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/addr.h>
37 #include <linux/sunrpc/sched.h>
38 #include <linux/sunrpc/svcsock.h>
39 #include <linux/sunrpc/xprtsock.h>
40 #include <linux/file.h>
41 #ifdef CONFIG_SUNRPC_BACKCHANNEL
42 #include <linux/sunrpc/bc_xprt.h>
46 #include <net/checksum.h>
50 #include <trace/events/sunrpc.h>
54 static void xs_close(struct rpc_xprt *xprt);
59 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
60 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
61 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
63 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
64 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
66 #define XS_TCP_LINGER_TO (15U * HZ)
67 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
70 * We can register our own files under /proc/sys/sunrpc by
71 * calling register_sysctl_table() again. The files in that
72 * directory become the union of all files registered there.
74 * We simply need to make sure that we don't collide with
75 * someone else's file names!
80 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
81 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
82 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
83 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
84 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
86 static struct ctl_table_header *sunrpc_table_header;
89 * FIXME: changing the UDP slot table size should also resize the UDP
90 * socket buffers for existing UDP transports
92 static struct ctl_table xs_tunables_table[] = {
94 .procname = "udp_slot_table_entries",
95 .data = &xprt_udp_slot_table_entries,
96 .maxlen = sizeof(unsigned int),
98 .proc_handler = proc_dointvec_minmax,
99 .extra1 = &min_slot_table_size,
100 .extra2 = &max_slot_table_size
103 .procname = "tcp_slot_table_entries",
104 .data = &xprt_tcp_slot_table_entries,
105 .maxlen = sizeof(unsigned int),
107 .proc_handler = proc_dointvec_minmax,
108 .extra1 = &min_slot_table_size,
109 .extra2 = &max_slot_table_size
112 .procname = "tcp_max_slot_table_entries",
113 .data = &xprt_max_tcp_slot_table_entries,
114 .maxlen = sizeof(unsigned int),
116 .proc_handler = proc_dointvec_minmax,
117 .extra1 = &min_slot_table_size,
118 .extra2 = &max_tcp_slot_table_limit
121 .procname = "min_resvport",
122 .data = &xprt_min_resvport,
123 .maxlen = sizeof(unsigned int),
125 .proc_handler = proc_dointvec_minmax,
126 .extra1 = &xprt_min_resvport_limit,
127 .extra2 = &xprt_max_resvport_limit
130 .procname = "max_resvport",
131 .data = &xprt_max_resvport,
132 .maxlen = sizeof(unsigned int),
134 .proc_handler = proc_dointvec_minmax,
135 .extra1 = &xprt_min_resvport_limit,
136 .extra2 = &xprt_max_resvport_limit
139 .procname = "tcp_fin_timeout",
140 .data = &xs_tcp_fin_timeout,
141 .maxlen = sizeof(xs_tcp_fin_timeout),
143 .proc_handler = proc_dointvec_jiffies,
148 static struct ctl_table sunrpc_table[] = {
150 .procname = "sunrpc",
152 .child = xs_tunables_table
160 * Wait duration for a reply from the RPC portmapper.
162 #define XS_BIND_TO (60U * HZ)
165 * Delay if a UDP socket connect error occurs. This is most likely some
166 * kind of resource problem on the local host.
168 #define XS_UDP_REEST_TO (2U * HZ)
171 * The reestablish timeout allows clients to delay for a bit before attempting
172 * to reconnect to a server that just dropped our connection.
174 * We implement an exponential backoff when trying to reestablish a TCP
175 * transport connection with the server. Some servers like to drop a TCP
176 * connection when they are overworked, so we start with a short timeout and
177 * increase over time if the server is down or not responding.
179 #define XS_TCP_INIT_REEST_TO (3U * HZ)
180 #define XS_TCP_MAX_REEST_TO (5U * 60 * HZ)
183 * TCP idle timeout; client drops the transport socket if it is idle
184 * for this long. Note that we also timeout UDP sockets to prevent
185 * holding port numbers when there is no RPC traffic.
187 #define XS_IDLE_DISC_TO (5U * 60 * HZ)
190 # undef RPC_DEBUG_DATA
191 # define RPCDBG_FACILITY RPCDBG_TRANS
194 #ifdef RPC_DEBUG_DATA
195 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
197 u8 *buf = (u8 *) packet;
200 dprintk("RPC: %s\n", msg);
201 for (j = 0; j < count && j < 128; j += 4) {
205 dprintk("0x%04x ", j);
207 dprintk("%02x%02x%02x%02x ",
208 buf[j], buf[j+1], buf[j+2], buf[j+3]);
213 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
220 struct rpc_xprt xprt;
225 struct socket * sock;
229 * State of TCP reply receive
238 unsigned long tcp_copied,
242 * Connection of transports
244 struct delayed_work connect_worker;
245 struct sockaddr_storage srcaddr;
246 unsigned short srcport;
249 * UDP socket buffer size parameters
255 * Saved socket callback addresses
257 void (*old_data_ready)(struct sock *, int);
258 void (*old_state_change)(struct sock *);
259 void (*old_write_space)(struct sock *);
260 void (*old_error_report)(struct sock *);
264 * TCP receive state flags
266 #define TCP_RCV_LAST_FRAG (1UL << 0)
267 #define TCP_RCV_COPY_FRAGHDR (1UL << 1)
268 #define TCP_RCV_COPY_XID (1UL << 2)
269 #define TCP_RCV_COPY_DATA (1UL << 3)
270 #define TCP_RCV_READ_CALLDIR (1UL << 4)
271 #define TCP_RCV_COPY_CALLDIR (1UL << 5)
276 #define TCP_RPC_REPLY (1UL << 6)
278 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
280 return (struct rpc_xprt *) sk->sk_user_data;
283 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
285 return (struct sockaddr *) &xprt->addr;
288 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
290 return (struct sockaddr_un *) &xprt->addr;
293 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
295 return (struct sockaddr_in *) &xprt->addr;
298 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
300 return (struct sockaddr_in6 *) &xprt->addr;
303 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
305 struct sockaddr *sap = xs_addr(xprt);
306 struct sockaddr_in6 *sin6;
307 struct sockaddr_in *sin;
308 struct sockaddr_un *sun;
311 switch (sap->sa_family) {
313 sun = xs_addr_un(xprt);
314 strlcpy(buf, sun->sun_path, sizeof(buf));
315 xprt->address_strings[RPC_DISPLAY_ADDR] =
316 kstrdup(buf, GFP_KERNEL);
319 (void)rpc_ntop(sap, buf, sizeof(buf));
320 xprt->address_strings[RPC_DISPLAY_ADDR] =
321 kstrdup(buf, GFP_KERNEL);
322 sin = xs_addr_in(xprt);
323 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
326 (void)rpc_ntop(sap, buf, sizeof(buf));
327 xprt->address_strings[RPC_DISPLAY_ADDR] =
328 kstrdup(buf, GFP_KERNEL);
329 sin6 = xs_addr_in6(xprt);
330 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
336 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
339 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
341 struct sockaddr *sap = xs_addr(xprt);
344 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
345 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
347 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
348 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
351 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
352 const char *protocol,
355 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
356 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
357 xs_format_common_peer_addresses(xprt);
358 xs_format_common_peer_ports(xprt);
361 static void xs_update_peer_port(struct rpc_xprt *xprt)
363 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
364 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
366 xs_format_common_peer_ports(xprt);
369 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
373 for (i = 0; i < RPC_DISPLAY_MAX; i++)
375 case RPC_DISPLAY_PROTO:
376 case RPC_DISPLAY_NETID:
379 kfree(xprt->address_strings[i]);
383 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
385 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
387 struct msghdr msg = {
389 .msg_namelen = addrlen,
390 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
393 .iov_base = vec->iov_base + base,
394 .iov_len = vec->iov_len - base,
397 if (iov.iov_len != 0)
398 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
399 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
402 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy)
404 ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
405 int offset, size_t size, int flags);
407 unsigned int remainder;
410 remainder = xdr->page_len - base;
411 base += xdr->page_base;
412 ppage = xdr->pages + (base >> PAGE_SHIFT);
414 do_sendpage = sock->ops->sendpage;
416 do_sendpage = sock_no_sendpage;
418 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
419 int flags = XS_SENDMSG_FLAGS;
422 if (remainder != 0 || more)
424 err = do_sendpage(sock, *ppage, base, len, flags);
425 if (remainder == 0 || err != len)
439 * xs_sendpages - write pages directly to a socket
440 * @sock: socket to send on
441 * @addr: UDP only -- address of destination
442 * @addrlen: UDP only -- length of destination address
443 * @xdr: buffer containing this request
444 * @base: starting position in the buffer
445 * @zerocopy: true if it is safe to use sendpage()
448 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy)
450 unsigned int remainder = xdr->len - base;
456 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
462 if (base < xdr->head[0].iov_len || addr != NULL) {
463 unsigned int len = xdr->head[0].iov_len - base;
465 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
466 if (remainder == 0 || err != len)
471 base -= xdr->head[0].iov_len;
473 if (base < xdr->page_len) {
474 unsigned int len = xdr->page_len - base;
476 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy);
477 if (remainder == 0 || err != len)
482 base -= xdr->page_len;
484 if (base >= xdr->tail[0].iov_len)
486 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
495 static void xs_nospace_callback(struct rpc_task *task)
497 struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
499 transport->inet->sk_write_pending--;
500 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
504 * xs_nospace - place task on wait queue if transmit was incomplete
505 * @task: task to put to sleep
508 static int xs_nospace(struct rpc_task *task)
510 struct rpc_rqst *req = task->tk_rqstp;
511 struct rpc_xprt *xprt = req->rq_xprt;
512 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
515 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
516 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
519 /* Protect against races with write_space */
520 spin_lock_bh(&xprt->transport_lock);
522 /* Don't race with disconnect */
523 if (xprt_connected(xprt)) {
524 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
526 * Notify TCP that we're limited by the application
529 set_bit(SOCK_NOSPACE, &transport->sock->flags);
530 transport->inet->sk_write_pending++;
531 /* ...and wait for more buffer space */
532 xprt_wait_for_buffer_space(task, xs_nospace_callback);
535 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
539 spin_unlock_bh(&xprt->transport_lock);
544 * Construct a stream transport record marker in @buf.
546 static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
548 u32 reclen = buf->len - sizeof(rpc_fraghdr);
549 rpc_fraghdr *base = buf->head[0].iov_base;
550 *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
554 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
555 * @task: RPC task that manages the state of an RPC request
558 * 0: The request has been sent
559 * EAGAIN: The socket was blocked, please call again later to
560 * complete the request
561 * ENOTCONN: Caller needs to invoke connect logic then call again
562 * other: Some other error occured, the request was not sent
564 static int xs_local_send_request(struct rpc_task *task)
566 struct rpc_rqst *req = task->tk_rqstp;
567 struct rpc_xprt *xprt = req->rq_xprt;
568 struct sock_xprt *transport =
569 container_of(xprt, struct sock_xprt, xprt);
570 struct xdr_buf *xdr = &req->rq_snd_buf;
573 xs_encode_stream_record_marker(&req->rq_snd_buf);
575 xs_pktdump("packet data:",
576 req->rq_svec->iov_base, req->rq_svec->iov_len);
578 status = xs_sendpages(transport->sock, NULL, 0,
579 xdr, req->rq_bytes_sent, true);
580 dprintk("RPC: %s(%u) = %d\n",
581 __func__, xdr->len - req->rq_bytes_sent, status);
582 if (likely(status >= 0)) {
583 req->rq_bytes_sent += status;
584 req->rq_xmit_bytes_sent += status;
585 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
586 req->rq_bytes_sent = 0;
594 status = xs_nospace(task);
597 dprintk("RPC: sendmsg returned unrecognized error %d\n",
608 * xs_udp_send_request - write an RPC request to a UDP socket
609 * @task: address of RPC task that manages the state of an RPC request
612 * 0: The request has been sent
613 * EAGAIN: The socket was blocked, please call again later to
614 * complete the request
615 * ENOTCONN: Caller needs to invoke connect logic then call again
616 * other: Some other error occurred, the request was not sent
618 static int xs_udp_send_request(struct rpc_task *task)
620 struct rpc_rqst *req = task->tk_rqstp;
621 struct rpc_xprt *xprt = req->rq_xprt;
622 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
623 struct xdr_buf *xdr = &req->rq_snd_buf;
626 xs_pktdump("packet data:",
627 req->rq_svec->iov_base,
628 req->rq_svec->iov_len);
630 if (!xprt_bound(xprt))
632 status = xs_sendpages(transport->sock,
635 req->rq_bytes_sent, true);
637 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
638 xdr->len - req->rq_bytes_sent, status);
641 req->rq_xmit_bytes_sent += status;
642 if (status >= req->rq_slen)
644 /* Still some bytes left; set up for a retry later. */
651 /* Should we call xs_close() here? */
654 status = xs_nospace(task);
657 dprintk("RPC: sendmsg returned unrecognized error %d\n",
662 /* When the server has died, an ICMP port unreachable message
663 * prompts ECONNREFUSED. */
664 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
671 * xs_tcp_shutdown - gracefully shut down a TCP socket
674 * Initiates a graceful shutdown of the TCP socket by calling the
675 * equivalent of shutdown(SHUT_WR);
677 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
679 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
680 struct socket *sock = transport->sock;
683 kernel_sock_shutdown(sock, SHUT_WR);
684 trace_rpc_socket_shutdown(xprt, sock);
689 * xs_tcp_send_request - write an RPC request to a TCP socket
690 * @task: address of RPC task that manages the state of an RPC request
693 * 0: The request has been sent
694 * EAGAIN: The socket was blocked, please call again later to
695 * complete the request
696 * ENOTCONN: Caller needs to invoke connect logic then call again
697 * other: Some other error occurred, the request was not sent
699 * XXX: In the case of soft timeouts, should we eventually give up
700 * if sendmsg is not able to make progress?
702 static int xs_tcp_send_request(struct rpc_task *task)
704 struct rpc_rqst *req = task->tk_rqstp;
705 struct rpc_xprt *xprt = req->rq_xprt;
706 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
707 struct xdr_buf *xdr = &req->rq_snd_buf;
708 bool zerocopy = true;
711 xs_encode_stream_record_marker(&req->rq_snd_buf);
713 xs_pktdump("packet data:",
714 req->rq_svec->iov_base,
715 req->rq_svec->iov_len);
716 /* Don't use zero copy if this is a resend. If the RPC call
717 * completes while the socket holds a reference to the pages,
718 * then we may end up resending corrupted data.
720 if (task->tk_flags & RPC_TASK_SENT)
723 /* Continue transmitting the packet/record. We must be careful
724 * to cope with writespace callbacks arriving _after_ we have
725 * called sendmsg(). */
727 status = xs_sendpages(transport->sock,
728 NULL, 0, xdr, req->rq_bytes_sent,
731 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
732 xdr->len - req->rq_bytes_sent, status);
734 if (unlikely(status < 0))
737 /* If we've sent the entire packet, immediately
738 * reset the count of bytes sent. */
739 req->rq_bytes_sent += status;
740 req->rq_xmit_bytes_sent += status;
741 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
742 req->rq_bytes_sent = 0;
755 /* Should we call xs_close() here? */
758 status = xs_nospace(task);
761 dprintk("RPC: sendmsg returned unrecognized error %d\n",
764 xs_tcp_shutdown(xprt);
768 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
775 * xs_tcp_release_xprt - clean up after a tcp transmission
779 * This cleans up if an error causes us to abort the transmission of a request.
780 * In this case, the socket may need to be reset in order to avoid confusing
783 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
785 struct rpc_rqst *req;
787 if (task != xprt->snd_task)
791 req = task->tk_rqstp;
794 if (req->rq_bytes_sent == 0)
796 if (req->rq_bytes_sent == req->rq_snd_buf.len)
798 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
800 xprt_release_xprt(xprt, task);
803 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
805 transport->old_data_ready = sk->sk_data_ready;
806 transport->old_state_change = sk->sk_state_change;
807 transport->old_write_space = sk->sk_write_space;
808 transport->old_error_report = sk->sk_error_report;
811 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
813 sk->sk_data_ready = transport->old_data_ready;
814 sk->sk_state_change = transport->old_state_change;
815 sk->sk_write_space = transport->old_write_space;
816 sk->sk_error_report = transport->old_error_report;
820 * xs_error_report - callback to handle TCP socket state errors
823 * Note: we don't call sock_error() since there may be a rpc_task
824 * using the socket, and so we don't want to clear sk->sk_err.
826 static void xs_error_report(struct sock *sk)
828 struct rpc_xprt *xprt;
831 read_lock_bh(&sk->sk_callback_lock);
832 if (!(xprt = xprt_from_sock(sk)))
838 dprintk("RPC: xs_error_report client %p, error=%d...\n",
840 trace_rpc_socket_error(xprt, sk->sk_socket, err);
841 xprt_wake_pending_tasks(xprt, err);
843 read_unlock_bh(&sk->sk_callback_lock);
846 static void xs_reset_transport(struct sock_xprt *transport)
848 struct socket *sock = transport->sock;
849 struct sock *sk = transport->inet;
854 transport->srcport = 0;
856 write_lock_bh(&sk->sk_callback_lock);
857 transport->inet = NULL;
858 transport->sock = NULL;
860 sk->sk_user_data = NULL;
862 xs_restore_old_callbacks(transport, sk);
863 write_unlock_bh(&sk->sk_callback_lock);
867 trace_rpc_socket_close(&transport->xprt, sock);
872 * xs_close - close a socket
875 * This is used when all requests are complete; ie, no DRC state remains
876 * on the server we want to save.
878 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
879 * xs_reset_transport() zeroing the socket from underneath a writer.
881 static void xs_close(struct rpc_xprt *xprt)
883 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
885 dprintk("RPC: xs_close xprt %p\n", xprt);
887 cancel_delayed_work_sync(&transport->connect_worker);
889 xs_reset_transport(transport);
890 xprt->reestablish_timeout = 0;
892 smp_mb__before_clear_bit();
893 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
894 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
895 clear_bit(XPRT_CLOSING, &xprt->state);
896 smp_mb__after_clear_bit();
897 xprt_disconnect_done(xprt);
900 static void xs_tcp_close(struct rpc_xprt *xprt)
902 if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state))
905 xs_tcp_shutdown(xprt);
908 static void xs_xprt_free(struct rpc_xprt *xprt)
910 xs_free_peer_addresses(xprt);
915 * xs_destroy - prepare to shutdown a transport
916 * @xprt: doomed transport
919 static void xs_destroy(struct rpc_xprt *xprt)
921 dprintk("RPC: xs_destroy xprt %p\n", xprt);
925 module_put(THIS_MODULE);
928 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
930 struct xdr_skb_reader desc = {
932 .offset = sizeof(rpc_fraghdr),
933 .count = skb->len - sizeof(rpc_fraghdr),
936 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
944 * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets
945 * @sk: socket with data to read
946 * @len: how much data to read
948 * Currently this assumes we can read the whole reply in a single gulp.
950 static void xs_local_data_ready(struct sock *sk, int len)
952 struct rpc_task *task;
953 struct rpc_xprt *xprt;
954 struct rpc_rqst *rovr;
956 int err, repsize, copied;
960 read_lock_bh(&sk->sk_callback_lock);
961 dprintk("RPC: %s...\n", __func__);
962 xprt = xprt_from_sock(sk);
966 skb = skb_recv_datagram(sk, 0, 1, &err);
970 repsize = skb->len - sizeof(rpc_fraghdr);
972 dprintk("RPC: impossible RPC reply size %d\n", repsize);
976 /* Copy the XID from the skb... */
977 xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
981 /* Look up and lock the request corresponding to the given XID */
982 spin_lock(&xprt->transport_lock);
983 rovr = xprt_lookup_rqst(xprt, *xp);
986 task = rovr->rq_task;
988 copied = rovr->rq_private_buf.buflen;
989 if (copied > repsize)
992 if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
993 dprintk("RPC: sk_buff copy failed\n");
997 xprt_complete_rqst(task, copied);
1000 spin_unlock(&xprt->transport_lock);
1002 skb_free_datagram(sk, skb);
1004 read_unlock_bh(&sk->sk_callback_lock);
1008 * xs_udp_data_ready - "data ready" callback for UDP sockets
1009 * @sk: socket with data to read
1010 * @len: how much data to read
1013 static void xs_udp_data_ready(struct sock *sk, int len)
1015 struct rpc_task *task;
1016 struct rpc_xprt *xprt;
1017 struct rpc_rqst *rovr;
1018 struct sk_buff *skb;
1019 int err, repsize, copied;
1023 read_lock_bh(&sk->sk_callback_lock);
1024 dprintk("RPC: xs_udp_data_ready...\n");
1025 if (!(xprt = xprt_from_sock(sk)))
1028 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
1031 repsize = skb->len - sizeof(struct udphdr);
1033 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
1037 /* Copy the XID from the skb... */
1038 xp = skb_header_pointer(skb, sizeof(struct udphdr),
1039 sizeof(_xid), &_xid);
1043 /* Look up and lock the request corresponding to the given XID */
1044 spin_lock(&xprt->transport_lock);
1045 rovr = xprt_lookup_rqst(xprt, *xp);
1048 task = rovr->rq_task;
1050 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1053 /* Suck it into the iovec, verify checksum if not done by hw. */
1054 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1055 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
1059 UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
1061 xprt_adjust_cwnd(xprt, task, copied);
1062 xprt_complete_rqst(task, copied);
1065 spin_unlock(&xprt->transport_lock);
1067 skb_free_datagram(sk, skb);
1069 read_unlock_bh(&sk->sk_callback_lock);
1073 * Helper function to force a TCP close if the server is sending
1074 * junk and/or it has put us in CLOSE_WAIT
1076 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1078 set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1079 xprt_force_disconnect(xprt);
1082 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1084 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1088 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1089 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1090 used = xdr_skb_read_bits(desc, p, len);
1091 transport->tcp_offset += used;
1095 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1096 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1097 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1099 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1100 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1102 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1103 transport->tcp_offset = 0;
1105 /* Sanity check of the record length */
1106 if (unlikely(transport->tcp_reclen < 8)) {
1107 dprintk("RPC: invalid TCP record fragment length\n");
1108 xs_tcp_force_close(xprt);
1111 dprintk("RPC: reading TCP record fragment of length %d\n",
1112 transport->tcp_reclen);
1115 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1117 if (transport->tcp_offset == transport->tcp_reclen) {
1118 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1119 transport->tcp_offset = 0;
1120 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1121 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1122 transport->tcp_flags |= TCP_RCV_COPY_XID;
1123 transport->tcp_copied = 0;
1128 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1133 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1134 dprintk("RPC: reading XID (%Zu bytes)\n", len);
1135 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1136 used = xdr_skb_read_bits(desc, p, len);
1137 transport->tcp_offset += used;
1140 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1141 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1142 transport->tcp_copied = 4;
1143 dprintk("RPC: reading %s XID %08x\n",
1144 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1146 ntohl(transport->tcp_xid));
1147 xs_tcp_check_fraghdr(transport);
1150 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1151 struct xdr_skb_reader *desc)
1158 * We want transport->tcp_offset to be 8 at the end of this routine
1159 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1160 * When this function is called for the first time,
1161 * transport->tcp_offset is 4 (after having already read the xid).
1163 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1164 len = sizeof(transport->tcp_calldir) - offset;
1165 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len);
1166 p = ((char *) &transport->tcp_calldir) + offset;
1167 used = xdr_skb_read_bits(desc, p, len);
1168 transport->tcp_offset += used;
1171 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1173 * We don't yet have the XDR buffer, so we will write the calldir
1174 * out after we get the buffer from the 'struct rpc_rqst'
1176 switch (ntohl(transport->tcp_calldir)) {
1178 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1179 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1180 transport->tcp_flags |= TCP_RPC_REPLY;
1183 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1184 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1185 transport->tcp_flags &= ~TCP_RPC_REPLY;
1188 dprintk("RPC: invalid request message type\n");
1189 xs_tcp_force_close(&transport->xprt);
1191 xs_tcp_check_fraghdr(transport);
1194 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1195 struct xdr_skb_reader *desc,
1196 struct rpc_rqst *req)
1198 struct sock_xprt *transport =
1199 container_of(xprt, struct sock_xprt, xprt);
1200 struct xdr_buf *rcvbuf;
1204 rcvbuf = &req->rq_private_buf;
1206 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1208 * Save the RPC direction in the XDR buffer
1210 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1211 &transport->tcp_calldir,
1212 sizeof(transport->tcp_calldir));
1213 transport->tcp_copied += sizeof(transport->tcp_calldir);
1214 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1218 if (len > transport->tcp_reclen - transport->tcp_offset) {
1219 struct xdr_skb_reader my_desc;
1221 len = transport->tcp_reclen - transport->tcp_offset;
1222 memcpy(&my_desc, desc, sizeof(my_desc));
1223 my_desc.count = len;
1224 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1225 &my_desc, xdr_skb_read_bits);
1229 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1230 desc, xdr_skb_read_bits);
1233 transport->tcp_copied += r;
1234 transport->tcp_offset += r;
1237 /* Error when copying to the receive buffer,
1238 * usually because we weren't able to allocate
1239 * additional buffer pages. All we can do now
1240 * is turn off TCP_RCV_COPY_DATA, so the request
1241 * will not receive any additional updates,
1243 * Any remaining data from this record will
1246 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1247 dprintk("RPC: XID %08x truncated request\n",
1248 ntohl(transport->tcp_xid));
1249 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1250 "tcp_offset = %u, tcp_reclen = %u\n",
1251 xprt, transport->tcp_copied,
1252 transport->tcp_offset, transport->tcp_reclen);
1256 dprintk("RPC: XID %08x read %Zd bytes\n",
1257 ntohl(transport->tcp_xid), r);
1258 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1259 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1260 transport->tcp_offset, transport->tcp_reclen);
1262 if (transport->tcp_copied == req->rq_private_buf.buflen)
1263 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1264 else if (transport->tcp_offset == transport->tcp_reclen) {
1265 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1266 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1271 * Finds the request corresponding to the RPC xid and invokes the common
1272 * tcp read code to read the data.
1274 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1275 struct xdr_skb_reader *desc)
1277 struct sock_xprt *transport =
1278 container_of(xprt, struct sock_xprt, xprt);
1279 struct rpc_rqst *req;
1281 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1283 /* Find and lock the request corresponding to this xid */
1284 spin_lock(&xprt->transport_lock);
1285 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1287 dprintk("RPC: XID %08x request not found!\n",
1288 ntohl(transport->tcp_xid));
1289 spin_unlock(&xprt->transport_lock);
1293 xs_tcp_read_common(xprt, desc, req);
1295 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1296 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1298 spin_unlock(&xprt->transport_lock);
1302 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1304 * Obtains an rpc_rqst previously allocated and invokes the common
1305 * tcp read code to read the data. The result is placed in the callback
1307 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1308 * connection and return -1.
1310 static inline int xs_tcp_read_callback(struct rpc_xprt *xprt,
1311 struct xdr_skb_reader *desc)
1313 struct sock_xprt *transport =
1314 container_of(xprt, struct sock_xprt, xprt);
1315 struct rpc_rqst *req;
1317 req = xprt_alloc_bc_request(xprt);
1319 printk(KERN_WARNING "Callback slot table overflowed\n");
1320 xprt_force_disconnect(xprt);
1324 req->rq_xid = transport->tcp_xid;
1325 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1326 xs_tcp_read_common(xprt, desc, req);
1328 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) {
1329 struct svc_serv *bc_serv = xprt->bc_serv;
1332 * Add callback request to callback list. The callback
1333 * service sleeps on the sv_cb_waitq waiting for new
1334 * requests. Wake it up after adding enqueing the
1337 dprintk("RPC: add callback request to list\n");
1338 spin_lock(&bc_serv->sv_cb_lock);
1339 list_add(&req->rq_bc_list, &bc_serv->sv_cb_list);
1340 spin_unlock(&bc_serv->sv_cb_lock);
1341 wake_up(&bc_serv->sv_cb_waitq);
1344 req->rq_private_buf.len = transport->tcp_copied;
1349 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1350 struct xdr_skb_reader *desc)
1352 struct sock_xprt *transport =
1353 container_of(xprt, struct sock_xprt, xprt);
1355 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1356 xs_tcp_read_reply(xprt, desc) :
1357 xs_tcp_read_callback(xprt, desc);
1360 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1361 struct xdr_skb_reader *desc)
1363 return xs_tcp_read_reply(xprt, desc);
1365 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1368 * Read data off the transport. This can be either an RPC_CALL or an
1369 * RPC_REPLY. Relay the processing to helper functions.
1371 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1372 struct xdr_skb_reader *desc)
1374 struct sock_xprt *transport =
1375 container_of(xprt, struct sock_xprt, xprt);
1377 if (_xs_tcp_read_data(xprt, desc) == 0)
1378 xs_tcp_check_fraghdr(transport);
1381 * The transport_lock protects the request handling.
1382 * There's no need to hold it to update the tcp_flags.
1384 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1388 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1392 len = transport->tcp_reclen - transport->tcp_offset;
1393 if (len > desc->count)
1396 desc->offset += len;
1397 transport->tcp_offset += len;
1398 dprintk("RPC: discarded %Zu bytes\n", len);
1399 xs_tcp_check_fraghdr(transport);
1402 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1404 struct rpc_xprt *xprt = rd_desc->arg.data;
1405 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1406 struct xdr_skb_reader desc = {
1412 dprintk("RPC: xs_tcp_data_recv started\n");
1414 /* Read in a new fragment marker if necessary */
1415 /* Can we ever really expect to get completely empty fragments? */
1416 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1417 xs_tcp_read_fraghdr(xprt, &desc);
1420 /* Read in the xid if necessary */
1421 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1422 xs_tcp_read_xid(transport, &desc);
1425 /* Read in the call/reply flag */
1426 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1427 xs_tcp_read_calldir(transport, &desc);
1430 /* Read in the request data */
1431 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1432 xs_tcp_read_data(xprt, &desc);
1435 /* Skip over any trailing bytes on short reads */
1436 xs_tcp_read_discard(transport, &desc);
1437 } while (desc.count);
1438 dprintk("RPC: xs_tcp_data_recv done\n");
1439 return len - desc.count;
1443 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1444 * @sk: socket with data to read
1445 * @bytes: how much data to read
1448 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1450 struct rpc_xprt *xprt;
1451 read_descriptor_t rd_desc;
1454 dprintk("RPC: xs_tcp_data_ready...\n");
1456 read_lock_bh(&sk->sk_callback_lock);
1457 if (!(xprt = xprt_from_sock(sk)))
1459 /* Any data means we had a useful conversation, so
1460 * the we don't need to delay the next reconnect
1462 if (xprt->reestablish_timeout)
1463 xprt->reestablish_timeout = 0;
1465 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1466 rd_desc.arg.data = xprt;
1468 rd_desc.count = 65536;
1469 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1472 read_unlock_bh(&sk->sk_callback_lock);
1476 * Do the equivalent of linger/linger2 handling for dealing with
1477 * broken servers that don't close the socket in a timely
1480 static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1481 unsigned long timeout)
1483 struct sock_xprt *transport;
1485 if (xprt_test_and_set_connecting(xprt))
1487 set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1488 transport = container_of(xprt, struct sock_xprt, xprt);
1489 queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1493 static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1495 struct sock_xprt *transport;
1497 transport = container_of(xprt, struct sock_xprt, xprt);
1499 if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1500 !cancel_delayed_work(&transport->connect_worker))
1502 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1503 xprt_clear_connecting(xprt);
1506 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1508 smp_mb__before_clear_bit();
1509 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1510 clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1511 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1512 clear_bit(XPRT_CLOSING, &xprt->state);
1513 smp_mb__after_clear_bit();
1516 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1518 xs_sock_reset_connection_flags(xprt);
1519 /* Mark transport as closed and wake up all pending tasks */
1520 xprt_disconnect_done(xprt);
1524 * xs_tcp_state_change - callback to handle TCP socket state changes
1525 * @sk: socket whose state has changed
1528 static void xs_tcp_state_change(struct sock *sk)
1530 struct rpc_xprt *xprt;
1532 read_lock_bh(&sk->sk_callback_lock);
1533 if (!(xprt = xprt_from_sock(sk)))
1535 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1536 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1537 sk->sk_state, xprt_connected(xprt),
1538 sock_flag(sk, SOCK_DEAD),
1539 sock_flag(sk, SOCK_ZAPPED),
1542 trace_rpc_socket_state_change(xprt, sk->sk_socket);
1543 switch (sk->sk_state) {
1544 case TCP_ESTABLISHED:
1545 spin_lock(&xprt->transport_lock);
1546 if (!xprt_test_and_set_connected(xprt)) {
1547 struct sock_xprt *transport = container_of(xprt,
1548 struct sock_xprt, xprt);
1550 /* Reset TCP record info */
1551 transport->tcp_offset = 0;
1552 transport->tcp_reclen = 0;
1553 transport->tcp_copied = 0;
1554 transport->tcp_flags =
1555 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1556 xprt->connect_cookie++;
1558 xprt_wake_pending_tasks(xprt, -EAGAIN);
1560 spin_unlock(&xprt->transport_lock);
1563 /* The client initiated a shutdown of the socket */
1564 xprt->connect_cookie++;
1565 xprt->reestablish_timeout = 0;
1566 set_bit(XPRT_CLOSING, &xprt->state);
1567 smp_mb__before_clear_bit();
1568 clear_bit(XPRT_CONNECTED, &xprt->state);
1569 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1570 smp_mb__after_clear_bit();
1571 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1573 case TCP_CLOSE_WAIT:
1574 /* The server initiated a shutdown of the socket */
1575 xprt->connect_cookie++;
1576 clear_bit(XPRT_CONNECTED, &xprt->state);
1577 xs_tcp_force_close(xprt);
1580 * If the server closed down the connection, make sure that
1581 * we back off before reconnecting
1583 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1584 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1587 set_bit(XPRT_CLOSING, &xprt->state);
1588 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1589 smp_mb__before_clear_bit();
1590 clear_bit(XPRT_CONNECTED, &xprt->state);
1591 smp_mb__after_clear_bit();
1594 xs_tcp_cancel_linger_timeout(xprt);
1595 xs_sock_mark_closed(xprt);
1598 read_unlock_bh(&sk->sk_callback_lock);
1601 static void xs_write_space(struct sock *sk)
1603 struct socket *sock;
1604 struct rpc_xprt *xprt;
1606 if (unlikely(!(sock = sk->sk_socket)))
1608 clear_bit(SOCK_NOSPACE, &sock->flags);
1610 if (unlikely(!(xprt = xprt_from_sock(sk))))
1612 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1615 xprt_write_space(xprt);
1619 * xs_udp_write_space - callback invoked when socket buffer space
1621 * @sk: socket whose state has changed
1623 * Called when more output buffer space is available for this socket.
1624 * We try not to wake our writers until they can make "significant"
1625 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1626 * with a bunch of small requests.
1628 static void xs_udp_write_space(struct sock *sk)
1630 read_lock_bh(&sk->sk_callback_lock);
1632 /* from net/core/sock.c:sock_def_write_space */
1633 if (sock_writeable(sk))
1636 read_unlock_bh(&sk->sk_callback_lock);
1640 * xs_tcp_write_space - callback invoked when socket buffer space
1642 * @sk: socket whose state has changed
1644 * Called when more output buffer space is available for this socket.
1645 * We try not to wake our writers until they can make "significant"
1646 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1647 * with a bunch of small requests.
1649 static void xs_tcp_write_space(struct sock *sk)
1651 read_lock_bh(&sk->sk_callback_lock);
1653 /* from net/core/stream.c:sk_stream_write_space */
1654 if (sk_stream_is_writeable(sk))
1657 read_unlock_bh(&sk->sk_callback_lock);
1660 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1662 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1663 struct sock *sk = transport->inet;
1665 if (transport->rcvsize) {
1666 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1667 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1669 if (transport->sndsize) {
1670 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1671 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1672 sk->sk_write_space(sk);
1677 * xs_udp_set_buffer_size - set send and receive limits
1678 * @xprt: generic transport
1679 * @sndsize: requested size of send buffer, in bytes
1680 * @rcvsize: requested size of receive buffer, in bytes
1682 * Set socket send and receive buffer size limits.
1684 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1686 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1688 transport->sndsize = 0;
1690 transport->sndsize = sndsize + 1024;
1691 transport->rcvsize = 0;
1693 transport->rcvsize = rcvsize + 1024;
1695 xs_udp_do_set_buffer_size(xprt);
1699 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1700 * @task: task that timed out
1702 * Adjust the congestion window after a retransmit timeout has occurred.
1704 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1706 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1709 static unsigned short xs_get_random_port(void)
1711 unsigned short range = xprt_max_resvport - xprt_min_resvport;
1712 unsigned short rand = (unsigned short) prandom_u32() % range;
1713 return rand + xprt_min_resvport;
1717 * xs_set_port - reset the port number in the remote endpoint address
1718 * @xprt: generic transport
1719 * @port: new port number
1722 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1724 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1726 rpc_set_port(xs_addr(xprt), port);
1727 xs_update_peer_port(xprt);
1730 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1732 unsigned short port = transport->srcport;
1734 if (port == 0 && transport->xprt.resvport)
1735 port = xs_get_random_port();
1739 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1741 if (transport->srcport != 0)
1742 transport->srcport = 0;
1743 if (!transport->xprt.resvport)
1745 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1746 return xprt_max_resvport;
1749 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1751 struct sockaddr_storage myaddr;
1753 unsigned short port = xs_get_srcport(transport);
1754 unsigned short last;
1756 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1758 rpc_set_port((struct sockaddr *)&myaddr, port);
1759 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1760 transport->xprt.addrlen);
1764 transport->srcport = port;
1768 port = xs_next_srcport(transport, port);
1771 } while (err == -EADDRINUSE && nloop != 2);
1773 if (myaddr.ss_family == AF_INET)
1774 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1775 &((struct sockaddr_in *)&myaddr)->sin_addr,
1776 port, err ? "failed" : "ok", err);
1778 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1779 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1780 port, err ? "failed" : "ok", err);
1785 * We don't support autobind on AF_LOCAL sockets
1787 static void xs_local_rpcbind(struct rpc_task *task)
1790 xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt));
1794 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1798 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1799 static struct lock_class_key xs_key[2];
1800 static struct lock_class_key xs_slock_key[2];
1802 static inline void xs_reclassify_socketu(struct socket *sock)
1804 struct sock *sk = sock->sk;
1806 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1807 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1810 static inline void xs_reclassify_socket4(struct socket *sock)
1812 struct sock *sk = sock->sk;
1814 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1815 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1818 static inline void xs_reclassify_socket6(struct socket *sock)
1820 struct sock *sk = sock->sk;
1822 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1823 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1826 static inline void xs_reclassify_socket(int family, struct socket *sock)
1828 WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1829 if (sock_owned_by_user(sock->sk))
1834 xs_reclassify_socketu(sock);
1837 xs_reclassify_socket4(sock);
1840 xs_reclassify_socket6(sock);
1845 static inline void xs_reclassify_socketu(struct socket *sock)
1849 static inline void xs_reclassify_socket4(struct socket *sock)
1853 static inline void xs_reclassify_socket6(struct socket *sock)
1857 static inline void xs_reclassify_socket(int family, struct socket *sock)
1862 static void xs_dummy_setup_socket(struct work_struct *work)
1866 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1867 struct sock_xprt *transport, int family, int type, int protocol)
1869 struct socket *sock;
1872 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1874 dprintk("RPC: can't create %d transport socket (%d).\n",
1878 xs_reclassify_socket(family, sock);
1880 err = xs_bind(transport, sock);
1888 return ERR_PTR(err);
1891 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1892 struct socket *sock)
1894 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1897 if (!transport->inet) {
1898 struct sock *sk = sock->sk;
1900 write_lock_bh(&sk->sk_callback_lock);
1902 xs_save_old_callbacks(transport, sk);
1904 sk->sk_user_data = xprt;
1905 sk->sk_data_ready = xs_local_data_ready;
1906 sk->sk_write_space = xs_udp_write_space;
1907 sk->sk_error_report = xs_error_report;
1908 sk->sk_allocation = GFP_ATOMIC;
1910 xprt_clear_connected(xprt);
1912 /* Reset to new socket */
1913 transport->sock = sock;
1914 transport->inet = sk;
1916 write_unlock_bh(&sk->sk_callback_lock);
1919 /* Tell the socket layer to start connecting... */
1920 xprt->stat.connect_count++;
1921 xprt->stat.connect_start = jiffies;
1922 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1926 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1927 * @xprt: RPC transport to connect
1928 * @transport: socket transport to connect
1929 * @create_sock: function to create a socket of the correct type
1931 static int xs_local_setup_socket(struct sock_xprt *transport)
1933 struct rpc_xprt *xprt = &transport->xprt;
1934 struct socket *sock;
1937 current->flags |= PF_FSTRANS;
1939 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1940 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1941 SOCK_STREAM, 0, &sock, 1);
1943 dprintk("RPC: can't create AF_LOCAL "
1944 "transport socket (%d).\n", -status);
1947 xs_reclassify_socketu(sock);
1949 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1950 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1952 status = xs_local_finish_connecting(xprt, sock);
1953 trace_rpc_socket_connect(xprt, sock, status);
1956 dprintk("RPC: xprt %p connected to %s\n",
1957 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1958 xprt_set_connected(xprt);
1961 dprintk("RPC: xprt %p: socket %s does not exist\n",
1962 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1965 dprintk("RPC: xprt %p: connection refused for %s\n",
1966 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1969 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1971 xprt->address_strings[RPC_DISPLAY_ADDR]);
1975 xprt_clear_connecting(xprt);
1976 xprt_wake_pending_tasks(xprt, status);
1977 current->flags &= ~PF_FSTRANS;
1981 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1983 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1986 if (RPC_IS_ASYNC(task)) {
1988 * We want the AF_LOCAL connect to be resolved in the
1989 * filesystem namespace of the process making the rpc
1990 * call. Thus we connect synchronously.
1992 * If we want to support asynchronous AF_LOCAL calls,
1993 * we'll need to figure out how to pass a namespace to
1996 rpc_exit(task, -ENOTCONN);
1999 ret = xs_local_setup_socket(transport);
2000 if (ret && !RPC_IS_SOFTCONN(task))
2001 msleep_interruptible(15000);
2004 #ifdef CONFIG_SUNRPC_SWAP
2005 static void xs_set_memalloc(struct rpc_xprt *xprt)
2007 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2011 sk_set_memalloc(transport->inet);
2015 * xs_swapper - Tag this transport as being used for swap.
2016 * @xprt: transport to tag
2017 * @enable: enable/disable
2020 int xs_swapper(struct rpc_xprt *xprt, int enable)
2022 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2028 xs_set_memalloc(xprt);
2029 } else if (xprt->swapper) {
2031 sk_clear_memalloc(transport->inet);
2036 EXPORT_SYMBOL_GPL(xs_swapper);
2038 static void xs_set_memalloc(struct rpc_xprt *xprt)
2043 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2045 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2047 if (!transport->inet) {
2048 struct sock *sk = sock->sk;
2050 write_lock_bh(&sk->sk_callback_lock);
2052 xs_save_old_callbacks(transport, sk);
2054 sk->sk_user_data = xprt;
2055 sk->sk_data_ready = xs_udp_data_ready;
2056 sk->sk_write_space = xs_udp_write_space;
2057 sk->sk_no_check = UDP_CSUM_NORCV;
2058 sk->sk_allocation = GFP_ATOMIC;
2060 xprt_set_connected(xprt);
2062 /* Reset to new socket */
2063 transport->sock = sock;
2064 transport->inet = sk;
2066 xs_set_memalloc(xprt);
2068 write_unlock_bh(&sk->sk_callback_lock);
2070 xs_udp_do_set_buffer_size(xprt);
2073 static void xs_udp_setup_socket(struct work_struct *work)
2075 struct sock_xprt *transport =
2076 container_of(work, struct sock_xprt, connect_worker.work);
2077 struct rpc_xprt *xprt = &transport->xprt;
2078 struct socket *sock = transport->sock;
2081 current->flags |= PF_FSTRANS;
2083 /* Start by resetting any existing state */
2084 xs_reset_transport(transport);
2085 sock = xs_create_sock(xprt, transport,
2086 xs_addr(xprt)->sa_family, SOCK_DGRAM, IPPROTO_UDP);
2090 dprintk("RPC: worker connecting xprt %p via %s to "
2091 "%s (port %s)\n", xprt,
2092 xprt->address_strings[RPC_DISPLAY_PROTO],
2093 xprt->address_strings[RPC_DISPLAY_ADDR],
2094 xprt->address_strings[RPC_DISPLAY_PORT]);
2096 xs_udp_finish_connecting(xprt, sock);
2097 trace_rpc_socket_connect(xprt, sock, 0);
2100 xprt_clear_connecting(xprt);
2101 xprt_wake_pending_tasks(xprt, status);
2102 current->flags &= ~PF_FSTRANS;
2106 * We need to preserve the port number so the reply cache on the server can
2107 * find our cached RPC replies when we get around to reconnecting.
2109 static void xs_abort_connection(struct sock_xprt *transport)
2112 struct sockaddr any;
2114 dprintk("RPC: disconnecting xprt %p to reuse port\n", transport);
2117 * Disconnect the transport socket by doing a connect operation
2118 * with AF_UNSPEC. This should return immediately...
2120 memset(&any, 0, sizeof(any));
2121 any.sa_family = AF_UNSPEC;
2122 result = kernel_connect(transport->sock, &any, sizeof(any), 0);
2123 trace_rpc_socket_reset_connection(&transport->xprt,
2124 transport->sock, result);
2126 xs_sock_reset_connection_flags(&transport->xprt);
2127 dprintk("RPC: AF_UNSPEC connect return code %d\n", result);
2130 static void xs_tcp_reuse_connection(struct sock_xprt *transport)
2132 unsigned int state = transport->inet->sk_state;
2134 if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED) {
2135 /* we don't need to abort the connection if the socket
2136 * hasn't undergone a shutdown
2138 if (transport->inet->sk_shutdown == 0)
2140 dprintk("RPC: %s: TCP_CLOSEd and sk_shutdown set to %d\n",
2141 __func__, transport->inet->sk_shutdown);
2143 if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT)) {
2144 /* we don't need to abort the connection if the socket
2145 * hasn't undergone a shutdown
2147 if (transport->inet->sk_shutdown == 0)
2149 dprintk("RPC: %s: ESTABLISHED/SYN_SENT "
2150 "sk_shutdown set to %d\n",
2151 __func__, transport->inet->sk_shutdown);
2153 xs_abort_connection(transport);
2156 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2158 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2159 int ret = -ENOTCONN;
2161 if (!transport->inet) {
2162 struct sock *sk = sock->sk;
2163 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2164 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2165 unsigned int opt_on = 1;
2167 /* TCP Keepalive options */
2168 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2169 (char *)&opt_on, sizeof(opt_on));
2170 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2171 (char *)&keepidle, sizeof(keepidle));
2172 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2173 (char *)&keepidle, sizeof(keepidle));
2174 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2175 (char *)&keepcnt, sizeof(keepcnt));
2177 write_lock_bh(&sk->sk_callback_lock);
2179 xs_save_old_callbacks(transport, sk);
2181 sk->sk_user_data = xprt;
2182 sk->sk_data_ready = xs_tcp_data_ready;
2183 sk->sk_state_change = xs_tcp_state_change;
2184 sk->sk_write_space = xs_tcp_write_space;
2185 sk->sk_error_report = xs_error_report;
2186 sk->sk_allocation = GFP_ATOMIC;
2188 /* socket options */
2189 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
2190 sock_reset_flag(sk, SOCK_LINGER);
2191 tcp_sk(sk)->linger2 = 0;
2192 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2194 xprt_clear_connected(xprt);
2196 /* Reset to new socket */
2197 transport->sock = sock;
2198 transport->inet = sk;
2200 write_unlock_bh(&sk->sk_callback_lock);
2203 if (!xprt_bound(xprt))
2206 xs_set_memalloc(xprt);
2208 /* Tell the socket layer to start connecting... */
2209 xprt->stat.connect_count++;
2210 xprt->stat.connect_start = jiffies;
2211 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2216 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2217 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2224 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2225 * @xprt: RPC transport to connect
2226 * @transport: socket transport to connect
2227 * @create_sock: function to create a socket of the correct type
2229 * Invoked by a work queue tasklet.
2231 static void xs_tcp_setup_socket(struct work_struct *work)
2233 struct sock_xprt *transport =
2234 container_of(work, struct sock_xprt, connect_worker.work);
2235 struct socket *sock = transport->sock;
2236 struct rpc_xprt *xprt = &transport->xprt;
2239 current->flags |= PF_FSTRANS;
2242 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
2243 sock = xs_create_sock(xprt, transport,
2244 xs_addr(xprt)->sa_family, SOCK_STREAM, IPPROTO_TCP);
2246 status = PTR_ERR(sock);
2252 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
2254 /* "close" the socket, preserving the local port */
2255 xs_tcp_reuse_connection(transport);
2261 dprintk("RPC: worker connecting xprt %p via %s to "
2262 "%s (port %s)\n", xprt,
2263 xprt->address_strings[RPC_DISPLAY_PROTO],
2264 xprt->address_strings[RPC_DISPLAY_ADDR],
2265 xprt->address_strings[RPC_DISPLAY_PORT]);
2267 status = xs_tcp_finish_connecting(xprt, sock);
2268 trace_rpc_socket_connect(xprt, sock, status);
2269 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2270 xprt, -status, xprt_connected(xprt),
2271 sock->sk->sk_state);
2274 printk("%s: connect returned unhandled error %d\n",
2276 case -EADDRNOTAVAIL:
2277 /* We're probably in TIME_WAIT. Get rid of existing socket,
2280 xs_tcp_force_close(xprt);
2285 xprt_clear_connecting(xprt);
2286 current->flags &= ~PF_FSTRANS;
2289 /* Happens, for instance, if the user specified a link
2290 * local IPv6 address without a scope-id.
2295 /* retry with existing socket, after a delay */
2301 xprt_clear_connecting(xprt);
2302 xprt_wake_pending_tasks(xprt, status);
2303 current->flags &= ~PF_FSTRANS;
2307 * xs_connect - connect a socket to a remote endpoint
2308 * @xprt: pointer to transport structure
2309 * @task: address of RPC task that manages state of connect request
2311 * TCP: If the remote end dropped the connection, delay reconnecting.
2313 * UDP socket connects are synchronous, but we use a work queue anyway
2314 * to guarantee that even unprivileged user processes can set up a
2315 * socket on a privileged port.
2317 * If a UDP socket connect fails, the delay behavior here prevents
2318 * retry floods (hard mounts).
2320 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2322 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2324 if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
2325 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2327 xprt, xprt->reestablish_timeout / HZ);
2328 queue_delayed_work(rpciod_workqueue,
2329 &transport->connect_worker,
2330 xprt->reestablish_timeout);
2331 xprt->reestablish_timeout <<= 1;
2332 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2333 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2334 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2335 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2337 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2338 queue_delayed_work(rpciod_workqueue,
2339 &transport->connect_worker, 0);
2344 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2345 * @xprt: rpc_xprt struct containing statistics
2349 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2353 if (xprt_connected(xprt))
2354 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2356 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2357 "%llu %llu %lu %llu %llu\n",
2358 xprt->stat.bind_count,
2359 xprt->stat.connect_count,
2360 xprt->stat.connect_time,
2364 xprt->stat.bad_xids,
2367 xprt->stat.max_slots,
2368 xprt->stat.sending_u,
2369 xprt->stat.pending_u);
2373 * xs_udp_print_stats - display UDP socket-specifc stats
2374 * @xprt: rpc_xprt struct containing statistics
2378 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2380 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2382 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2385 xprt->stat.bind_count,
2388 xprt->stat.bad_xids,
2391 xprt->stat.max_slots,
2392 xprt->stat.sending_u,
2393 xprt->stat.pending_u);
2397 * xs_tcp_print_stats - display TCP socket-specifc stats
2398 * @xprt: rpc_xprt struct containing statistics
2402 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2404 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2407 if (xprt_connected(xprt))
2408 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2410 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2411 "%llu %llu %lu %llu %llu\n",
2413 xprt->stat.bind_count,
2414 xprt->stat.connect_count,
2415 xprt->stat.connect_time,
2419 xprt->stat.bad_xids,
2422 xprt->stat.max_slots,
2423 xprt->stat.sending_u,
2424 xprt->stat.pending_u);
2428 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2429 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2430 * to use the server side send routines.
2432 static void *bc_malloc(struct rpc_task *task, size_t size)
2435 struct rpc_buffer *buf;
2437 WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2438 if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2441 page = alloc_page(GFP_KERNEL);
2445 buf = page_address(page);
2446 buf->len = PAGE_SIZE;
2452 * Free the space allocated in the bc_alloc routine
2454 static void bc_free(void *buffer)
2456 struct rpc_buffer *buf;
2461 buf = container_of(buffer, struct rpc_buffer, data);
2462 free_page((unsigned long)buf);
2466 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2467 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2469 static int bc_sendto(struct rpc_rqst *req)
2472 struct xdr_buf *xbufp = &req->rq_snd_buf;
2473 struct rpc_xprt *xprt = req->rq_xprt;
2474 struct sock_xprt *transport =
2475 container_of(xprt, struct sock_xprt, xprt);
2476 struct socket *sock = transport->sock;
2477 unsigned long headoff;
2478 unsigned long tailoff;
2480 xs_encode_stream_record_marker(xbufp);
2482 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2483 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2484 len = svc_send_common(sock, xbufp,
2485 virt_to_page(xbufp->head[0].iov_base), headoff,
2486 xbufp->tail[0].iov_base, tailoff);
2488 if (len != xbufp->len) {
2489 printk(KERN_NOTICE "Error sending entire callback!\n");
2497 * The send routine. Borrows from svc_send
2499 static int bc_send_request(struct rpc_task *task)
2501 struct rpc_rqst *req = task->tk_rqstp;
2502 struct svc_xprt *xprt;
2505 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2507 * Get the server socket associated with this callback xprt
2509 xprt = req->rq_xprt->bc_xprt;
2512 * Grab the mutex to serialize data as the connection is shared
2513 * with the fore channel
2515 if (!mutex_trylock(&xprt->xpt_mutex)) {
2516 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2517 if (!mutex_trylock(&xprt->xpt_mutex))
2519 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2521 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2524 len = bc_sendto(req);
2525 mutex_unlock(&xprt->xpt_mutex);
2534 * The close routine. Since this is client initiated, we do nothing
2537 static void bc_close(struct rpc_xprt *xprt)
2542 * The xprt destroy routine. Again, because this connection is client
2543 * initiated, we do nothing
2546 static void bc_destroy(struct rpc_xprt *xprt)
2548 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2551 module_put(THIS_MODULE);
2554 static struct rpc_xprt_ops xs_local_ops = {
2555 .reserve_xprt = xprt_reserve_xprt,
2556 .release_xprt = xs_tcp_release_xprt,
2557 .alloc_slot = xprt_alloc_slot,
2558 .rpcbind = xs_local_rpcbind,
2559 .set_port = xs_local_set_port,
2560 .connect = xs_local_connect,
2561 .buf_alloc = rpc_malloc,
2562 .buf_free = rpc_free,
2563 .send_request = xs_local_send_request,
2564 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2566 .destroy = xs_destroy,
2567 .print_stats = xs_local_print_stats,
2570 static struct rpc_xprt_ops xs_udp_ops = {
2571 .set_buffer_size = xs_udp_set_buffer_size,
2572 .reserve_xprt = xprt_reserve_xprt_cong,
2573 .release_xprt = xprt_release_xprt_cong,
2574 .alloc_slot = xprt_alloc_slot,
2575 .rpcbind = rpcb_getport_async,
2576 .set_port = xs_set_port,
2577 .connect = xs_connect,
2578 .buf_alloc = rpc_malloc,
2579 .buf_free = rpc_free,
2580 .send_request = xs_udp_send_request,
2581 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
2582 .timer = xs_udp_timer,
2583 .release_request = xprt_release_rqst_cong,
2585 .destroy = xs_destroy,
2586 .print_stats = xs_udp_print_stats,
2589 static struct rpc_xprt_ops xs_tcp_ops = {
2590 .reserve_xprt = xprt_reserve_xprt,
2591 .release_xprt = xs_tcp_release_xprt,
2592 .alloc_slot = xprt_lock_and_alloc_slot,
2593 .rpcbind = rpcb_getport_async,
2594 .set_port = xs_set_port,
2595 .connect = xs_connect,
2596 .buf_alloc = rpc_malloc,
2597 .buf_free = rpc_free,
2598 .send_request = xs_tcp_send_request,
2599 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2600 .close = xs_tcp_close,
2601 .destroy = xs_destroy,
2602 .print_stats = xs_tcp_print_stats,
2606 * The rpc_xprt_ops for the server backchannel
2609 static struct rpc_xprt_ops bc_tcp_ops = {
2610 .reserve_xprt = xprt_reserve_xprt,
2611 .release_xprt = xprt_release_xprt,
2612 .alloc_slot = xprt_alloc_slot,
2613 .buf_alloc = bc_malloc,
2614 .buf_free = bc_free,
2615 .send_request = bc_send_request,
2616 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2618 .destroy = bc_destroy,
2619 .print_stats = xs_tcp_print_stats,
2622 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2624 static const struct sockaddr_in sin = {
2625 .sin_family = AF_INET,
2626 .sin_addr.s_addr = htonl(INADDR_ANY),
2628 static const struct sockaddr_in6 sin6 = {
2629 .sin6_family = AF_INET6,
2630 .sin6_addr = IN6ADDR_ANY_INIT,
2637 memcpy(sap, &sin, sizeof(sin));
2640 memcpy(sap, &sin6, sizeof(sin6));
2643 dprintk("RPC: %s: Bad address family\n", __func__);
2644 return -EAFNOSUPPORT;
2649 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2650 unsigned int slot_table_size,
2651 unsigned int max_slot_table_size)
2653 struct rpc_xprt *xprt;
2654 struct sock_xprt *new;
2656 if (args->addrlen > sizeof(xprt->addr)) {
2657 dprintk("RPC: xs_setup_xprt: address too large\n");
2658 return ERR_PTR(-EBADF);
2661 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2662 max_slot_table_size);
2664 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2666 return ERR_PTR(-ENOMEM);
2669 new = container_of(xprt, struct sock_xprt, xprt);
2670 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2671 xprt->addrlen = args->addrlen;
2673 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2676 err = xs_init_anyaddr(args->dstaddr->sa_family,
2677 (struct sockaddr *)&new->srcaddr);
2680 return ERR_PTR(err);
2687 static const struct rpc_timeout xs_local_default_timeout = {
2688 .to_initval = 10 * HZ,
2689 .to_maxval = 10 * HZ,
2694 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2695 * @args: rpc transport creation arguments
2697 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2699 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2701 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2702 struct sock_xprt *transport;
2703 struct rpc_xprt *xprt;
2704 struct rpc_xprt *ret;
2706 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2707 xprt_max_tcp_slot_table_entries);
2710 transport = container_of(xprt, struct sock_xprt, xprt);
2713 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2714 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2716 xprt->bind_timeout = XS_BIND_TO;
2717 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2718 xprt->idle_timeout = XS_IDLE_DISC_TO;
2720 xprt->ops = &xs_local_ops;
2721 xprt->timeout = &xs_local_default_timeout;
2723 INIT_DELAYED_WORK(&transport->connect_worker,
2724 xs_dummy_setup_socket);
2726 switch (sun->sun_family) {
2728 if (sun->sun_path[0] != '/') {
2729 dprintk("RPC: bad AF_LOCAL address: %s\n",
2731 ret = ERR_PTR(-EINVAL);
2734 xprt_set_bound(xprt);
2735 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2736 ret = ERR_PTR(xs_local_setup_socket(transport));
2741 ret = ERR_PTR(-EAFNOSUPPORT);
2745 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2746 xprt->address_strings[RPC_DISPLAY_ADDR]);
2748 if (try_module_get(THIS_MODULE))
2750 ret = ERR_PTR(-EINVAL);
2756 static const struct rpc_timeout xs_udp_default_timeout = {
2757 .to_initval = 5 * HZ,
2758 .to_maxval = 30 * HZ,
2759 .to_increment = 5 * HZ,
2764 * xs_setup_udp - Set up transport to use a UDP socket
2765 * @args: rpc transport creation arguments
2768 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2770 struct sockaddr *addr = args->dstaddr;
2771 struct rpc_xprt *xprt;
2772 struct sock_xprt *transport;
2773 struct rpc_xprt *ret;
2775 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2776 xprt_udp_slot_table_entries);
2779 transport = container_of(xprt, struct sock_xprt, xprt);
2781 xprt->prot = IPPROTO_UDP;
2783 /* XXX: header size can vary due to auth type, IPv6, etc. */
2784 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2786 xprt->bind_timeout = XS_BIND_TO;
2787 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2788 xprt->idle_timeout = XS_IDLE_DISC_TO;
2790 xprt->ops = &xs_udp_ops;
2792 xprt->timeout = &xs_udp_default_timeout;
2794 switch (addr->sa_family) {
2796 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2797 xprt_set_bound(xprt);
2799 INIT_DELAYED_WORK(&transport->connect_worker,
2800 xs_udp_setup_socket);
2801 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2804 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2805 xprt_set_bound(xprt);
2807 INIT_DELAYED_WORK(&transport->connect_worker,
2808 xs_udp_setup_socket);
2809 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2812 ret = ERR_PTR(-EAFNOSUPPORT);
2816 if (xprt_bound(xprt))
2817 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2818 xprt->address_strings[RPC_DISPLAY_ADDR],
2819 xprt->address_strings[RPC_DISPLAY_PORT],
2820 xprt->address_strings[RPC_DISPLAY_PROTO]);
2822 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2823 xprt->address_strings[RPC_DISPLAY_ADDR],
2824 xprt->address_strings[RPC_DISPLAY_PROTO]);
2826 if (try_module_get(THIS_MODULE))
2828 ret = ERR_PTR(-EINVAL);
2834 static const struct rpc_timeout xs_tcp_default_timeout = {
2835 .to_initval = 60 * HZ,
2836 .to_maxval = 60 * HZ,
2841 * xs_setup_tcp - Set up transport to use a TCP socket
2842 * @args: rpc transport creation arguments
2845 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2847 struct sockaddr *addr = args->dstaddr;
2848 struct rpc_xprt *xprt;
2849 struct sock_xprt *transport;
2850 struct rpc_xprt *ret;
2851 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2853 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2854 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2856 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2857 max_slot_table_size);
2860 transport = container_of(xprt, struct sock_xprt, xprt);
2862 xprt->prot = IPPROTO_TCP;
2863 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2864 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2866 xprt->bind_timeout = XS_BIND_TO;
2867 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2868 xprt->idle_timeout = XS_IDLE_DISC_TO;
2870 xprt->ops = &xs_tcp_ops;
2871 xprt->timeout = &xs_tcp_default_timeout;
2873 switch (addr->sa_family) {
2875 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2876 xprt_set_bound(xprt);
2878 INIT_DELAYED_WORK(&transport->connect_worker,
2879 xs_tcp_setup_socket);
2880 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2883 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2884 xprt_set_bound(xprt);
2886 INIT_DELAYED_WORK(&transport->connect_worker,
2887 xs_tcp_setup_socket);
2888 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2891 ret = ERR_PTR(-EAFNOSUPPORT);
2895 if (xprt_bound(xprt))
2896 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2897 xprt->address_strings[RPC_DISPLAY_ADDR],
2898 xprt->address_strings[RPC_DISPLAY_PORT],
2899 xprt->address_strings[RPC_DISPLAY_PROTO]);
2901 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2902 xprt->address_strings[RPC_DISPLAY_ADDR],
2903 xprt->address_strings[RPC_DISPLAY_PROTO]);
2905 if (try_module_get(THIS_MODULE))
2907 ret = ERR_PTR(-EINVAL);
2914 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2915 * @args: rpc transport creation arguments
2918 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2920 struct sockaddr *addr = args->dstaddr;
2921 struct rpc_xprt *xprt;
2922 struct sock_xprt *transport;
2923 struct svc_sock *bc_sock;
2924 struct rpc_xprt *ret;
2926 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2927 xprt_tcp_slot_table_entries);
2930 transport = container_of(xprt, struct sock_xprt, xprt);
2932 xprt->prot = IPPROTO_TCP;
2933 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2934 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2935 xprt->timeout = &xs_tcp_default_timeout;
2938 xprt_set_bound(xprt);
2939 xprt->bind_timeout = 0;
2940 xprt->reestablish_timeout = 0;
2941 xprt->idle_timeout = 0;
2943 xprt->ops = &bc_tcp_ops;
2945 switch (addr->sa_family) {
2947 xs_format_peer_addresses(xprt, "tcp",
2951 xs_format_peer_addresses(xprt, "tcp",
2952 RPCBIND_NETID_TCP6);
2955 ret = ERR_PTR(-EAFNOSUPPORT);
2959 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2960 xprt->address_strings[RPC_DISPLAY_ADDR],
2961 xprt->address_strings[RPC_DISPLAY_PORT],
2962 xprt->address_strings[RPC_DISPLAY_PROTO]);
2965 * Once we've associated a backchannel xprt with a connection,
2966 * we want to keep it around as long as the connection lasts,
2967 * in case we need to start using it for a backchannel again;
2968 * this reference won't be dropped until bc_xprt is destroyed.
2971 args->bc_xprt->xpt_bc_xprt = xprt;
2972 xprt->bc_xprt = args->bc_xprt;
2973 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2974 transport->sock = bc_sock->sk_sock;
2975 transport->inet = bc_sock->sk_sk;
2978 * Since we don't want connections for the backchannel, we set
2979 * the xprt status to connected
2981 xprt_set_connected(xprt);
2983 if (try_module_get(THIS_MODULE))
2986 ret = ERR_PTR(-EINVAL);
2992 static struct xprt_class xs_local_transport = {
2993 .list = LIST_HEAD_INIT(xs_local_transport.list),
2994 .name = "named UNIX socket",
2995 .owner = THIS_MODULE,
2996 .ident = XPRT_TRANSPORT_LOCAL,
2997 .setup = xs_setup_local,
3000 static struct xprt_class xs_udp_transport = {
3001 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3003 .owner = THIS_MODULE,
3004 .ident = XPRT_TRANSPORT_UDP,
3005 .setup = xs_setup_udp,
3008 static struct xprt_class xs_tcp_transport = {
3009 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3011 .owner = THIS_MODULE,
3012 .ident = XPRT_TRANSPORT_TCP,
3013 .setup = xs_setup_tcp,
3016 static struct xprt_class xs_bc_tcp_transport = {
3017 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3018 .name = "tcp NFSv4.1 backchannel",
3019 .owner = THIS_MODULE,
3020 .ident = XPRT_TRANSPORT_BC_TCP,
3021 .setup = xs_setup_bc_tcp,
3025 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3028 int init_socket_xprt(void)
3031 if (!sunrpc_table_header)
3032 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3035 xprt_register_transport(&xs_local_transport);
3036 xprt_register_transport(&xs_udp_transport);
3037 xprt_register_transport(&xs_tcp_transport);
3038 xprt_register_transport(&xs_bc_tcp_transport);
3044 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3047 void cleanup_socket_xprt(void)
3050 if (sunrpc_table_header) {
3051 unregister_sysctl_table(sunrpc_table_header);
3052 sunrpc_table_header = NULL;
3056 xprt_unregister_transport(&xs_local_transport);
3057 xprt_unregister_transport(&xs_udp_transport);
3058 xprt_unregister_transport(&xs_tcp_transport);
3059 xprt_unregister_transport(&xs_bc_tcp_transport);
3062 static int param_set_uint_minmax(const char *val,
3063 const struct kernel_param *kp,
3064 unsigned int min, unsigned int max)
3071 ret = strict_strtoul(val, 0, &num);
3072 if (ret == -EINVAL || num < min || num > max)
3074 *((unsigned int *)kp->arg) = num;
3078 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3080 return param_set_uint_minmax(val, kp,
3085 static struct kernel_param_ops param_ops_portnr = {
3086 .set = param_set_portnr,
3087 .get = param_get_uint,
3090 #define param_check_portnr(name, p) \
3091 __param_check(name, p, unsigned int);
3093 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3094 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3096 static int param_set_slot_table_size(const char *val,
3097 const struct kernel_param *kp)
3099 return param_set_uint_minmax(val, kp,
3101 RPC_MAX_SLOT_TABLE);
3104 static struct kernel_param_ops param_ops_slot_table_size = {
3105 .set = param_set_slot_table_size,
3106 .get = param_get_uint,
3109 #define param_check_slot_table_size(name, p) \
3110 __param_check(name, p, unsigned int);
3112 static int param_set_max_slot_table_size(const char *val,
3113 const struct kernel_param *kp)
3115 return param_set_uint_minmax(val, kp,
3117 RPC_MAX_SLOT_TABLE_LIMIT);
3120 static struct kernel_param_ops param_ops_max_slot_table_size = {
3121 .set = param_set_max_slot_table_size,
3122 .get = param_get_uint,
3125 #define param_check_max_slot_table_size(name, p) \
3126 __param_check(name, p, unsigned int);
3128 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3129 slot_table_size, 0644);
3130 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3131 max_slot_table_size, 0644);
3132 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3133 slot_table_size, 0644);