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 cancel_delayed_work_sync(&transport->connect_worker);
857 xs_reset_transport(transport);
858 xprt->reestablish_timeout = 0;
860 xprt_disconnect_done(xprt);
863 static void xs_tcp_close(struct rpc_xprt *xprt)
865 if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state))
868 xs_tcp_shutdown(xprt);
871 static void xs_xprt_free(struct rpc_xprt *xprt)
873 xs_free_peer_addresses(xprt);
878 * xs_destroy - prepare to shutdown a transport
879 * @xprt: doomed transport
882 static void xs_destroy(struct rpc_xprt *xprt)
884 dprintk("RPC: xs_destroy xprt %p\n", xprt);
888 module_put(THIS_MODULE);
891 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
893 struct xdr_skb_reader desc = {
895 .offset = sizeof(rpc_fraghdr),
896 .count = skb->len - sizeof(rpc_fraghdr),
899 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
907 * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets
908 * @sk: socket with data to read
909 * @len: how much data to read
911 * Currently this assumes we can read the whole reply in a single gulp.
913 static void xs_local_data_ready(struct sock *sk)
915 struct rpc_task *task;
916 struct rpc_xprt *xprt;
917 struct rpc_rqst *rovr;
919 int err, repsize, copied;
923 read_lock_bh(&sk->sk_callback_lock);
924 dprintk("RPC: %s...\n", __func__);
925 xprt = xprt_from_sock(sk);
929 skb = skb_recv_datagram(sk, 0, 1, &err);
933 repsize = skb->len - sizeof(rpc_fraghdr);
935 dprintk("RPC: impossible RPC reply size %d\n", repsize);
939 /* Copy the XID from the skb... */
940 xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
944 /* Look up and lock the request corresponding to the given XID */
945 spin_lock(&xprt->transport_lock);
946 rovr = xprt_lookup_rqst(xprt, *xp);
949 task = rovr->rq_task;
951 copied = rovr->rq_private_buf.buflen;
952 if (copied > repsize)
955 if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
956 dprintk("RPC: sk_buff copy failed\n");
960 xprt_complete_rqst(task, copied);
963 spin_unlock(&xprt->transport_lock);
965 skb_free_datagram(sk, skb);
967 read_unlock_bh(&sk->sk_callback_lock);
971 * xs_udp_data_ready - "data ready" callback for UDP sockets
972 * @sk: socket with data to read
973 * @len: how much data to read
976 static void xs_udp_data_ready(struct sock *sk)
978 struct rpc_task *task;
979 struct rpc_xprt *xprt;
980 struct rpc_rqst *rovr;
982 int err, repsize, copied;
986 read_lock_bh(&sk->sk_callback_lock);
987 dprintk("RPC: xs_udp_data_ready...\n");
988 if (!(xprt = xprt_from_sock(sk)))
991 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
994 repsize = skb->len - sizeof(struct udphdr);
996 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
1000 /* Copy the XID from the skb... */
1001 xp = skb_header_pointer(skb, sizeof(struct udphdr),
1002 sizeof(_xid), &_xid);
1006 /* Look up and lock the request corresponding to the given XID */
1007 spin_lock(&xprt->transport_lock);
1008 rovr = xprt_lookup_rqst(xprt, *xp);
1011 task = rovr->rq_task;
1013 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1016 /* Suck it into the iovec, verify checksum if not done by hw. */
1017 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1018 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
1022 UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
1024 xprt_adjust_cwnd(xprt, task, copied);
1025 xprt_complete_rqst(task, copied);
1028 spin_unlock(&xprt->transport_lock);
1030 skb_free_datagram(sk, skb);
1032 read_unlock_bh(&sk->sk_callback_lock);
1036 * Helper function to force a TCP close if the server is sending
1037 * junk and/or it has put us in CLOSE_WAIT
1039 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1041 set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1042 xprt_force_disconnect(xprt);
1045 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1047 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1051 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1052 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1053 used = xdr_skb_read_bits(desc, p, len);
1054 transport->tcp_offset += used;
1058 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1059 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1060 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1062 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1063 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1065 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1066 transport->tcp_offset = 0;
1068 /* Sanity check of the record length */
1069 if (unlikely(transport->tcp_reclen < 8)) {
1070 dprintk("RPC: invalid TCP record fragment length\n");
1071 xs_tcp_force_close(xprt);
1074 dprintk("RPC: reading TCP record fragment of length %d\n",
1075 transport->tcp_reclen);
1078 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1080 if (transport->tcp_offset == transport->tcp_reclen) {
1081 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1082 transport->tcp_offset = 0;
1083 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1084 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1085 transport->tcp_flags |= TCP_RCV_COPY_XID;
1086 transport->tcp_copied = 0;
1091 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1096 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1097 dprintk("RPC: reading XID (%Zu bytes)\n", len);
1098 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1099 used = xdr_skb_read_bits(desc, p, len);
1100 transport->tcp_offset += used;
1103 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1104 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1105 transport->tcp_copied = 4;
1106 dprintk("RPC: reading %s XID %08x\n",
1107 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1109 ntohl(transport->tcp_xid));
1110 xs_tcp_check_fraghdr(transport);
1113 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1114 struct xdr_skb_reader *desc)
1121 * We want transport->tcp_offset to be 8 at the end of this routine
1122 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1123 * When this function is called for the first time,
1124 * transport->tcp_offset is 4 (after having already read the xid).
1126 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1127 len = sizeof(transport->tcp_calldir) - offset;
1128 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len);
1129 p = ((char *) &transport->tcp_calldir) + offset;
1130 used = xdr_skb_read_bits(desc, p, len);
1131 transport->tcp_offset += used;
1134 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1136 * We don't yet have the XDR buffer, so we will write the calldir
1137 * out after we get the buffer from the 'struct rpc_rqst'
1139 switch (ntohl(transport->tcp_calldir)) {
1141 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1142 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1143 transport->tcp_flags |= TCP_RPC_REPLY;
1146 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1147 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1148 transport->tcp_flags &= ~TCP_RPC_REPLY;
1151 dprintk("RPC: invalid request message type\n");
1152 xs_tcp_force_close(&transport->xprt);
1154 xs_tcp_check_fraghdr(transport);
1157 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1158 struct xdr_skb_reader *desc,
1159 struct rpc_rqst *req)
1161 struct sock_xprt *transport =
1162 container_of(xprt, struct sock_xprt, xprt);
1163 struct xdr_buf *rcvbuf;
1167 rcvbuf = &req->rq_private_buf;
1169 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1171 * Save the RPC direction in the XDR buffer
1173 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1174 &transport->tcp_calldir,
1175 sizeof(transport->tcp_calldir));
1176 transport->tcp_copied += sizeof(transport->tcp_calldir);
1177 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1181 if (len > transport->tcp_reclen - transport->tcp_offset) {
1182 struct xdr_skb_reader my_desc;
1184 len = transport->tcp_reclen - transport->tcp_offset;
1185 memcpy(&my_desc, desc, sizeof(my_desc));
1186 my_desc.count = len;
1187 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1188 &my_desc, xdr_skb_read_bits);
1192 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1193 desc, xdr_skb_read_bits);
1196 transport->tcp_copied += r;
1197 transport->tcp_offset += r;
1200 /* Error when copying to the receive buffer,
1201 * usually because we weren't able to allocate
1202 * additional buffer pages. All we can do now
1203 * is turn off TCP_RCV_COPY_DATA, so the request
1204 * will not receive any additional updates,
1206 * Any remaining data from this record will
1209 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1210 dprintk("RPC: XID %08x truncated request\n",
1211 ntohl(transport->tcp_xid));
1212 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1213 "tcp_offset = %u, tcp_reclen = %u\n",
1214 xprt, transport->tcp_copied,
1215 transport->tcp_offset, transport->tcp_reclen);
1219 dprintk("RPC: XID %08x read %Zd bytes\n",
1220 ntohl(transport->tcp_xid), r);
1221 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1222 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1223 transport->tcp_offset, transport->tcp_reclen);
1225 if (transport->tcp_copied == req->rq_private_buf.buflen)
1226 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1227 else if (transport->tcp_offset == transport->tcp_reclen) {
1228 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1229 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1234 * Finds the request corresponding to the RPC xid and invokes the common
1235 * tcp read code to read the data.
1237 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1238 struct xdr_skb_reader *desc)
1240 struct sock_xprt *transport =
1241 container_of(xprt, struct sock_xprt, xprt);
1242 struct rpc_rqst *req;
1244 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1246 /* Find and lock the request corresponding to this xid */
1247 spin_lock(&xprt->transport_lock);
1248 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1250 dprintk("RPC: XID %08x request not found!\n",
1251 ntohl(transport->tcp_xid));
1252 spin_unlock(&xprt->transport_lock);
1256 xs_tcp_read_common(xprt, desc, req);
1258 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1259 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1261 spin_unlock(&xprt->transport_lock);
1265 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1267 * Obtains an rpc_rqst previously allocated and invokes the common
1268 * tcp read code to read the data. The result is placed in the callback
1270 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1271 * connection and return -1.
1273 static int xs_tcp_read_callback(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 /* Look up and lock the request corresponding to the given XID */
1281 spin_lock(&xprt->transport_lock);
1282 req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
1284 spin_unlock(&xprt->transport_lock);
1285 printk(KERN_WARNING "Callback slot table overflowed\n");
1286 xprt_force_disconnect(xprt);
1290 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1291 xs_tcp_read_common(xprt, desc, req);
1293 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1294 xprt_complete_bc_request(req, transport->tcp_copied);
1295 spin_unlock(&xprt->transport_lock);
1300 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1301 struct xdr_skb_reader *desc)
1303 struct sock_xprt *transport =
1304 container_of(xprt, struct sock_xprt, xprt);
1306 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1307 xs_tcp_read_reply(xprt, desc) :
1308 xs_tcp_read_callback(xprt, desc);
1311 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1312 struct xdr_skb_reader *desc)
1314 return xs_tcp_read_reply(xprt, desc);
1316 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1319 * Read data off the transport. This can be either an RPC_CALL or an
1320 * RPC_REPLY. Relay the processing to helper functions.
1322 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1323 struct xdr_skb_reader *desc)
1325 struct sock_xprt *transport =
1326 container_of(xprt, struct sock_xprt, xprt);
1328 if (_xs_tcp_read_data(xprt, desc) == 0)
1329 xs_tcp_check_fraghdr(transport);
1332 * The transport_lock protects the request handling.
1333 * There's no need to hold it to update the tcp_flags.
1335 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1339 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1343 len = transport->tcp_reclen - transport->tcp_offset;
1344 if (len > desc->count)
1347 desc->offset += len;
1348 transport->tcp_offset += len;
1349 dprintk("RPC: discarded %Zu bytes\n", len);
1350 xs_tcp_check_fraghdr(transport);
1353 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1355 struct rpc_xprt *xprt = rd_desc->arg.data;
1356 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1357 struct xdr_skb_reader desc = {
1363 dprintk("RPC: xs_tcp_data_recv started\n");
1365 trace_xs_tcp_data_recv(transport);
1366 /* Read in a new fragment marker if necessary */
1367 /* Can we ever really expect to get completely empty fragments? */
1368 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1369 xs_tcp_read_fraghdr(xprt, &desc);
1372 /* Read in the xid if necessary */
1373 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1374 xs_tcp_read_xid(transport, &desc);
1377 /* Read in the call/reply flag */
1378 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1379 xs_tcp_read_calldir(transport, &desc);
1382 /* Read in the request data */
1383 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1384 xs_tcp_read_data(xprt, &desc);
1387 /* Skip over any trailing bytes on short reads */
1388 xs_tcp_read_discard(transport, &desc);
1389 } while (desc.count);
1390 trace_xs_tcp_data_recv(transport);
1391 dprintk("RPC: xs_tcp_data_recv done\n");
1392 return len - desc.count;
1396 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1397 * @sk: socket with data to read
1398 * @bytes: how much data to read
1401 static void xs_tcp_data_ready(struct sock *sk)
1403 struct rpc_xprt *xprt;
1404 read_descriptor_t rd_desc;
1406 unsigned long total = 0;
1408 dprintk("RPC: xs_tcp_data_ready...\n");
1410 read_lock_bh(&sk->sk_callback_lock);
1411 if (!(xprt = xprt_from_sock(sk))) {
1415 /* Any data means we had a useful conversation, so
1416 * the we don't need to delay the next reconnect
1418 if (xprt->reestablish_timeout)
1419 xprt->reestablish_timeout = 0;
1421 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1422 rd_desc.arg.data = xprt;
1424 rd_desc.count = 65536;
1425 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1430 trace_xs_tcp_data_ready(xprt, read, total);
1431 read_unlock_bh(&sk->sk_callback_lock);
1435 * Do the equivalent of linger/linger2 handling for dealing with
1436 * broken servers that don't close the socket in a timely
1439 static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1440 unsigned long timeout)
1442 struct sock_xprt *transport;
1444 if (xprt_test_and_set_connecting(xprt))
1446 set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1447 transport = container_of(xprt, struct sock_xprt, xprt);
1448 queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1452 static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1454 struct sock_xprt *transport;
1456 transport = container_of(xprt, struct sock_xprt, xprt);
1458 if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1459 !cancel_delayed_work(&transport->connect_worker))
1461 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1462 xprt_clear_connecting(xprt);
1465 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1467 xs_sock_reset_connection_flags(xprt);
1468 /* Mark transport as closed and wake up all pending tasks */
1469 xprt_disconnect_done(xprt);
1473 * xs_tcp_state_change - callback to handle TCP socket state changes
1474 * @sk: socket whose state has changed
1477 static void xs_tcp_state_change(struct sock *sk)
1479 struct rpc_xprt *xprt;
1481 read_lock_bh(&sk->sk_callback_lock);
1482 if (!(xprt = xprt_from_sock(sk)))
1484 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1485 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1486 sk->sk_state, xprt_connected(xprt),
1487 sock_flag(sk, SOCK_DEAD),
1488 sock_flag(sk, SOCK_ZAPPED),
1491 trace_rpc_socket_state_change(xprt, sk->sk_socket);
1492 switch (sk->sk_state) {
1493 case TCP_ESTABLISHED:
1494 spin_lock(&xprt->transport_lock);
1495 if (!xprt_test_and_set_connected(xprt)) {
1496 struct sock_xprt *transport = container_of(xprt,
1497 struct sock_xprt, xprt);
1499 /* Reset TCP record info */
1500 transport->tcp_offset = 0;
1501 transport->tcp_reclen = 0;
1502 transport->tcp_copied = 0;
1503 transport->tcp_flags =
1504 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1505 xprt->connect_cookie++;
1507 xprt_wake_pending_tasks(xprt, -EAGAIN);
1509 spin_unlock(&xprt->transport_lock);
1512 /* The client initiated a shutdown of the socket */
1513 xprt->connect_cookie++;
1514 xprt->reestablish_timeout = 0;
1515 set_bit(XPRT_CLOSING, &xprt->state);
1516 smp_mb__before_atomic();
1517 clear_bit(XPRT_CONNECTED, &xprt->state);
1518 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1519 smp_mb__after_atomic();
1520 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1522 case TCP_CLOSE_WAIT:
1523 /* The server initiated a shutdown of the socket */
1524 xprt->connect_cookie++;
1525 clear_bit(XPRT_CONNECTED, &xprt->state);
1526 xs_tcp_force_close(xprt);
1529 * If the server closed down the connection, make sure that
1530 * we back off before reconnecting
1532 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1533 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1536 set_bit(XPRT_CLOSING, &xprt->state);
1537 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1538 smp_mb__before_atomic();
1539 clear_bit(XPRT_CONNECTED, &xprt->state);
1540 smp_mb__after_atomic();
1543 xs_tcp_cancel_linger_timeout(xprt);
1544 xs_sock_mark_closed(xprt);
1547 read_unlock_bh(&sk->sk_callback_lock);
1550 static void xs_write_space(struct sock *sk)
1552 struct socket *sock;
1553 struct rpc_xprt *xprt;
1555 if (unlikely(!(sock = sk->sk_socket)))
1557 clear_bit(SOCK_NOSPACE, &sock->flags);
1559 if (unlikely(!(xprt = xprt_from_sock(sk))))
1561 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1564 xprt_write_space(xprt);
1568 * xs_udp_write_space - callback invoked when socket buffer space
1570 * @sk: socket whose state has changed
1572 * Called when more output buffer space is available for this socket.
1573 * We try not to wake our writers until they can make "significant"
1574 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1575 * with a bunch of small requests.
1577 static void xs_udp_write_space(struct sock *sk)
1579 read_lock_bh(&sk->sk_callback_lock);
1581 /* from net/core/sock.c:sock_def_write_space */
1582 if (sock_writeable(sk))
1585 read_unlock_bh(&sk->sk_callback_lock);
1589 * xs_tcp_write_space - callback invoked when socket buffer space
1591 * @sk: socket whose state has changed
1593 * Called when more output buffer space is available for this socket.
1594 * We try not to wake our writers until they can make "significant"
1595 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1596 * with a bunch of small requests.
1598 static void xs_tcp_write_space(struct sock *sk)
1600 read_lock_bh(&sk->sk_callback_lock);
1602 /* from net/core/stream.c:sk_stream_write_space */
1603 if (sk_stream_is_writeable(sk))
1606 read_unlock_bh(&sk->sk_callback_lock);
1609 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1611 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1612 struct sock *sk = transport->inet;
1614 if (transport->rcvsize) {
1615 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1616 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1618 if (transport->sndsize) {
1619 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1620 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1621 sk->sk_write_space(sk);
1626 * xs_udp_set_buffer_size - set send and receive limits
1627 * @xprt: generic transport
1628 * @sndsize: requested size of send buffer, in bytes
1629 * @rcvsize: requested size of receive buffer, in bytes
1631 * Set socket send and receive buffer size limits.
1633 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1635 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1637 transport->sndsize = 0;
1639 transport->sndsize = sndsize + 1024;
1640 transport->rcvsize = 0;
1642 transport->rcvsize = rcvsize + 1024;
1644 xs_udp_do_set_buffer_size(xprt);
1648 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1649 * @task: task that timed out
1651 * Adjust the congestion window after a retransmit timeout has occurred.
1653 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1655 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1658 static unsigned short xs_get_random_port(void)
1660 unsigned short range = xprt_max_resvport - xprt_min_resvport;
1661 unsigned short rand = (unsigned short) prandom_u32() % range;
1662 return rand + xprt_min_resvport;
1666 * xs_set_reuseaddr_port - set the socket's port and address reuse options
1669 * Note that this function has to be called on all sockets that share the
1670 * same port, and it must be called before binding.
1672 static void xs_sock_set_reuseport(struct socket *sock)
1676 kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT, &opt, sizeof(opt));
1679 static unsigned short xs_sock_getport(struct socket *sock)
1681 struct sockaddr_storage buf;
1683 unsigned short port = 0;
1685 if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0)
1687 switch (buf.ss_family) {
1689 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1692 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1699 * xs_set_port - reset the port number in the remote endpoint address
1700 * @xprt: generic transport
1701 * @port: new port number
1704 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1706 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1708 rpc_set_port(xs_addr(xprt), port);
1709 xs_update_peer_port(xprt);
1712 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1714 if (transport->srcport == 0)
1715 transport->srcport = xs_sock_getport(sock);
1718 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1720 unsigned short port = transport->srcport;
1722 if (port == 0 && transport->xprt.resvport)
1723 port = xs_get_random_port();
1727 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1729 if (transport->srcport != 0)
1730 transport->srcport = 0;
1731 if (!transport->xprt.resvport)
1733 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1734 return xprt_max_resvport;
1737 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1739 struct sockaddr_storage myaddr;
1741 unsigned short port = xs_get_srcport(transport);
1742 unsigned short last;
1745 * If we are asking for any ephemeral port (i.e. port == 0 &&
1746 * transport->xprt.resvport == 0), don't bind. Let the local
1747 * port selection happen implicitly when the socket is used
1748 * (for example at connect time).
1750 * This ensures that we can continue to establish TCP
1751 * connections even when all local ephemeral ports are already
1752 * a part of some TCP connection. This makes no difference
1753 * for UDP sockets, but also doens't harm them.
1755 * If we're asking for any reserved port (i.e. port == 0 &&
1756 * transport->xprt.resvport == 1) xs_get_srcport above will
1757 * ensure that port is non-zero and we will bind as needed.
1762 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1764 rpc_set_port((struct sockaddr *)&myaddr, port);
1765 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1766 transport->xprt.addrlen);
1768 transport->srcport = port;
1772 port = xs_next_srcport(transport, port);
1775 } while (err == -EADDRINUSE && nloop != 2);
1777 if (myaddr.ss_family == AF_INET)
1778 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1779 &((struct sockaddr_in *)&myaddr)->sin_addr,
1780 port, err ? "failed" : "ok", err);
1782 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1783 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1784 port, err ? "failed" : "ok", err);
1789 * We don't support autobind on AF_LOCAL sockets
1791 static void xs_local_rpcbind(struct rpc_task *task)
1794 xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt));
1798 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1802 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1803 static struct lock_class_key xs_key[2];
1804 static struct lock_class_key xs_slock_key[2];
1806 static inline void xs_reclassify_socketu(struct socket *sock)
1808 struct sock *sk = sock->sk;
1810 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1811 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1814 static inline void xs_reclassify_socket4(struct socket *sock)
1816 struct sock *sk = sock->sk;
1818 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1819 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1822 static inline void xs_reclassify_socket6(struct socket *sock)
1824 struct sock *sk = sock->sk;
1826 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1827 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1830 static inline void xs_reclassify_socket(int family, struct socket *sock)
1832 WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1833 if (sock_owned_by_user(sock->sk))
1838 xs_reclassify_socketu(sock);
1841 xs_reclassify_socket4(sock);
1844 xs_reclassify_socket6(sock);
1849 static inline void xs_reclassify_socketu(struct socket *sock)
1853 static inline void xs_reclassify_socket4(struct socket *sock)
1857 static inline void xs_reclassify_socket6(struct socket *sock)
1861 static inline void xs_reclassify_socket(int family, struct socket *sock)
1866 static void xs_dummy_setup_socket(struct work_struct *work)
1870 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1871 struct sock_xprt *transport, int family, int type,
1872 int protocol, bool reuseport)
1874 struct socket *sock;
1877 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1879 dprintk("RPC: can't create %d transport socket (%d).\n",
1883 xs_reclassify_socket(family, sock);
1886 xs_sock_set_reuseport(sock);
1888 err = xs_bind(transport, sock);
1896 return ERR_PTR(err);
1899 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1900 struct socket *sock)
1902 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1905 if (!transport->inet) {
1906 struct sock *sk = sock->sk;
1908 write_lock_bh(&sk->sk_callback_lock);
1910 xs_save_old_callbacks(transport, sk);
1912 sk->sk_user_data = xprt;
1913 sk->sk_data_ready = xs_local_data_ready;
1914 sk->sk_write_space = xs_udp_write_space;
1915 sk->sk_error_report = xs_error_report;
1916 sk->sk_allocation = GFP_ATOMIC;
1918 xprt_clear_connected(xprt);
1920 /* Reset to new socket */
1921 transport->sock = sock;
1922 transport->inet = sk;
1924 write_unlock_bh(&sk->sk_callback_lock);
1927 /* Tell the socket layer to start connecting... */
1928 xprt->stat.connect_count++;
1929 xprt->stat.connect_start = jiffies;
1930 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1934 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1935 * @xprt: RPC transport to connect
1936 * @transport: socket transport to connect
1937 * @create_sock: function to create a socket of the correct type
1939 static int xs_local_setup_socket(struct sock_xprt *transport)
1941 struct rpc_xprt *xprt = &transport->xprt;
1942 struct socket *sock;
1945 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1946 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1947 SOCK_STREAM, 0, &sock, 1);
1949 dprintk("RPC: can't create AF_LOCAL "
1950 "transport socket (%d).\n", -status);
1953 xs_reclassify_socketu(sock);
1955 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1956 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1958 status = xs_local_finish_connecting(xprt, sock);
1959 trace_rpc_socket_connect(xprt, sock, status);
1962 dprintk("RPC: xprt %p connected to %s\n",
1963 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1964 xprt_set_connected(xprt);
1968 dprintk("RPC: xprt %p: socket %s does not exist\n",
1969 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1972 dprintk("RPC: xprt %p: connection refused for %s\n",
1973 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1976 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1978 xprt->address_strings[RPC_DISPLAY_ADDR]);
1982 xprt_clear_connecting(xprt);
1983 xprt_wake_pending_tasks(xprt, status);
1987 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1989 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1992 if (RPC_IS_ASYNC(task)) {
1994 * We want the AF_LOCAL connect to be resolved in the
1995 * filesystem namespace of the process making the rpc
1996 * call. Thus we connect synchronously.
1998 * If we want to support asynchronous AF_LOCAL calls,
1999 * we'll need to figure out how to pass a namespace to
2002 rpc_exit(task, -ENOTCONN);
2005 ret = xs_local_setup_socket(transport);
2006 if (ret && !RPC_IS_SOFTCONN(task))
2007 msleep_interruptible(15000);
2010 #ifdef CONFIG_SUNRPC_SWAP
2011 static void xs_set_memalloc(struct rpc_xprt *xprt)
2013 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2017 sk_set_memalloc(transport->inet);
2021 * xs_swapper - Tag this transport as being used for swap.
2022 * @xprt: transport to tag
2023 * @enable: enable/disable
2026 int xs_swapper(struct rpc_xprt *xprt, int enable)
2028 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2034 xs_set_memalloc(xprt);
2035 } else if (xprt->swapper) {
2037 sk_clear_memalloc(transport->inet);
2042 EXPORT_SYMBOL_GPL(xs_swapper);
2044 static void xs_set_memalloc(struct rpc_xprt *xprt)
2049 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2051 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2053 if (!transport->inet) {
2054 struct sock *sk = sock->sk;
2056 write_lock_bh(&sk->sk_callback_lock);
2058 xs_save_old_callbacks(transport, sk);
2060 sk->sk_user_data = xprt;
2061 sk->sk_data_ready = xs_udp_data_ready;
2062 sk->sk_write_space = xs_udp_write_space;
2063 sk->sk_allocation = GFP_ATOMIC;
2065 xprt_set_connected(xprt);
2067 /* Reset to new socket */
2068 transport->sock = sock;
2069 transport->inet = sk;
2071 xs_set_memalloc(xprt);
2073 write_unlock_bh(&sk->sk_callback_lock);
2075 xs_udp_do_set_buffer_size(xprt);
2078 static void xs_udp_setup_socket(struct work_struct *work)
2080 struct sock_xprt *transport =
2081 container_of(work, struct sock_xprt, connect_worker.work);
2082 struct rpc_xprt *xprt = &transport->xprt;
2083 struct socket *sock = transport->sock;
2086 /* Start by resetting any existing state */
2087 xs_reset_transport(transport);
2088 sock = xs_create_sock(xprt, transport,
2089 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2090 IPPROTO_UDP, false);
2094 dprintk("RPC: worker connecting xprt %p via %s to "
2095 "%s (port %s)\n", xprt,
2096 xprt->address_strings[RPC_DISPLAY_PROTO],
2097 xprt->address_strings[RPC_DISPLAY_ADDR],
2098 xprt->address_strings[RPC_DISPLAY_PORT]);
2100 xs_udp_finish_connecting(xprt, sock);
2101 trace_rpc_socket_connect(xprt, sock, 0);
2104 xprt_clear_connecting(xprt);
2105 xprt_wake_pending_tasks(xprt, status);
2109 * We need to preserve the port number so the reply cache on the server can
2110 * find our cached RPC replies when we get around to reconnecting.
2112 static void xs_abort_connection(struct sock_xprt *transport)
2115 struct sockaddr any;
2117 dprintk("RPC: disconnecting xprt %p to reuse port\n", transport);
2120 * Disconnect the transport socket by doing a connect operation
2121 * with AF_UNSPEC. This should return immediately...
2123 memset(&any, 0, sizeof(any));
2124 any.sa_family = AF_UNSPEC;
2125 result = kernel_connect(transport->sock, &any, sizeof(any), 0);
2126 trace_rpc_socket_reset_connection(&transport->xprt,
2127 transport->sock, result);
2129 xs_sock_reset_connection_flags(&transport->xprt);
2130 dprintk("RPC: AF_UNSPEC connect return code %d\n", result);
2133 static void xs_tcp_reuse_connection(struct sock_xprt *transport)
2135 unsigned int state = transport->inet->sk_state;
2137 if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED) {
2138 /* we don't need to abort the connection if the socket
2139 * hasn't undergone a shutdown
2141 if (transport->inet->sk_shutdown == 0)
2143 dprintk("RPC: %s: TCP_CLOSEd and sk_shutdown set to %d\n",
2144 __func__, transport->inet->sk_shutdown);
2146 if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT)) {
2147 /* we don't need to abort the connection if the socket
2148 * hasn't undergone a shutdown
2150 if (transport->inet->sk_shutdown == 0)
2152 dprintk("RPC: %s: ESTABLISHED/SYN_SENT "
2153 "sk_shutdown set to %d\n",
2154 __func__, transport->inet->sk_shutdown);
2156 xs_abort_connection(transport);
2159 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2161 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2162 int ret = -ENOTCONN;
2164 if (!transport->inet) {
2165 struct sock *sk = sock->sk;
2166 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2167 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2168 unsigned int opt_on = 1;
2170 /* TCP Keepalive options */
2171 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2172 (char *)&opt_on, sizeof(opt_on));
2173 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2174 (char *)&keepidle, sizeof(keepidle));
2175 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2176 (char *)&keepidle, sizeof(keepidle));
2177 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2178 (char *)&keepcnt, sizeof(keepcnt));
2180 write_lock_bh(&sk->sk_callback_lock);
2182 xs_save_old_callbacks(transport, sk);
2184 sk->sk_user_data = xprt;
2185 sk->sk_data_ready = xs_tcp_data_ready;
2186 sk->sk_state_change = xs_tcp_state_change;
2187 sk->sk_write_space = xs_tcp_write_space;
2188 sk->sk_error_report = xs_error_report;
2189 sk->sk_allocation = GFP_ATOMIC;
2191 /* socket options */
2192 sock_reset_flag(sk, SOCK_LINGER);
2193 tcp_sk(sk)->linger2 = 0;
2194 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2196 xprt_clear_connected(xprt);
2198 /* Reset to new socket */
2199 transport->sock = sock;
2200 transport->inet = sk;
2202 write_unlock_bh(&sk->sk_callback_lock);
2205 if (!xprt_bound(xprt))
2208 xs_set_memalloc(xprt);
2210 /* Tell the socket layer to start connecting... */
2211 xprt->stat.connect_count++;
2212 xprt->stat.connect_start = jiffies;
2213 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2216 xs_set_srcport(transport, sock);
2219 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2220 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2227 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2228 * @xprt: RPC transport to connect
2229 * @transport: socket transport to connect
2230 * @create_sock: function to create a socket of the correct type
2232 * Invoked by a work queue tasklet.
2234 static void xs_tcp_setup_socket(struct work_struct *work)
2236 struct sock_xprt *transport =
2237 container_of(work, struct sock_xprt, connect_worker.work);
2238 struct socket *sock = transport->sock;
2239 struct rpc_xprt *xprt = &transport->xprt;
2243 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
2244 sock = xs_create_sock(xprt, transport,
2245 xs_addr(xprt)->sa_family, SOCK_STREAM,
2248 status = PTR_ERR(sock);
2254 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
2256 /* "close" the socket, preserving the local port */
2257 set_bit(XPRT_CONNECTION_REUSE, &xprt->state);
2258 xs_tcp_reuse_connection(transport);
2259 clear_bit(XPRT_CONNECTION_REUSE, &xprt->state);
2265 dprintk("RPC: worker connecting xprt %p via %s to "
2266 "%s (port %s)\n", xprt,
2267 xprt->address_strings[RPC_DISPLAY_PROTO],
2268 xprt->address_strings[RPC_DISPLAY_ADDR],
2269 xprt->address_strings[RPC_DISPLAY_PORT]);
2271 status = xs_tcp_finish_connecting(xprt, sock);
2272 trace_rpc_socket_connect(xprt, sock, status);
2273 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2274 xprt, -status, xprt_connected(xprt),
2275 sock->sk->sk_state);
2278 printk("%s: connect returned unhandled error %d\n",
2280 case -EADDRNOTAVAIL:
2281 /* We're probably in TIME_WAIT. Get rid of existing socket,
2284 xs_tcp_force_close(xprt);
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_clear_connecting(xprt);
2307 xprt_wake_pending_tasks(xprt, status);
2311 * xs_connect - connect a socket to a remote endpoint
2312 * @xprt: pointer to transport structure
2313 * @task: address of RPC task that manages state of connect request
2315 * TCP: If the remote end dropped the connection, delay reconnecting.
2317 * UDP socket connects are synchronous, but we use a work queue anyway
2318 * to guarantee that even unprivileged user processes can set up a
2319 * socket on a privileged port.
2321 * If a UDP socket connect fails, the delay behavior here prevents
2322 * retry floods (hard mounts).
2324 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2326 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2328 if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
2329 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2331 xprt, xprt->reestablish_timeout / HZ);
2332 queue_delayed_work(rpciod_workqueue,
2333 &transport->connect_worker,
2334 xprt->reestablish_timeout);
2335 xprt->reestablish_timeout <<= 1;
2336 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2337 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2338 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2339 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2341 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2342 queue_delayed_work(rpciod_workqueue,
2343 &transport->connect_worker, 0);
2348 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2349 * @xprt: rpc_xprt struct containing statistics
2353 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2357 if (xprt_connected(xprt))
2358 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2360 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2361 "%llu %llu %lu %llu %llu\n",
2362 xprt->stat.bind_count,
2363 xprt->stat.connect_count,
2364 xprt->stat.connect_time,
2368 xprt->stat.bad_xids,
2371 xprt->stat.max_slots,
2372 xprt->stat.sending_u,
2373 xprt->stat.pending_u);
2377 * xs_udp_print_stats - display UDP socket-specifc stats
2378 * @xprt: rpc_xprt struct containing statistics
2382 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2384 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2386 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2389 xprt->stat.bind_count,
2392 xprt->stat.bad_xids,
2395 xprt->stat.max_slots,
2396 xprt->stat.sending_u,
2397 xprt->stat.pending_u);
2401 * xs_tcp_print_stats - display TCP socket-specifc stats
2402 * @xprt: rpc_xprt struct containing statistics
2406 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2408 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2411 if (xprt_connected(xprt))
2412 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2414 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2415 "%llu %llu %lu %llu %llu\n",
2417 xprt->stat.bind_count,
2418 xprt->stat.connect_count,
2419 xprt->stat.connect_time,
2423 xprt->stat.bad_xids,
2426 xprt->stat.max_slots,
2427 xprt->stat.sending_u,
2428 xprt->stat.pending_u);
2432 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2433 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2434 * to use the server side send routines.
2436 static void *bc_malloc(struct rpc_task *task, size_t size)
2439 struct rpc_buffer *buf;
2441 WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2442 if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2445 page = alloc_page(GFP_KERNEL);
2449 buf = page_address(page);
2450 buf->len = PAGE_SIZE;
2456 * Free the space allocated in the bc_alloc routine
2458 static void bc_free(void *buffer)
2460 struct rpc_buffer *buf;
2465 buf = container_of(buffer, struct rpc_buffer, data);
2466 free_page((unsigned long)buf);
2470 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2471 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2473 static int bc_sendto(struct rpc_rqst *req)
2476 struct xdr_buf *xbufp = &req->rq_snd_buf;
2477 struct rpc_xprt *xprt = req->rq_xprt;
2478 struct sock_xprt *transport =
2479 container_of(xprt, struct sock_xprt, xprt);
2480 struct socket *sock = transport->sock;
2481 unsigned long headoff;
2482 unsigned long tailoff;
2484 xs_encode_stream_record_marker(xbufp);
2486 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2487 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2488 len = svc_send_common(sock, xbufp,
2489 virt_to_page(xbufp->head[0].iov_base), headoff,
2490 xbufp->tail[0].iov_base, tailoff);
2492 if (len != xbufp->len) {
2493 printk(KERN_NOTICE "Error sending entire callback!\n");
2501 * The send routine. Borrows from svc_send
2503 static int bc_send_request(struct rpc_task *task)
2505 struct rpc_rqst *req = task->tk_rqstp;
2506 struct svc_xprt *xprt;
2509 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2511 * Get the server socket associated with this callback xprt
2513 xprt = req->rq_xprt->bc_xprt;
2516 * Grab the mutex to serialize data as the connection is shared
2517 * with the fore channel
2519 if (!mutex_trylock(&xprt->xpt_mutex)) {
2520 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2521 if (!mutex_trylock(&xprt->xpt_mutex))
2523 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2525 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2528 len = bc_sendto(req);
2529 mutex_unlock(&xprt->xpt_mutex);
2538 * The close routine. Since this is client initiated, we do nothing
2541 static void bc_close(struct rpc_xprt *xprt)
2546 * The xprt destroy routine. Again, because this connection is client
2547 * initiated, we do nothing
2550 static void bc_destroy(struct rpc_xprt *xprt)
2552 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2555 module_put(THIS_MODULE);
2558 static struct rpc_xprt_ops xs_local_ops = {
2559 .reserve_xprt = xprt_reserve_xprt,
2560 .release_xprt = xs_tcp_release_xprt,
2561 .alloc_slot = xprt_alloc_slot,
2562 .rpcbind = xs_local_rpcbind,
2563 .set_port = xs_local_set_port,
2564 .connect = xs_local_connect,
2565 .buf_alloc = rpc_malloc,
2566 .buf_free = rpc_free,
2567 .send_request = xs_local_send_request,
2568 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2570 .destroy = xs_destroy,
2571 .print_stats = xs_local_print_stats,
2574 static struct rpc_xprt_ops xs_udp_ops = {
2575 .set_buffer_size = xs_udp_set_buffer_size,
2576 .reserve_xprt = xprt_reserve_xprt_cong,
2577 .release_xprt = xprt_release_xprt_cong,
2578 .alloc_slot = xprt_alloc_slot,
2579 .rpcbind = rpcb_getport_async,
2580 .set_port = xs_set_port,
2581 .connect = xs_connect,
2582 .buf_alloc = rpc_malloc,
2583 .buf_free = rpc_free,
2584 .send_request = xs_udp_send_request,
2585 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
2586 .timer = xs_udp_timer,
2587 .release_request = xprt_release_rqst_cong,
2589 .destroy = xs_destroy,
2590 .print_stats = xs_udp_print_stats,
2593 static struct rpc_xprt_ops xs_tcp_ops = {
2594 .reserve_xprt = xprt_reserve_xprt,
2595 .release_xprt = xs_tcp_release_xprt,
2596 .alloc_slot = xprt_lock_and_alloc_slot,
2597 .rpcbind = rpcb_getport_async,
2598 .set_port = xs_set_port,
2599 .connect = xs_connect,
2600 .buf_alloc = rpc_malloc,
2601 .buf_free = rpc_free,
2602 .send_request = xs_tcp_send_request,
2603 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2604 .close = xs_tcp_close,
2605 .destroy = xs_destroy,
2606 .print_stats = xs_tcp_print_stats,
2610 * The rpc_xprt_ops for the server backchannel
2613 static struct rpc_xprt_ops bc_tcp_ops = {
2614 .reserve_xprt = xprt_reserve_xprt,
2615 .release_xprt = xprt_release_xprt,
2616 .alloc_slot = xprt_alloc_slot,
2617 .buf_alloc = bc_malloc,
2618 .buf_free = bc_free,
2619 .send_request = bc_send_request,
2620 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2622 .destroy = bc_destroy,
2623 .print_stats = xs_tcp_print_stats,
2626 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2628 static const struct sockaddr_in sin = {
2629 .sin_family = AF_INET,
2630 .sin_addr.s_addr = htonl(INADDR_ANY),
2632 static const struct sockaddr_in6 sin6 = {
2633 .sin6_family = AF_INET6,
2634 .sin6_addr = IN6ADDR_ANY_INIT,
2641 memcpy(sap, &sin, sizeof(sin));
2644 memcpy(sap, &sin6, sizeof(sin6));
2647 dprintk("RPC: %s: Bad address family\n", __func__);
2648 return -EAFNOSUPPORT;
2653 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2654 unsigned int slot_table_size,
2655 unsigned int max_slot_table_size)
2657 struct rpc_xprt *xprt;
2658 struct sock_xprt *new;
2660 if (args->addrlen > sizeof(xprt->addr)) {
2661 dprintk("RPC: xs_setup_xprt: address too large\n");
2662 return ERR_PTR(-EBADF);
2665 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2666 max_slot_table_size);
2668 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2670 return ERR_PTR(-ENOMEM);
2673 new = container_of(xprt, struct sock_xprt, xprt);
2674 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2675 xprt->addrlen = args->addrlen;
2677 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2680 err = xs_init_anyaddr(args->dstaddr->sa_family,
2681 (struct sockaddr *)&new->srcaddr);
2684 return ERR_PTR(err);
2691 static const struct rpc_timeout xs_local_default_timeout = {
2692 .to_initval = 10 * HZ,
2693 .to_maxval = 10 * HZ,
2698 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2699 * @args: rpc transport creation arguments
2701 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2703 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2705 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2706 struct sock_xprt *transport;
2707 struct rpc_xprt *xprt;
2708 struct rpc_xprt *ret;
2710 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2711 xprt_max_tcp_slot_table_entries);
2714 transport = container_of(xprt, struct sock_xprt, xprt);
2717 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2718 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2720 xprt->bind_timeout = XS_BIND_TO;
2721 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2722 xprt->idle_timeout = XS_IDLE_DISC_TO;
2724 xprt->ops = &xs_local_ops;
2725 xprt->timeout = &xs_local_default_timeout;
2727 INIT_DELAYED_WORK(&transport->connect_worker,
2728 xs_dummy_setup_socket);
2730 switch (sun->sun_family) {
2732 if (sun->sun_path[0] != '/') {
2733 dprintk("RPC: bad AF_LOCAL address: %s\n",
2735 ret = ERR_PTR(-EINVAL);
2738 xprt_set_bound(xprt);
2739 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2740 ret = ERR_PTR(xs_local_setup_socket(transport));
2745 ret = ERR_PTR(-EAFNOSUPPORT);
2749 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2750 xprt->address_strings[RPC_DISPLAY_ADDR]);
2752 if (try_module_get(THIS_MODULE))
2754 ret = ERR_PTR(-EINVAL);
2760 static const struct rpc_timeout xs_udp_default_timeout = {
2761 .to_initval = 5 * HZ,
2762 .to_maxval = 30 * HZ,
2763 .to_increment = 5 * HZ,
2768 * xs_setup_udp - Set up transport to use a UDP socket
2769 * @args: rpc transport creation arguments
2772 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2774 struct sockaddr *addr = args->dstaddr;
2775 struct rpc_xprt *xprt;
2776 struct sock_xprt *transport;
2777 struct rpc_xprt *ret;
2779 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2780 xprt_udp_slot_table_entries);
2783 transport = container_of(xprt, struct sock_xprt, xprt);
2785 xprt->prot = IPPROTO_UDP;
2787 /* XXX: header size can vary due to auth type, IPv6, etc. */
2788 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2790 xprt->bind_timeout = XS_BIND_TO;
2791 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2792 xprt->idle_timeout = XS_IDLE_DISC_TO;
2794 xprt->ops = &xs_udp_ops;
2796 xprt->timeout = &xs_udp_default_timeout;
2798 switch (addr->sa_family) {
2800 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2801 xprt_set_bound(xprt);
2803 INIT_DELAYED_WORK(&transport->connect_worker,
2804 xs_udp_setup_socket);
2805 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2808 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2809 xprt_set_bound(xprt);
2811 INIT_DELAYED_WORK(&transport->connect_worker,
2812 xs_udp_setup_socket);
2813 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2816 ret = ERR_PTR(-EAFNOSUPPORT);
2820 if (xprt_bound(xprt))
2821 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2822 xprt->address_strings[RPC_DISPLAY_ADDR],
2823 xprt->address_strings[RPC_DISPLAY_PORT],
2824 xprt->address_strings[RPC_DISPLAY_PROTO]);
2826 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2827 xprt->address_strings[RPC_DISPLAY_ADDR],
2828 xprt->address_strings[RPC_DISPLAY_PROTO]);
2830 if (try_module_get(THIS_MODULE))
2832 ret = ERR_PTR(-EINVAL);
2838 static const struct rpc_timeout xs_tcp_default_timeout = {
2839 .to_initval = 60 * HZ,
2840 .to_maxval = 60 * HZ,
2845 * xs_setup_tcp - Set up transport to use a TCP socket
2846 * @args: rpc transport creation arguments
2849 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2851 struct sockaddr *addr = args->dstaddr;
2852 struct rpc_xprt *xprt;
2853 struct sock_xprt *transport;
2854 struct rpc_xprt *ret;
2855 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2857 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2858 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2860 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2861 max_slot_table_size);
2864 transport = container_of(xprt, struct sock_xprt, xprt);
2866 xprt->prot = IPPROTO_TCP;
2867 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2868 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2870 xprt->bind_timeout = XS_BIND_TO;
2871 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2872 xprt->idle_timeout = XS_IDLE_DISC_TO;
2874 xprt->ops = &xs_tcp_ops;
2875 xprt->timeout = &xs_tcp_default_timeout;
2877 switch (addr->sa_family) {
2879 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2880 xprt_set_bound(xprt);
2882 INIT_DELAYED_WORK(&transport->connect_worker,
2883 xs_tcp_setup_socket);
2884 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2887 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2888 xprt_set_bound(xprt);
2890 INIT_DELAYED_WORK(&transport->connect_worker,
2891 xs_tcp_setup_socket);
2892 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2895 ret = ERR_PTR(-EAFNOSUPPORT);
2899 if (xprt_bound(xprt))
2900 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2901 xprt->address_strings[RPC_DISPLAY_ADDR],
2902 xprt->address_strings[RPC_DISPLAY_PORT],
2903 xprt->address_strings[RPC_DISPLAY_PROTO]);
2905 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2906 xprt->address_strings[RPC_DISPLAY_ADDR],
2907 xprt->address_strings[RPC_DISPLAY_PROTO]);
2909 if (try_module_get(THIS_MODULE))
2911 ret = ERR_PTR(-EINVAL);
2918 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2919 * @args: rpc transport creation arguments
2922 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2924 struct sockaddr *addr = args->dstaddr;
2925 struct rpc_xprt *xprt;
2926 struct sock_xprt *transport;
2927 struct svc_sock *bc_sock;
2928 struct rpc_xprt *ret;
2930 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2931 xprt_tcp_slot_table_entries);
2934 transport = container_of(xprt, struct sock_xprt, xprt);
2936 xprt->prot = IPPROTO_TCP;
2937 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2938 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2939 xprt->timeout = &xs_tcp_default_timeout;
2942 xprt_set_bound(xprt);
2943 xprt->bind_timeout = 0;
2944 xprt->reestablish_timeout = 0;
2945 xprt->idle_timeout = 0;
2947 xprt->ops = &bc_tcp_ops;
2949 switch (addr->sa_family) {
2951 xs_format_peer_addresses(xprt, "tcp",
2955 xs_format_peer_addresses(xprt, "tcp",
2956 RPCBIND_NETID_TCP6);
2959 ret = ERR_PTR(-EAFNOSUPPORT);
2963 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2964 xprt->address_strings[RPC_DISPLAY_ADDR],
2965 xprt->address_strings[RPC_DISPLAY_PORT],
2966 xprt->address_strings[RPC_DISPLAY_PROTO]);
2969 * Once we've associated a backchannel xprt with a connection,
2970 * we want to keep it around as long as the connection lasts,
2971 * in case we need to start using it for a backchannel again;
2972 * this reference won't be dropped until bc_xprt is destroyed.
2975 args->bc_xprt->xpt_bc_xprt = xprt;
2976 xprt->bc_xprt = args->bc_xprt;
2977 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2978 transport->sock = bc_sock->sk_sock;
2979 transport->inet = bc_sock->sk_sk;
2982 * Since we don't want connections for the backchannel, we set
2983 * the xprt status to connected
2985 xprt_set_connected(xprt);
2987 if (try_module_get(THIS_MODULE))
2990 args->bc_xprt->xpt_bc_xprt = NULL;
2992 ret = ERR_PTR(-EINVAL);
2998 static struct xprt_class xs_local_transport = {
2999 .list = LIST_HEAD_INIT(xs_local_transport.list),
3000 .name = "named UNIX socket",
3001 .owner = THIS_MODULE,
3002 .ident = XPRT_TRANSPORT_LOCAL,
3003 .setup = xs_setup_local,
3006 static struct xprt_class xs_udp_transport = {
3007 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3009 .owner = THIS_MODULE,
3010 .ident = XPRT_TRANSPORT_UDP,
3011 .setup = xs_setup_udp,
3014 static struct xprt_class xs_tcp_transport = {
3015 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3017 .owner = THIS_MODULE,
3018 .ident = XPRT_TRANSPORT_TCP,
3019 .setup = xs_setup_tcp,
3022 static struct xprt_class xs_bc_tcp_transport = {
3023 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3024 .name = "tcp NFSv4.1 backchannel",
3025 .owner = THIS_MODULE,
3026 .ident = XPRT_TRANSPORT_BC_TCP,
3027 .setup = xs_setup_bc_tcp,
3031 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3034 int init_socket_xprt(void)
3036 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3037 if (!sunrpc_table_header)
3038 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3041 xprt_register_transport(&xs_local_transport);
3042 xprt_register_transport(&xs_udp_transport);
3043 xprt_register_transport(&xs_tcp_transport);
3044 xprt_register_transport(&xs_bc_tcp_transport);
3050 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3053 void cleanup_socket_xprt(void)
3055 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3056 if (sunrpc_table_header) {
3057 unregister_sysctl_table(sunrpc_table_header);
3058 sunrpc_table_header = NULL;
3062 xprt_unregister_transport(&xs_local_transport);
3063 xprt_unregister_transport(&xs_udp_transport);
3064 xprt_unregister_transport(&xs_tcp_transport);
3065 xprt_unregister_transport(&xs_bc_tcp_transport);
3068 static int param_set_uint_minmax(const char *val,
3069 const struct kernel_param *kp,
3070 unsigned int min, unsigned int max)
3077 ret = kstrtouint(val, 0, &num);
3078 if (ret == -EINVAL || num < min || num > max)
3080 *((unsigned int *)kp->arg) = num;
3084 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3086 return param_set_uint_minmax(val, kp,
3091 static struct kernel_param_ops param_ops_portnr = {
3092 .set = param_set_portnr,
3093 .get = param_get_uint,
3096 #define param_check_portnr(name, p) \
3097 __param_check(name, p, unsigned int);
3099 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3100 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3102 static int param_set_slot_table_size(const char *val,
3103 const struct kernel_param *kp)
3105 return param_set_uint_minmax(val, kp,
3107 RPC_MAX_SLOT_TABLE);
3110 static struct kernel_param_ops param_ops_slot_table_size = {
3111 .set = param_set_slot_table_size,
3112 .get = param_get_uint,
3115 #define param_check_slot_table_size(name, p) \
3116 __param_check(name, p, unsigned int);
3118 static int param_set_max_slot_table_size(const char *val,
3119 const struct kernel_param *kp)
3121 return param_set_uint_minmax(val, kp,
3123 RPC_MAX_SLOT_TABLE_LIMIT);
3126 static struct kernel_param_ops param_ops_max_slot_table_size = {
3127 .set = param_set_max_slot_table_size,
3128 .get = param_get_uint,
3131 #define param_check_max_slot_table_size(name, p) \
3132 __param_check(name, p, unsigned int);
3134 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3135 slot_table_size, 0644);
3136 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3137 max_slot_table_size, 0644);
3138 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3139 slot_table_size, 0644);