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);
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;
598 status = xs_nospace(task);
601 dprintk("RPC: sendmsg returned unrecognized error %d\n",
612 * xs_udp_send_request - write an RPC request to a UDP socket
613 * @task: address of RPC task that manages the state of an RPC request
616 * 0: The request has been sent
617 * EAGAIN: The socket was blocked, please call again later to
618 * complete the request
619 * ENOTCONN: Caller needs to invoke connect logic then call again
620 * other: Some other error occurred, the request was not sent
622 static int xs_udp_send_request(struct rpc_task *task)
624 struct rpc_rqst *req = task->tk_rqstp;
625 struct rpc_xprt *xprt = req->rq_xprt;
626 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
627 struct xdr_buf *xdr = &req->rq_snd_buf;
630 xs_pktdump("packet data:",
631 req->rq_svec->iov_base,
632 req->rq_svec->iov_len);
634 if (!xprt_bound(xprt))
636 status = xs_sendpages(transport->sock,
639 req->rq_bytes_sent, true);
641 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
642 xdr->len - req->rq_bytes_sent, status);
645 req->rq_xmit_bytes_sent += status;
646 if (status >= req->rq_slen)
648 /* Still some bytes left; set up for a retry later. */
655 /* Should we call xs_close() here? */
658 status = xs_nospace(task);
661 dprintk("RPC: sendmsg returned unrecognized error %d\n",
666 /* When the server has died, an ICMP port unreachable message
667 * prompts ECONNREFUSED. */
668 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
675 * xs_tcp_shutdown - gracefully shut down a TCP socket
678 * Initiates a graceful shutdown of the TCP socket by calling the
679 * equivalent of shutdown(SHUT_WR);
681 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
683 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
684 struct socket *sock = transport->sock;
687 kernel_sock_shutdown(sock, SHUT_WR);
688 trace_rpc_socket_shutdown(xprt, sock);
693 * xs_tcp_send_request - write an RPC request to a TCP socket
694 * @task: address of RPC task that manages the state of an RPC request
697 * 0: The request has been sent
698 * EAGAIN: The socket was blocked, please call again later to
699 * complete the request
700 * ENOTCONN: Caller needs to invoke connect logic then call again
701 * other: Some other error occurred, the request was not sent
703 * XXX: In the case of soft timeouts, should we eventually give up
704 * if sendmsg is not able to make progress?
706 static int xs_tcp_send_request(struct rpc_task *task)
708 struct rpc_rqst *req = task->tk_rqstp;
709 struct rpc_xprt *xprt = req->rq_xprt;
710 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
711 struct xdr_buf *xdr = &req->rq_snd_buf;
712 bool zerocopy = true;
715 xs_encode_stream_record_marker(&req->rq_snd_buf);
717 xs_pktdump("packet data:",
718 req->rq_svec->iov_base,
719 req->rq_svec->iov_len);
720 /* Don't use zero copy if this is a resend. If the RPC call
721 * completes while the socket holds a reference to the pages,
722 * then we may end up resending corrupted data.
724 if (task->tk_flags & RPC_TASK_SENT)
727 /* Continue transmitting the packet/record. We must be careful
728 * to cope with writespace callbacks arriving _after_ we have
729 * called sendmsg(). */
731 status = xs_sendpages(transport->sock,
732 NULL, 0, xdr, req->rq_bytes_sent,
735 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
736 xdr->len - req->rq_bytes_sent, status);
738 if (unlikely(status < 0))
741 /* If we've sent the entire packet, immediately
742 * reset the count of bytes sent. */
743 req->rq_bytes_sent += status;
744 req->rq_xmit_bytes_sent += status;
745 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
746 req->rq_bytes_sent = 0;
759 /* Should we call xs_close() here? */
762 status = xs_nospace(task);
765 dprintk("RPC: sendmsg returned unrecognized error %d\n",
768 xs_tcp_shutdown(xprt);
772 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
779 * xs_tcp_release_xprt - clean up after a tcp transmission
783 * This cleans up if an error causes us to abort the transmission of a request.
784 * In this case, the socket may need to be reset in order to avoid confusing
787 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
789 struct rpc_rqst *req;
791 if (task != xprt->snd_task)
795 req = task->tk_rqstp;
798 if (req->rq_bytes_sent == 0)
800 if (req->rq_bytes_sent == req->rq_snd_buf.len)
802 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
804 xprt_release_xprt(xprt, task);
807 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
809 transport->old_data_ready = sk->sk_data_ready;
810 transport->old_state_change = sk->sk_state_change;
811 transport->old_write_space = sk->sk_write_space;
812 transport->old_error_report = sk->sk_error_report;
815 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
817 sk->sk_data_ready = transport->old_data_ready;
818 sk->sk_state_change = transport->old_state_change;
819 sk->sk_write_space = transport->old_write_space;
820 sk->sk_error_report = transport->old_error_report;
824 * xs_error_report - callback to handle TCP socket state errors
827 * Note: we don't call sock_error() since there may be a rpc_task
828 * using the socket, and so we don't want to clear sk->sk_err.
830 static void xs_error_report(struct sock *sk)
832 struct rpc_xprt *xprt;
835 read_lock_bh(&sk->sk_callback_lock);
836 if (!(xprt = xprt_from_sock(sk)))
842 dprintk("RPC: xs_error_report client %p, error=%d...\n",
844 trace_rpc_socket_error(xprt, sk->sk_socket, err);
845 xprt_wake_pending_tasks(xprt, err);
847 read_unlock_bh(&sk->sk_callback_lock);
850 static void xs_reset_transport(struct sock_xprt *transport)
852 struct socket *sock = transport->sock;
853 struct sock *sk = transport->inet;
858 transport->srcport = 0;
860 write_lock_bh(&sk->sk_callback_lock);
861 transport->inet = NULL;
862 transport->sock = NULL;
864 sk->sk_user_data = NULL;
866 xs_restore_old_callbacks(transport, sk);
867 write_unlock_bh(&sk->sk_callback_lock);
871 trace_rpc_socket_close(&transport->xprt, sock);
876 * xs_close - close a socket
879 * This is used when all requests are complete; ie, no DRC state remains
880 * on the server we want to save.
882 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
883 * xs_reset_transport() zeroing the socket from underneath a writer.
885 static void xs_close(struct rpc_xprt *xprt)
887 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
889 dprintk("RPC: xs_close xprt %p\n", xprt);
891 cancel_delayed_work_sync(&transport->connect_worker);
893 xs_reset_transport(transport);
894 xprt->reestablish_timeout = 0;
896 smp_mb__before_clear_bit();
897 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
898 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
899 clear_bit(XPRT_CLOSING, &xprt->state);
900 smp_mb__after_clear_bit();
901 xprt_disconnect_done(xprt);
904 static void xs_tcp_close(struct rpc_xprt *xprt)
906 if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state))
909 xs_tcp_shutdown(xprt);
913 * xs_destroy - prepare to shutdown a transport
914 * @xprt: doomed transport
917 static void xs_destroy(struct rpc_xprt *xprt)
919 dprintk("RPC: xs_destroy xprt %p\n", xprt);
922 xs_free_peer_addresses(xprt);
924 module_put(THIS_MODULE);
927 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
929 struct xdr_skb_reader desc = {
931 .offset = sizeof(rpc_fraghdr),
932 .count = skb->len - sizeof(rpc_fraghdr),
935 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
943 * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets
944 * @sk: socket with data to read
945 * @len: how much data to read
947 * Currently this assumes we can read the whole reply in a single gulp.
949 static void xs_local_data_ready(struct sock *sk, int len)
951 struct rpc_task *task;
952 struct rpc_xprt *xprt;
953 struct rpc_rqst *rovr;
955 int err, repsize, copied;
959 read_lock_bh(&sk->sk_callback_lock);
960 dprintk("RPC: %s...\n", __func__);
961 xprt = xprt_from_sock(sk);
965 skb = skb_recv_datagram(sk, 0, 1, &err);
969 repsize = skb->len - sizeof(rpc_fraghdr);
971 dprintk("RPC: impossible RPC reply size %d\n", repsize);
975 /* Copy the XID from the skb... */
976 xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
980 /* Look up and lock the request corresponding to the given XID */
981 spin_lock(&xprt->transport_lock);
982 rovr = xprt_lookup_rqst(xprt, *xp);
985 task = rovr->rq_task;
987 copied = rovr->rq_private_buf.buflen;
988 if (copied > repsize)
991 if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
992 dprintk("RPC: sk_buff copy failed\n");
996 xprt_complete_rqst(task, copied);
999 spin_unlock(&xprt->transport_lock);
1001 skb_free_datagram(sk, skb);
1003 read_unlock_bh(&sk->sk_callback_lock);
1007 * xs_udp_data_ready - "data ready" callback for UDP sockets
1008 * @sk: socket with data to read
1009 * @len: how much data to read
1012 static void xs_udp_data_ready(struct sock *sk, int len)
1014 struct rpc_task *task;
1015 struct rpc_xprt *xprt;
1016 struct rpc_rqst *rovr;
1017 struct sk_buff *skb;
1018 int err, repsize, copied;
1022 read_lock_bh(&sk->sk_callback_lock);
1023 dprintk("RPC: xs_udp_data_ready...\n");
1024 if (!(xprt = xprt_from_sock(sk)))
1027 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
1030 repsize = skb->len - sizeof(struct udphdr);
1032 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
1036 /* Copy the XID from the skb... */
1037 xp = skb_header_pointer(skb, sizeof(struct udphdr),
1038 sizeof(_xid), &_xid);
1042 /* Look up and lock the request corresponding to the given XID */
1043 spin_lock(&xprt->transport_lock);
1044 rovr = xprt_lookup_rqst(xprt, *xp);
1047 task = rovr->rq_task;
1049 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1052 /* Suck it into the iovec, verify checksum if not done by hw. */
1053 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1054 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
1058 UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
1060 xprt_adjust_cwnd(xprt, task, copied);
1061 xprt_complete_rqst(task, copied);
1064 spin_unlock(&xprt->transport_lock);
1066 skb_free_datagram(sk, skb);
1068 read_unlock_bh(&sk->sk_callback_lock);
1072 * Helper function to force a TCP close if the server is sending
1073 * junk and/or it has put us in CLOSE_WAIT
1075 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1077 set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1078 xprt_force_disconnect(xprt);
1081 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1083 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1087 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1088 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1089 used = xdr_skb_read_bits(desc, p, len);
1090 transport->tcp_offset += used;
1094 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1095 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1096 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1098 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1099 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1101 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1102 transport->tcp_offset = 0;
1104 /* Sanity check of the record length */
1105 if (unlikely(transport->tcp_reclen < 8)) {
1106 dprintk("RPC: invalid TCP record fragment length\n");
1107 xs_tcp_force_close(xprt);
1110 dprintk("RPC: reading TCP record fragment of length %d\n",
1111 transport->tcp_reclen);
1114 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1116 if (transport->tcp_offset == transport->tcp_reclen) {
1117 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1118 transport->tcp_offset = 0;
1119 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1120 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1121 transport->tcp_flags |= TCP_RCV_COPY_XID;
1122 transport->tcp_copied = 0;
1127 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1132 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1133 dprintk("RPC: reading XID (%Zu bytes)\n", len);
1134 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1135 used = xdr_skb_read_bits(desc, p, len);
1136 transport->tcp_offset += used;
1139 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1140 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1141 transport->tcp_copied = 4;
1142 dprintk("RPC: reading %s XID %08x\n",
1143 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1145 ntohl(transport->tcp_xid));
1146 xs_tcp_check_fraghdr(transport);
1149 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1150 struct xdr_skb_reader *desc)
1157 * We want transport->tcp_offset to be 8 at the end of this routine
1158 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1159 * When this function is called for the first time,
1160 * transport->tcp_offset is 4 (after having already read the xid).
1162 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1163 len = sizeof(transport->tcp_calldir) - offset;
1164 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len);
1165 p = ((char *) &transport->tcp_calldir) + offset;
1166 used = xdr_skb_read_bits(desc, p, len);
1167 transport->tcp_offset += used;
1170 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1172 * We don't yet have the XDR buffer, so we will write the calldir
1173 * out after we get the buffer from the 'struct rpc_rqst'
1175 switch (ntohl(transport->tcp_calldir)) {
1177 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1178 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1179 transport->tcp_flags |= TCP_RPC_REPLY;
1182 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1183 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1184 transport->tcp_flags &= ~TCP_RPC_REPLY;
1187 dprintk("RPC: invalid request message type\n");
1188 xs_tcp_force_close(&transport->xprt);
1190 xs_tcp_check_fraghdr(transport);
1193 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1194 struct xdr_skb_reader *desc,
1195 struct rpc_rqst *req)
1197 struct sock_xprt *transport =
1198 container_of(xprt, struct sock_xprt, xprt);
1199 struct xdr_buf *rcvbuf;
1203 rcvbuf = &req->rq_private_buf;
1205 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1207 * Save the RPC direction in the XDR buffer
1209 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1210 &transport->tcp_calldir,
1211 sizeof(transport->tcp_calldir));
1212 transport->tcp_copied += sizeof(transport->tcp_calldir);
1213 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1217 if (len > transport->tcp_reclen - transport->tcp_offset) {
1218 struct xdr_skb_reader my_desc;
1220 len = transport->tcp_reclen - transport->tcp_offset;
1221 memcpy(&my_desc, desc, sizeof(my_desc));
1222 my_desc.count = len;
1223 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1224 &my_desc, xdr_skb_read_bits);
1228 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1229 desc, xdr_skb_read_bits);
1232 transport->tcp_copied += r;
1233 transport->tcp_offset += r;
1236 /* Error when copying to the receive buffer,
1237 * usually because we weren't able to allocate
1238 * additional buffer pages. All we can do now
1239 * is turn off TCP_RCV_COPY_DATA, so the request
1240 * will not receive any additional updates,
1242 * Any remaining data from this record will
1245 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1246 dprintk("RPC: XID %08x truncated request\n",
1247 ntohl(transport->tcp_xid));
1248 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1249 "tcp_offset = %u, tcp_reclen = %u\n",
1250 xprt, transport->tcp_copied,
1251 transport->tcp_offset, transport->tcp_reclen);
1255 dprintk("RPC: XID %08x read %Zd bytes\n",
1256 ntohl(transport->tcp_xid), r);
1257 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1258 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1259 transport->tcp_offset, transport->tcp_reclen);
1261 if (transport->tcp_copied == req->rq_private_buf.buflen)
1262 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1263 else if (transport->tcp_offset == transport->tcp_reclen) {
1264 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1265 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1270 * Finds the request corresponding to the RPC xid and invokes the common
1271 * tcp read code to read the data.
1273 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1274 struct xdr_skb_reader *desc)
1276 struct sock_xprt *transport =
1277 container_of(xprt, struct sock_xprt, xprt);
1278 struct rpc_rqst *req;
1280 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1282 /* Find and lock the request corresponding to this xid */
1283 spin_lock(&xprt->transport_lock);
1284 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1286 dprintk("RPC: XID %08x request not found!\n",
1287 ntohl(transport->tcp_xid));
1288 spin_unlock(&xprt->transport_lock);
1292 xs_tcp_read_common(xprt, desc, req);
1294 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1295 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1297 spin_unlock(&xprt->transport_lock);
1301 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1303 * Obtains an rpc_rqst previously allocated and invokes the common
1304 * tcp read code to read the data. The result is placed in the callback
1306 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1307 * connection and return -1.
1309 static int xs_tcp_read_callback(struct rpc_xprt *xprt,
1310 struct xdr_skb_reader *desc)
1312 struct sock_xprt *transport =
1313 container_of(xprt, struct sock_xprt, xprt);
1314 struct rpc_rqst *req;
1316 /* Look up and lock the request corresponding to the given XID */
1317 spin_lock(&xprt->transport_lock);
1318 req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
1320 spin_unlock(&xprt->transport_lock);
1321 printk(KERN_WARNING "Callback slot table overflowed\n");
1322 xprt_force_disconnect(xprt);
1326 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1327 xs_tcp_read_common(xprt, desc, req);
1329 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1330 xprt_complete_bc_request(req, transport->tcp_copied);
1331 spin_unlock(&xprt->transport_lock);
1336 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1337 struct xdr_skb_reader *desc)
1339 struct sock_xprt *transport =
1340 container_of(xprt, struct sock_xprt, xprt);
1342 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1343 xs_tcp_read_reply(xprt, desc) :
1344 xs_tcp_read_callback(xprt, desc);
1347 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1348 struct xdr_skb_reader *desc)
1350 return xs_tcp_read_reply(xprt, desc);
1352 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1355 * Read data off the transport. This can be either an RPC_CALL or an
1356 * RPC_REPLY. Relay the processing to helper functions.
1358 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1359 struct xdr_skb_reader *desc)
1361 struct sock_xprt *transport =
1362 container_of(xprt, struct sock_xprt, xprt);
1364 if (_xs_tcp_read_data(xprt, desc) == 0)
1365 xs_tcp_check_fraghdr(transport);
1368 * The transport_lock protects the request handling.
1369 * There's no need to hold it to update the tcp_flags.
1371 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1375 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1379 len = transport->tcp_reclen - transport->tcp_offset;
1380 if (len > desc->count)
1383 desc->offset += len;
1384 transport->tcp_offset += len;
1385 dprintk("RPC: discarded %Zu bytes\n", len);
1386 xs_tcp_check_fraghdr(transport);
1389 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1391 struct rpc_xprt *xprt = rd_desc->arg.data;
1392 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1393 struct xdr_skb_reader desc = {
1399 dprintk("RPC: xs_tcp_data_recv started\n");
1401 /* Read in a new fragment marker if necessary */
1402 /* Can we ever really expect to get completely empty fragments? */
1403 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1404 xs_tcp_read_fraghdr(xprt, &desc);
1407 /* Read in the xid if necessary */
1408 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1409 xs_tcp_read_xid(transport, &desc);
1412 /* Read in the call/reply flag */
1413 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1414 xs_tcp_read_calldir(transport, &desc);
1417 /* Read in the request data */
1418 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1419 xs_tcp_read_data(xprt, &desc);
1422 /* Skip over any trailing bytes on short reads */
1423 xs_tcp_read_discard(transport, &desc);
1424 } while (desc.count);
1425 dprintk("RPC: xs_tcp_data_recv done\n");
1426 return len - desc.count;
1430 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1431 * @sk: socket with data to read
1432 * @bytes: how much data to read
1435 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1437 struct rpc_xprt *xprt;
1438 read_descriptor_t rd_desc;
1441 dprintk("RPC: xs_tcp_data_ready...\n");
1443 read_lock_bh(&sk->sk_callback_lock);
1444 if (!(xprt = xprt_from_sock(sk)))
1446 /* Any data means we had a useful conversation, so
1447 * the we don't need to delay the next reconnect
1449 if (xprt->reestablish_timeout)
1450 xprt->reestablish_timeout = 0;
1452 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1453 rd_desc.arg.data = xprt;
1455 rd_desc.count = 65536;
1456 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1459 read_unlock_bh(&sk->sk_callback_lock);
1463 * Do the equivalent of linger/linger2 handling for dealing with
1464 * broken servers that don't close the socket in a timely
1467 static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1468 unsigned long timeout)
1470 struct sock_xprt *transport;
1472 if (xprt_test_and_set_connecting(xprt))
1474 set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1475 transport = container_of(xprt, struct sock_xprt, xprt);
1476 queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1480 static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1482 struct sock_xprt *transport;
1484 transport = container_of(xprt, struct sock_xprt, xprt);
1486 if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1487 !cancel_delayed_work(&transport->connect_worker))
1489 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1490 xprt_clear_connecting(xprt);
1493 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1495 smp_mb__before_clear_bit();
1496 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1497 clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1498 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1499 clear_bit(XPRT_CLOSING, &xprt->state);
1500 smp_mb__after_clear_bit();
1503 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1505 xs_sock_reset_connection_flags(xprt);
1506 /* Mark transport as closed and wake up all pending tasks */
1507 xprt_disconnect_done(xprt);
1511 * xs_tcp_state_change - callback to handle TCP socket state changes
1512 * @sk: socket whose state has changed
1515 static void xs_tcp_state_change(struct sock *sk)
1517 struct rpc_xprt *xprt;
1519 read_lock_bh(&sk->sk_callback_lock);
1520 if (!(xprt = xprt_from_sock(sk)))
1522 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1523 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1524 sk->sk_state, xprt_connected(xprt),
1525 sock_flag(sk, SOCK_DEAD),
1526 sock_flag(sk, SOCK_ZAPPED),
1529 trace_rpc_socket_state_change(xprt, sk->sk_socket);
1530 switch (sk->sk_state) {
1531 case TCP_ESTABLISHED:
1532 spin_lock(&xprt->transport_lock);
1533 if (!xprt_test_and_set_connected(xprt)) {
1534 struct sock_xprt *transport = container_of(xprt,
1535 struct sock_xprt, xprt);
1537 /* Reset TCP record info */
1538 transport->tcp_offset = 0;
1539 transport->tcp_reclen = 0;
1540 transport->tcp_copied = 0;
1541 transport->tcp_flags =
1542 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1543 xprt->connect_cookie++;
1545 xprt_wake_pending_tasks(xprt, -EAGAIN);
1547 spin_unlock(&xprt->transport_lock);
1550 /* The client initiated a shutdown of the socket */
1551 xprt->connect_cookie++;
1552 xprt->reestablish_timeout = 0;
1553 set_bit(XPRT_CLOSING, &xprt->state);
1554 smp_mb__before_clear_bit();
1555 clear_bit(XPRT_CONNECTED, &xprt->state);
1556 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1557 smp_mb__after_clear_bit();
1558 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1560 case TCP_CLOSE_WAIT:
1561 /* The server initiated a shutdown of the socket */
1562 xprt->connect_cookie++;
1563 clear_bit(XPRT_CONNECTED, &xprt->state);
1564 xs_tcp_force_close(xprt);
1567 * If the server closed down the connection, make sure that
1568 * we back off before reconnecting
1570 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1571 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1574 set_bit(XPRT_CLOSING, &xprt->state);
1575 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1576 smp_mb__before_clear_bit();
1577 clear_bit(XPRT_CONNECTED, &xprt->state);
1578 smp_mb__after_clear_bit();
1581 xs_tcp_cancel_linger_timeout(xprt);
1582 xs_sock_mark_closed(xprt);
1585 read_unlock_bh(&sk->sk_callback_lock);
1588 static void xs_write_space(struct sock *sk)
1590 struct socket *sock;
1591 struct rpc_xprt *xprt;
1593 if (unlikely(!(sock = sk->sk_socket)))
1595 clear_bit(SOCK_NOSPACE, &sock->flags);
1597 if (unlikely(!(xprt = xprt_from_sock(sk))))
1599 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1602 xprt_write_space(xprt);
1606 * xs_udp_write_space - callback invoked when socket buffer space
1608 * @sk: socket whose state has changed
1610 * Called when more output buffer space is available for this socket.
1611 * We try not to wake our writers until they can make "significant"
1612 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1613 * with a bunch of small requests.
1615 static void xs_udp_write_space(struct sock *sk)
1617 read_lock_bh(&sk->sk_callback_lock);
1619 /* from net/core/sock.c:sock_def_write_space */
1620 if (sock_writeable(sk))
1623 read_unlock_bh(&sk->sk_callback_lock);
1627 * xs_tcp_write_space - callback invoked when socket buffer space
1629 * @sk: socket whose state has changed
1631 * Called when more output buffer space is available for this socket.
1632 * We try not to wake our writers until they can make "significant"
1633 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1634 * with a bunch of small requests.
1636 static void xs_tcp_write_space(struct sock *sk)
1638 read_lock_bh(&sk->sk_callback_lock);
1640 /* from net/core/stream.c:sk_stream_write_space */
1641 if (sk_stream_is_writeable(sk))
1644 read_unlock_bh(&sk->sk_callback_lock);
1647 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1649 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1650 struct sock *sk = transport->inet;
1652 if (transport->rcvsize) {
1653 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1654 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1656 if (transport->sndsize) {
1657 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1658 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1659 sk->sk_write_space(sk);
1664 * xs_udp_set_buffer_size - set send and receive limits
1665 * @xprt: generic transport
1666 * @sndsize: requested size of send buffer, in bytes
1667 * @rcvsize: requested size of receive buffer, in bytes
1669 * Set socket send and receive buffer size limits.
1671 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1673 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1675 transport->sndsize = 0;
1677 transport->sndsize = sndsize + 1024;
1678 transport->rcvsize = 0;
1680 transport->rcvsize = rcvsize + 1024;
1682 xs_udp_do_set_buffer_size(xprt);
1686 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1687 * @task: task that timed out
1689 * Adjust the congestion window after a retransmit timeout has occurred.
1691 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1693 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1696 static unsigned short xs_get_random_port(void)
1698 unsigned short range = xprt_max_resvport - xprt_min_resvport;
1699 unsigned short rand = (unsigned short) prandom_u32() % range;
1700 return rand + xprt_min_resvport;
1704 * xs_set_port - reset the port number in the remote endpoint address
1705 * @xprt: generic transport
1706 * @port: new port number
1709 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1711 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1713 rpc_set_port(xs_addr(xprt), port);
1714 xs_update_peer_port(xprt);
1717 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1719 unsigned short port = transport->srcport;
1721 if (port == 0 && transport->xprt.resvport)
1722 port = xs_get_random_port();
1726 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1728 if (transport->srcport != 0)
1729 transport->srcport = 0;
1730 if (!transport->xprt.resvport)
1732 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1733 return xprt_max_resvport;
1736 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1738 struct sockaddr_storage myaddr;
1740 unsigned short port = xs_get_srcport(transport);
1741 unsigned short last;
1743 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1745 rpc_set_port((struct sockaddr *)&myaddr, port);
1746 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1747 transport->xprt.addrlen);
1751 transport->srcport = port;
1755 port = xs_next_srcport(transport, port);
1758 } while (err == -EADDRINUSE && nloop != 2);
1760 if (myaddr.ss_family == AF_INET)
1761 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1762 &((struct sockaddr_in *)&myaddr)->sin_addr,
1763 port, err ? "failed" : "ok", err);
1765 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1766 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1767 port, err ? "failed" : "ok", err);
1772 * We don't support autobind on AF_LOCAL sockets
1774 static void xs_local_rpcbind(struct rpc_task *task)
1777 xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt));
1781 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1785 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1786 static struct lock_class_key xs_key[2];
1787 static struct lock_class_key xs_slock_key[2];
1789 static inline void xs_reclassify_socketu(struct socket *sock)
1791 struct sock *sk = sock->sk;
1793 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1794 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1797 static inline void xs_reclassify_socket4(struct socket *sock)
1799 struct sock *sk = sock->sk;
1801 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1802 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1805 static inline void xs_reclassify_socket6(struct socket *sock)
1807 struct sock *sk = sock->sk;
1809 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1810 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1813 static inline void xs_reclassify_socket(int family, struct socket *sock)
1815 WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1816 if (sock_owned_by_user(sock->sk))
1821 xs_reclassify_socketu(sock);
1824 xs_reclassify_socket4(sock);
1827 xs_reclassify_socket6(sock);
1832 static inline void xs_reclassify_socketu(struct socket *sock)
1836 static inline void xs_reclassify_socket4(struct socket *sock)
1840 static inline void xs_reclassify_socket6(struct socket *sock)
1844 static inline void xs_reclassify_socket(int family, struct socket *sock)
1849 static void xs_dummy_setup_socket(struct work_struct *work)
1853 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1854 struct sock_xprt *transport, int family, int type, int protocol)
1856 struct socket *sock;
1859 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1861 dprintk("RPC: can't create %d transport socket (%d).\n",
1865 xs_reclassify_socket(family, sock);
1867 err = xs_bind(transport, sock);
1875 return ERR_PTR(err);
1878 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1879 struct socket *sock)
1881 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1884 if (!transport->inet) {
1885 struct sock *sk = sock->sk;
1887 write_lock_bh(&sk->sk_callback_lock);
1889 xs_save_old_callbacks(transport, sk);
1891 sk->sk_user_data = xprt;
1892 sk->sk_data_ready = xs_local_data_ready;
1893 sk->sk_write_space = xs_udp_write_space;
1894 sk->sk_error_report = xs_error_report;
1895 sk->sk_allocation = GFP_ATOMIC;
1897 xprt_clear_connected(xprt);
1899 /* Reset to new socket */
1900 transport->sock = sock;
1901 transport->inet = sk;
1903 write_unlock_bh(&sk->sk_callback_lock);
1906 /* Tell the socket layer to start connecting... */
1907 xprt->stat.connect_count++;
1908 xprt->stat.connect_start = jiffies;
1909 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1913 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1914 * @xprt: RPC transport to connect
1915 * @transport: socket transport to connect
1916 * @create_sock: function to create a socket of the correct type
1918 static int xs_local_setup_socket(struct sock_xprt *transport)
1920 struct rpc_xprt *xprt = &transport->xprt;
1921 struct socket *sock;
1924 current->flags |= PF_FSTRANS;
1926 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1927 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1928 SOCK_STREAM, 0, &sock, 1);
1930 dprintk("RPC: can't create AF_LOCAL "
1931 "transport socket (%d).\n", -status);
1934 xs_reclassify_socketu(sock);
1936 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1937 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1939 status = xs_local_finish_connecting(xprt, sock);
1940 trace_rpc_socket_connect(xprt, sock, status);
1943 dprintk("RPC: xprt %p connected to %s\n",
1944 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1945 xprt_set_connected(xprt);
1948 dprintk("RPC: xprt %p: socket %s does not exist\n",
1949 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1952 dprintk("RPC: xprt %p: connection refused for %s\n",
1953 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1956 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1958 xprt->address_strings[RPC_DISPLAY_ADDR]);
1962 xprt_clear_connecting(xprt);
1963 xprt_wake_pending_tasks(xprt, status);
1964 current->flags &= ~PF_FSTRANS;
1968 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1970 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1973 if (RPC_IS_ASYNC(task)) {
1975 * We want the AF_LOCAL connect to be resolved in the
1976 * filesystem namespace of the process making the rpc
1977 * call. Thus we connect synchronously.
1979 * If we want to support asynchronous AF_LOCAL calls,
1980 * we'll need to figure out how to pass a namespace to
1983 rpc_exit(task, -ENOTCONN);
1986 ret = xs_local_setup_socket(transport);
1987 if (ret && !RPC_IS_SOFTCONN(task))
1988 msleep_interruptible(15000);
1991 #ifdef CONFIG_SUNRPC_SWAP
1992 static void xs_set_memalloc(struct rpc_xprt *xprt)
1994 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1998 sk_set_memalloc(transport->inet);
2002 * xs_swapper - Tag this transport as being used for swap.
2003 * @xprt: transport to tag
2004 * @enable: enable/disable
2007 int xs_swapper(struct rpc_xprt *xprt, int enable)
2009 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2015 xs_set_memalloc(xprt);
2016 } else if (xprt->swapper) {
2018 sk_clear_memalloc(transport->inet);
2023 EXPORT_SYMBOL_GPL(xs_swapper);
2025 static void xs_set_memalloc(struct rpc_xprt *xprt)
2030 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2032 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2034 if (!transport->inet) {
2035 struct sock *sk = sock->sk;
2037 write_lock_bh(&sk->sk_callback_lock);
2039 xs_save_old_callbacks(transport, sk);
2041 sk->sk_user_data = xprt;
2042 sk->sk_data_ready = xs_udp_data_ready;
2043 sk->sk_write_space = xs_udp_write_space;
2044 sk->sk_no_check = UDP_CSUM_NORCV;
2045 sk->sk_allocation = GFP_ATOMIC;
2047 xprt_set_connected(xprt);
2049 /* Reset to new socket */
2050 transport->sock = sock;
2051 transport->inet = sk;
2053 xs_set_memalloc(xprt);
2055 write_unlock_bh(&sk->sk_callback_lock);
2057 xs_udp_do_set_buffer_size(xprt);
2060 static void xs_udp_setup_socket(struct work_struct *work)
2062 struct sock_xprt *transport =
2063 container_of(work, struct sock_xprt, connect_worker.work);
2064 struct rpc_xprt *xprt = &transport->xprt;
2065 struct socket *sock = transport->sock;
2068 current->flags |= PF_FSTRANS;
2070 /* Start by resetting any existing state */
2071 xs_reset_transport(transport);
2072 sock = xs_create_sock(xprt, transport,
2073 xs_addr(xprt)->sa_family, SOCK_DGRAM, IPPROTO_UDP);
2077 dprintk("RPC: worker connecting xprt %p via %s to "
2078 "%s (port %s)\n", xprt,
2079 xprt->address_strings[RPC_DISPLAY_PROTO],
2080 xprt->address_strings[RPC_DISPLAY_ADDR],
2081 xprt->address_strings[RPC_DISPLAY_PORT]);
2083 xs_udp_finish_connecting(xprt, sock);
2084 trace_rpc_socket_connect(xprt, sock, 0);
2087 xprt_clear_connecting(xprt);
2088 xprt_wake_pending_tasks(xprt, status);
2089 current->flags &= ~PF_FSTRANS;
2093 * We need to preserve the port number so the reply cache on the server can
2094 * find our cached RPC replies when we get around to reconnecting.
2096 static void xs_abort_connection(struct sock_xprt *transport)
2099 struct sockaddr any;
2101 dprintk("RPC: disconnecting xprt %p to reuse port\n", transport);
2104 * Disconnect the transport socket by doing a connect operation
2105 * with AF_UNSPEC. This should return immediately...
2107 memset(&any, 0, sizeof(any));
2108 any.sa_family = AF_UNSPEC;
2109 result = kernel_connect(transport->sock, &any, sizeof(any), 0);
2110 trace_rpc_socket_reset_connection(&transport->xprt,
2111 transport->sock, result);
2113 xs_sock_reset_connection_flags(&transport->xprt);
2114 dprintk("RPC: AF_UNSPEC connect return code %d\n", result);
2117 static void xs_tcp_reuse_connection(struct sock_xprt *transport)
2119 unsigned int state = transport->inet->sk_state;
2121 if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED) {
2122 /* we don't need to abort the connection if the socket
2123 * hasn't undergone a shutdown
2125 if (transport->inet->sk_shutdown == 0)
2127 dprintk("RPC: %s: TCP_CLOSEd and sk_shutdown set to %d\n",
2128 __func__, transport->inet->sk_shutdown);
2130 if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT)) {
2131 /* we don't need to abort the connection if the socket
2132 * hasn't undergone a shutdown
2134 if (transport->inet->sk_shutdown == 0)
2136 dprintk("RPC: %s: ESTABLISHED/SYN_SENT "
2137 "sk_shutdown set to %d\n",
2138 __func__, transport->inet->sk_shutdown);
2140 xs_abort_connection(transport);
2143 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2145 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2146 int ret = -ENOTCONN;
2148 if (!transport->inet) {
2149 struct sock *sk = sock->sk;
2150 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2151 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2152 unsigned int opt_on = 1;
2154 /* TCP Keepalive options */
2155 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2156 (char *)&opt_on, sizeof(opt_on));
2157 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2158 (char *)&keepidle, sizeof(keepidle));
2159 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2160 (char *)&keepidle, sizeof(keepidle));
2161 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2162 (char *)&keepcnt, sizeof(keepcnt));
2164 write_lock_bh(&sk->sk_callback_lock);
2166 xs_save_old_callbacks(transport, sk);
2168 sk->sk_user_data = xprt;
2169 sk->sk_data_ready = xs_tcp_data_ready;
2170 sk->sk_state_change = xs_tcp_state_change;
2171 sk->sk_write_space = xs_tcp_write_space;
2172 sk->sk_error_report = xs_error_report;
2173 sk->sk_allocation = GFP_ATOMIC;
2175 /* socket options */
2176 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
2177 sock_reset_flag(sk, SOCK_LINGER);
2178 tcp_sk(sk)->linger2 = 0;
2179 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2181 xprt_clear_connected(xprt);
2183 /* Reset to new socket */
2184 transport->sock = sock;
2185 transport->inet = sk;
2187 write_unlock_bh(&sk->sk_callback_lock);
2190 if (!xprt_bound(xprt))
2193 xs_set_memalloc(xprt);
2195 /* Tell the socket layer to start connecting... */
2196 xprt->stat.connect_count++;
2197 xprt->stat.connect_start = jiffies;
2198 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2203 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2204 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2211 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2212 * @xprt: RPC transport to connect
2213 * @transport: socket transport to connect
2214 * @create_sock: function to create a socket of the correct type
2216 * Invoked by a work queue tasklet.
2218 static void xs_tcp_setup_socket(struct work_struct *work)
2220 struct sock_xprt *transport =
2221 container_of(work, struct sock_xprt, connect_worker.work);
2222 struct socket *sock = transport->sock;
2223 struct rpc_xprt *xprt = &transport->xprt;
2226 current->flags |= PF_FSTRANS;
2229 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
2230 sock = xs_create_sock(xprt, transport,
2231 xs_addr(xprt)->sa_family, SOCK_STREAM, IPPROTO_TCP);
2233 status = PTR_ERR(sock);
2239 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
2241 /* "close" the socket, preserving the local port */
2242 xs_tcp_reuse_connection(transport);
2248 dprintk("RPC: worker connecting xprt %p via %s to "
2249 "%s (port %s)\n", xprt,
2250 xprt->address_strings[RPC_DISPLAY_PROTO],
2251 xprt->address_strings[RPC_DISPLAY_ADDR],
2252 xprt->address_strings[RPC_DISPLAY_PORT]);
2254 status = xs_tcp_finish_connecting(xprt, sock);
2255 trace_rpc_socket_connect(xprt, sock, status);
2256 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2257 xprt, -status, xprt_connected(xprt),
2258 sock->sk->sk_state);
2261 printk("%s: connect returned unhandled error %d\n",
2263 case -EADDRNOTAVAIL:
2264 /* We're probably in TIME_WAIT. Get rid of existing socket,
2267 xs_tcp_force_close(xprt);
2272 xprt_clear_connecting(xprt);
2273 current->flags &= ~PF_FSTRANS;
2276 /* Happens, for instance, if the user specified a link
2277 * local IPv6 address without a scope-id.
2282 /* retry with existing socket, after a delay */
2288 xprt_clear_connecting(xprt);
2289 xprt_wake_pending_tasks(xprt, status);
2290 current->flags &= ~PF_FSTRANS;
2294 * xs_connect - connect a socket to a remote endpoint
2295 * @xprt: pointer to transport structure
2296 * @task: address of RPC task that manages state of connect request
2298 * TCP: If the remote end dropped the connection, delay reconnecting.
2300 * UDP socket connects are synchronous, but we use a work queue anyway
2301 * to guarantee that even unprivileged user processes can set up a
2302 * socket on a privileged port.
2304 * If a UDP socket connect fails, the delay behavior here prevents
2305 * retry floods (hard mounts).
2307 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2309 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2311 if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
2312 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2314 xprt, xprt->reestablish_timeout / HZ);
2315 queue_delayed_work(rpciod_workqueue,
2316 &transport->connect_worker,
2317 xprt->reestablish_timeout);
2318 xprt->reestablish_timeout <<= 1;
2319 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2320 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2321 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2322 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2324 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2325 queue_delayed_work(rpciod_workqueue,
2326 &transport->connect_worker, 0);
2331 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2332 * @xprt: rpc_xprt struct containing statistics
2336 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2340 if (xprt_connected(xprt))
2341 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2343 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2344 "%llu %llu %lu %llu %llu\n",
2345 xprt->stat.bind_count,
2346 xprt->stat.connect_count,
2347 xprt->stat.connect_time,
2351 xprt->stat.bad_xids,
2354 xprt->stat.max_slots,
2355 xprt->stat.sending_u,
2356 xprt->stat.pending_u);
2360 * xs_udp_print_stats - display UDP socket-specifc stats
2361 * @xprt: rpc_xprt struct containing statistics
2365 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2367 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2369 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2372 xprt->stat.bind_count,
2375 xprt->stat.bad_xids,
2378 xprt->stat.max_slots,
2379 xprt->stat.sending_u,
2380 xprt->stat.pending_u);
2384 * xs_tcp_print_stats - display TCP socket-specifc stats
2385 * @xprt: rpc_xprt struct containing statistics
2389 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2391 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2394 if (xprt_connected(xprt))
2395 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2397 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2398 "%llu %llu %lu %llu %llu\n",
2400 xprt->stat.bind_count,
2401 xprt->stat.connect_count,
2402 xprt->stat.connect_time,
2406 xprt->stat.bad_xids,
2409 xprt->stat.max_slots,
2410 xprt->stat.sending_u,
2411 xprt->stat.pending_u);
2415 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2416 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2417 * to use the server side send routines.
2419 static void *bc_malloc(struct rpc_task *task, size_t size)
2422 struct rpc_buffer *buf;
2424 WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2425 if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2428 page = alloc_page(GFP_KERNEL);
2432 buf = page_address(page);
2433 buf->len = PAGE_SIZE;
2439 * Free the space allocated in the bc_alloc routine
2441 static void bc_free(void *buffer)
2443 struct rpc_buffer *buf;
2448 buf = container_of(buffer, struct rpc_buffer, data);
2449 free_page((unsigned long)buf);
2453 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2454 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2456 static int bc_sendto(struct rpc_rqst *req)
2459 struct xdr_buf *xbufp = &req->rq_snd_buf;
2460 struct rpc_xprt *xprt = req->rq_xprt;
2461 struct sock_xprt *transport =
2462 container_of(xprt, struct sock_xprt, xprt);
2463 struct socket *sock = transport->sock;
2464 unsigned long headoff;
2465 unsigned long tailoff;
2467 xs_encode_stream_record_marker(xbufp);
2469 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2470 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2471 len = svc_send_common(sock, xbufp,
2472 virt_to_page(xbufp->head[0].iov_base), headoff,
2473 xbufp->tail[0].iov_base, tailoff);
2475 if (len != xbufp->len) {
2476 printk(KERN_NOTICE "Error sending entire callback!\n");
2484 * The send routine. Borrows from svc_send
2486 static int bc_send_request(struct rpc_task *task)
2488 struct rpc_rqst *req = task->tk_rqstp;
2489 struct svc_xprt *xprt;
2492 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2494 * Get the server socket associated with this callback xprt
2496 xprt = req->rq_xprt->bc_xprt;
2499 * Grab the mutex to serialize data as the connection is shared
2500 * with the fore channel
2502 if (!mutex_trylock(&xprt->xpt_mutex)) {
2503 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2504 if (!mutex_trylock(&xprt->xpt_mutex))
2506 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2508 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2511 len = bc_sendto(req);
2512 mutex_unlock(&xprt->xpt_mutex);
2521 * The close routine. Since this is client initiated, we do nothing
2524 static void bc_close(struct rpc_xprt *xprt)
2529 * The xprt destroy routine. Again, because this connection is client
2530 * initiated, we do nothing
2533 static void bc_destroy(struct rpc_xprt *xprt)
2537 static struct rpc_xprt_ops xs_local_ops = {
2538 .reserve_xprt = xprt_reserve_xprt,
2539 .release_xprt = xs_tcp_release_xprt,
2540 .alloc_slot = xprt_alloc_slot,
2541 .rpcbind = xs_local_rpcbind,
2542 .set_port = xs_local_set_port,
2543 .connect = xs_local_connect,
2544 .buf_alloc = rpc_malloc,
2545 .buf_free = rpc_free,
2546 .send_request = xs_local_send_request,
2547 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2549 .destroy = xs_destroy,
2550 .print_stats = xs_local_print_stats,
2553 static struct rpc_xprt_ops xs_udp_ops = {
2554 .set_buffer_size = xs_udp_set_buffer_size,
2555 .reserve_xprt = xprt_reserve_xprt_cong,
2556 .release_xprt = xprt_release_xprt_cong,
2557 .alloc_slot = xprt_alloc_slot,
2558 .rpcbind = rpcb_getport_async,
2559 .set_port = xs_set_port,
2560 .connect = xs_connect,
2561 .buf_alloc = rpc_malloc,
2562 .buf_free = rpc_free,
2563 .send_request = xs_udp_send_request,
2564 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
2565 .timer = xs_udp_timer,
2566 .release_request = xprt_release_rqst_cong,
2568 .destroy = xs_destroy,
2569 .print_stats = xs_udp_print_stats,
2572 static struct rpc_xprt_ops xs_tcp_ops = {
2573 .reserve_xprt = xprt_reserve_xprt,
2574 .release_xprt = xs_tcp_release_xprt,
2575 .alloc_slot = xprt_lock_and_alloc_slot,
2576 .rpcbind = rpcb_getport_async,
2577 .set_port = xs_set_port,
2578 .connect = xs_connect,
2579 .buf_alloc = rpc_malloc,
2580 .buf_free = rpc_free,
2581 .send_request = xs_tcp_send_request,
2582 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2583 .close = xs_tcp_close,
2584 .destroy = xs_destroy,
2585 .print_stats = xs_tcp_print_stats,
2589 * The rpc_xprt_ops for the server backchannel
2592 static struct rpc_xprt_ops bc_tcp_ops = {
2593 .reserve_xprt = xprt_reserve_xprt,
2594 .release_xprt = xprt_release_xprt,
2595 .alloc_slot = xprt_alloc_slot,
2596 .buf_alloc = bc_malloc,
2597 .buf_free = bc_free,
2598 .send_request = bc_send_request,
2599 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2601 .destroy = bc_destroy,
2602 .print_stats = xs_tcp_print_stats,
2605 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2607 static const struct sockaddr_in sin = {
2608 .sin_family = AF_INET,
2609 .sin_addr.s_addr = htonl(INADDR_ANY),
2611 static const struct sockaddr_in6 sin6 = {
2612 .sin6_family = AF_INET6,
2613 .sin6_addr = IN6ADDR_ANY_INIT,
2620 memcpy(sap, &sin, sizeof(sin));
2623 memcpy(sap, &sin6, sizeof(sin6));
2626 dprintk("RPC: %s: Bad address family\n", __func__);
2627 return -EAFNOSUPPORT;
2632 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2633 unsigned int slot_table_size,
2634 unsigned int max_slot_table_size)
2636 struct rpc_xprt *xprt;
2637 struct sock_xprt *new;
2639 if (args->addrlen > sizeof(xprt->addr)) {
2640 dprintk("RPC: xs_setup_xprt: address too large\n");
2641 return ERR_PTR(-EBADF);
2644 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2645 max_slot_table_size);
2647 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2649 return ERR_PTR(-ENOMEM);
2652 new = container_of(xprt, struct sock_xprt, xprt);
2653 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2654 xprt->addrlen = args->addrlen;
2656 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2659 err = xs_init_anyaddr(args->dstaddr->sa_family,
2660 (struct sockaddr *)&new->srcaddr);
2663 return ERR_PTR(err);
2670 static const struct rpc_timeout xs_local_default_timeout = {
2671 .to_initval = 10 * HZ,
2672 .to_maxval = 10 * HZ,
2677 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2678 * @args: rpc transport creation arguments
2680 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2682 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2684 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2685 struct sock_xprt *transport;
2686 struct rpc_xprt *xprt;
2687 struct rpc_xprt *ret;
2689 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2690 xprt_max_tcp_slot_table_entries);
2693 transport = container_of(xprt, struct sock_xprt, xprt);
2696 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2697 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2699 xprt->bind_timeout = XS_BIND_TO;
2700 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2701 xprt->idle_timeout = XS_IDLE_DISC_TO;
2703 xprt->ops = &xs_local_ops;
2704 xprt->timeout = &xs_local_default_timeout;
2706 INIT_DELAYED_WORK(&transport->connect_worker,
2707 xs_dummy_setup_socket);
2709 switch (sun->sun_family) {
2711 if (sun->sun_path[0] != '/') {
2712 dprintk("RPC: bad AF_LOCAL address: %s\n",
2714 ret = ERR_PTR(-EINVAL);
2717 xprt_set_bound(xprt);
2718 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2719 ret = ERR_PTR(xs_local_setup_socket(transport));
2724 ret = ERR_PTR(-EAFNOSUPPORT);
2728 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2729 xprt->address_strings[RPC_DISPLAY_ADDR]);
2731 if (try_module_get(THIS_MODULE))
2733 ret = ERR_PTR(-EINVAL);
2739 static const struct rpc_timeout xs_udp_default_timeout = {
2740 .to_initval = 5 * HZ,
2741 .to_maxval = 30 * HZ,
2742 .to_increment = 5 * HZ,
2747 * xs_setup_udp - Set up transport to use a UDP socket
2748 * @args: rpc transport creation arguments
2751 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2753 struct sockaddr *addr = args->dstaddr;
2754 struct rpc_xprt *xprt;
2755 struct sock_xprt *transport;
2756 struct rpc_xprt *ret;
2758 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2759 xprt_udp_slot_table_entries);
2762 transport = container_of(xprt, struct sock_xprt, xprt);
2764 xprt->prot = IPPROTO_UDP;
2766 /* XXX: header size can vary due to auth type, IPv6, etc. */
2767 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2769 xprt->bind_timeout = XS_BIND_TO;
2770 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2771 xprt->idle_timeout = XS_IDLE_DISC_TO;
2773 xprt->ops = &xs_udp_ops;
2775 xprt->timeout = &xs_udp_default_timeout;
2777 switch (addr->sa_family) {
2779 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2780 xprt_set_bound(xprt);
2782 INIT_DELAYED_WORK(&transport->connect_worker,
2783 xs_udp_setup_socket);
2784 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2787 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2788 xprt_set_bound(xprt);
2790 INIT_DELAYED_WORK(&transport->connect_worker,
2791 xs_udp_setup_socket);
2792 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2795 ret = ERR_PTR(-EAFNOSUPPORT);
2799 if (xprt_bound(xprt))
2800 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2801 xprt->address_strings[RPC_DISPLAY_ADDR],
2802 xprt->address_strings[RPC_DISPLAY_PORT],
2803 xprt->address_strings[RPC_DISPLAY_PROTO]);
2805 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2806 xprt->address_strings[RPC_DISPLAY_ADDR],
2807 xprt->address_strings[RPC_DISPLAY_PROTO]);
2809 if (try_module_get(THIS_MODULE))
2811 ret = ERR_PTR(-EINVAL);
2817 static const struct rpc_timeout xs_tcp_default_timeout = {
2818 .to_initval = 60 * HZ,
2819 .to_maxval = 60 * HZ,
2824 * xs_setup_tcp - Set up transport to use a TCP socket
2825 * @args: rpc transport creation arguments
2828 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2830 struct sockaddr *addr = args->dstaddr;
2831 struct rpc_xprt *xprt;
2832 struct sock_xprt *transport;
2833 struct rpc_xprt *ret;
2834 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2836 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2837 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2839 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2840 max_slot_table_size);
2843 transport = container_of(xprt, struct sock_xprt, xprt);
2845 xprt->prot = IPPROTO_TCP;
2846 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2847 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2849 xprt->bind_timeout = XS_BIND_TO;
2850 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2851 xprt->idle_timeout = XS_IDLE_DISC_TO;
2853 xprt->ops = &xs_tcp_ops;
2854 xprt->timeout = &xs_tcp_default_timeout;
2856 switch (addr->sa_family) {
2858 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2859 xprt_set_bound(xprt);
2861 INIT_DELAYED_WORK(&transport->connect_worker,
2862 xs_tcp_setup_socket);
2863 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2866 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2867 xprt_set_bound(xprt);
2869 INIT_DELAYED_WORK(&transport->connect_worker,
2870 xs_tcp_setup_socket);
2871 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2874 ret = ERR_PTR(-EAFNOSUPPORT);
2878 if (xprt_bound(xprt))
2879 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2880 xprt->address_strings[RPC_DISPLAY_ADDR],
2881 xprt->address_strings[RPC_DISPLAY_PORT],
2882 xprt->address_strings[RPC_DISPLAY_PROTO]);
2884 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2885 xprt->address_strings[RPC_DISPLAY_ADDR],
2886 xprt->address_strings[RPC_DISPLAY_PROTO]);
2889 if (try_module_get(THIS_MODULE))
2891 ret = ERR_PTR(-EINVAL);
2898 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2899 * @args: rpc transport creation arguments
2902 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2904 struct sockaddr *addr = args->dstaddr;
2905 struct rpc_xprt *xprt;
2906 struct sock_xprt *transport;
2907 struct svc_sock *bc_sock;
2908 struct rpc_xprt *ret;
2910 if (args->bc_xprt->xpt_bc_xprt) {
2912 * This server connection already has a backchannel
2913 * transport; we can't create a new one, as we wouldn't
2914 * be able to match replies based on xid any more. So,
2915 * reuse the already-existing one:
2917 return args->bc_xprt->xpt_bc_xprt;
2919 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2920 xprt_tcp_slot_table_entries);
2923 transport = container_of(xprt, struct sock_xprt, xprt);
2925 xprt->prot = IPPROTO_TCP;
2926 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2927 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2928 xprt->timeout = &xs_tcp_default_timeout;
2931 xprt_set_bound(xprt);
2932 xprt->bind_timeout = 0;
2933 xprt->reestablish_timeout = 0;
2934 xprt->idle_timeout = 0;
2936 xprt->ops = &bc_tcp_ops;
2938 switch (addr->sa_family) {
2940 xs_format_peer_addresses(xprt, "tcp",
2944 xs_format_peer_addresses(xprt, "tcp",
2945 RPCBIND_NETID_TCP6);
2948 ret = ERR_PTR(-EAFNOSUPPORT);
2952 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2953 xprt->address_strings[RPC_DISPLAY_ADDR],
2954 xprt->address_strings[RPC_DISPLAY_PORT],
2955 xprt->address_strings[RPC_DISPLAY_PROTO]);
2958 * Once we've associated a backchannel xprt with a connection,
2959 * we want to keep it around as long as the connection lasts,
2960 * in case we need to start using it for a backchannel again;
2961 * this reference won't be dropped until bc_xprt is destroyed.
2964 args->bc_xprt->xpt_bc_xprt = xprt;
2965 xprt->bc_xprt = args->bc_xprt;
2966 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2967 transport->sock = bc_sock->sk_sock;
2968 transport->inet = bc_sock->sk_sk;
2971 * Since we don't want connections for the backchannel, we set
2972 * the xprt status to connected
2974 xprt_set_connected(xprt);
2977 if (try_module_get(THIS_MODULE))
2980 ret = ERR_PTR(-EINVAL);
2986 static struct xprt_class xs_local_transport = {
2987 .list = LIST_HEAD_INIT(xs_local_transport.list),
2988 .name = "named UNIX socket",
2989 .owner = THIS_MODULE,
2990 .ident = XPRT_TRANSPORT_LOCAL,
2991 .setup = xs_setup_local,
2994 static struct xprt_class xs_udp_transport = {
2995 .list = LIST_HEAD_INIT(xs_udp_transport.list),
2997 .owner = THIS_MODULE,
2998 .ident = XPRT_TRANSPORT_UDP,
2999 .setup = xs_setup_udp,
3002 static struct xprt_class xs_tcp_transport = {
3003 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3005 .owner = THIS_MODULE,
3006 .ident = XPRT_TRANSPORT_TCP,
3007 .setup = xs_setup_tcp,
3010 static struct xprt_class xs_bc_tcp_transport = {
3011 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3012 .name = "tcp NFSv4.1 backchannel",
3013 .owner = THIS_MODULE,
3014 .ident = XPRT_TRANSPORT_BC_TCP,
3015 .setup = xs_setup_bc_tcp,
3019 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3022 int init_socket_xprt(void)
3025 if (!sunrpc_table_header)
3026 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3029 xprt_register_transport(&xs_local_transport);
3030 xprt_register_transport(&xs_udp_transport);
3031 xprt_register_transport(&xs_tcp_transport);
3032 xprt_register_transport(&xs_bc_tcp_transport);
3038 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3041 void cleanup_socket_xprt(void)
3044 if (sunrpc_table_header) {
3045 unregister_sysctl_table(sunrpc_table_header);
3046 sunrpc_table_header = NULL;
3050 xprt_unregister_transport(&xs_local_transport);
3051 xprt_unregister_transport(&xs_udp_transport);
3052 xprt_unregister_transport(&xs_tcp_transport);
3053 xprt_unregister_transport(&xs_bc_tcp_transport);
3056 static int param_set_uint_minmax(const char *val,
3057 const struct kernel_param *kp,
3058 unsigned int min, unsigned int max)
3065 ret = strict_strtoul(val, 0, &num);
3066 if (ret == -EINVAL || num < min || num > max)
3068 *((unsigned int *)kp->arg) = num;
3072 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3074 return param_set_uint_minmax(val, kp,
3079 static struct kernel_param_ops param_ops_portnr = {
3080 .set = param_set_portnr,
3081 .get = param_get_uint,
3084 #define param_check_portnr(name, p) \
3085 __param_check(name, p, unsigned int);
3087 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3088 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3090 static int param_set_slot_table_size(const char *val,
3091 const struct kernel_param *kp)
3093 return param_set_uint_minmax(val, kp,
3095 RPC_MAX_SLOT_TABLE);
3098 static struct kernel_param_ops param_ops_slot_table_size = {
3099 .set = param_set_slot_table_size,
3100 .get = param_get_uint,
3103 #define param_check_slot_table_size(name, p) \
3104 __param_check(name, p, unsigned int);
3106 static int param_set_max_slot_table_size(const char *val,
3107 const struct kernel_param *kp)
3109 return param_set_uint_minmax(val, kp,
3111 RPC_MAX_SLOT_TABLE_LIMIT);
3114 static struct kernel_param_ops param_ops_max_slot_table_size = {
3115 .set = param_set_max_slot_table_size,
3116 .get = param_get_uint,
3119 #define param_check_max_slot_table_size(name, p) \
3120 __param_check(name, p, unsigned int);
3122 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3123 slot_table_size, 0644);
3124 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3125 max_slot_table_size, 0644);
3126 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3127 slot_table_size, 0644);