, transport->tcp_reclen);
980 dprintk("RPC: XID %08x read %Zd bytes\n",
981 ntohl(transport->tcp_xid), r);
982 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
983 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
984 transport->tcp_offset, transport->tcp_reclen);
986 if (transport->tcp_copied == req->rq_private_buf.buflen)
987 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
988 else if (transport->tcp_offset == transport->tcp_reclen) {
989 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
990 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
994 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
995 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
996 spin_unlock(&xprt->transport_lock);
997 xs_tcp_check_fraghdr(transport);
1000 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1004 len = transport->tcp_reclen - transport->tcp_offset;
1005 if (len > desc->count)
1008 desc->offset += len;
1009 transport->tcp_offset += len;
1010 dprintk("RPC: discarded %Zu bytes\n", len);
1011 xs_tcp_check_fraghdr(transport);
1014 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1016 struct rpc_xprt *xprt = rd_desc->arg.data;
1017 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1018 struct xdr_skb_reader desc = {
1024 dprintk("RPC: xs_tcp_data_recv started\n");
1026 /* Read in a new fragment marker if necessary */
1027 /* Can we ever really expect to get completely empty fragments? */
1028 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1029 xs_tcp_read_fraghdr(xprt, &desc);
1032 /* Read in the xid if necessary */
1033 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1034 xs_tcp_read_xid(transport, &desc);
1037 /* Read in the request data */
1038 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1039 xs_tcp_read_request(xprt, &desc);
1042 /* Skip over any trailing bytes on short reads */
1043 xs_tcp_read_discard(transport, &desc);
1044 } while (desc.count);
1045 dprintk("RPC: xs_tcp_data_recv done\n");
1046 return len - desc.count;
1050 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1051 * @sk: socket with data to read
1052 * @bytes: how much data to read
1055 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1057 struct rpc_xprt *xprt;
1058 read_descriptor_t rd_desc;
1060 dprintk("RPC: xs_tcp_data_ready...\n");
1062 read_lock(&sk->sk_callback_lock);
1063 if (!(xprt = xprt_from_sock(sk)))
1068 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1069 rd_desc.arg.data = xprt;
1070 rd_desc.count = 65536;
1071 tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1073 read_unlock(&sk->sk_callback_lock);
1077 * xs_tcp_state_change - callback to handle TCP socket state changes
1078 * @sk: socket whose state has changed
1081 static void xs_tcp_state_change(struct sock *sk)
1083 struct rpc_xprt *xprt;
1085 read_lock(&sk->sk_callback_lock);
1086 if (!(xprt = xprt_from_sock(sk)))
1088 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1089 dprintk("RPC: state %x conn %d dead %d zapped %d\n",
1090 sk->sk_state, xprt_connected(xprt),
1091 sock_flag(sk, SOCK_DEAD),
1092 sock_flag(sk, SOCK_ZAPPED));
1094 switch (sk->sk_state) {
1095 case TCP_ESTABLISHED:
1096 spin_lock_bh(&xprt->transport_lock);
1097 if (!xprt_test_and_set_connected(xprt)) {
1098 struct sock_xprt *transport = container_of(xprt,
1099 struct sock_xprt, xprt);
1101 /* Reset TCP record info */
1102 transport->tcp_offset = 0;
1103 transport->tcp_reclen = 0;
1104 transport->tcp_copied = 0;
1105 transport->tcp_flags =
1106 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1108 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1109 xprt_wake_pending_tasks(xprt, 0);
1111 spin_unlock_bh(&xprt->transport_lock);
1116 case TCP_CLOSE_WAIT:
1117 /* Try to schedule an autoclose RPC calls */
1118 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
1119 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
1120 queue_work(rpciod_workqueue, &xprt->task_cleanup);
1122 xprt_disconnect(xprt);
1125 read_unlock(&sk->sk_callback_lock);
1129 * xs_udp_write_space - callback invoked when socket buffer space
1131 * @sk: socket whose state has changed
1133 * Called when more output buffer space is available for this socket.
1134 * We try not to wake our writers until they can make "significant"
1135 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1136 * with a bunch of small requests.
1138 static void xs_udp_write_space(struct sock *sk)
1140 read_lock(&sk->sk_callback_lock);
1142 /* from net/core/sock.c:sock_def_write_space */
1143 if (sock_writeable(sk)) {
1144 struct socket *sock;
1145 struct rpc_xprt *xprt;
1147 if (unlikely(!(sock = sk->sk_socket)))
1149 if (unlikely(!(xprt = xprt_from_sock(sk))))
1151 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
1154 xprt_write_space(xprt);
1158 read_unlock(&sk->sk_callback_lock);
1162 * xs_tcp_write_space - callback invoked when socket buffer space
1164 * @sk: socket whose state has changed
1166 * Called when more output buffer space is available for this socket.
1167 * We try not to wake our writers until they can make "significant"
1168 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1169 * with a bunch of small requests.
1171 static void xs_tcp_write_space(struct sock *sk)
1173 read_lock(&sk->sk_callback_lock);
1175 /* from net/core/stream.c:sk_stream_write_space */
1176 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
1177 struct socket *sock;
1178 struct rpc_xprt *xprt;
1180 if (unlikely(!(sock = sk->sk_socket)))
1182 if (unlikely(!(xprt = xprt_from_sock(sk))))
1184 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
1187 xprt_write_space(xprt);
1191 read_unlock(&sk->sk_callback_lock);
1194 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1196 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1197 struct sock *sk = transport->inet;
1199 if (transport->rcvsize) {
1200 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1201 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1203 if (transport->sndsize) {
1204 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1205 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1206 sk->sk_write_space(sk);
1211 * xs_udp_set_buffer_size - set send and receive limits
1212 * @xprt: generic transport
1213 * @sndsize: requested size of send buffer, in bytes
1214 * @rcvsize: requested size of receive buffer, in bytes
1216 * Set socket send and receive buffer size limits.
1218 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1220 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1222 transport->sndsize = 0;
1224 transport->sndsize = sndsize + 1024;
1225 transport->rcvsize = 0;
1227 transport->rcvsize = rcvsize + 1024;
1229 xs_udp_do_set_buffer_size(xprt);
1233 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1234 * @task: task that timed out
1236 * Adjust the congestion window after a retransmit timeout has occurred.
1238 static void xs_udp_timer(struct rpc_task *task)
1240 xprt_adjust_cwnd(task, -ETIMEDOUT);
1243 static unsigned short xs_get_random_port(void)
1245 unsigned short range = xprt_max_resvport - xprt_min_resvport;
1246 unsigned short rand = (unsigned short) net_random() % range;
1247 return rand + xprt_min_resvport;
1251 * xs_set_port - reset the port number in the remote endpoint address
1252 * @xprt: generic transport
1253 * @port: new port number
1256 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1258 struct sockaddr *addr = xs_addr(xprt);
1260 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1262 switch (addr->sa_family) {
1264 ((struct sockaddr_in *)addr)->sin_port = htons(port);
1267 ((struct sockaddr_in6 *)addr)->sin6_port = htons(port);
1274 static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
1276 struct sockaddr_in myaddr = {
1277 .sin_family = AF_INET,
1279 struct sockaddr_in *sa;
1281 unsigned short port = transport->port;
1283 if (!transport->xprt.resvport)
1285 sa = (struct sockaddr_in *)&transport->addr;
1286 myaddr.sin_addr = sa->sin_addr;
1288 myaddr.sin_port = htons(port);
1289 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1291 if (!transport->xprt.resvport)
1294 transport->port = port;
1297 if (port <= xprt_min_resvport)
1298 port = xprt_max_resvport;
1301 } while (err == -EADDRINUSE && port != transport->port);
1302 dprintk("RPC: %s "NIPQUAD_FMT":%u: %s (%d)\n",
1303 __FUNCTION__, NIPQUAD(myaddr.sin_addr),
1304 port, err ? "failed" : "ok", err);
1308 static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1310 struct sockaddr_in6 myaddr = {
1311 .sin6_family = AF_INET6,
1313 struct sockaddr_in6 *sa;
1315 unsigned short port = transport->port;
1317 if (!transport->xprt.resvport)
1319 sa = (struct sockaddr_in6 *)&transport->addr;
1320 myaddr.sin6_addr = sa->sin6_addr;
1322 myaddr.sin6_port = htons(port);
1323 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1325 if (!transport->xprt.resvport)
1328 transport->port = port;
1331 if (port <= xprt_min_resvport)
1332 port = xprt_max_resvport;
1335 } while (err == -EADDRINUSE && port != transport->port);
1336 dprintk("RPC: xs_bind6 "NIP6_FMT":%u: %s (%d)\n",
1337 NIP6(myaddr.sin6_addr), port, err ? "failed" : "ok", err);
1341 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1342 static struct lock_class_key xs_key[2];
1343 static struct lock_class_key xs_slock_key[2];
1345 static inline void xs_reclassify_socket4(struct socket *sock)
1347 struct sock *sk = sock->sk;
1349 BUG_ON(sk->sk_lock.owner != NULL);
1350 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1351 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1354 static inline void xs_reclassify_socket6(struct socket *sock)
1356 struct sock *sk = sock->sk;
1358 BUG_ON(sk->sk_lock.owner != NULL);
1359 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1360 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1363 static inline void xs_reclassify_socket4(struct socket *sock)
1367 static inline void xs_reclassify_socket6(struct socket *sock)
1372 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1374 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1376 if (!transport->inet) {
1377 struct sock *sk = sock->sk;
1379 write_lock_bh(&sk->sk_callback_lock);
1381 sk->sk_user_data = xprt;
1382 transport->old_data_ready = sk->sk_data_ready;
1383 transport->old_state_change = sk->sk_state_change;
1384 transport->old_write_space = sk->sk_write_space;
1385 sk->sk_data_ready = xs_udp_data_ready;
1386 sk->sk_write_space = xs_udp_write_space;
1387 sk->sk_no_check = UDP_CSUM_NORCV;
1388 sk->sk_allocation = GFP_ATOMIC;
1390 xprt_set_connected(xprt);
1392 /* Reset to new socket */
1393 transport->sock = sock;
1394 transport->inet = sk;
1396 write_unlock_bh(&sk->sk_callback_lock);
1398 xs_udp_do_set_buffer_size(xprt);
1402 * xs_udp_connect_worker4 - set up a UDP socket
1403 * @work: RPC transport to connect
1405 * Invoked by a work queue tasklet.
1407 static void xs_udp_connect_worker4(struct work_struct *work)
1409 struct sock_xprt *transport =
1410 container_of(work, struct sock_xprt, connect_worker.work);
1411 struct rpc_xprt *xprt = &transport->xprt;
1412 struct socket *sock = transport->sock;
1413 int err, status = -EIO;
1415 if (xprt->shutdown || !xprt_bound(xprt))
1418 /* Start by resetting any existing state */
1421 if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1422 dprintk("RPC: can't create UDP transport socket (%d).\n", -err);
1425 xs_reclassify_socket4(sock);
1427 if (xs_bind4(transport, sock)) {
1432 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1433 xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1435 xs_udp_finish_connecting(xprt, sock);
1438 xprt_wake_pending_tasks(xprt, status);
1439 xprt_clear_connecting(xprt);
1443 * xs_udp_connect_worker6 - set up a UDP socket
1444 * @work: RPC transport to connect
1446 * Invoked by a work queue tasklet.
1448 static void xs_udp_connect_worker6(struct work_struct *work)
1450 struct sock_xprt *transport =
1451 container_of(work, struct sock_xprt, connect_worker.work);
1452 struct rpc_xprt *xprt = &transport->xprt;
1453 struct socket *sock = transport->sock;
1454 int err, status = -EIO;
1456 if (xprt->shutdown || !xprt_bound(xprt))
1459 /* Start by resetting any existing state */
1462 if ((err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1463 dprintk("RPC: can't create UDP transport socket (%d).\n", -err);
1466 xs_reclassify_socket6(sock);
1468 if (xs_bind6(transport, sock) < 0) {
1473 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1474 xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1476 xs_udp_finish_connecting(xprt, sock);
1479 xprt_wake_pending_tasks(xprt, status);
1480 xprt_clear_connecting(xprt);
1484 * We need to preserve the port number so the reply cache on the server can
1485 * find our cached RPC replies when we get around to reconnecting.
1487 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1490 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1491 struct sockaddr any;
1493 dprintk("RPC: disconnecting xprt %p to reuse port\n", xprt);
1496 * Disconnect the transport socket by doing a connect operation
1497 * with AF_UNSPEC. This should return immediately...
1499 memset(&any, 0, sizeof(any));
1500 any.sa_family = AF_UNSPEC;
1501 result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1503 dprintk("RPC: AF_UNSPEC connect return code %d\n",
1507 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1509 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1511 if (!transport->inet) {
1512 struct sock *sk = sock->sk;
1514 write_lock_bh(&sk->sk_callback_lock);
1516 sk->sk_user_data = xprt;
1517 transport->old_data_ready = sk->sk_data_ready;
1518 transport->old_state_change = sk->sk_state_change;
1519 transport->old_write_space = sk->sk_write_space;
1520 sk->sk_data_ready = xs_tcp_data_ready;
1521 sk->sk_state_change = xs_tcp_state_change;
1522 sk->sk_write_space = xs_tcp_write_space;
1523 sk->sk_allocation = GFP_ATOMIC;
1525 /* socket options */
1526 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1527 sock_reset_flag(sk, SOCK_LINGER);
1528 tcp_sk(sk)->linger2 = 0;
1529 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1531 xprt_clear_connected(xprt);
1533 /* Reset to new socket */
1534 transport->sock = sock;
1535 transport->inet = sk;
1537 write_unlock_bh(&sk->sk_callback_lock);
1540 /* Tell the socket layer to start connecting... */
1541 xprt->stat.connect_count++;
1542 xprt->stat.connect_start = jiffies;
1543 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
1547 * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1548 * @work: RPC transport to connect
1550 * Invoked by a work queue tasklet.
1552 static void xs_tcp_connect_worker4(struct work_struct *work)
1554 struct sock_xprt *transport =
1555 container_of(work, struct sock_xprt, connect_worker.work);
1556 struct rpc_xprt *xprt = &transport->xprt;
1557 struct socket *sock = transport->sock;
1558 int err, status = -EIO;
1560 if (xprt->shutdown || !xprt_bound(xprt))
1564 /* start from scratch */
1565 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1566 dprintk("RPC: can't create TCP transport socket (%d).\n", -err);
1569 xs_reclassify_socket4(sock);
1571 if (xs_bind4(transport, sock) < 0) {
1576 /* "close" the socket, preserving the local port */
1577 xs_tcp_reuse_connection(xprt);
1579 dprintk("RPC: worker connecting xprt %p to address: %s\n",
1580 xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1582 status = xs_tcp_finish_connecting(xprt, sock);
1583 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
1584 xprt, -status, xprt_connected(xprt),
1585 sock->sk->sk_state);
1593 /* retry with existing socket, after a delay */
1596 /* get rid of existing socket, and retry */
1602 xprt_wake_pending_tasks(xprt, status);
1604 xprt_clear_connecting(xprt);
1608 * xs_tcp_connect_worker6 - 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_worker6(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_INET6, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1627 dprintk("RPC: can't create TCP transport socket (%d).\n", -err);
1630 xs_reclassify_socket6(sock);
1632 if (xs_bind6(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), sock->sk->sk_state);
1653 /* retry with existing socket, after a delay */
1656 /* get rid of existing socket, and retry */
1662 xprt_wake_pending_tasks(xprt, status);
1664 xprt_clear_connecting(xprt);
1668 * xs_connect - connect a socket to a remote endpoint
1669 * @task: address of RPC task that manages state of connect request
1671 * TCP: If the remote end dropped the connection, delay reconnecting.
1673 * UDP socket connects are synchronous, but we use a work queue anyway
1674 * to guarantee that even unprivileged user processes can set up a
1675 * socket on a privileged port.
1677 * If a UDP socket connect fails, the delay behavior here prevents
1678 * retry floods (hard mounts).
1680 static void xs_connect(struct rpc_task *task)
1682 struct rpc_xprt *xprt = task->tk_xprt;
1683 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1685 if (xprt_test_and_set_connecting(xprt))
1688 if (transport->sock != NULL) {
1689 dprintk("RPC: xs_connect delayed xprt %p for %lu "
1691 xprt, xprt->reestablish_timeout / HZ);
1692 queue_delayed_work(rpciod_workqueue,
1693 &transport->connect_worker,
1694 xprt->reestablish_timeout);
1695 xprt->reestablish_timeout <<= 1;
1696 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1697 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1699 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
1700 queue_delayed_work(rpciod_workqueue,
1701 &transport->connect_worker, 0);
1706 * xs_udp_print_stats - display UDP socket-specifc stats
1707 * @xprt: rpc_xprt struct containing statistics
1711 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1713 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1715 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1717 xprt->stat.bind_count,
1720 xprt->stat.bad_xids,
1722 xprt->stat.bklog_u);
1726 * xs_tcp_print_stats - display TCP socket-specifc stats
1727 * @xprt: rpc_xprt struct containing statistics
1731 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1733 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1736 if (xprt_connected(xprt))
1737 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1739 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1741 xprt->stat.bind_count,
1742 xprt->stat.connect_count,
1743 xprt->stat.connect_time,
1747 xprt->stat.bad_xids,
1749 xprt->stat.bklog_u);
1752 static struct rpc_xprt_ops xs_udp_ops = {
1753 .set_buffer_size = xs_udp_set_buffer_size,
1754 .reserve_xprt = xprt_reserve_xprt_cong,
1755 .release_xprt = xprt_release_xprt_cong,
1756 .rpcbind = rpcb_getport_async,
1757 .set_port = xs_set_port,
1758 .connect = xs_connect,
1759 .buf_alloc = rpc_malloc,
1760 .buf_free = rpc_free,
1761 .send_request = xs_udp_send_request,
1762 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
1763 .timer = xs_udp_timer,
1764 .release_request = xprt_release_rqst_cong,
1766 .destroy = xs_destroy,
1767 .print_stats = xs_udp_print_stats,
1770 static struct rpc_xprt_ops xs_tcp_ops = {
1771 .reserve_xprt = xprt_reserve_xprt,
1772 .release_xprt = xs_tcp_release_xprt,
1773 .rpcbind = rpcb_getport_async,
1774 .set_port = xs_set_port,
1775 .connect = xs_connect,
1776 .buf_alloc = rpc_malloc,
1777 .buf_free = rpc_free,
1778 .send_request = xs_tcp_send_request,
1779 .set_retrans_timeout = xprt_set_retrans_timeout_def,
1781 .destroy = xs_destroy,
1782 .print_stats = xs_tcp_print_stats,
1785 static struct rpc_xprt *xs_setup_xprt(struct rpc_xprtsock_create *args, unsigned int slot_table_size)
1787 struct rpc_xprt *xprt;
1788 struct sock_xprt *new;
1790 if (args->addrlen > sizeof(xprt->addr)) {
1791 dprintk("RPC: xs_setup_xprt: address too large\n");
1792 return ERR_PTR(-EBADF);
1795 new = kzalloc(sizeof(*new), GFP_KERNEL);
1797 dprintk("RPC: xs_setup_xprt: couldn't allocate "
1799 return ERR_PTR(-ENOMEM);
1803 xprt->max_reqs = slot_table_size;
1804 xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1805 if (xprt->slot == NULL) {
1807 dprintk("RPC: xs_setup_xprt: couldn't allocate slot "
1809 return ERR_PTR(-ENOMEM);
1812 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
1813 xprt->addrlen = args->addrlen;
1815 memcpy(&new->addr, args->srcaddr, args->addrlen);
1816 new->port = xs_get_random_port();
1822 * xs_setup_udp - Set up transport to use a UDP socket
1823 * @args: rpc transport creation arguments
1826 struct rpc_xprt *xs_setup_udp(struct rpc_xprtsock_create *args)
1828 struct sockaddr *addr = args->dstaddr;
1829 struct rpc_xprt *xprt;
1830 struct sock_xprt *transport;
1832 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
1835 transport = container_of(xprt, struct sock_xprt, xprt);
1837 xprt->prot = IPPROTO_UDP;
1839 /* XXX: header size can vary due to auth type, IPv6, etc. */
1840 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1842 xprt->bind_timeout = XS_BIND_TO;
1843 xprt->connect_timeout = XS_UDP_CONN_TO;
1844 xprt->reestablish_timeout = XS_UDP_REEST_TO;
1845 xprt->idle_timeout = XS_IDLE_DISC_TO;
1847 xprt->ops = &xs_udp_ops;
1850 xprt->timeout = *args->timeout;
1852 xprt_set_timeout(&xprt->timeout, 5, 5 * HZ);
1854 switch (addr->sa_family) {
1856 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1857 xprt_set_bound(xprt);
1859 INIT_DELAYED_WORK(&transport->connect_worker,
1860 xs_udp_connect_worker4);
1861 xs_format_ipv4_peer_addresses(xprt);
1864 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
1865 xprt_set_bound(xprt);
1867 INIT_DELAYED_WORK(&transport->connect_worker,
1868 xs_udp_connect_worker6);
1869 xs_format_ipv6_peer_addresses(xprt);
1873 return ERR_PTR(-EAFNOSUPPORT);
1876 dprintk("RPC: set up transport to address %s\n",
1877 xprt->address_strings[RPC_DISPLAY_ALL]);
1883 * xs_setup_tcp - Set up transport to use a TCP socket
1884 * @args: rpc transport creation arguments
1887 struct rpc_xprt *xs_setup_tcp(struct rpc_xprtsock_create *args)
1889 struct sockaddr *addr = args->dstaddr;
1890 struct rpc_xprt *xprt;
1891 struct sock_xprt *transport;
1893 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
1896 transport = container_of(xprt, struct sock_xprt, xprt);
1898 xprt->prot = IPPROTO_TCP;
1899 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
1900 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
1902 xprt->bind_timeout = XS_BIND_TO;
1903 xprt->connect_timeout = XS_TCP_CONN_TO;
1904 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1905 xprt->idle_timeout = XS_IDLE_DISC_TO;
1907 xprt->ops = &xs_tcp_ops;
1910 xprt->timeout = *args->timeout;
1912 xprt_set_timeout(&xprt->timeout, 2, 60 * HZ);
1914 switch (addr->sa_family) {
1916 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
1917 xprt_set_bound(xprt);
1919 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker4);
1920 xs_format_ipv4_peer_addresses(xprt);
1923 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
1924 xprt_set_bound(xprt);
1926 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker6);
1927 xs_format_ipv6_peer_addresses(xprt);
1931 return ERR_PTR(-EAFNOSUPPORT);
1934 dprintk("RPC: set up transport to address %s\n",
1935 xprt->address_strings[RPC_DISPLAY_ALL]);
1941 * init_socket_xprt - set up xprtsock's sysctls
1944 int init_socket_xprt(void)
1947 if (!sunrpc_table_header)
1948 sunrpc_table_header = register_sysctl_table(sunrpc_table);
1955 * cleanup_socket_xprt - remove xprtsock's sysctls
1958 void cleanup_socket_xprt(void)
1961 if (sunrpc_table_header) {
1962 unregister_sysctl_table(sunrpc_table_header);
1963 sunrpc_table_header = NULL;