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!
78 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
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)
189 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
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)
219 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
221 return (struct rpc_xprt *) sk->sk_user_data;
224 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
226 return (struct sockaddr *) &xprt->addr;
229 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
231 return (struct sockaddr_un *) &xprt->addr;
234 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
236 return (struct sockaddr_in *) &xprt->addr;
239 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
241 return (struct sockaddr_in6 *) &xprt->addr;
244 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
246 struct sockaddr *sap = xs_addr(xprt);
247 struct sockaddr_in6 *sin6;
248 struct sockaddr_in *sin;
249 struct sockaddr_un *sun;
252 switch (sap->sa_family) {
254 sun = xs_addr_un(xprt);
255 strlcpy(buf, sun->sun_path, sizeof(buf));
256 xprt->address_strings[RPC_DISPLAY_ADDR] =
257 kstrdup(buf, GFP_KERNEL);
260 (void)rpc_ntop(sap, buf, sizeof(buf));
261 xprt->address_strings[RPC_DISPLAY_ADDR] =
262 kstrdup(buf, GFP_KERNEL);
263 sin = xs_addr_in(xprt);
264 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
267 (void)rpc_ntop(sap, buf, sizeof(buf));
268 xprt->address_strings[RPC_DISPLAY_ADDR] =
269 kstrdup(buf, GFP_KERNEL);
270 sin6 = xs_addr_in6(xprt);
271 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
277 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
280 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
282 struct sockaddr *sap = xs_addr(xprt);
285 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
286 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
288 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
289 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
292 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
293 const char *protocol,
296 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
297 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
298 xs_format_common_peer_addresses(xprt);
299 xs_format_common_peer_ports(xprt);
302 static void xs_update_peer_port(struct rpc_xprt *xprt)
304 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
305 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
307 xs_format_common_peer_ports(xprt);
310 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
314 for (i = 0; i < RPC_DISPLAY_MAX; i++)
316 case RPC_DISPLAY_PROTO:
317 case RPC_DISPLAY_NETID:
320 kfree(xprt->address_strings[i]);
324 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
326 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
328 struct msghdr msg = {
330 .msg_namelen = addrlen,
331 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
334 .iov_base = vec->iov_base + base,
335 .iov_len = vec->iov_len - base,
338 if (iov.iov_len != 0)
339 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
340 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
343 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p)
345 ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
346 int offset, size_t size, int flags);
348 unsigned int remainder;
351 remainder = xdr->page_len - base;
352 base += xdr->page_base;
353 ppage = xdr->pages + (base >> PAGE_SHIFT);
355 do_sendpage = sock->ops->sendpage;
357 do_sendpage = sock_no_sendpage;
359 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
360 int flags = XS_SENDMSG_FLAGS;
363 if (remainder != 0 || more)
365 err = do_sendpage(sock, *ppage, base, len, flags);
366 if (remainder == 0 || err != len)
380 * xs_sendpages - write pages directly to a socket
381 * @sock: socket to send on
382 * @addr: UDP only -- address of destination
383 * @addrlen: UDP only -- length of destination address
384 * @xdr: buffer containing this request
385 * @base: starting position in the buffer
386 * @zerocopy: true if it is safe to use sendpage()
387 * @sent_p: return the total number of bytes successfully queued for sending
390 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p)
392 unsigned int remainder = xdr->len - base;
399 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
405 if (base < xdr->head[0].iov_len || addr != NULL) {
406 unsigned int len = xdr->head[0].iov_len - base;
408 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
409 if (remainder == 0 || err != len)
414 base -= xdr->head[0].iov_len;
416 if (base < xdr->page_len) {
417 unsigned int len = xdr->page_len - base;
419 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy, &sent);
421 if (remainder == 0 || sent != len)
425 base -= xdr->page_len;
427 if (base >= xdr->tail[0].iov_len)
429 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
438 static void xs_nospace_callback(struct rpc_task *task)
440 struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
442 transport->inet->sk_write_pending--;
443 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
447 * xs_nospace - place task on wait queue if transmit was incomplete
448 * @task: task to put to sleep
451 static int xs_nospace(struct rpc_task *task)
453 struct rpc_rqst *req = task->tk_rqstp;
454 struct rpc_xprt *xprt = req->rq_xprt;
455 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
456 struct sock *sk = transport->inet;
459 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
460 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
463 /* Protect against races with write_space */
464 spin_lock_bh(&xprt->transport_lock);
466 /* Don't race with disconnect */
467 if (xprt_connected(xprt)) {
468 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
470 * Notify TCP that we're limited by the application
473 set_bit(SOCK_NOSPACE, &transport->sock->flags);
474 sk->sk_write_pending++;
475 /* ...and wait for more buffer space */
476 xprt_wait_for_buffer_space(task, xs_nospace_callback);
479 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
483 spin_unlock_bh(&xprt->transport_lock);
485 /* Race breaker in case memory is freed before above code is called */
486 sk->sk_write_space(sk);
491 * Construct a stream transport record marker in @buf.
493 static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
495 u32 reclen = buf->len - sizeof(rpc_fraghdr);
496 rpc_fraghdr *base = buf->head[0].iov_base;
497 *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
501 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
502 * @task: RPC task that manages the state of an RPC request
505 * 0: The request has been sent
506 * EAGAIN: The socket was blocked, please call again later to
507 * complete the request
508 * ENOTCONN: Caller needs to invoke connect logic then call again
509 * other: Some other error occured, the request was not sent
511 static int xs_local_send_request(struct rpc_task *task)
513 struct rpc_rqst *req = task->tk_rqstp;
514 struct rpc_xprt *xprt = req->rq_xprt;
515 struct sock_xprt *transport =
516 container_of(xprt, struct sock_xprt, xprt);
517 struct xdr_buf *xdr = &req->rq_snd_buf;
521 xs_encode_stream_record_marker(&req->rq_snd_buf);
523 xs_pktdump("packet data:",
524 req->rq_svec->iov_base, req->rq_svec->iov_len);
526 status = xs_sendpages(transport->sock, NULL, 0, xdr, req->rq_bytes_sent,
528 dprintk("RPC: %s(%u) = %d\n",
529 __func__, xdr->len - req->rq_bytes_sent, status);
530 if (likely(sent > 0) || status == 0) {
531 req->rq_bytes_sent += sent;
532 req->rq_xmit_bytes_sent += sent;
533 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
534 req->rq_bytes_sent = 0;
543 status = xs_nospace(task);
546 dprintk("RPC: sendmsg returned unrecognized error %d\n",
557 * xs_udp_send_request - write an RPC request to a UDP socket
558 * @task: address of RPC task that manages the state of an RPC request
561 * 0: The request has been sent
562 * EAGAIN: The socket was blocked, please call again later to
563 * complete the request
564 * ENOTCONN: Caller needs to invoke connect logic then call again
565 * other: Some other error occurred, the request was not sent
567 static int xs_udp_send_request(struct rpc_task *task)
569 struct rpc_rqst *req = task->tk_rqstp;
570 struct rpc_xprt *xprt = req->rq_xprt;
571 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
572 struct xdr_buf *xdr = &req->rq_snd_buf;
576 xs_pktdump("packet data:",
577 req->rq_svec->iov_base,
578 req->rq_svec->iov_len);
580 if (!xprt_bound(xprt))
582 status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
583 xdr, req->rq_bytes_sent, true, &sent);
585 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
586 xdr->len - req->rq_bytes_sent, status);
588 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
589 if (status == -EPERM)
592 if (sent > 0 || status == 0) {
593 req->rq_xmit_bytes_sent += sent;
594 if (sent >= req->rq_slen)
596 /* Still some bytes left; set up for a retry later. */
604 /* Should we call xs_close() here? */
607 status = xs_nospace(task);
610 dprintk("RPC: sendmsg returned unrecognized error %d\n",
617 /* When the server has died, an ICMP port unreachable message
618 * prompts ECONNREFUSED. */
619 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
626 * xs_tcp_shutdown - gracefully shut down a TCP socket
629 * Initiates a graceful shutdown of the TCP socket by calling the
630 * equivalent of shutdown(SHUT_WR);
632 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
634 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
635 struct socket *sock = transport->sock;
638 kernel_sock_shutdown(sock, SHUT_WR);
639 trace_rpc_socket_shutdown(xprt, sock);
644 * xs_tcp_send_request - write an RPC request to a TCP socket
645 * @task: address of RPC task that manages the state of an RPC request
648 * 0: The request has been sent
649 * EAGAIN: The socket was blocked, please call again later to
650 * complete the request
651 * ENOTCONN: Caller needs to invoke connect logic then call again
652 * other: Some other error occurred, the request was not sent
654 * XXX: In the case of soft timeouts, should we eventually give up
655 * if sendmsg is not able to make progress?
657 static int xs_tcp_send_request(struct rpc_task *task)
659 struct rpc_rqst *req = task->tk_rqstp;
660 struct rpc_xprt *xprt = req->rq_xprt;
661 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
662 struct xdr_buf *xdr = &req->rq_snd_buf;
663 bool zerocopy = true;
667 xs_encode_stream_record_marker(&req->rq_snd_buf);
669 xs_pktdump("packet data:",
670 req->rq_svec->iov_base,
671 req->rq_svec->iov_len);
672 /* Don't use zero copy if this is a resend. If the RPC call
673 * completes while the socket holds a reference to the pages,
674 * then we may end up resending corrupted data.
676 if (task->tk_flags & RPC_TASK_SENT)
679 /* Continue transmitting the packet/record. We must be careful
680 * to cope with writespace callbacks arriving _after_ we have
681 * called sendmsg(). */
684 status = xs_sendpages(transport->sock, NULL, 0, xdr,
685 req->rq_bytes_sent, zerocopy, &sent);
687 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
688 xdr->len - req->rq_bytes_sent, status);
690 if (unlikely(sent == 0 && status < 0))
693 /* If we've sent the entire packet, immediately
694 * reset the count of bytes sent. */
695 req->rq_bytes_sent += sent;
696 req->rq_xmit_bytes_sent += sent;
697 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
698 req->rq_bytes_sent = 0;
711 /* Should we call xs_close() here? */
715 status = xs_nospace(task);
718 dprintk("RPC: sendmsg returned unrecognized error %d\n",
721 xs_tcp_shutdown(xprt);
726 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
733 * xs_tcp_release_xprt - clean up after a tcp transmission
737 * This cleans up if an error causes us to abort the transmission of a request.
738 * In this case, the socket may need to be reset in order to avoid confusing
741 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
743 struct rpc_rqst *req;
745 if (task != xprt->snd_task)
749 req = task->tk_rqstp;
752 if (req->rq_bytes_sent == 0)
754 if (req->rq_bytes_sent == req->rq_snd_buf.len)
756 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
758 xprt_release_xprt(xprt, task);
761 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
763 transport->old_data_ready = sk->sk_data_ready;
764 transport->old_state_change = sk->sk_state_change;
765 transport->old_write_space = sk->sk_write_space;
766 transport->old_error_report = sk->sk_error_report;
769 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
771 sk->sk_data_ready = transport->old_data_ready;
772 sk->sk_state_change = transport->old_state_change;
773 sk->sk_write_space = transport->old_write_space;
774 sk->sk_error_report = transport->old_error_report;
778 * xs_error_report - callback to handle TCP socket state errors
781 * Note: we don't call sock_error() since there may be a rpc_task
782 * using the socket, and so we don't want to clear sk->sk_err.
784 static void xs_error_report(struct sock *sk)
786 struct rpc_xprt *xprt;
789 read_lock_bh(&sk->sk_callback_lock);
790 if (!(xprt = xprt_from_sock(sk)))
796 dprintk("RPC: xs_error_report client %p, error=%d...\n",
798 trace_rpc_socket_error(xprt, sk->sk_socket, err);
799 if (test_bit(XPRT_CONNECTION_REUSE, &xprt->state))
801 xprt_wake_pending_tasks(xprt, err);
803 read_unlock_bh(&sk->sk_callback_lock);
806 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
808 smp_mb__before_atomic();
809 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
810 clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
811 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
812 clear_bit(XPRT_CLOSING, &xprt->state);
813 smp_mb__after_atomic();
816 static void xs_reset_transport(struct sock_xprt *transport)
818 struct socket *sock = transport->sock;
819 struct sock *sk = transport->inet;
820 struct rpc_xprt *xprt = &transport->xprt;
825 write_lock_bh(&sk->sk_callback_lock);
826 transport->inet = NULL;
827 transport->sock = NULL;
829 sk->sk_user_data = NULL;
831 xs_restore_old_callbacks(transport, sk);
832 write_unlock_bh(&sk->sk_callback_lock);
833 xs_sock_reset_connection_flags(xprt);
835 trace_rpc_socket_close(xprt, sock);
840 * xs_close - close a socket
843 * This is used when all requests are complete; ie, no DRC state remains
844 * on the server we want to save.
846 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
847 * xs_reset_transport() zeroing the socket from underneath a writer.
849 static void xs_close(struct rpc_xprt *xprt)
851 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
853 dprintk("RPC: xs_close xprt %p\n", xprt);
855 xs_reset_transport(transport);
856 xprt->reestablish_timeout = 0;
858 xprt_disconnect_done(xprt);
861 static void xs_tcp_close(struct rpc_xprt *xprt)
863 if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state))
866 xs_tcp_shutdown(xprt);
869 static void xs_xprt_free(struct rpc_xprt *xprt)
871 xs_free_peer_addresses(xprt);
876 * xs_destroy - prepare to shutdown a transport
877 * @xprt: doomed transport
880 static void xs_destroy(struct rpc_xprt *xprt)
882 dprintk("RPC: xs_destroy xprt %p\n", xprt);
886 module_put(THIS_MODULE);
889 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
891 struct xdr_skb_reader desc = {
893 .offset = sizeof(rpc_fraghdr),
894 .count = skb->len - sizeof(rpc_fraghdr),
897 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
905 * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets
906 * @sk: socket with data to read
907 * @len: how much data to read
909 * Currently this assumes we can read the whole reply in a single gulp.
911 static void xs_local_data_ready(struct sock *sk)
913 struct rpc_task *task;
914 struct rpc_xprt *xprt;
915 struct rpc_rqst *rovr;
917 int err, repsize, copied;
921 read_lock_bh(&sk->sk_callback_lock);
922 dprintk("RPC: %s...\n", __func__);
923 xprt = xprt_from_sock(sk);
927 skb = skb_recv_datagram(sk, 0, 1, &err);
931 repsize = skb->len - sizeof(rpc_fraghdr);
933 dprintk("RPC: impossible RPC reply size %d\n", repsize);
937 /* Copy the XID from the skb... */
938 xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
942 /* Look up and lock the request corresponding to the given XID */
943 spin_lock(&xprt->transport_lock);
944 rovr = xprt_lookup_rqst(xprt, *xp);
947 task = rovr->rq_task;
949 copied = rovr->rq_private_buf.buflen;
950 if (copied > repsize)
953 if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
954 dprintk("RPC: sk_buff copy failed\n");
958 xprt_complete_rqst(task, copied);
961 spin_unlock(&xprt->transport_lock);
963 skb_free_datagram(sk, skb);
965 read_unlock_bh(&sk->sk_callback_lock);
969 * xs_udp_data_ready - "data ready" callback for UDP sockets
970 * @sk: socket with data to read
971 * @len: how much data to read
974 static void xs_udp_data_ready(struct sock *sk)
976 struct rpc_task *task;
977 struct rpc_xprt *xprt;
978 struct rpc_rqst *rovr;
980 int err, repsize, copied;
984 read_lock_bh(&sk->sk_callback_lock);
985 dprintk("RPC: xs_udp_data_ready...\n");
986 if (!(xprt = xprt_from_sock(sk)))
989 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
992 repsize = skb->len - sizeof(struct udphdr);
994 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
998 /* Copy the XID from the skb... */
999 xp = skb_header_pointer(skb, sizeof(struct udphdr),
1000 sizeof(_xid), &_xid);
1004 /* Look up and lock the request corresponding to the given XID */
1005 spin_lock(&xprt->transport_lock);
1006 rovr = xprt_lookup_rqst(xprt, *xp);
1009 task = rovr->rq_task;
1011 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1014 /* Suck it into the iovec, verify checksum if not done by hw. */
1015 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1016 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
1020 UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
1022 xprt_adjust_cwnd(xprt, task, copied);
1023 xprt_complete_rqst(task, copied);
1026 spin_unlock(&xprt->transport_lock);
1028 skb_free_datagram(sk, skb);
1030 read_unlock_bh(&sk->sk_callback_lock);
1034 * Helper function to force a TCP close if the server is sending
1035 * junk and/or it has put us in CLOSE_WAIT
1037 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1039 set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1040 xprt_force_disconnect(xprt);
1043 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1045 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1049 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1050 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1051 used = xdr_skb_read_bits(desc, p, len);
1052 transport->tcp_offset += used;
1056 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1057 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1058 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1060 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1061 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1063 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1064 transport->tcp_offset = 0;
1066 /* Sanity check of the record length */
1067 if (unlikely(transport->tcp_reclen < 8)) {
1068 dprintk("RPC: invalid TCP record fragment length\n");
1069 xs_tcp_force_close(xprt);
1072 dprintk("RPC: reading TCP record fragment of length %d\n",
1073 transport->tcp_reclen);
1076 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1078 if (transport->tcp_offset == transport->tcp_reclen) {
1079 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1080 transport->tcp_offset = 0;
1081 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1082 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1083 transport->tcp_flags |= TCP_RCV_COPY_XID;
1084 transport->tcp_copied = 0;
1089 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1094 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1095 dprintk("RPC: reading XID (%Zu bytes)\n", len);
1096 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1097 used = xdr_skb_read_bits(desc, p, len);
1098 transport->tcp_offset += used;
1101 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1102 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1103 transport->tcp_copied = 4;
1104 dprintk("RPC: reading %s XID %08x\n",
1105 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1107 ntohl(transport->tcp_xid));
1108 xs_tcp_check_fraghdr(transport);
1111 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1112 struct xdr_skb_reader *desc)
1119 * We want transport->tcp_offset to be 8 at the end of this routine
1120 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1121 * When this function is called for the first time,
1122 * transport->tcp_offset is 4 (after having already read the xid).
1124 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1125 len = sizeof(transport->tcp_calldir) - offset;
1126 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len);
1127 p = ((char *) &transport->tcp_calldir) + offset;
1128 used = xdr_skb_read_bits(desc, p, len);
1129 transport->tcp_offset += used;
1132 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1134 * We don't yet have the XDR buffer, so we will write the calldir
1135 * out after we get the buffer from the 'struct rpc_rqst'
1137 switch (ntohl(transport->tcp_calldir)) {
1139 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1140 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1141 transport->tcp_flags |= TCP_RPC_REPLY;
1144 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1145 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1146 transport->tcp_flags &= ~TCP_RPC_REPLY;
1149 dprintk("RPC: invalid request message type\n");
1150 xs_tcp_force_close(&transport->xprt);
1152 xs_tcp_check_fraghdr(transport);
1155 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1156 struct xdr_skb_reader *desc,
1157 struct rpc_rqst *req)
1159 struct sock_xprt *transport =
1160 container_of(xprt, struct sock_xprt, xprt);
1161 struct xdr_buf *rcvbuf;
1165 rcvbuf = &req->rq_private_buf;
1167 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1169 * Save the RPC direction in the XDR buffer
1171 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1172 &transport->tcp_calldir,
1173 sizeof(transport->tcp_calldir));
1174 transport->tcp_copied += sizeof(transport->tcp_calldir);
1175 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1179 if (len > transport->tcp_reclen - transport->tcp_offset) {
1180 struct xdr_skb_reader my_desc;
1182 len = transport->tcp_reclen - transport->tcp_offset;
1183 memcpy(&my_desc, desc, sizeof(my_desc));
1184 my_desc.count = len;
1185 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1186 &my_desc, xdr_skb_read_bits);
1190 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1191 desc, xdr_skb_read_bits);
1194 transport->tcp_copied += r;
1195 transport->tcp_offset += r;
1198 /* Error when copying to the receive buffer,
1199 * usually because we weren't able to allocate
1200 * additional buffer pages. All we can do now
1201 * is turn off TCP_RCV_COPY_DATA, so the request
1202 * will not receive any additional updates,
1204 * Any remaining data from this record will
1207 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1208 dprintk("RPC: XID %08x truncated request\n",
1209 ntohl(transport->tcp_xid));
1210 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1211 "tcp_offset = %u, tcp_reclen = %u\n",
1212 xprt, transport->tcp_copied,
1213 transport->tcp_offset, transport->tcp_reclen);
1217 dprintk("RPC: XID %08x read %Zd bytes\n",
1218 ntohl(transport->tcp_xid), r);
1219 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1220 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1221 transport->tcp_offset, transport->tcp_reclen);
1223 if (transport->tcp_copied == req->rq_private_buf.buflen)
1224 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1225 else if (transport->tcp_offset == transport->tcp_reclen) {
1226 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1227 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1232 * Finds the request corresponding to the RPC xid and invokes the common
1233 * tcp read code to read the data.
1235 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1236 struct xdr_skb_reader *desc)
1238 struct sock_xprt *transport =
1239 container_of(xprt, struct sock_xprt, xprt);
1240 struct rpc_rqst *req;
1242 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1244 /* Find and lock the request corresponding to this xid */
1245 spin_lock(&xprt->transport_lock);
1246 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1248 dprintk("RPC: XID %08x request not found!\n",
1249 ntohl(transport->tcp_xid));
1250 spin_unlock(&xprt->transport_lock);
1254 xs_tcp_read_common(xprt, desc, req);
1256 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1257 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1259 spin_unlock(&xprt->transport_lock);
1263 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1265 * Obtains an rpc_rqst previously allocated and invokes the common
1266 * tcp read code to read the data. The result is placed in the callback
1268 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1269 * connection and return -1.
1271 static int xs_tcp_read_callback(struct rpc_xprt *xprt,
1272 struct xdr_skb_reader *desc)
1274 struct sock_xprt *transport =
1275 container_of(xprt, struct sock_xprt, xprt);
1276 struct rpc_rqst *req;
1278 /* Look up and lock the request corresponding to the given XID */
1279 spin_lock(&xprt->transport_lock);
1280 req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
1282 spin_unlock(&xprt->transport_lock);
1283 printk(KERN_WARNING "Callback slot table overflowed\n");
1284 xprt_force_disconnect(xprt);
1288 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1289 xs_tcp_read_common(xprt, desc, req);
1291 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1292 xprt_complete_bc_request(req, transport->tcp_copied);
1293 spin_unlock(&xprt->transport_lock);
1298 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1299 struct xdr_skb_reader *desc)
1301 struct sock_xprt *transport =
1302 container_of(xprt, struct sock_xprt, xprt);
1304 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1305 xs_tcp_read_reply(xprt, desc) :
1306 xs_tcp_read_callback(xprt, desc);
1309 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1310 struct xdr_skb_reader *desc)
1312 return xs_tcp_read_reply(xprt, desc);
1314 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1317 * Read data off the transport. This can be either an RPC_CALL or an
1318 * RPC_REPLY. Relay the processing to helper functions.
1320 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1321 struct xdr_skb_reader *desc)
1323 struct sock_xprt *transport =
1324 container_of(xprt, struct sock_xprt, xprt);
1326 if (_xs_tcp_read_data(xprt, desc) == 0)
1327 xs_tcp_check_fraghdr(transport);
1330 * The transport_lock protects the request handling.
1331 * There's no need to hold it to update the tcp_flags.
1333 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1337 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1341 len = transport->tcp_reclen - transport->tcp_offset;
1342 if (len > desc->count)
1345 desc->offset += len;
1346 transport->tcp_offset += len;
1347 dprintk("RPC: discarded %Zu bytes\n", len);
1348 xs_tcp_check_fraghdr(transport);
1351 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1353 struct rpc_xprt *xprt = rd_desc->arg.data;
1354 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1355 struct xdr_skb_reader desc = {
1361 dprintk("RPC: xs_tcp_data_recv started\n");
1363 trace_xs_tcp_data_recv(transport);
1364 /* Read in a new fragment marker if necessary */
1365 /* Can we ever really expect to get completely empty fragments? */
1366 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1367 xs_tcp_read_fraghdr(xprt, &desc);
1370 /* Read in the xid if necessary */
1371 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1372 xs_tcp_read_xid(transport, &desc);
1375 /* Read in the call/reply flag */
1376 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1377 xs_tcp_read_calldir(transport, &desc);
1380 /* Read in the request data */
1381 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1382 xs_tcp_read_data(xprt, &desc);
1385 /* Skip over any trailing bytes on short reads */
1386 xs_tcp_read_discard(transport, &desc);
1387 } while (desc.count);
1388 trace_xs_tcp_data_recv(transport);
1389 dprintk("RPC: xs_tcp_data_recv done\n");
1390 return len - desc.count;
1394 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1395 * @sk: socket with data to read
1396 * @bytes: how much data to read
1399 static void xs_tcp_data_ready(struct sock *sk)
1401 struct rpc_xprt *xprt;
1402 read_descriptor_t rd_desc;
1404 unsigned long total = 0;
1406 dprintk("RPC: xs_tcp_data_ready...\n");
1408 read_lock_bh(&sk->sk_callback_lock);
1409 if (!(xprt = xprt_from_sock(sk))) {
1413 /* Any data means we had a useful conversation, so
1414 * the we don't need to delay the next reconnect
1416 if (xprt->reestablish_timeout)
1417 xprt->reestablish_timeout = 0;
1419 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1420 rd_desc.arg.data = xprt;
1422 rd_desc.count = 65536;
1423 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1428 trace_xs_tcp_data_ready(xprt, read, total);
1429 read_unlock_bh(&sk->sk_callback_lock);
1433 * Do the equivalent of linger/linger2 handling for dealing with
1434 * broken servers that don't close the socket in a timely
1437 static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1438 unsigned long timeout)
1440 struct sock_xprt *transport;
1442 if (xprt_test_and_set_connecting(xprt))
1444 set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1445 transport = container_of(xprt, struct sock_xprt, xprt);
1446 queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1450 static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1452 struct sock_xprt *transport;
1454 transport = container_of(xprt, struct sock_xprt, xprt);
1456 if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1457 !cancel_delayed_work(&transport->connect_worker))
1459 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1460 xprt_clear_connecting(xprt);
1463 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1465 xs_sock_reset_connection_flags(xprt);
1466 /* Mark transport as closed and wake up all pending tasks */
1467 xprt_disconnect_done(xprt);
1471 * xs_tcp_state_change - callback to handle TCP socket state changes
1472 * @sk: socket whose state has changed
1475 static void xs_tcp_state_change(struct sock *sk)
1477 struct rpc_xprt *xprt;
1479 read_lock_bh(&sk->sk_callback_lock);
1480 if (!(xprt = xprt_from_sock(sk)))
1482 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1483 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1484 sk->sk_state, xprt_connected(xprt),
1485 sock_flag(sk, SOCK_DEAD),
1486 sock_flag(sk, SOCK_ZAPPED),
1489 trace_rpc_socket_state_change(xprt, sk->sk_socket);
1490 switch (sk->sk_state) {
1491 case TCP_ESTABLISHED:
1492 spin_lock(&xprt->transport_lock);
1493 if (!xprt_test_and_set_connected(xprt)) {
1494 struct sock_xprt *transport = container_of(xprt,
1495 struct sock_xprt, xprt);
1497 /* Reset TCP record info */
1498 transport->tcp_offset = 0;
1499 transport->tcp_reclen = 0;
1500 transport->tcp_copied = 0;
1501 transport->tcp_flags =
1502 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1503 xprt->connect_cookie++;
1505 xprt_wake_pending_tasks(xprt, -EAGAIN);
1507 spin_unlock(&xprt->transport_lock);
1510 /* The client initiated a shutdown of the socket */
1511 xprt->connect_cookie++;
1512 xprt->reestablish_timeout = 0;
1513 set_bit(XPRT_CLOSING, &xprt->state);
1514 smp_mb__before_atomic();
1515 clear_bit(XPRT_CONNECTED, &xprt->state);
1516 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1517 smp_mb__after_atomic();
1518 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1520 case TCP_CLOSE_WAIT:
1521 /* The server initiated a shutdown of the socket */
1522 xprt->connect_cookie++;
1523 clear_bit(XPRT_CONNECTED, &xprt->state);
1524 xs_tcp_force_close(xprt);
1527 * If the server closed down the connection, make sure that
1528 * we back off before reconnecting
1530 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1531 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1534 set_bit(XPRT_CLOSING, &xprt->state);
1535 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1536 smp_mb__before_atomic();
1537 clear_bit(XPRT_CONNECTED, &xprt->state);
1538 smp_mb__after_atomic();
1541 xs_tcp_cancel_linger_timeout(xprt);
1542 xs_sock_mark_closed(xprt);
1545 read_unlock_bh(&sk->sk_callback_lock);
1548 static void xs_write_space(struct sock *sk)
1550 struct socket *sock;
1551 struct rpc_xprt *xprt;
1553 if (unlikely(!(sock = sk->sk_socket)))
1555 clear_bit(SOCK_NOSPACE, &sock->flags);
1557 if (unlikely(!(xprt = xprt_from_sock(sk))))
1559 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1562 xprt_write_space(xprt);
1566 * xs_udp_write_space - callback invoked when socket buffer space
1568 * @sk: socket whose state has changed
1570 * Called when more output buffer space is available for this socket.
1571 * We try not to wake our writers until they can make "significant"
1572 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1573 * with a bunch of small requests.
1575 static void xs_udp_write_space(struct sock *sk)
1577 read_lock_bh(&sk->sk_callback_lock);
1579 /* from net/core/sock.c:sock_def_write_space */
1580 if (sock_writeable(sk))
1583 read_unlock_bh(&sk->sk_callback_lock);
1587 * xs_tcp_write_space - callback invoked when socket buffer space
1589 * @sk: socket whose state has changed
1591 * Called when more output buffer space is available for this socket.
1592 * We try not to wake our writers until they can make "significant"
1593 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1594 * with a bunch of small requests.
1596 static void xs_tcp_write_space(struct sock *sk)
1598 read_lock_bh(&sk->sk_callback_lock);
1600 /* from net/core/stream.c:sk_stream_write_space */
1601 if (sk_stream_is_writeable(sk))
1604 read_unlock_bh(&sk->sk_callback_lock);
1607 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1609 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1610 struct sock *sk = transport->inet;
1612 if (transport->rcvsize) {
1613 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1614 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1616 if (transport->sndsize) {
1617 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1618 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1619 sk->sk_write_space(sk);
1624 * xs_udp_set_buffer_size - set send and receive limits
1625 * @xprt: generic transport
1626 * @sndsize: requested size of send buffer, in bytes
1627 * @rcvsize: requested size of receive buffer, in bytes
1629 * Set socket send and receive buffer size limits.
1631 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1633 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1635 transport->sndsize = 0;
1637 transport->sndsize = sndsize + 1024;
1638 transport->rcvsize = 0;
1640 transport->rcvsize = rcvsize + 1024;
1642 xs_udp_do_set_buffer_size(xprt);
1646 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1647 * @task: task that timed out
1649 * Adjust the congestion window after a retransmit timeout has occurred.
1651 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1653 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1656 static unsigned short xs_get_random_port(void)
1658 unsigned short range = xprt_max_resvport - xprt_min_resvport;
1659 unsigned short rand = (unsigned short) prandom_u32() % range;
1660 return rand + xprt_min_resvport;
1664 * xs_set_reuseaddr_port - set the socket's port and address reuse options
1667 * Note that this function has to be called on all sockets that share the
1668 * same port, and it must be called before binding.
1670 static void xs_sock_set_reuseport(struct socket *sock)
1674 kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT, &opt, sizeof(opt));
1677 static unsigned short xs_sock_getport(struct socket *sock)
1679 struct sockaddr_storage buf;
1681 unsigned short port = 0;
1683 if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0)
1685 switch (buf.ss_family) {
1687 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1690 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1697 * xs_set_port - reset the port number in the remote endpoint address
1698 * @xprt: generic transport
1699 * @port: new port number
1702 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1704 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1706 rpc_set_port(xs_addr(xprt), port);
1707 xs_update_peer_port(xprt);
1710 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1712 if (transport->srcport == 0)
1713 transport->srcport = xs_sock_getport(sock);
1716 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1718 unsigned short port = transport->srcport;
1720 if (port == 0 && transport->xprt.resvport)
1721 port = xs_get_random_port();
1725 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1727 if (transport->srcport != 0)
1728 transport->srcport = 0;
1729 if (!transport->xprt.resvport)
1731 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1732 return xprt_max_resvport;
1735 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1737 struct sockaddr_storage myaddr;
1739 unsigned short port = xs_get_srcport(transport);
1740 unsigned short last;
1743 * If we are asking for any ephemeral port (i.e. port == 0 &&
1744 * transport->xprt.resvport == 0), don't bind. Let the local
1745 * port selection happen implicitly when the socket is used
1746 * (for example at connect time).
1748 * This ensures that we can continue to establish TCP
1749 * connections even when all local ephemeral ports are already
1750 * a part of some TCP connection. This makes no difference
1751 * for UDP sockets, but also doens't harm them.
1753 * If we're asking for any reserved port (i.e. port == 0 &&
1754 * transport->xprt.resvport == 1) xs_get_srcport above will
1755 * ensure that port is non-zero and we will bind as needed.
1760 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1762 rpc_set_port((struct sockaddr *)&myaddr, port);
1763 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1764 transport->xprt.addrlen);
1766 transport->srcport = port;
1770 port = xs_next_srcport(transport, port);
1773 } while (err == -EADDRINUSE && nloop != 2);
1775 if (myaddr.ss_family == AF_INET)
1776 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1777 &((struct sockaddr_in *)&myaddr)->sin_addr,
1778 port, err ? "failed" : "ok", err);
1780 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1781 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1782 port, err ? "failed" : "ok", err);
1787 * We don't support autobind on AF_LOCAL sockets
1789 static void xs_local_rpcbind(struct rpc_task *task)
1792 xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt));
1796 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1800 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1801 static struct lock_class_key xs_key[2];
1802 static struct lock_class_key xs_slock_key[2];
1804 static inline void xs_reclassify_socketu(struct socket *sock)
1806 struct sock *sk = sock->sk;
1808 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1809 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1812 static inline void xs_reclassify_socket4(struct socket *sock)
1814 struct sock *sk = sock->sk;
1816 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1817 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1820 static inline void xs_reclassify_socket6(struct socket *sock)
1822 struct sock *sk = sock->sk;
1824 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1825 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1828 static inline void xs_reclassify_socket(int family, struct socket *sock)
1830 WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1831 if (sock_owned_by_user(sock->sk))
1836 xs_reclassify_socketu(sock);
1839 xs_reclassify_socket4(sock);
1842 xs_reclassify_socket6(sock);
1847 static inline void xs_reclassify_socketu(struct socket *sock)
1851 static inline void xs_reclassify_socket4(struct socket *sock)
1855 static inline void xs_reclassify_socket6(struct socket *sock)
1859 static inline void xs_reclassify_socket(int family, struct socket *sock)
1864 static void xs_dummy_setup_socket(struct work_struct *work)
1868 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1869 struct sock_xprt *transport, int family, int type,
1870 int protocol, bool reuseport)
1872 struct socket *sock;
1875 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1877 dprintk("RPC: can't create %d transport socket (%d).\n",
1881 xs_reclassify_socket(family, sock);
1884 xs_sock_set_reuseport(sock);
1886 err = xs_bind(transport, sock);
1894 return ERR_PTR(err);
1897 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1898 struct socket *sock)
1900 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1903 if (!transport->inet) {
1904 struct sock *sk = sock->sk;
1906 write_lock_bh(&sk->sk_callback_lock);
1908 xs_save_old_callbacks(transport, sk);
1910 sk->sk_user_data = xprt;
1911 sk->sk_data_ready = xs_local_data_ready;
1912 sk->sk_write_space = xs_udp_write_space;
1913 sk->sk_error_report = xs_error_report;
1914 sk->sk_allocation = GFP_ATOMIC;
1916 xprt_clear_connected(xprt);
1918 /* Reset to new socket */
1919 transport->sock = sock;
1920 transport->inet = sk;
1922 write_unlock_bh(&sk->sk_callback_lock);
1925 /* Tell the socket layer to start connecting... */
1926 xprt->stat.connect_count++;
1927 xprt->stat.connect_start = jiffies;
1928 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1932 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1933 * @xprt: RPC transport to connect
1934 * @transport: socket transport to connect
1935 * @create_sock: function to create a socket of the correct type
1937 static int xs_local_setup_socket(struct sock_xprt *transport)
1939 struct rpc_xprt *xprt = &transport->xprt;
1940 struct socket *sock;
1943 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1944 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1945 SOCK_STREAM, 0, &sock, 1);
1947 dprintk("RPC: can't create AF_LOCAL "
1948 "transport socket (%d).\n", -status);
1951 xs_reclassify_socketu(sock);
1953 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1954 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1956 status = xs_local_finish_connecting(xprt, sock);
1957 trace_rpc_socket_connect(xprt, sock, status);
1960 dprintk("RPC: xprt %p connected to %s\n",
1961 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1962 xprt_set_connected(xprt);
1966 dprintk("RPC: xprt %p: socket %s does not exist\n",
1967 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1970 dprintk("RPC: xprt %p: connection refused for %s\n",
1971 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1974 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1976 xprt->address_strings[RPC_DISPLAY_ADDR]);
1980 xprt_clear_connecting(xprt);
1981 xprt_wake_pending_tasks(xprt, status);
1985 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1987 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1990 if (RPC_IS_ASYNC(task)) {
1992 * We want the AF_LOCAL connect to be resolved in the
1993 * filesystem namespace of the process making the rpc
1994 * call. Thus we connect synchronously.
1996 * If we want to support asynchronous AF_LOCAL calls,
1997 * we'll need to figure out how to pass a namespace to
2000 rpc_exit(task, -ENOTCONN);
2003 ret = xs_local_setup_socket(transport);
2004 if (ret && !RPC_IS_SOFTCONN(task))
2005 msleep_interruptible(15000);
2008 #ifdef CONFIG_SUNRPC_SWAP
2009 static void xs_set_memalloc(struct rpc_xprt *xprt)
2011 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2015 sk_set_memalloc(transport->inet);
2019 * xs_swapper - Tag this transport as being used for swap.
2020 * @xprt: transport to tag
2021 * @enable: enable/disable
2024 int xs_swapper(struct rpc_xprt *xprt, int enable)
2026 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2032 xs_set_memalloc(xprt);
2033 } else if (xprt->swapper) {
2035 sk_clear_memalloc(transport->inet);
2040 EXPORT_SYMBOL_GPL(xs_swapper);
2042 static void xs_set_memalloc(struct rpc_xprt *xprt)
2047 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2049 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2051 if (!transport->inet) {
2052 struct sock *sk = sock->sk;
2054 write_lock_bh(&sk->sk_callback_lock);
2056 xs_save_old_callbacks(transport, sk);
2058 sk->sk_user_data = xprt;
2059 sk->sk_data_ready = xs_udp_data_ready;
2060 sk->sk_write_space = xs_udp_write_space;
2061 sk->sk_allocation = GFP_ATOMIC;
2063 xprt_set_connected(xprt);
2065 /* Reset to new socket */
2066 transport->sock = sock;
2067 transport->inet = sk;
2069 xs_set_memalloc(xprt);
2071 write_unlock_bh(&sk->sk_callback_lock);
2073 xs_udp_do_set_buffer_size(xprt);
2076 static void xs_udp_setup_socket(struct work_struct *work)
2078 struct sock_xprt *transport =
2079 container_of(work, struct sock_xprt, connect_worker.work);
2080 struct rpc_xprt *xprt = &transport->xprt;
2081 struct socket *sock = transport->sock;
2084 /* Start by resetting any existing state */
2085 xs_reset_transport(transport);
2086 sock = xs_create_sock(xprt, transport,
2087 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2088 IPPROTO_UDP, false);
2092 dprintk("RPC: worker connecting xprt %p via %s to "
2093 "%s (port %s)\n", xprt,
2094 xprt->address_strings[RPC_DISPLAY_PROTO],
2095 xprt->address_strings[RPC_DISPLAY_ADDR],
2096 xprt->address_strings[RPC_DISPLAY_PORT]);
2098 xs_udp_finish_connecting(xprt, sock);
2099 trace_rpc_socket_connect(xprt, sock, 0);
2102 xprt_unlock_connect(xprt, transport);
2103 xprt_clear_connecting(xprt);
2104 xprt_wake_pending_tasks(xprt, status);
2108 * We need to preserve the port number so the reply cache on the server can
2109 * find our cached RPC replies when we get around to reconnecting.
2111 static void xs_abort_connection(struct sock_xprt *transport)
2114 struct sockaddr any;
2116 dprintk("RPC: disconnecting xprt %p to reuse port\n", transport);
2119 * Disconnect the transport socket by doing a connect operation
2120 * with AF_UNSPEC. This should return immediately...
2122 memset(&any, 0, sizeof(any));
2123 any.sa_family = AF_UNSPEC;
2124 result = kernel_connect(transport->sock, &any, sizeof(any), 0);
2125 trace_rpc_socket_reset_connection(&transport->xprt,
2126 transport->sock, result);
2128 xs_sock_reset_connection_flags(&transport->xprt);
2129 dprintk("RPC: AF_UNSPEC connect return code %d\n", result);
2132 static void xs_tcp_reuse_connection(struct sock_xprt *transport)
2134 unsigned int state = transport->inet->sk_state;
2136 if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED) {
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: TCP_CLOSEd and sk_shutdown set to %d\n",
2143 __func__, transport->inet->sk_shutdown);
2145 if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT)) {
2146 /* we don't need to abort the connection if the socket
2147 * hasn't undergone a shutdown
2149 if (transport->inet->sk_shutdown == 0)
2151 dprintk("RPC: %s: ESTABLISHED/SYN_SENT "
2152 "sk_shutdown set to %d\n",
2153 __func__, transport->inet->sk_shutdown);
2155 xs_abort_connection(transport);
2158 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2160 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2161 int ret = -ENOTCONN;
2163 if (!transport->inet) {
2164 struct sock *sk = sock->sk;
2165 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2166 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2167 unsigned int opt_on = 1;
2169 /* TCP Keepalive options */
2170 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2171 (char *)&opt_on, sizeof(opt_on));
2172 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2173 (char *)&keepidle, sizeof(keepidle));
2174 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2175 (char *)&keepidle, sizeof(keepidle));
2176 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2177 (char *)&keepcnt, sizeof(keepcnt));
2179 write_lock_bh(&sk->sk_callback_lock);
2181 xs_save_old_callbacks(transport, sk);
2183 sk->sk_user_data = xprt;
2184 sk->sk_data_ready = xs_tcp_data_ready;
2185 sk->sk_state_change = xs_tcp_state_change;
2186 sk->sk_write_space = xs_tcp_write_space;
2187 sk->sk_error_report = xs_error_report;
2188 sk->sk_allocation = GFP_ATOMIC;
2190 /* socket options */
2191 sock_reset_flag(sk, SOCK_LINGER);
2192 tcp_sk(sk)->linger2 = 0;
2193 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2195 xprt_clear_connected(xprt);
2197 /* Reset to new socket */
2198 transport->sock = sock;
2199 transport->inet = sk;
2201 write_unlock_bh(&sk->sk_callback_lock);
2204 if (!xprt_bound(xprt))
2207 xs_set_memalloc(xprt);
2209 /* Tell the socket layer to start connecting... */
2210 xprt->stat.connect_count++;
2211 xprt->stat.connect_start = jiffies;
2212 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2215 xs_set_srcport(transport, sock);
2218 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2219 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2226 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2227 * @xprt: RPC transport to connect
2228 * @transport: socket transport to connect
2229 * @create_sock: function to create a socket of the correct type
2231 * Invoked by a work queue tasklet.
2233 static void xs_tcp_setup_socket(struct work_struct *work)
2235 struct sock_xprt *transport =
2236 container_of(work, struct sock_xprt, connect_worker.work);
2237 struct socket *sock = transport->sock;
2238 struct rpc_xprt *xprt = &transport->xprt;
2242 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
2243 sock = xs_create_sock(xprt, transport,
2244 xs_addr(xprt)->sa_family, SOCK_STREAM,
2247 status = PTR_ERR(sock);
2253 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
2255 /* "close" the socket, preserving the local port */
2256 set_bit(XPRT_CONNECTION_REUSE, &xprt->state);
2257 xs_tcp_reuse_connection(transport);
2258 clear_bit(XPRT_CONNECTION_REUSE, &xprt->state);
2264 dprintk("RPC: worker connecting xprt %p via %s to "
2265 "%s (port %s)\n", xprt,
2266 xprt->address_strings[RPC_DISPLAY_PROTO],
2267 xprt->address_strings[RPC_DISPLAY_ADDR],
2268 xprt->address_strings[RPC_DISPLAY_PORT]);
2270 status = xs_tcp_finish_connecting(xprt, sock);
2271 trace_rpc_socket_connect(xprt, sock, status);
2272 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2273 xprt, -status, xprt_connected(xprt),
2274 sock->sk->sk_state);
2277 printk("%s: connect returned unhandled error %d\n",
2279 case -EADDRNOTAVAIL:
2280 /* We're probably in TIME_WAIT. Get rid of existing socket,
2283 xs_tcp_force_close(xprt);
2288 xprt_unlock_connect(xprt, transport);
2289 xprt_clear_connecting(xprt);
2292 /* Happens, for instance, if the user specified a link
2293 * local IPv6 address without a scope-id.
2300 /* retry with existing socket, after a delay */
2306 xprt_unlock_connect(xprt, transport);
2307 xprt_clear_connecting(xprt);
2308 xprt_wake_pending_tasks(xprt, status);
2312 * xs_connect - connect a socket to a remote endpoint
2313 * @xprt: pointer to transport structure
2314 * @task: address of RPC task that manages state of connect request
2316 * TCP: If the remote end dropped the connection, delay reconnecting.
2318 * UDP socket connects are synchronous, but we use a work queue anyway
2319 * to guarantee that even unprivileged user processes can set up a
2320 * socket on a privileged port.
2322 * If a UDP socket connect fails, the delay behavior here prevents
2323 * retry floods (hard mounts).
2325 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2327 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2329 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2331 if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
2332 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2334 xprt, xprt->reestablish_timeout / HZ);
2335 queue_delayed_work(rpciod_workqueue,
2336 &transport->connect_worker,
2337 xprt->reestablish_timeout);
2338 xprt->reestablish_timeout <<= 1;
2339 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2340 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2341 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2342 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2344 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2345 queue_delayed_work(rpciod_workqueue,
2346 &transport->connect_worker, 0);
2351 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2352 * @xprt: rpc_xprt struct containing statistics
2356 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2360 if (xprt_connected(xprt))
2361 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2363 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2364 "%llu %llu %lu %llu %llu\n",
2365 xprt->stat.bind_count,
2366 xprt->stat.connect_count,
2367 xprt->stat.connect_time,
2371 xprt->stat.bad_xids,
2374 xprt->stat.max_slots,
2375 xprt->stat.sending_u,
2376 xprt->stat.pending_u);
2380 * xs_udp_print_stats - display UDP socket-specifc stats
2381 * @xprt: rpc_xprt struct containing statistics
2385 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2387 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2389 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2392 xprt->stat.bind_count,
2395 xprt->stat.bad_xids,
2398 xprt->stat.max_slots,
2399 xprt->stat.sending_u,
2400 xprt->stat.pending_u);
2404 * xs_tcp_print_stats - display TCP socket-specifc stats
2405 * @xprt: rpc_xprt struct containing statistics
2409 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2411 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2414 if (xprt_connected(xprt))
2415 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2417 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2418 "%llu %llu %lu %llu %llu\n",
2420 xprt->stat.bind_count,
2421 xprt->stat.connect_count,
2422 xprt->stat.connect_time,
2426 xprt->stat.bad_xids,
2429 xprt->stat.max_slots,
2430 xprt->stat.sending_u,
2431 xprt->stat.pending_u);
2435 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2436 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2437 * to use the server side send routines.
2439 static void *bc_malloc(struct rpc_task *task, size_t size)
2442 struct rpc_buffer *buf;
2444 WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2445 if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2448 page = alloc_page(GFP_KERNEL);
2452 buf = page_address(page);
2453 buf->len = PAGE_SIZE;
2459 * Free the space allocated in the bc_alloc routine
2461 static void bc_free(void *buffer)
2463 struct rpc_buffer *buf;
2468 buf = container_of(buffer, struct rpc_buffer, data);
2469 free_page((unsigned long)buf);
2473 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2474 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2476 static int bc_sendto(struct rpc_rqst *req)
2479 struct xdr_buf *xbufp = &req->rq_snd_buf;
2480 struct rpc_xprt *xprt = req->rq_xprt;
2481 struct sock_xprt *transport =
2482 container_of(xprt, struct sock_xprt, xprt);
2483 struct socket *sock = transport->sock;
2484 unsigned long headoff;
2485 unsigned long tailoff;
2487 xs_encode_stream_record_marker(xbufp);
2489 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2490 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2491 len = svc_send_common(sock, xbufp,
2492 virt_to_page(xbufp->head[0].iov_base), headoff,
2493 xbufp->tail[0].iov_base, tailoff);
2495 if (len != xbufp->len) {
2496 printk(KERN_NOTICE "Error sending entire callback!\n");
2504 * The send routine. Borrows from svc_send
2506 static int bc_send_request(struct rpc_task *task)
2508 struct rpc_rqst *req = task->tk_rqstp;
2509 struct svc_xprt *xprt;
2512 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2514 * Get the server socket associated with this callback xprt
2516 xprt = req->rq_xprt->bc_xprt;
2519 * Grab the mutex to serialize data as the connection is shared
2520 * with the fore channel
2522 if (!mutex_trylock(&xprt->xpt_mutex)) {
2523 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2524 if (!mutex_trylock(&xprt->xpt_mutex))
2526 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2528 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2531 len = bc_sendto(req);
2532 mutex_unlock(&xprt->xpt_mutex);
2541 * The close routine. Since this is client initiated, we do nothing
2544 static void bc_close(struct rpc_xprt *xprt)
2549 * The xprt destroy routine. Again, because this connection is client
2550 * initiated, we do nothing
2553 static void bc_destroy(struct rpc_xprt *xprt)
2555 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2558 module_put(THIS_MODULE);
2561 static struct rpc_xprt_ops xs_local_ops = {
2562 .reserve_xprt = xprt_reserve_xprt,
2563 .release_xprt = xs_tcp_release_xprt,
2564 .alloc_slot = xprt_alloc_slot,
2565 .rpcbind = xs_local_rpcbind,
2566 .set_port = xs_local_set_port,
2567 .connect = xs_local_connect,
2568 .buf_alloc = rpc_malloc,
2569 .buf_free = rpc_free,
2570 .send_request = xs_local_send_request,
2571 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2573 .destroy = xs_destroy,
2574 .print_stats = xs_local_print_stats,
2577 static struct rpc_xprt_ops xs_udp_ops = {
2578 .set_buffer_size = xs_udp_set_buffer_size,
2579 .reserve_xprt = xprt_reserve_xprt_cong,
2580 .release_xprt = xprt_release_xprt_cong,
2581 .alloc_slot = xprt_alloc_slot,
2582 .rpcbind = rpcb_getport_async,
2583 .set_port = xs_set_port,
2584 .connect = xs_connect,
2585 .buf_alloc = rpc_malloc,
2586 .buf_free = rpc_free,
2587 .send_request = xs_udp_send_request,
2588 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
2589 .timer = xs_udp_timer,
2590 .release_request = xprt_release_rqst_cong,
2592 .destroy = xs_destroy,
2593 .print_stats = xs_udp_print_stats,
2596 static struct rpc_xprt_ops xs_tcp_ops = {
2597 .reserve_xprt = xprt_reserve_xprt,
2598 .release_xprt = xs_tcp_release_xprt,
2599 .alloc_slot = xprt_lock_and_alloc_slot,
2600 .rpcbind = rpcb_getport_async,
2601 .set_port = xs_set_port,
2602 .connect = xs_connect,
2603 .buf_alloc = rpc_malloc,
2604 .buf_free = rpc_free,
2605 .send_request = xs_tcp_send_request,
2606 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2607 .close = xs_tcp_close,
2608 .destroy = xs_destroy,
2609 .print_stats = xs_tcp_print_stats,
2613 * The rpc_xprt_ops for the server backchannel
2616 static struct rpc_xprt_ops bc_tcp_ops = {
2617 .reserve_xprt = xprt_reserve_xprt,
2618 .release_xprt = xprt_release_xprt,
2619 .alloc_slot = xprt_alloc_slot,
2620 .buf_alloc = bc_malloc,
2621 .buf_free = bc_free,
2622 .send_request = bc_send_request,
2623 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2625 .destroy = bc_destroy,
2626 .print_stats = xs_tcp_print_stats,
2629 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2631 static const struct sockaddr_in sin = {
2632 .sin_family = AF_INET,
2633 .sin_addr.s_addr = htonl(INADDR_ANY),
2635 static const struct sockaddr_in6 sin6 = {
2636 .sin6_family = AF_INET6,
2637 .sin6_addr = IN6ADDR_ANY_INIT,
2644 memcpy(sap, &sin, sizeof(sin));
2647 memcpy(sap, &sin6, sizeof(sin6));
2650 dprintk("RPC: %s: Bad address family\n", __func__);
2651 return -EAFNOSUPPORT;
2656 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2657 unsigned int slot_table_size,
2658 unsigned int max_slot_table_size)
2660 struct rpc_xprt *xprt;
2661 struct sock_xprt *new;
2663 if (args->addrlen > sizeof(xprt->addr)) {
2664 dprintk("RPC: xs_setup_xprt: address too large\n");
2665 return ERR_PTR(-EBADF);
2668 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2669 max_slot_table_size);
2671 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2673 return ERR_PTR(-ENOMEM);
2676 new = container_of(xprt, struct sock_xprt, xprt);
2677 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2678 xprt->addrlen = args->addrlen;
2680 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2683 err = xs_init_anyaddr(args->dstaddr->sa_family,
2684 (struct sockaddr *)&new->srcaddr);
2687 return ERR_PTR(err);
2694 static const struct rpc_timeout xs_local_default_timeout = {
2695 .to_initval = 10 * HZ,
2696 .to_maxval = 10 * HZ,
2701 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2702 * @args: rpc transport creation arguments
2704 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2706 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2708 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2709 struct sock_xprt *transport;
2710 struct rpc_xprt *xprt;
2711 struct rpc_xprt *ret;
2713 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2714 xprt_max_tcp_slot_table_entries);
2717 transport = container_of(xprt, struct sock_xprt, xprt);
2720 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2721 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2723 xprt->bind_timeout = XS_BIND_TO;
2724 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2725 xprt->idle_timeout = XS_IDLE_DISC_TO;
2727 xprt->ops = &xs_local_ops;
2728 xprt->timeout = &xs_local_default_timeout;
2730 INIT_DELAYED_WORK(&transport->connect_worker,
2731 xs_dummy_setup_socket);
2733 switch (sun->sun_family) {
2735 if (sun->sun_path[0] != '/') {
2736 dprintk("RPC: bad AF_LOCAL address: %s\n",
2738 ret = ERR_PTR(-EINVAL);
2741 xprt_set_bound(xprt);
2742 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2743 ret = ERR_PTR(xs_local_setup_socket(transport));
2748 ret = ERR_PTR(-EAFNOSUPPORT);
2752 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2753 xprt->address_strings[RPC_DISPLAY_ADDR]);
2755 if (try_module_get(THIS_MODULE))
2757 ret = ERR_PTR(-EINVAL);
2763 static const struct rpc_timeout xs_udp_default_timeout = {
2764 .to_initval = 5 * HZ,
2765 .to_maxval = 30 * HZ,
2766 .to_increment = 5 * HZ,
2771 * xs_setup_udp - Set up transport to use a UDP socket
2772 * @args: rpc transport creation arguments
2775 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2777 struct sockaddr *addr = args->dstaddr;
2778 struct rpc_xprt *xprt;
2779 struct sock_xprt *transport;
2780 struct rpc_xprt *ret;
2782 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2783 xprt_udp_slot_table_entries);
2786 transport = container_of(xprt, struct sock_xprt, xprt);
2788 xprt->prot = IPPROTO_UDP;
2790 /* XXX: header size can vary due to auth type, IPv6, etc. */
2791 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2793 xprt->bind_timeout = XS_BIND_TO;
2794 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2795 xprt->idle_timeout = XS_IDLE_DISC_TO;
2797 xprt->ops = &xs_udp_ops;
2799 xprt->timeout = &xs_udp_default_timeout;
2801 switch (addr->sa_family) {
2803 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2804 xprt_set_bound(xprt);
2806 INIT_DELAYED_WORK(&transport->connect_worker,
2807 xs_udp_setup_socket);
2808 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2811 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2812 xprt_set_bound(xprt);
2814 INIT_DELAYED_WORK(&transport->connect_worker,
2815 xs_udp_setup_socket);
2816 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2819 ret = ERR_PTR(-EAFNOSUPPORT);
2823 if (xprt_bound(xprt))
2824 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2825 xprt->address_strings[RPC_DISPLAY_ADDR],
2826 xprt->address_strings[RPC_DISPLAY_PORT],
2827 xprt->address_strings[RPC_DISPLAY_PROTO]);
2829 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2830 xprt->address_strings[RPC_DISPLAY_ADDR],
2831 xprt->address_strings[RPC_DISPLAY_PROTO]);
2833 if (try_module_get(THIS_MODULE))
2835 ret = ERR_PTR(-EINVAL);
2841 static const struct rpc_timeout xs_tcp_default_timeout = {
2842 .to_initval = 60 * HZ,
2843 .to_maxval = 60 * HZ,
2848 * xs_setup_tcp - Set up transport to use a TCP socket
2849 * @args: rpc transport creation arguments
2852 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2854 struct sockaddr *addr = args->dstaddr;
2855 struct rpc_xprt *xprt;
2856 struct sock_xprt *transport;
2857 struct rpc_xprt *ret;
2858 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2860 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2861 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2863 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2864 max_slot_table_size);
2867 transport = container_of(xprt, struct sock_xprt, xprt);
2869 xprt->prot = IPPROTO_TCP;
2870 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2871 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2873 xprt->bind_timeout = XS_BIND_TO;
2874 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2875 xprt->idle_timeout = XS_IDLE_DISC_TO;
2877 xprt->ops = &xs_tcp_ops;
2878 xprt->timeout = &xs_tcp_default_timeout;
2880 switch (addr->sa_family) {
2882 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2883 xprt_set_bound(xprt);
2885 INIT_DELAYED_WORK(&transport->connect_worker,
2886 xs_tcp_setup_socket);
2887 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2890 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2891 xprt_set_bound(xprt);
2893 INIT_DELAYED_WORK(&transport->connect_worker,
2894 xs_tcp_setup_socket);
2895 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2898 ret = ERR_PTR(-EAFNOSUPPORT);
2902 if (xprt_bound(xprt))
2903 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2904 xprt->address_strings[RPC_DISPLAY_ADDR],
2905 xprt->address_strings[RPC_DISPLAY_PORT],
2906 xprt->address_strings[RPC_DISPLAY_PROTO]);
2908 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2909 xprt->address_strings[RPC_DISPLAY_ADDR],
2910 xprt->address_strings[RPC_DISPLAY_PROTO]);
2912 if (try_module_get(THIS_MODULE))
2914 ret = ERR_PTR(-EINVAL);
2921 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2922 * @args: rpc transport creation arguments
2925 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2927 struct sockaddr *addr = args->dstaddr;
2928 struct rpc_xprt *xprt;
2929 struct sock_xprt *transport;
2930 struct svc_sock *bc_sock;
2931 struct rpc_xprt *ret;
2933 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2934 xprt_tcp_slot_table_entries);
2937 transport = container_of(xprt, struct sock_xprt, xprt);
2939 xprt->prot = IPPROTO_TCP;
2940 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2941 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2942 xprt->timeout = &xs_tcp_default_timeout;
2945 xprt_set_bound(xprt);
2946 xprt->bind_timeout = 0;
2947 xprt->reestablish_timeout = 0;
2948 xprt->idle_timeout = 0;
2950 xprt->ops = &bc_tcp_ops;
2952 switch (addr->sa_family) {
2954 xs_format_peer_addresses(xprt, "tcp",
2958 xs_format_peer_addresses(xprt, "tcp",
2959 RPCBIND_NETID_TCP6);
2962 ret = ERR_PTR(-EAFNOSUPPORT);
2966 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2967 xprt->address_strings[RPC_DISPLAY_ADDR],
2968 xprt->address_strings[RPC_DISPLAY_PORT],
2969 xprt->address_strings[RPC_DISPLAY_PROTO]);
2972 * Once we've associated a backchannel xprt with a connection,
2973 * we want to keep it around as long as the connection lasts,
2974 * in case we need to start using it for a backchannel again;
2975 * this reference won't be dropped until bc_xprt is destroyed.
2978 args->bc_xprt->xpt_bc_xprt = xprt;
2979 xprt->bc_xprt = args->bc_xprt;
2980 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2981 transport->sock = bc_sock->sk_sock;
2982 transport->inet = bc_sock->sk_sk;
2985 * Since we don't want connections for the backchannel, we set
2986 * the xprt status to connected
2988 xprt_set_connected(xprt);
2990 if (try_module_get(THIS_MODULE))
2993 args->bc_xprt->xpt_bc_xprt = NULL;
2995 ret = ERR_PTR(-EINVAL);
3001 static struct xprt_class xs_local_transport = {
3002 .list = LIST_HEAD_INIT(xs_local_transport.list),
3003 .name = "named UNIX socket",
3004 .owner = THIS_MODULE,
3005 .ident = XPRT_TRANSPORT_LOCAL,
3006 .setup = xs_setup_local,
3009 static struct xprt_class xs_udp_transport = {
3010 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3012 .owner = THIS_MODULE,
3013 .ident = XPRT_TRANSPORT_UDP,
3014 .setup = xs_setup_udp,
3017 static struct xprt_class xs_tcp_transport = {
3018 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3020 .owner = THIS_MODULE,
3021 .ident = XPRT_TRANSPORT_TCP,
3022 .setup = xs_setup_tcp,
3025 static struct xprt_class xs_bc_tcp_transport = {
3026 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3027 .name = "tcp NFSv4.1 backchannel",
3028 .owner = THIS_MODULE,
3029 .ident = XPRT_TRANSPORT_BC_TCP,
3030 .setup = xs_setup_bc_tcp,
3034 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3037 int init_socket_xprt(void)
3039 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3040 if (!sunrpc_table_header)
3041 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3044 xprt_register_transport(&xs_local_transport);
3045 xprt_register_transport(&xs_udp_transport);
3046 xprt_register_transport(&xs_tcp_transport);
3047 xprt_register_transport(&xs_bc_tcp_transport);
3053 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3056 void cleanup_socket_xprt(void)
3058 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3059 if (sunrpc_table_header) {
3060 unregister_sysctl_table(sunrpc_table_header);
3061 sunrpc_table_header = NULL;
3065 xprt_unregister_transport(&xs_local_transport);
3066 xprt_unregister_transport(&xs_udp_transport);
3067 xprt_unregister_transport(&xs_tcp_transport);
3068 xprt_unregister_transport(&xs_bc_tcp_transport);
3071 static int param_set_uint_minmax(const char *val,
3072 const struct kernel_param *kp,
3073 unsigned int min, unsigned int max)
3080 ret = kstrtouint(val, 0, &num);
3081 if (ret == -EINVAL || num < min || num > max)
3083 *((unsigned int *)kp->arg) = num;
3087 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3089 return param_set_uint_minmax(val, kp,
3094 static struct kernel_param_ops param_ops_portnr = {
3095 .set = param_set_portnr,
3096 .get = param_get_uint,
3099 #define param_check_portnr(name, p) \
3100 __param_check(name, p, unsigned int);
3102 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3103 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3105 static int param_set_slot_table_size(const char *val,
3106 const struct kernel_param *kp)
3108 return param_set_uint_minmax(val, kp,
3110 RPC_MAX_SLOT_TABLE);
3113 static struct kernel_param_ops param_ops_slot_table_size = {
3114 .set = param_set_slot_table_size,
3115 .get = param_get_uint,
3118 #define param_check_slot_table_size(name, p) \
3119 __param_check(name, p, unsigned int);
3121 static int param_set_max_slot_table_size(const char *val,
3122 const struct kernel_param *kp)
3124 return param_set_uint_minmax(val, kp,
3126 RPC_MAX_SLOT_TABLE_LIMIT);
3129 static struct kernel_param_ops param_ops_max_slot_table_size = {
3130 .set = param_set_max_slot_table_size,
3131 .get = param_get_uint,
3134 #define param_check_max_slot_table_size(name, p) \
3135 __param_check(name, p, unsigned int);
3137 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3138 slot_table_size, 0644);
3139 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3140 max_slot_table_size, 0644);
3141 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3142 slot_table_size, 0644);