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 *);
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);
513 struct sock *sk = transport->inet;
516 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
517 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
520 /* Protect against races with write_space */
521 spin_lock_bh(&xprt->transport_lock);
523 /* Don't race with disconnect */
524 if (xprt_connected(xprt)) {
525 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
527 * Notify TCP that we're limited by the application
530 set_bit(SOCK_NOSPACE, &transport->sock->flags);
531 sk->sk_write_pending++;
532 /* ...and wait for more buffer space */
533 xprt_wait_for_buffer_space(task, xs_nospace_callback);
536 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
540 spin_unlock_bh(&xprt->transport_lock);
542 /* Race breaker in case memory is freed before above code is called */
543 sk->sk_write_space(sk);
548 * Construct a stream transport record marker in @buf.
550 static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
552 u32 reclen = buf->len - sizeof(rpc_fraghdr);
553 rpc_fraghdr *base = buf->head[0].iov_base;
554 *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
558 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
559 * @task: RPC task that manages the state of an RPC request
562 * 0: The request has been sent
563 * EAGAIN: The socket was blocked, please call again later to
564 * complete the request
565 * ENOTCONN: Caller needs to invoke connect logic then call again
566 * other: Some other error occured, the request was not sent
568 static int xs_local_send_request(struct rpc_task *task)
570 struct rpc_rqst *req = task->tk_rqstp;
571 struct rpc_xprt *xprt = req->rq_xprt;
572 struct sock_xprt *transport =
573 container_of(xprt, struct sock_xprt, xprt);
574 struct xdr_buf *xdr = &req->rq_snd_buf;
577 xs_encode_stream_record_marker(&req->rq_snd_buf);
579 xs_pktdump("packet data:",
580 req->rq_svec->iov_base, req->rq_svec->iov_len);
582 status = xs_sendpages(transport->sock, NULL, 0,
583 xdr, req->rq_bytes_sent, true);
584 dprintk("RPC: %s(%u) = %d\n",
585 __func__, xdr->len - req->rq_bytes_sent, status);
586 if (likely(status >= 0)) {
587 req->rq_bytes_sent += status;
588 req->rq_xmit_bytes_sent += status;
589 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
590 req->rq_bytes_sent = 0;
599 status = xs_nospace(task);
602 dprintk("RPC: sendmsg returned unrecognized error %d\n",
613 * xs_udp_send_request - write an RPC request to a UDP socket
614 * @task: address of RPC task that manages the state of an RPC request
617 * 0: The request has been sent
618 * EAGAIN: The socket was blocked, please call again later to
619 * complete the request
620 * ENOTCONN: Caller needs to invoke connect logic then call again
621 * other: Some other error occurred, the request was not sent
623 static int xs_udp_send_request(struct rpc_task *task)
625 struct rpc_rqst *req = task->tk_rqstp;
626 struct rpc_xprt *xprt = req->rq_xprt;
627 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
628 struct xdr_buf *xdr = &req->rq_snd_buf;
631 xs_pktdump("packet data:",
632 req->rq_svec->iov_base,
633 req->rq_svec->iov_len);
635 if (!xprt_bound(xprt))
637 status = xs_sendpages(transport->sock,
640 req->rq_bytes_sent, true);
642 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
643 xdr->len - req->rq_bytes_sent, status);
646 req->rq_xmit_bytes_sent += status;
647 if (status >= req->rq_slen)
649 /* Still some bytes left; set up for a retry later. */
656 /* Should we call xs_close() here? */
659 status = xs_nospace(task);
662 dprintk("RPC: sendmsg returned unrecognized error %d\n",
668 /* When the server has died, an ICMP port unreachable message
669 * prompts ECONNREFUSED. */
670 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
677 * xs_tcp_shutdown - gracefully shut down a TCP socket
680 * Initiates a graceful shutdown of the TCP socket by calling the
681 * equivalent of shutdown(SHUT_WR);
683 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
685 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
686 struct socket *sock = transport->sock;
689 kernel_sock_shutdown(sock, SHUT_WR);
690 trace_rpc_socket_shutdown(xprt, sock);
695 * xs_tcp_send_request - write an RPC request to a TCP socket
696 * @task: address of RPC task that manages the state of an RPC request
699 * 0: The request has been sent
700 * EAGAIN: The socket was blocked, please call again later to
701 * complete the request
702 * ENOTCONN: Caller needs to invoke connect logic then call again
703 * other: Some other error occurred, the request was not sent
705 * XXX: In the case of soft timeouts, should we eventually give up
706 * if sendmsg is not able to make progress?
708 static int xs_tcp_send_request(struct rpc_task *task)
710 struct rpc_rqst *req = task->tk_rqstp;
711 struct rpc_xprt *xprt = req->rq_xprt;
712 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
713 struct xdr_buf *xdr = &req->rq_snd_buf;
714 bool zerocopy = true;
717 xs_encode_stream_record_marker(&req->rq_snd_buf);
719 xs_pktdump("packet data:",
720 req->rq_svec->iov_base,
721 req->rq_svec->iov_len);
722 /* Don't use zero copy if this is a resend. If the RPC call
723 * completes while the socket holds a reference to the pages,
724 * then we may end up resending corrupted data.
726 if (task->tk_flags & RPC_TASK_SENT)
729 /* Continue transmitting the packet/record. We must be careful
730 * to cope with writespace callbacks arriving _after_ we have
731 * called sendmsg(). */
733 status = xs_sendpages(transport->sock,
734 NULL, 0, xdr, req->rq_bytes_sent,
737 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
738 xdr->len - req->rq_bytes_sent, status);
740 if (unlikely(status < 0))
743 /* If we've sent the entire packet, immediately
744 * reset the count of bytes sent. */
745 req->rq_bytes_sent += status;
746 req->rq_xmit_bytes_sent += status;
747 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
748 req->rq_bytes_sent = 0;
761 /* Should we call xs_close() here? */
765 status = xs_nospace(task);
768 dprintk("RPC: sendmsg returned unrecognized error %d\n",
771 xs_tcp_shutdown(xprt);
775 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
782 * xs_tcp_release_xprt - clean up after a tcp transmission
786 * This cleans up if an error causes us to abort the transmission of a request.
787 * In this case, the socket may need to be reset in order to avoid confusing
790 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
792 struct rpc_rqst *req;
794 if (task != xprt->snd_task)
798 req = task->tk_rqstp;
801 if (req->rq_bytes_sent == 0)
803 if (req->rq_bytes_sent == req->rq_snd_buf.len)
805 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
807 xprt_release_xprt(xprt, task);
810 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
812 transport->old_data_ready = sk->sk_data_ready;
813 transport->old_state_change = sk->sk_state_change;
814 transport->old_write_space = sk->sk_write_space;
815 transport->old_error_report = sk->sk_error_report;
818 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
820 sk->sk_data_ready = transport->old_data_ready;
821 sk->sk_state_change = transport->old_state_change;
822 sk->sk_write_space = transport->old_write_space;
823 sk->sk_error_report = transport->old_error_report;
827 * xs_error_report - callback to handle TCP socket state errors
830 * Note: we don't call sock_error() since there may be a rpc_task
831 * using the socket, and so we don't want to clear sk->sk_err.
833 static void xs_error_report(struct sock *sk)
835 struct rpc_xprt *xprt;
838 read_lock_bh(&sk->sk_callback_lock);
839 if (!(xprt = xprt_from_sock(sk)))
845 dprintk("RPC: xs_error_report client %p, error=%d...\n",
847 trace_rpc_socket_error(xprt, sk->sk_socket, err);
848 xprt_wake_pending_tasks(xprt, err);
850 read_unlock_bh(&sk->sk_callback_lock);
853 static void xs_reset_transport(struct sock_xprt *transport)
855 struct socket *sock = transport->sock;
856 struct sock *sk = transport->inet;
861 transport->srcport = 0;
863 write_lock_bh(&sk->sk_callback_lock);
864 transport->inet = NULL;
865 transport->sock = NULL;
867 sk->sk_user_data = NULL;
869 xs_restore_old_callbacks(transport, sk);
870 write_unlock_bh(&sk->sk_callback_lock);
872 trace_rpc_socket_close(&transport->xprt, sock);
877 * xs_close - close a socket
880 * This is used when all requests are complete; ie, no DRC state remains
881 * on the server we want to save.
883 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
884 * xs_reset_transport() zeroing the socket from underneath a writer.
886 static void xs_close(struct rpc_xprt *xprt)
888 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
890 dprintk("RPC: xs_close xprt %p\n", xprt);
892 cancel_delayed_work_sync(&transport->connect_worker);
894 xs_reset_transport(transport);
895 xprt->reestablish_timeout = 0;
897 smp_mb__before_atomic();
898 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
899 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
900 clear_bit(XPRT_CLOSING, &xprt->state);
901 smp_mb__after_atomic();
902 xprt_disconnect_done(xprt);
905 static void xs_tcp_close(struct rpc_xprt *xprt)
907 if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state))
910 xs_tcp_shutdown(xprt);
913 static void xs_xprt_free(struct rpc_xprt *xprt)
915 xs_free_peer_addresses(xprt);
920 * xs_destroy - prepare to shutdown a transport
921 * @xprt: doomed transport
924 static void xs_destroy(struct rpc_xprt *xprt)
926 dprintk("RPC: xs_destroy xprt %p\n", xprt);
930 module_put(THIS_MODULE);
933 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
935 struct xdr_skb_reader desc = {
937 .offset = sizeof(rpc_fraghdr),
938 .count = skb->len - sizeof(rpc_fraghdr),
941 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
949 * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets
950 * @sk: socket with data to read
951 * @len: how much data to read
953 * Currently this assumes we can read the whole reply in a single gulp.
955 static void xs_local_data_ready(struct sock *sk)
957 struct rpc_task *task;
958 struct rpc_xprt *xprt;
959 struct rpc_rqst *rovr;
961 int err, repsize, copied;
965 read_lock_bh(&sk->sk_callback_lock);
966 dprintk("RPC: %s...\n", __func__);
967 xprt = xprt_from_sock(sk);
971 skb = skb_recv_datagram(sk, 0, 1, &err);
975 repsize = skb->len - sizeof(rpc_fraghdr);
977 dprintk("RPC: impossible RPC reply size %d\n", repsize);
981 /* Copy the XID from the skb... */
982 xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
986 /* Look up and lock the request corresponding to the given XID */
987 spin_lock(&xprt->transport_lock);
988 rovr = xprt_lookup_rqst(xprt, *xp);
991 task = rovr->rq_task;
993 copied = rovr->rq_private_buf.buflen;
994 if (copied > repsize)
997 if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
998 dprintk("RPC: sk_buff copy failed\n");
1002 xprt_complete_rqst(task, copied);
1005 spin_unlock(&xprt->transport_lock);
1007 skb_free_datagram(sk, skb);
1009 read_unlock_bh(&sk->sk_callback_lock);
1013 * xs_udp_data_ready - "data ready" callback for UDP sockets
1014 * @sk: socket with data to read
1015 * @len: how much data to read
1018 static void xs_udp_data_ready(struct sock *sk)
1020 struct rpc_task *task;
1021 struct rpc_xprt *xprt;
1022 struct rpc_rqst *rovr;
1023 struct sk_buff *skb;
1024 int err, repsize, copied;
1028 read_lock_bh(&sk->sk_callback_lock);
1029 dprintk("RPC: xs_udp_data_ready...\n");
1030 if (!(xprt = xprt_from_sock(sk)))
1033 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
1036 repsize = skb->len - sizeof(struct udphdr);
1038 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
1042 /* Copy the XID from the skb... */
1043 xp = skb_header_pointer(skb, sizeof(struct udphdr),
1044 sizeof(_xid), &_xid);
1048 /* Look up and lock the request corresponding to the given XID */
1049 spin_lock(&xprt->transport_lock);
1050 rovr = xprt_lookup_rqst(xprt, *xp);
1053 task = rovr->rq_task;
1055 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1058 /* Suck it into the iovec, verify checksum if not done by hw. */
1059 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1060 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
1064 UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
1066 xprt_adjust_cwnd(xprt, task, copied);
1067 xprt_complete_rqst(task, copied);
1070 spin_unlock(&xprt->transport_lock);
1072 skb_free_datagram(sk, skb);
1074 read_unlock_bh(&sk->sk_callback_lock);
1078 * Helper function to force a TCP close if the server is sending
1079 * junk and/or it has put us in CLOSE_WAIT
1081 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1083 set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1084 xprt_force_disconnect(xprt);
1087 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1089 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1093 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1094 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1095 used = xdr_skb_read_bits(desc, p, len);
1096 transport->tcp_offset += used;
1100 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1101 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1102 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1104 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1105 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1107 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1108 transport->tcp_offset = 0;
1110 /* Sanity check of the record length */
1111 if (unlikely(transport->tcp_reclen < 8)) {
1112 dprintk("RPC: invalid TCP record fragment length\n");
1113 xs_tcp_force_close(xprt);
1116 dprintk("RPC: reading TCP record fragment of length %d\n",
1117 transport->tcp_reclen);
1120 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1122 if (transport->tcp_offset == transport->tcp_reclen) {
1123 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1124 transport->tcp_offset = 0;
1125 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1126 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1127 transport->tcp_flags |= TCP_RCV_COPY_XID;
1128 transport->tcp_copied = 0;
1133 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1138 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1139 dprintk("RPC: reading XID (%Zu bytes)\n", len);
1140 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1141 used = xdr_skb_read_bits(desc, p, len);
1142 transport->tcp_offset += used;
1145 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1146 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1147 transport->tcp_copied = 4;
1148 dprintk("RPC: reading %s XID %08x\n",
1149 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1151 ntohl(transport->tcp_xid));
1152 xs_tcp_check_fraghdr(transport);
1155 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1156 struct xdr_skb_reader *desc)
1163 * We want transport->tcp_offset to be 8 at the end of this routine
1164 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1165 * When this function is called for the first time,
1166 * transport->tcp_offset is 4 (after having already read the xid).
1168 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1169 len = sizeof(transport->tcp_calldir) - offset;
1170 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len);
1171 p = ((char *) &transport->tcp_calldir) + offset;
1172 used = xdr_skb_read_bits(desc, p, len);
1173 transport->tcp_offset += used;
1176 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1178 * We don't yet have the XDR buffer, so we will write the calldir
1179 * out after we get the buffer from the 'struct rpc_rqst'
1181 switch (ntohl(transport->tcp_calldir)) {
1183 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1184 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1185 transport->tcp_flags |= TCP_RPC_REPLY;
1188 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1189 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1190 transport->tcp_flags &= ~TCP_RPC_REPLY;
1193 dprintk("RPC: invalid request message type\n");
1194 xs_tcp_force_close(&transport->xprt);
1196 xs_tcp_check_fraghdr(transport);
1199 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1200 struct xdr_skb_reader *desc,
1201 struct rpc_rqst *req)
1203 struct sock_xprt *transport =
1204 container_of(xprt, struct sock_xprt, xprt);
1205 struct xdr_buf *rcvbuf;
1209 rcvbuf = &req->rq_private_buf;
1211 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1213 * Save the RPC direction in the XDR buffer
1215 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1216 &transport->tcp_calldir,
1217 sizeof(transport->tcp_calldir));
1218 transport->tcp_copied += sizeof(transport->tcp_calldir);
1219 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1223 if (len > transport->tcp_reclen - transport->tcp_offset) {
1224 struct xdr_skb_reader my_desc;
1226 len = transport->tcp_reclen - transport->tcp_offset;
1227 memcpy(&my_desc, desc, sizeof(my_desc));
1228 my_desc.count = len;
1229 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1230 &my_desc, xdr_skb_read_bits);
1234 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1235 desc, xdr_skb_read_bits);
1238 transport->tcp_copied += r;
1239 transport->tcp_offset += r;
1242 /* Error when copying to the receive buffer,
1243 * usually because we weren't able to allocate
1244 * additional buffer pages. All we can do now
1245 * is turn off TCP_RCV_COPY_DATA, so the request
1246 * will not receive any additional updates,
1248 * Any remaining data from this record will
1251 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1252 dprintk("RPC: XID %08x truncated request\n",
1253 ntohl(transport->tcp_xid));
1254 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1255 "tcp_offset = %u, tcp_reclen = %u\n",
1256 xprt, transport->tcp_copied,
1257 transport->tcp_offset, transport->tcp_reclen);
1261 dprintk("RPC: XID %08x read %Zd bytes\n",
1262 ntohl(transport->tcp_xid), r);
1263 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1264 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1265 transport->tcp_offset, transport->tcp_reclen);
1267 if (transport->tcp_copied == req->rq_private_buf.buflen)
1268 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1269 else if (transport->tcp_offset == transport->tcp_reclen) {
1270 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1271 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1276 * Finds the request corresponding to the RPC xid and invokes the common
1277 * tcp read code to read the data.
1279 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1280 struct xdr_skb_reader *desc)
1282 struct sock_xprt *transport =
1283 container_of(xprt, struct sock_xprt, xprt);
1284 struct rpc_rqst *req;
1286 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1288 /* Find and lock the request corresponding to this xid */
1289 spin_lock(&xprt->transport_lock);
1290 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1292 dprintk("RPC: XID %08x request not found!\n",
1293 ntohl(transport->tcp_xid));
1294 spin_unlock(&xprt->transport_lock);
1298 xs_tcp_read_common(xprt, desc, req);
1300 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1301 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1303 spin_unlock(&xprt->transport_lock);
1307 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1309 * Obtains an rpc_rqst previously allocated and invokes the common
1310 * tcp read code to read the data. The result is placed in the callback
1312 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1313 * connection and return -1.
1315 static int xs_tcp_read_callback(struct rpc_xprt *xprt,
1316 struct xdr_skb_reader *desc)
1318 struct sock_xprt *transport =
1319 container_of(xprt, struct sock_xprt, xprt);
1320 struct rpc_rqst *req;
1322 /* Look up and lock the request corresponding to the given XID */
1323 spin_lock(&xprt->transport_lock);
1324 req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
1326 spin_unlock(&xprt->transport_lock);
1327 printk(KERN_WARNING "Callback slot table overflowed\n");
1328 xprt_force_disconnect(xprt);
1332 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1333 xs_tcp_read_common(xprt, desc, req);
1335 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1336 xprt_complete_bc_request(req, transport->tcp_copied);
1337 spin_unlock(&xprt->transport_lock);
1342 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1343 struct xdr_skb_reader *desc)
1345 struct sock_xprt *transport =
1346 container_of(xprt, struct sock_xprt, xprt);
1348 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1349 xs_tcp_read_reply(xprt, desc) :
1350 xs_tcp_read_callback(xprt, desc);
1353 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1354 struct xdr_skb_reader *desc)
1356 return xs_tcp_read_reply(xprt, desc);
1358 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1361 * Read data off the transport. This can be either an RPC_CALL or an
1362 * RPC_REPLY. Relay the processing to helper functions.
1364 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1365 struct xdr_skb_reader *desc)
1367 struct sock_xprt *transport =
1368 container_of(xprt, struct sock_xprt, xprt);
1370 if (_xs_tcp_read_data(xprt, desc) == 0)
1371 xs_tcp_check_fraghdr(transport);
1374 * The transport_lock protects the request handling.
1375 * There's no need to hold it to update the tcp_flags.
1377 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1381 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1385 len = transport->tcp_reclen - transport->tcp_offset;
1386 if (len > desc->count)
1389 desc->offset += len;
1390 transport->tcp_offset += len;
1391 dprintk("RPC: discarded %Zu bytes\n", len);
1392 xs_tcp_check_fraghdr(transport);
1395 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1397 struct rpc_xprt *xprt = rd_desc->arg.data;
1398 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1399 struct xdr_skb_reader desc = {
1405 dprintk("RPC: xs_tcp_data_recv started\n");
1407 /* Read in a new fragment marker if necessary */
1408 /* Can we ever really expect to get completely empty fragments? */
1409 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1410 xs_tcp_read_fraghdr(xprt, &desc);
1413 /* Read in the xid if necessary */
1414 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1415 xs_tcp_read_xid(transport, &desc);
1418 /* Read in the call/reply flag */
1419 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1420 xs_tcp_read_calldir(transport, &desc);
1423 /* Read in the request data */
1424 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1425 xs_tcp_read_data(xprt, &desc);
1428 /* Skip over any trailing bytes on short reads */
1429 xs_tcp_read_discard(transport, &desc);
1430 } while (desc.count);
1431 dprintk("RPC: xs_tcp_data_recv done\n");
1432 return len - desc.count;
1436 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1437 * @sk: socket with data to read
1438 * @bytes: how much data to read
1441 static void xs_tcp_data_ready(struct sock *sk)
1443 struct rpc_xprt *xprt;
1444 read_descriptor_t rd_desc;
1447 dprintk("RPC: xs_tcp_data_ready...\n");
1449 read_lock_bh(&sk->sk_callback_lock);
1450 if (!(xprt = xprt_from_sock(sk)))
1452 /* Any data means we had a useful conversation, so
1453 * the we don't need to delay the next reconnect
1455 if (xprt->reestablish_timeout)
1456 xprt->reestablish_timeout = 0;
1458 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1459 rd_desc.arg.data = xprt;
1461 rd_desc.count = 65536;
1462 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1465 read_unlock_bh(&sk->sk_callback_lock);
1469 * Do the equivalent of linger/linger2 handling for dealing with
1470 * broken servers that don't close the socket in a timely
1473 static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1474 unsigned long timeout)
1476 struct sock_xprt *transport;
1478 if (xprt_test_and_set_connecting(xprt))
1480 set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1481 transport = container_of(xprt, struct sock_xprt, xprt);
1482 queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1486 static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1488 struct sock_xprt *transport;
1490 transport = container_of(xprt, struct sock_xprt, xprt);
1492 if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1493 !cancel_delayed_work(&transport->connect_worker))
1495 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1496 xprt_clear_connecting(xprt);
1499 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1501 smp_mb__before_atomic();
1502 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1503 clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1504 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1505 clear_bit(XPRT_CLOSING, &xprt->state);
1506 smp_mb__after_atomic();
1509 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1511 xs_sock_reset_connection_flags(xprt);
1512 /* Mark transport as closed and wake up all pending tasks */
1513 xprt_disconnect_done(xprt);
1517 * xs_tcp_state_change - callback to handle TCP socket state changes
1518 * @sk: socket whose state has changed
1521 static void xs_tcp_state_change(struct sock *sk)
1523 struct rpc_xprt *xprt;
1525 read_lock_bh(&sk->sk_callback_lock);
1526 if (!(xprt = xprt_from_sock(sk)))
1528 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1529 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1530 sk->sk_state, xprt_connected(xprt),
1531 sock_flag(sk, SOCK_DEAD),
1532 sock_flag(sk, SOCK_ZAPPED),
1535 trace_rpc_socket_state_change(xprt, sk->sk_socket);
1536 switch (sk->sk_state) {
1537 case TCP_ESTABLISHED:
1538 spin_lock(&xprt->transport_lock);
1539 if (!xprt_test_and_set_connected(xprt)) {
1540 struct sock_xprt *transport = container_of(xprt,
1541 struct sock_xprt, xprt);
1543 /* Reset TCP record info */
1544 transport->tcp_offset = 0;
1545 transport->tcp_reclen = 0;
1546 transport->tcp_copied = 0;
1547 transport->tcp_flags =
1548 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1549 xprt->connect_cookie++;
1551 xprt_wake_pending_tasks(xprt, -EAGAIN);
1553 spin_unlock(&xprt->transport_lock);
1556 /* The client initiated a shutdown of the socket */
1557 xprt->connect_cookie++;
1558 xprt->reestablish_timeout = 0;
1559 set_bit(XPRT_CLOSING, &xprt->state);
1560 smp_mb__before_atomic();
1561 clear_bit(XPRT_CONNECTED, &xprt->state);
1562 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1563 smp_mb__after_atomic();
1564 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1566 case TCP_CLOSE_WAIT:
1567 /* The server initiated a shutdown of the socket */
1568 xprt->connect_cookie++;
1569 clear_bit(XPRT_CONNECTED, &xprt->state);
1570 xs_tcp_force_close(xprt);
1573 * If the server closed down the connection, make sure that
1574 * we back off before reconnecting
1576 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1577 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1580 set_bit(XPRT_CLOSING, &xprt->state);
1581 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1582 smp_mb__before_atomic();
1583 clear_bit(XPRT_CONNECTED, &xprt->state);
1584 smp_mb__after_atomic();
1587 xs_tcp_cancel_linger_timeout(xprt);
1588 xs_sock_mark_closed(xprt);
1591 read_unlock_bh(&sk->sk_callback_lock);
1594 static void xs_write_space(struct sock *sk)
1596 struct socket *sock;
1597 struct rpc_xprt *xprt;
1599 if (unlikely(!(sock = sk->sk_socket)))
1601 clear_bit(SOCK_NOSPACE, &sock->flags);
1603 if (unlikely(!(xprt = xprt_from_sock(sk))))
1605 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1608 xprt_write_space(xprt);
1612 * xs_udp_write_space - callback invoked when socket buffer space
1614 * @sk: socket whose state has changed
1616 * Called when more output buffer space is available for this socket.
1617 * We try not to wake our writers until they can make "significant"
1618 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1619 * with a bunch of small requests.
1621 static void xs_udp_write_space(struct sock *sk)
1623 read_lock_bh(&sk->sk_callback_lock);
1625 /* from net/core/sock.c:sock_def_write_space */
1626 if (sock_writeable(sk))
1629 read_unlock_bh(&sk->sk_callback_lock);
1633 * xs_tcp_write_space - callback invoked when socket buffer space
1635 * @sk: socket whose state has changed
1637 * Called when more output buffer space is available for this socket.
1638 * We try not to wake our writers until they can make "significant"
1639 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1640 * with a bunch of small requests.
1642 static void xs_tcp_write_space(struct sock *sk)
1644 read_lock_bh(&sk->sk_callback_lock);
1646 /* from net/core/stream.c:sk_stream_write_space */
1647 if (sk_stream_is_writeable(sk))
1650 read_unlock_bh(&sk->sk_callback_lock);
1653 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1655 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1656 struct sock *sk = transport->inet;
1658 if (transport->rcvsize) {
1659 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1660 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1662 if (transport->sndsize) {
1663 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1664 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1665 sk->sk_write_space(sk);
1670 * xs_udp_set_buffer_size - set send and receive limits
1671 * @xprt: generic transport
1672 * @sndsize: requested size of send buffer, in bytes
1673 * @rcvsize: requested size of receive buffer, in bytes
1675 * Set socket send and receive buffer size limits.
1677 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1679 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1681 transport->sndsize = 0;
1683 transport->sndsize = sndsize + 1024;
1684 transport->rcvsize = 0;
1686 transport->rcvsize = rcvsize + 1024;
1688 xs_udp_do_set_buffer_size(xprt);
1692 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1693 * @task: task that timed out
1695 * Adjust the congestion window after a retransmit timeout has occurred.
1697 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1699 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1702 static unsigned short xs_get_random_port(void)
1704 unsigned short range = xprt_max_resvport - xprt_min_resvport;
1705 unsigned short rand = (unsigned short) prandom_u32() % range;
1706 return rand + xprt_min_resvport;
1710 * xs_set_port - reset the port number in the remote endpoint address
1711 * @xprt: generic transport
1712 * @port: new port number
1715 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1717 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1719 rpc_set_port(xs_addr(xprt), port);
1720 xs_update_peer_port(xprt);
1723 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1725 unsigned short port = transport->srcport;
1727 if (port == 0 && transport->xprt.resvport)
1728 port = xs_get_random_port();
1732 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1734 if (transport->srcport != 0)
1735 transport->srcport = 0;
1736 if (!transport->xprt.resvport)
1738 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1739 return xprt_max_resvport;
1742 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1744 struct sockaddr_storage myaddr;
1746 unsigned short port = xs_get_srcport(transport);
1747 unsigned short last;
1749 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1751 rpc_set_port((struct sockaddr *)&myaddr, port);
1752 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1753 transport->xprt.addrlen);
1757 transport->srcport = port;
1761 port = xs_next_srcport(transport, port);
1764 } while (err == -EADDRINUSE && nloop != 2);
1766 if (myaddr.ss_family == AF_INET)
1767 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1768 &((struct sockaddr_in *)&myaddr)->sin_addr,
1769 port, err ? "failed" : "ok", err);
1771 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1772 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1773 port, err ? "failed" : "ok", err);
1778 * We don't support autobind on AF_LOCAL sockets
1780 static void xs_local_rpcbind(struct rpc_task *task)
1783 xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt));
1787 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1791 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1792 static struct lock_class_key xs_key[2];
1793 static struct lock_class_key xs_slock_key[2];
1795 static inline void xs_reclassify_socketu(struct socket *sock)
1797 struct sock *sk = sock->sk;
1799 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1800 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1803 static inline void xs_reclassify_socket4(struct socket *sock)
1805 struct sock *sk = sock->sk;
1807 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1808 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1811 static inline void xs_reclassify_socket6(struct socket *sock)
1813 struct sock *sk = sock->sk;
1815 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1816 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1819 static inline void xs_reclassify_socket(int family, struct socket *sock)
1821 WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1822 if (sock_owned_by_user(sock->sk))
1827 xs_reclassify_socketu(sock);
1830 xs_reclassify_socket4(sock);
1833 xs_reclassify_socket6(sock);
1838 static inline void xs_reclassify_socketu(struct socket *sock)
1842 static inline void xs_reclassify_socket4(struct socket *sock)
1846 static inline void xs_reclassify_socket6(struct socket *sock)
1850 static inline void xs_reclassify_socket(int family, struct socket *sock)
1855 static void xs_dummy_setup_socket(struct work_struct *work)
1859 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1860 struct sock_xprt *transport, int family, int type, int protocol)
1862 struct socket *sock;
1865 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1867 dprintk("RPC: can't create %d transport socket (%d).\n",
1871 xs_reclassify_socket(family, sock);
1873 err = xs_bind(transport, sock);
1881 return ERR_PTR(err);
1884 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1885 struct socket *sock)
1887 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1890 if (!transport->inet) {
1891 struct sock *sk = sock->sk;
1893 write_lock_bh(&sk->sk_callback_lock);
1895 xs_save_old_callbacks(transport, sk);
1897 sk->sk_user_data = xprt;
1898 sk->sk_data_ready = xs_local_data_ready;
1899 sk->sk_write_space = xs_udp_write_space;
1900 sk->sk_error_report = xs_error_report;
1901 sk->sk_allocation = GFP_ATOMIC;
1903 xprt_clear_connected(xprt);
1905 /* Reset to new socket */
1906 transport->sock = sock;
1907 transport->inet = sk;
1909 write_unlock_bh(&sk->sk_callback_lock);
1912 /* Tell the socket layer to start connecting... */
1913 xprt->stat.connect_count++;
1914 xprt->stat.connect_start = jiffies;
1915 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1919 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1920 * @xprt: RPC transport to connect
1921 * @transport: socket transport to connect
1922 * @create_sock: function to create a socket of the correct type
1924 static int xs_local_setup_socket(struct sock_xprt *transport)
1926 struct rpc_xprt *xprt = &transport->xprt;
1927 struct socket *sock;
1930 current->flags |= PF_FSTRANS;
1932 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1933 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1934 SOCK_STREAM, 0, &sock, 1);
1936 dprintk("RPC: can't create AF_LOCAL "
1937 "transport socket (%d).\n", -status);
1940 xs_reclassify_socketu(sock);
1942 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1943 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1945 status = xs_local_finish_connecting(xprt, sock);
1946 trace_rpc_socket_connect(xprt, sock, status);
1949 dprintk("RPC: xprt %p connected to %s\n",
1950 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1951 xprt_set_connected(xprt);
1955 dprintk("RPC: xprt %p: socket %s does not exist\n",
1956 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1959 dprintk("RPC: xprt %p: connection refused for %s\n",
1960 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1963 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1965 xprt->address_strings[RPC_DISPLAY_ADDR]);
1969 xprt_clear_connecting(xprt);
1970 xprt_wake_pending_tasks(xprt, status);
1971 current->flags &= ~PF_FSTRANS;
1975 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1977 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1980 if (RPC_IS_ASYNC(task)) {
1982 * We want the AF_LOCAL connect to be resolved in the
1983 * filesystem namespace of the process making the rpc
1984 * call. Thus we connect synchronously.
1986 * If we want to support asynchronous AF_LOCAL calls,
1987 * we'll need to figure out how to pass a namespace to
1990 rpc_exit(task, -ENOTCONN);
1993 ret = xs_local_setup_socket(transport);
1994 if (ret && !RPC_IS_SOFTCONN(task))
1995 msleep_interruptible(15000);
1998 #ifdef CONFIG_SUNRPC_SWAP
1999 static void xs_set_memalloc(struct rpc_xprt *xprt)
2001 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2005 sk_set_memalloc(transport->inet);
2009 * xs_swapper - Tag this transport as being used for swap.
2010 * @xprt: transport to tag
2011 * @enable: enable/disable
2014 int xs_swapper(struct rpc_xprt *xprt, int enable)
2016 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2022 xs_set_memalloc(xprt);
2023 } else if (xprt->swapper) {
2025 sk_clear_memalloc(transport->inet);
2030 EXPORT_SYMBOL_GPL(xs_swapper);
2032 static void xs_set_memalloc(struct rpc_xprt *xprt)
2037 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2039 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2041 if (!transport->inet) {
2042 struct sock *sk = sock->sk;
2044 write_lock_bh(&sk->sk_callback_lock);
2046 xs_save_old_callbacks(transport, sk);
2048 sk->sk_user_data = xprt;
2049 sk->sk_data_ready = xs_udp_data_ready;
2050 sk->sk_write_space = xs_udp_write_space;
2051 sk->sk_allocation = GFP_ATOMIC;
2053 xprt_set_connected(xprt);
2055 /* Reset to new socket */
2056 transport->sock = sock;
2057 transport->inet = sk;
2059 xs_set_memalloc(xprt);
2061 write_unlock_bh(&sk->sk_callback_lock);
2063 xs_udp_do_set_buffer_size(xprt);
2066 static void xs_udp_setup_socket(struct work_struct *work)
2068 struct sock_xprt *transport =
2069 container_of(work, struct sock_xprt, connect_worker.work);
2070 struct rpc_xprt *xprt = &transport->xprt;
2071 struct socket *sock = transport->sock;
2074 current->flags |= PF_FSTRANS;
2076 /* Start by resetting any existing state */
2077 xs_reset_transport(transport);
2078 sock = xs_create_sock(xprt, transport,
2079 xs_addr(xprt)->sa_family, SOCK_DGRAM, IPPROTO_UDP);
2083 dprintk("RPC: worker connecting xprt %p via %s to "
2084 "%s (port %s)\n", xprt,
2085 xprt->address_strings[RPC_DISPLAY_PROTO],
2086 xprt->address_strings[RPC_DISPLAY_ADDR],
2087 xprt->address_strings[RPC_DISPLAY_PORT]);
2089 xs_udp_finish_connecting(xprt, sock);
2090 trace_rpc_socket_connect(xprt, sock, 0);
2093 xprt_clear_connecting(xprt);
2094 xprt_wake_pending_tasks(xprt, status);
2095 current->flags &= ~PF_FSTRANS;
2099 * We need to preserve the port number so the reply cache on the server can
2100 * find our cached RPC replies when we get around to reconnecting.
2102 static void xs_abort_connection(struct sock_xprt *transport)
2105 struct sockaddr any;
2107 dprintk("RPC: disconnecting xprt %p to reuse port\n", transport);
2110 * Disconnect the transport socket by doing a connect operation
2111 * with AF_UNSPEC. This should return immediately...
2113 memset(&any, 0, sizeof(any));
2114 any.sa_family = AF_UNSPEC;
2115 result = kernel_connect(transport->sock, &any, sizeof(any), 0);
2116 trace_rpc_socket_reset_connection(&transport->xprt,
2117 transport->sock, result);
2119 xs_sock_reset_connection_flags(&transport->xprt);
2120 dprintk("RPC: AF_UNSPEC connect return code %d\n", result);
2123 static void xs_tcp_reuse_connection(struct sock_xprt *transport)
2125 unsigned int state = transport->inet->sk_state;
2127 if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED) {
2128 /* we don't need to abort the connection if the socket
2129 * hasn't undergone a shutdown
2131 if (transport->inet->sk_shutdown == 0)
2133 dprintk("RPC: %s: TCP_CLOSEd and sk_shutdown set to %d\n",
2134 __func__, transport->inet->sk_shutdown);
2136 if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT)) {
2137 /* we don't need to abort the connection if the socket
2138 * hasn't undergone a shutdown
2140 if (transport->inet->sk_shutdown == 0)
2142 dprintk("RPC: %s: ESTABLISHED/SYN_SENT "
2143 "sk_shutdown set to %d\n",
2144 __func__, transport->inet->sk_shutdown);
2146 xs_abort_connection(transport);
2149 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2151 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2152 int ret = -ENOTCONN;
2154 if (!transport->inet) {
2155 struct sock *sk = sock->sk;
2156 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2157 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2158 unsigned int opt_on = 1;
2160 /* TCP Keepalive options */
2161 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2162 (char *)&opt_on, sizeof(opt_on));
2163 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2164 (char *)&keepidle, sizeof(keepidle));
2165 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2166 (char *)&keepidle, sizeof(keepidle));
2167 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2168 (char *)&keepcnt, sizeof(keepcnt));
2170 write_lock_bh(&sk->sk_callback_lock);
2172 xs_save_old_callbacks(transport, sk);
2174 sk->sk_user_data = xprt;
2175 sk->sk_data_ready = xs_tcp_data_ready;
2176 sk->sk_state_change = xs_tcp_state_change;
2177 sk->sk_write_space = xs_tcp_write_space;
2178 sk->sk_error_report = xs_error_report;
2179 sk->sk_allocation = GFP_ATOMIC;
2181 /* socket options */
2182 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
2183 sock_reset_flag(sk, SOCK_LINGER);
2184 tcp_sk(sk)->linger2 = 0;
2185 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2187 xprt_clear_connected(xprt);
2189 /* Reset to new socket */
2190 transport->sock = sock;
2191 transport->inet = sk;
2193 write_unlock_bh(&sk->sk_callback_lock);
2196 if (!xprt_bound(xprt))
2199 xs_set_memalloc(xprt);
2201 /* Tell the socket layer to start connecting... */
2202 xprt->stat.connect_count++;
2203 xprt->stat.connect_start = jiffies;
2204 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2209 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2210 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2217 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2218 * @xprt: RPC transport to connect
2219 * @transport: socket transport to connect
2220 * @create_sock: function to create a socket of the correct type
2222 * Invoked by a work queue tasklet.
2224 static void xs_tcp_setup_socket(struct work_struct *work)
2226 struct sock_xprt *transport =
2227 container_of(work, struct sock_xprt, connect_worker.work);
2228 struct socket *sock = transport->sock;
2229 struct rpc_xprt *xprt = &transport->xprt;
2232 current->flags |= PF_FSTRANS;
2235 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
2236 sock = xs_create_sock(xprt, transport,
2237 xs_addr(xprt)->sa_family, SOCK_STREAM, IPPROTO_TCP);
2239 status = PTR_ERR(sock);
2245 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
2247 /* "close" the socket, preserving the local port */
2248 xs_tcp_reuse_connection(transport);
2254 dprintk("RPC: worker connecting xprt %p via %s to "
2255 "%s (port %s)\n", xprt,
2256 xprt->address_strings[RPC_DISPLAY_PROTO],
2257 xprt->address_strings[RPC_DISPLAY_ADDR],
2258 xprt->address_strings[RPC_DISPLAY_PORT]);
2260 status = xs_tcp_finish_connecting(xprt, sock);
2261 trace_rpc_socket_connect(xprt, sock, status);
2262 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2263 xprt, -status, xprt_connected(xprt),
2264 sock->sk->sk_state);
2267 printk("%s: connect returned unhandled error %d\n",
2269 case -EADDRNOTAVAIL:
2270 /* We're probably in TIME_WAIT. Get rid of existing socket,
2273 xs_tcp_force_close(xprt);
2278 xprt_clear_connecting(xprt);
2279 current->flags &= ~PF_FSTRANS;
2282 /* Happens, for instance, if the user specified a link
2283 * local IPv6 address without a scope-id.
2289 /* retry with existing socket, after a delay */
2295 xprt_clear_connecting(xprt);
2296 xprt_wake_pending_tasks(xprt, status);
2297 current->flags &= ~PF_FSTRANS;
2301 * xs_connect - connect a socket to a remote endpoint
2302 * @xprt: pointer to transport structure
2303 * @task: address of RPC task that manages state of connect request
2305 * TCP: If the remote end dropped the connection, delay reconnecting.
2307 * UDP socket connects are synchronous, but we use a work queue anyway
2308 * to guarantee that even unprivileged user processes can set up a
2309 * socket on a privileged port.
2311 * If a UDP socket connect fails, the delay behavior here prevents
2312 * retry floods (hard mounts).
2314 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2316 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2318 if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
2319 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2321 xprt, xprt->reestablish_timeout / HZ);
2322 queue_delayed_work(rpciod_workqueue,
2323 &transport->connect_worker,
2324 xprt->reestablish_timeout);
2325 xprt->reestablish_timeout <<= 1;
2326 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2327 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2328 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2329 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2331 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2332 queue_delayed_work(rpciod_workqueue,
2333 &transport->connect_worker, 0);
2338 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2339 * @xprt: rpc_xprt struct containing statistics
2343 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2347 if (xprt_connected(xprt))
2348 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2350 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2351 "%llu %llu %lu %llu %llu\n",
2352 xprt->stat.bind_count,
2353 xprt->stat.connect_count,
2354 xprt->stat.connect_time,
2358 xprt->stat.bad_xids,
2361 xprt->stat.max_slots,
2362 xprt->stat.sending_u,
2363 xprt->stat.pending_u);
2367 * xs_udp_print_stats - display UDP socket-specifc stats
2368 * @xprt: rpc_xprt struct containing statistics
2372 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2374 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2376 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2379 xprt->stat.bind_count,
2382 xprt->stat.bad_xids,
2385 xprt->stat.max_slots,
2386 xprt->stat.sending_u,
2387 xprt->stat.pending_u);
2391 * xs_tcp_print_stats - display TCP socket-specifc stats
2392 * @xprt: rpc_xprt struct containing statistics
2396 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2398 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2401 if (xprt_connected(xprt))
2402 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2404 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2405 "%llu %llu %lu %llu %llu\n",
2407 xprt->stat.bind_count,
2408 xprt->stat.connect_count,
2409 xprt->stat.connect_time,
2413 xprt->stat.bad_xids,
2416 xprt->stat.max_slots,
2417 xprt->stat.sending_u,
2418 xprt->stat.pending_u);
2422 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2423 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2424 * to use the server side send routines.
2426 static void *bc_malloc(struct rpc_task *task, size_t size)
2429 struct rpc_buffer *buf;
2431 WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2432 if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2435 page = alloc_page(GFP_KERNEL);
2439 buf = page_address(page);
2440 buf->len = PAGE_SIZE;
2446 * Free the space allocated in the bc_alloc routine
2448 static void bc_free(void *buffer)
2450 struct rpc_buffer *buf;
2455 buf = container_of(buffer, struct rpc_buffer, data);
2456 free_page((unsigned long)buf);
2460 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2461 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2463 static int bc_sendto(struct rpc_rqst *req)
2466 struct xdr_buf *xbufp = &req->rq_snd_buf;
2467 struct rpc_xprt *xprt = req->rq_xprt;
2468 struct sock_xprt *transport =
2469 container_of(xprt, struct sock_xprt, xprt);
2470 struct socket *sock = transport->sock;
2471 unsigned long headoff;
2472 unsigned long tailoff;
2474 xs_encode_stream_record_marker(xbufp);
2476 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2477 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2478 len = svc_send_common(sock, xbufp,
2479 virt_to_page(xbufp->head[0].iov_base), headoff,
2480 xbufp->tail[0].iov_base, tailoff);
2482 if (len != xbufp->len) {
2483 printk(KERN_NOTICE "Error sending entire callback!\n");
2491 * The send routine. Borrows from svc_send
2493 static int bc_send_request(struct rpc_task *task)
2495 struct rpc_rqst *req = task->tk_rqstp;
2496 struct svc_xprt *xprt;
2499 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2501 * Get the server socket associated with this callback xprt
2503 xprt = req->rq_xprt->bc_xprt;
2506 * Grab the mutex to serialize data as the connection is shared
2507 * with the fore channel
2509 if (!mutex_trylock(&xprt->xpt_mutex)) {
2510 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2511 if (!mutex_trylock(&xprt->xpt_mutex))
2513 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2515 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2518 len = bc_sendto(req);
2519 mutex_unlock(&xprt->xpt_mutex);
2528 * The close routine. Since this is client initiated, we do nothing
2531 static void bc_close(struct rpc_xprt *xprt)
2536 * The xprt destroy routine. Again, because this connection is client
2537 * initiated, we do nothing
2540 static void bc_destroy(struct rpc_xprt *xprt)
2542 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2545 module_put(THIS_MODULE);
2548 static struct rpc_xprt_ops xs_local_ops = {
2549 .reserve_xprt = xprt_reserve_xprt,
2550 .release_xprt = xs_tcp_release_xprt,
2551 .alloc_slot = xprt_alloc_slot,
2552 .rpcbind = xs_local_rpcbind,
2553 .set_port = xs_local_set_port,
2554 .connect = xs_local_connect,
2555 .buf_alloc = rpc_malloc,
2556 .buf_free = rpc_free,
2557 .send_request = xs_local_send_request,
2558 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2560 .destroy = xs_destroy,
2561 .print_stats = xs_local_print_stats,
2564 static struct rpc_xprt_ops xs_udp_ops = {
2565 .set_buffer_size = xs_udp_set_buffer_size,
2566 .reserve_xprt = xprt_reserve_xprt_cong,
2567 .release_xprt = xprt_release_xprt_cong,
2568 .alloc_slot = xprt_alloc_slot,
2569 .rpcbind = rpcb_getport_async,
2570 .set_port = xs_set_port,
2571 .connect = xs_connect,
2572 .buf_alloc = rpc_malloc,
2573 .buf_free = rpc_free,
2574 .send_request = xs_udp_send_request,
2575 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
2576 .timer = xs_udp_timer,
2577 .release_request = xprt_release_rqst_cong,
2579 .destroy = xs_destroy,
2580 .print_stats = xs_udp_print_stats,
2583 static struct rpc_xprt_ops xs_tcp_ops = {
2584 .reserve_xprt = xprt_reserve_xprt,
2585 .release_xprt = xs_tcp_release_xprt,
2586 .alloc_slot = xprt_lock_and_alloc_slot,
2587 .rpcbind = rpcb_getport_async,
2588 .set_port = xs_set_port,
2589 .connect = xs_connect,
2590 .buf_alloc = rpc_malloc,
2591 .buf_free = rpc_free,
2592 .send_request = xs_tcp_send_request,
2593 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2594 .close = xs_tcp_close,
2595 .destroy = xs_destroy,
2596 .print_stats = xs_tcp_print_stats,
2600 * The rpc_xprt_ops for the server backchannel
2603 static struct rpc_xprt_ops bc_tcp_ops = {
2604 .reserve_xprt = xprt_reserve_xprt,
2605 .release_xprt = xprt_release_xprt,
2606 .alloc_slot = xprt_alloc_slot,
2607 .buf_alloc = bc_malloc,
2608 .buf_free = bc_free,
2609 .send_request = bc_send_request,
2610 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2612 .destroy = bc_destroy,
2613 .print_stats = xs_tcp_print_stats,
2616 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2618 static const struct sockaddr_in sin = {
2619 .sin_family = AF_INET,
2620 .sin_addr.s_addr = htonl(INADDR_ANY),
2622 static const struct sockaddr_in6 sin6 = {
2623 .sin6_family = AF_INET6,
2624 .sin6_addr = IN6ADDR_ANY_INIT,
2631 memcpy(sap, &sin, sizeof(sin));
2634 memcpy(sap, &sin6, sizeof(sin6));
2637 dprintk("RPC: %s: Bad address family\n", __func__);
2638 return -EAFNOSUPPORT;
2643 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2644 unsigned int slot_table_size,
2645 unsigned int max_slot_table_size)
2647 struct rpc_xprt *xprt;
2648 struct sock_xprt *new;
2650 if (args->addrlen > sizeof(xprt->addr)) {
2651 dprintk("RPC: xs_setup_xprt: address too large\n");
2652 return ERR_PTR(-EBADF);
2655 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2656 max_slot_table_size);
2658 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2660 return ERR_PTR(-ENOMEM);
2663 new = container_of(xprt, struct sock_xprt, xprt);
2664 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2665 xprt->addrlen = args->addrlen;
2667 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2670 err = xs_init_anyaddr(args->dstaddr->sa_family,
2671 (struct sockaddr *)&new->srcaddr);
2674 return ERR_PTR(err);
2681 static const struct rpc_timeout xs_local_default_timeout = {
2682 .to_initval = 10 * HZ,
2683 .to_maxval = 10 * HZ,
2688 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2689 * @args: rpc transport creation arguments
2691 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2693 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2695 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2696 struct sock_xprt *transport;
2697 struct rpc_xprt *xprt;
2698 struct rpc_xprt *ret;
2700 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2701 xprt_max_tcp_slot_table_entries);
2704 transport = container_of(xprt, struct sock_xprt, xprt);
2707 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2708 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2710 xprt->bind_timeout = XS_BIND_TO;
2711 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2712 xprt->idle_timeout = XS_IDLE_DISC_TO;
2714 xprt->ops = &xs_local_ops;
2715 xprt->timeout = &xs_local_default_timeout;
2717 INIT_DELAYED_WORK(&transport->connect_worker,
2718 xs_dummy_setup_socket);
2720 switch (sun->sun_family) {
2722 if (sun->sun_path[0] != '/') {
2723 dprintk("RPC: bad AF_LOCAL address: %s\n",
2725 ret = ERR_PTR(-EINVAL);
2728 xprt_set_bound(xprt);
2729 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2730 ret = ERR_PTR(xs_local_setup_socket(transport));
2735 ret = ERR_PTR(-EAFNOSUPPORT);
2739 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2740 xprt->address_strings[RPC_DISPLAY_ADDR]);
2742 if (try_module_get(THIS_MODULE))
2744 ret = ERR_PTR(-EINVAL);
2750 static const struct rpc_timeout xs_udp_default_timeout = {
2751 .to_initval = 5 * HZ,
2752 .to_maxval = 30 * HZ,
2753 .to_increment = 5 * HZ,
2758 * xs_setup_udp - Set up transport to use a UDP socket
2759 * @args: rpc transport creation arguments
2762 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2764 struct sockaddr *addr = args->dstaddr;
2765 struct rpc_xprt *xprt;
2766 struct sock_xprt *transport;
2767 struct rpc_xprt *ret;
2769 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2770 xprt_udp_slot_table_entries);
2773 transport = container_of(xprt, struct sock_xprt, xprt);
2775 xprt->prot = IPPROTO_UDP;
2777 /* XXX: header size can vary due to auth type, IPv6, etc. */
2778 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2780 xprt->bind_timeout = XS_BIND_TO;
2781 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2782 xprt->idle_timeout = XS_IDLE_DISC_TO;
2784 xprt->ops = &xs_udp_ops;
2786 xprt->timeout = &xs_udp_default_timeout;
2788 switch (addr->sa_family) {
2790 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2791 xprt_set_bound(xprt);
2793 INIT_DELAYED_WORK(&transport->connect_worker,
2794 xs_udp_setup_socket);
2795 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2798 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2799 xprt_set_bound(xprt);
2801 INIT_DELAYED_WORK(&transport->connect_worker,
2802 xs_udp_setup_socket);
2803 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2806 ret = ERR_PTR(-EAFNOSUPPORT);
2810 if (xprt_bound(xprt))
2811 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2812 xprt->address_strings[RPC_DISPLAY_ADDR],
2813 xprt->address_strings[RPC_DISPLAY_PORT],
2814 xprt->address_strings[RPC_DISPLAY_PROTO]);
2816 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2817 xprt->address_strings[RPC_DISPLAY_ADDR],
2818 xprt->address_strings[RPC_DISPLAY_PROTO]);
2820 if (try_module_get(THIS_MODULE))
2822 ret = ERR_PTR(-EINVAL);
2828 static const struct rpc_timeout xs_tcp_default_timeout = {
2829 .to_initval = 60 * HZ,
2830 .to_maxval = 60 * HZ,
2835 * xs_setup_tcp - Set up transport to use a TCP socket
2836 * @args: rpc transport creation arguments
2839 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2841 struct sockaddr *addr = args->dstaddr;
2842 struct rpc_xprt *xprt;
2843 struct sock_xprt *transport;
2844 struct rpc_xprt *ret;
2845 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2847 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2848 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2850 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2851 max_slot_table_size);
2854 transport = container_of(xprt, struct sock_xprt, xprt);
2856 xprt->prot = IPPROTO_TCP;
2857 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2858 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2860 xprt->bind_timeout = XS_BIND_TO;
2861 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2862 xprt->idle_timeout = XS_IDLE_DISC_TO;
2864 xprt->ops = &xs_tcp_ops;
2865 xprt->timeout = &xs_tcp_default_timeout;
2867 switch (addr->sa_family) {
2869 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2870 xprt_set_bound(xprt);
2872 INIT_DELAYED_WORK(&transport->connect_worker,
2873 xs_tcp_setup_socket);
2874 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2877 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2878 xprt_set_bound(xprt);
2880 INIT_DELAYED_WORK(&transport->connect_worker,
2881 xs_tcp_setup_socket);
2882 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2885 ret = ERR_PTR(-EAFNOSUPPORT);
2889 if (xprt_bound(xprt))
2890 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2891 xprt->address_strings[RPC_DISPLAY_ADDR],
2892 xprt->address_strings[RPC_DISPLAY_PORT],
2893 xprt->address_strings[RPC_DISPLAY_PROTO]);
2895 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2896 xprt->address_strings[RPC_DISPLAY_ADDR],
2897 xprt->address_strings[RPC_DISPLAY_PROTO]);
2899 if (try_module_get(THIS_MODULE))
2901 ret = ERR_PTR(-EINVAL);
2908 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2909 * @args: rpc transport creation arguments
2912 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2914 struct sockaddr *addr = args->dstaddr;
2915 struct rpc_xprt *xprt;
2916 struct sock_xprt *transport;
2917 struct svc_sock *bc_sock;
2918 struct rpc_xprt *ret;
2920 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2921 xprt_tcp_slot_table_entries);
2924 transport = container_of(xprt, struct sock_xprt, xprt);
2926 xprt->prot = IPPROTO_TCP;
2927 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2928 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2929 xprt->timeout = &xs_tcp_default_timeout;
2932 xprt_set_bound(xprt);
2933 xprt->bind_timeout = 0;
2934 xprt->reestablish_timeout = 0;
2935 xprt->idle_timeout = 0;
2937 xprt->ops = &bc_tcp_ops;
2939 switch (addr->sa_family) {
2941 xs_format_peer_addresses(xprt, "tcp",
2945 xs_format_peer_addresses(xprt, "tcp",
2946 RPCBIND_NETID_TCP6);
2949 ret = ERR_PTR(-EAFNOSUPPORT);
2953 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2954 xprt->address_strings[RPC_DISPLAY_ADDR],
2955 xprt->address_strings[RPC_DISPLAY_PORT],
2956 xprt->address_strings[RPC_DISPLAY_PROTO]);
2959 * Once we've associated a backchannel xprt with a connection,
2960 * we want to keep it around as long as the connection lasts,
2961 * in case we need to start using it for a backchannel again;
2962 * this reference won't be dropped until bc_xprt is destroyed.
2965 args->bc_xprt->xpt_bc_xprt = xprt;
2966 xprt->bc_xprt = args->bc_xprt;
2967 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2968 transport->sock = bc_sock->sk_sock;
2969 transport->inet = bc_sock->sk_sk;
2972 * Since we don't want connections for the backchannel, we set
2973 * the xprt status to connected
2975 xprt_set_connected(xprt);
2977 if (try_module_get(THIS_MODULE))
2980 args->bc_xprt->xpt_bc_xprt = NULL;
2982 ret = ERR_PTR(-EINVAL);
2988 static struct xprt_class xs_local_transport = {
2989 .list = LIST_HEAD_INIT(xs_local_transport.list),
2990 .name = "named UNIX socket",
2991 .owner = THIS_MODULE,
2992 .ident = XPRT_TRANSPORT_LOCAL,
2993 .setup = xs_setup_local,
2996 static struct xprt_class xs_udp_transport = {
2997 .list = LIST_HEAD_INIT(xs_udp_transport.list),
2999 .owner = THIS_MODULE,
3000 .ident = XPRT_TRANSPORT_UDP,
3001 .setup = xs_setup_udp,
3004 static struct xprt_class xs_tcp_transport = {
3005 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3007 .owner = THIS_MODULE,
3008 .ident = XPRT_TRANSPORT_TCP,
3009 .setup = xs_setup_tcp,
3012 static struct xprt_class xs_bc_tcp_transport = {
3013 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3014 .name = "tcp NFSv4.1 backchannel",
3015 .owner = THIS_MODULE,
3016 .ident = XPRT_TRANSPORT_BC_TCP,
3017 .setup = xs_setup_bc_tcp,
3021 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3024 int init_socket_xprt(void)
3027 if (!sunrpc_table_header)
3028 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3031 xprt_register_transport(&xs_local_transport);
3032 xprt_register_transport(&xs_udp_transport);
3033 xprt_register_transport(&xs_tcp_transport);
3034 xprt_register_transport(&xs_bc_tcp_transport);
3040 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3043 void cleanup_socket_xprt(void)
3046 if (sunrpc_table_header) {
3047 unregister_sysctl_table(sunrpc_table_header);
3048 sunrpc_table_header = NULL;
3052 xprt_unregister_transport(&xs_local_transport);
3053 xprt_unregister_transport(&xs_udp_transport);
3054 xprt_unregister_transport(&xs_tcp_transport);
3055 xprt_unregister_transport(&xs_bc_tcp_transport);
3058 static int param_set_uint_minmax(const char *val,
3059 const struct kernel_param *kp,
3060 unsigned int min, unsigned int max)
3067 ret = kstrtouint(val, 0, &num);
3068 if (ret == -EINVAL || num < min || num > max)
3070 *((unsigned int *)kp->arg) = num;
3074 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3076 return param_set_uint_minmax(val, kp,
3081 static struct kernel_param_ops param_ops_portnr = {
3082 .set = param_set_portnr,
3083 .get = param_get_uint,
3086 #define param_check_portnr(name, p) \
3087 __param_check(name, p, unsigned int);
3089 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3090 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3092 static int param_set_slot_table_size(const char *val,
3093 const struct kernel_param *kp)
3095 return param_set_uint_minmax(val, kp,
3097 RPC_MAX_SLOT_TABLE);
3100 static struct kernel_param_ops param_ops_slot_table_size = {
3101 .set = param_set_slot_table_size,
3102 .get = param_get_uint,
3105 #define param_check_slot_table_size(name, p) \
3106 __param_check(name, p, unsigned int);
3108 static int param_set_max_slot_table_size(const char *val,
3109 const struct kernel_param *kp)
3111 return param_set_uint_minmax(val, kp,
3113 RPC_MAX_SLOT_TABLE_LIMIT);
3116 static struct kernel_param_ops param_ops_max_slot_table_size = {
3117 .set = param_set_max_slot_table_size,
3118 .get = param_get_uint,
3121 #define param_check_max_slot_table_size(name, p) \
3122 __param_check(name, p, unsigned int);
3124 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3125 slot_table_size, 0644);
3126 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3127 max_slot_table_size, 0644);
3128 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3129 slot_table_size, 0644);