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/uio.h>
53 #include <trace/events/sunrpc.h>
57 static void xs_close(struct rpc_xprt *xprt);
58 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
64 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
65 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
66 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
68 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
69 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
71 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
73 #define XS_TCP_LINGER_TO (15U * HZ)
74 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
77 * We can register our own files under /proc/sys/sunrpc by
78 * calling register_sysctl_table() again. The files in that
79 * directory become the union of all files registered there.
81 * We simply need to make sure that we don't collide with
82 * someone else's file names!
85 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
86 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
87 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
88 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
89 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
91 static struct ctl_table_header *sunrpc_table_header;
94 * FIXME: changing the UDP slot table size should also resize the UDP
95 * socket buffers for existing UDP transports
97 static struct ctl_table xs_tunables_table[] = {
99 .procname = "udp_slot_table_entries",
100 .data = &xprt_udp_slot_table_entries,
101 .maxlen = sizeof(unsigned int),
103 .proc_handler = proc_dointvec_minmax,
104 .extra1 = &min_slot_table_size,
105 .extra2 = &max_slot_table_size
108 .procname = "tcp_slot_table_entries",
109 .data = &xprt_tcp_slot_table_entries,
110 .maxlen = sizeof(unsigned int),
112 .proc_handler = proc_dointvec_minmax,
113 .extra1 = &min_slot_table_size,
114 .extra2 = &max_slot_table_size
117 .procname = "tcp_max_slot_table_entries",
118 .data = &xprt_max_tcp_slot_table_entries,
119 .maxlen = sizeof(unsigned int),
121 .proc_handler = proc_dointvec_minmax,
122 .extra1 = &min_slot_table_size,
123 .extra2 = &max_tcp_slot_table_limit
126 .procname = "min_resvport",
127 .data = &xprt_min_resvport,
128 .maxlen = sizeof(unsigned int),
130 .proc_handler = proc_dointvec_minmax,
131 .extra1 = &xprt_min_resvport_limit,
132 .extra2 = &xprt_max_resvport_limit
135 .procname = "max_resvport",
136 .data = &xprt_max_resvport,
137 .maxlen = sizeof(unsigned int),
139 .proc_handler = proc_dointvec_minmax,
140 .extra1 = &xprt_min_resvport_limit,
141 .extra2 = &xprt_max_resvport_limit
144 .procname = "tcp_fin_timeout",
145 .data = &xs_tcp_fin_timeout,
146 .maxlen = sizeof(xs_tcp_fin_timeout),
148 .proc_handler = proc_dointvec_jiffies,
153 static struct ctl_table sunrpc_table[] = {
155 .procname = "sunrpc",
157 .child = xs_tunables_table
165 * Wait duration for a reply from the RPC portmapper.
167 #define XS_BIND_TO (60U * HZ)
170 * Delay if a UDP socket connect error occurs. This is most likely some
171 * kind of resource problem on the local host.
173 #define XS_UDP_REEST_TO (2U * HZ)
176 * The reestablish timeout allows clients to delay for a bit before attempting
177 * to reconnect to a server that just dropped our connection.
179 * We implement an exponential backoff when trying to reestablish a TCP
180 * transport connection with the server. Some servers like to drop a TCP
181 * connection when they are overworked, so we start with a short timeout and
182 * increase over time if the server is down or not responding.
184 #define XS_TCP_INIT_REEST_TO (3U * HZ)
187 * TCP idle timeout; client drops the transport socket if it is idle
188 * for this long. Note that we also timeout UDP sockets to prevent
189 * holding port numbers when there is no RPC traffic.
191 #define XS_IDLE_DISC_TO (5U * 60 * HZ)
193 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
194 # undef RPC_DEBUG_DATA
195 # define RPCDBG_FACILITY RPCDBG_TRANS
198 #ifdef RPC_DEBUG_DATA
199 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
201 u8 *buf = (u8 *) packet;
204 dprintk("RPC: %s\n", msg);
205 for (j = 0; j < count && j < 128; j += 4) {
209 dprintk("0x%04x ", j);
211 dprintk("%02x%02x%02x%02x ",
212 buf[j], buf[j+1], buf[j+2], buf[j+3]);
217 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
223 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
225 return (struct rpc_xprt *) sk->sk_user_data;
228 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
230 return (struct sockaddr *) &xprt->addr;
233 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
235 return (struct sockaddr_un *) &xprt->addr;
238 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
240 return (struct sockaddr_in *) &xprt->addr;
243 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
245 return (struct sockaddr_in6 *) &xprt->addr;
248 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
250 struct sockaddr *sap = xs_addr(xprt);
251 struct sockaddr_in6 *sin6;
252 struct sockaddr_in *sin;
253 struct sockaddr_un *sun;
256 switch (sap->sa_family) {
258 sun = xs_addr_un(xprt);
259 strlcpy(buf, sun->sun_path, sizeof(buf));
260 xprt->address_strings[RPC_DISPLAY_ADDR] =
261 kstrdup(buf, GFP_KERNEL);
264 (void)rpc_ntop(sap, buf, sizeof(buf));
265 xprt->address_strings[RPC_DISPLAY_ADDR] =
266 kstrdup(buf, GFP_KERNEL);
267 sin = xs_addr_in(xprt);
268 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
271 (void)rpc_ntop(sap, buf, sizeof(buf));
272 xprt->address_strings[RPC_DISPLAY_ADDR] =
273 kstrdup(buf, GFP_KERNEL);
274 sin6 = xs_addr_in6(xprt);
275 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
281 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
284 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
286 struct sockaddr *sap = xs_addr(xprt);
289 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
290 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
292 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
293 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
296 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
297 const char *protocol,
300 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
301 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
302 xs_format_common_peer_addresses(xprt);
303 xs_format_common_peer_ports(xprt);
306 static void xs_update_peer_port(struct rpc_xprt *xprt)
308 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
309 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
311 xs_format_common_peer_ports(xprt);
314 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
318 for (i = 0; i < RPC_DISPLAY_MAX; i++)
320 case RPC_DISPLAY_PROTO:
321 case RPC_DISPLAY_NETID:
324 kfree(xprt->address_strings[i]);
329 xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp)
333 if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES))
335 n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT;
336 for (i = 0; i < n; i++) {
339 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp);
340 if (!buf->pages[i]) {
342 return i > buf->page_base ? i - buf->page_base : 0;
349 xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek)
353 iov_iter_advance(&msg->msg_iter, seek);
354 ret = sock_recvmsg(sock, msg, flags);
355 return ret > 0 ? ret + seek : ret;
359 xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags,
360 struct kvec *kvec, size_t count, size_t seek)
362 iov_iter_kvec(&msg->msg_iter, READ, kvec, 1, count);
363 return xs_sock_recvmsg(sock, msg, flags, seek);
367 xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags,
368 struct bio_vec *bvec, unsigned long nr, size_t count,
371 iov_iter_bvec(&msg->msg_iter, READ, bvec, nr, count);
372 return xs_sock_recvmsg(sock, msg, flags, seek);
376 xs_read_discard(struct socket *sock, struct msghdr *msg, int flags,
379 iov_iter_discard(&msg->msg_iter, READ, count);
380 return sock_recvmsg(sock, msg, flags);
384 xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
385 struct xdr_buf *buf, size_t count, size_t seek, size_t *read)
387 size_t want, seek_init = seek, offset = 0;
390 if (seek < buf->head[0].iov_len) {
391 want = min_t(size_t, count, buf->head[0].iov_len);
392 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek);
396 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
402 seek -= buf->head[0].iov_len;
403 offset += buf->head[0].iov_len;
406 want = xs_alloc_sparse_pages(buf,
407 min_t(size_t, count - offset, buf->page_len),
410 ret = xs_read_bvec(sock, msg, flags, buf->bvec,
411 xdr_buf_pagecount(buf),
412 want + buf->page_base,
413 seek + buf->page_base);
416 offset += ret - buf->page_base;
417 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
427 if (seek < buf->tail[0].iov_len) {
428 want = min_t(size_t, count - offset, buf->tail[0].iov_len);
429 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek);
433 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
438 offset += buf->tail[0].iov_len;
441 *read = offset - seek_init;
449 xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf)
451 if (!transport->recv.copied) {
452 if (buf->head[0].iov_len >= transport->recv.offset)
453 memcpy(buf->head[0].iov_base,
454 &transport->recv.xid,
455 transport->recv.offset);
456 transport->recv.copied = transport->recv.offset;
461 xs_read_stream_request_done(struct sock_xprt *transport)
463 return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT);
467 xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg,
468 int flags, struct rpc_rqst *req)
470 struct xdr_buf *buf = &req->rq_private_buf;
474 xs_read_header(transport, buf);
476 want = transport->recv.len - transport->recv.offset;
477 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf,
478 transport->recv.copied + want, transport->recv.copied,
480 transport->recv.offset += read;
481 transport->recv.copied += read;
482 if (transport->recv.offset == transport->recv.len) {
483 if (xs_read_stream_request_done(transport))
484 msg->msg_flags |= MSG_EOR;
493 msg->msg_flags |= MSG_TRUNC;
498 return ret < 0 ? ret : read;
502 xs_read_stream_headersize(bool isfrag)
505 return sizeof(__be32);
506 return 3 * sizeof(__be32);
510 xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg,
511 int flags, size_t want, size_t seek)
514 .iov_base = &transport->recv.fraghdr,
517 return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek);
520 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
522 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
524 struct rpc_xprt *xprt = &transport->xprt;
525 struct rpc_rqst *req;
528 /* Look up and lock the request corresponding to the given XID */
529 req = xprt_lookup_bc_request(xprt, transport->recv.xid);
531 printk(KERN_WARNING "Callback slot table overflowed\n");
535 ret = xs_read_stream_request(transport, msg, flags, req);
536 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
537 xprt_complete_bc_request(req, transport->recv.copied);
541 #else /* CONFIG_SUNRPC_BACKCHANNEL */
543 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
547 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
550 xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags)
552 struct rpc_xprt *xprt = &transport->xprt;
553 struct rpc_rqst *req;
556 /* Look up and lock the request corresponding to the given XID */
557 spin_lock(&xprt->queue_lock);
558 req = xprt_lookup_rqst(xprt, transport->recv.xid);
560 msg->msg_flags |= MSG_TRUNC;
564 spin_unlock(&xprt->queue_lock);
566 ret = xs_read_stream_request(transport, msg, flags, req);
568 spin_lock(&xprt->queue_lock);
569 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
570 xprt_complete_rqst(req->rq_task, transport->recv.copied);
571 xprt_unpin_rqst(req);
573 spin_unlock(&xprt->queue_lock);
578 xs_read_stream(struct sock_xprt *transport, int flags)
580 struct msghdr msg = { 0 };
581 size_t want, read = 0;
584 if (transport->recv.len == 0) {
585 want = xs_read_stream_headersize(transport->recv.copied != 0);
586 ret = xs_read_stream_header(transport, &msg, flags, want,
587 transport->recv.offset);
590 transport->recv.offset = ret;
591 if (transport->recv.offset != want)
592 return transport->recv.offset;
593 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) &
594 RPC_FRAGMENT_SIZE_MASK;
595 transport->recv.offset -= sizeof(transport->recv.fraghdr);
599 switch (be32_to_cpu(transport->recv.calldir)) {
601 msg.msg_flags |= MSG_TRUNC;
604 ret = xs_read_stream_call(transport, &msg, flags);
607 ret = xs_read_stream_reply(transport, &msg, flags);
609 if (msg.msg_flags & MSG_TRUNC) {
610 transport->recv.calldir = cpu_to_be32(-1);
611 transport->recv.copied = -1;
616 if (transport->recv.offset < transport->recv.len) {
617 if (!(msg.msg_flags & MSG_TRUNC))
620 ret = xs_read_discard(transport->sock, &msg, flags,
621 transport->recv.len - transport->recv.offset);
624 transport->recv.offset += ret;
626 if (transport->recv.offset != transport->recv.len)
629 if (xs_read_stream_request_done(transport)) {
630 trace_xs_stream_read_request(transport);
631 transport->recv.copied = 0;
633 transport->recv.offset = 0;
634 transport->recv.len = 0;
637 return ret != 0 ? ret : -ESHUTDOWN;
640 static void xs_stream_data_receive(struct sock_xprt *transport)
645 mutex_lock(&transport->recv_mutex);
646 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
647 if (transport->sock == NULL)
650 ret = xs_read_stream(transport, MSG_DONTWAIT);
657 mutex_unlock(&transport->recv_mutex);
658 trace_xs_stream_read_data(&transport->xprt, ret, read);
661 static void xs_stream_data_receive_workfn(struct work_struct *work)
663 struct sock_xprt *transport =
664 container_of(work, struct sock_xprt, recv_worker);
665 xs_stream_data_receive(transport);
669 xs_stream_reset_connect(struct sock_xprt *transport)
671 transport->recv.offset = 0;
672 transport->recv.len = 0;
673 transport->recv.copied = 0;
674 transport->xmit.offset = 0;
675 transport->xprt.stat.connect_count++;
676 transport->xprt.stat.connect_start = jiffies;
679 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
681 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
683 struct msghdr msg = {
685 .msg_namelen = addrlen,
686 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
689 .iov_base = vec->iov_base + base,
690 .iov_len = vec->iov_len - base,
693 if (iov.iov_len != 0)
694 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
695 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
698 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p)
700 ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
701 int offset, size_t size, int flags);
703 unsigned int remainder;
706 remainder = xdr->page_len - base;
707 base += xdr->page_base;
708 ppage = xdr->pages + (base >> PAGE_SHIFT);
710 do_sendpage = sock->ops->sendpage;
712 do_sendpage = sock_no_sendpage;
714 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
715 int flags = XS_SENDMSG_FLAGS;
721 flags |= MSG_SENDPAGE_NOTLAST | MSG_MORE;
722 err = do_sendpage(sock, *ppage, base, len, flags);
723 if (remainder == 0 || err != len)
737 * xs_sendpages - write pages directly to a socket
738 * @sock: socket to send on
739 * @addr: UDP only -- address of destination
740 * @addrlen: UDP only -- length of destination address
741 * @xdr: buffer containing this request
742 * @base: starting position in the buffer
743 * @zerocopy: true if it is safe to use sendpage()
744 * @sent_p: return the total number of bytes successfully queued for sending
747 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p)
749 unsigned int remainder = xdr->len - base;
761 if (base < xdr->head[0].iov_len || addr != NULL) {
762 unsigned int len = xdr->head[0].iov_len - base;
764 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
765 if (remainder == 0 || err != len)
770 base -= xdr->head[0].iov_len;
772 if (base < xdr->page_len) {
773 unsigned int len = xdr->page_len - base;
775 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy, &sent);
777 if (remainder == 0 || sent != len)
781 base -= xdr->page_len;
783 if (base >= xdr->tail[0].iov_len)
785 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
795 * xs_nospace - handle transmit was incomplete
796 * @req: pointer to RPC request
799 static int xs_nospace(struct rpc_rqst *req)
801 struct rpc_xprt *xprt = req->rq_xprt;
802 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
803 struct sock *sk = transport->inet;
806 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
807 req->rq_task->tk_pid,
808 req->rq_slen - transport->xmit.offset,
811 /* Protect against races with write_space */
812 spin_lock_bh(&xprt->transport_lock);
814 /* Don't race with disconnect */
815 if (xprt_connected(xprt)) {
816 /* wait for more buffer space */
817 sk->sk_write_pending++;
818 xprt_wait_for_buffer_space(xprt);
822 spin_unlock_bh(&xprt->transport_lock);
824 /* Race breaker in case memory is freed before above code is called */
825 if (ret == -EAGAIN) {
826 struct socket_wq *wq;
829 wq = rcu_dereference(sk->sk_wq);
830 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
833 sk->sk_write_space(sk);
839 xs_stream_prepare_request(struct rpc_rqst *req)
841 req->rq_task->tk_status = xdr_alloc_bvec(&req->rq_rcv_buf, GFP_NOIO);
845 * Determine if the previous message in the stream was aborted before it
846 * could complete transmission.
849 xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req)
851 return transport->xmit.offset != 0 && req->rq_bytes_sent == 0;
855 * Construct a stream transport record marker in @buf.
857 static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
859 u32 reclen = buf->len - sizeof(rpc_fraghdr);
860 rpc_fraghdr *base = buf->head[0].iov_base;
861 *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
865 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
866 * @req: pointer to RPC request
869 * 0: The request has been sent
870 * EAGAIN: The socket was blocked, please call again later to
871 * complete the request
872 * ENOTCONN: Caller needs to invoke connect logic then call again
873 * other: Some other error occured, the request was not sent
875 static int xs_local_send_request(struct rpc_rqst *req)
877 struct rpc_xprt *xprt = req->rq_xprt;
878 struct sock_xprt *transport =
879 container_of(xprt, struct sock_xprt, xprt);
880 struct xdr_buf *xdr = &req->rq_snd_buf;
884 /* Close the stream if the previous transmission was incomplete */
885 if (xs_send_request_was_aborted(transport, req)) {
890 xs_encode_stream_record_marker(&req->rq_snd_buf);
892 xs_pktdump("packet data:",
893 req->rq_svec->iov_base, req->rq_svec->iov_len);
895 req->rq_xtime = ktime_get();
896 status = xs_sendpages(transport->sock, NULL, 0, xdr,
897 transport->xmit.offset,
899 dprintk("RPC: %s(%u) = %d\n",
900 __func__, xdr->len - transport->xmit.offset, status);
902 if (status == -EAGAIN && sock_writeable(transport->inet))
905 if (likely(sent > 0) || status == 0) {
906 transport->xmit.offset += sent;
907 req->rq_bytes_sent = transport->xmit.offset;
908 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
909 req->rq_xmit_bytes_sent += transport->xmit.offset;
910 req->rq_bytes_sent = 0;
911 transport->xmit.offset = 0;
921 status = xs_nospace(req);
924 dprintk("RPC: sendmsg returned unrecognized error %d\n",
936 * xs_udp_send_request - write an RPC request to a UDP socket
937 * @req: pointer to RPC request
940 * 0: The request has been sent
941 * EAGAIN: The socket was blocked, please call again later to
942 * complete the request
943 * ENOTCONN: Caller needs to invoke connect logic then call again
944 * other: Some other error occurred, the request was not sent
946 static int xs_udp_send_request(struct rpc_rqst *req)
948 struct rpc_xprt *xprt = req->rq_xprt;
949 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
950 struct xdr_buf *xdr = &req->rq_snd_buf;
954 xs_pktdump("packet data:",
955 req->rq_svec->iov_base,
956 req->rq_svec->iov_len);
958 if (!xprt_bound(xprt))
961 if (!xprt_request_get_cong(xprt, req))
964 req->rq_xtime = ktime_get();
965 status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
966 xdr, 0, true, &sent);
968 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
971 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
972 if (status == -EPERM)
975 if (status == -EAGAIN && sock_writeable(transport->inet))
978 if (sent > 0 || status == 0) {
979 req->rq_xmit_bytes_sent += sent;
980 if (sent >= req->rq_slen)
982 /* Still some bytes left; set up for a retry later. */
990 /* Should we call xs_close() here? */
993 status = xs_nospace(req);
1000 /* When the server has died, an ICMP port unreachable message
1001 * prompts ECONNREFUSED. */
1004 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1012 * xs_tcp_send_request - write an RPC request to a TCP socket
1013 * @req: pointer to RPC request
1016 * 0: The request has been sent
1017 * EAGAIN: The socket was blocked, please call again later to
1018 * complete the request
1019 * ENOTCONN: Caller needs to invoke connect logic then call again
1020 * other: Some other error occurred, the request was not sent
1022 * XXX: In the case of soft timeouts, should we eventually give up
1023 * if sendmsg is not able to make progress?
1025 static int xs_tcp_send_request(struct rpc_rqst *req)
1027 struct rpc_xprt *xprt = req->rq_xprt;
1028 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1029 struct xdr_buf *xdr = &req->rq_snd_buf;
1030 bool zerocopy = true;
1031 bool vm_wait = false;
1035 /* Close the stream if the previous transmission was incomplete */
1036 if (xs_send_request_was_aborted(transport, req)) {
1037 if (transport->sock != NULL)
1038 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
1042 xs_encode_stream_record_marker(&req->rq_snd_buf);
1044 xs_pktdump("packet data:",
1045 req->rq_svec->iov_base,
1046 req->rq_svec->iov_len);
1047 /* Don't use zero copy if this is a resend. If the RPC call
1048 * completes while the socket holds a reference to the pages,
1049 * then we may end up resending corrupted data.
1051 if (req->rq_task->tk_flags & RPC_TASK_SENT)
1054 if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
1055 xs_tcp_set_socket_timeouts(xprt, transport->sock);
1057 /* Continue transmitting the packet/record. We must be careful
1058 * to cope with writespace callbacks arriving _after_ we have
1059 * called sendmsg(). */
1060 req->rq_xtime = ktime_get();
1063 status = xs_sendpages(transport->sock, NULL, 0, xdr,
1064 transport->xmit.offset,
1067 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
1068 xdr->len - transport->xmit.offset, status);
1070 /* If we've sent the entire packet, immediately
1071 * reset the count of bytes sent. */
1072 transport->xmit.offset += sent;
1073 req->rq_bytes_sent = transport->xmit.offset;
1074 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
1075 req->rq_xmit_bytes_sent += transport->xmit.offset;
1076 req->rq_bytes_sent = 0;
1077 transport->xmit.offset = 0;
1081 WARN_ON_ONCE(sent == 0 && status == 0);
1083 if (status == -EAGAIN ) {
1085 * Return EAGAIN if we're sure we're hitting the
1086 * socket send buffer limits.
1088 if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
1091 * Did we hit a memory allocation failure?
1097 /* Retry, knowing now that we're below the
1098 * socket send buffer limit
1112 /* Should we call xs_close() here? */
1115 status = xs_nospace(req);
1125 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1132 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1134 transport->old_data_ready = sk->sk_data_ready;
1135 transport->old_state_change = sk->sk_state_change;
1136 transport->old_write_space = sk->sk_write_space;
1137 transport->old_error_report = sk->sk_error_report;
1140 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1142 sk->sk_data_ready = transport->old_data_ready;
1143 sk->sk_state_change = transport->old_state_change;
1144 sk->sk_write_space = transport->old_write_space;
1145 sk->sk_error_report = transport->old_error_report;
1148 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
1150 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1152 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1155 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1157 smp_mb__before_atomic();
1158 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1159 clear_bit(XPRT_CLOSING, &xprt->state);
1160 xs_sock_reset_state_flags(xprt);
1161 smp_mb__after_atomic();
1165 * xs_error_report - callback to handle TCP socket state errors
1168 * Note: we don't call sock_error() since there may be a rpc_task
1169 * using the socket, and so we don't want to clear sk->sk_err.
1171 static void xs_error_report(struct sock *sk)
1173 struct rpc_xprt *xprt;
1176 read_lock_bh(&sk->sk_callback_lock);
1177 if (!(xprt = xprt_from_sock(sk)))
1183 dprintk("RPC: xs_error_report client %p, error=%d...\n",
1185 trace_rpc_socket_error(xprt, sk->sk_socket, err);
1186 xprt_wake_pending_tasks(xprt, err);
1188 read_unlock_bh(&sk->sk_callback_lock);
1191 static void xs_reset_transport(struct sock_xprt *transport)
1193 struct socket *sock = transport->sock;
1194 struct sock *sk = transport->inet;
1195 struct rpc_xprt *xprt = &transport->xprt;
1200 if (atomic_read(&transport->xprt.swapper))
1201 sk_clear_memalloc(sk);
1203 kernel_sock_shutdown(sock, SHUT_RDWR);
1205 mutex_lock(&transport->recv_mutex);
1206 write_lock_bh(&sk->sk_callback_lock);
1207 transport->inet = NULL;
1208 transport->sock = NULL;
1210 sk->sk_user_data = NULL;
1212 xs_restore_old_callbacks(transport, sk);
1213 xprt_clear_connected(xprt);
1214 write_unlock_bh(&sk->sk_callback_lock);
1215 xs_sock_reset_connection_flags(xprt);
1216 mutex_unlock(&transport->recv_mutex);
1218 trace_rpc_socket_close(xprt, sock);
1221 xprt_disconnect_done(xprt);
1225 * xs_close - close a socket
1228 * This is used when all requests are complete; ie, no DRC state remains
1229 * on the server we want to save.
1231 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1232 * xs_reset_transport() zeroing the socket from underneath a writer.
1234 static void xs_close(struct rpc_xprt *xprt)
1236 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1238 dprintk("RPC: xs_close xprt %p\n", xprt);
1240 xs_reset_transport(transport);
1241 xprt->reestablish_timeout = 0;
1244 static void xs_inject_disconnect(struct rpc_xprt *xprt)
1246 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
1248 xprt_disconnect_done(xprt);
1251 static void xs_xprt_free(struct rpc_xprt *xprt)
1253 xs_free_peer_addresses(xprt);
1258 * xs_destroy - prepare to shutdown a transport
1259 * @xprt: doomed transport
1262 static void xs_destroy(struct rpc_xprt *xprt)
1264 struct sock_xprt *transport = container_of(xprt,
1265 struct sock_xprt, xprt);
1266 dprintk("RPC: xs_destroy xprt %p\n", xprt);
1268 cancel_delayed_work_sync(&transport->connect_worker);
1270 cancel_work_sync(&transport->recv_worker);
1272 module_put(THIS_MODULE);
1276 * xs_udp_data_read_skb - receive callback for UDP sockets
1282 static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1284 struct sk_buff *skb)
1286 struct rpc_task *task;
1287 struct rpc_rqst *rovr;
1288 int repsize, copied;
1294 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
1298 /* Copy the XID from the skb... */
1299 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1303 /* Look up and lock the request corresponding to the given XID */
1304 spin_lock(&xprt->queue_lock);
1305 rovr = xprt_lookup_rqst(xprt, *xp);
1308 xprt_pin_rqst(rovr);
1309 xprt_update_rtt(rovr->rq_task);
1310 spin_unlock(&xprt->queue_lock);
1311 task = rovr->rq_task;
1313 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1316 /* Suck it into the iovec, verify checksum if not done by hw. */
1317 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1318 spin_lock(&xprt->queue_lock);
1319 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1324 spin_lock_bh(&xprt->transport_lock);
1325 xprt_adjust_cwnd(xprt, task, copied);
1326 spin_unlock_bh(&xprt->transport_lock);
1327 spin_lock(&xprt->queue_lock);
1328 xprt_complete_rqst(task, copied);
1329 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1331 xprt_unpin_rqst(rovr);
1333 spin_unlock(&xprt->queue_lock);
1336 static void xs_udp_data_receive(struct sock_xprt *transport)
1338 struct sk_buff *skb;
1342 mutex_lock(&transport->recv_mutex);
1343 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1344 sk = transport->inet;
1348 skb = skb_recv_udp(sk, 0, 1, &err);
1351 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1356 mutex_unlock(&transport->recv_mutex);
1359 static void xs_udp_data_receive_workfn(struct work_struct *work)
1361 struct sock_xprt *transport =
1362 container_of(work, struct sock_xprt, recv_worker);
1363 xs_udp_data_receive(transport);
1367 * xs_data_ready - "data ready" callback for UDP sockets
1368 * @sk: socket with data to read
1371 static void xs_data_ready(struct sock *sk)
1373 struct rpc_xprt *xprt;
1375 read_lock_bh(&sk->sk_callback_lock);
1376 dprintk("RPC: xs_data_ready...\n");
1377 xprt = xprt_from_sock(sk);
1379 struct sock_xprt *transport = container_of(xprt,
1380 struct sock_xprt, xprt);
1381 transport->old_data_ready(sk);
1382 /* Any data means we had a useful conversation, so
1383 * then we don't need to delay the next reconnect
1385 if (xprt->reestablish_timeout)
1386 xprt->reestablish_timeout = 0;
1387 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1388 queue_work(xprtiod_workqueue, &transport->recv_worker);
1390 read_unlock_bh(&sk->sk_callback_lock);
1394 * Helper function to force a TCP close if the server is sending
1395 * junk and/or it has put us in CLOSE_WAIT
1397 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1399 xprt_force_disconnect(xprt);
1402 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1403 static int xs_tcp_bc_up(struct svc_serv *serv, struct net *net)
1407 ret = svc_create_xprt(serv, "tcp-bc", net, PF_INET, 0,
1408 SVC_SOCK_ANONYMOUS);
1414 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1418 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1421 * xs_tcp_state_change - callback to handle TCP socket state changes
1422 * @sk: socket whose state has changed
1425 static void xs_tcp_state_change(struct sock *sk)
1427 struct rpc_xprt *xprt;
1428 struct sock_xprt *transport;
1430 read_lock_bh(&sk->sk_callback_lock);
1431 if (!(xprt = xprt_from_sock(sk)))
1433 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1434 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1435 sk->sk_state, xprt_connected(xprt),
1436 sock_flag(sk, SOCK_DEAD),
1437 sock_flag(sk, SOCK_ZAPPED),
1440 transport = container_of(xprt, struct sock_xprt, xprt);
1441 trace_rpc_socket_state_change(xprt, sk->sk_socket);
1442 switch (sk->sk_state) {
1443 case TCP_ESTABLISHED:
1444 spin_lock(&xprt->transport_lock);
1445 if (!xprt_test_and_set_connected(xprt)) {
1446 xprt->connect_cookie++;
1447 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1448 xprt_clear_connecting(xprt);
1450 xprt->stat.connect_count++;
1451 xprt->stat.connect_time += (long)jiffies -
1452 xprt->stat.connect_start;
1453 xprt_wake_pending_tasks(xprt, -EAGAIN);
1455 spin_unlock(&xprt->transport_lock);
1458 /* The client initiated a shutdown of the socket */
1459 xprt->connect_cookie++;
1460 xprt->reestablish_timeout = 0;
1461 set_bit(XPRT_CLOSING, &xprt->state);
1462 smp_mb__before_atomic();
1463 clear_bit(XPRT_CONNECTED, &xprt->state);
1464 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1465 smp_mb__after_atomic();
1467 case TCP_CLOSE_WAIT:
1468 /* The server initiated a shutdown of the socket */
1469 xprt->connect_cookie++;
1470 clear_bit(XPRT_CONNECTED, &xprt->state);
1471 xs_tcp_force_close(xprt);
1475 * If the server closed down the connection, make sure that
1476 * we back off before reconnecting
1478 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1479 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1482 set_bit(XPRT_CLOSING, &xprt->state);
1483 smp_mb__before_atomic();
1484 clear_bit(XPRT_CONNECTED, &xprt->state);
1485 smp_mb__after_atomic();
1488 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1489 &transport->sock_state))
1490 xprt_clear_connecting(xprt);
1491 clear_bit(XPRT_CLOSING, &xprt->state);
1492 /* Trigger the socket release */
1493 xs_tcp_force_close(xprt);
1496 read_unlock_bh(&sk->sk_callback_lock);
1499 static void xs_write_space(struct sock *sk)
1501 struct socket_wq *wq;
1502 struct rpc_xprt *xprt;
1506 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1508 if (unlikely(!(xprt = xprt_from_sock(sk))))
1511 wq = rcu_dereference(sk->sk_wq);
1512 if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1515 if (xprt_write_space(xprt))
1516 sk->sk_write_pending--;
1522 * xs_udp_write_space - callback invoked when socket buffer space
1524 * @sk: socket whose state has changed
1526 * Called when more output buffer space is available for this socket.
1527 * We try not to wake our writers until they can make "significant"
1528 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1529 * with a bunch of small requests.
1531 static void xs_udp_write_space(struct sock *sk)
1533 read_lock_bh(&sk->sk_callback_lock);
1535 /* from net/core/sock.c:sock_def_write_space */
1536 if (sock_writeable(sk))
1539 read_unlock_bh(&sk->sk_callback_lock);
1543 * xs_tcp_write_space - callback invoked when socket buffer space
1545 * @sk: socket whose state has changed
1547 * Called when more output buffer space is available for this socket.
1548 * We try not to wake our writers until they can make "significant"
1549 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1550 * with a bunch of small requests.
1552 static void xs_tcp_write_space(struct sock *sk)
1554 read_lock_bh(&sk->sk_callback_lock);
1556 /* from net/core/stream.c:sk_stream_write_space */
1557 if (sk_stream_is_writeable(sk))
1560 read_unlock_bh(&sk->sk_callback_lock);
1563 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1565 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1566 struct sock *sk = transport->inet;
1568 if (transport->rcvsize) {
1569 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1570 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1572 if (transport->sndsize) {
1573 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1574 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1575 sk->sk_write_space(sk);
1580 * xs_udp_set_buffer_size - set send and receive limits
1581 * @xprt: generic transport
1582 * @sndsize: requested size of send buffer, in bytes
1583 * @rcvsize: requested size of receive buffer, in bytes
1585 * Set socket send and receive buffer size limits.
1587 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1589 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1591 transport->sndsize = 0;
1593 transport->sndsize = sndsize + 1024;
1594 transport->rcvsize = 0;
1596 transport->rcvsize = rcvsize + 1024;
1598 xs_udp_do_set_buffer_size(xprt);
1602 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1603 * @task: task that timed out
1605 * Adjust the congestion window after a retransmit timeout has occurred.
1607 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1609 spin_lock_bh(&xprt->transport_lock);
1610 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1611 spin_unlock_bh(&xprt->transport_lock);
1614 static int xs_get_random_port(void)
1616 unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1617 unsigned short range;
1618 unsigned short rand;
1622 range = max - min + 1;
1623 rand = (unsigned short) prandom_u32() % range;
1628 * xs_set_reuseaddr_port - set the socket's port and address reuse options
1631 * Note that this function has to be called on all sockets that share the
1632 * same port, and it must be called before binding.
1634 static void xs_sock_set_reuseport(struct socket *sock)
1638 kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1639 (char *)&opt, sizeof(opt));
1642 static unsigned short xs_sock_getport(struct socket *sock)
1644 struct sockaddr_storage buf;
1645 unsigned short port = 0;
1647 if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1649 switch (buf.ss_family) {
1651 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1654 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1661 * xs_set_port - reset the port number in the remote endpoint address
1662 * @xprt: generic transport
1663 * @port: new port number
1666 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1668 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1670 rpc_set_port(xs_addr(xprt), port);
1671 xs_update_peer_port(xprt);
1674 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1676 if (transport->srcport == 0)
1677 transport->srcport = xs_sock_getport(sock);
1680 static int xs_get_srcport(struct sock_xprt *transport)
1682 int port = transport->srcport;
1684 if (port == 0 && transport->xprt.resvport)
1685 port = xs_get_random_port();
1689 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1691 if (transport->srcport != 0)
1692 transport->srcport = 0;
1693 if (!transport->xprt.resvport)
1695 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1696 return xprt_max_resvport;
1699 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1701 struct sockaddr_storage myaddr;
1703 int port = xs_get_srcport(transport);
1704 unsigned short last;
1707 * If we are asking for any ephemeral port (i.e. port == 0 &&
1708 * transport->xprt.resvport == 0), don't bind. Let the local
1709 * port selection happen implicitly when the socket is used
1710 * (for example at connect time).
1712 * This ensures that we can continue to establish TCP
1713 * connections even when all local ephemeral ports are already
1714 * a part of some TCP connection. This makes no difference
1715 * for UDP sockets, but also doens't harm them.
1717 * If we're asking for any reserved port (i.e. port == 0 &&
1718 * transport->xprt.resvport == 1) xs_get_srcport above will
1719 * ensure that port is non-zero and we will bind as needed.
1724 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1726 rpc_set_port((struct sockaddr *)&myaddr, port);
1727 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1728 transport->xprt.addrlen);
1730 transport->srcport = port;
1734 port = xs_next_srcport(transport, port);
1737 } while (err == -EADDRINUSE && nloop != 2);
1739 if (myaddr.ss_family == AF_INET)
1740 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1741 &((struct sockaddr_in *)&myaddr)->sin_addr,
1742 port, err ? "failed" : "ok", err);
1744 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1745 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1746 port, err ? "failed" : "ok", err);
1751 * We don't support autobind on AF_LOCAL sockets
1753 static void xs_local_rpcbind(struct rpc_task *task)
1755 xprt_set_bound(task->tk_xprt);
1758 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1762 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1763 static struct lock_class_key xs_key[2];
1764 static struct lock_class_key xs_slock_key[2];
1766 static inline void xs_reclassify_socketu(struct socket *sock)
1768 struct sock *sk = sock->sk;
1770 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1771 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1774 static inline void xs_reclassify_socket4(struct socket *sock)
1776 struct sock *sk = sock->sk;
1778 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1779 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1782 static inline void xs_reclassify_socket6(struct socket *sock)
1784 struct sock *sk = sock->sk;
1786 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1787 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1790 static inline void xs_reclassify_socket(int family, struct socket *sock)
1792 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1797 xs_reclassify_socketu(sock);
1800 xs_reclassify_socket4(sock);
1803 xs_reclassify_socket6(sock);
1808 static inline void xs_reclassify_socket(int family, struct socket *sock)
1813 static void xs_dummy_setup_socket(struct work_struct *work)
1817 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1818 struct sock_xprt *transport, int family, int type,
1819 int protocol, bool reuseport)
1821 struct socket *sock;
1824 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1826 dprintk("RPC: can't create %d transport socket (%d).\n",
1830 xs_reclassify_socket(family, sock);
1833 xs_sock_set_reuseport(sock);
1835 err = xs_bind(transport, sock);
1843 return ERR_PTR(err);
1846 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1847 struct socket *sock)
1849 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1852 if (!transport->inet) {
1853 struct sock *sk = sock->sk;
1855 write_lock_bh(&sk->sk_callback_lock);
1857 xs_save_old_callbacks(transport, sk);
1859 sk->sk_user_data = xprt;
1860 sk->sk_data_ready = xs_data_ready;
1861 sk->sk_write_space = xs_udp_write_space;
1862 sock_set_flag(sk, SOCK_FASYNC);
1863 sk->sk_error_report = xs_error_report;
1864 sk->sk_allocation = GFP_NOIO;
1866 xprt_clear_connected(xprt);
1868 /* Reset to new socket */
1869 transport->sock = sock;
1870 transport->inet = sk;
1872 write_unlock_bh(&sk->sk_callback_lock);
1875 xs_stream_reset_connect(transport);
1877 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1881 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1882 * @transport: socket transport to connect
1884 static int xs_local_setup_socket(struct sock_xprt *transport)
1886 struct rpc_xprt *xprt = &transport->xprt;
1887 struct socket *sock;
1890 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1891 SOCK_STREAM, 0, &sock, 1);
1893 dprintk("RPC: can't create AF_LOCAL "
1894 "transport socket (%d).\n", -status);
1897 xs_reclassify_socket(AF_LOCAL, sock);
1899 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1900 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1902 status = xs_local_finish_connecting(xprt, sock);
1903 trace_rpc_socket_connect(xprt, sock, status);
1906 dprintk("RPC: xprt %p connected to %s\n",
1907 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1908 xprt->stat.connect_count++;
1909 xprt->stat.connect_time += (long)jiffies -
1910 xprt->stat.connect_start;
1911 xprt_set_connected(xprt);
1915 dprintk("RPC: xprt %p: socket %s does not exist\n",
1916 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1919 dprintk("RPC: xprt %p: connection refused for %s\n",
1920 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1923 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1925 xprt->address_strings[RPC_DISPLAY_ADDR]);
1929 xprt_clear_connecting(xprt);
1930 xprt_wake_pending_tasks(xprt, status);
1934 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1936 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1939 if (RPC_IS_ASYNC(task)) {
1941 * We want the AF_LOCAL connect to be resolved in the
1942 * filesystem namespace of the process making the rpc
1943 * call. Thus we connect synchronously.
1945 * If we want to support asynchronous AF_LOCAL calls,
1946 * we'll need to figure out how to pass a namespace to
1949 rpc_exit(task, -ENOTCONN);
1952 ret = xs_local_setup_socket(transport);
1953 if (ret && !RPC_IS_SOFTCONN(task))
1954 msleep_interruptible(15000);
1957 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
1959 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
1960 * know that we have exclusive access to the socket), to guard against
1961 * races with xs_reset_transport.
1963 static void xs_set_memalloc(struct rpc_xprt *xprt)
1965 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1969 * If there's no sock, then we have nothing to set. The
1970 * reconnecting process will get it for us.
1972 if (!transport->inet)
1974 if (atomic_read(&xprt->swapper))
1975 sk_set_memalloc(transport->inet);
1979 * xs_enable_swap - Tag this transport as being used for swap.
1980 * @xprt: transport to tag
1982 * Take a reference to this transport on behalf of the rpc_clnt, and
1983 * optionally mark it for swapping if it wasn't already.
1986 xs_enable_swap(struct rpc_xprt *xprt)
1988 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
1990 if (atomic_inc_return(&xprt->swapper) != 1)
1992 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
1993 return -ERESTARTSYS;
1995 sk_set_memalloc(xs->inet);
1996 xprt_release_xprt(xprt, NULL);
2001 * xs_disable_swap - Untag this transport as being used for swap.
2002 * @xprt: transport to tag
2004 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2005 * swapper refcount goes to 0, untag the socket as a memalloc socket.
2008 xs_disable_swap(struct rpc_xprt *xprt)
2010 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2012 if (!atomic_dec_and_test(&xprt->swapper))
2014 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2017 sk_clear_memalloc(xs->inet);
2018 xprt_release_xprt(xprt, NULL);
2021 static void xs_set_memalloc(struct rpc_xprt *xprt)
2026 xs_enable_swap(struct rpc_xprt *xprt)
2032 xs_disable_swap(struct rpc_xprt *xprt)
2037 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2039 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2041 if (!transport->inet) {
2042 struct sock *sk = sock->sk;
2044 write_lock_bh(&sk->sk_callback_lock);
2046 xs_save_old_callbacks(transport, sk);
2048 sk->sk_user_data = xprt;
2049 sk->sk_data_ready = xs_data_ready;
2050 sk->sk_write_space = xs_udp_write_space;
2051 sock_set_flag(sk, SOCK_FASYNC);
2052 sk->sk_allocation = GFP_NOIO;
2054 xprt_set_connected(xprt);
2056 /* Reset to new socket */
2057 transport->sock = sock;
2058 transport->inet = sk;
2060 xs_set_memalloc(xprt);
2062 write_unlock_bh(&sk->sk_callback_lock);
2064 xs_udp_do_set_buffer_size(xprt);
2066 xprt->stat.connect_start = jiffies;
2069 static void xs_udp_setup_socket(struct work_struct *work)
2071 struct sock_xprt *transport =
2072 container_of(work, struct sock_xprt, connect_worker.work);
2073 struct rpc_xprt *xprt = &transport->xprt;
2074 struct socket *sock;
2077 sock = xs_create_sock(xprt, transport,
2078 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2079 IPPROTO_UDP, false);
2083 dprintk("RPC: worker connecting xprt %p via %s to "
2084 "%s (port %s)\n", xprt,
2085 xprt->address_strings[RPC_DISPLAY_PROTO],
2086 xprt->address_strings[RPC_DISPLAY_ADDR],
2087 xprt->address_strings[RPC_DISPLAY_PORT]);
2089 xs_udp_finish_connecting(xprt, sock);
2090 trace_rpc_socket_connect(xprt, sock, 0);
2093 xprt_clear_connecting(xprt);
2094 xprt_unlock_connect(xprt, transport);
2095 xprt_wake_pending_tasks(xprt, status);
2099 * xs_tcp_shutdown - gracefully shut down a TCP socket
2102 * Initiates a graceful shutdown of the TCP socket by calling the
2103 * equivalent of shutdown(SHUT_RDWR);
2105 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2107 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2108 struct socket *sock = transport->sock;
2109 int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2115 kernel_sock_shutdown(sock, SHUT_RDWR);
2116 trace_rpc_socket_shutdown(xprt, sock);
2120 xs_reset_transport(transport);
2124 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2125 struct socket *sock)
2127 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2128 unsigned int keepidle;
2129 unsigned int keepcnt;
2130 unsigned int opt_on = 1;
2133 spin_lock_bh(&xprt->transport_lock);
2134 keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2135 keepcnt = xprt->timeout->to_retries + 1;
2136 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2137 (xprt->timeout->to_retries + 1);
2138 clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2139 spin_unlock_bh(&xprt->transport_lock);
2141 /* TCP Keepalive options */
2142 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2143 (char *)&opt_on, sizeof(opt_on));
2144 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2145 (char *)&keepidle, sizeof(keepidle));
2146 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2147 (char *)&keepidle, sizeof(keepidle));
2148 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2149 (char *)&keepcnt, sizeof(keepcnt));
2151 /* TCP user timeout (see RFC5482) */
2152 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2153 (char *)&timeo, sizeof(timeo));
2156 static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2157 unsigned long connect_timeout,
2158 unsigned long reconnect_timeout)
2160 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2161 struct rpc_timeout to;
2162 unsigned long initval;
2164 spin_lock_bh(&xprt->transport_lock);
2165 if (reconnect_timeout < xprt->max_reconnect_timeout)
2166 xprt->max_reconnect_timeout = reconnect_timeout;
2167 if (connect_timeout < xprt->connect_timeout) {
2168 memcpy(&to, xprt->timeout, sizeof(to));
2169 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2170 /* Arbitrary lower limit */
2171 if (initval < XS_TCP_INIT_REEST_TO << 1)
2172 initval = XS_TCP_INIT_REEST_TO << 1;
2173 to.to_initval = initval;
2174 to.to_maxval = initval;
2175 memcpy(&transport->tcp_timeout, &to,
2176 sizeof(transport->tcp_timeout));
2177 xprt->timeout = &transport->tcp_timeout;
2178 xprt->connect_timeout = connect_timeout;
2180 set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2181 spin_unlock_bh(&xprt->transport_lock);
2184 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2186 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2187 int ret = -ENOTCONN;
2189 if (!transport->inet) {
2190 struct sock *sk = sock->sk;
2191 unsigned int addr_pref = IPV6_PREFER_SRC_PUBLIC;
2193 /* Avoid temporary address, they are bad for long-lived
2194 * connections such as NFS mounts.
2195 * RFC4941, section 3.6 suggests that:
2196 * Individual applications, which have specific
2197 * knowledge about the normal duration of connections,
2198 * MAY override this as appropriate.
2200 kernel_setsockopt(sock, SOL_IPV6, IPV6_ADDR_PREFERENCES,
2201 (char *)&addr_pref, sizeof(addr_pref));
2203 xs_tcp_set_socket_timeouts(xprt, sock);
2205 write_lock_bh(&sk->sk_callback_lock);
2207 xs_save_old_callbacks(transport, sk);
2209 sk->sk_user_data = xprt;
2210 sk->sk_data_ready = xs_data_ready;
2211 sk->sk_state_change = xs_tcp_state_change;
2212 sk->sk_write_space = xs_tcp_write_space;
2213 sock_set_flag(sk, SOCK_FASYNC);
2214 sk->sk_error_report = xs_error_report;
2215 sk->sk_allocation = GFP_NOIO;
2217 /* socket options */
2218 sock_reset_flag(sk, SOCK_LINGER);
2219 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2221 xprt_clear_connected(xprt);
2223 /* Reset to new socket */
2224 transport->sock = sock;
2225 transport->inet = sk;
2227 write_unlock_bh(&sk->sk_callback_lock);
2230 if (!xprt_bound(xprt))
2233 xs_set_memalloc(xprt);
2235 /* Reset TCP record info */
2236 xs_stream_reset_connect(transport);
2238 /* Tell the socket layer to start connecting... */
2239 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2240 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2243 xs_set_srcport(transport, sock);
2247 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2248 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2250 case -EADDRNOTAVAIL:
2251 /* Source port number is unavailable. Try a new one! */
2252 transport->srcport = 0;
2259 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2261 * Invoked by a work queue tasklet.
2263 static void xs_tcp_setup_socket(struct work_struct *work)
2265 struct sock_xprt *transport =
2266 container_of(work, struct sock_xprt, connect_worker.work);
2267 struct socket *sock = transport->sock;
2268 struct rpc_xprt *xprt = &transport->xprt;
2272 sock = xs_create_sock(xprt, transport,
2273 xs_addr(xprt)->sa_family, SOCK_STREAM,
2276 status = PTR_ERR(sock);
2281 dprintk("RPC: worker connecting xprt %p via %s to "
2282 "%s (port %s)\n", xprt,
2283 xprt->address_strings[RPC_DISPLAY_PROTO],
2284 xprt->address_strings[RPC_DISPLAY_ADDR],
2285 xprt->address_strings[RPC_DISPLAY_PORT]);
2287 status = xs_tcp_finish_connecting(xprt, sock);
2288 trace_rpc_socket_connect(xprt, sock, status);
2289 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2290 xprt, -status, xprt_connected(xprt),
2291 sock->sk->sk_state);
2294 printk("%s: connect returned unhandled error %d\n",
2297 case -EADDRNOTAVAIL:
2298 /* We're probably in TIME_WAIT. Get rid of existing socket,
2301 xs_tcp_force_close(xprt);
2306 xprt_unlock_connect(xprt, transport);
2309 /* Happens, for instance, if the user specified a link
2310 * local IPv6 address without a scope-id.
2320 * xs_tcp_force_close() wakes tasks with -EIO.
2321 * We need to wake them first to ensure the
2322 * correct error code.
2324 xprt_wake_pending_tasks(xprt, status);
2325 xs_tcp_force_close(xprt);
2330 xprt_clear_connecting(xprt);
2331 xprt_unlock_connect(xprt, transport);
2332 xprt_wake_pending_tasks(xprt, status);
2335 static unsigned long xs_reconnect_delay(const struct rpc_xprt *xprt)
2337 unsigned long start, now = jiffies;
2339 start = xprt->stat.connect_start + xprt->reestablish_timeout;
2340 if (time_after(start, now))
2345 static void xs_reconnect_backoff(struct rpc_xprt *xprt)
2347 xprt->reestablish_timeout <<= 1;
2348 if (xprt->reestablish_timeout > xprt->max_reconnect_timeout)
2349 xprt->reestablish_timeout = xprt->max_reconnect_timeout;
2350 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2351 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2355 * xs_connect - connect a socket to a remote endpoint
2356 * @xprt: pointer to transport structure
2357 * @task: address of RPC task that manages state of connect request
2359 * TCP: If the remote end dropped the connection, delay reconnecting.
2361 * UDP socket connects are synchronous, but we use a work queue anyway
2362 * to guarantee that even unprivileged user processes can set up a
2363 * socket on a privileged port.
2365 * If a UDP socket connect fails, the delay behavior here prevents
2366 * retry floods (hard mounts).
2368 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2370 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2371 unsigned long delay = 0;
2373 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2375 if (transport->sock != NULL) {
2376 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2378 xprt, xprt->reestablish_timeout / HZ);
2380 /* Start by resetting any existing state */
2381 xs_reset_transport(transport);
2383 delay = xs_reconnect_delay(xprt);
2384 xs_reconnect_backoff(xprt);
2387 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2389 queue_delayed_work(xprtiod_workqueue,
2390 &transport->connect_worker,
2395 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2396 * @xprt: rpc_xprt struct containing statistics
2400 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2404 if (xprt_connected(xprt))
2405 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2407 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2408 "%llu %llu %lu %llu %llu\n",
2409 xprt->stat.bind_count,
2410 xprt->stat.connect_count,
2411 xprt->stat.connect_time / HZ,
2415 xprt->stat.bad_xids,
2418 xprt->stat.max_slots,
2419 xprt->stat.sending_u,
2420 xprt->stat.pending_u);
2424 * xs_udp_print_stats - display UDP socket-specifc stats
2425 * @xprt: rpc_xprt struct containing statistics
2429 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2431 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2433 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2436 xprt->stat.bind_count,
2439 xprt->stat.bad_xids,
2442 xprt->stat.max_slots,
2443 xprt->stat.sending_u,
2444 xprt->stat.pending_u);
2448 * xs_tcp_print_stats - display TCP socket-specifc stats
2449 * @xprt: rpc_xprt struct containing statistics
2453 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2455 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2458 if (xprt_connected(xprt))
2459 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2461 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2462 "%llu %llu %lu %llu %llu\n",
2464 xprt->stat.bind_count,
2465 xprt->stat.connect_count,
2466 xprt->stat.connect_time / HZ,
2470 xprt->stat.bad_xids,
2473 xprt->stat.max_slots,
2474 xprt->stat.sending_u,
2475 xprt->stat.pending_u);
2479 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2480 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2481 * to use the server side send routines.
2483 static int bc_malloc(struct rpc_task *task)
2485 struct rpc_rqst *rqst = task->tk_rqstp;
2486 size_t size = rqst->rq_callsize;
2488 struct rpc_buffer *buf;
2490 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2491 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2496 page = alloc_page(GFP_KERNEL);
2500 buf = page_address(page);
2501 buf->len = PAGE_SIZE;
2503 rqst->rq_buffer = buf->data;
2504 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2509 * Free the space allocated in the bc_alloc routine
2511 static void bc_free(struct rpc_task *task)
2513 void *buffer = task->tk_rqstp->rq_buffer;
2514 struct rpc_buffer *buf;
2516 buf = container_of(buffer, struct rpc_buffer, data);
2517 free_page((unsigned long)buf);
2521 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2522 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2524 static int bc_sendto(struct rpc_rqst *req)
2527 struct xdr_buf *xbufp = &req->rq_snd_buf;
2528 struct rpc_xprt *xprt = req->rq_xprt;
2529 struct sock_xprt *transport =
2530 container_of(xprt, struct sock_xprt, xprt);
2531 struct socket *sock = transport->sock;
2532 unsigned long headoff;
2533 unsigned long tailoff;
2535 xs_encode_stream_record_marker(xbufp);
2537 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2538 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2539 len = svc_send_common(sock, xbufp,
2540 virt_to_page(xbufp->head[0].iov_base), headoff,
2541 xbufp->tail[0].iov_base, tailoff);
2543 if (len != xbufp->len) {
2544 printk(KERN_NOTICE "Error sending entire callback!\n");
2552 * The send routine. Borrows from svc_send
2554 static int bc_send_request(struct rpc_rqst *req)
2556 struct svc_xprt *xprt;
2559 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2561 * Get the server socket associated with this callback xprt
2563 xprt = req->rq_xprt->bc_xprt;
2566 * Grab the mutex to serialize data as the connection is shared
2567 * with the fore channel
2569 mutex_lock(&xprt->xpt_mutex);
2570 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2573 len = bc_sendto(req);
2574 mutex_unlock(&xprt->xpt_mutex);
2583 * The close routine. Since this is client initiated, we do nothing
2586 static void bc_close(struct rpc_xprt *xprt)
2591 * The xprt destroy routine. Again, because this connection is client
2592 * initiated, we do nothing
2595 static void bc_destroy(struct rpc_xprt *xprt)
2597 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2600 module_put(THIS_MODULE);
2603 static const struct rpc_xprt_ops xs_local_ops = {
2604 .reserve_xprt = xprt_reserve_xprt,
2605 .release_xprt = xprt_release_xprt,
2606 .alloc_slot = xprt_alloc_slot,
2607 .free_slot = xprt_free_slot,
2608 .rpcbind = xs_local_rpcbind,
2609 .set_port = xs_local_set_port,
2610 .connect = xs_local_connect,
2611 .buf_alloc = rpc_malloc,
2612 .buf_free = rpc_free,
2613 .prepare_request = xs_stream_prepare_request,
2614 .send_request = xs_local_send_request,
2615 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2617 .destroy = xs_destroy,
2618 .print_stats = xs_local_print_stats,
2619 .enable_swap = xs_enable_swap,
2620 .disable_swap = xs_disable_swap,
2623 static const struct rpc_xprt_ops xs_udp_ops = {
2624 .set_buffer_size = xs_udp_set_buffer_size,
2625 .reserve_xprt = xprt_reserve_xprt_cong,
2626 .release_xprt = xprt_release_xprt_cong,
2627 .alloc_slot = xprt_alloc_slot,
2628 .free_slot = xprt_free_slot,
2629 .rpcbind = rpcb_getport_async,
2630 .set_port = xs_set_port,
2631 .connect = xs_connect,
2632 .buf_alloc = rpc_malloc,
2633 .buf_free = rpc_free,
2634 .send_request = xs_udp_send_request,
2635 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
2636 .timer = xs_udp_timer,
2637 .release_request = xprt_release_rqst_cong,
2639 .destroy = xs_destroy,
2640 .print_stats = xs_udp_print_stats,
2641 .enable_swap = xs_enable_swap,
2642 .disable_swap = xs_disable_swap,
2643 .inject_disconnect = xs_inject_disconnect,
2646 static const struct rpc_xprt_ops xs_tcp_ops = {
2647 .reserve_xprt = xprt_reserve_xprt,
2648 .release_xprt = xprt_release_xprt,
2649 .alloc_slot = xprt_alloc_slot,
2650 .free_slot = xprt_free_slot,
2651 .rpcbind = rpcb_getport_async,
2652 .set_port = xs_set_port,
2653 .connect = xs_connect,
2654 .buf_alloc = rpc_malloc,
2655 .buf_free = rpc_free,
2656 .prepare_request = xs_stream_prepare_request,
2657 .send_request = xs_tcp_send_request,
2658 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2659 .close = xs_tcp_shutdown,
2660 .destroy = xs_destroy,
2661 .set_connect_timeout = xs_tcp_set_connect_timeout,
2662 .print_stats = xs_tcp_print_stats,
2663 .enable_swap = xs_enable_swap,
2664 .disable_swap = xs_disable_swap,
2665 .inject_disconnect = xs_inject_disconnect,
2666 #ifdef CONFIG_SUNRPC_BACKCHANNEL
2667 .bc_setup = xprt_setup_bc,
2668 .bc_up = xs_tcp_bc_up,
2669 .bc_maxpayload = xs_tcp_bc_maxpayload,
2670 .bc_free_rqst = xprt_free_bc_rqst,
2671 .bc_destroy = xprt_destroy_bc,
2676 * The rpc_xprt_ops for the server backchannel
2679 static const struct rpc_xprt_ops bc_tcp_ops = {
2680 .reserve_xprt = xprt_reserve_xprt,
2681 .release_xprt = xprt_release_xprt,
2682 .alloc_slot = xprt_alloc_slot,
2683 .free_slot = xprt_free_slot,
2684 .buf_alloc = bc_malloc,
2685 .buf_free = bc_free,
2686 .send_request = bc_send_request,
2687 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2689 .destroy = bc_destroy,
2690 .print_stats = xs_tcp_print_stats,
2691 .enable_swap = xs_enable_swap,
2692 .disable_swap = xs_disable_swap,
2693 .inject_disconnect = xs_inject_disconnect,
2696 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2698 static const struct sockaddr_in sin = {
2699 .sin_family = AF_INET,
2700 .sin_addr.s_addr = htonl(INADDR_ANY),
2702 static const struct sockaddr_in6 sin6 = {
2703 .sin6_family = AF_INET6,
2704 .sin6_addr = IN6ADDR_ANY_INIT,
2711 memcpy(sap, &sin, sizeof(sin));
2714 memcpy(sap, &sin6, sizeof(sin6));
2717 dprintk("RPC: %s: Bad address family\n", __func__);
2718 return -EAFNOSUPPORT;
2723 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2724 unsigned int slot_table_size,
2725 unsigned int max_slot_table_size)
2727 struct rpc_xprt *xprt;
2728 struct sock_xprt *new;
2730 if (args->addrlen > sizeof(xprt->addr)) {
2731 dprintk("RPC: xs_setup_xprt: address too large\n");
2732 return ERR_PTR(-EBADF);
2735 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2736 max_slot_table_size);
2738 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2740 return ERR_PTR(-ENOMEM);
2743 new = container_of(xprt, struct sock_xprt, xprt);
2744 mutex_init(&new->recv_mutex);
2745 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2746 xprt->addrlen = args->addrlen;
2748 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2751 err = xs_init_anyaddr(args->dstaddr->sa_family,
2752 (struct sockaddr *)&new->srcaddr);
2755 return ERR_PTR(err);
2762 static const struct rpc_timeout xs_local_default_timeout = {
2763 .to_initval = 10 * HZ,
2764 .to_maxval = 10 * HZ,
2769 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2770 * @args: rpc transport creation arguments
2772 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2774 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2776 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2777 struct sock_xprt *transport;
2778 struct rpc_xprt *xprt;
2779 struct rpc_xprt *ret;
2781 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2782 xprt_max_tcp_slot_table_entries);
2785 transport = container_of(xprt, struct sock_xprt, xprt);
2788 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2789 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2791 xprt->bind_timeout = XS_BIND_TO;
2792 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2793 xprt->idle_timeout = XS_IDLE_DISC_TO;
2795 xprt->ops = &xs_local_ops;
2796 xprt->timeout = &xs_local_default_timeout;
2798 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2799 INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket);
2801 switch (sun->sun_family) {
2803 if (sun->sun_path[0] != '/') {
2804 dprintk("RPC: bad AF_LOCAL address: %s\n",
2806 ret = ERR_PTR(-EINVAL);
2809 xprt_set_bound(xprt);
2810 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2811 ret = ERR_PTR(xs_local_setup_socket(transport));
2816 ret = ERR_PTR(-EAFNOSUPPORT);
2820 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2821 xprt->address_strings[RPC_DISPLAY_ADDR]);
2823 if (try_module_get(THIS_MODULE))
2825 ret = ERR_PTR(-EINVAL);
2831 static const struct rpc_timeout xs_udp_default_timeout = {
2832 .to_initval = 5 * HZ,
2833 .to_maxval = 30 * HZ,
2834 .to_increment = 5 * HZ,
2839 * xs_setup_udp - Set up transport to use a UDP socket
2840 * @args: rpc transport creation arguments
2843 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2845 struct sockaddr *addr = args->dstaddr;
2846 struct rpc_xprt *xprt;
2847 struct sock_xprt *transport;
2848 struct rpc_xprt *ret;
2850 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2851 xprt_udp_slot_table_entries);
2854 transport = container_of(xprt, struct sock_xprt, xprt);
2856 xprt->prot = IPPROTO_UDP;
2858 /* XXX: header size can vary due to auth type, IPv6, etc. */
2859 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2861 xprt->bind_timeout = XS_BIND_TO;
2862 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2863 xprt->idle_timeout = XS_IDLE_DISC_TO;
2865 xprt->ops = &xs_udp_ops;
2867 xprt->timeout = &xs_udp_default_timeout;
2869 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
2870 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2872 switch (addr->sa_family) {
2874 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2875 xprt_set_bound(xprt);
2877 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2880 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2881 xprt_set_bound(xprt);
2883 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2886 ret = ERR_PTR(-EAFNOSUPPORT);
2890 if (xprt_bound(xprt))
2891 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2892 xprt->address_strings[RPC_DISPLAY_ADDR],
2893 xprt->address_strings[RPC_DISPLAY_PORT],
2894 xprt->address_strings[RPC_DISPLAY_PROTO]);
2896 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2897 xprt->address_strings[RPC_DISPLAY_ADDR],
2898 xprt->address_strings[RPC_DISPLAY_PROTO]);
2900 if (try_module_get(THIS_MODULE))
2902 ret = ERR_PTR(-EINVAL);
2908 static const struct rpc_timeout xs_tcp_default_timeout = {
2909 .to_initval = 60 * HZ,
2910 .to_maxval = 60 * HZ,
2915 * xs_setup_tcp - Set up transport to use a TCP socket
2916 * @args: rpc transport creation arguments
2919 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2921 struct sockaddr *addr = args->dstaddr;
2922 struct rpc_xprt *xprt;
2923 struct sock_xprt *transport;
2924 struct rpc_xprt *ret;
2925 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2927 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2928 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2930 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2931 max_slot_table_size);
2934 transport = container_of(xprt, struct sock_xprt, xprt);
2936 xprt->prot = IPPROTO_TCP;
2937 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2938 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2940 xprt->bind_timeout = XS_BIND_TO;
2941 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2942 xprt->idle_timeout = XS_IDLE_DISC_TO;
2944 xprt->ops = &xs_tcp_ops;
2945 xprt->timeout = &xs_tcp_default_timeout;
2947 xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
2948 xprt->connect_timeout = xprt->timeout->to_initval *
2949 (xprt->timeout->to_retries + 1);
2951 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2952 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
2954 switch (addr->sa_family) {
2956 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2957 xprt_set_bound(xprt);
2959 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2962 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2963 xprt_set_bound(xprt);
2965 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2968 ret = ERR_PTR(-EAFNOSUPPORT);
2972 if (xprt_bound(xprt))
2973 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2974 xprt->address_strings[RPC_DISPLAY_ADDR],
2975 xprt->address_strings[RPC_DISPLAY_PORT],
2976 xprt->address_strings[RPC_DISPLAY_PROTO]);
2978 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2979 xprt->address_strings[RPC_DISPLAY_ADDR],
2980 xprt->address_strings[RPC_DISPLAY_PROTO]);
2982 if (try_module_get(THIS_MODULE))
2984 ret = ERR_PTR(-EINVAL);
2991 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2992 * @args: rpc transport creation arguments
2995 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2997 struct sockaddr *addr = args->dstaddr;
2998 struct rpc_xprt *xprt;
2999 struct sock_xprt *transport;
3000 struct svc_sock *bc_sock;
3001 struct rpc_xprt *ret;
3003 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3004 xprt_tcp_slot_table_entries);
3007 transport = container_of(xprt, struct sock_xprt, xprt);
3009 xprt->prot = IPPROTO_TCP;
3010 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3011 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3012 xprt->timeout = &xs_tcp_default_timeout;
3015 xprt_set_bound(xprt);
3016 xprt->bind_timeout = 0;
3017 xprt->reestablish_timeout = 0;
3018 xprt->idle_timeout = 0;
3020 xprt->ops = &bc_tcp_ops;
3022 switch (addr->sa_family) {
3024 xs_format_peer_addresses(xprt, "tcp",
3028 xs_format_peer_addresses(xprt, "tcp",
3029 RPCBIND_NETID_TCP6);
3032 ret = ERR_PTR(-EAFNOSUPPORT);
3036 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3037 xprt->address_strings[RPC_DISPLAY_ADDR],
3038 xprt->address_strings[RPC_DISPLAY_PORT],
3039 xprt->address_strings[RPC_DISPLAY_PROTO]);
3042 * Once we've associated a backchannel xprt with a connection,
3043 * we want to keep it around as long as the connection lasts,
3044 * in case we need to start using it for a backchannel again;
3045 * this reference won't be dropped until bc_xprt is destroyed.
3048 args->bc_xprt->xpt_bc_xprt = xprt;
3049 xprt->bc_xprt = args->bc_xprt;
3050 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3051 transport->sock = bc_sock->sk_sock;
3052 transport->inet = bc_sock->sk_sk;
3055 * Since we don't want connections for the backchannel, we set
3056 * the xprt status to connected
3058 xprt_set_connected(xprt);
3060 if (try_module_get(THIS_MODULE))
3063 args->bc_xprt->xpt_bc_xprt = NULL;
3064 args->bc_xprt->xpt_bc_xps = NULL;
3066 ret = ERR_PTR(-EINVAL);
3072 static struct xprt_class xs_local_transport = {
3073 .list = LIST_HEAD_INIT(xs_local_transport.list),
3074 .name = "named UNIX socket",
3075 .owner = THIS_MODULE,
3076 .ident = XPRT_TRANSPORT_LOCAL,
3077 .setup = xs_setup_local,
3080 static struct xprt_class xs_udp_transport = {
3081 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3083 .owner = THIS_MODULE,
3084 .ident = XPRT_TRANSPORT_UDP,
3085 .setup = xs_setup_udp,
3088 static struct xprt_class xs_tcp_transport = {
3089 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3091 .owner = THIS_MODULE,
3092 .ident = XPRT_TRANSPORT_TCP,
3093 .setup = xs_setup_tcp,
3096 static struct xprt_class xs_bc_tcp_transport = {
3097 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3098 .name = "tcp NFSv4.1 backchannel",
3099 .owner = THIS_MODULE,
3100 .ident = XPRT_TRANSPORT_BC_TCP,
3101 .setup = xs_setup_bc_tcp,
3105 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3108 int init_socket_xprt(void)
3110 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3111 if (!sunrpc_table_header)
3112 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3115 xprt_register_transport(&xs_local_transport);
3116 xprt_register_transport(&xs_udp_transport);
3117 xprt_register_transport(&xs_tcp_transport);
3118 xprt_register_transport(&xs_bc_tcp_transport);
3124 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3127 void cleanup_socket_xprt(void)
3129 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3130 if (sunrpc_table_header) {
3131 unregister_sysctl_table(sunrpc_table_header);
3132 sunrpc_table_header = NULL;
3136 xprt_unregister_transport(&xs_local_transport);
3137 xprt_unregister_transport(&xs_udp_transport);
3138 xprt_unregister_transport(&xs_tcp_transport);
3139 xprt_unregister_transport(&xs_bc_tcp_transport);
3142 static int param_set_uint_minmax(const char *val,
3143 const struct kernel_param *kp,
3144 unsigned int min, unsigned int max)
3151 ret = kstrtouint(val, 0, &num);
3154 if (num < min || num > max)
3156 *((unsigned int *)kp->arg) = num;
3160 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3162 return param_set_uint_minmax(val, kp,
3167 static const struct kernel_param_ops param_ops_portnr = {
3168 .set = param_set_portnr,
3169 .get = param_get_uint,
3172 #define param_check_portnr(name, p) \
3173 __param_check(name, p, unsigned int);
3175 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3176 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3178 static int param_set_slot_table_size(const char *val,
3179 const struct kernel_param *kp)
3181 return param_set_uint_minmax(val, kp,
3183 RPC_MAX_SLOT_TABLE);
3186 static const struct kernel_param_ops param_ops_slot_table_size = {
3187 .set = param_set_slot_table_size,
3188 .get = param_get_uint,
3191 #define param_check_slot_table_size(name, p) \
3192 __param_check(name, p, unsigned int);
3194 static int param_set_max_slot_table_size(const char *val,
3195 const struct kernel_param *kp)
3197 return param_set_uint_minmax(val, kp,
3199 RPC_MAX_SLOT_TABLE_LIMIT);
3202 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3203 .set = param_set_max_slot_table_size,
3204 .get = param_get_uint,
3207 #define param_check_max_slot_table_size(name, p) \
3208 __param_check(name, p, unsigned int);
3210 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3211 slot_table_size, 0644);
3212 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3213 max_slot_table_size, 0644);
3214 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3215 slot_table_size, 0644);