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;
95 * FIXME: changing the UDP slot table size should also resize the UDP
96 * socket buffers for existing UDP transports
98 static struct ctl_table xs_tunables_table[] = {
100 .procname = "udp_slot_table_entries",
101 .data = &xprt_udp_slot_table_entries,
102 .maxlen = sizeof(unsigned int),
104 .proc_handler = proc_dointvec_minmax,
105 .extra1 = &min_slot_table_size,
106 .extra2 = &max_slot_table_size
109 .procname = "tcp_slot_table_entries",
110 .data = &xprt_tcp_slot_table_entries,
111 .maxlen = sizeof(unsigned int),
113 .proc_handler = proc_dointvec_minmax,
114 .extra1 = &min_slot_table_size,
115 .extra2 = &max_slot_table_size
118 .procname = "tcp_max_slot_table_entries",
119 .data = &xprt_max_tcp_slot_table_entries,
120 .maxlen = sizeof(unsigned int),
122 .proc_handler = proc_dointvec_minmax,
123 .extra1 = &min_slot_table_size,
124 .extra2 = &max_tcp_slot_table_limit
127 .procname = "min_resvport",
128 .data = &xprt_min_resvport,
129 .maxlen = sizeof(unsigned int),
131 .proc_handler = proc_dointvec_minmax,
132 .extra1 = &xprt_min_resvport_limit,
133 .extra2 = &xprt_max_resvport_limit
136 .procname = "max_resvport",
137 .data = &xprt_max_resvport,
138 .maxlen = sizeof(unsigned int),
140 .proc_handler = proc_dointvec_minmax,
141 .extra1 = &xprt_min_resvport_limit,
142 .extra2 = &xprt_max_resvport_limit
145 .procname = "tcp_fin_timeout",
146 .data = &xs_tcp_fin_timeout,
147 .maxlen = sizeof(xs_tcp_fin_timeout),
149 .proc_handler = proc_dointvec_jiffies,
154 static struct ctl_table sunrpc_table[] = {
156 .procname = "sunrpc",
158 .child = xs_tunables_table
164 * Wait duration for a reply from the RPC portmapper.
166 #define XS_BIND_TO (60U * HZ)
169 * Delay if a UDP socket connect error occurs. This is most likely some
170 * kind of resource problem on the local host.
172 #define XS_UDP_REEST_TO (2U * HZ)
175 * The reestablish timeout allows clients to delay for a bit before attempting
176 * to reconnect to a server that just dropped our connection.
178 * We implement an exponential backoff when trying to reestablish a TCP
179 * transport connection with the server. Some servers like to drop a TCP
180 * connection when they are overworked, so we start with a short timeout and
181 * increase over time if the server is down or not responding.
183 #define XS_TCP_INIT_REEST_TO (3U * HZ)
186 * TCP idle timeout; client drops the transport socket if it is idle
187 * for this long. Note that we also timeout UDP sockets to prevent
188 * holding port numbers when there is no RPC traffic.
190 #define XS_IDLE_DISC_TO (5U * 60 * HZ)
192 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
193 # undef RPC_DEBUG_DATA
194 # define RPCDBG_FACILITY RPCDBG_TRANS
197 #ifdef RPC_DEBUG_DATA
198 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
200 u8 *buf = (u8 *) packet;
203 dprintk("RPC: %s\n", msg);
204 for (j = 0; j < count && j < 128; j += 4) {
208 dprintk("0x%04x ", j);
210 dprintk("%02x%02x%02x%02x ",
211 buf[j], buf[j+1], buf[j+2], buf[j+3]);
216 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
222 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
224 return (struct rpc_xprt *) sk->sk_user_data;
227 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
229 return (struct sockaddr *) &xprt->addr;
232 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
234 return (struct sockaddr_un *) &xprt->addr;
237 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
239 return (struct sockaddr_in *) &xprt->addr;
242 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
244 return (struct sockaddr_in6 *) &xprt->addr;
247 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
249 struct sockaddr *sap = xs_addr(xprt);
250 struct sockaddr_in6 *sin6;
251 struct sockaddr_in *sin;
252 struct sockaddr_un *sun;
255 switch (sap->sa_family) {
257 sun = xs_addr_un(xprt);
258 strlcpy(buf, sun->sun_path, sizeof(buf));
259 xprt->address_strings[RPC_DISPLAY_ADDR] =
260 kstrdup(buf, GFP_KERNEL);
263 (void)rpc_ntop(sap, buf, sizeof(buf));
264 xprt->address_strings[RPC_DISPLAY_ADDR] =
265 kstrdup(buf, GFP_KERNEL);
266 sin = xs_addr_in(xprt);
267 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
270 (void)rpc_ntop(sap, buf, sizeof(buf));
271 xprt->address_strings[RPC_DISPLAY_ADDR] =
272 kstrdup(buf, GFP_KERNEL);
273 sin6 = xs_addr_in6(xprt);
274 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
280 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
283 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
285 struct sockaddr *sap = xs_addr(xprt);
288 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
289 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
291 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
292 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
295 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
296 const char *protocol,
299 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
300 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
301 xs_format_common_peer_addresses(xprt);
302 xs_format_common_peer_ports(xprt);
305 static void xs_update_peer_port(struct rpc_xprt *xprt)
307 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
308 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
310 xs_format_common_peer_ports(xprt);
313 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
317 for (i = 0; i < RPC_DISPLAY_MAX; i++)
319 case RPC_DISPLAY_PROTO:
320 case RPC_DISPLAY_NETID:
323 kfree(xprt->address_strings[i]);
328 xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp)
332 if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES))
334 n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT;
335 for (i = 0; i < n; i++) {
338 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp);
339 if (!buf->pages[i]) {
341 return i > buf->page_base ? i - buf->page_base : 0;
348 xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek)
352 iov_iter_advance(&msg->msg_iter, seek);
353 ret = sock_recvmsg(sock, msg, flags);
354 return ret > 0 ? ret + seek : ret;
358 xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags,
359 struct kvec *kvec, size_t count, size_t seek)
361 iov_iter_kvec(&msg->msg_iter, READ, kvec, 1, count);
362 return xs_sock_recvmsg(sock, msg, flags, seek);
366 xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags,
367 struct bio_vec *bvec, unsigned long nr, size_t count,
370 iov_iter_bvec(&msg->msg_iter, READ, bvec, nr, count);
371 return xs_sock_recvmsg(sock, msg, flags, seek);
375 xs_read_discard(struct socket *sock, struct msghdr *msg, int flags,
378 iov_iter_discard(&msg->msg_iter, READ, count);
379 return sock_recvmsg(sock, msg, flags);
382 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
384 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
386 struct bvec_iter bi = {
391 bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
392 for_each_bvec(bv, bvec, bi, bi)
393 flush_dcache_page(bv.bv_page);
397 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
403 xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
404 struct xdr_buf *buf, size_t count, size_t seek, size_t *read)
406 size_t want, seek_init = seek, offset = 0;
409 want = min_t(size_t, count, buf->head[0].iov_len);
411 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek);
415 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
425 want = xs_alloc_sparse_pages(buf,
426 min_t(size_t, count - offset, buf->page_len),
429 ret = xs_read_bvec(sock, msg, flags, buf->bvec,
430 xdr_buf_pagecount(buf),
431 want + buf->page_base,
432 seek + buf->page_base);
435 xs_flush_bvec(buf->bvec, ret, seek + buf->page_base);
436 offset += ret - buf->page_base;
437 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
447 want = min_t(size_t, count - offset, buf->tail[0].iov_len);
449 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek);
453 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
457 } else if (offset < seek_init)
461 *read = offset - seek_init;
469 xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf)
471 if (!transport->recv.copied) {
472 if (buf->head[0].iov_len >= transport->recv.offset)
473 memcpy(buf->head[0].iov_base,
474 &transport->recv.xid,
475 transport->recv.offset);
476 transport->recv.copied = transport->recv.offset;
481 xs_read_stream_request_done(struct sock_xprt *transport)
483 return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT);
487 xs_read_stream_check_eor(struct sock_xprt *transport,
490 if (xs_read_stream_request_done(transport))
491 msg->msg_flags |= MSG_EOR;
495 xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg,
496 int flags, struct rpc_rqst *req)
498 struct xdr_buf *buf = &req->rq_private_buf;
502 xs_read_header(transport, buf);
504 want = transport->recv.len - transport->recv.offset;
506 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf,
507 transport->recv.copied + want,
508 transport->recv.copied,
510 transport->recv.offset += read;
511 transport->recv.copied += read;
514 if (transport->recv.offset == transport->recv.len)
515 xs_read_stream_check_eor(transport, msg);
525 msg->msg_flags |= MSG_TRUNC;
530 return ret < 0 ? ret : read;
534 xs_read_stream_headersize(bool isfrag)
537 return sizeof(__be32);
538 return 3 * sizeof(__be32);
542 xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg,
543 int flags, size_t want, size_t seek)
546 .iov_base = &transport->recv.fraghdr,
549 return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek);
552 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
554 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
556 struct rpc_xprt *xprt = &transport->xprt;
557 struct rpc_rqst *req;
560 /* Look up and lock the request corresponding to the given XID */
561 req = xprt_lookup_bc_request(xprt, transport->recv.xid);
563 printk(KERN_WARNING "Callback slot table overflowed\n");
566 if (transport->recv.copied && !req->rq_private_buf.len)
569 ret = xs_read_stream_request(transport, msg, flags, req);
570 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
571 xprt_complete_bc_request(req, transport->recv.copied);
573 req->rq_private_buf.len = transport->recv.copied;
577 #else /* CONFIG_SUNRPC_BACKCHANNEL */
579 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
583 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
586 xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags)
588 struct rpc_xprt *xprt = &transport->xprt;
589 struct rpc_rqst *req;
592 /* Look up and lock the request corresponding to the given XID */
593 spin_lock(&xprt->queue_lock);
594 req = xprt_lookup_rqst(xprt, transport->recv.xid);
595 if (!req || (transport->recv.copied && !req->rq_private_buf.len)) {
596 msg->msg_flags |= MSG_TRUNC;
600 spin_unlock(&xprt->queue_lock);
602 ret = xs_read_stream_request(transport, msg, flags, req);
604 spin_lock(&xprt->queue_lock);
605 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
606 xprt_complete_rqst(req->rq_task, transport->recv.copied);
608 req->rq_private_buf.len = transport->recv.copied;
609 xprt_unpin_rqst(req);
611 spin_unlock(&xprt->queue_lock);
616 xs_read_stream(struct sock_xprt *transport, int flags)
618 struct msghdr msg = { 0 };
619 size_t want, read = 0;
622 if (transport->recv.len == 0) {
623 want = xs_read_stream_headersize(transport->recv.copied != 0);
624 ret = xs_read_stream_header(transport, &msg, flags, want,
625 transport->recv.offset);
628 transport->recv.offset = ret;
629 if (transport->recv.offset != want)
630 return transport->recv.offset;
631 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) &
632 RPC_FRAGMENT_SIZE_MASK;
633 transport->recv.offset -= sizeof(transport->recv.fraghdr);
637 switch (be32_to_cpu(transport->recv.calldir)) {
639 msg.msg_flags |= MSG_TRUNC;
642 ret = xs_read_stream_call(transport, &msg, flags);
645 ret = xs_read_stream_reply(transport, &msg, flags);
647 if (msg.msg_flags & MSG_TRUNC) {
648 transport->recv.calldir = cpu_to_be32(-1);
649 transport->recv.copied = -1;
654 if (transport->recv.offset < transport->recv.len) {
655 if (!(msg.msg_flags & MSG_TRUNC))
658 ret = xs_read_discard(transport->sock, &msg, flags,
659 transport->recv.len - transport->recv.offset);
662 transport->recv.offset += ret;
664 if (transport->recv.offset != transport->recv.len)
667 if (xs_read_stream_request_done(transport)) {
668 trace_xs_stream_read_request(transport);
669 transport->recv.copied = 0;
671 transport->recv.offset = 0;
672 transport->recv.len = 0;
675 return ret != 0 ? ret : -ESHUTDOWN;
678 static __poll_t xs_poll_socket(struct sock_xprt *transport)
680 return transport->sock->ops->poll(transport->file, transport->sock,
684 static bool xs_poll_socket_readable(struct sock_xprt *transport)
686 __poll_t events = xs_poll_socket(transport);
688 return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP);
691 static void xs_poll_check_readable(struct sock_xprt *transport)
694 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
695 if (!xs_poll_socket_readable(transport))
697 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
698 queue_work(xprtiod_workqueue, &transport->recv_worker);
701 static void xs_stream_data_receive(struct sock_xprt *transport)
706 mutex_lock(&transport->recv_mutex);
707 if (transport->sock == NULL)
710 ret = xs_read_stream(transport, MSG_DONTWAIT);
716 if (ret == -ESHUTDOWN)
717 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
719 xs_poll_check_readable(transport);
721 mutex_unlock(&transport->recv_mutex);
722 trace_xs_stream_read_data(&transport->xprt, ret, read);
725 static void xs_stream_data_receive_workfn(struct work_struct *work)
727 struct sock_xprt *transport =
728 container_of(work, struct sock_xprt, recv_worker);
729 unsigned int pflags = memalloc_nofs_save();
731 xs_stream_data_receive(transport);
732 memalloc_nofs_restore(pflags);
736 xs_stream_reset_connect(struct sock_xprt *transport)
738 transport->recv.offset = 0;
739 transport->recv.len = 0;
740 transport->recv.copied = 0;
741 transport->xmit.offset = 0;
745 xs_stream_start_connect(struct sock_xprt *transport)
747 transport->xprt.stat.connect_count++;
748 transport->xprt.stat.connect_start = jiffies;
751 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
754 * xs_nospace - handle transmit was incomplete
755 * @req: pointer to RPC request
758 static int xs_nospace(struct rpc_rqst *req)
760 struct rpc_xprt *xprt = req->rq_xprt;
761 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
762 struct sock *sk = transport->inet;
765 trace_rpc_socket_nospace(req, transport);
767 /* Protect against races with write_space */
768 spin_lock(&xprt->transport_lock);
770 /* Don't race with disconnect */
771 if (xprt_connected(xprt)) {
772 /* wait for more buffer space */
773 sk->sk_write_pending++;
774 xprt_wait_for_buffer_space(xprt);
778 spin_unlock(&xprt->transport_lock);
780 /* Race breaker in case memory is freed before above code is called */
781 if (ret == -EAGAIN) {
782 struct socket_wq *wq;
785 wq = rcu_dereference(sk->sk_wq);
786 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
789 sk->sk_write_space(sk);
795 xs_stream_prepare_request(struct rpc_rqst *req)
797 xdr_free_bvec(&req->rq_rcv_buf);
798 req->rq_task->tk_status = xdr_alloc_bvec(&req->rq_rcv_buf, GFP_KERNEL);
802 * Determine if the previous message in the stream was aborted before it
803 * could complete transmission.
806 xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req)
808 return transport->xmit.offset != 0 && req->rq_bytes_sent == 0;
812 * Return the stream record marker field for a record of length < 2^31-1
815 xs_stream_record_marker(struct xdr_buf *xdr)
819 return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len);
823 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
824 * @req: pointer to RPC request
827 * 0: The request has been sent
828 * EAGAIN: The socket was blocked, please call again later to
829 * complete the request
830 * ENOTCONN: Caller needs to invoke connect logic then call again
831 * other: Some other error occured, the request was not sent
833 static int xs_local_send_request(struct rpc_rqst *req)
835 struct rpc_xprt *xprt = req->rq_xprt;
836 struct sock_xprt *transport =
837 container_of(xprt, struct sock_xprt, xprt);
838 struct xdr_buf *xdr = &req->rq_snd_buf;
839 rpc_fraghdr rm = xs_stream_record_marker(xdr);
840 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
841 struct msghdr msg = {
842 .msg_flags = XS_SENDMSG_FLAGS,
847 /* Close the stream if the previous transmission was incomplete */
848 if (xs_send_request_was_aborted(transport, req)) {
853 xs_pktdump("packet data:",
854 req->rq_svec->iov_base, req->rq_svec->iov_len);
856 req->rq_xtime = ktime_get();
857 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
858 transport->xmit.offset, rm, &sent);
859 dprintk("RPC: %s(%u) = %d\n",
860 __func__, xdr->len - transport->xmit.offset, status);
862 if (status == -EAGAIN && sock_writeable(transport->inet))
865 if (likely(sent > 0) || status == 0) {
866 transport->xmit.offset += sent;
867 req->rq_bytes_sent = transport->xmit.offset;
868 if (likely(req->rq_bytes_sent >= msglen)) {
869 req->rq_xmit_bytes_sent += transport->xmit.offset;
870 transport->xmit.offset = 0;
880 status = xs_nospace(req);
883 dprintk("RPC: sendmsg returned unrecognized error %d\n",
895 * xs_udp_send_request - write an RPC request to a UDP socket
896 * @req: pointer to RPC request
899 * 0: The request has been sent
900 * EAGAIN: The socket was blocked, please call again later to
901 * complete the request
902 * ENOTCONN: Caller needs to invoke connect logic then call again
903 * other: Some other error occurred, the request was not sent
905 static int xs_udp_send_request(struct rpc_rqst *req)
907 struct rpc_xprt *xprt = req->rq_xprt;
908 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
909 struct xdr_buf *xdr = &req->rq_snd_buf;
910 struct msghdr msg = {
911 .msg_name = xs_addr(xprt),
912 .msg_namelen = xprt->addrlen,
913 .msg_flags = XS_SENDMSG_FLAGS,
918 xs_pktdump("packet data:",
919 req->rq_svec->iov_base,
920 req->rq_svec->iov_len);
922 if (!xprt_bound(xprt))
925 if (!xprt_request_get_cong(xprt, req))
928 req->rq_xtime = ktime_get();
929 status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, 0, &sent);
931 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
934 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
935 if (status == -EPERM)
938 if (status == -EAGAIN && sock_writeable(transport->inet))
941 if (sent > 0 || status == 0) {
942 req->rq_xmit_bytes_sent += sent;
943 if (sent >= req->rq_slen)
945 /* Still some bytes left; set up for a retry later. */
953 /* Should we call xs_close() here? */
956 status = xs_nospace(req);
963 /* When the server has died, an ICMP port unreachable message
964 * prompts ECONNREFUSED. */
967 dprintk("RPC: sendmsg returned unrecognized error %d\n",
975 * xs_tcp_send_request - write an RPC request to a TCP socket
976 * @req: pointer to RPC request
979 * 0: The request has been sent
980 * EAGAIN: The socket was blocked, please call again later to
981 * complete the request
982 * ENOTCONN: Caller needs to invoke connect logic then call again
983 * other: Some other error occurred, the request was not sent
985 * XXX: In the case of soft timeouts, should we eventually give up
986 * if sendmsg is not able to make progress?
988 static int xs_tcp_send_request(struct rpc_rqst *req)
990 struct rpc_xprt *xprt = req->rq_xprt;
991 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
992 struct xdr_buf *xdr = &req->rq_snd_buf;
993 rpc_fraghdr rm = xs_stream_record_marker(xdr);
994 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
995 struct msghdr msg = {
996 .msg_flags = XS_SENDMSG_FLAGS,
998 bool vm_wait = false;
1002 /* Close the stream if the previous transmission was incomplete */
1003 if (xs_send_request_was_aborted(transport, req)) {
1004 if (transport->sock != NULL)
1005 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
1009 xs_pktdump("packet data:",
1010 req->rq_svec->iov_base,
1011 req->rq_svec->iov_len);
1013 if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
1014 xs_tcp_set_socket_timeouts(xprt, transport->sock);
1016 /* Continue transmitting the packet/record. We must be careful
1017 * to cope with writespace callbacks arriving _after_ we have
1018 * called sendmsg(). */
1019 req->rq_xtime = ktime_get();
1021 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
1022 transport->xmit.offset, rm, &sent);
1024 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
1025 xdr->len - transport->xmit.offset, status);
1027 /* If we've sent the entire packet, immediately
1028 * reset the count of bytes sent. */
1029 transport->xmit.offset += sent;
1030 req->rq_bytes_sent = transport->xmit.offset;
1031 if (likely(req->rq_bytes_sent >= msglen)) {
1032 req->rq_xmit_bytes_sent += transport->xmit.offset;
1033 transport->xmit.offset = 0;
1037 WARN_ON_ONCE(sent == 0 && status == 0);
1039 if (status == -EAGAIN ) {
1041 * Return EAGAIN if we're sure we're hitting the
1042 * socket send buffer limits.
1044 if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
1047 * Did we hit a memory allocation failure?
1053 /* Retry, knowing now that we're below the
1054 * socket send buffer limit
1068 /* Should we call xs_close() here? */
1071 status = xs_nospace(req);
1081 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1088 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1090 transport->old_data_ready = sk->sk_data_ready;
1091 transport->old_state_change = sk->sk_state_change;
1092 transport->old_write_space = sk->sk_write_space;
1093 transport->old_error_report = sk->sk_error_report;
1096 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1098 sk->sk_data_ready = transport->old_data_ready;
1099 sk->sk_state_change = transport->old_state_change;
1100 sk->sk_write_space = transport->old_write_space;
1101 sk->sk_error_report = transport->old_error_report;
1104 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
1106 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1108 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1109 clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state);
1110 clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state);
1111 clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state);
1114 static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr)
1116 set_bit(nr, &transport->sock_state);
1117 queue_work(xprtiod_workqueue, &transport->error_worker);
1120 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1122 smp_mb__before_atomic();
1123 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1124 clear_bit(XPRT_CLOSING, &xprt->state);
1125 xs_sock_reset_state_flags(xprt);
1126 smp_mb__after_atomic();
1130 * xs_error_report - callback to handle TCP socket state errors
1133 * Note: we don't call sock_error() since there may be a rpc_task
1134 * using the socket, and so we don't want to clear sk->sk_err.
1136 static void xs_error_report(struct sock *sk)
1138 struct sock_xprt *transport;
1139 struct rpc_xprt *xprt;
1141 read_lock_bh(&sk->sk_callback_lock);
1142 if (!(xprt = xprt_from_sock(sk)))
1145 transport = container_of(xprt, struct sock_xprt, xprt);
1146 transport->xprt_err = -sk->sk_err;
1147 if (transport->xprt_err == 0)
1149 dprintk("RPC: xs_error_report client %p, error=%d...\n",
1150 xprt, -transport->xprt_err);
1151 trace_rpc_socket_error(xprt, sk->sk_socket, transport->xprt_err);
1153 /* barrier ensures xprt_err is set before XPRT_SOCK_WAKE_ERROR */
1154 smp_mb__before_atomic();
1155 xs_run_error_worker(transport, XPRT_SOCK_WAKE_ERROR);
1157 read_unlock_bh(&sk->sk_callback_lock);
1160 static void xs_reset_transport(struct sock_xprt *transport)
1162 struct socket *sock = transport->sock;
1163 struct sock *sk = transport->inet;
1164 struct rpc_xprt *xprt = &transport->xprt;
1165 struct file *filp = transport->file;
1170 if (atomic_read(&transport->xprt.swapper))
1171 sk_clear_memalloc(sk);
1173 kernel_sock_shutdown(sock, SHUT_RDWR);
1175 mutex_lock(&transport->recv_mutex);
1176 write_lock_bh(&sk->sk_callback_lock);
1177 transport->inet = NULL;
1178 transport->sock = NULL;
1179 transport->file = NULL;
1181 sk->sk_user_data = NULL;
1183 xs_restore_old_callbacks(transport, sk);
1184 xprt_clear_connected(xprt);
1185 write_unlock_bh(&sk->sk_callback_lock);
1186 xs_sock_reset_connection_flags(xprt);
1187 /* Reset stream record info */
1188 xs_stream_reset_connect(transport);
1189 mutex_unlock(&transport->recv_mutex);
1191 trace_rpc_socket_close(xprt, sock);
1194 xprt_disconnect_done(xprt);
1198 * xs_close - close a socket
1201 * This is used when all requests are complete; ie, no DRC state remains
1202 * on the server we want to save.
1204 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1205 * xs_reset_transport() zeroing the socket from underneath a writer.
1207 static void xs_close(struct rpc_xprt *xprt)
1209 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1211 dprintk("RPC: xs_close xprt %p\n", xprt);
1213 xs_reset_transport(transport);
1214 xprt->reestablish_timeout = 0;
1217 static void xs_inject_disconnect(struct rpc_xprt *xprt)
1219 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
1221 xprt_disconnect_done(xprt);
1224 static void xs_xprt_free(struct rpc_xprt *xprt)
1226 xs_free_peer_addresses(xprt);
1231 * xs_destroy - prepare to shutdown a transport
1232 * @xprt: doomed transport
1235 static void xs_destroy(struct rpc_xprt *xprt)
1237 struct sock_xprt *transport = container_of(xprt,
1238 struct sock_xprt, xprt);
1239 dprintk("RPC: xs_destroy xprt %p\n", xprt);
1241 cancel_delayed_work_sync(&transport->connect_worker);
1243 cancel_work_sync(&transport->recv_worker);
1244 cancel_work_sync(&transport->error_worker);
1246 module_put(THIS_MODULE);
1250 * xs_udp_data_read_skb - receive callback for UDP sockets
1256 static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1258 struct sk_buff *skb)
1260 struct rpc_task *task;
1261 struct rpc_rqst *rovr;
1262 int repsize, copied;
1268 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
1272 /* Copy the XID from the skb... */
1273 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1277 /* Look up and lock the request corresponding to the given XID */
1278 spin_lock(&xprt->queue_lock);
1279 rovr = xprt_lookup_rqst(xprt, *xp);
1282 xprt_pin_rqst(rovr);
1283 xprt_update_rtt(rovr->rq_task);
1284 spin_unlock(&xprt->queue_lock);
1285 task = rovr->rq_task;
1287 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1290 /* Suck it into the iovec, verify checksum if not done by hw. */
1291 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1292 spin_lock(&xprt->queue_lock);
1293 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1298 spin_lock(&xprt->transport_lock);
1299 xprt_adjust_cwnd(xprt, task, copied);
1300 spin_unlock(&xprt->transport_lock);
1301 spin_lock(&xprt->queue_lock);
1302 xprt_complete_rqst(task, copied);
1303 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1305 xprt_unpin_rqst(rovr);
1307 spin_unlock(&xprt->queue_lock);
1310 static void xs_udp_data_receive(struct sock_xprt *transport)
1312 struct sk_buff *skb;
1316 mutex_lock(&transport->recv_mutex);
1317 sk = transport->inet;
1321 skb = skb_recv_udp(sk, 0, 1, &err);
1324 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1328 xs_poll_check_readable(transport);
1330 mutex_unlock(&transport->recv_mutex);
1333 static void xs_udp_data_receive_workfn(struct work_struct *work)
1335 struct sock_xprt *transport =
1336 container_of(work, struct sock_xprt, recv_worker);
1337 unsigned int pflags = memalloc_nofs_save();
1339 xs_udp_data_receive(transport);
1340 memalloc_nofs_restore(pflags);
1344 * xs_data_ready - "data ready" callback for UDP sockets
1345 * @sk: socket with data to read
1348 static void xs_data_ready(struct sock *sk)
1350 struct rpc_xprt *xprt;
1352 read_lock_bh(&sk->sk_callback_lock);
1353 dprintk("RPC: xs_data_ready...\n");
1354 xprt = xprt_from_sock(sk);
1356 struct sock_xprt *transport = container_of(xprt,
1357 struct sock_xprt, xprt);
1358 transport->old_data_ready(sk);
1359 /* Any data means we had a useful conversation, so
1360 * then we don't need to delay the next reconnect
1362 if (xprt->reestablish_timeout)
1363 xprt->reestablish_timeout = 0;
1364 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1365 queue_work(xprtiod_workqueue, &transport->recv_worker);
1367 read_unlock_bh(&sk->sk_callback_lock);
1371 * Helper function to force a TCP close if the server is sending
1372 * junk and/or it has put us in CLOSE_WAIT
1374 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1376 xprt_force_disconnect(xprt);
1379 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1380 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1384 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1387 * xs_tcp_state_change - callback to handle TCP socket state changes
1388 * @sk: socket whose state has changed
1391 static void xs_tcp_state_change(struct sock *sk)
1393 struct rpc_xprt *xprt;
1394 struct sock_xprt *transport;
1396 read_lock_bh(&sk->sk_callback_lock);
1397 if (!(xprt = xprt_from_sock(sk)))
1399 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1400 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1401 sk->sk_state, xprt_connected(xprt),
1402 sock_flag(sk, SOCK_DEAD),
1403 sock_flag(sk, SOCK_ZAPPED),
1406 transport = container_of(xprt, struct sock_xprt, xprt);
1407 trace_rpc_socket_state_change(xprt, sk->sk_socket);
1408 switch (sk->sk_state) {
1409 case TCP_ESTABLISHED:
1410 if (!xprt_test_and_set_connected(xprt)) {
1411 xprt->connect_cookie++;
1412 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1413 xprt_clear_connecting(xprt);
1415 xprt->stat.connect_count++;
1416 xprt->stat.connect_time += (long)jiffies -
1417 xprt->stat.connect_start;
1418 xs_run_error_worker(transport, XPRT_SOCK_WAKE_PENDING);
1422 /* The client initiated a shutdown of the socket */
1423 xprt->connect_cookie++;
1424 xprt->reestablish_timeout = 0;
1425 set_bit(XPRT_CLOSING, &xprt->state);
1426 smp_mb__before_atomic();
1427 clear_bit(XPRT_CONNECTED, &xprt->state);
1428 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1429 smp_mb__after_atomic();
1431 case TCP_CLOSE_WAIT:
1432 /* The server initiated a shutdown of the socket */
1433 xprt->connect_cookie++;
1434 clear_bit(XPRT_CONNECTED, &xprt->state);
1435 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1439 * If the server closed down the connection, make sure that
1440 * we back off before reconnecting
1442 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1443 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1446 set_bit(XPRT_CLOSING, &xprt->state);
1447 smp_mb__before_atomic();
1448 clear_bit(XPRT_CONNECTED, &xprt->state);
1449 smp_mb__after_atomic();
1452 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1453 &transport->sock_state))
1454 xprt_clear_connecting(xprt);
1455 clear_bit(XPRT_CLOSING, &xprt->state);
1456 /* Trigger the socket release */
1457 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1460 read_unlock_bh(&sk->sk_callback_lock);
1463 static void xs_write_space(struct sock *sk)
1465 struct socket_wq *wq;
1466 struct sock_xprt *transport;
1467 struct rpc_xprt *xprt;
1471 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1473 if (unlikely(!(xprt = xprt_from_sock(sk))))
1475 transport = container_of(xprt, struct sock_xprt, xprt);
1477 wq = rcu_dereference(sk->sk_wq);
1478 if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1481 xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE);
1482 sk->sk_write_pending--;
1488 * xs_udp_write_space - callback invoked when socket buffer space
1490 * @sk: socket whose state has changed
1492 * Called when more output buffer space is available for this socket.
1493 * We try not to wake our writers until they can make "significant"
1494 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1495 * with a bunch of small requests.
1497 static void xs_udp_write_space(struct sock *sk)
1499 read_lock_bh(&sk->sk_callback_lock);
1501 /* from net/core/sock.c:sock_def_write_space */
1502 if (sock_writeable(sk))
1505 read_unlock_bh(&sk->sk_callback_lock);
1509 * xs_tcp_write_space - callback invoked when socket buffer space
1511 * @sk: socket whose state has changed
1513 * Called when more output buffer space is available for this socket.
1514 * We try not to wake our writers until they can make "significant"
1515 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1516 * with a bunch of small requests.
1518 static void xs_tcp_write_space(struct sock *sk)
1520 read_lock_bh(&sk->sk_callback_lock);
1522 /* from net/core/stream.c:sk_stream_write_space */
1523 if (sk_stream_is_writeable(sk))
1526 read_unlock_bh(&sk->sk_callback_lock);
1529 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1531 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1532 struct sock *sk = transport->inet;
1534 if (transport->rcvsize) {
1535 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1536 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1538 if (transport->sndsize) {
1539 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1540 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1541 sk->sk_write_space(sk);
1546 * xs_udp_set_buffer_size - set send and receive limits
1547 * @xprt: generic transport
1548 * @sndsize: requested size of send buffer, in bytes
1549 * @rcvsize: requested size of receive buffer, in bytes
1551 * Set socket send and receive buffer size limits.
1553 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1555 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1557 transport->sndsize = 0;
1559 transport->sndsize = sndsize + 1024;
1560 transport->rcvsize = 0;
1562 transport->rcvsize = rcvsize + 1024;
1564 xs_udp_do_set_buffer_size(xprt);
1568 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1569 * @xprt: controlling transport
1570 * @task: task that timed out
1572 * Adjust the congestion window after a retransmit timeout has occurred.
1574 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1576 spin_lock(&xprt->transport_lock);
1577 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1578 spin_unlock(&xprt->transport_lock);
1581 static int xs_get_random_port(void)
1583 unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1584 unsigned short range;
1585 unsigned short rand;
1589 range = max - min + 1;
1590 rand = (unsigned short) prandom_u32() % range;
1594 static unsigned short xs_sock_getport(struct socket *sock)
1596 struct sockaddr_storage buf;
1597 unsigned short port = 0;
1599 if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1601 switch (buf.ss_family) {
1603 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1606 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1613 * xs_set_port - reset the port number in the remote endpoint address
1614 * @xprt: generic transport
1615 * @port: new port number
1618 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1620 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1622 rpc_set_port(xs_addr(xprt), port);
1623 xs_update_peer_port(xprt);
1626 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1628 if (transport->srcport == 0 && transport->xprt.reuseport)
1629 transport->srcport = xs_sock_getport(sock);
1632 static int xs_get_srcport(struct sock_xprt *transport)
1634 int port = transport->srcport;
1636 if (port == 0 && transport->xprt.resvport)
1637 port = xs_get_random_port();
1641 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1643 if (transport->srcport != 0)
1644 transport->srcport = 0;
1645 if (!transport->xprt.resvport)
1647 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1648 return xprt_max_resvport;
1651 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1653 struct sockaddr_storage myaddr;
1655 int port = xs_get_srcport(transport);
1656 unsigned short last;
1659 * If we are asking for any ephemeral port (i.e. port == 0 &&
1660 * transport->xprt.resvport == 0), don't bind. Let the local
1661 * port selection happen implicitly when the socket is used
1662 * (for example at connect time).
1664 * This ensures that we can continue to establish TCP
1665 * connections even when all local ephemeral ports are already
1666 * a part of some TCP connection. This makes no difference
1667 * for UDP sockets, but also doens't harm them.
1669 * If we're asking for any reserved port (i.e. port == 0 &&
1670 * transport->xprt.resvport == 1) xs_get_srcport above will
1671 * ensure that port is non-zero and we will bind as needed.
1676 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1678 rpc_set_port((struct sockaddr *)&myaddr, port);
1679 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1680 transport->xprt.addrlen);
1682 transport->srcport = port;
1686 port = xs_next_srcport(transport, port);
1689 } while (err == -EADDRINUSE && nloop != 2);
1691 if (myaddr.ss_family == AF_INET)
1692 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1693 &((struct sockaddr_in *)&myaddr)->sin_addr,
1694 port, err ? "failed" : "ok", err);
1696 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1697 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1698 port, err ? "failed" : "ok", err);
1703 * We don't support autobind on AF_LOCAL sockets
1705 static void xs_local_rpcbind(struct rpc_task *task)
1707 xprt_set_bound(task->tk_xprt);
1710 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1714 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1715 static struct lock_class_key xs_key[2];
1716 static struct lock_class_key xs_slock_key[2];
1718 static inline void xs_reclassify_socketu(struct socket *sock)
1720 struct sock *sk = sock->sk;
1722 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1723 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1726 static inline void xs_reclassify_socket4(struct socket *sock)
1728 struct sock *sk = sock->sk;
1730 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1731 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1734 static inline void xs_reclassify_socket6(struct socket *sock)
1736 struct sock *sk = sock->sk;
1738 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1739 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1742 static inline void xs_reclassify_socket(int family, struct socket *sock)
1744 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1749 xs_reclassify_socketu(sock);
1752 xs_reclassify_socket4(sock);
1755 xs_reclassify_socket6(sock);
1760 static inline void xs_reclassify_socket(int family, struct socket *sock)
1765 static void xs_dummy_setup_socket(struct work_struct *work)
1769 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1770 struct sock_xprt *transport, int family, int type,
1771 int protocol, bool reuseport)
1774 struct socket *sock;
1777 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1779 dprintk("RPC: can't create %d transport socket (%d).\n",
1783 xs_reclassify_socket(family, sock);
1786 sock_set_reuseport(sock->sk);
1788 err = xs_bind(transport, sock);
1794 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1796 return ERR_CAST(filp);
1797 transport->file = filp;
1801 return ERR_PTR(err);
1804 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1805 struct socket *sock)
1807 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1810 if (!transport->inet) {
1811 struct sock *sk = sock->sk;
1813 write_lock_bh(&sk->sk_callback_lock);
1815 xs_save_old_callbacks(transport, sk);
1817 sk->sk_user_data = xprt;
1818 sk->sk_data_ready = xs_data_ready;
1819 sk->sk_write_space = xs_udp_write_space;
1820 sock_set_flag(sk, SOCK_FASYNC);
1821 sk->sk_error_report = xs_error_report;
1823 xprt_clear_connected(xprt);
1825 /* Reset to new socket */
1826 transport->sock = sock;
1827 transport->inet = sk;
1829 write_unlock_bh(&sk->sk_callback_lock);
1832 xs_stream_start_connect(transport);
1834 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1838 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1839 * @transport: socket transport to connect
1841 static int xs_local_setup_socket(struct sock_xprt *transport)
1843 struct rpc_xprt *xprt = &transport->xprt;
1845 struct socket *sock;
1848 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1849 SOCK_STREAM, 0, &sock, 1);
1851 dprintk("RPC: can't create AF_LOCAL "
1852 "transport socket (%d).\n", -status);
1855 xs_reclassify_socket(AF_LOCAL, sock);
1857 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1859 status = PTR_ERR(filp);
1862 transport->file = filp;
1864 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1865 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1867 status = xs_local_finish_connecting(xprt, sock);
1868 trace_rpc_socket_connect(xprt, sock, status);
1871 dprintk("RPC: xprt %p connected to %s\n",
1872 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1873 xprt->stat.connect_count++;
1874 xprt->stat.connect_time += (long)jiffies -
1875 xprt->stat.connect_start;
1876 xprt_set_connected(xprt);
1880 dprintk("RPC: xprt %p: socket %s does not exist\n",
1881 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1884 dprintk("RPC: xprt %p: connection refused for %s\n",
1885 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1888 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1890 xprt->address_strings[RPC_DISPLAY_ADDR]);
1894 xprt_clear_connecting(xprt);
1895 xprt_wake_pending_tasks(xprt, status);
1899 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1901 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1904 if (RPC_IS_ASYNC(task)) {
1906 * We want the AF_LOCAL connect to be resolved in the
1907 * filesystem namespace of the process making the rpc
1908 * call. Thus we connect synchronously.
1910 * If we want to support asynchronous AF_LOCAL calls,
1911 * we'll need to figure out how to pass a namespace to
1914 task->tk_rpc_status = -ENOTCONN;
1915 rpc_exit(task, -ENOTCONN);
1918 ret = xs_local_setup_socket(transport);
1919 if (ret && !RPC_IS_SOFTCONN(task))
1920 msleep_interruptible(15000);
1923 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
1925 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
1926 * know that we have exclusive access to the socket), to guard against
1927 * races with xs_reset_transport.
1929 static void xs_set_memalloc(struct rpc_xprt *xprt)
1931 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1935 * If there's no sock, then we have nothing to set. The
1936 * reconnecting process will get it for us.
1938 if (!transport->inet)
1940 if (atomic_read(&xprt->swapper))
1941 sk_set_memalloc(transport->inet);
1945 * xs_enable_swap - Tag this transport as being used for swap.
1946 * @xprt: transport to tag
1948 * Take a reference to this transport on behalf of the rpc_clnt, and
1949 * optionally mark it for swapping if it wasn't already.
1952 xs_enable_swap(struct rpc_xprt *xprt)
1954 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
1956 if (atomic_inc_return(&xprt->swapper) != 1)
1958 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
1959 return -ERESTARTSYS;
1961 sk_set_memalloc(xs->inet);
1962 xprt_release_xprt(xprt, NULL);
1967 * xs_disable_swap - Untag this transport as being used for swap.
1968 * @xprt: transport to tag
1970 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
1971 * swapper refcount goes to 0, untag the socket as a memalloc socket.
1974 xs_disable_swap(struct rpc_xprt *xprt)
1976 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
1978 if (!atomic_dec_and_test(&xprt->swapper))
1980 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
1983 sk_clear_memalloc(xs->inet);
1984 xprt_release_xprt(xprt, NULL);
1987 static void xs_set_memalloc(struct rpc_xprt *xprt)
1992 xs_enable_swap(struct rpc_xprt *xprt)
1998 xs_disable_swap(struct rpc_xprt *xprt)
2003 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2005 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2007 if (!transport->inet) {
2008 struct sock *sk = sock->sk;
2010 write_lock_bh(&sk->sk_callback_lock);
2012 xs_save_old_callbacks(transport, sk);
2014 sk->sk_user_data = xprt;
2015 sk->sk_data_ready = xs_data_ready;
2016 sk->sk_write_space = xs_udp_write_space;
2017 sock_set_flag(sk, SOCK_FASYNC);
2019 xprt_set_connected(xprt);
2021 /* Reset to new socket */
2022 transport->sock = sock;
2023 transport->inet = sk;
2025 xs_set_memalloc(xprt);
2027 write_unlock_bh(&sk->sk_callback_lock);
2029 xs_udp_do_set_buffer_size(xprt);
2031 xprt->stat.connect_start = jiffies;
2034 static void xs_udp_setup_socket(struct work_struct *work)
2036 struct sock_xprt *transport =
2037 container_of(work, struct sock_xprt, connect_worker.work);
2038 struct rpc_xprt *xprt = &transport->xprt;
2039 struct socket *sock;
2042 sock = xs_create_sock(xprt, transport,
2043 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2044 IPPROTO_UDP, false);
2048 dprintk("RPC: worker connecting xprt %p via %s to "
2049 "%s (port %s)\n", xprt,
2050 xprt->address_strings[RPC_DISPLAY_PROTO],
2051 xprt->address_strings[RPC_DISPLAY_ADDR],
2052 xprt->address_strings[RPC_DISPLAY_PORT]);
2054 xs_udp_finish_connecting(xprt, sock);
2055 trace_rpc_socket_connect(xprt, sock, 0);
2058 xprt_clear_connecting(xprt);
2059 xprt_unlock_connect(xprt, transport);
2060 xprt_wake_pending_tasks(xprt, status);
2064 * xs_tcp_shutdown - gracefully shut down a TCP socket
2067 * Initiates a graceful shutdown of the TCP socket by calling the
2068 * equivalent of shutdown(SHUT_RDWR);
2070 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2072 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2073 struct socket *sock = transport->sock;
2074 int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2080 kernel_sock_shutdown(sock, SHUT_RDWR);
2081 trace_rpc_socket_shutdown(xprt, sock);
2085 xs_reset_transport(transport);
2089 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2090 struct socket *sock)
2092 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2093 unsigned int keepidle;
2094 unsigned int keepcnt;
2097 spin_lock(&xprt->transport_lock);
2098 keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2099 keepcnt = xprt->timeout->to_retries + 1;
2100 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2101 (xprt->timeout->to_retries + 1);
2102 clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2103 spin_unlock(&xprt->transport_lock);
2105 /* TCP Keepalive options */
2106 sock_set_keepalive(sock->sk);
2107 tcp_sock_set_keepidle(sock->sk, keepidle);
2108 tcp_sock_set_keepintvl(sock->sk, keepidle);
2109 tcp_sock_set_keepcnt(sock->sk, keepcnt);
2111 /* TCP user timeout (see RFC5482) */
2112 tcp_sock_set_user_timeout(sock->sk, timeo);
2115 static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2116 unsigned long connect_timeout,
2117 unsigned long reconnect_timeout)
2119 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2120 struct rpc_timeout to;
2121 unsigned long initval;
2123 spin_lock(&xprt->transport_lock);
2124 if (reconnect_timeout < xprt->max_reconnect_timeout)
2125 xprt->max_reconnect_timeout = reconnect_timeout;
2126 if (connect_timeout < xprt->connect_timeout) {
2127 memcpy(&to, xprt->timeout, sizeof(to));
2128 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2129 /* Arbitrary lower limit */
2130 if (initval < XS_TCP_INIT_REEST_TO << 1)
2131 initval = XS_TCP_INIT_REEST_TO << 1;
2132 to.to_initval = initval;
2133 to.to_maxval = initval;
2134 memcpy(&transport->tcp_timeout, &to,
2135 sizeof(transport->tcp_timeout));
2136 xprt->timeout = &transport->tcp_timeout;
2137 xprt->connect_timeout = connect_timeout;
2139 set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2140 spin_unlock(&xprt->transport_lock);
2143 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2145 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2146 int ret = -ENOTCONN;
2148 if (!transport->inet) {
2149 struct sock *sk = sock->sk;
2151 /* Avoid temporary address, they are bad for long-lived
2152 * connections such as NFS mounts.
2153 * RFC4941, section 3.6 suggests that:
2154 * Individual applications, which have specific
2155 * knowledge about the normal duration of connections,
2156 * MAY override this as appropriate.
2158 if (xs_addr(xprt)->sa_family == PF_INET6) {
2159 ip6_sock_set_addr_preferences(sk,
2160 IPV6_PREFER_SRC_PUBLIC);
2163 xs_tcp_set_socket_timeouts(xprt, sock);
2165 write_lock_bh(&sk->sk_callback_lock);
2167 xs_save_old_callbacks(transport, sk);
2169 sk->sk_user_data = xprt;
2170 sk->sk_data_ready = xs_data_ready;
2171 sk->sk_state_change = xs_tcp_state_change;
2172 sk->sk_write_space = xs_tcp_write_space;
2173 sock_set_flag(sk, SOCK_FASYNC);
2174 sk->sk_error_report = xs_error_report;
2176 /* socket options */
2177 sock_reset_flag(sk, SOCK_LINGER);
2178 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2180 xprt_clear_connected(xprt);
2182 /* Reset to new socket */
2183 transport->sock = sock;
2184 transport->inet = sk;
2186 write_unlock_bh(&sk->sk_callback_lock);
2189 if (!xprt_bound(xprt))
2192 xs_set_memalloc(xprt);
2194 xs_stream_start_connect(transport);
2196 /* Tell the socket layer to start connecting... */
2197 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2198 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2201 xs_set_srcport(transport, sock);
2205 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2206 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2208 case -EADDRNOTAVAIL:
2209 /* Source port number is unavailable. Try a new one! */
2210 transport->srcport = 0;
2217 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2218 * @work: queued work item
2220 * Invoked by a work queue tasklet.
2222 static void xs_tcp_setup_socket(struct work_struct *work)
2224 struct sock_xprt *transport =
2225 container_of(work, struct sock_xprt, connect_worker.work);
2226 struct socket *sock = transport->sock;
2227 struct rpc_xprt *xprt = &transport->xprt;
2231 sock = xs_create_sock(xprt, transport,
2232 xs_addr(xprt)->sa_family, SOCK_STREAM,
2235 status = PTR_ERR(sock);
2240 dprintk("RPC: worker connecting xprt %p via %s to "
2241 "%s (port %s)\n", xprt,
2242 xprt->address_strings[RPC_DISPLAY_PROTO],
2243 xprt->address_strings[RPC_DISPLAY_ADDR],
2244 xprt->address_strings[RPC_DISPLAY_PORT]);
2246 status = xs_tcp_finish_connecting(xprt, sock);
2247 trace_rpc_socket_connect(xprt, sock, status);
2248 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2249 xprt, -status, xprt_connected(xprt),
2250 sock->sk->sk_state);
2253 printk("%s: connect returned unhandled error %d\n",
2256 case -EADDRNOTAVAIL:
2257 /* We're probably in TIME_WAIT. Get rid of existing socket,
2260 xs_tcp_force_close(xprt);
2265 xprt_unlock_connect(xprt, transport);
2268 /* Happens, for instance, if the user specified a link
2269 * local IPv6 address without a scope-id.
2279 * xs_tcp_force_close() wakes tasks with -EIO.
2280 * We need to wake them first to ensure the
2281 * correct error code.
2283 xprt_wake_pending_tasks(xprt, status);
2284 xs_tcp_force_close(xprt);
2289 xprt_clear_connecting(xprt);
2290 xprt_unlock_connect(xprt, transport);
2291 xprt_wake_pending_tasks(xprt, status);
2295 * xs_connect - connect a socket to a remote endpoint
2296 * @xprt: pointer to transport structure
2297 * @task: address of RPC task that manages state of connect request
2299 * TCP: If the remote end dropped the connection, delay reconnecting.
2301 * UDP socket connects are synchronous, but we use a work queue anyway
2302 * to guarantee that even unprivileged user processes can set up a
2303 * socket on a privileged port.
2305 * If a UDP socket connect fails, the delay behavior here prevents
2306 * retry floods (hard mounts).
2308 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2310 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2311 unsigned long delay = 0;
2313 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2315 if (transport->sock != NULL) {
2316 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2318 xprt, xprt->reestablish_timeout / HZ);
2320 /* Start by resetting any existing state */
2321 xs_reset_transport(transport);
2323 delay = xprt_reconnect_delay(xprt);
2324 xprt_reconnect_backoff(xprt, XS_TCP_INIT_REEST_TO);
2327 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2329 queue_delayed_work(xprtiod_workqueue,
2330 &transport->connect_worker,
2334 static void xs_wake_disconnect(struct sock_xprt *transport)
2336 if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state))
2337 xs_tcp_force_close(&transport->xprt);
2340 static void xs_wake_write(struct sock_xprt *transport)
2342 if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state))
2343 xprt_write_space(&transport->xprt);
2346 static void xs_wake_error(struct sock_xprt *transport)
2350 if (!test_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2352 mutex_lock(&transport->recv_mutex);
2353 if (transport->sock == NULL)
2355 if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2357 sockerr = xchg(&transport->xprt_err, 0);
2359 xprt_wake_pending_tasks(&transport->xprt, sockerr);
2361 mutex_unlock(&transport->recv_mutex);
2364 static void xs_wake_pending(struct sock_xprt *transport)
2366 if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING, &transport->sock_state))
2367 xprt_wake_pending_tasks(&transport->xprt, -EAGAIN);
2370 static void xs_error_handle(struct work_struct *work)
2372 struct sock_xprt *transport = container_of(work,
2373 struct sock_xprt, error_worker);
2375 xs_wake_disconnect(transport);
2376 xs_wake_write(transport);
2377 xs_wake_error(transport);
2378 xs_wake_pending(transport);
2382 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2383 * @xprt: rpc_xprt struct containing statistics
2387 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2391 if (xprt_connected(xprt))
2392 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2394 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2395 "%llu %llu %lu %llu %llu\n",
2396 xprt->stat.bind_count,
2397 xprt->stat.connect_count,
2398 xprt->stat.connect_time / HZ,
2402 xprt->stat.bad_xids,
2405 xprt->stat.max_slots,
2406 xprt->stat.sending_u,
2407 xprt->stat.pending_u);
2411 * xs_udp_print_stats - display UDP socket-specifc stats
2412 * @xprt: rpc_xprt struct containing statistics
2416 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2418 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2420 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2423 xprt->stat.bind_count,
2426 xprt->stat.bad_xids,
2429 xprt->stat.max_slots,
2430 xprt->stat.sending_u,
2431 xprt->stat.pending_u);
2435 * xs_tcp_print_stats - display TCP socket-specifc stats
2436 * @xprt: rpc_xprt struct containing statistics
2440 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2442 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2445 if (xprt_connected(xprt))
2446 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2448 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2449 "%llu %llu %lu %llu %llu\n",
2451 xprt->stat.bind_count,
2452 xprt->stat.connect_count,
2453 xprt->stat.connect_time / HZ,
2457 xprt->stat.bad_xids,
2460 xprt->stat.max_slots,
2461 xprt->stat.sending_u,
2462 xprt->stat.pending_u);
2466 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2467 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2468 * to use the server side send routines.
2470 static int bc_malloc(struct rpc_task *task)
2472 struct rpc_rqst *rqst = task->tk_rqstp;
2473 size_t size = rqst->rq_callsize;
2475 struct rpc_buffer *buf;
2477 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2478 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2483 page = alloc_page(GFP_KERNEL);
2487 buf = page_address(page);
2488 buf->len = PAGE_SIZE;
2490 rqst->rq_buffer = buf->data;
2491 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2496 * Free the space allocated in the bc_alloc routine
2498 static void bc_free(struct rpc_task *task)
2500 void *buffer = task->tk_rqstp->rq_buffer;
2501 struct rpc_buffer *buf;
2503 buf = container_of(buffer, struct rpc_buffer, data);
2504 free_page((unsigned long)buf);
2507 static int bc_sendto(struct rpc_rqst *req)
2509 struct xdr_buf *xdr = &req->rq_snd_buf;
2510 struct sock_xprt *transport =
2511 container_of(req->rq_xprt, struct sock_xprt, xprt);
2512 struct msghdr msg = {
2515 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
2517 unsigned int sent = 0;
2520 req->rq_xtime = ktime_get();
2521 err = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, marker, &sent);
2523 if (err < 0 || sent != (xdr->len + sizeof(marker)))
2529 * bc_send_request - Send a backchannel Call on a TCP socket
2530 * @req: rpc_rqst containing Call message to be sent
2532 * xpt_mutex ensures @rqstp's whole message is written to the socket
2533 * without interruption.
2536 * %0 if the message was sent successfully
2537 * %ENOTCONN if the message was not sent
2539 static int bc_send_request(struct rpc_rqst *req)
2541 struct svc_xprt *xprt;
2545 * Get the server socket associated with this callback xprt
2547 xprt = req->rq_xprt->bc_xprt;
2550 * Grab the mutex to serialize data as the connection is shared
2551 * with the fore channel
2553 mutex_lock(&xprt->xpt_mutex);
2554 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2557 len = bc_sendto(req);
2558 mutex_unlock(&xprt->xpt_mutex);
2567 * The close routine. Since this is client initiated, we do nothing
2570 static void bc_close(struct rpc_xprt *xprt)
2572 xprt_disconnect_done(xprt);
2576 * The xprt destroy routine. Again, because this connection is client
2577 * initiated, we do nothing
2580 static void bc_destroy(struct rpc_xprt *xprt)
2582 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2585 module_put(THIS_MODULE);
2588 static const struct rpc_xprt_ops xs_local_ops = {
2589 .reserve_xprt = xprt_reserve_xprt,
2590 .release_xprt = xprt_release_xprt,
2591 .alloc_slot = xprt_alloc_slot,
2592 .free_slot = xprt_free_slot,
2593 .rpcbind = xs_local_rpcbind,
2594 .set_port = xs_local_set_port,
2595 .connect = xs_local_connect,
2596 .buf_alloc = rpc_malloc,
2597 .buf_free = rpc_free,
2598 .prepare_request = xs_stream_prepare_request,
2599 .send_request = xs_local_send_request,
2600 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2602 .destroy = xs_destroy,
2603 .print_stats = xs_local_print_stats,
2604 .enable_swap = xs_enable_swap,
2605 .disable_swap = xs_disable_swap,
2608 static const struct rpc_xprt_ops xs_udp_ops = {
2609 .set_buffer_size = xs_udp_set_buffer_size,
2610 .reserve_xprt = xprt_reserve_xprt_cong,
2611 .release_xprt = xprt_release_xprt_cong,
2612 .alloc_slot = xprt_alloc_slot,
2613 .free_slot = xprt_free_slot,
2614 .rpcbind = rpcb_getport_async,
2615 .set_port = xs_set_port,
2616 .connect = xs_connect,
2617 .buf_alloc = rpc_malloc,
2618 .buf_free = rpc_free,
2619 .send_request = xs_udp_send_request,
2620 .wait_for_reply_request = xprt_wait_for_reply_request_rtt,
2621 .timer = xs_udp_timer,
2622 .release_request = xprt_release_rqst_cong,
2624 .destroy = xs_destroy,
2625 .print_stats = xs_udp_print_stats,
2626 .enable_swap = xs_enable_swap,
2627 .disable_swap = xs_disable_swap,
2628 .inject_disconnect = xs_inject_disconnect,
2631 static const struct rpc_xprt_ops xs_tcp_ops = {
2632 .reserve_xprt = xprt_reserve_xprt,
2633 .release_xprt = xprt_release_xprt,
2634 .alloc_slot = xprt_alloc_slot,
2635 .free_slot = xprt_free_slot,
2636 .rpcbind = rpcb_getport_async,
2637 .set_port = xs_set_port,
2638 .connect = xs_connect,
2639 .buf_alloc = rpc_malloc,
2640 .buf_free = rpc_free,
2641 .prepare_request = xs_stream_prepare_request,
2642 .send_request = xs_tcp_send_request,
2643 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2644 .close = xs_tcp_shutdown,
2645 .destroy = xs_destroy,
2646 .set_connect_timeout = xs_tcp_set_connect_timeout,
2647 .print_stats = xs_tcp_print_stats,
2648 .enable_swap = xs_enable_swap,
2649 .disable_swap = xs_disable_swap,
2650 .inject_disconnect = xs_inject_disconnect,
2651 #ifdef CONFIG_SUNRPC_BACKCHANNEL
2652 .bc_setup = xprt_setup_bc,
2653 .bc_maxpayload = xs_tcp_bc_maxpayload,
2654 .bc_num_slots = xprt_bc_max_slots,
2655 .bc_free_rqst = xprt_free_bc_rqst,
2656 .bc_destroy = xprt_destroy_bc,
2661 * The rpc_xprt_ops for the server backchannel
2664 static const struct rpc_xprt_ops bc_tcp_ops = {
2665 .reserve_xprt = xprt_reserve_xprt,
2666 .release_xprt = xprt_release_xprt,
2667 .alloc_slot = xprt_alloc_slot,
2668 .free_slot = xprt_free_slot,
2669 .buf_alloc = bc_malloc,
2670 .buf_free = bc_free,
2671 .send_request = bc_send_request,
2672 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2674 .destroy = bc_destroy,
2675 .print_stats = xs_tcp_print_stats,
2676 .enable_swap = xs_enable_swap,
2677 .disable_swap = xs_disable_swap,
2678 .inject_disconnect = xs_inject_disconnect,
2681 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2683 static const struct sockaddr_in sin = {
2684 .sin_family = AF_INET,
2685 .sin_addr.s_addr = htonl(INADDR_ANY),
2687 static const struct sockaddr_in6 sin6 = {
2688 .sin6_family = AF_INET6,
2689 .sin6_addr = IN6ADDR_ANY_INIT,
2696 memcpy(sap, &sin, sizeof(sin));
2699 memcpy(sap, &sin6, sizeof(sin6));
2702 dprintk("RPC: %s: Bad address family\n", __func__);
2703 return -EAFNOSUPPORT;
2708 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2709 unsigned int slot_table_size,
2710 unsigned int max_slot_table_size)
2712 struct rpc_xprt *xprt;
2713 struct sock_xprt *new;
2715 if (args->addrlen > sizeof(xprt->addr)) {
2716 dprintk("RPC: xs_setup_xprt: address too large\n");
2717 return ERR_PTR(-EBADF);
2720 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2721 max_slot_table_size);
2723 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2725 return ERR_PTR(-ENOMEM);
2728 new = container_of(xprt, struct sock_xprt, xprt);
2729 mutex_init(&new->recv_mutex);
2730 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2731 xprt->addrlen = args->addrlen;
2733 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2736 err = xs_init_anyaddr(args->dstaddr->sa_family,
2737 (struct sockaddr *)&new->srcaddr);
2740 return ERR_PTR(err);
2747 static const struct rpc_timeout xs_local_default_timeout = {
2748 .to_initval = 10 * HZ,
2749 .to_maxval = 10 * HZ,
2754 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2755 * @args: rpc transport creation arguments
2757 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2759 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2761 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2762 struct sock_xprt *transport;
2763 struct rpc_xprt *xprt;
2764 struct rpc_xprt *ret;
2766 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2767 xprt_max_tcp_slot_table_entries);
2770 transport = container_of(xprt, struct sock_xprt, xprt);
2773 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2775 xprt->bind_timeout = XS_BIND_TO;
2776 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2777 xprt->idle_timeout = XS_IDLE_DISC_TO;
2779 xprt->ops = &xs_local_ops;
2780 xprt->timeout = &xs_local_default_timeout;
2782 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2783 INIT_WORK(&transport->error_worker, xs_error_handle);
2784 INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket);
2786 switch (sun->sun_family) {
2788 if (sun->sun_path[0] != '/') {
2789 dprintk("RPC: bad AF_LOCAL address: %s\n",
2791 ret = ERR_PTR(-EINVAL);
2794 xprt_set_bound(xprt);
2795 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2796 ret = ERR_PTR(xs_local_setup_socket(transport));
2801 ret = ERR_PTR(-EAFNOSUPPORT);
2805 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2806 xprt->address_strings[RPC_DISPLAY_ADDR]);
2808 if (try_module_get(THIS_MODULE))
2810 ret = ERR_PTR(-EINVAL);
2816 static const struct rpc_timeout xs_udp_default_timeout = {
2817 .to_initval = 5 * HZ,
2818 .to_maxval = 30 * HZ,
2819 .to_increment = 5 * HZ,
2824 * xs_setup_udp - Set up transport to use a UDP socket
2825 * @args: rpc transport creation arguments
2828 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2830 struct sockaddr *addr = args->dstaddr;
2831 struct rpc_xprt *xprt;
2832 struct sock_xprt *transport;
2833 struct rpc_xprt *ret;
2835 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2836 xprt_udp_slot_table_entries);
2839 transport = container_of(xprt, struct sock_xprt, xprt);
2841 xprt->prot = IPPROTO_UDP;
2842 /* XXX: header size can vary due to auth type, IPv6, etc. */
2843 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2845 xprt->bind_timeout = XS_BIND_TO;
2846 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2847 xprt->idle_timeout = XS_IDLE_DISC_TO;
2849 xprt->ops = &xs_udp_ops;
2851 xprt->timeout = &xs_udp_default_timeout;
2853 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
2854 INIT_WORK(&transport->error_worker, xs_error_handle);
2855 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2857 switch (addr->sa_family) {
2859 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2860 xprt_set_bound(xprt);
2862 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2865 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2866 xprt_set_bound(xprt);
2868 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2871 ret = ERR_PTR(-EAFNOSUPPORT);
2875 if (xprt_bound(xprt))
2876 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2877 xprt->address_strings[RPC_DISPLAY_ADDR],
2878 xprt->address_strings[RPC_DISPLAY_PORT],
2879 xprt->address_strings[RPC_DISPLAY_PROTO]);
2881 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2882 xprt->address_strings[RPC_DISPLAY_ADDR],
2883 xprt->address_strings[RPC_DISPLAY_PROTO]);
2885 if (try_module_get(THIS_MODULE))
2887 ret = ERR_PTR(-EINVAL);
2893 static const struct rpc_timeout xs_tcp_default_timeout = {
2894 .to_initval = 60 * HZ,
2895 .to_maxval = 60 * HZ,
2900 * xs_setup_tcp - Set up transport to use a TCP socket
2901 * @args: rpc transport creation arguments
2904 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2906 struct sockaddr *addr = args->dstaddr;
2907 struct rpc_xprt *xprt;
2908 struct sock_xprt *transport;
2909 struct rpc_xprt *ret;
2910 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2912 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2913 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2915 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2916 max_slot_table_size);
2919 transport = container_of(xprt, struct sock_xprt, xprt);
2921 xprt->prot = IPPROTO_TCP;
2922 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2924 xprt->bind_timeout = XS_BIND_TO;
2925 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2926 xprt->idle_timeout = XS_IDLE_DISC_TO;
2928 xprt->ops = &xs_tcp_ops;
2929 xprt->timeout = &xs_tcp_default_timeout;
2931 xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
2932 xprt->connect_timeout = xprt->timeout->to_initval *
2933 (xprt->timeout->to_retries + 1);
2935 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2936 INIT_WORK(&transport->error_worker, xs_error_handle);
2937 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
2939 switch (addr->sa_family) {
2941 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2942 xprt_set_bound(xprt);
2944 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2947 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2948 xprt_set_bound(xprt);
2950 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2953 ret = ERR_PTR(-EAFNOSUPPORT);
2957 if (xprt_bound(xprt))
2958 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2959 xprt->address_strings[RPC_DISPLAY_ADDR],
2960 xprt->address_strings[RPC_DISPLAY_PORT],
2961 xprt->address_strings[RPC_DISPLAY_PROTO]);
2963 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2964 xprt->address_strings[RPC_DISPLAY_ADDR],
2965 xprt->address_strings[RPC_DISPLAY_PROTO]);
2967 if (try_module_get(THIS_MODULE))
2969 ret = ERR_PTR(-EINVAL);
2976 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2977 * @args: rpc transport creation arguments
2980 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2982 struct sockaddr *addr = args->dstaddr;
2983 struct rpc_xprt *xprt;
2984 struct sock_xprt *transport;
2985 struct svc_sock *bc_sock;
2986 struct rpc_xprt *ret;
2988 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2989 xprt_tcp_slot_table_entries);
2992 transport = container_of(xprt, struct sock_xprt, xprt);
2994 xprt->prot = IPPROTO_TCP;
2995 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2996 xprt->timeout = &xs_tcp_default_timeout;
2999 xprt_set_bound(xprt);
3000 xprt->bind_timeout = 0;
3001 xprt->reestablish_timeout = 0;
3002 xprt->idle_timeout = 0;
3004 xprt->ops = &bc_tcp_ops;
3006 switch (addr->sa_family) {
3008 xs_format_peer_addresses(xprt, "tcp",
3012 xs_format_peer_addresses(xprt, "tcp",
3013 RPCBIND_NETID_TCP6);
3016 ret = ERR_PTR(-EAFNOSUPPORT);
3020 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3021 xprt->address_strings[RPC_DISPLAY_ADDR],
3022 xprt->address_strings[RPC_DISPLAY_PORT],
3023 xprt->address_strings[RPC_DISPLAY_PROTO]);
3026 * Once we've associated a backchannel xprt with a connection,
3027 * we want to keep it around as long as the connection lasts,
3028 * in case we need to start using it for a backchannel again;
3029 * this reference won't be dropped until bc_xprt is destroyed.
3032 args->bc_xprt->xpt_bc_xprt = xprt;
3033 xprt->bc_xprt = args->bc_xprt;
3034 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3035 transport->sock = bc_sock->sk_sock;
3036 transport->inet = bc_sock->sk_sk;
3039 * Since we don't want connections for the backchannel, we set
3040 * the xprt status to connected
3042 xprt_set_connected(xprt);
3044 if (try_module_get(THIS_MODULE))
3047 args->bc_xprt->xpt_bc_xprt = NULL;
3048 args->bc_xprt->xpt_bc_xps = NULL;
3050 ret = ERR_PTR(-EINVAL);
3056 static struct xprt_class xs_local_transport = {
3057 .list = LIST_HEAD_INIT(xs_local_transport.list),
3058 .name = "named UNIX socket",
3059 .owner = THIS_MODULE,
3060 .ident = XPRT_TRANSPORT_LOCAL,
3061 .setup = xs_setup_local,
3064 static struct xprt_class xs_udp_transport = {
3065 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3067 .owner = THIS_MODULE,
3068 .ident = XPRT_TRANSPORT_UDP,
3069 .setup = xs_setup_udp,
3072 static struct xprt_class xs_tcp_transport = {
3073 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3075 .owner = THIS_MODULE,
3076 .ident = XPRT_TRANSPORT_TCP,
3077 .setup = xs_setup_tcp,
3080 static struct xprt_class xs_bc_tcp_transport = {
3081 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3082 .name = "tcp NFSv4.1 backchannel",
3083 .owner = THIS_MODULE,
3084 .ident = XPRT_TRANSPORT_BC_TCP,
3085 .setup = xs_setup_bc_tcp,
3089 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3092 int init_socket_xprt(void)
3094 if (!sunrpc_table_header)
3095 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3097 xprt_register_transport(&xs_local_transport);
3098 xprt_register_transport(&xs_udp_transport);
3099 xprt_register_transport(&xs_tcp_transport);
3100 xprt_register_transport(&xs_bc_tcp_transport);
3106 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3109 void cleanup_socket_xprt(void)
3111 if (sunrpc_table_header) {
3112 unregister_sysctl_table(sunrpc_table_header);
3113 sunrpc_table_header = NULL;
3116 xprt_unregister_transport(&xs_local_transport);
3117 xprt_unregister_transport(&xs_udp_transport);
3118 xprt_unregister_transport(&xs_tcp_transport);
3119 xprt_unregister_transport(&xs_bc_tcp_transport);
3122 static int param_set_uint_minmax(const char *val,
3123 const struct kernel_param *kp,
3124 unsigned int min, unsigned int max)
3131 ret = kstrtouint(val, 0, &num);
3134 if (num < min || num > max)
3136 *((unsigned int *)kp->arg) = num;
3140 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3142 return param_set_uint_minmax(val, kp,
3147 static const struct kernel_param_ops param_ops_portnr = {
3148 .set = param_set_portnr,
3149 .get = param_get_uint,
3152 #define param_check_portnr(name, p) \
3153 __param_check(name, p, unsigned int);
3155 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3156 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3158 static int param_set_slot_table_size(const char *val,
3159 const struct kernel_param *kp)
3161 return param_set_uint_minmax(val, kp,
3163 RPC_MAX_SLOT_TABLE);
3166 static const struct kernel_param_ops param_ops_slot_table_size = {
3167 .set = param_set_slot_table_size,
3168 .get = param_get_uint,
3171 #define param_check_slot_table_size(name, p) \
3172 __param_check(name, p, unsigned int);
3174 static int param_set_max_slot_table_size(const char *val,
3175 const struct kernel_param *kp)
3177 return param_set_uint_minmax(val, kp,
3179 RPC_MAX_SLOT_TABLE_LIMIT);
3182 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3183 .set = param_set_max_slot_table_size,
3184 .get = param_get_uint,
3187 #define param_check_max_slot_table_size(name, p) \
3188 __param_check(name, p, unsigned int);
3190 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3191 slot_table_size, 0644);
3192 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3193 max_slot_table_size, 0644);
3194 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3195 slot_table_size, 0644);