1 // SPDX-License-Identifier: GPL-2.0
3 * linux/net/sunrpc/xprtsock.c
5 * Client-side transport implementation for sockets.
7 * TCP callback races fixes (C) 1998 Red Hat
8 * TCP send fixes (C) 1998 Red Hat
9 * TCP NFS related read + write fixes
10 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
12 * Rewrite of larges part of the code in order to stabilize TCP stuff.
13 * Fix behaviour when socket buffer is full.
14 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
16 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
18 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
19 * <gilles.quillard@bull.net>
22 #include <linux/types.h>
23 #include <linux/string.h>
24 #include <linux/slab.h>
25 #include <linux/module.h>
26 #include <linux/capability.h>
27 #include <linux/pagemap.h>
28 #include <linux/errno.h>
29 #include <linux/socket.h>
31 #include <linux/net.h>
34 #include <linux/udp.h>
35 #include <linux/tcp.h>
36 #include <linux/sunrpc/clnt.h>
37 #include <linux/sunrpc/addr.h>
38 #include <linux/sunrpc/sched.h>
39 #include <linux/sunrpc/svcsock.h>
40 #include <linux/sunrpc/xprtsock.h>
41 #include <linux/file.h>
42 #ifdef CONFIG_SUNRPC_BACKCHANNEL
43 #include <linux/sunrpc/bc_xprt.h>
47 #include <net/checksum.h>
50 #include <linux/bvec.h>
51 #include <linux/highmem.h>
52 #include <linux/uio.h>
53 #include <linux/sched/mm.h>
55 #include <trace/events/sunrpc.h>
60 static void xs_close(struct rpc_xprt *xprt);
61 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
67 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
68 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
69 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
71 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
72 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
74 #define XS_TCP_LINGER_TO (15U * HZ)
75 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
78 * We can register our own files under /proc/sys/sunrpc by
79 * calling register_sysctl_table() again. The files in that
80 * directory become the union of all files registered there.
82 * We simply need to make sure that we don't collide with
83 * someone else's file names!
86 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
87 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
88 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
89 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
90 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
92 static struct ctl_table_header *sunrpc_table_header;
94 static struct xprt_class xs_local_transport;
95 static struct xprt_class xs_udp_transport;
96 static struct xprt_class xs_tcp_transport;
97 static struct xprt_class xs_bc_tcp_transport;
100 * FIXME: changing the UDP slot table size should also resize the UDP
101 * socket buffers for existing UDP transports
103 static struct ctl_table xs_tunables_table[] = {
105 .procname = "udp_slot_table_entries",
106 .data = &xprt_udp_slot_table_entries,
107 .maxlen = sizeof(unsigned int),
109 .proc_handler = proc_dointvec_minmax,
110 .extra1 = &min_slot_table_size,
111 .extra2 = &max_slot_table_size
114 .procname = "tcp_slot_table_entries",
115 .data = &xprt_tcp_slot_table_entries,
116 .maxlen = sizeof(unsigned int),
118 .proc_handler = proc_dointvec_minmax,
119 .extra1 = &min_slot_table_size,
120 .extra2 = &max_slot_table_size
123 .procname = "tcp_max_slot_table_entries",
124 .data = &xprt_max_tcp_slot_table_entries,
125 .maxlen = sizeof(unsigned int),
127 .proc_handler = proc_dointvec_minmax,
128 .extra1 = &min_slot_table_size,
129 .extra2 = &max_tcp_slot_table_limit
132 .procname = "min_resvport",
133 .data = &xprt_min_resvport,
134 .maxlen = sizeof(unsigned int),
136 .proc_handler = proc_dointvec_minmax,
137 .extra1 = &xprt_min_resvport_limit,
138 .extra2 = &xprt_max_resvport_limit
141 .procname = "max_resvport",
142 .data = &xprt_max_resvport,
143 .maxlen = sizeof(unsigned int),
145 .proc_handler = proc_dointvec_minmax,
146 .extra1 = &xprt_min_resvport_limit,
147 .extra2 = &xprt_max_resvport_limit
150 .procname = "tcp_fin_timeout",
151 .data = &xs_tcp_fin_timeout,
152 .maxlen = sizeof(xs_tcp_fin_timeout),
154 .proc_handler = proc_dointvec_jiffies,
159 static struct ctl_table sunrpc_table[] = {
161 .procname = "sunrpc",
163 .child = xs_tunables_table
169 * Wait duration for a reply from the RPC portmapper.
171 #define XS_BIND_TO (60U * HZ)
174 * Delay if a UDP socket connect error occurs. This is most likely some
175 * kind of resource problem on the local host.
177 #define XS_UDP_REEST_TO (2U * HZ)
180 * The reestablish timeout allows clients to delay for a bit before attempting
181 * to reconnect to a server that just dropped our connection.
183 * We implement an exponential backoff when trying to reestablish a TCP
184 * transport connection with the server. Some servers like to drop a TCP
185 * connection when they are overworked, so we start with a short timeout and
186 * increase over time if the server is down or not responding.
188 #define XS_TCP_INIT_REEST_TO (3U * HZ)
191 * TCP idle timeout; client drops the transport socket if it is idle
192 * for this long. Note that we also timeout UDP sockets to prevent
193 * holding port numbers when there is no RPC traffic.
195 #define XS_IDLE_DISC_TO (5U * 60 * HZ)
197 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
198 # undef RPC_DEBUG_DATA
199 # define RPCDBG_FACILITY RPCDBG_TRANS
202 #ifdef RPC_DEBUG_DATA
203 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
205 u8 *buf = (u8 *) packet;
208 dprintk("RPC: %s\n", msg);
209 for (j = 0; j < count && j < 128; j += 4) {
213 dprintk("0x%04x ", j);
215 dprintk("%02x%02x%02x%02x ",
216 buf[j], buf[j+1], buf[j+2], buf[j+3]);
221 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
227 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
229 return (struct rpc_xprt *) sk->sk_user_data;
232 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
234 return (struct sockaddr *) &xprt->addr;
237 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
239 return (struct sockaddr_un *) &xprt->addr;
242 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
244 return (struct sockaddr_in *) &xprt->addr;
247 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
249 return (struct sockaddr_in6 *) &xprt->addr;
252 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
254 struct sockaddr *sap = xs_addr(xprt);
255 struct sockaddr_in6 *sin6;
256 struct sockaddr_in *sin;
257 struct sockaddr_un *sun;
260 switch (sap->sa_family) {
262 sun = xs_addr_un(xprt);
263 strlcpy(buf, sun->sun_path, sizeof(buf));
264 xprt->address_strings[RPC_DISPLAY_ADDR] =
265 kstrdup(buf, GFP_KERNEL);
268 (void)rpc_ntop(sap, buf, sizeof(buf));
269 xprt->address_strings[RPC_DISPLAY_ADDR] =
270 kstrdup(buf, GFP_KERNEL);
271 sin = xs_addr_in(xprt);
272 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
275 (void)rpc_ntop(sap, buf, sizeof(buf));
276 xprt->address_strings[RPC_DISPLAY_ADDR] =
277 kstrdup(buf, GFP_KERNEL);
278 sin6 = xs_addr_in6(xprt);
279 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
285 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
288 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
290 struct sockaddr *sap = xs_addr(xprt);
293 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
294 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
296 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
297 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
300 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
301 const char *protocol,
304 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
305 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
306 xs_format_common_peer_addresses(xprt);
307 xs_format_common_peer_ports(xprt);
310 static void xs_update_peer_port(struct rpc_xprt *xprt)
312 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
313 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
315 xs_format_common_peer_ports(xprt);
318 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
322 for (i = 0; i < RPC_DISPLAY_MAX; i++)
324 case RPC_DISPLAY_PROTO:
325 case RPC_DISPLAY_NETID:
328 kfree(xprt->address_strings[i]);
333 xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp)
337 if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES))
339 n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT;
340 for (i = 0; i < n; i++) {
343 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp);
344 if (!buf->pages[i]) {
346 return i > buf->page_base ? i - buf->page_base : 0;
353 xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek)
357 iov_iter_advance(&msg->msg_iter, seek);
358 ret = sock_recvmsg(sock, msg, flags);
359 return ret > 0 ? ret + seek : ret;
363 xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags,
364 struct kvec *kvec, size_t count, size_t seek)
366 iov_iter_kvec(&msg->msg_iter, READ, kvec, 1, count);
367 return xs_sock_recvmsg(sock, msg, flags, seek);
371 xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags,
372 struct bio_vec *bvec, unsigned long nr, size_t count,
375 iov_iter_bvec(&msg->msg_iter, READ, bvec, nr, count);
376 return xs_sock_recvmsg(sock, msg, flags, seek);
380 xs_read_discard(struct socket *sock, struct msghdr *msg, int flags,
383 iov_iter_discard(&msg->msg_iter, READ, count);
384 return sock_recvmsg(sock, msg, flags);
387 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
389 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
391 struct bvec_iter bi = {
396 bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
397 for_each_bvec(bv, bvec, bi, bi)
398 flush_dcache_page(bv.bv_page);
402 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
408 xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
409 struct xdr_buf *buf, size_t count, size_t seek, size_t *read)
411 size_t want, seek_init = seek, offset = 0;
414 want = min_t(size_t, count, buf->head[0].iov_len);
416 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek);
420 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
430 want = xs_alloc_sparse_pages(buf,
431 min_t(size_t, count - offset, buf->page_len),
434 ret = xs_read_bvec(sock, msg, flags, buf->bvec,
435 xdr_buf_pagecount(buf),
436 want + buf->page_base,
437 seek + buf->page_base);
440 xs_flush_bvec(buf->bvec, ret, seek + buf->page_base);
441 ret -= buf->page_base;
443 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
453 want = min_t(size_t, count - offset, buf->tail[0].iov_len);
455 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek);
459 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
463 } else if (offset < seek_init)
467 *read = offset - seek_init;
475 xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf)
477 if (!transport->recv.copied) {
478 if (buf->head[0].iov_len >= transport->recv.offset)
479 memcpy(buf->head[0].iov_base,
480 &transport->recv.xid,
481 transport->recv.offset);
482 transport->recv.copied = transport->recv.offset;
487 xs_read_stream_request_done(struct sock_xprt *transport)
489 return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT);
493 xs_read_stream_check_eor(struct sock_xprt *transport,
496 if (xs_read_stream_request_done(transport))
497 msg->msg_flags |= MSG_EOR;
501 xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg,
502 int flags, struct rpc_rqst *req)
504 struct xdr_buf *buf = &req->rq_private_buf;
508 xs_read_header(transport, buf);
510 want = transport->recv.len - transport->recv.offset;
512 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf,
513 transport->recv.copied + want,
514 transport->recv.copied,
516 transport->recv.offset += read;
517 transport->recv.copied += read;
520 if (transport->recv.offset == transport->recv.len)
521 xs_read_stream_check_eor(transport, msg);
531 msg->msg_flags |= MSG_TRUNC;
536 return ret < 0 ? ret : read;
540 xs_read_stream_headersize(bool isfrag)
543 return sizeof(__be32);
544 return 3 * sizeof(__be32);
548 xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg,
549 int flags, size_t want, size_t seek)
552 .iov_base = &transport->recv.fraghdr,
555 return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek);
558 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
560 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
562 struct rpc_xprt *xprt = &transport->xprt;
563 struct rpc_rqst *req;
566 /* Is this transport associated with the backchannel? */
570 /* Look up and lock the request corresponding to the given XID */
571 req = xprt_lookup_bc_request(xprt, transport->recv.xid);
573 printk(KERN_WARNING "Callback slot table overflowed\n");
576 if (transport->recv.copied && !req->rq_private_buf.len)
579 ret = xs_read_stream_request(transport, msg, flags, req);
580 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
581 xprt_complete_bc_request(req, transport->recv.copied);
583 req->rq_private_buf.len = transport->recv.copied;
587 #else /* CONFIG_SUNRPC_BACKCHANNEL */
589 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
593 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
596 xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags)
598 struct rpc_xprt *xprt = &transport->xprt;
599 struct rpc_rqst *req;
602 /* Look up and lock the request corresponding to the given XID */
603 spin_lock(&xprt->queue_lock);
604 req = xprt_lookup_rqst(xprt, transport->recv.xid);
605 if (!req || (transport->recv.copied && !req->rq_private_buf.len)) {
606 msg->msg_flags |= MSG_TRUNC;
610 spin_unlock(&xprt->queue_lock);
612 ret = xs_read_stream_request(transport, msg, flags, req);
614 spin_lock(&xprt->queue_lock);
615 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
616 xprt_complete_rqst(req->rq_task, transport->recv.copied);
618 req->rq_private_buf.len = transport->recv.copied;
619 xprt_unpin_rqst(req);
621 spin_unlock(&xprt->queue_lock);
626 xs_read_stream(struct sock_xprt *transport, int flags)
628 struct msghdr msg = { 0 };
629 size_t want, read = 0;
632 if (transport->recv.len == 0) {
633 want = xs_read_stream_headersize(transport->recv.copied != 0);
634 ret = xs_read_stream_header(transport, &msg, flags, want,
635 transport->recv.offset);
638 transport->recv.offset = ret;
639 if (transport->recv.offset != want)
640 return transport->recv.offset;
641 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) &
642 RPC_FRAGMENT_SIZE_MASK;
643 transport->recv.offset -= sizeof(transport->recv.fraghdr);
647 switch (be32_to_cpu(transport->recv.calldir)) {
649 msg.msg_flags |= MSG_TRUNC;
652 ret = xs_read_stream_call(transport, &msg, flags);
655 ret = xs_read_stream_reply(transport, &msg, flags);
657 if (msg.msg_flags & MSG_TRUNC) {
658 transport->recv.calldir = cpu_to_be32(-1);
659 transport->recv.copied = -1;
664 if (transport->recv.offset < transport->recv.len) {
665 if (!(msg.msg_flags & MSG_TRUNC))
668 ret = xs_read_discard(transport->sock, &msg, flags,
669 transport->recv.len - transport->recv.offset);
672 transport->recv.offset += ret;
674 if (transport->recv.offset != transport->recv.len)
677 if (xs_read_stream_request_done(transport)) {
678 trace_xs_stream_read_request(transport);
679 transport->recv.copied = 0;
681 transport->recv.offset = 0;
682 transport->recv.len = 0;
685 return ret != 0 ? ret : -ESHUTDOWN;
688 static __poll_t xs_poll_socket(struct sock_xprt *transport)
690 return transport->sock->ops->poll(transport->file, transport->sock,
694 static bool xs_poll_socket_readable(struct sock_xprt *transport)
696 __poll_t events = xs_poll_socket(transport);
698 return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP);
701 static void xs_poll_check_readable(struct sock_xprt *transport)
704 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
705 if (!xs_poll_socket_readable(transport))
707 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
708 queue_work(xprtiod_workqueue, &transport->recv_worker);
711 static void xs_stream_data_receive(struct sock_xprt *transport)
716 mutex_lock(&transport->recv_mutex);
717 if (transport->sock == NULL)
720 ret = xs_read_stream(transport, MSG_DONTWAIT);
726 if (ret == -ESHUTDOWN)
727 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
729 xs_poll_check_readable(transport);
731 mutex_unlock(&transport->recv_mutex);
732 trace_xs_stream_read_data(&transport->xprt, ret, read);
735 static void xs_stream_data_receive_workfn(struct work_struct *work)
737 struct sock_xprt *transport =
738 container_of(work, struct sock_xprt, recv_worker);
739 unsigned int pflags = memalloc_nofs_save();
741 xs_stream_data_receive(transport);
742 memalloc_nofs_restore(pflags);
746 xs_stream_reset_connect(struct sock_xprt *transport)
748 transport->recv.offset = 0;
749 transport->recv.len = 0;
750 transport->recv.copied = 0;
751 transport->xmit.offset = 0;
755 xs_stream_start_connect(struct sock_xprt *transport)
757 transport->xprt.stat.connect_count++;
758 transport->xprt.stat.connect_start = jiffies;
761 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
764 * xs_nospace - handle transmit was incomplete
765 * @req: pointer to RPC request
766 * @transport: pointer to struct sock_xprt
769 static int xs_nospace(struct rpc_rqst *req, struct sock_xprt *transport)
771 struct rpc_xprt *xprt = &transport->xprt;
772 struct sock *sk = transport->inet;
775 trace_rpc_socket_nospace(req, transport);
777 /* Protect against races with write_space */
778 spin_lock(&xprt->transport_lock);
780 /* Don't race with disconnect */
781 if (xprt_connected(xprt)) {
782 struct socket_wq *wq;
785 wq = rcu_dereference(sk->sk_wq);
786 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
789 /* wait for more buffer space */
790 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
791 sk->sk_write_pending++;
792 xprt_wait_for_buffer_space(xprt);
796 spin_unlock(&xprt->transport_lock);
800 static int xs_sock_nospace(struct rpc_rqst *req)
802 struct sock_xprt *transport =
803 container_of(req->rq_xprt, struct sock_xprt, xprt);
804 struct sock *sk = transport->inet;
808 if (!sock_writeable(sk))
809 ret = xs_nospace(req, transport);
814 static int xs_stream_nospace(struct rpc_rqst *req)
816 struct sock_xprt *transport =
817 container_of(req->rq_xprt, struct sock_xprt, xprt);
818 struct sock *sk = transport->inet;
822 if (!sk_stream_memory_free(sk))
823 ret = xs_nospace(req, transport);
829 xs_stream_prepare_request(struct rpc_rqst *req)
831 xdr_free_bvec(&req->rq_rcv_buf);
832 req->rq_task->tk_status = xdr_alloc_bvec(&req->rq_rcv_buf, GFP_KERNEL);
836 * Determine if the previous message in the stream was aborted before it
837 * could complete transmission.
840 xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req)
842 return transport->xmit.offset != 0 && req->rq_bytes_sent == 0;
846 * Return the stream record marker field for a record of length < 2^31-1
849 xs_stream_record_marker(struct xdr_buf *xdr)
853 return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len);
857 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
858 * @req: pointer to RPC request
861 * 0: The request has been sent
862 * EAGAIN: The socket was blocked, please call again later to
863 * complete the request
864 * ENOTCONN: Caller needs to invoke connect logic then call again
865 * other: Some other error occurred, the request was not sent
867 static int xs_local_send_request(struct rpc_rqst *req)
869 struct rpc_xprt *xprt = req->rq_xprt;
870 struct sock_xprt *transport =
871 container_of(xprt, struct sock_xprt, xprt);
872 struct xdr_buf *xdr = &req->rq_snd_buf;
873 rpc_fraghdr rm = xs_stream_record_marker(xdr);
874 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
875 struct msghdr msg = {
876 .msg_flags = XS_SENDMSG_FLAGS,
881 /* Close the stream if the previous transmission was incomplete */
882 if (xs_send_request_was_aborted(transport, req)) {
883 xprt_force_disconnect(xprt);
887 xs_pktdump("packet data:",
888 req->rq_svec->iov_base, req->rq_svec->iov_len);
890 req->rq_xtime = ktime_get();
891 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
892 transport->xmit.offset, rm, &sent);
893 dprintk("RPC: %s(%u) = %d\n",
894 __func__, xdr->len - transport->xmit.offset, status);
896 if (status == -EAGAIN && sock_writeable(transport->inet))
899 if (likely(sent > 0) || status == 0) {
900 transport->xmit.offset += sent;
901 req->rq_bytes_sent = transport->xmit.offset;
902 if (likely(req->rq_bytes_sent >= msglen)) {
903 req->rq_xmit_bytes_sent += transport->xmit.offset;
904 transport->xmit.offset = 0;
914 status = xs_stream_nospace(req);
917 dprintk("RPC: sendmsg returned unrecognized error %d\n",
921 xprt_force_disconnect(xprt);
929 * xs_udp_send_request - write an RPC request to a UDP socket
930 * @req: pointer to RPC request
933 * 0: The request has been sent
934 * EAGAIN: The socket was blocked, please call again later to
935 * complete the request
936 * ENOTCONN: Caller needs to invoke connect logic then call again
937 * other: Some other error occurred, the request was not sent
939 static int xs_udp_send_request(struct rpc_rqst *req)
941 struct rpc_xprt *xprt = req->rq_xprt;
942 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
943 struct xdr_buf *xdr = &req->rq_snd_buf;
944 struct msghdr msg = {
945 .msg_name = xs_addr(xprt),
946 .msg_namelen = xprt->addrlen,
947 .msg_flags = XS_SENDMSG_FLAGS,
952 xs_pktdump("packet data:",
953 req->rq_svec->iov_base,
954 req->rq_svec->iov_len);
956 if (!xprt_bound(xprt))
959 if (!xprt_request_get_cong(xprt, req))
962 req->rq_xtime = ktime_get();
963 status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, 0, &sent);
965 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
968 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
969 if (status == -EPERM)
972 if (status == -EAGAIN && sock_writeable(transport->inet))
975 if (sent > 0 || status == 0) {
976 req->rq_xmit_bytes_sent += sent;
977 if (sent >= req->rq_slen)
979 /* Still some bytes left; set up for a retry later. */
987 /* Should we call xs_close() here? */
990 status = xs_sock_nospace(req);
997 /* When the server has died, an ICMP port unreachable message
998 * prompts ECONNREFUSED. */
1001 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1009 * xs_tcp_send_request - write an RPC request to a TCP socket
1010 * @req: pointer to RPC request
1013 * 0: The request has been sent
1014 * EAGAIN: The socket was blocked, please call again later to
1015 * complete the request
1016 * ENOTCONN: Caller needs to invoke connect logic then call again
1017 * other: Some other error occurred, the request was not sent
1019 * XXX: In the case of soft timeouts, should we eventually give up
1020 * if sendmsg is not able to make progress?
1022 static int xs_tcp_send_request(struct rpc_rqst *req)
1024 struct rpc_xprt *xprt = req->rq_xprt;
1025 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1026 struct xdr_buf *xdr = &req->rq_snd_buf;
1027 rpc_fraghdr rm = xs_stream_record_marker(xdr);
1028 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
1029 struct msghdr msg = {
1030 .msg_flags = XS_SENDMSG_FLAGS,
1032 bool vm_wait = false;
1036 /* Close the stream if the previous transmission was incomplete */
1037 if (xs_send_request_was_aborted(transport, req)) {
1038 if (transport->sock != NULL)
1039 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
1042 if (!transport->inet)
1045 xs_pktdump("packet data:",
1046 req->rq_svec->iov_base,
1047 req->rq_svec->iov_len);
1049 if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
1050 xs_tcp_set_socket_timeouts(xprt, transport->sock);
1052 /* Continue transmitting the packet/record. We must be careful
1053 * to cope with writespace callbacks arriving _after_ we have
1054 * called sendmsg(). */
1055 req->rq_xtime = ktime_get();
1056 tcp_sock_set_cork(transport->inet, true);
1058 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
1059 transport->xmit.offset, rm, &sent);
1061 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
1062 xdr->len - transport->xmit.offset, status);
1064 /* If we've sent the entire packet, immediately
1065 * reset the count of bytes sent. */
1066 transport->xmit.offset += sent;
1067 req->rq_bytes_sent = transport->xmit.offset;
1068 if (likely(req->rq_bytes_sent >= msglen)) {
1069 req->rq_xmit_bytes_sent += transport->xmit.offset;
1070 transport->xmit.offset = 0;
1071 if (atomic_long_read(&xprt->xmit_queuelen) == 1)
1072 tcp_sock_set_cork(transport->inet, false);
1076 WARN_ON_ONCE(sent == 0 && status == 0);
1078 if (status == -EAGAIN ) {
1080 * Return EAGAIN if we're sure we're hitting the
1081 * socket send buffer limits.
1083 if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
1086 * Did we hit a memory allocation failure?
1092 /* Retry, knowing now that we're below the
1093 * socket send buffer limit
1107 /* Should we call xs_close() here? */
1110 status = xs_stream_nospace(req);
1120 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1127 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1129 transport->old_data_ready = sk->sk_data_ready;
1130 transport->old_state_change = sk->sk_state_change;
1131 transport->old_write_space = sk->sk_write_space;
1132 transport->old_error_report = sk->sk_error_report;
1135 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1137 sk->sk_data_ready = transport->old_data_ready;
1138 sk->sk_state_change = transport->old_state_change;
1139 sk->sk_write_space = transport->old_write_space;
1140 sk->sk_error_report = transport->old_error_report;
1143 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
1145 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1147 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1148 clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state);
1149 clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state);
1150 clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state);
1153 static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr)
1155 set_bit(nr, &transport->sock_state);
1156 queue_work(xprtiod_workqueue, &transport->error_worker);
1159 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1161 smp_mb__before_atomic();
1162 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1163 clear_bit(XPRT_CLOSING, &xprt->state);
1164 xs_sock_reset_state_flags(xprt);
1165 smp_mb__after_atomic();
1169 * xs_error_report - callback to handle TCP socket state errors
1172 * Note: we don't call sock_error() since there may be a rpc_task
1173 * using the socket, and so we don't want to clear sk->sk_err.
1175 static void xs_error_report(struct sock *sk)
1177 struct sock_xprt *transport;
1178 struct rpc_xprt *xprt;
1180 read_lock_bh(&sk->sk_callback_lock);
1181 if (!(xprt = xprt_from_sock(sk)))
1184 transport = container_of(xprt, struct sock_xprt, xprt);
1185 transport->xprt_err = -sk->sk_err;
1186 if (transport->xprt_err == 0)
1188 dprintk("RPC: xs_error_report client %p, error=%d...\n",
1189 xprt, -transport->xprt_err);
1190 trace_rpc_socket_error(xprt, sk->sk_socket, transport->xprt_err);
1192 /* barrier ensures xprt_err is set before XPRT_SOCK_WAKE_ERROR */
1193 smp_mb__before_atomic();
1194 xs_run_error_worker(transport, XPRT_SOCK_WAKE_ERROR);
1196 read_unlock_bh(&sk->sk_callback_lock);
1199 static void xs_reset_transport(struct sock_xprt *transport)
1201 struct socket *sock = transport->sock;
1202 struct sock *sk = transport->inet;
1203 struct rpc_xprt *xprt = &transport->xprt;
1204 struct file *filp = transport->file;
1209 * Make sure we're calling this in a context from which it is safe
1210 * to call __fput_sync(). In practice that means rpciod and the
1213 if (!(current->flags & PF_WQ_WORKER)) {
1215 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
1219 if (atomic_read(&transport->xprt.swapper))
1220 sk_clear_memalloc(sk);
1222 kernel_sock_shutdown(sock, SHUT_RDWR);
1224 mutex_lock(&transport->recv_mutex);
1225 write_lock_bh(&sk->sk_callback_lock);
1226 transport->inet = NULL;
1227 transport->sock = NULL;
1228 transport->file = NULL;
1230 sk->sk_user_data = NULL;
1232 xs_restore_old_callbacks(transport, sk);
1233 xprt_clear_connected(xprt);
1234 write_unlock_bh(&sk->sk_callback_lock);
1235 xs_sock_reset_connection_flags(xprt);
1236 /* Reset stream record info */
1237 xs_stream_reset_connect(transport);
1238 mutex_unlock(&transport->recv_mutex);
1240 trace_rpc_socket_close(xprt, sock);
1243 xprt_disconnect_done(xprt);
1247 * xs_close - close a socket
1250 * This is used when all requests are complete; ie, no DRC state remains
1251 * on the server we want to save.
1253 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1254 * xs_reset_transport() zeroing the socket from underneath a writer.
1256 static void xs_close(struct rpc_xprt *xprt)
1258 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1260 dprintk("RPC: xs_close xprt %p\n", xprt);
1262 xs_reset_transport(transport);
1263 xprt->reestablish_timeout = 0;
1266 static void xs_inject_disconnect(struct rpc_xprt *xprt)
1268 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
1270 xprt_disconnect_done(xprt);
1273 static void xs_xprt_free(struct rpc_xprt *xprt)
1275 xs_free_peer_addresses(xprt);
1280 * xs_destroy - prepare to shutdown a transport
1281 * @xprt: doomed transport
1284 static void xs_destroy(struct rpc_xprt *xprt)
1286 struct sock_xprt *transport = container_of(xprt,
1287 struct sock_xprt, xprt);
1288 dprintk("RPC: xs_destroy xprt %p\n", xprt);
1290 cancel_delayed_work_sync(&transport->connect_worker);
1292 cancel_work_sync(&transport->recv_worker);
1293 cancel_work_sync(&transport->error_worker);
1295 module_put(THIS_MODULE);
1299 * xs_udp_data_read_skb - receive callback for UDP sockets
1305 static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1307 struct sk_buff *skb)
1309 struct rpc_task *task;
1310 struct rpc_rqst *rovr;
1311 int repsize, copied;
1317 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
1321 /* Copy the XID from the skb... */
1322 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1326 /* Look up and lock the request corresponding to the given XID */
1327 spin_lock(&xprt->queue_lock);
1328 rovr = xprt_lookup_rqst(xprt, *xp);
1331 xprt_pin_rqst(rovr);
1332 xprt_update_rtt(rovr->rq_task);
1333 spin_unlock(&xprt->queue_lock);
1334 task = rovr->rq_task;
1336 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1339 /* Suck it into the iovec, verify checksum if not done by hw. */
1340 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1341 spin_lock(&xprt->queue_lock);
1342 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1347 spin_lock(&xprt->transport_lock);
1348 xprt_adjust_cwnd(xprt, task, copied);
1349 spin_unlock(&xprt->transport_lock);
1350 spin_lock(&xprt->queue_lock);
1351 xprt_complete_rqst(task, copied);
1352 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1354 xprt_unpin_rqst(rovr);
1356 spin_unlock(&xprt->queue_lock);
1359 static void xs_udp_data_receive(struct sock_xprt *transport)
1361 struct sk_buff *skb;
1365 mutex_lock(&transport->recv_mutex);
1366 sk = transport->inet;
1370 skb = skb_recv_udp(sk, 0, 1, &err);
1373 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1377 xs_poll_check_readable(transport);
1379 mutex_unlock(&transport->recv_mutex);
1382 static void xs_udp_data_receive_workfn(struct work_struct *work)
1384 struct sock_xprt *transport =
1385 container_of(work, struct sock_xprt, recv_worker);
1386 unsigned int pflags = memalloc_nofs_save();
1388 xs_udp_data_receive(transport);
1389 memalloc_nofs_restore(pflags);
1393 * xs_data_ready - "data ready" callback for UDP sockets
1394 * @sk: socket with data to read
1397 static void xs_data_ready(struct sock *sk)
1399 struct rpc_xprt *xprt;
1401 read_lock_bh(&sk->sk_callback_lock);
1402 dprintk("RPC: xs_data_ready...\n");
1403 xprt = xprt_from_sock(sk);
1405 struct sock_xprt *transport = container_of(xprt,
1406 struct sock_xprt, xprt);
1407 transport->old_data_ready(sk);
1408 /* Any data means we had a useful conversation, so
1409 * then we don't need to delay the next reconnect
1411 if (xprt->reestablish_timeout)
1412 xprt->reestablish_timeout = 0;
1413 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1414 queue_work(xprtiod_workqueue, &transport->recv_worker);
1416 read_unlock_bh(&sk->sk_callback_lock);
1420 * Helper function to force a TCP close if the server is sending
1421 * junk and/or it has put us in CLOSE_WAIT
1423 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1425 xprt_force_disconnect(xprt);
1428 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1429 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1433 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1436 * xs_tcp_state_change - callback to handle TCP socket state changes
1437 * @sk: socket whose state has changed
1440 static void xs_tcp_state_change(struct sock *sk)
1442 struct rpc_xprt *xprt;
1443 struct sock_xprt *transport;
1445 read_lock_bh(&sk->sk_callback_lock);
1446 if (!(xprt = xprt_from_sock(sk)))
1448 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1449 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1450 sk->sk_state, xprt_connected(xprt),
1451 sock_flag(sk, SOCK_DEAD),
1452 sock_flag(sk, SOCK_ZAPPED),
1455 transport = container_of(xprt, struct sock_xprt, xprt);
1456 trace_rpc_socket_state_change(xprt, sk->sk_socket);
1457 switch (sk->sk_state) {
1458 case TCP_ESTABLISHED:
1459 if (!xprt_test_and_set_connected(xprt)) {
1460 xprt->connect_cookie++;
1461 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1462 xprt_clear_connecting(xprt);
1464 xprt->stat.connect_count++;
1465 xprt->stat.connect_time += (long)jiffies -
1466 xprt->stat.connect_start;
1467 xs_run_error_worker(transport, XPRT_SOCK_WAKE_PENDING);
1471 /* The client initiated a shutdown of the socket */
1472 xprt->connect_cookie++;
1473 xprt->reestablish_timeout = 0;
1474 set_bit(XPRT_CLOSING, &xprt->state);
1475 smp_mb__before_atomic();
1476 clear_bit(XPRT_CONNECTED, &xprt->state);
1477 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1478 smp_mb__after_atomic();
1480 case TCP_CLOSE_WAIT:
1481 /* The server initiated a shutdown of the socket */
1482 xprt->connect_cookie++;
1483 clear_bit(XPRT_CONNECTED, &xprt->state);
1484 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1488 * If the server closed down the connection, make sure that
1489 * we back off before reconnecting
1491 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1492 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1495 set_bit(XPRT_CLOSING, &xprt->state);
1496 smp_mb__before_atomic();
1497 clear_bit(XPRT_CONNECTED, &xprt->state);
1498 smp_mb__after_atomic();
1501 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1502 &transport->sock_state))
1503 xprt_clear_connecting(xprt);
1504 clear_bit(XPRT_CLOSING, &xprt->state);
1505 /* Trigger the socket release */
1506 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1509 read_unlock_bh(&sk->sk_callback_lock);
1512 static void xs_write_space(struct sock *sk)
1514 struct socket_wq *wq;
1515 struct sock_xprt *transport;
1516 struct rpc_xprt *xprt;
1520 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1522 if (unlikely(!(xprt = xprt_from_sock(sk))))
1524 transport = container_of(xprt, struct sock_xprt, xprt);
1526 wq = rcu_dereference(sk->sk_wq);
1527 if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1530 xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE);
1531 sk->sk_write_pending--;
1537 * xs_udp_write_space - callback invoked when socket buffer space
1539 * @sk: socket whose state has changed
1541 * Called when more output buffer space is available for this socket.
1542 * We try not to wake our writers until they can make "significant"
1543 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1544 * with a bunch of small requests.
1546 static void xs_udp_write_space(struct sock *sk)
1548 read_lock_bh(&sk->sk_callback_lock);
1550 /* from net/core/sock.c:sock_def_write_space */
1551 if (sock_writeable(sk))
1554 read_unlock_bh(&sk->sk_callback_lock);
1558 * xs_tcp_write_space - callback invoked when socket buffer space
1560 * @sk: socket whose state has changed
1562 * Called when more output buffer space is available for this socket.
1563 * We try not to wake our writers until they can make "significant"
1564 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1565 * with a bunch of small requests.
1567 static void xs_tcp_write_space(struct sock *sk)
1569 read_lock_bh(&sk->sk_callback_lock);
1571 /* from net/core/stream.c:sk_stream_write_space */
1572 if (sk_stream_is_writeable(sk))
1575 read_unlock_bh(&sk->sk_callback_lock);
1578 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1580 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1581 struct sock *sk = transport->inet;
1583 if (transport->rcvsize) {
1584 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1585 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1587 if (transport->sndsize) {
1588 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1589 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1590 sk->sk_write_space(sk);
1595 * xs_udp_set_buffer_size - set send and receive limits
1596 * @xprt: generic transport
1597 * @sndsize: requested size of send buffer, in bytes
1598 * @rcvsize: requested size of receive buffer, in bytes
1600 * Set socket send and receive buffer size limits.
1602 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1604 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1606 transport->sndsize = 0;
1608 transport->sndsize = sndsize + 1024;
1609 transport->rcvsize = 0;
1611 transport->rcvsize = rcvsize + 1024;
1613 xs_udp_do_set_buffer_size(xprt);
1617 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1618 * @xprt: controlling transport
1619 * @task: task that timed out
1621 * Adjust the congestion window after a retransmit timeout has occurred.
1623 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1625 spin_lock(&xprt->transport_lock);
1626 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1627 spin_unlock(&xprt->transport_lock);
1630 static int xs_get_random_port(void)
1632 unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1633 unsigned short range;
1634 unsigned short rand;
1638 range = max - min + 1;
1639 rand = (unsigned short) prandom_u32() % range;
1643 static unsigned short xs_sock_getport(struct socket *sock)
1645 struct sockaddr_storage buf;
1646 unsigned short port = 0;
1648 if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1650 switch (buf.ss_family) {
1652 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1655 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1662 * xs_set_port - reset the port number in the remote endpoint address
1663 * @xprt: generic transport
1664 * @port: new port number
1667 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1669 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1671 rpc_set_port(xs_addr(xprt), port);
1672 xs_update_peer_port(xprt);
1675 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1677 if (transport->srcport == 0 && transport->xprt.reuseport)
1678 transport->srcport = xs_sock_getport(sock);
1681 static int xs_get_srcport(struct sock_xprt *transport)
1683 int port = transport->srcport;
1685 if (port == 0 && transport->xprt.resvport)
1686 port = xs_get_random_port();
1690 unsigned short get_srcport(struct rpc_xprt *xprt)
1692 struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt);
1693 return xs_sock_getport(sock->sock);
1695 EXPORT_SYMBOL(get_srcport);
1697 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1699 if (transport->srcport != 0)
1700 transport->srcport = 0;
1701 if (!transport->xprt.resvport)
1703 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1704 return xprt_max_resvport;
1707 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1709 struct sockaddr_storage myaddr;
1711 int port = xs_get_srcport(transport);
1712 unsigned short last;
1715 * If we are asking for any ephemeral port (i.e. port == 0 &&
1716 * transport->xprt.resvport == 0), don't bind. Let the local
1717 * port selection happen implicitly when the socket is used
1718 * (for example at connect time).
1720 * This ensures that we can continue to establish TCP
1721 * connections even when all local ephemeral ports are already
1722 * a part of some TCP connection. This makes no difference
1723 * for UDP sockets, but also doesn't harm them.
1725 * If we're asking for any reserved port (i.e. port == 0 &&
1726 * transport->xprt.resvport == 1) xs_get_srcport above will
1727 * ensure that port is non-zero and we will bind as needed.
1732 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1734 rpc_set_port((struct sockaddr *)&myaddr, port);
1735 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1736 transport->xprt.addrlen);
1738 if (transport->xprt.reuseport)
1739 transport->srcport = port;
1743 port = xs_next_srcport(transport, port);
1746 } while (err == -EADDRINUSE && nloop != 2);
1748 if (myaddr.ss_family == AF_INET)
1749 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1750 &((struct sockaddr_in *)&myaddr)->sin_addr,
1751 port, err ? "failed" : "ok", err);
1753 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1754 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1755 port, err ? "failed" : "ok", err);
1760 * We don't support autobind on AF_LOCAL sockets
1762 static void xs_local_rpcbind(struct rpc_task *task)
1764 xprt_set_bound(task->tk_xprt);
1767 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1771 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1772 static struct lock_class_key xs_key[2];
1773 static struct lock_class_key xs_slock_key[2];
1775 static inline void xs_reclassify_socketu(struct socket *sock)
1777 struct sock *sk = sock->sk;
1779 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1780 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1783 static inline void xs_reclassify_socket4(struct socket *sock)
1785 struct sock *sk = sock->sk;
1787 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1788 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1791 static inline void xs_reclassify_socket6(struct socket *sock)
1793 struct sock *sk = sock->sk;
1795 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1796 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1799 static inline void xs_reclassify_socket(int family, struct socket *sock)
1801 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1806 xs_reclassify_socketu(sock);
1809 xs_reclassify_socket4(sock);
1812 xs_reclassify_socket6(sock);
1817 static inline void xs_reclassify_socket(int family, struct socket *sock)
1822 static void xs_dummy_setup_socket(struct work_struct *work)
1826 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1827 struct sock_xprt *transport, int family, int type,
1828 int protocol, bool reuseport)
1831 struct socket *sock;
1834 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1836 dprintk("RPC: can't create %d transport socket (%d).\n",
1840 xs_reclassify_socket(family, sock);
1843 sock_set_reuseport(sock->sk);
1845 err = xs_bind(transport, sock);
1851 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1853 return ERR_CAST(filp);
1854 transport->file = filp;
1858 return ERR_PTR(err);
1861 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1862 struct socket *sock)
1864 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1867 if (!transport->inet) {
1868 struct sock *sk = sock->sk;
1870 write_lock_bh(&sk->sk_callback_lock);
1872 xs_save_old_callbacks(transport, sk);
1874 sk->sk_user_data = xprt;
1875 sk->sk_data_ready = xs_data_ready;
1876 sk->sk_write_space = xs_udp_write_space;
1877 sock_set_flag(sk, SOCK_FASYNC);
1878 sk->sk_error_report = xs_error_report;
1880 xprt_clear_connected(xprt);
1882 /* Reset to new socket */
1883 transport->sock = sock;
1884 transport->inet = sk;
1886 write_unlock_bh(&sk->sk_callback_lock);
1889 xs_stream_start_connect(transport);
1891 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1895 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1896 * @transport: socket transport to connect
1898 static int xs_local_setup_socket(struct sock_xprt *transport)
1900 struct rpc_xprt *xprt = &transport->xprt;
1902 struct socket *sock;
1905 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1906 SOCK_STREAM, 0, &sock, 1);
1908 dprintk("RPC: can't create AF_LOCAL "
1909 "transport socket (%d).\n", -status);
1912 xs_reclassify_socket(AF_LOCAL, sock);
1914 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1916 status = PTR_ERR(filp);
1919 transport->file = filp;
1921 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1922 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1924 status = xs_local_finish_connecting(xprt, sock);
1925 trace_rpc_socket_connect(xprt, sock, status);
1928 dprintk("RPC: xprt %p connected to %s\n",
1929 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1930 xprt->stat.connect_count++;
1931 xprt->stat.connect_time += (long)jiffies -
1932 xprt->stat.connect_start;
1933 xprt_set_connected(xprt);
1938 dprintk("RPC: xprt %p: socket %s does not exist\n",
1939 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1942 dprintk("RPC: xprt %p: connection refused for %s\n",
1943 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1946 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1948 xprt->address_strings[RPC_DISPLAY_ADDR]);
1952 xprt_clear_connecting(xprt);
1953 xprt_wake_pending_tasks(xprt, status);
1957 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1959 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1962 if (RPC_IS_ASYNC(task)) {
1964 * We want the AF_LOCAL connect to be resolved in the
1965 * filesystem namespace of the process making the rpc
1966 * call. Thus we connect synchronously.
1968 * If we want to support asynchronous AF_LOCAL calls,
1969 * we'll need to figure out how to pass a namespace to
1972 task->tk_rpc_status = -ENOTCONN;
1973 rpc_exit(task, -ENOTCONN);
1976 ret = xs_local_setup_socket(transport);
1977 if (ret && !RPC_IS_SOFTCONN(task))
1978 msleep_interruptible(15000);
1981 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
1983 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
1984 * know that we have exclusive access to the socket), to guard against
1985 * races with xs_reset_transport.
1987 static void xs_set_memalloc(struct rpc_xprt *xprt)
1989 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1993 * If there's no sock, then we have nothing to set. The
1994 * reconnecting process will get it for us.
1996 if (!transport->inet)
1998 if (atomic_read(&xprt->swapper))
1999 sk_set_memalloc(transport->inet);
2003 * xs_enable_swap - Tag this transport as being used for swap.
2004 * @xprt: transport to tag
2006 * Take a reference to this transport on behalf of the rpc_clnt, and
2007 * optionally mark it for swapping if it wasn't already.
2010 xs_enable_swap(struct rpc_xprt *xprt)
2012 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2014 if (atomic_inc_return(&xprt->swapper) != 1)
2016 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2017 return -ERESTARTSYS;
2019 sk_set_memalloc(xs->inet);
2020 xprt_release_xprt(xprt, NULL);
2025 * xs_disable_swap - Untag this transport as being used for swap.
2026 * @xprt: transport to tag
2028 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2029 * swapper refcount goes to 0, untag the socket as a memalloc socket.
2032 xs_disable_swap(struct rpc_xprt *xprt)
2034 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2036 if (!atomic_dec_and_test(&xprt->swapper))
2038 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2041 sk_clear_memalloc(xs->inet);
2042 xprt_release_xprt(xprt, NULL);
2045 static void xs_set_memalloc(struct rpc_xprt *xprt)
2050 xs_enable_swap(struct rpc_xprt *xprt)
2056 xs_disable_swap(struct rpc_xprt *xprt)
2061 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2063 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2065 if (!transport->inet) {
2066 struct sock *sk = sock->sk;
2068 write_lock_bh(&sk->sk_callback_lock);
2070 xs_save_old_callbacks(transport, sk);
2072 sk->sk_user_data = xprt;
2073 sk->sk_data_ready = xs_data_ready;
2074 sk->sk_write_space = xs_udp_write_space;
2075 sock_set_flag(sk, SOCK_FASYNC);
2077 xprt_set_connected(xprt);
2079 /* Reset to new socket */
2080 transport->sock = sock;
2081 transport->inet = sk;
2083 xs_set_memalloc(xprt);
2085 write_unlock_bh(&sk->sk_callback_lock);
2087 xs_udp_do_set_buffer_size(xprt);
2089 xprt->stat.connect_start = jiffies;
2092 static void xs_udp_setup_socket(struct work_struct *work)
2094 struct sock_xprt *transport =
2095 container_of(work, struct sock_xprt, connect_worker.work);
2096 struct rpc_xprt *xprt = &transport->xprt;
2097 struct socket *sock;
2100 sock = xs_create_sock(xprt, transport,
2101 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2102 IPPROTO_UDP, false);
2106 dprintk("RPC: worker connecting xprt %p via %s to "
2107 "%s (port %s)\n", xprt,
2108 xprt->address_strings[RPC_DISPLAY_PROTO],
2109 xprt->address_strings[RPC_DISPLAY_ADDR],
2110 xprt->address_strings[RPC_DISPLAY_PORT]);
2112 xs_udp_finish_connecting(xprt, sock);
2113 trace_rpc_socket_connect(xprt, sock, 0);
2116 xprt_clear_connecting(xprt);
2117 xprt_unlock_connect(xprt, transport);
2118 xprt_wake_pending_tasks(xprt, status);
2122 * xs_tcp_shutdown - gracefully shut down a TCP socket
2125 * Initiates a graceful shutdown of the TCP socket by calling the
2126 * equivalent of shutdown(SHUT_RDWR);
2128 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2130 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2131 struct socket *sock = transport->sock;
2132 int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2136 if (!xprt->reuseport) {
2144 case TCP_ESTABLISHED:
2145 case TCP_CLOSE_WAIT:
2146 kernel_sock_shutdown(sock, SHUT_RDWR);
2147 trace_rpc_socket_shutdown(xprt, sock);
2150 xs_reset_transport(transport);
2154 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2155 struct socket *sock)
2157 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2158 unsigned int keepidle;
2159 unsigned int keepcnt;
2162 spin_lock(&xprt->transport_lock);
2163 keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2164 keepcnt = xprt->timeout->to_retries + 1;
2165 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2166 (xprt->timeout->to_retries + 1);
2167 clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2168 spin_unlock(&xprt->transport_lock);
2170 /* TCP Keepalive options */
2171 sock_set_keepalive(sock->sk);
2172 tcp_sock_set_keepidle(sock->sk, keepidle);
2173 tcp_sock_set_keepintvl(sock->sk, keepidle);
2174 tcp_sock_set_keepcnt(sock->sk, keepcnt);
2176 /* TCP user timeout (see RFC5482) */
2177 tcp_sock_set_user_timeout(sock->sk, timeo);
2180 static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2181 unsigned long connect_timeout,
2182 unsigned long reconnect_timeout)
2184 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2185 struct rpc_timeout to;
2186 unsigned long initval;
2188 spin_lock(&xprt->transport_lock);
2189 if (reconnect_timeout < xprt->max_reconnect_timeout)
2190 xprt->max_reconnect_timeout = reconnect_timeout;
2191 if (connect_timeout < xprt->connect_timeout) {
2192 memcpy(&to, xprt->timeout, sizeof(to));
2193 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2194 /* Arbitrary lower limit */
2195 if (initval < XS_TCP_INIT_REEST_TO << 1)
2196 initval = XS_TCP_INIT_REEST_TO << 1;
2197 to.to_initval = initval;
2198 to.to_maxval = initval;
2199 memcpy(&transport->tcp_timeout, &to,
2200 sizeof(transport->tcp_timeout));
2201 xprt->timeout = &transport->tcp_timeout;
2202 xprt->connect_timeout = connect_timeout;
2204 set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2205 spin_unlock(&xprt->transport_lock);
2208 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2210 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2211 int ret = -ENOTCONN;
2213 if (!transport->inet) {
2214 struct sock *sk = sock->sk;
2216 /* Avoid temporary address, they are bad for long-lived
2217 * connections such as NFS mounts.
2218 * RFC4941, section 3.6 suggests that:
2219 * Individual applications, which have specific
2220 * knowledge about the normal duration of connections,
2221 * MAY override this as appropriate.
2223 if (xs_addr(xprt)->sa_family == PF_INET6) {
2224 ip6_sock_set_addr_preferences(sk,
2225 IPV6_PREFER_SRC_PUBLIC);
2228 xs_tcp_set_socket_timeouts(xprt, sock);
2229 tcp_sock_set_nodelay(sk);
2231 write_lock_bh(&sk->sk_callback_lock);
2233 xs_save_old_callbacks(transport, sk);
2235 sk->sk_user_data = xprt;
2236 sk->sk_data_ready = xs_data_ready;
2237 sk->sk_state_change = xs_tcp_state_change;
2238 sk->sk_write_space = xs_tcp_write_space;
2239 sock_set_flag(sk, SOCK_FASYNC);
2240 sk->sk_error_report = xs_error_report;
2242 /* socket options */
2243 sock_reset_flag(sk, SOCK_LINGER);
2245 xprt_clear_connected(xprt);
2247 /* Reset to new socket */
2248 transport->sock = sock;
2249 transport->inet = sk;
2251 write_unlock_bh(&sk->sk_callback_lock);
2254 if (!xprt_bound(xprt))
2257 xs_set_memalloc(xprt);
2259 xs_stream_start_connect(transport);
2261 /* Tell the socket layer to start connecting... */
2262 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2263 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2266 xs_set_srcport(transport, sock);
2270 set_bit(XPRT_SOCK_CONNECT_SENT, &transport->sock_state);
2271 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2272 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2274 case -EADDRNOTAVAIL:
2275 /* Source port number is unavailable. Try a new one! */
2276 transport->srcport = 0;
2283 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2284 * @work: queued work item
2286 * Invoked by a work queue tasklet.
2288 static void xs_tcp_setup_socket(struct work_struct *work)
2290 struct sock_xprt *transport =
2291 container_of(work, struct sock_xprt, connect_worker.work);
2292 struct socket *sock = transport->sock;
2293 struct rpc_xprt *xprt = &transport->xprt;
2296 if (xprt_connected(xprt))
2298 if (test_and_clear_bit(XPRT_SOCK_CONNECT_SENT,
2299 &transport->sock_state) ||
2301 xs_reset_transport(transport);
2302 sock = xs_create_sock(xprt, transport, xs_addr(xprt)->sa_family,
2303 SOCK_STREAM, IPPROTO_TCP, true);
2305 status = PTR_ERR(sock);
2310 dprintk("RPC: worker connecting xprt %p via %s to "
2311 "%s (port %s)\n", xprt,
2312 xprt->address_strings[RPC_DISPLAY_PROTO],
2313 xprt->address_strings[RPC_DISPLAY_ADDR],
2314 xprt->address_strings[RPC_DISPLAY_PORT]);
2316 status = xs_tcp_finish_connecting(xprt, sock);
2317 trace_rpc_socket_connect(xprt, sock, status);
2318 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2319 xprt, -status, xprt_connected(xprt),
2320 sock->sk->sk_state);
2323 printk("%s: connect returned unhandled error %d\n",
2326 case -EADDRNOTAVAIL:
2327 /* We're probably in TIME_WAIT. Get rid of existing socket,
2330 xs_tcp_force_close(xprt);
2335 xprt_unlock_connect(xprt, transport);
2338 /* Happens, for instance, if the user specified a link
2339 * local IPv6 address without a scope-id.
2348 /* xs_tcp_force_close() wakes tasks with a fixed error code.
2349 * We need to wake them first to ensure the correct error code.
2351 xprt_wake_pending_tasks(xprt, status);
2352 xs_tcp_force_close(xprt);
2357 xprt_clear_connecting(xprt);
2358 xprt_unlock_connect(xprt, transport);
2359 xprt_wake_pending_tasks(xprt, status);
2363 * xs_connect - connect a socket to a remote endpoint
2364 * @xprt: pointer to transport structure
2365 * @task: address of RPC task that manages state of connect request
2367 * TCP: If the remote end dropped the connection, delay reconnecting.
2369 * UDP socket connects are synchronous, but we use a work queue anyway
2370 * to guarantee that even unprivileged user processes can set up a
2371 * socket on a privileged port.
2373 * If a UDP socket connect fails, the delay behavior here prevents
2374 * retry floods (hard mounts).
2376 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2378 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2379 unsigned long delay = 0;
2381 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2383 if (transport->sock != NULL) {
2384 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2385 "seconds\n", xprt, xprt->reestablish_timeout / HZ);
2387 delay = xprt_reconnect_delay(xprt);
2388 xprt_reconnect_backoff(xprt, XS_TCP_INIT_REEST_TO);
2391 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2393 queue_delayed_work(xprtiod_workqueue,
2394 &transport->connect_worker,
2398 static void xs_wake_disconnect(struct sock_xprt *transport)
2400 if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state))
2401 xs_tcp_force_close(&transport->xprt);
2404 static void xs_wake_write(struct sock_xprt *transport)
2406 if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state))
2407 xprt_write_space(&transport->xprt);
2410 static void xs_wake_error(struct sock_xprt *transport)
2414 if (!test_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2416 mutex_lock(&transport->recv_mutex);
2417 if (transport->sock == NULL)
2419 if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2421 sockerr = xchg(&transport->xprt_err, 0);
2423 xprt_wake_pending_tasks(&transport->xprt, sockerr);
2425 mutex_unlock(&transport->recv_mutex);
2428 static void xs_wake_pending(struct sock_xprt *transport)
2430 if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING, &transport->sock_state))
2431 xprt_wake_pending_tasks(&transport->xprt, -EAGAIN);
2434 static void xs_error_handle(struct work_struct *work)
2436 struct sock_xprt *transport = container_of(work,
2437 struct sock_xprt, error_worker);
2439 xs_wake_disconnect(transport);
2440 xs_wake_write(transport);
2441 xs_wake_error(transport);
2442 xs_wake_pending(transport);
2446 * xs_local_print_stats - display AF_LOCAL socket-specific stats
2447 * @xprt: rpc_xprt struct containing statistics
2451 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2455 if (xprt_connected(xprt))
2456 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2458 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2459 "%llu %llu %lu %llu %llu\n",
2460 xprt->stat.bind_count,
2461 xprt->stat.connect_count,
2462 xprt->stat.connect_time / HZ,
2466 xprt->stat.bad_xids,
2469 xprt->stat.max_slots,
2470 xprt->stat.sending_u,
2471 xprt->stat.pending_u);
2475 * xs_udp_print_stats - display UDP socket-specific stats
2476 * @xprt: rpc_xprt struct containing statistics
2480 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2482 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2484 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2487 xprt->stat.bind_count,
2490 xprt->stat.bad_xids,
2493 xprt->stat.max_slots,
2494 xprt->stat.sending_u,
2495 xprt->stat.pending_u);
2499 * xs_tcp_print_stats - display TCP socket-specific stats
2500 * @xprt: rpc_xprt struct containing statistics
2504 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2506 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2509 if (xprt_connected(xprt))
2510 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2512 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2513 "%llu %llu %lu %llu %llu\n",
2515 xprt->stat.bind_count,
2516 xprt->stat.connect_count,
2517 xprt->stat.connect_time / HZ,
2521 xprt->stat.bad_xids,
2524 xprt->stat.max_slots,
2525 xprt->stat.sending_u,
2526 xprt->stat.pending_u);
2530 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2531 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2532 * to use the server side send routines.
2534 static int bc_malloc(struct rpc_task *task)
2536 struct rpc_rqst *rqst = task->tk_rqstp;
2537 size_t size = rqst->rq_callsize;
2539 struct rpc_buffer *buf;
2541 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2542 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2547 page = alloc_page(GFP_KERNEL);
2551 buf = page_address(page);
2552 buf->len = PAGE_SIZE;
2554 rqst->rq_buffer = buf->data;
2555 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2560 * Free the space allocated in the bc_alloc routine
2562 static void bc_free(struct rpc_task *task)
2564 void *buffer = task->tk_rqstp->rq_buffer;
2565 struct rpc_buffer *buf;
2567 buf = container_of(buffer, struct rpc_buffer, data);
2568 free_page((unsigned long)buf);
2571 static int bc_sendto(struct rpc_rqst *req)
2573 struct xdr_buf *xdr = &req->rq_snd_buf;
2574 struct sock_xprt *transport =
2575 container_of(req->rq_xprt, struct sock_xprt, xprt);
2576 struct msghdr msg = {
2579 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
2581 unsigned int sent = 0;
2584 req->rq_xtime = ktime_get();
2585 err = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, marker, &sent);
2587 if (err < 0 || sent != (xdr->len + sizeof(marker)))
2593 * bc_send_request - Send a backchannel Call on a TCP socket
2594 * @req: rpc_rqst containing Call message to be sent
2596 * xpt_mutex ensures @rqstp's whole message is written to the socket
2597 * without interruption.
2600 * %0 if the message was sent successfully
2601 * %ENOTCONN if the message was not sent
2603 static int bc_send_request(struct rpc_rqst *req)
2605 struct svc_xprt *xprt;
2609 * Get the server socket associated with this callback xprt
2611 xprt = req->rq_xprt->bc_xprt;
2614 * Grab the mutex to serialize data as the connection is shared
2615 * with the fore channel
2617 mutex_lock(&xprt->xpt_mutex);
2618 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2621 len = bc_sendto(req);
2622 mutex_unlock(&xprt->xpt_mutex);
2631 * The close routine. Since this is client initiated, we do nothing
2634 static void bc_close(struct rpc_xprt *xprt)
2636 xprt_disconnect_done(xprt);
2640 * The xprt destroy routine. Again, because this connection is client
2641 * initiated, we do nothing
2644 static void bc_destroy(struct rpc_xprt *xprt)
2646 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2649 module_put(THIS_MODULE);
2652 static const struct rpc_xprt_ops xs_local_ops = {
2653 .reserve_xprt = xprt_reserve_xprt,
2654 .release_xprt = xprt_release_xprt,
2655 .alloc_slot = xprt_alloc_slot,
2656 .free_slot = xprt_free_slot,
2657 .rpcbind = xs_local_rpcbind,
2658 .set_port = xs_local_set_port,
2659 .connect = xs_local_connect,
2660 .buf_alloc = rpc_malloc,
2661 .buf_free = rpc_free,
2662 .prepare_request = xs_stream_prepare_request,
2663 .send_request = xs_local_send_request,
2664 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2666 .destroy = xs_destroy,
2667 .print_stats = xs_local_print_stats,
2668 .enable_swap = xs_enable_swap,
2669 .disable_swap = xs_disable_swap,
2672 static const struct rpc_xprt_ops xs_udp_ops = {
2673 .set_buffer_size = xs_udp_set_buffer_size,
2674 .reserve_xprt = xprt_reserve_xprt_cong,
2675 .release_xprt = xprt_release_xprt_cong,
2676 .alloc_slot = xprt_alloc_slot,
2677 .free_slot = xprt_free_slot,
2678 .rpcbind = rpcb_getport_async,
2679 .set_port = xs_set_port,
2680 .connect = xs_connect,
2681 .buf_alloc = rpc_malloc,
2682 .buf_free = rpc_free,
2683 .send_request = xs_udp_send_request,
2684 .wait_for_reply_request = xprt_wait_for_reply_request_rtt,
2685 .timer = xs_udp_timer,
2686 .release_request = xprt_release_rqst_cong,
2688 .destroy = xs_destroy,
2689 .print_stats = xs_udp_print_stats,
2690 .enable_swap = xs_enable_swap,
2691 .disable_swap = xs_disable_swap,
2692 .inject_disconnect = xs_inject_disconnect,
2695 static const struct rpc_xprt_ops xs_tcp_ops = {
2696 .reserve_xprt = xprt_reserve_xprt,
2697 .release_xprt = xprt_release_xprt,
2698 .alloc_slot = xprt_alloc_slot,
2699 .free_slot = xprt_free_slot,
2700 .rpcbind = rpcb_getport_async,
2701 .set_port = xs_set_port,
2702 .connect = xs_connect,
2703 .buf_alloc = rpc_malloc,
2704 .buf_free = rpc_free,
2705 .prepare_request = xs_stream_prepare_request,
2706 .send_request = xs_tcp_send_request,
2707 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2708 .close = xs_tcp_shutdown,
2709 .destroy = xs_destroy,
2710 .set_connect_timeout = xs_tcp_set_connect_timeout,
2711 .print_stats = xs_tcp_print_stats,
2712 .enable_swap = xs_enable_swap,
2713 .disable_swap = xs_disable_swap,
2714 .inject_disconnect = xs_inject_disconnect,
2715 #ifdef CONFIG_SUNRPC_BACKCHANNEL
2716 .bc_setup = xprt_setup_bc,
2717 .bc_maxpayload = xs_tcp_bc_maxpayload,
2718 .bc_num_slots = xprt_bc_max_slots,
2719 .bc_free_rqst = xprt_free_bc_rqst,
2720 .bc_destroy = xprt_destroy_bc,
2725 * The rpc_xprt_ops for the server backchannel
2728 static const struct rpc_xprt_ops bc_tcp_ops = {
2729 .reserve_xprt = xprt_reserve_xprt,
2730 .release_xprt = xprt_release_xprt,
2731 .alloc_slot = xprt_alloc_slot,
2732 .free_slot = xprt_free_slot,
2733 .buf_alloc = bc_malloc,
2734 .buf_free = bc_free,
2735 .send_request = bc_send_request,
2736 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2738 .destroy = bc_destroy,
2739 .print_stats = xs_tcp_print_stats,
2740 .enable_swap = xs_enable_swap,
2741 .disable_swap = xs_disable_swap,
2742 .inject_disconnect = xs_inject_disconnect,
2745 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2747 static const struct sockaddr_in sin = {
2748 .sin_family = AF_INET,
2749 .sin_addr.s_addr = htonl(INADDR_ANY),
2751 static const struct sockaddr_in6 sin6 = {
2752 .sin6_family = AF_INET6,
2753 .sin6_addr = IN6ADDR_ANY_INIT,
2760 memcpy(sap, &sin, sizeof(sin));
2763 memcpy(sap, &sin6, sizeof(sin6));
2766 dprintk("RPC: %s: Bad address family\n", __func__);
2767 return -EAFNOSUPPORT;
2772 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2773 unsigned int slot_table_size,
2774 unsigned int max_slot_table_size)
2776 struct rpc_xprt *xprt;
2777 struct sock_xprt *new;
2779 if (args->addrlen > sizeof(xprt->addr)) {
2780 dprintk("RPC: xs_setup_xprt: address too large\n");
2781 return ERR_PTR(-EBADF);
2784 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2785 max_slot_table_size);
2787 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2789 return ERR_PTR(-ENOMEM);
2792 new = container_of(xprt, struct sock_xprt, xprt);
2793 mutex_init(&new->recv_mutex);
2794 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2795 xprt->addrlen = args->addrlen;
2797 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2800 err = xs_init_anyaddr(args->dstaddr->sa_family,
2801 (struct sockaddr *)&new->srcaddr);
2804 return ERR_PTR(err);
2811 static const struct rpc_timeout xs_local_default_timeout = {
2812 .to_initval = 10 * HZ,
2813 .to_maxval = 10 * HZ,
2818 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2819 * @args: rpc transport creation arguments
2821 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2823 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2825 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2826 struct sock_xprt *transport;
2827 struct rpc_xprt *xprt;
2828 struct rpc_xprt *ret;
2830 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2831 xprt_max_tcp_slot_table_entries);
2834 transport = container_of(xprt, struct sock_xprt, xprt);
2837 xprt->xprt_class = &xs_local_transport;
2838 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2840 xprt->bind_timeout = XS_BIND_TO;
2841 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2842 xprt->idle_timeout = XS_IDLE_DISC_TO;
2844 xprt->ops = &xs_local_ops;
2845 xprt->timeout = &xs_local_default_timeout;
2847 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2848 INIT_WORK(&transport->error_worker, xs_error_handle);
2849 INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket);
2851 switch (sun->sun_family) {
2853 if (sun->sun_path[0] != '/') {
2854 dprintk("RPC: bad AF_LOCAL address: %s\n",
2856 ret = ERR_PTR(-EINVAL);
2859 xprt_set_bound(xprt);
2860 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2863 ret = ERR_PTR(-EAFNOSUPPORT);
2867 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2868 xprt->address_strings[RPC_DISPLAY_ADDR]);
2870 if (try_module_get(THIS_MODULE))
2872 ret = ERR_PTR(-EINVAL);
2878 static const struct rpc_timeout xs_udp_default_timeout = {
2879 .to_initval = 5 * HZ,
2880 .to_maxval = 30 * HZ,
2881 .to_increment = 5 * HZ,
2886 * xs_setup_udp - Set up transport to use a UDP socket
2887 * @args: rpc transport creation arguments
2890 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2892 struct sockaddr *addr = args->dstaddr;
2893 struct rpc_xprt *xprt;
2894 struct sock_xprt *transport;
2895 struct rpc_xprt *ret;
2897 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2898 xprt_udp_slot_table_entries);
2901 transport = container_of(xprt, struct sock_xprt, xprt);
2903 xprt->prot = IPPROTO_UDP;
2904 xprt->xprt_class = &xs_udp_transport;
2905 /* XXX: header size can vary due to auth type, IPv6, etc. */
2906 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2908 xprt->bind_timeout = XS_BIND_TO;
2909 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2910 xprt->idle_timeout = XS_IDLE_DISC_TO;
2912 xprt->ops = &xs_udp_ops;
2914 xprt->timeout = &xs_udp_default_timeout;
2916 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
2917 INIT_WORK(&transport->error_worker, xs_error_handle);
2918 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2920 switch (addr->sa_family) {
2922 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2923 xprt_set_bound(xprt);
2925 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2928 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2929 xprt_set_bound(xprt);
2931 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2934 ret = ERR_PTR(-EAFNOSUPPORT);
2938 if (xprt_bound(xprt))
2939 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2940 xprt->address_strings[RPC_DISPLAY_ADDR],
2941 xprt->address_strings[RPC_DISPLAY_PORT],
2942 xprt->address_strings[RPC_DISPLAY_PROTO]);
2944 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2945 xprt->address_strings[RPC_DISPLAY_ADDR],
2946 xprt->address_strings[RPC_DISPLAY_PROTO]);
2948 if (try_module_get(THIS_MODULE))
2950 ret = ERR_PTR(-EINVAL);
2956 static const struct rpc_timeout xs_tcp_default_timeout = {
2957 .to_initval = 60 * HZ,
2958 .to_maxval = 60 * HZ,
2963 * xs_setup_tcp - Set up transport to use a TCP socket
2964 * @args: rpc transport creation arguments
2967 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2969 struct sockaddr *addr = args->dstaddr;
2970 struct rpc_xprt *xprt;
2971 struct sock_xprt *transport;
2972 struct rpc_xprt *ret;
2973 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2975 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2976 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2978 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2979 max_slot_table_size);
2982 transport = container_of(xprt, struct sock_xprt, xprt);
2984 xprt->prot = IPPROTO_TCP;
2985 xprt->xprt_class = &xs_tcp_transport;
2986 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2988 xprt->bind_timeout = XS_BIND_TO;
2989 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2990 xprt->idle_timeout = XS_IDLE_DISC_TO;
2992 xprt->ops = &xs_tcp_ops;
2993 xprt->timeout = &xs_tcp_default_timeout;
2995 xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
2996 xprt->connect_timeout = xprt->timeout->to_initval *
2997 (xprt->timeout->to_retries + 1);
2999 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3000 INIT_WORK(&transport->error_worker, xs_error_handle);
3001 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
3003 switch (addr->sa_family) {
3005 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3006 xprt_set_bound(xprt);
3008 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3011 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3012 xprt_set_bound(xprt);
3014 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3017 ret = ERR_PTR(-EAFNOSUPPORT);
3021 if (xprt_bound(xprt))
3022 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3023 xprt->address_strings[RPC_DISPLAY_ADDR],
3024 xprt->address_strings[RPC_DISPLAY_PORT],
3025 xprt->address_strings[RPC_DISPLAY_PROTO]);
3027 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
3028 xprt->address_strings[RPC_DISPLAY_ADDR],
3029 xprt->address_strings[RPC_DISPLAY_PROTO]);
3031 if (try_module_get(THIS_MODULE))
3033 ret = ERR_PTR(-EINVAL);
3040 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3041 * @args: rpc transport creation arguments
3044 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3046 struct sockaddr *addr = args->dstaddr;
3047 struct rpc_xprt *xprt;
3048 struct sock_xprt *transport;
3049 struct svc_sock *bc_sock;
3050 struct rpc_xprt *ret;
3052 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3053 xprt_tcp_slot_table_entries);
3056 transport = container_of(xprt, struct sock_xprt, xprt);
3058 xprt->prot = IPPROTO_TCP;
3059 xprt->xprt_class = &xs_bc_tcp_transport;
3060 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3061 xprt->timeout = &xs_tcp_default_timeout;
3064 xprt_set_bound(xprt);
3065 xprt->bind_timeout = 0;
3066 xprt->reestablish_timeout = 0;
3067 xprt->idle_timeout = 0;
3069 xprt->ops = &bc_tcp_ops;
3071 switch (addr->sa_family) {
3073 xs_format_peer_addresses(xprt, "tcp",
3077 xs_format_peer_addresses(xprt, "tcp",
3078 RPCBIND_NETID_TCP6);
3081 ret = ERR_PTR(-EAFNOSUPPORT);
3085 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3086 xprt->address_strings[RPC_DISPLAY_ADDR],
3087 xprt->address_strings[RPC_DISPLAY_PORT],
3088 xprt->address_strings[RPC_DISPLAY_PROTO]);
3091 * Once we've associated a backchannel xprt with a connection,
3092 * we want to keep it around as long as the connection lasts,
3093 * in case we need to start using it for a backchannel again;
3094 * this reference won't be dropped until bc_xprt is destroyed.
3097 args->bc_xprt->xpt_bc_xprt = xprt;
3098 xprt->bc_xprt = args->bc_xprt;
3099 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3100 transport->sock = bc_sock->sk_sock;
3101 transport->inet = bc_sock->sk_sk;
3104 * Since we don't want connections for the backchannel, we set
3105 * the xprt status to connected
3107 xprt_set_connected(xprt);
3109 if (try_module_get(THIS_MODULE))
3112 args->bc_xprt->xpt_bc_xprt = NULL;
3113 args->bc_xprt->xpt_bc_xps = NULL;
3115 ret = ERR_PTR(-EINVAL);
3121 static struct xprt_class xs_local_transport = {
3122 .list = LIST_HEAD_INIT(xs_local_transport.list),
3123 .name = "named UNIX socket",
3124 .owner = THIS_MODULE,
3125 .ident = XPRT_TRANSPORT_LOCAL,
3126 .setup = xs_setup_local,
3130 static struct xprt_class xs_udp_transport = {
3131 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3133 .owner = THIS_MODULE,
3134 .ident = XPRT_TRANSPORT_UDP,
3135 .setup = xs_setup_udp,
3136 .netid = { "udp", "udp6", "" },
3139 static struct xprt_class xs_tcp_transport = {
3140 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3142 .owner = THIS_MODULE,
3143 .ident = XPRT_TRANSPORT_TCP,
3144 .setup = xs_setup_tcp,
3145 .netid = { "tcp", "tcp6", "" },
3148 static struct xprt_class xs_bc_tcp_transport = {
3149 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3150 .name = "tcp NFSv4.1 backchannel",
3151 .owner = THIS_MODULE,
3152 .ident = XPRT_TRANSPORT_BC_TCP,
3153 .setup = xs_setup_bc_tcp,
3158 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3161 int init_socket_xprt(void)
3163 if (!sunrpc_table_header)
3164 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3166 xprt_register_transport(&xs_local_transport);
3167 xprt_register_transport(&xs_udp_transport);
3168 xprt_register_transport(&xs_tcp_transport);
3169 xprt_register_transport(&xs_bc_tcp_transport);
3175 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3178 void cleanup_socket_xprt(void)
3180 if (sunrpc_table_header) {
3181 unregister_sysctl_table(sunrpc_table_header);
3182 sunrpc_table_header = NULL;
3185 xprt_unregister_transport(&xs_local_transport);
3186 xprt_unregister_transport(&xs_udp_transport);
3187 xprt_unregister_transport(&xs_tcp_transport);
3188 xprt_unregister_transport(&xs_bc_tcp_transport);
3191 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3193 return param_set_uint_minmax(val, kp,
3198 static const struct kernel_param_ops param_ops_portnr = {
3199 .set = param_set_portnr,
3200 .get = param_get_uint,
3203 #define param_check_portnr(name, p) \
3204 __param_check(name, p, unsigned int);
3206 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3207 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3209 static int param_set_slot_table_size(const char *val,
3210 const struct kernel_param *kp)
3212 return param_set_uint_minmax(val, kp,
3214 RPC_MAX_SLOT_TABLE);
3217 static const struct kernel_param_ops param_ops_slot_table_size = {
3218 .set = param_set_slot_table_size,
3219 .get = param_get_uint,
3222 #define param_check_slot_table_size(name, p) \
3223 __param_check(name, p, unsigned int);
3225 static int param_set_max_slot_table_size(const char *val,
3226 const struct kernel_param *kp)
3228 return param_set_uint_minmax(val, kp,
3230 RPC_MAX_SLOT_TABLE_LIMIT);
3233 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3234 .set = param_set_max_slot_table_size,
3235 .get = param_get_uint,
3238 #define param_check_max_slot_table_size(name, p) \
3239 __param_check(name, p, unsigned int);
3241 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3242 slot_table_size, 0644);
3243 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3244 max_slot_table_size, 0644);
3245 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3246 slot_table_size, 0644);