2 * linux/net/sunrpc/xprtsock.c
4 * Client-side transport implementation for sockets.
6 * TCP callback races fixes (C) 1998 Red Hat
7 * TCP send fixes (C) 1998 Red Hat
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/string.h>
23 #include <linux/slab.h>
24 #include <linux/module.h>
25 #include <linux/capability.h>
26 #include <linux/pagemap.h>
27 #include <linux/errno.h>
28 #include <linux/socket.h>
30 #include <linux/net.h>
33 #include <linux/udp.h>
34 #include <linux/tcp.h>
35 #include <linux/sunrpc/clnt.h>
36 #include <linux/sunrpc/sched.h>
37 #include <linux/sunrpc/svcsock.h>
38 #include <linux/sunrpc/xprtsock.h>
39 #include <linux/file.h>
40 #ifdef CONFIG_SUNRPC_BACKCHANNEL
41 #include <linux/sunrpc/bc_xprt.h>
45 #include <net/checksum.h>
51 static void xs_close(struct rpc_xprt *xprt);
56 unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
57 unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
58 unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
60 unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
61 unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
63 #define XS_TCP_LINGER_TO (15U * HZ)
64 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
67 * We can register our own files under /proc/sys/sunrpc by
68 * calling register_sysctl_table() again. The files in that
69 * directory become the union of all files registered there.
71 * We simply need to make sure that we don't collide with
72 * someone else's file names!
77 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
78 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
79 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
80 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
81 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
83 static struct ctl_table_header *sunrpc_table_header;
86 * FIXME: changing the UDP slot table size should also resize the UDP
87 * socket buffers for existing UDP transports
89 static ctl_table xs_tunables_table[] = {
91 .procname = "udp_slot_table_entries",
92 .data = &xprt_udp_slot_table_entries,
93 .maxlen = sizeof(unsigned int),
95 .proc_handler = proc_dointvec_minmax,
96 .extra1 = &min_slot_table_size,
97 .extra2 = &max_slot_table_size
100 .procname = "tcp_slot_table_entries",
101 .data = &xprt_tcp_slot_table_entries,
102 .maxlen = sizeof(unsigned int),
104 .proc_handler = proc_dointvec_minmax,
105 .extra1 = &min_slot_table_size,
106 .extra2 = &max_slot_table_size
109 .procname = "tcp_max_slot_table_entries",
110 .data = &xprt_max_tcp_slot_table_entries,
111 .maxlen = sizeof(unsigned int),
113 .proc_handler = proc_dointvec_minmax,
114 .extra1 = &min_slot_table_size,
115 .extra2 = &max_tcp_slot_table_limit
118 .procname = "min_resvport",
119 .data = &xprt_min_resvport,
120 .maxlen = sizeof(unsigned int),
122 .proc_handler = proc_dointvec_minmax,
123 .extra1 = &xprt_min_resvport_limit,
124 .extra2 = &xprt_max_resvport_limit
127 .procname = "max_resvport",
128 .data = &xprt_max_resvport,
129 .maxlen = sizeof(unsigned int),
131 .proc_handler = proc_dointvec_minmax,
132 .extra1 = &xprt_min_resvport_limit,
133 .extra2 = &xprt_max_resvport_limit
136 .procname = "tcp_fin_timeout",
137 .data = &xs_tcp_fin_timeout,
138 .maxlen = sizeof(xs_tcp_fin_timeout),
140 .proc_handler = proc_dointvec_jiffies,
145 static ctl_table sunrpc_table[] = {
147 .procname = "sunrpc",
149 .child = xs_tunables_table
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
235 unsigned long tcp_copied,
239 * Connection of transports
241 struct delayed_work connect_worker;
242 struct sockaddr_storage srcaddr;
243 unsigned short srcport;
246 * UDP socket buffer size parameters
252 * Saved socket callback addresses
254 void (*old_data_ready)(struct sock *, int);
255 void (*old_state_change)(struct sock *);
256 void (*old_write_space)(struct sock *);
260 * TCP receive state flags
262 #define TCP_RCV_LAST_FRAG (1UL << 0)
263 #define TCP_RCV_COPY_FRAGHDR (1UL << 1)
264 #define TCP_RCV_COPY_XID (1UL << 2)
265 #define TCP_RCV_COPY_DATA (1UL << 3)
266 #define TCP_RCV_READ_CALLDIR (1UL << 4)
267 #define TCP_RCV_COPY_CALLDIR (1UL << 5)
272 #define TCP_RPC_REPLY (1UL << 6)
274 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
276 return (struct sockaddr *) &xprt->addr;
279 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
281 return (struct sockaddr_un *) &xprt->addr;
284 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
286 return (struct sockaddr_in *) &xprt->addr;
289 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
291 return (struct sockaddr_in6 *) &xprt->addr;
294 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
296 struct sockaddr *sap = xs_addr(xprt);
297 struct sockaddr_in6 *sin6;
298 struct sockaddr_in *sin;
299 struct sockaddr_un *sun;
302 switch (sap->sa_family) {
304 sun = xs_addr_un(xprt);
305 strlcpy(buf, sun->sun_path, sizeof(buf));
306 xprt->address_strings[RPC_DISPLAY_ADDR] =
307 kstrdup(buf, GFP_KERNEL);
310 (void)rpc_ntop(sap, buf, sizeof(buf));
311 xprt->address_strings[RPC_DISPLAY_ADDR] =
312 kstrdup(buf, GFP_KERNEL);
313 sin = xs_addr_in(xprt);
314 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
317 (void)rpc_ntop(sap, buf, sizeof(buf));
318 xprt->address_strings[RPC_DISPLAY_ADDR] =
319 kstrdup(buf, GFP_KERNEL);
320 sin6 = xs_addr_in6(xprt);
321 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
327 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
330 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
332 struct sockaddr *sap = xs_addr(xprt);
335 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
336 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
338 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
339 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
342 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
343 const char *protocol,
346 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
347 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
348 xs_format_common_peer_addresses(xprt);
349 xs_format_common_peer_ports(xprt);
352 static void xs_update_peer_port(struct rpc_xprt *xprt)
354 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
355 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
357 xs_format_common_peer_ports(xprt);
360 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
364 for (i = 0; i < RPC_DISPLAY_MAX; i++)
366 case RPC_DISPLAY_PROTO:
367 case RPC_DISPLAY_NETID:
370 kfree(xprt->address_strings[i]);
374 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
376 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
378 struct msghdr msg = {
380 .msg_namelen = addrlen,
381 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
384 .iov_base = vec->iov_base + base,
385 .iov_len = vec->iov_len - base,
388 if (iov.iov_len != 0)
389 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
390 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
393 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy)
395 ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
396 int offset, size_t size, int flags);
398 unsigned int remainder;
401 remainder = xdr->page_len - base;
402 base += xdr->page_base;
403 ppage = xdr->pages + (base >> PAGE_SHIFT);
405 do_sendpage = sock->ops->sendpage;
407 do_sendpage = sock_no_sendpage;
409 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
410 int flags = XS_SENDMSG_FLAGS;
413 if (remainder != 0 || more)
415 err = do_sendpage(sock, *ppage, base, len, flags);
416 if (remainder == 0 || err != len)
430 * xs_sendpages - write pages directly to a socket
431 * @sock: socket to send on
432 * @addr: UDP only -- address of destination
433 * @addrlen: UDP only -- length of destination address
434 * @xdr: buffer containing this request
435 * @base: starting position in the buffer
436 * @zerocopy: true if it is safe to use sendpage()
439 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy)
441 unsigned int remainder = xdr->len - base;
447 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
453 if (base < xdr->head[0].iov_len || addr != NULL) {
454 unsigned int len = xdr->head[0].iov_len - base;
456 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
457 if (remainder == 0 || err != len)
462 base -= xdr->head[0].iov_len;
464 if (base < xdr->page_len) {
465 unsigned int len = xdr->page_len - base;
467 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy);
468 if (remainder == 0 || err != len)
473 base -= xdr->page_len;
475 if (base >= xdr->tail[0].iov_len)
477 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
486 static void xs_nospace_callback(struct rpc_task *task)
488 struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
490 transport->inet->sk_write_pending--;
491 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
495 * xs_nospace - place task on wait queue if transmit was incomplete
496 * @task: task to put to sleep
499 static int xs_nospace(struct rpc_task *task)
501 struct rpc_rqst *req = task->tk_rqstp;
502 struct rpc_xprt *xprt = req->rq_xprt;
503 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
504 struct sock *sk = transport->inet;
507 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
508 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
511 /* Protect against races with write_space */
512 spin_lock_bh(&xprt->transport_lock);
514 /* Don't race with disconnect */
515 if (xprt_connected(xprt)) {
516 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
518 * Notify TCP that we're limited by the application
521 set_bit(SOCK_NOSPACE, &transport->sock->flags);
522 sk->sk_write_pending++;
523 /* ...and wait for more buffer space */
524 xprt_wait_for_buffer_space(task, xs_nospace_callback);
527 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
531 spin_unlock_bh(&xprt->transport_lock);
533 /* Race breaker in case memory is freed before above code is called */
534 sk->sk_write_space(sk);
539 * Construct a stream transport record marker in @buf.
541 static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
543 u32 reclen = buf->len - sizeof(rpc_fraghdr);
544 rpc_fraghdr *base = buf->head[0].iov_base;
545 *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
549 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
550 * @task: RPC task that manages the state of an RPC request
553 * 0: The request has been sent
554 * EAGAIN: The socket was blocked, please call again later to
555 * complete the request
556 * ENOTCONN: Caller needs to invoke connect logic then call again
557 * other: Some other error occured, the request was not sent
559 static int xs_local_send_request(struct rpc_task *task)
561 struct rpc_rqst *req = task->tk_rqstp;
562 struct rpc_xprt *xprt = req->rq_xprt;
563 struct sock_xprt *transport =
564 container_of(xprt, struct sock_xprt, xprt);
565 struct xdr_buf *xdr = &req->rq_snd_buf;
568 xs_encode_stream_record_marker(&req->rq_snd_buf);
570 xs_pktdump("packet data:",
571 req->rq_svec->iov_base, req->rq_svec->iov_len);
573 status = xs_sendpages(transport->sock, NULL, 0,
574 xdr, req->rq_bytes_sent, true);
575 dprintk("RPC: %s(%u) = %d\n",
576 __func__, xdr->len - req->rq_bytes_sent, status);
577 if (likely(status >= 0)) {
578 req->rq_bytes_sent += status;
579 req->rq_xmit_bytes_sent += status;
580 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
581 req->rq_bytes_sent = 0;
589 status = xs_nospace(task);
592 dprintk("RPC: sendmsg returned unrecognized error %d\n",
603 * xs_udp_send_request - write an RPC request to a UDP socket
604 * @task: address of RPC task that manages the state of an RPC request
607 * 0: The request has been sent
608 * EAGAIN: The socket was blocked, please call again later to
609 * complete the request
610 * ENOTCONN: Caller needs to invoke connect logic then call again
611 * other: Some other error occurred, the request was not sent
613 static int xs_udp_send_request(struct rpc_task *task)
615 struct rpc_rqst *req = task->tk_rqstp;
616 struct rpc_xprt *xprt = req->rq_xprt;
617 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
618 struct xdr_buf *xdr = &req->rq_snd_buf;
621 xs_pktdump("packet data:",
622 req->rq_svec->iov_base,
623 req->rq_svec->iov_len);
625 if (!xprt_bound(xprt))
627 status = xs_sendpages(transport->sock,
630 req->rq_bytes_sent, true);
632 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
633 xdr->len - req->rq_bytes_sent, status);
636 req->rq_xmit_bytes_sent += status;
637 if (status >= req->rq_slen)
639 /* Still some bytes left; set up for a retry later. */
646 /* Should we call xs_close() here? */
649 status = xs_nospace(task);
652 dprintk("RPC: sendmsg returned unrecognized error %d\n",
657 /* When the server has died, an ICMP port unreachable message
658 * prompts ECONNREFUSED. */
659 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
666 * xs_tcp_shutdown - gracefully shut down a TCP socket
669 * Initiates a graceful shutdown of the TCP socket by calling the
670 * equivalent of shutdown(SHUT_WR);
672 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
674 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
675 struct socket *sock = transport->sock;
678 kernel_sock_shutdown(sock, SHUT_WR);
682 * xs_tcp_send_request - write an RPC request to a TCP socket
683 * @task: address of RPC task that manages the state of an RPC request
686 * 0: The request has been sent
687 * EAGAIN: The socket was blocked, please call again later to
688 * complete the request
689 * ENOTCONN: Caller needs to invoke connect logic then call again
690 * other: Some other error occurred, the request was not sent
692 * XXX: In the case of soft timeouts, should we eventually give up
693 * if sendmsg is not able to make progress?
695 static int xs_tcp_send_request(struct rpc_task *task)
697 struct rpc_rqst *req = task->tk_rqstp;
698 struct rpc_xprt *xprt = req->rq_xprt;
699 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
700 struct xdr_buf *xdr = &req->rq_snd_buf;
701 bool zerocopy = true;
704 xs_encode_stream_record_marker(&req->rq_snd_buf);
706 xs_pktdump("packet data:",
707 req->rq_svec->iov_base,
708 req->rq_svec->iov_len);
709 /* Don't use zero copy if this is a resend. If the RPC call
710 * completes while the socket holds a reference to the pages,
711 * then we may end up resending corrupted data.
713 if (task->tk_flags & RPC_TASK_SENT)
716 /* Continue transmitting the packet/record. We must be careful
717 * to cope with writespace callbacks arriving _after_ we have
718 * called sendmsg(). */
720 status = xs_sendpages(transport->sock,
721 NULL, 0, xdr, req->rq_bytes_sent,
724 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
725 xdr->len - req->rq_bytes_sent, status);
727 if (unlikely(status < 0))
730 /* If we've sent the entire packet, immediately
731 * reset the count of bytes sent. */
732 req->rq_bytes_sent += status;
733 req->rq_xmit_bytes_sent += status;
734 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
735 req->rq_bytes_sent = 0;
748 /* Should we call xs_close() here? */
751 status = xs_nospace(task);
754 dprintk("RPC: sendmsg returned unrecognized error %d\n",
757 xs_tcp_shutdown(xprt);
761 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
768 * xs_tcp_release_xprt - clean up after a tcp transmission
772 * This cleans up if an error causes us to abort the transmission of a request.
773 * In this case, the socket may need to be reset in order to avoid confusing
776 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
778 struct rpc_rqst *req;
780 if (task != xprt->snd_task)
784 req = task->tk_rqstp;
787 if (req->rq_bytes_sent == 0)
789 if (req->rq_bytes_sent == req->rq_snd_buf.len)
791 set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
793 xprt_release_xprt(xprt, task);
796 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
798 transport->old_data_ready = sk->sk_data_ready;
799 transport->old_state_change = sk->sk_state_change;
800 transport->old_write_space = sk->sk_write_space;
803 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
805 sk->sk_data_ready = transport->old_data_ready;
806 sk->sk_state_change = transport->old_state_change;
807 sk->sk_write_space = transport->old_write_space;
810 static void xs_reset_transport(struct sock_xprt *transport)
812 struct socket *sock = transport->sock;
813 struct sock *sk = transport->inet;
814 struct rpc_xprt *xprt = &transport->xprt;
819 transport->srcport = 0;
821 write_lock_bh(&sk->sk_callback_lock);
822 transport->inet = NULL;
823 transport->sock = NULL;
825 sk->sk_user_data = NULL;
827 xs_restore_old_callbacks(transport, sk);
828 xprt_clear_connected(xprt);
829 write_unlock_bh(&sk->sk_callback_lock);
837 * xs_close - close a socket
840 * This is used when all requests are complete; ie, no DRC state remains
841 * on the server we want to save.
843 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
844 * xs_reset_transport() zeroing the socket from underneath a writer.
846 static void xs_close(struct rpc_xprt *xprt)
848 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
850 dprintk("RPC: xs_close xprt %p\n", xprt);
852 xs_reset_transport(transport);
853 xprt->reestablish_timeout = 0;
855 smp_mb__before_clear_bit();
856 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
857 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
858 clear_bit(XPRT_CLOSING, &xprt->state);
859 smp_mb__after_clear_bit();
860 xprt_disconnect_done(xprt);
863 static void xs_tcp_close(struct rpc_xprt *xprt)
865 if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state))
868 xs_tcp_shutdown(xprt);
872 * xs_destroy - prepare to shutdown a transport
873 * @xprt: doomed transport
876 static void xs_destroy(struct rpc_xprt *xprt)
878 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
880 dprintk("RPC: xs_destroy xprt %p\n", xprt);
882 cancel_delayed_work_sync(&transport->connect_worker);
885 xs_free_peer_addresses(xprt);
887 module_put(THIS_MODULE);
890 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
892 return (struct rpc_xprt *) sk->sk_user_data;
895 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
897 struct xdr_skb_reader desc = {
899 .offset = sizeof(rpc_fraghdr),
900 .count = skb->len - sizeof(rpc_fraghdr),
903 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
911 * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets
912 * @sk: socket with data to read
913 * @len: how much data to read
915 * Currently this assumes we can read the whole reply in a single gulp.
917 static void xs_local_data_ready(struct sock *sk, int len)
919 struct rpc_task *task;
920 struct rpc_xprt *xprt;
921 struct rpc_rqst *rovr;
923 int err, repsize, copied;
927 read_lock_bh(&sk->sk_callback_lock);
928 dprintk("RPC: %s...\n", __func__);
929 xprt = xprt_from_sock(sk);
933 skb = skb_recv_datagram(sk, 0, 1, &err);
940 repsize = skb->len - sizeof(rpc_fraghdr);
942 dprintk("RPC: impossible RPC reply size %d\n", repsize);
946 /* Copy the XID from the skb... */
947 xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
951 /* Look up and lock the request corresponding to the given XID */
952 spin_lock(&xprt->transport_lock);
953 rovr = xprt_lookup_rqst(xprt, *xp);
956 task = rovr->rq_task;
958 copied = rovr->rq_private_buf.buflen;
959 if (copied > repsize)
962 if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
963 dprintk("RPC: sk_buff copy failed\n");
967 xprt_complete_rqst(task, copied);
970 spin_unlock(&xprt->transport_lock);
972 skb_free_datagram(sk, skb);
974 read_unlock_bh(&sk->sk_callback_lock);
978 * xs_udp_data_ready - "data ready" callback for UDP sockets
979 * @sk: socket with data to read
980 * @len: how much data to read
983 static void xs_udp_data_ready(struct sock *sk, int len)
985 struct rpc_task *task;
986 struct rpc_xprt *xprt;
987 struct rpc_rqst *rovr;
989 int err, repsize, copied;
993 read_lock_bh(&sk->sk_callback_lock);
994 dprintk("RPC: xs_udp_data_ready...\n");
995 if (!(xprt = xprt_from_sock(sk)))
998 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
1004 repsize = skb->len - sizeof(struct udphdr);
1006 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
1010 /* Copy the XID from the skb... */
1011 xp = skb_header_pointer(skb, sizeof(struct udphdr),
1012 sizeof(_xid), &_xid);
1016 /* Look up and lock the request corresponding to the given XID */
1017 spin_lock(&xprt->transport_lock);
1018 rovr = xprt_lookup_rqst(xprt, *xp);
1021 task = rovr->rq_task;
1023 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1026 /* Suck it into the iovec, verify checksum if not done by hw. */
1027 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1028 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
1032 UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
1034 /* Something worked... */
1035 dst_confirm(skb_dst(skb));
1037 xprt_adjust_cwnd(task, copied);
1038 xprt_complete_rqst(task, copied);
1041 spin_unlock(&xprt->transport_lock);
1043 skb_free_datagram(sk, skb);
1045 read_unlock_bh(&sk->sk_callback_lock);
1049 * Helper function to force a TCP close if the server is sending
1050 * junk and/or it has put us in CLOSE_WAIT
1052 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1054 set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1055 xprt_force_disconnect(xprt);
1058 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1060 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1064 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1065 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1066 used = xdr_skb_read_bits(desc, p, len);
1067 transport->tcp_offset += used;
1071 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1072 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1073 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1075 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1076 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1078 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1079 transport->tcp_offset = 0;
1081 /* Sanity check of the record length */
1082 if (unlikely(transport->tcp_reclen < 8)) {
1083 dprintk("RPC: invalid TCP record fragment length\n");
1084 xs_tcp_force_close(xprt);
1087 dprintk("RPC: reading TCP record fragment of length %d\n",
1088 transport->tcp_reclen);
1091 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1093 if (transport->tcp_offset == transport->tcp_reclen) {
1094 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1095 transport->tcp_offset = 0;
1096 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1097 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1098 transport->tcp_flags |= TCP_RCV_COPY_XID;
1099 transport->tcp_copied = 0;
1104 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1109 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1110 dprintk("RPC: reading XID (%Zu bytes)\n", len);
1111 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1112 used = xdr_skb_read_bits(desc, p, len);
1113 transport->tcp_offset += used;
1116 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1117 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1118 transport->tcp_copied = 4;
1119 dprintk("RPC: reading %s XID %08x\n",
1120 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1122 ntohl(transport->tcp_xid));
1123 xs_tcp_check_fraghdr(transport);
1126 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1127 struct xdr_skb_reader *desc)
1134 * We want transport->tcp_offset to be 8 at the end of this routine
1135 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1136 * When this function is called for the first time,
1137 * transport->tcp_offset is 4 (after having already read the xid).
1139 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1140 len = sizeof(transport->tcp_calldir) - offset;
1141 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len);
1142 p = ((char *) &transport->tcp_calldir) + offset;
1143 used = xdr_skb_read_bits(desc, p, len);
1144 transport->tcp_offset += used;
1147 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1149 * We don't yet have the XDR buffer, so we will write the calldir
1150 * out after we get the buffer from the 'struct rpc_rqst'
1152 switch (ntohl(transport->tcp_calldir)) {
1154 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1155 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1156 transport->tcp_flags |= TCP_RPC_REPLY;
1159 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1160 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1161 transport->tcp_flags &= ~TCP_RPC_REPLY;
1164 dprintk("RPC: invalid request message type\n");
1165 xs_tcp_force_close(&transport->xprt);
1167 xs_tcp_check_fraghdr(transport);
1170 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1171 struct xdr_skb_reader *desc,
1172 struct rpc_rqst *req)
1174 struct sock_xprt *transport =
1175 container_of(xprt, struct sock_xprt, xprt);
1176 struct xdr_buf *rcvbuf;
1180 rcvbuf = &req->rq_private_buf;
1182 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1184 * Save the RPC direction in the XDR buffer
1186 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1187 &transport->tcp_calldir,
1188 sizeof(transport->tcp_calldir));
1189 transport->tcp_copied += sizeof(transport->tcp_calldir);
1190 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1194 if (len > transport->tcp_reclen - transport->tcp_offset) {
1195 struct xdr_skb_reader my_desc;
1197 len = transport->tcp_reclen - transport->tcp_offset;
1198 memcpy(&my_desc, desc, sizeof(my_desc));
1199 my_desc.count = len;
1200 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1201 &my_desc, xdr_skb_read_bits);
1205 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1206 desc, xdr_skb_read_bits);
1209 transport->tcp_copied += r;
1210 transport->tcp_offset += r;
1213 /* Error when copying to the receive buffer,
1214 * usually because we weren't able to allocate
1215 * additional buffer pages. All we can do now
1216 * is turn off TCP_RCV_COPY_DATA, so the request
1217 * will not receive any additional updates,
1219 * Any remaining data from this record will
1222 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1223 dprintk("RPC: XID %08x truncated request\n",
1224 ntohl(transport->tcp_xid));
1225 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1226 "tcp_offset = %u, tcp_reclen = %u\n",
1227 xprt, transport->tcp_copied,
1228 transport->tcp_offset, transport->tcp_reclen);
1232 dprintk("RPC: XID %08x read %Zd bytes\n",
1233 ntohl(transport->tcp_xid), r);
1234 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1235 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1236 transport->tcp_offset, transport->tcp_reclen);
1238 if (transport->tcp_copied == req->rq_private_buf.buflen)
1239 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1240 else if (transport->tcp_offset == transport->tcp_reclen) {
1241 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1242 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1247 * Finds the request corresponding to the RPC xid and invokes the common
1248 * tcp read code to read the data.
1250 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1251 struct xdr_skb_reader *desc)
1253 struct sock_xprt *transport =
1254 container_of(xprt, struct sock_xprt, xprt);
1255 struct rpc_rqst *req;
1257 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1259 /* Find and lock the request corresponding to this xid */
1260 spin_lock(&xprt->transport_lock);
1261 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1263 dprintk("RPC: XID %08x request not found!\n",
1264 ntohl(transport->tcp_xid));
1265 spin_unlock(&xprt->transport_lock);
1269 xs_tcp_read_common(xprt, desc, req);
1271 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1272 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1274 spin_unlock(&xprt->transport_lock);
1278 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1280 * Obtains an rpc_rqst previously allocated and invokes the common
1281 * tcp read code to read the data. The result is placed in the callback
1283 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1284 * connection and return -1.
1286 static inline int xs_tcp_read_callback(struct rpc_xprt *xprt,
1287 struct xdr_skb_reader *desc)
1289 struct sock_xprt *transport =
1290 container_of(xprt, struct sock_xprt, xprt);
1291 struct rpc_rqst *req;
1293 req = xprt_alloc_bc_request(xprt);
1295 printk(KERN_WARNING "Callback slot table overflowed\n");
1296 xprt_force_disconnect(xprt);
1300 req->rq_xid = transport->tcp_xid;
1301 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1302 xs_tcp_read_common(xprt, desc, req);
1304 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) {
1305 struct svc_serv *bc_serv = xprt->bc_serv;
1308 * Add callback request to callback list. The callback
1309 * service sleeps on the sv_cb_waitq waiting for new
1310 * requests. Wake it up after adding enqueing the
1313 dprintk("RPC: add callback request to list\n");
1314 spin_lock(&bc_serv->sv_cb_lock);
1315 list_add(&req->rq_bc_list, &bc_serv->sv_cb_list);
1316 spin_unlock(&bc_serv->sv_cb_lock);
1317 wake_up(&bc_serv->sv_cb_waitq);
1320 req->rq_private_buf.len = transport->tcp_copied;
1325 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1326 struct xdr_skb_reader *desc)
1328 struct sock_xprt *transport =
1329 container_of(xprt, struct sock_xprt, xprt);
1331 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1332 xs_tcp_read_reply(xprt, desc) :
1333 xs_tcp_read_callback(xprt, desc);
1336 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1337 struct xdr_skb_reader *desc)
1339 return xs_tcp_read_reply(xprt, desc);
1341 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1344 * Read data off the transport. This can be either an RPC_CALL or an
1345 * RPC_REPLY. Relay the processing to helper functions.
1347 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1348 struct xdr_skb_reader *desc)
1350 struct sock_xprt *transport =
1351 container_of(xprt, struct sock_xprt, xprt);
1353 if (_xs_tcp_read_data(xprt, desc) == 0)
1354 xs_tcp_check_fraghdr(transport);
1357 * The transport_lock protects the request handling.
1358 * There's no need to hold it to update the tcp_flags.
1360 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1364 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1368 len = transport->tcp_reclen - transport->tcp_offset;
1369 if (len > desc->count)
1372 desc->offset += len;
1373 transport->tcp_offset += len;
1374 dprintk("RPC: discarded %Zu bytes\n", len);
1375 xs_tcp_check_fraghdr(transport);
1378 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1380 struct rpc_xprt *xprt = rd_desc->arg.data;
1381 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1382 struct xdr_skb_reader desc = {
1388 dprintk("RPC: xs_tcp_data_recv started\n");
1390 /* Read in a new fragment marker if necessary */
1391 /* Can we ever really expect to get completely empty fragments? */
1392 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1393 xs_tcp_read_fraghdr(xprt, &desc);
1396 /* Read in the xid if necessary */
1397 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1398 xs_tcp_read_xid(transport, &desc);
1401 /* Read in the call/reply flag */
1402 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1403 xs_tcp_read_calldir(transport, &desc);
1406 /* Read in the request data */
1407 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1408 xs_tcp_read_data(xprt, &desc);
1411 /* Skip over any trailing bytes on short reads */
1412 xs_tcp_read_discard(transport, &desc);
1413 } while (desc.count);
1414 dprintk("RPC: xs_tcp_data_recv done\n");
1415 return len - desc.count;
1419 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1420 * @sk: socket with data to read
1421 * @bytes: how much data to read
1424 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1426 struct rpc_xprt *xprt;
1427 read_descriptor_t rd_desc;
1430 dprintk("RPC: xs_tcp_data_ready...\n");
1432 read_lock_bh(&sk->sk_callback_lock);
1433 if (!(xprt = xprt_from_sock(sk)))
1438 /* Any data means we had a useful conversation, so
1439 * the we don't need to delay the next reconnect
1441 if (xprt->reestablish_timeout)
1442 xprt->reestablish_timeout = 0;
1444 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1445 rd_desc.arg.data = xprt;
1447 rd_desc.count = 65536;
1448 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1451 read_unlock_bh(&sk->sk_callback_lock);
1455 * Do the equivalent of linger/linger2 handling for dealing with
1456 * broken servers that don't close the socket in a timely
1459 static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1460 unsigned long timeout)
1462 struct sock_xprt *transport;
1464 if (xprt_test_and_set_connecting(xprt))
1466 set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1467 transport = container_of(xprt, struct sock_xprt, xprt);
1468 queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1472 static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1474 struct sock_xprt *transport;
1476 transport = container_of(xprt, struct sock_xprt, xprt);
1478 if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1479 !cancel_delayed_work(&transport->connect_worker))
1481 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1482 xprt_clear_connecting(xprt);
1485 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1487 smp_mb__before_clear_bit();
1488 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1489 clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1490 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1491 clear_bit(XPRT_CLOSING, &xprt->state);
1492 smp_mb__after_clear_bit();
1495 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1497 xs_sock_reset_connection_flags(xprt);
1498 /* Mark transport as closed and wake up all pending tasks */
1499 xprt_disconnect_done(xprt);
1503 * xs_tcp_state_change - callback to handle TCP socket state changes
1504 * @sk: socket whose state has changed
1507 static void xs_tcp_state_change(struct sock *sk)
1509 struct rpc_xprt *xprt;
1511 read_lock_bh(&sk->sk_callback_lock);
1512 if (!(xprt = xprt_from_sock(sk)))
1514 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1515 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1516 sk->sk_state, xprt_connected(xprt),
1517 sock_flag(sk, SOCK_DEAD),
1518 sock_flag(sk, SOCK_ZAPPED),
1521 switch (sk->sk_state) {
1522 case TCP_ESTABLISHED:
1523 spin_lock(&xprt->transport_lock);
1524 if (!xprt_test_and_set_connected(xprt)) {
1525 struct sock_xprt *transport = container_of(xprt,
1526 struct sock_xprt, xprt);
1528 /* Reset TCP record info */
1529 transport->tcp_offset = 0;
1530 transport->tcp_reclen = 0;
1531 transport->tcp_copied = 0;
1532 transport->tcp_flags =
1533 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1535 xprt_wake_pending_tasks(xprt, -EAGAIN);
1537 spin_unlock(&xprt->transport_lock);
1540 /* The client initiated a shutdown of the socket */
1541 xprt->connect_cookie++;
1542 xprt->reestablish_timeout = 0;
1543 set_bit(XPRT_CLOSING, &xprt->state);
1544 smp_mb__before_clear_bit();
1545 clear_bit(XPRT_CONNECTED, &xprt->state);
1546 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1547 smp_mb__after_clear_bit();
1548 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1550 case TCP_CLOSE_WAIT:
1551 /* The server initiated a shutdown of the socket */
1552 xprt->connect_cookie++;
1553 clear_bit(XPRT_CONNECTED, &xprt->state);
1554 xs_tcp_force_close(xprt);
1557 * If the server closed down the connection, make sure that
1558 * we back off before reconnecting
1560 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1561 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1564 set_bit(XPRT_CLOSING, &xprt->state);
1565 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1566 smp_mb__before_clear_bit();
1567 clear_bit(XPRT_CONNECTED, &xprt->state);
1568 smp_mb__after_clear_bit();
1571 xs_tcp_cancel_linger_timeout(xprt);
1572 xs_sock_mark_closed(xprt);
1575 read_unlock_bh(&sk->sk_callback_lock);
1578 static void xs_write_space(struct sock *sk)
1580 struct socket *sock;
1581 struct rpc_xprt *xprt;
1583 if (unlikely(!(sock = sk->sk_socket)))
1585 clear_bit(SOCK_NOSPACE, &sock->flags);
1587 if (unlikely(!(xprt = xprt_from_sock(sk))))
1589 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1592 xprt_write_space(xprt);
1596 * xs_udp_write_space - callback invoked when socket buffer space
1598 * @sk: socket whose state has changed
1600 * Called when more output buffer space is available for this socket.
1601 * We try not to wake our writers until they can make "significant"
1602 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1603 * with a bunch of small requests.
1605 static void xs_udp_write_space(struct sock *sk)
1607 read_lock_bh(&sk->sk_callback_lock);
1609 /* from net/core/sock.c:sock_def_write_space */
1610 if (sock_writeable(sk))
1613 read_unlock_bh(&sk->sk_callback_lock);
1617 * xs_tcp_write_space - callback invoked when socket buffer space
1619 * @sk: socket whose state has changed
1621 * Called when more output buffer space is available for this socket.
1622 * We try not to wake our writers until they can make "significant"
1623 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1624 * with a bunch of small requests.
1626 static void xs_tcp_write_space(struct sock *sk)
1628 read_lock_bh(&sk->sk_callback_lock);
1630 /* from net/core/stream.c:sk_stream_write_space */
1631 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk))
1634 read_unlock_bh(&sk->sk_callback_lock);
1637 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1639 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1640 struct sock *sk = transport->inet;
1642 if (transport->rcvsize) {
1643 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1644 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1646 if (transport->sndsize) {
1647 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1648 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1649 sk->sk_write_space(sk);
1654 * xs_udp_set_buffer_size - set send and receive limits
1655 * @xprt: generic transport
1656 * @sndsize: requested size of send buffer, in bytes
1657 * @rcvsize: requested size of receive buffer, in bytes
1659 * Set socket send and receive buffer size limits.
1661 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1663 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1665 transport->sndsize = 0;
1667 transport->sndsize = sndsize + 1024;
1668 transport->rcvsize = 0;
1670 transport->rcvsize = rcvsize + 1024;
1672 xs_udp_do_set_buffer_size(xprt);
1676 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1677 * @task: task that timed out
1679 * Adjust the congestion window after a retransmit timeout has occurred.
1681 static void xs_udp_timer(struct rpc_task *task)
1683 xprt_adjust_cwnd(task, -ETIMEDOUT);
1686 static unsigned short xs_get_random_port(void)
1688 unsigned short range = xprt_max_resvport - xprt_min_resvport;
1689 unsigned short rand = (unsigned short) net_random() % range;
1690 return rand + xprt_min_resvport;
1694 * xs_set_port - reset the port number in the remote endpoint address
1695 * @xprt: generic transport
1696 * @port: new port number
1699 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1701 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1703 rpc_set_port(xs_addr(xprt), port);
1704 xs_update_peer_port(xprt);
1707 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1709 unsigned short port = transport->srcport;
1711 if (port == 0 && transport->xprt.resvport)
1712 port = xs_get_random_port();
1716 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1718 if (transport->srcport != 0)
1719 transport->srcport = 0;
1720 if (!transport->xprt.resvport)
1722 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1723 return xprt_max_resvport;
1726 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1728 struct sockaddr_storage myaddr;
1730 unsigned short port = xs_get_srcport(transport);
1731 unsigned short last;
1733 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1735 rpc_set_port((struct sockaddr *)&myaddr, port);
1736 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1737 transport->xprt.addrlen);
1741 transport->srcport = port;
1745 port = xs_next_srcport(transport, port);
1748 } while (err == -EADDRINUSE && nloop != 2);
1750 if (myaddr.ss_family == AF_INET)
1751 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1752 &((struct sockaddr_in *)&myaddr)->sin_addr,
1753 port, err ? "failed" : "ok", err);
1755 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1756 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1757 port, err ? "failed" : "ok", err);
1762 * We don't support autobind on AF_LOCAL sockets
1764 static void xs_local_rpcbind(struct rpc_task *task)
1766 xprt_set_bound(task->tk_xprt);
1769 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1773 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1774 static struct lock_class_key xs_key[2];
1775 static struct lock_class_key xs_slock_key[2];
1777 static inline void xs_reclassify_socketu(struct socket *sock)
1779 struct sock *sk = sock->sk;
1781 BUG_ON(sock_owned_by_user(sk));
1782 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1783 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1786 static inline void xs_reclassify_socket4(struct socket *sock)
1788 struct sock *sk = sock->sk;
1790 BUG_ON(sock_owned_by_user(sk));
1791 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1792 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1795 static inline void xs_reclassify_socket6(struct socket *sock)
1797 struct sock *sk = sock->sk;
1799 BUG_ON(sock_owned_by_user(sk));
1800 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1801 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1804 static inline void xs_reclassify_socket(int family, struct socket *sock)
1808 xs_reclassify_socketu(sock);
1811 xs_reclassify_socket4(sock);
1814 xs_reclassify_socket6(sock);
1819 static inline void xs_reclassify_socketu(struct socket *sock)
1823 static inline void xs_reclassify_socket4(struct socket *sock)
1827 static inline void xs_reclassify_socket6(struct socket *sock)
1831 static inline void xs_reclassify_socket(int family, struct socket *sock)
1836 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1837 struct sock_xprt *transport, int family, int type, int protocol)
1839 struct socket *sock;
1842 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1844 dprintk("RPC: can't create %d transport socket (%d).\n",
1848 xs_reclassify_socket(family, sock);
1850 err = xs_bind(transport, sock);
1858 return ERR_PTR(err);
1861 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1862 struct socket *sock)
1864 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1867 if (!transport->inet) {
1868 struct sock *sk = sock->sk;
1870 write_lock_bh(&sk->sk_callback_lock);
1872 xs_save_old_callbacks(transport, sk);
1874 sk->sk_user_data = xprt;
1875 sk->sk_data_ready = xs_local_data_ready;
1876 sk->sk_write_space = xs_udp_write_space;
1877 sk->sk_allocation = GFP_ATOMIC;
1879 xprt_clear_connected(xprt);
1881 /* Reset to new socket */
1882 transport->sock = sock;
1883 transport->inet = sk;
1885 write_unlock_bh(&sk->sk_callback_lock);
1888 /* Tell the socket layer to start connecting... */
1889 xprt->stat.connect_count++;
1890 xprt->stat.connect_start = jiffies;
1891 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1895 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1896 * @xprt: RPC transport to connect
1897 * @transport: socket transport to connect
1898 * @create_sock: function to create a socket of the correct type
1900 * Invoked by a work queue tasklet.
1902 static void xs_local_setup_socket(struct work_struct *work)
1904 struct sock_xprt *transport =
1905 container_of(work, struct sock_xprt, connect_worker.work);
1906 struct rpc_xprt *xprt = &transport->xprt;
1907 struct socket *sock;
1913 current->flags |= PF_FSTRANS;
1915 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1916 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1917 SOCK_STREAM, 0, &sock, 1);
1919 dprintk("RPC: can't create AF_LOCAL "
1920 "transport socket (%d).\n", -status);
1923 xs_reclassify_socketu(sock);
1925 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1926 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1928 status = xs_local_finish_connecting(xprt, sock);
1931 dprintk("RPC: xprt %p connected to %s\n",
1932 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1933 xprt_set_connected(xprt);
1936 dprintk("RPC: xprt %p: socket %s does not exist\n",
1937 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1940 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1942 xprt->address_strings[RPC_DISPLAY_ADDR]);
1946 xprt_clear_connecting(xprt);
1947 xprt_wake_pending_tasks(xprt, status);
1948 current->flags &= ~PF_FSTRANS;
1951 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1953 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1955 if (!transport->inet) {
1956 struct sock *sk = sock->sk;
1958 write_lock_bh(&sk->sk_callback_lock);
1960 xs_save_old_callbacks(transport, sk);
1962 sk->sk_user_data = xprt;
1963 sk->sk_data_ready = xs_udp_data_ready;
1964 sk->sk_write_space = xs_udp_write_space;
1965 sk->sk_no_check = UDP_CSUM_NORCV;
1966 sk->sk_allocation = GFP_ATOMIC;
1968 xprt_set_connected(xprt);
1970 /* Reset to new socket */
1971 transport->sock = sock;
1972 transport->inet = sk;
1974 write_unlock_bh(&sk->sk_callback_lock);
1976 xs_udp_do_set_buffer_size(xprt);
1979 static void xs_udp_setup_socket(struct work_struct *work)
1981 struct sock_xprt *transport =
1982 container_of(work, struct sock_xprt, connect_worker.work);
1983 struct rpc_xprt *xprt = &transport->xprt;
1984 struct socket *sock = transport->sock;
1990 current->flags |= PF_FSTRANS;
1992 /* Start by resetting any existing state */
1993 xs_reset_transport(transport);
1994 sock = xs_create_sock(xprt, transport,
1995 xs_addr(xprt)->sa_family, SOCK_DGRAM, IPPROTO_UDP);
1999 dprintk("RPC: worker connecting xprt %p via %s to "
2000 "%s (port %s)\n", xprt,
2001 xprt->address_strings[RPC_DISPLAY_PROTO],
2002 xprt->address_strings[RPC_DISPLAY_ADDR],
2003 xprt->address_strings[RPC_DISPLAY_PORT]);
2005 xs_udp_finish_connecting(xprt, sock);
2008 xprt_clear_connecting(xprt);
2009 xprt_wake_pending_tasks(xprt, status);
2010 current->flags &= ~PF_FSTRANS;
2014 * We need to preserve the port number so the reply cache on the server can
2015 * find our cached RPC replies when we get around to reconnecting.
2017 static void xs_abort_connection(struct sock_xprt *transport)
2020 struct sockaddr any;
2022 dprintk("RPC: disconnecting xprt %p to reuse port\n", transport);
2025 * Disconnect the transport socket by doing a connect operation
2026 * with AF_UNSPEC. This should return immediately...
2028 memset(&any, 0, sizeof(any));
2029 any.sa_family = AF_UNSPEC;
2030 result = kernel_connect(transport->sock, &any, sizeof(any), 0);
2032 xs_sock_reset_connection_flags(&transport->xprt);
2033 dprintk("RPC: AF_UNSPEC connect return code %d\n", result);
2036 static void xs_tcp_reuse_connection(struct sock_xprt *transport)
2038 unsigned int state = transport->inet->sk_state;
2040 if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED) {
2041 /* we don't need to abort the connection if the socket
2042 * hasn't undergone a shutdown
2044 if (transport->inet->sk_shutdown == 0)
2046 dprintk("RPC: %s: TCP_CLOSEd and sk_shutdown set to %d\n",
2047 __func__, transport->inet->sk_shutdown);
2049 if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT)) {
2050 /* we don't need to abort the connection if the socket
2051 * hasn't undergone a shutdown
2053 if (transport->inet->sk_shutdown == 0)
2055 dprintk("RPC: %s: ESTABLISHED/SYN_SENT "
2056 "sk_shutdown set to %d\n",
2057 __func__, transport->inet->sk_shutdown);
2059 xs_abort_connection(transport);
2062 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2064 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2065 int ret = -ENOTCONN;
2067 if (!transport->inet) {
2068 struct sock *sk = sock->sk;
2070 write_lock_bh(&sk->sk_callback_lock);
2072 xs_save_old_callbacks(transport, sk);
2074 sk->sk_user_data = xprt;
2075 sk->sk_data_ready = xs_tcp_data_ready;
2076 sk->sk_state_change = xs_tcp_state_change;
2077 sk->sk_write_space = xs_tcp_write_space;
2078 sk->sk_allocation = GFP_ATOMIC;
2080 /* socket options */
2081 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
2082 sock_reset_flag(sk, SOCK_LINGER);
2083 tcp_sk(sk)->linger2 = 0;
2084 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2086 xprt_clear_connected(xprt);
2088 /* Reset to new socket */
2089 transport->sock = sock;
2090 transport->inet = sk;
2092 write_unlock_bh(&sk->sk_callback_lock);
2095 if (!xprt_bound(xprt))
2098 /* Tell the socket layer to start connecting... */
2099 xprt->stat.connect_count++;
2100 xprt->stat.connect_start = jiffies;
2101 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2106 xprt->connect_cookie++;
2107 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2108 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2115 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2116 * @xprt: RPC transport to connect
2117 * @transport: socket transport to connect
2118 * @create_sock: function to create a socket of the correct type
2120 * Invoked by a work queue tasklet.
2122 static void xs_tcp_setup_socket(struct work_struct *work)
2124 struct sock_xprt *transport =
2125 container_of(work, struct sock_xprt, connect_worker.work);
2126 struct socket *sock = transport->sock;
2127 struct rpc_xprt *xprt = &transport->xprt;
2133 current->flags |= PF_FSTRANS;
2136 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
2137 sock = xs_create_sock(xprt, transport,
2138 xs_addr(xprt)->sa_family, SOCK_STREAM, IPPROTO_TCP);
2140 status = PTR_ERR(sock);
2146 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
2148 /* "close" the socket, preserving the local port */
2149 xs_tcp_reuse_connection(transport);
2155 dprintk("RPC: worker connecting xprt %p via %s to "
2156 "%s (port %s)\n", xprt,
2157 xprt->address_strings[RPC_DISPLAY_PROTO],
2158 xprt->address_strings[RPC_DISPLAY_ADDR],
2159 xprt->address_strings[RPC_DISPLAY_PORT]);
2161 status = xs_tcp_finish_connecting(xprt, sock);
2162 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2163 xprt, -status, xprt_connected(xprt),
2164 sock->sk->sk_state);
2167 printk("%s: connect returned unhandled error %d\n",
2169 case -EADDRNOTAVAIL:
2170 /* We're probably in TIME_WAIT. Get rid of existing socket,
2173 xs_tcp_force_close(xprt);
2178 /* retry with existing socket, after a delay */
2182 xprt_clear_connecting(xprt);
2183 current->flags &= ~PF_FSTRANS;
2186 /* Happens, for instance, if the user specified a link
2187 * local IPv6 address without a scope-id.
2194 xprt_clear_connecting(xprt);
2195 xprt_wake_pending_tasks(xprt, status);
2196 current->flags &= ~PF_FSTRANS;
2200 * xs_connect - connect a socket to a remote endpoint
2201 * @task: address of RPC task that manages state of connect request
2203 * TCP: If the remote end dropped the connection, delay reconnecting.
2205 * UDP socket connects are synchronous, but we use a work queue anyway
2206 * to guarantee that even unprivileged user processes can set up a
2207 * socket on a privileged port.
2209 * If a UDP socket connect fails, the delay behavior here prevents
2210 * retry floods (hard mounts).
2212 static void xs_connect(struct rpc_task *task)
2214 struct rpc_xprt *xprt = task->tk_xprt;
2215 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2217 if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
2218 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2220 xprt, xprt->reestablish_timeout / HZ);
2221 queue_delayed_work(rpciod_workqueue,
2222 &transport->connect_worker,
2223 xprt->reestablish_timeout);
2224 xprt->reestablish_timeout <<= 1;
2225 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2226 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2227 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2228 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2230 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2231 queue_delayed_work(rpciod_workqueue,
2232 &transport->connect_worker, 0);
2237 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2238 * @xprt: rpc_xprt struct containing statistics
2242 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2246 if (xprt_connected(xprt))
2247 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2249 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2251 xprt->stat.bind_count,
2252 xprt->stat.connect_count,
2253 xprt->stat.connect_time,
2257 xprt->stat.bad_xids,
2259 xprt->stat.bklog_u);
2263 * xs_udp_print_stats - display UDP socket-specifc stats
2264 * @xprt: rpc_xprt struct containing statistics
2268 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2270 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2272 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
2274 xprt->stat.bind_count,
2277 xprt->stat.bad_xids,
2279 xprt->stat.bklog_u);
2283 * xs_tcp_print_stats - display TCP socket-specifc stats
2284 * @xprt: rpc_xprt struct containing statistics
2288 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2290 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2293 if (xprt_connected(xprt))
2294 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2296 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
2298 xprt->stat.bind_count,
2299 xprt->stat.connect_count,
2300 xprt->stat.connect_time,
2304 xprt->stat.bad_xids,
2306 xprt->stat.bklog_u);
2310 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2311 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2312 * to use the server side send routines.
2314 static void *bc_malloc(struct rpc_task *task, size_t size)
2317 struct rpc_buffer *buf;
2319 BUG_ON(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2320 page = alloc_page(GFP_KERNEL);
2325 buf = page_address(page);
2326 buf->len = PAGE_SIZE;
2332 * Free the space allocated in the bc_alloc routine
2334 static void bc_free(void *buffer)
2336 struct rpc_buffer *buf;
2341 buf = container_of(buffer, struct rpc_buffer, data);
2342 free_page((unsigned long)buf);
2346 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2347 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2349 static int bc_sendto(struct rpc_rqst *req)
2352 struct xdr_buf *xbufp = &req->rq_snd_buf;
2353 struct rpc_xprt *xprt = req->rq_xprt;
2354 struct sock_xprt *transport =
2355 container_of(xprt, struct sock_xprt, xprt);
2356 struct socket *sock = transport->sock;
2357 unsigned long headoff;
2358 unsigned long tailoff;
2360 xs_encode_stream_record_marker(xbufp);
2362 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2363 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2364 len = svc_send_common(sock, xbufp,
2365 virt_to_page(xbufp->head[0].iov_base), headoff,
2366 xbufp->tail[0].iov_base, tailoff);
2368 if (len != xbufp->len) {
2369 printk(KERN_NOTICE "Error sending entire callback!\n");
2377 * The send routine. Borrows from svc_send
2379 static int bc_send_request(struct rpc_task *task)
2381 struct rpc_rqst *req = task->tk_rqstp;
2382 struct svc_xprt *xprt;
2383 struct svc_sock *svsk;
2386 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2388 * Get the server socket associated with this callback xprt
2390 xprt = req->rq_xprt->bc_xprt;
2391 svsk = container_of(xprt, struct svc_sock, sk_xprt);
2394 * Grab the mutex to serialize data as the connection is shared
2395 * with the fore channel
2397 if (!mutex_trylock(&xprt->xpt_mutex)) {
2398 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2399 if (!mutex_trylock(&xprt->xpt_mutex))
2401 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2403 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2406 len = bc_sendto(req);
2407 mutex_unlock(&xprt->xpt_mutex);
2416 * The close routine. Since this is client initiated, we do nothing
2419 static void bc_close(struct rpc_xprt *xprt)
2424 * The xprt destroy routine. Again, because this connection is client
2425 * initiated, we do nothing
2428 static void bc_destroy(struct rpc_xprt *xprt)
2432 static struct rpc_xprt_ops xs_local_ops = {
2433 .reserve_xprt = xprt_reserve_xprt,
2434 .release_xprt = xs_tcp_release_xprt,
2435 .alloc_slot = xprt_alloc_slot,
2436 .rpcbind = xs_local_rpcbind,
2437 .set_port = xs_local_set_port,
2438 .connect = xs_connect,
2439 .buf_alloc = rpc_malloc,
2440 .buf_free = rpc_free,
2441 .send_request = xs_local_send_request,
2442 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2444 .destroy = xs_destroy,
2445 .print_stats = xs_local_print_stats,
2448 static struct rpc_xprt_ops xs_udp_ops = {
2449 .set_buffer_size = xs_udp_set_buffer_size,
2450 .reserve_xprt = xprt_reserve_xprt_cong,
2451 .release_xprt = xprt_release_xprt_cong,
2452 .alloc_slot = xprt_alloc_slot,
2453 .rpcbind = rpcb_getport_async,
2454 .set_port = xs_set_port,
2455 .connect = xs_connect,
2456 .buf_alloc = rpc_malloc,
2457 .buf_free = rpc_free,
2458 .send_request = xs_udp_send_request,
2459 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
2460 .timer = xs_udp_timer,
2461 .release_request = xprt_release_rqst_cong,
2463 .destroy = xs_destroy,
2464 .print_stats = xs_udp_print_stats,
2467 static struct rpc_xprt_ops xs_tcp_ops = {
2468 .reserve_xprt = xprt_reserve_xprt,
2469 .release_xprt = xs_tcp_release_xprt,
2470 .alloc_slot = xprt_lock_and_alloc_slot,
2471 .rpcbind = rpcb_getport_async,
2472 .set_port = xs_set_port,
2473 .connect = xs_connect,
2474 .buf_alloc = rpc_malloc,
2475 .buf_free = rpc_free,
2476 .send_request = xs_tcp_send_request,
2477 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2478 .close = xs_tcp_close,
2479 .destroy = xs_destroy,
2480 .print_stats = xs_tcp_print_stats,
2484 * The rpc_xprt_ops for the server backchannel
2487 static struct rpc_xprt_ops bc_tcp_ops = {
2488 .reserve_xprt = xprt_reserve_xprt,
2489 .release_xprt = xprt_release_xprt,
2490 .alloc_slot = xprt_alloc_slot,
2491 .buf_alloc = bc_malloc,
2492 .buf_free = bc_free,
2493 .send_request = bc_send_request,
2494 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2496 .destroy = bc_destroy,
2497 .print_stats = xs_tcp_print_stats,
2500 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2502 static const struct sockaddr_in sin = {
2503 .sin_family = AF_INET,
2504 .sin_addr.s_addr = htonl(INADDR_ANY),
2506 static const struct sockaddr_in6 sin6 = {
2507 .sin6_family = AF_INET6,
2508 .sin6_addr = IN6ADDR_ANY_INIT,
2515 memcpy(sap, &sin, sizeof(sin));
2518 memcpy(sap, &sin6, sizeof(sin6));
2521 dprintk("RPC: %s: Bad address family\n", __func__);
2522 return -EAFNOSUPPORT;
2527 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2528 unsigned int slot_table_size,
2529 unsigned int max_slot_table_size)
2531 struct rpc_xprt *xprt;
2532 struct sock_xprt *new;
2534 if (args->addrlen > sizeof(xprt->addr)) {
2535 dprintk("RPC: xs_setup_xprt: address too large\n");
2536 return ERR_PTR(-EBADF);
2539 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2540 max_slot_table_size);
2542 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2544 return ERR_PTR(-ENOMEM);
2547 new = container_of(xprt, struct sock_xprt, xprt);
2548 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2549 xprt->addrlen = args->addrlen;
2551 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2554 err = xs_init_anyaddr(args->dstaddr->sa_family,
2555 (struct sockaddr *)&new->srcaddr);
2558 return ERR_PTR(err);
2565 static const struct rpc_timeout xs_local_default_timeout = {
2566 .to_initval = 10 * HZ,
2567 .to_maxval = 10 * HZ,
2572 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2573 * @args: rpc transport creation arguments
2575 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2577 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2579 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2580 struct sock_xprt *transport;
2581 struct rpc_xprt *xprt;
2582 struct rpc_xprt *ret;
2584 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2585 xprt_max_tcp_slot_table_entries);
2588 transport = container_of(xprt, struct sock_xprt, xprt);
2591 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2592 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2594 xprt->bind_timeout = XS_BIND_TO;
2595 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2596 xprt->idle_timeout = XS_IDLE_DISC_TO;
2598 xprt->ops = &xs_local_ops;
2599 xprt->timeout = &xs_local_default_timeout;
2601 switch (sun->sun_family) {
2603 if (sun->sun_path[0] != '/') {
2604 dprintk("RPC: bad AF_LOCAL address: %s\n",
2606 ret = ERR_PTR(-EINVAL);
2609 xprt_set_bound(xprt);
2610 INIT_DELAYED_WORK(&transport->connect_worker,
2611 xs_local_setup_socket);
2612 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2615 ret = ERR_PTR(-EAFNOSUPPORT);
2619 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2620 xprt->address_strings[RPC_DISPLAY_ADDR]);
2622 if (try_module_get(THIS_MODULE))
2624 ret = ERR_PTR(-EINVAL);
2630 static const struct rpc_timeout xs_udp_default_timeout = {
2631 .to_initval = 5 * HZ,
2632 .to_maxval = 30 * HZ,
2633 .to_increment = 5 * HZ,
2638 * xs_setup_udp - Set up transport to use a UDP socket
2639 * @args: rpc transport creation arguments
2642 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2644 struct sockaddr *addr = args->dstaddr;
2645 struct rpc_xprt *xprt;
2646 struct sock_xprt *transport;
2647 struct rpc_xprt *ret;
2649 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2650 xprt_udp_slot_table_entries);
2653 transport = container_of(xprt, struct sock_xprt, xprt);
2655 xprt->prot = IPPROTO_UDP;
2657 /* XXX: header size can vary due to auth type, IPv6, etc. */
2658 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2660 xprt->bind_timeout = XS_BIND_TO;
2661 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2662 xprt->idle_timeout = XS_IDLE_DISC_TO;
2664 xprt->ops = &xs_udp_ops;
2666 xprt->timeout = &xs_udp_default_timeout;
2668 switch (addr->sa_family) {
2670 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2671 xprt_set_bound(xprt);
2673 INIT_DELAYED_WORK(&transport->connect_worker,
2674 xs_udp_setup_socket);
2675 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2678 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2679 xprt_set_bound(xprt);
2681 INIT_DELAYED_WORK(&transport->connect_worker,
2682 xs_udp_setup_socket);
2683 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2686 ret = ERR_PTR(-EAFNOSUPPORT);
2690 if (xprt_bound(xprt))
2691 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2692 xprt->address_strings[RPC_DISPLAY_ADDR],
2693 xprt->address_strings[RPC_DISPLAY_PORT],
2694 xprt->address_strings[RPC_DISPLAY_PROTO]);
2696 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2697 xprt->address_strings[RPC_DISPLAY_ADDR],
2698 xprt->address_strings[RPC_DISPLAY_PROTO]);
2700 if (try_module_get(THIS_MODULE))
2702 ret = ERR_PTR(-EINVAL);
2708 static const struct rpc_timeout xs_tcp_default_timeout = {
2709 .to_initval = 60 * HZ,
2710 .to_maxval = 60 * HZ,
2715 * xs_setup_tcp - Set up transport to use a TCP socket
2716 * @args: rpc transport creation arguments
2719 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2721 struct sockaddr *addr = args->dstaddr;
2722 struct rpc_xprt *xprt;
2723 struct sock_xprt *transport;
2724 struct rpc_xprt *ret;
2726 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2727 xprt_max_tcp_slot_table_entries);
2730 transport = container_of(xprt, struct sock_xprt, xprt);
2732 xprt->prot = IPPROTO_TCP;
2733 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2734 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2736 xprt->bind_timeout = XS_BIND_TO;
2737 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2738 xprt->idle_timeout = XS_IDLE_DISC_TO;
2740 xprt->ops = &xs_tcp_ops;
2741 xprt->timeout = &xs_tcp_default_timeout;
2743 switch (addr->sa_family) {
2745 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2746 xprt_set_bound(xprt);
2748 INIT_DELAYED_WORK(&transport->connect_worker,
2749 xs_tcp_setup_socket);
2750 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2753 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2754 xprt_set_bound(xprt);
2756 INIT_DELAYED_WORK(&transport->connect_worker,
2757 xs_tcp_setup_socket);
2758 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2761 ret = ERR_PTR(-EAFNOSUPPORT);
2765 if (xprt_bound(xprt))
2766 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2767 xprt->address_strings[RPC_DISPLAY_ADDR],
2768 xprt->address_strings[RPC_DISPLAY_PORT],
2769 xprt->address_strings[RPC_DISPLAY_PROTO]);
2771 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2772 xprt->address_strings[RPC_DISPLAY_ADDR],
2773 xprt->address_strings[RPC_DISPLAY_PROTO]);
2776 if (try_module_get(THIS_MODULE))
2778 ret = ERR_PTR(-EINVAL);
2785 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2786 * @args: rpc transport creation arguments
2789 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2791 struct sockaddr *addr = args->dstaddr;
2792 struct rpc_xprt *xprt;
2793 struct sock_xprt *transport;
2794 struct svc_sock *bc_sock;
2795 struct rpc_xprt *ret;
2797 if (args->bc_xprt->xpt_bc_xprt) {
2799 * This server connection already has a backchannel
2800 * export; we can't create a new one, as we wouldn't be
2801 * able to match replies based on xid any more. So,
2802 * reuse the already-existing one:
2804 return args->bc_xprt->xpt_bc_xprt;
2806 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2807 xprt_tcp_slot_table_entries);
2810 transport = container_of(xprt, struct sock_xprt, xprt);
2812 xprt->prot = IPPROTO_TCP;
2813 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2814 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2815 xprt->timeout = &xs_tcp_default_timeout;
2818 xprt_set_bound(xprt);
2819 xprt->bind_timeout = 0;
2820 xprt->reestablish_timeout = 0;
2821 xprt->idle_timeout = 0;
2823 xprt->ops = &bc_tcp_ops;
2825 switch (addr->sa_family) {
2827 xs_format_peer_addresses(xprt, "tcp",
2831 xs_format_peer_addresses(xprt, "tcp",
2832 RPCBIND_NETID_TCP6);
2835 ret = ERR_PTR(-EAFNOSUPPORT);
2839 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2840 xprt->address_strings[RPC_DISPLAY_ADDR],
2841 xprt->address_strings[RPC_DISPLAY_PORT],
2842 xprt->address_strings[RPC_DISPLAY_PROTO]);
2845 * Once we've associated a backchannel xprt with a connection,
2846 * we want to keep it around as long as long as the connection
2847 * lasts, in case we need to start using it for a backchannel
2848 * again; this reference won't be dropped until bc_xprt is
2852 args->bc_xprt->xpt_bc_xprt = xprt;
2853 xprt->bc_xprt = args->bc_xprt;
2854 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2855 transport->sock = bc_sock->sk_sock;
2856 transport->inet = bc_sock->sk_sk;
2859 * Since we don't want connections for the backchannel, we set
2860 * the xprt status to connected
2862 xprt_set_connected(xprt);
2865 if (try_module_get(THIS_MODULE))
2868 ret = ERR_PTR(-EINVAL);
2874 static struct xprt_class xs_local_transport = {
2875 .list = LIST_HEAD_INIT(xs_local_transport.list),
2876 .name = "named UNIX socket",
2877 .owner = THIS_MODULE,
2878 .ident = XPRT_TRANSPORT_LOCAL,
2879 .setup = xs_setup_local,
2882 static struct xprt_class xs_udp_transport = {
2883 .list = LIST_HEAD_INIT(xs_udp_transport.list),
2885 .owner = THIS_MODULE,
2886 .ident = XPRT_TRANSPORT_UDP,
2887 .setup = xs_setup_udp,
2890 static struct xprt_class xs_tcp_transport = {
2891 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
2893 .owner = THIS_MODULE,
2894 .ident = XPRT_TRANSPORT_TCP,
2895 .setup = xs_setup_tcp,
2898 static struct xprt_class xs_bc_tcp_transport = {
2899 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
2900 .name = "tcp NFSv4.1 backchannel",
2901 .owner = THIS_MODULE,
2902 .ident = XPRT_TRANSPORT_BC_TCP,
2903 .setup = xs_setup_bc_tcp,
2907 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2910 int init_socket_xprt(void)
2913 if (!sunrpc_table_header)
2914 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2917 xprt_register_transport(&xs_local_transport);
2918 xprt_register_transport(&xs_udp_transport);
2919 xprt_register_transport(&xs_tcp_transport);
2920 xprt_register_transport(&xs_bc_tcp_transport);
2926 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2929 void cleanup_socket_xprt(void)
2932 if (sunrpc_table_header) {
2933 unregister_sysctl_table(sunrpc_table_header);
2934 sunrpc_table_header = NULL;
2938 xprt_unregister_transport(&xs_local_transport);
2939 xprt_unregister_transport(&xs_udp_transport);
2940 xprt_unregister_transport(&xs_tcp_transport);
2941 xprt_unregister_transport(&xs_bc_tcp_transport);
2944 static int param_set_uint_minmax(const char *val,
2945 const struct kernel_param *kp,
2946 unsigned int min, unsigned int max)
2953 ret = strict_strtoul(val, 0, &num);
2954 if (ret == -EINVAL || num < min || num > max)
2956 *((unsigned int *)kp->arg) = num;
2960 static int param_set_portnr(const char *val, const struct kernel_param *kp)
2962 return param_set_uint_minmax(val, kp,
2967 static struct kernel_param_ops param_ops_portnr = {
2968 .set = param_set_portnr,
2969 .get = param_get_uint,
2972 #define param_check_portnr(name, p) \
2973 __param_check(name, p, unsigned int);
2975 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
2976 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
2978 static int param_set_slot_table_size(const char *val,
2979 const struct kernel_param *kp)
2981 return param_set_uint_minmax(val, kp,
2983 RPC_MAX_SLOT_TABLE);
2986 static struct kernel_param_ops param_ops_slot_table_size = {
2987 .set = param_set_slot_table_size,
2988 .get = param_get_uint,
2991 #define param_check_slot_table_size(name, p) \
2992 __param_check(name, p, unsigned int);
2994 static int param_set_max_slot_table_size(const char *val,
2995 const struct kernel_param *kp)
2997 return param_set_uint_minmax(val, kp,
2999 RPC_MAX_SLOT_TABLE_LIMIT);
3002 static struct kernel_param_ops param_ops_max_slot_table_size = {
3003 .set = param_set_max_slot_table_size,
3004 .get = param_get_uint,
3007 #define param_check_max_slot_table_size(name, p) \
3008 __param_check(name, p, unsigned int);
3010 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3011 slot_table_size, 0644);
3012 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3013 max_slot_table_size, 0644);
3014 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3015 slot_table_size, 0644);