1 // SPDX-License-Identifier: GPL-2.0
3 * linux/net/sunrpc/xprtsock.c
5 * Client-side transport implementation for sockets.
7 * TCP callback races fixes (C) 1998 Red Hat
8 * TCP send fixes (C) 1998 Red Hat
9 * TCP NFS related read + write fixes
10 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
12 * Rewrite of larges part of the code in order to stabilize TCP stuff.
13 * Fix behaviour when socket buffer is full.
14 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
16 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
18 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
19 * <gilles.quillard@bull.net>
22 #include <linux/types.h>
23 #include <linux/string.h>
24 #include <linux/slab.h>
25 #include <linux/module.h>
26 #include <linux/capability.h>
27 #include <linux/pagemap.h>
28 #include <linux/errno.h>
29 #include <linux/socket.h>
31 #include <linux/net.h>
34 #include <linux/udp.h>
35 #include <linux/tcp.h>
36 #include <linux/sunrpc/clnt.h>
37 #include <linux/sunrpc/addr.h>
38 #include <linux/sunrpc/sched.h>
39 #include <linux/sunrpc/svcsock.h>
40 #include <linux/sunrpc/xprtsock.h>
41 #include <linux/file.h>
42 #ifdef CONFIG_SUNRPC_BACKCHANNEL
43 #include <linux/sunrpc/bc_xprt.h>
47 #include <net/checksum.h>
50 #include <linux/bvec.h>
51 #include <linux/highmem.h>
52 #include <linux/uio.h>
53 #include <linux/sched/mm.h>
55 #include <trace/events/sunrpc.h>
60 static void xs_close(struct rpc_xprt *xprt);
61 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
67 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
68 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
69 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
71 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
72 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
74 #define XS_TCP_LINGER_TO (15U * HZ)
75 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
78 * We can register our own files under /proc/sys/sunrpc by
79 * calling register_sysctl_table() again. The files in that
80 * directory become the union of all files registered there.
82 * We simply need to make sure that we don't collide with
83 * someone else's file names!
86 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
87 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
88 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
89 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
90 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
92 static struct ctl_table_header *sunrpc_table_header;
94 static struct xprt_class xs_local_transport;
95 static struct xprt_class xs_udp_transport;
96 static struct xprt_class xs_tcp_transport;
97 static struct xprt_class xs_bc_tcp_transport;
100 * FIXME: changing the UDP slot table size should also resize the UDP
101 * socket buffers for existing UDP transports
103 static struct ctl_table xs_tunables_table[] = {
105 .procname = "udp_slot_table_entries",
106 .data = &xprt_udp_slot_table_entries,
107 .maxlen = sizeof(unsigned int),
109 .proc_handler = proc_dointvec_minmax,
110 .extra1 = &min_slot_table_size,
111 .extra2 = &max_slot_table_size
114 .procname = "tcp_slot_table_entries",
115 .data = &xprt_tcp_slot_table_entries,
116 .maxlen = sizeof(unsigned int),
118 .proc_handler = proc_dointvec_minmax,
119 .extra1 = &min_slot_table_size,
120 .extra2 = &max_slot_table_size
123 .procname = "tcp_max_slot_table_entries",
124 .data = &xprt_max_tcp_slot_table_entries,
125 .maxlen = sizeof(unsigned int),
127 .proc_handler = proc_dointvec_minmax,
128 .extra1 = &min_slot_table_size,
129 .extra2 = &max_tcp_slot_table_limit
132 .procname = "min_resvport",
133 .data = &xprt_min_resvport,
134 .maxlen = sizeof(unsigned int),
136 .proc_handler = proc_dointvec_minmax,
137 .extra1 = &xprt_min_resvport_limit,
138 .extra2 = &xprt_max_resvport_limit
141 .procname = "max_resvport",
142 .data = &xprt_max_resvport,
143 .maxlen = sizeof(unsigned int),
145 .proc_handler = proc_dointvec_minmax,
146 .extra1 = &xprt_min_resvport_limit,
147 .extra2 = &xprt_max_resvport_limit
150 .procname = "tcp_fin_timeout",
151 .data = &xs_tcp_fin_timeout,
152 .maxlen = sizeof(xs_tcp_fin_timeout),
154 .proc_handler = proc_dointvec_jiffies,
159 static struct ctl_table sunrpc_table[] = {
161 .procname = "sunrpc",
163 .child = xs_tunables_table
169 * Wait duration for a reply from the RPC portmapper.
171 #define XS_BIND_TO (60U * HZ)
174 * Delay if a UDP socket connect error occurs. This is most likely some
175 * kind of resource problem on the local host.
177 #define XS_UDP_REEST_TO (2U * HZ)
180 * The reestablish timeout allows clients to delay for a bit before attempting
181 * to reconnect to a server that just dropped our connection.
183 * We implement an exponential backoff when trying to reestablish a TCP
184 * transport connection with the server. Some servers like to drop a TCP
185 * connection when they are overworked, so we start with a short timeout and
186 * increase over time if the server is down or not responding.
188 #define XS_TCP_INIT_REEST_TO (3U * HZ)
191 * TCP idle timeout; client drops the transport socket if it is idle
192 * for this long. Note that we also timeout UDP sockets to prevent
193 * holding port numbers when there is no RPC traffic.
195 #define XS_IDLE_DISC_TO (5U * 60 * HZ)
197 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
198 # undef RPC_DEBUG_DATA
199 # define RPCDBG_FACILITY RPCDBG_TRANS
202 #ifdef RPC_DEBUG_DATA
203 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
205 u8 *buf = (u8 *) packet;
208 dprintk("RPC: %s\n", msg);
209 for (j = 0; j < count && j < 128; j += 4) {
213 dprintk("0x%04x ", j);
215 dprintk("%02x%02x%02x%02x ",
216 buf[j], buf[j+1], buf[j+2], buf[j+3]);
221 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
227 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
229 return (struct rpc_xprt *) sk->sk_user_data;
232 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
234 return (struct sockaddr *) &xprt->addr;
237 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
239 return (struct sockaddr_un *) &xprt->addr;
242 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
244 return (struct sockaddr_in *) &xprt->addr;
247 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
249 return (struct sockaddr_in6 *) &xprt->addr;
252 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
254 struct sockaddr *sap = xs_addr(xprt);
255 struct sockaddr_in6 *sin6;
256 struct sockaddr_in *sin;
257 struct sockaddr_un *sun;
260 switch (sap->sa_family) {
262 sun = xs_addr_un(xprt);
263 strlcpy(buf, sun->sun_path, sizeof(buf));
264 xprt->address_strings[RPC_DISPLAY_ADDR] =
265 kstrdup(buf, GFP_KERNEL);
268 (void)rpc_ntop(sap, buf, sizeof(buf));
269 xprt->address_strings[RPC_DISPLAY_ADDR] =
270 kstrdup(buf, GFP_KERNEL);
271 sin = xs_addr_in(xprt);
272 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
275 (void)rpc_ntop(sap, buf, sizeof(buf));
276 xprt->address_strings[RPC_DISPLAY_ADDR] =
277 kstrdup(buf, GFP_KERNEL);
278 sin6 = xs_addr_in6(xprt);
279 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
285 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
288 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
290 struct sockaddr *sap = xs_addr(xprt);
293 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
294 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
296 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
297 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
300 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
301 const char *protocol,
304 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
305 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
306 xs_format_common_peer_addresses(xprt);
307 xs_format_common_peer_ports(xprt);
310 static void xs_update_peer_port(struct rpc_xprt *xprt)
312 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
313 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
315 xs_format_common_peer_ports(xprt);
318 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
322 for (i = 0; i < RPC_DISPLAY_MAX; i++)
324 case RPC_DISPLAY_PROTO:
325 case RPC_DISPLAY_NETID:
328 kfree(xprt->address_strings[i]);
333 xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp)
337 if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES))
339 n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT;
340 for (i = 0; i < n; i++) {
343 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp);
344 if (!buf->pages[i]) {
346 return i > buf->page_base ? i - buf->page_base : 0;
353 xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek)
357 iov_iter_advance(&msg->msg_iter, seek);
358 ret = sock_recvmsg(sock, msg, flags);
359 return ret > 0 ? ret + seek : ret;
363 xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags,
364 struct kvec *kvec, size_t count, size_t seek)
366 iov_iter_kvec(&msg->msg_iter, READ, kvec, 1, count);
367 return xs_sock_recvmsg(sock, msg, flags, seek);
371 xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags,
372 struct bio_vec *bvec, unsigned long nr, size_t count,
375 iov_iter_bvec(&msg->msg_iter, READ, bvec, nr, count);
376 return xs_sock_recvmsg(sock, msg, flags, seek);
380 xs_read_discard(struct socket *sock, struct msghdr *msg, int flags,
383 iov_iter_discard(&msg->msg_iter, READ, count);
384 return sock_recvmsg(sock, msg, flags);
387 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
389 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
391 struct bvec_iter bi = {
396 bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
397 for_each_bvec(bv, bvec, bi, bi)
398 flush_dcache_page(bv.bv_page);
402 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
408 xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
409 struct xdr_buf *buf, size_t count, size_t seek, size_t *read)
411 size_t want, seek_init = seek, offset = 0;
414 want = min_t(size_t, count, buf->head[0].iov_len);
416 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek);
420 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
430 want = xs_alloc_sparse_pages(buf,
431 min_t(size_t, count - offset, buf->page_len),
434 ret = xs_read_bvec(sock, msg, flags, buf->bvec,
435 xdr_buf_pagecount(buf),
436 want + buf->page_base,
437 seek + buf->page_base);
440 xs_flush_bvec(buf->bvec, ret, seek + buf->page_base);
441 ret -= buf->page_base;
443 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
453 want = min_t(size_t, count - offset, buf->tail[0].iov_len);
455 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek);
459 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
463 } else if (offset < seek_init)
467 *read = offset - seek_init;
475 xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf)
477 if (!transport->recv.copied) {
478 if (buf->head[0].iov_len >= transport->recv.offset)
479 memcpy(buf->head[0].iov_base,
480 &transport->recv.xid,
481 transport->recv.offset);
482 transport->recv.copied = transport->recv.offset;
487 xs_read_stream_request_done(struct sock_xprt *transport)
489 return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT);
493 xs_read_stream_check_eor(struct sock_xprt *transport,
496 if (xs_read_stream_request_done(transport))
497 msg->msg_flags |= MSG_EOR;
501 xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg,
502 int flags, struct rpc_rqst *req)
504 struct xdr_buf *buf = &req->rq_private_buf;
508 xs_read_header(transport, buf);
510 want = transport->recv.len - transport->recv.offset;
512 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf,
513 transport->recv.copied + want,
514 transport->recv.copied,
516 transport->recv.offset += read;
517 transport->recv.copied += read;
520 if (transport->recv.offset == transport->recv.len)
521 xs_read_stream_check_eor(transport, msg);
531 msg->msg_flags |= MSG_TRUNC;
536 return ret < 0 ? ret : read;
540 xs_read_stream_headersize(bool isfrag)
543 return sizeof(__be32);
544 return 3 * sizeof(__be32);
548 xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg,
549 int flags, size_t want, size_t seek)
552 .iov_base = &transport->recv.fraghdr,
555 return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek);
558 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
560 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
562 struct rpc_xprt *xprt = &transport->xprt;
563 struct rpc_rqst *req;
566 /* Is this transport associated with the backchannel? */
570 /* Look up and lock the request corresponding to the given XID */
571 req = xprt_lookup_bc_request(xprt, transport->recv.xid);
573 printk(KERN_WARNING "Callback slot table overflowed\n");
576 if (transport->recv.copied && !req->rq_private_buf.len)
579 ret = xs_read_stream_request(transport, msg, flags, req);
580 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
581 xprt_complete_bc_request(req, transport->recv.copied);
583 req->rq_private_buf.len = transport->recv.copied;
587 #else /* CONFIG_SUNRPC_BACKCHANNEL */
589 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
593 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
596 xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags)
598 struct rpc_xprt *xprt = &transport->xprt;
599 struct rpc_rqst *req;
602 /* Look up and lock the request corresponding to the given XID */
603 spin_lock(&xprt->queue_lock);
604 req = xprt_lookup_rqst(xprt, transport->recv.xid);
605 if (!req || (transport->recv.copied && !req->rq_private_buf.len)) {
606 msg->msg_flags |= MSG_TRUNC;
610 spin_unlock(&xprt->queue_lock);
612 ret = xs_read_stream_request(transport, msg, flags, req);
614 spin_lock(&xprt->queue_lock);
615 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
616 xprt_complete_rqst(req->rq_task, transport->recv.copied);
618 req->rq_private_buf.len = transport->recv.copied;
619 xprt_unpin_rqst(req);
621 spin_unlock(&xprt->queue_lock);
626 xs_read_stream(struct sock_xprt *transport, int flags)
628 struct msghdr msg = { 0 };
629 size_t want, read = 0;
632 if (transport->recv.len == 0) {
633 want = xs_read_stream_headersize(transport->recv.copied != 0);
634 ret = xs_read_stream_header(transport, &msg, flags, want,
635 transport->recv.offset);
638 transport->recv.offset = ret;
639 if (transport->recv.offset != want)
640 return transport->recv.offset;
641 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) &
642 RPC_FRAGMENT_SIZE_MASK;
643 transport->recv.offset -= sizeof(transport->recv.fraghdr);
647 switch (be32_to_cpu(transport->recv.calldir)) {
649 msg.msg_flags |= MSG_TRUNC;
652 ret = xs_read_stream_call(transport, &msg, flags);
655 ret = xs_read_stream_reply(transport, &msg, flags);
657 if (msg.msg_flags & MSG_TRUNC) {
658 transport->recv.calldir = cpu_to_be32(-1);
659 transport->recv.copied = -1;
664 if (transport->recv.offset < transport->recv.len) {
665 if (!(msg.msg_flags & MSG_TRUNC))
668 ret = xs_read_discard(transport->sock, &msg, flags,
669 transport->recv.len - transport->recv.offset);
672 transport->recv.offset += ret;
674 if (transport->recv.offset != transport->recv.len)
677 if (xs_read_stream_request_done(transport)) {
678 trace_xs_stream_read_request(transport);
679 transport->recv.copied = 0;
681 transport->recv.offset = 0;
682 transport->recv.len = 0;
685 return ret != 0 ? ret : -ESHUTDOWN;
688 static __poll_t xs_poll_socket(struct sock_xprt *transport)
690 return transport->sock->ops->poll(transport->file, transport->sock,
694 static bool xs_poll_socket_readable(struct sock_xprt *transport)
696 __poll_t events = xs_poll_socket(transport);
698 return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP);
701 static void xs_poll_check_readable(struct sock_xprt *transport)
704 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
705 if (!xs_poll_socket_readable(transport))
707 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
708 queue_work(xprtiod_workqueue, &transport->recv_worker);
711 static void xs_stream_data_receive(struct sock_xprt *transport)
716 mutex_lock(&transport->recv_mutex);
717 if (transport->sock == NULL)
720 ret = xs_read_stream(transport, MSG_DONTWAIT);
726 if (ret == -ESHUTDOWN)
727 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
729 xs_poll_check_readable(transport);
731 mutex_unlock(&transport->recv_mutex);
732 trace_xs_stream_read_data(&transport->xprt, ret, read);
735 static void xs_stream_data_receive_workfn(struct work_struct *work)
737 struct sock_xprt *transport =
738 container_of(work, struct sock_xprt, recv_worker);
739 unsigned int pflags = memalloc_nofs_save();
741 xs_stream_data_receive(transport);
742 memalloc_nofs_restore(pflags);
746 xs_stream_reset_connect(struct sock_xprt *transport)
748 transport->recv.offset = 0;
749 transport->recv.len = 0;
750 transport->recv.copied = 0;
751 transport->xmit.offset = 0;
755 xs_stream_start_connect(struct sock_xprt *transport)
757 transport->xprt.stat.connect_count++;
758 transport->xprt.stat.connect_start = jiffies;
761 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
764 * xs_nospace - handle transmit was incomplete
765 * @req: pointer to RPC request
768 static int xs_nospace(struct rpc_rqst *req)
770 struct rpc_xprt *xprt = req->rq_xprt;
771 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
772 struct sock *sk = transport->inet;
775 trace_rpc_socket_nospace(req, transport);
777 /* Protect against races with write_space */
778 spin_lock(&xprt->transport_lock);
780 /* Don't race with disconnect */
781 if (xprt_connected(xprt)) {
782 /* wait for more buffer space */
783 sk->sk_write_pending++;
784 xprt_wait_for_buffer_space(xprt);
788 spin_unlock(&xprt->transport_lock);
790 /* Race breaker in case memory is freed before above code is called */
791 if (ret == -EAGAIN) {
792 struct socket_wq *wq;
795 wq = rcu_dereference(sk->sk_wq);
796 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
799 sk->sk_write_space(sk);
805 xs_stream_prepare_request(struct rpc_rqst *req)
807 xdr_free_bvec(&req->rq_rcv_buf);
808 req->rq_task->tk_status = xdr_alloc_bvec(&req->rq_rcv_buf, GFP_KERNEL);
812 * Determine if the previous message in the stream was aborted before it
813 * could complete transmission.
816 xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req)
818 return transport->xmit.offset != 0 && req->rq_bytes_sent == 0;
822 * Return the stream record marker field for a record of length < 2^31-1
825 xs_stream_record_marker(struct xdr_buf *xdr)
829 return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len);
833 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
834 * @req: pointer to RPC request
837 * 0: The request has been sent
838 * EAGAIN: The socket was blocked, please call again later to
839 * complete the request
840 * ENOTCONN: Caller needs to invoke connect logic then call again
841 * other: Some other error occurred, the request was not sent
843 static int xs_local_send_request(struct rpc_rqst *req)
845 struct rpc_xprt *xprt = req->rq_xprt;
846 struct sock_xprt *transport =
847 container_of(xprt, struct sock_xprt, xprt);
848 struct xdr_buf *xdr = &req->rq_snd_buf;
849 rpc_fraghdr rm = xs_stream_record_marker(xdr);
850 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
851 struct msghdr msg = {
852 .msg_flags = XS_SENDMSG_FLAGS,
857 /* Close the stream if the previous transmission was incomplete */
858 if (xs_send_request_was_aborted(transport, req)) {
863 xs_pktdump("packet data:",
864 req->rq_svec->iov_base, req->rq_svec->iov_len);
866 req->rq_xtime = ktime_get();
867 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
868 transport->xmit.offset, rm, &sent);
869 dprintk("RPC: %s(%u) = %d\n",
870 __func__, xdr->len - transport->xmit.offset, status);
872 if (status == -EAGAIN && sock_writeable(transport->inet))
875 if (likely(sent > 0) || status == 0) {
876 transport->xmit.offset += sent;
877 req->rq_bytes_sent = transport->xmit.offset;
878 if (likely(req->rq_bytes_sent >= msglen)) {
879 req->rq_xmit_bytes_sent += transport->xmit.offset;
880 transport->xmit.offset = 0;
890 status = xs_nospace(req);
893 dprintk("RPC: sendmsg returned unrecognized error %d\n",
905 * xs_udp_send_request - write an RPC request to a UDP socket
906 * @req: pointer to RPC request
909 * 0: The request has been sent
910 * EAGAIN: The socket was blocked, please call again later to
911 * complete the request
912 * ENOTCONN: Caller needs to invoke connect logic then call again
913 * other: Some other error occurred, the request was not sent
915 static int xs_udp_send_request(struct rpc_rqst *req)
917 struct rpc_xprt *xprt = req->rq_xprt;
918 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
919 struct xdr_buf *xdr = &req->rq_snd_buf;
920 struct msghdr msg = {
921 .msg_name = xs_addr(xprt),
922 .msg_namelen = xprt->addrlen,
923 .msg_flags = XS_SENDMSG_FLAGS,
928 xs_pktdump("packet data:",
929 req->rq_svec->iov_base,
930 req->rq_svec->iov_len);
932 if (!xprt_bound(xprt))
935 if (!xprt_request_get_cong(xprt, req))
938 req->rq_xtime = ktime_get();
939 status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, 0, &sent);
941 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
944 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
945 if (status == -EPERM)
948 if (status == -EAGAIN && sock_writeable(transport->inet))
951 if (sent > 0 || status == 0) {
952 req->rq_xmit_bytes_sent += sent;
953 if (sent >= req->rq_slen)
955 /* Still some bytes left; set up for a retry later. */
963 /* Should we call xs_close() here? */
966 status = xs_nospace(req);
973 /* When the server has died, an ICMP port unreachable message
974 * prompts ECONNREFUSED. */
977 dprintk("RPC: sendmsg returned unrecognized error %d\n",
985 * xs_tcp_send_request - write an RPC request to a TCP socket
986 * @req: pointer to RPC request
989 * 0: The request has been sent
990 * EAGAIN: The socket was blocked, please call again later to
991 * complete the request
992 * ENOTCONN: Caller needs to invoke connect logic then call again
993 * other: Some other error occurred, the request was not sent
995 * XXX: In the case of soft timeouts, should we eventually give up
996 * if sendmsg is not able to make progress?
998 static int xs_tcp_send_request(struct rpc_rqst *req)
1000 struct rpc_xprt *xprt = req->rq_xprt;
1001 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1002 struct xdr_buf *xdr = &req->rq_snd_buf;
1003 rpc_fraghdr rm = xs_stream_record_marker(xdr);
1004 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
1005 struct msghdr msg = {
1006 .msg_flags = XS_SENDMSG_FLAGS,
1008 bool vm_wait = false;
1012 /* Close the stream if the previous transmission was incomplete */
1013 if (xs_send_request_was_aborted(transport, req)) {
1014 if (transport->sock != NULL)
1015 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
1018 if (!transport->inet)
1021 xs_pktdump("packet data:",
1022 req->rq_svec->iov_base,
1023 req->rq_svec->iov_len);
1025 if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
1026 xs_tcp_set_socket_timeouts(xprt, transport->sock);
1028 /* Continue transmitting the packet/record. We must be careful
1029 * to cope with writespace callbacks arriving _after_ we have
1030 * called sendmsg(). */
1031 req->rq_xtime = ktime_get();
1032 tcp_sock_set_cork(transport->inet, true);
1034 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
1035 transport->xmit.offset, rm, &sent);
1037 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
1038 xdr->len - transport->xmit.offset, status);
1040 /* If we've sent the entire packet, immediately
1041 * reset the count of bytes sent. */
1042 transport->xmit.offset += sent;
1043 req->rq_bytes_sent = transport->xmit.offset;
1044 if (likely(req->rq_bytes_sent >= msglen)) {
1045 req->rq_xmit_bytes_sent += transport->xmit.offset;
1046 transport->xmit.offset = 0;
1047 if (atomic_long_read(&xprt->xmit_queuelen) == 1)
1048 tcp_sock_set_cork(transport->inet, false);
1052 WARN_ON_ONCE(sent == 0 && status == 0);
1054 if (status == -EAGAIN ) {
1056 * Return EAGAIN if we're sure we're hitting the
1057 * socket send buffer limits.
1059 if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
1062 * Did we hit a memory allocation failure?
1068 /* Retry, knowing now that we're below the
1069 * socket send buffer limit
1083 /* Should we call xs_close() here? */
1086 status = xs_nospace(req);
1096 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1103 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1105 transport->old_data_ready = sk->sk_data_ready;
1106 transport->old_state_change = sk->sk_state_change;
1107 transport->old_write_space = sk->sk_write_space;
1108 transport->old_error_report = sk->sk_error_report;
1111 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1113 sk->sk_data_ready = transport->old_data_ready;
1114 sk->sk_state_change = transport->old_state_change;
1115 sk->sk_write_space = transport->old_write_space;
1116 sk->sk_error_report = transport->old_error_report;
1119 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
1121 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1123 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1124 clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state);
1125 clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state);
1126 clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state);
1129 static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr)
1131 set_bit(nr, &transport->sock_state);
1132 queue_work(xprtiod_workqueue, &transport->error_worker);
1135 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1137 xprt->connect_cookie++;
1138 smp_mb__before_atomic();
1139 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1140 clear_bit(XPRT_CLOSING, &xprt->state);
1141 xs_sock_reset_state_flags(xprt);
1142 smp_mb__after_atomic();
1146 * xs_error_report - callback to handle TCP socket state errors
1149 * Note: we don't call sock_error() since there may be a rpc_task
1150 * using the socket, and so we don't want to clear sk->sk_err.
1152 static void xs_error_report(struct sock *sk)
1154 struct sock_xprt *transport;
1155 struct rpc_xprt *xprt;
1157 if (!(xprt = xprt_from_sock(sk)))
1160 transport = container_of(xprt, struct sock_xprt, xprt);
1161 transport->xprt_err = -sk->sk_err;
1162 if (transport->xprt_err == 0)
1164 dprintk("RPC: xs_error_report client %p, error=%d...\n",
1165 xprt, -transport->xprt_err);
1166 trace_rpc_socket_error(xprt, sk->sk_socket, transport->xprt_err);
1168 /* barrier ensures xprt_err is set before XPRT_SOCK_WAKE_ERROR */
1169 smp_mb__before_atomic();
1170 xs_run_error_worker(transport, XPRT_SOCK_WAKE_ERROR);
1173 static void xs_reset_transport(struct sock_xprt *transport)
1175 struct socket *sock = transport->sock;
1176 struct sock *sk = transport->inet;
1177 struct rpc_xprt *xprt = &transport->xprt;
1178 struct file *filp = transport->file;
1183 if (atomic_read(&transport->xprt.swapper))
1184 sk_clear_memalloc(sk);
1186 kernel_sock_shutdown(sock, SHUT_RDWR);
1188 mutex_lock(&transport->recv_mutex);
1190 transport->inet = NULL;
1191 transport->sock = NULL;
1192 transport->file = NULL;
1194 sk->sk_user_data = NULL;
1196 xs_restore_old_callbacks(transport, sk);
1197 xprt_clear_connected(xprt);
1198 xs_sock_reset_connection_flags(xprt);
1199 /* Reset stream record info */
1200 xs_stream_reset_connect(transport);
1202 mutex_unlock(&transport->recv_mutex);
1204 trace_rpc_socket_close(xprt, sock);
1207 xprt_disconnect_done(xprt);
1211 * xs_close - close a socket
1214 * This is used when all requests are complete; ie, no DRC state remains
1215 * on the server we want to save.
1217 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1218 * xs_reset_transport() zeroing the socket from underneath a writer.
1220 static void xs_close(struct rpc_xprt *xprt)
1222 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1224 dprintk("RPC: xs_close xprt %p\n", xprt);
1226 xs_reset_transport(transport);
1227 xprt->reestablish_timeout = 0;
1230 static void xs_inject_disconnect(struct rpc_xprt *xprt)
1232 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
1234 xprt_disconnect_done(xprt);
1237 static void xs_xprt_free(struct rpc_xprt *xprt)
1239 xs_free_peer_addresses(xprt);
1244 * xs_destroy - prepare to shutdown a transport
1245 * @xprt: doomed transport
1248 static void xs_destroy(struct rpc_xprt *xprt)
1250 struct sock_xprt *transport = container_of(xprt,
1251 struct sock_xprt, xprt);
1252 dprintk("RPC: xs_destroy xprt %p\n", xprt);
1254 cancel_delayed_work_sync(&transport->connect_worker);
1256 cancel_work_sync(&transport->recv_worker);
1257 cancel_work_sync(&transport->error_worker);
1259 module_put(THIS_MODULE);
1263 * xs_udp_data_read_skb - receive callback for UDP sockets
1269 static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1271 struct sk_buff *skb)
1273 struct rpc_task *task;
1274 struct rpc_rqst *rovr;
1275 int repsize, copied;
1281 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
1285 /* Copy the XID from the skb... */
1286 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1290 /* Look up and lock the request corresponding to the given XID */
1291 spin_lock(&xprt->queue_lock);
1292 rovr = xprt_lookup_rqst(xprt, *xp);
1295 xprt_pin_rqst(rovr);
1296 xprt_update_rtt(rovr->rq_task);
1297 spin_unlock(&xprt->queue_lock);
1298 task = rovr->rq_task;
1300 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1303 /* Suck it into the iovec, verify checksum if not done by hw. */
1304 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1305 spin_lock(&xprt->queue_lock);
1306 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1311 spin_lock(&xprt->transport_lock);
1312 xprt_adjust_cwnd(xprt, task, copied);
1313 spin_unlock(&xprt->transport_lock);
1314 spin_lock(&xprt->queue_lock);
1315 xprt_complete_rqst(task, copied);
1316 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1318 xprt_unpin_rqst(rovr);
1320 spin_unlock(&xprt->queue_lock);
1323 static void xs_udp_data_receive(struct sock_xprt *transport)
1325 struct sk_buff *skb;
1329 mutex_lock(&transport->recv_mutex);
1330 sk = transport->inet;
1334 skb = skb_recv_udp(sk, 0, 1, &err);
1337 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1341 xs_poll_check_readable(transport);
1343 mutex_unlock(&transport->recv_mutex);
1346 static void xs_udp_data_receive_workfn(struct work_struct *work)
1348 struct sock_xprt *transport =
1349 container_of(work, struct sock_xprt, recv_worker);
1350 unsigned int pflags = memalloc_nofs_save();
1352 xs_udp_data_receive(transport);
1353 memalloc_nofs_restore(pflags);
1357 * xs_data_ready - "data ready" callback for UDP sockets
1358 * @sk: socket with data to read
1361 static void xs_data_ready(struct sock *sk)
1363 struct rpc_xprt *xprt;
1365 dprintk("RPC: xs_data_ready...\n");
1366 xprt = xprt_from_sock(sk);
1368 struct sock_xprt *transport = container_of(xprt,
1369 struct sock_xprt, xprt);
1370 transport->old_data_ready(sk);
1371 /* Any data means we had a useful conversation, so
1372 * then we don't need to delay the next reconnect
1374 if (xprt->reestablish_timeout)
1375 xprt->reestablish_timeout = 0;
1376 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1377 queue_work(xprtiod_workqueue, &transport->recv_worker);
1382 * Helper function to force a TCP close if the server is sending
1383 * junk and/or it has put us in CLOSE_WAIT
1385 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1387 xprt_force_disconnect(xprt);
1390 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1391 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1395 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1398 * xs_tcp_state_change - callback to handle TCP socket state changes
1399 * @sk: socket whose state has changed
1402 static void xs_tcp_state_change(struct sock *sk)
1404 struct rpc_xprt *xprt;
1405 struct sock_xprt *transport;
1407 if (!(xprt = xprt_from_sock(sk)))
1409 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1410 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1411 sk->sk_state, xprt_connected(xprt),
1412 sock_flag(sk, SOCK_DEAD),
1413 sock_flag(sk, SOCK_ZAPPED),
1416 transport = container_of(xprt, struct sock_xprt, xprt);
1417 trace_rpc_socket_state_change(xprt, sk->sk_socket);
1418 switch (sk->sk_state) {
1419 case TCP_ESTABLISHED:
1420 if (!xprt_test_and_set_connected(xprt)) {
1421 xprt->connect_cookie++;
1422 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1423 xprt_clear_connecting(xprt);
1425 xprt->stat.connect_count++;
1426 xprt->stat.connect_time += (long)jiffies -
1427 xprt->stat.connect_start;
1428 xs_run_error_worker(transport, XPRT_SOCK_WAKE_PENDING);
1432 /* The client initiated a shutdown of the socket */
1433 xprt->connect_cookie++;
1434 xprt->reestablish_timeout = 0;
1435 set_bit(XPRT_CLOSING, &xprt->state);
1436 smp_mb__before_atomic();
1437 clear_bit(XPRT_CONNECTED, &xprt->state);
1438 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1439 smp_mb__after_atomic();
1441 case TCP_CLOSE_WAIT:
1442 /* The server initiated a shutdown of the socket */
1443 xprt->connect_cookie++;
1444 clear_bit(XPRT_CONNECTED, &xprt->state);
1445 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1449 * If the server closed down the connection, make sure that
1450 * we back off before reconnecting
1452 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1453 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1456 set_bit(XPRT_CLOSING, &xprt->state);
1457 smp_mb__before_atomic();
1458 clear_bit(XPRT_CONNECTED, &xprt->state);
1459 smp_mb__after_atomic();
1462 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1463 &transport->sock_state))
1464 xprt_clear_connecting(xprt);
1465 clear_bit(XPRT_CLOSING, &xprt->state);
1466 /* Trigger the socket release */
1467 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1471 static void xs_write_space(struct sock *sk)
1473 struct socket_wq *wq;
1474 struct sock_xprt *transport;
1475 struct rpc_xprt *xprt;
1479 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1481 if (unlikely(!(xprt = xprt_from_sock(sk))))
1483 transport = container_of(xprt, struct sock_xprt, xprt);
1485 wq = rcu_dereference(sk->sk_wq);
1486 if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1489 xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE);
1490 sk->sk_write_pending--;
1496 * xs_udp_write_space - callback invoked when socket buffer space
1498 * @sk: socket whose state has changed
1500 * Called when more output buffer space is available for this socket.
1501 * We try not to wake our writers until they can make "significant"
1502 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1503 * with a bunch of small requests.
1505 static void xs_udp_write_space(struct sock *sk)
1507 /* from net/core/sock.c:sock_def_write_space */
1508 if (sock_writeable(sk))
1513 * xs_tcp_write_space - callback invoked when socket buffer space
1515 * @sk: socket whose state has changed
1517 * Called when more output buffer space is available for this socket.
1518 * We try not to wake our writers until they can make "significant"
1519 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1520 * with a bunch of small requests.
1522 static void xs_tcp_write_space(struct sock *sk)
1524 /* from net/core/stream.c:sk_stream_write_space */
1525 if (sk_stream_is_writeable(sk))
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 unsigned short get_srcport(struct rpc_xprt *xprt)
1643 struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt);
1644 unsigned short ret = 0;
1645 mutex_lock(&sock->recv_mutex);
1647 ret = xs_sock_getport(sock->sock);
1648 mutex_unlock(&sock->recv_mutex);
1651 EXPORT_SYMBOL(get_srcport);
1653 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1655 if (transport->srcport != 0)
1656 transport->srcport = 0;
1657 if (!transport->xprt.resvport)
1659 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1660 return xprt_max_resvport;
1663 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1665 struct sockaddr_storage myaddr;
1667 int port = xs_get_srcport(transport);
1668 unsigned short last;
1671 * If we are asking for any ephemeral port (i.e. port == 0 &&
1672 * transport->xprt.resvport == 0), don't bind. Let the local
1673 * port selection happen implicitly when the socket is used
1674 * (for example at connect time).
1676 * This ensures that we can continue to establish TCP
1677 * connections even when all local ephemeral ports are already
1678 * a part of some TCP connection. This makes no difference
1679 * for UDP sockets, but also doesn't harm them.
1681 * If we're asking for any reserved port (i.e. port == 0 &&
1682 * transport->xprt.resvport == 1) xs_get_srcport above will
1683 * ensure that port is non-zero and we will bind as needed.
1688 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1690 rpc_set_port((struct sockaddr *)&myaddr, port);
1691 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1692 transport->xprt.addrlen);
1694 if (transport->xprt.reuseport)
1695 transport->srcport = port;
1699 port = xs_next_srcport(transport, port);
1702 } while (err == -EADDRINUSE && nloop != 2);
1704 if (myaddr.ss_family == AF_INET)
1705 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1706 &((struct sockaddr_in *)&myaddr)->sin_addr,
1707 port, err ? "failed" : "ok", err);
1709 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1710 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1711 port, err ? "failed" : "ok", err);
1716 * We don't support autobind on AF_LOCAL sockets
1718 static void xs_local_rpcbind(struct rpc_task *task)
1720 xprt_set_bound(task->tk_xprt);
1723 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1727 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1728 static struct lock_class_key xs_key[3];
1729 static struct lock_class_key xs_slock_key[3];
1731 static inline void xs_reclassify_socketu(struct socket *sock)
1733 struct sock *sk = sock->sk;
1735 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1736 &xs_slock_key[0], "sk_lock-AF_LOCAL-RPC", &xs_key[0]);
1739 static inline void xs_reclassify_socket4(struct socket *sock)
1741 struct sock *sk = sock->sk;
1743 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1744 &xs_slock_key[1], "sk_lock-AF_INET-RPC", &xs_key[1]);
1747 static inline void xs_reclassify_socket6(struct socket *sock)
1749 struct sock *sk = sock->sk;
1751 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1752 &xs_slock_key[2], "sk_lock-AF_INET6-RPC", &xs_key[2]);
1755 static inline void xs_reclassify_socket(int family, struct socket *sock)
1757 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1762 xs_reclassify_socketu(sock);
1765 xs_reclassify_socket4(sock);
1768 xs_reclassify_socket6(sock);
1773 static inline void xs_reclassify_socket(int family, struct socket *sock)
1778 static void xs_dummy_setup_socket(struct work_struct *work)
1782 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1783 struct sock_xprt *transport, int family, int type,
1784 int protocol, bool reuseport)
1787 struct socket *sock;
1790 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1792 dprintk("RPC: can't create %d transport socket (%d).\n",
1796 xs_reclassify_socket(family, sock);
1799 sock_set_reuseport(sock->sk);
1801 err = xs_bind(transport, sock);
1807 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1809 return ERR_CAST(filp);
1810 transport->file = filp;
1814 return ERR_PTR(err);
1817 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1818 struct socket *sock)
1820 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1823 if (!transport->inet) {
1824 struct sock *sk = sock->sk;
1828 xs_save_old_callbacks(transport, sk);
1830 sk->sk_user_data = xprt;
1831 sk->sk_data_ready = xs_data_ready;
1832 sk->sk_write_space = xs_udp_write_space;
1833 sock_set_flag(sk, SOCK_FASYNC);
1834 sk->sk_error_report = xs_error_report;
1836 xprt_clear_connected(xprt);
1838 /* Reset to new socket */
1839 transport->sock = sock;
1840 transport->inet = sk;
1845 xs_stream_start_connect(transport);
1847 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1851 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1852 * @transport: socket transport to connect
1854 static int xs_local_setup_socket(struct sock_xprt *transport)
1856 struct rpc_xprt *xprt = &transport->xprt;
1858 struct socket *sock;
1861 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1862 SOCK_STREAM, 0, &sock, 1);
1864 dprintk("RPC: can't create AF_LOCAL "
1865 "transport socket (%d).\n", -status);
1868 xs_reclassify_socket(AF_LOCAL, sock);
1870 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1872 status = PTR_ERR(filp);
1875 transport->file = filp;
1877 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1878 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1880 status = xs_local_finish_connecting(xprt, sock);
1881 trace_rpc_socket_connect(xprt, sock, status);
1884 dprintk("RPC: xprt %p connected to %s\n",
1885 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1886 xprt->stat.connect_count++;
1887 xprt->stat.connect_time += (long)jiffies -
1888 xprt->stat.connect_start;
1889 xprt_set_connected(xprt);
1894 dprintk("RPC: xprt %p: socket %s does not exist\n",
1895 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1898 dprintk("RPC: xprt %p: connection refused for %s\n",
1899 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1902 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1904 xprt->address_strings[RPC_DISPLAY_ADDR]);
1908 xprt_clear_connecting(xprt);
1909 xprt_wake_pending_tasks(xprt, status);
1913 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1915 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1918 if (RPC_IS_ASYNC(task)) {
1920 * We want the AF_LOCAL connect to be resolved in the
1921 * filesystem namespace of the process making the rpc
1922 * call. Thus we connect synchronously.
1924 * If we want to support asynchronous AF_LOCAL calls,
1925 * we'll need to figure out how to pass a namespace to
1928 task->tk_rpc_status = -ENOTCONN;
1929 rpc_exit(task, -ENOTCONN);
1932 ret = xs_local_setup_socket(transport);
1933 if (ret && !RPC_IS_SOFTCONN(task))
1934 msleep_interruptible(15000);
1937 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
1939 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
1940 * know that we have exclusive access to the socket), to guard against
1941 * races with xs_reset_transport.
1943 static void xs_set_memalloc(struct rpc_xprt *xprt)
1945 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1949 * If there's no sock, then we have nothing to set. The
1950 * reconnecting process will get it for us.
1952 if (!transport->inet)
1954 if (atomic_read(&xprt->swapper))
1955 sk_set_memalloc(transport->inet);
1959 * xs_enable_swap - Tag this transport as being used for swap.
1960 * @xprt: transport to tag
1962 * Take a reference to this transport on behalf of the rpc_clnt, and
1963 * optionally mark it for swapping if it wasn't already.
1966 xs_enable_swap(struct rpc_xprt *xprt)
1968 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
1970 if (atomic_inc_return(&xprt->swapper) != 1)
1972 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
1973 return -ERESTARTSYS;
1975 sk_set_memalloc(xs->inet);
1976 xprt_release_xprt(xprt, NULL);
1981 * xs_disable_swap - Untag this transport as being used for swap.
1982 * @xprt: transport to tag
1984 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
1985 * swapper refcount goes to 0, untag the socket as a memalloc socket.
1988 xs_disable_swap(struct rpc_xprt *xprt)
1990 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
1992 if (!atomic_dec_and_test(&xprt->swapper))
1994 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
1997 sk_clear_memalloc(xs->inet);
1998 xprt_release_xprt(xprt, NULL);
2001 static void xs_set_memalloc(struct rpc_xprt *xprt)
2006 xs_enable_swap(struct rpc_xprt *xprt)
2012 xs_disable_swap(struct rpc_xprt *xprt)
2017 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2019 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2021 if (!transport->inet) {
2022 struct sock *sk = sock->sk;
2026 xs_save_old_callbacks(transport, sk);
2028 sk->sk_user_data = xprt;
2029 sk->sk_data_ready = xs_data_ready;
2030 sk->sk_write_space = xs_udp_write_space;
2031 sock_set_flag(sk, SOCK_FASYNC);
2033 xprt_set_connected(xprt);
2035 /* Reset to new socket */
2036 transport->sock = sock;
2037 transport->inet = sk;
2039 xs_set_memalloc(xprt);
2043 xs_udp_do_set_buffer_size(xprt);
2045 xprt->stat.connect_start = jiffies;
2048 static void xs_udp_setup_socket(struct work_struct *work)
2050 struct sock_xprt *transport =
2051 container_of(work, struct sock_xprt, connect_worker.work);
2052 struct rpc_xprt *xprt = &transport->xprt;
2053 struct socket *sock;
2056 sock = xs_create_sock(xprt, transport,
2057 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2058 IPPROTO_UDP, false);
2062 dprintk("RPC: worker connecting xprt %p via %s to "
2063 "%s (port %s)\n", xprt,
2064 xprt->address_strings[RPC_DISPLAY_PROTO],
2065 xprt->address_strings[RPC_DISPLAY_ADDR],
2066 xprt->address_strings[RPC_DISPLAY_PORT]);
2068 xs_udp_finish_connecting(xprt, sock);
2069 trace_rpc_socket_connect(xprt, sock, 0);
2072 xprt_clear_connecting(xprt);
2073 xprt_unlock_connect(xprt, transport);
2074 xprt_wake_pending_tasks(xprt, status);
2078 * xs_tcp_shutdown - gracefully shut down a TCP socket
2081 * Initiates a graceful shutdown of the TCP socket by calling the
2082 * equivalent of shutdown(SHUT_RDWR);
2084 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2086 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2087 struct socket *sock = transport->sock;
2088 int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2092 if (!xprt->reuseport) {
2100 case TCP_ESTABLISHED:
2101 case TCP_CLOSE_WAIT:
2102 kernel_sock_shutdown(sock, SHUT_RDWR);
2103 trace_rpc_socket_shutdown(xprt, sock);
2106 xs_reset_transport(transport);
2110 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2111 struct socket *sock)
2113 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2114 unsigned int keepidle;
2115 unsigned int keepcnt;
2118 spin_lock(&xprt->transport_lock);
2119 keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2120 keepcnt = xprt->timeout->to_retries + 1;
2121 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2122 (xprt->timeout->to_retries + 1);
2123 clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2124 spin_unlock(&xprt->transport_lock);
2126 /* TCP Keepalive options */
2127 sock_set_keepalive(sock->sk);
2128 tcp_sock_set_keepidle(sock->sk, keepidle);
2129 tcp_sock_set_keepintvl(sock->sk, keepidle);
2130 tcp_sock_set_keepcnt(sock->sk, keepcnt);
2132 /* TCP user timeout (see RFC5482) */
2133 tcp_sock_set_user_timeout(sock->sk, timeo);
2136 static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2137 unsigned long connect_timeout,
2138 unsigned long reconnect_timeout)
2140 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2141 struct rpc_timeout to;
2142 unsigned long initval;
2144 spin_lock(&xprt->transport_lock);
2145 if (reconnect_timeout < xprt->max_reconnect_timeout)
2146 xprt->max_reconnect_timeout = reconnect_timeout;
2147 if (connect_timeout < xprt->connect_timeout) {
2148 memcpy(&to, xprt->timeout, sizeof(to));
2149 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2150 /* Arbitrary lower limit */
2151 if (initval < XS_TCP_INIT_REEST_TO << 1)
2152 initval = XS_TCP_INIT_REEST_TO << 1;
2153 to.to_initval = initval;
2154 to.to_maxval = initval;
2155 memcpy(&transport->tcp_timeout, &to,
2156 sizeof(transport->tcp_timeout));
2157 xprt->timeout = &transport->tcp_timeout;
2158 xprt->connect_timeout = connect_timeout;
2160 set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2161 spin_unlock(&xprt->transport_lock);
2164 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2166 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2168 if (!transport->inet) {
2169 struct sock *sk = sock->sk;
2171 /* Avoid temporary address, they are bad for long-lived
2172 * connections such as NFS mounts.
2173 * RFC4941, section 3.6 suggests that:
2174 * Individual applications, which have specific
2175 * knowledge about the normal duration of connections,
2176 * MAY override this as appropriate.
2178 if (xs_addr(xprt)->sa_family == PF_INET6) {
2179 ip6_sock_set_addr_preferences(sk,
2180 IPV6_PREFER_SRC_PUBLIC);
2183 xs_tcp_set_socket_timeouts(xprt, sock);
2184 tcp_sock_set_nodelay(sk);
2188 xs_save_old_callbacks(transport, sk);
2190 sk->sk_user_data = xprt;
2191 sk->sk_data_ready = xs_data_ready;
2192 sk->sk_state_change = xs_tcp_state_change;
2193 sk->sk_write_space = xs_tcp_write_space;
2194 sock_set_flag(sk, SOCK_FASYNC);
2195 sk->sk_error_report = xs_error_report;
2197 /* socket options */
2198 sock_reset_flag(sk, SOCK_LINGER);
2200 xprt_clear_connected(xprt);
2202 /* Reset to new socket */
2203 transport->sock = sock;
2204 transport->inet = sk;
2209 if (!xprt_bound(xprt))
2212 xs_set_memalloc(xprt);
2214 xs_stream_start_connect(transport);
2216 /* Tell the socket layer to start connecting... */
2217 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2218 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2222 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2223 * @work: queued work item
2225 * Invoked by a work queue tasklet.
2227 static void xs_tcp_setup_socket(struct work_struct *work)
2229 struct sock_xprt *transport =
2230 container_of(work, struct sock_xprt, connect_worker.work);
2231 struct socket *sock = transport->sock;
2232 struct rpc_xprt *xprt = &transport->xprt;
2236 sock = xs_create_sock(xprt, transport,
2237 xs_addr(xprt)->sa_family, SOCK_STREAM,
2240 xprt_wake_pending_tasks(xprt, PTR_ERR(sock));
2245 dprintk("RPC: worker connecting xprt %p via %s to "
2246 "%s (port %s)\n", xprt,
2247 xprt->address_strings[RPC_DISPLAY_PROTO],
2248 xprt->address_strings[RPC_DISPLAY_ADDR],
2249 xprt->address_strings[RPC_DISPLAY_PORT]);
2251 status = xs_tcp_finish_connecting(xprt, sock);
2252 trace_rpc_socket_connect(xprt, sock, status);
2253 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2254 xprt, -status, xprt_connected(xprt),
2255 sock->sk->sk_state);
2258 xs_set_srcport(transport, sock);
2262 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2263 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2267 case -EADDRNOTAVAIL:
2268 /* Source port number is unavailable. Try a new one! */
2269 transport->srcport = 0;
2273 /* Happens, for instance, if the user specified a link
2274 * local IPv6 address without a scope-id.
2285 printk("%s: connect returned unhandled error %d\n",
2290 /* xs_tcp_force_close() wakes tasks with a fixed error code.
2291 * We need to wake them first to ensure the correct error code.
2293 xprt_wake_pending_tasks(xprt, status);
2294 xs_tcp_force_close(xprt);
2296 xprt_clear_connecting(xprt);
2298 xprt_unlock_connect(xprt, transport);
2302 * xs_connect - connect a socket to a remote endpoint
2303 * @xprt: pointer to transport structure
2304 * @task: address of RPC task that manages state of connect request
2306 * TCP: If the remote end dropped the connection, delay reconnecting.
2308 * UDP socket connects are synchronous, but we use a work queue anyway
2309 * to guarantee that even unprivileged user processes can set up a
2310 * socket on a privileged port.
2312 * If a UDP socket connect fails, the delay behavior here prevents
2313 * retry floods (hard mounts).
2315 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2317 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2318 unsigned long delay = 0;
2320 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2322 if (transport->sock != NULL && !xprt_connecting(xprt)) {
2323 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2325 xprt, xprt->reestablish_timeout / HZ);
2327 /* Start by resetting any existing state */
2328 xs_reset_transport(transport);
2330 delay = xprt_reconnect_delay(xprt);
2331 xprt_reconnect_backoff(xprt, XS_TCP_INIT_REEST_TO);
2334 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2336 queue_delayed_work(xprtiod_workqueue,
2337 &transport->connect_worker,
2341 static void xs_wake_disconnect(struct sock_xprt *transport)
2343 if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state))
2344 xs_tcp_force_close(&transport->xprt);
2347 static void xs_wake_write(struct sock_xprt *transport)
2349 if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state))
2350 xprt_write_space(&transport->xprt);
2353 static void xs_wake_error(struct sock_xprt *transport)
2357 if (!test_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2359 mutex_lock(&transport->recv_mutex);
2360 if (transport->sock == NULL)
2362 if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2364 sockerr = xchg(&transport->xprt_err, 0);
2366 xprt_wake_pending_tasks(&transport->xprt, sockerr);
2368 mutex_unlock(&transport->recv_mutex);
2371 static void xs_wake_pending(struct sock_xprt *transport)
2373 if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING, &transport->sock_state))
2374 xprt_wake_pending_tasks(&transport->xprt, -EAGAIN);
2377 static void xs_error_handle(struct work_struct *work)
2379 struct sock_xprt *transport = container_of(work,
2380 struct sock_xprt, error_worker);
2382 xs_wake_disconnect(transport);
2383 xs_wake_write(transport);
2384 xs_wake_error(transport);
2385 xs_wake_pending(transport);
2389 * xs_local_print_stats - display AF_LOCAL socket-specific stats
2390 * @xprt: rpc_xprt struct containing statistics
2394 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2398 if (xprt_connected(xprt))
2399 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2401 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2402 "%llu %llu %lu %llu %llu\n",
2403 xprt->stat.bind_count,
2404 xprt->stat.connect_count,
2405 xprt->stat.connect_time / HZ,
2409 xprt->stat.bad_xids,
2412 xprt->stat.max_slots,
2413 xprt->stat.sending_u,
2414 xprt->stat.pending_u);
2418 * xs_udp_print_stats - display UDP socket-specific stats
2419 * @xprt: rpc_xprt struct containing statistics
2423 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2425 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2427 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2430 xprt->stat.bind_count,
2433 xprt->stat.bad_xids,
2436 xprt->stat.max_slots,
2437 xprt->stat.sending_u,
2438 xprt->stat.pending_u);
2442 * xs_tcp_print_stats - display TCP socket-specific stats
2443 * @xprt: rpc_xprt struct containing statistics
2447 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2449 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2452 if (xprt_connected(xprt))
2453 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2455 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2456 "%llu %llu %lu %llu %llu\n",
2458 xprt->stat.bind_count,
2459 xprt->stat.connect_count,
2460 xprt->stat.connect_time / HZ,
2464 xprt->stat.bad_xids,
2467 xprt->stat.max_slots,
2468 xprt->stat.sending_u,
2469 xprt->stat.pending_u);
2473 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2474 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2475 * to use the server side send routines.
2477 static int bc_malloc(struct rpc_task *task)
2479 struct rpc_rqst *rqst = task->tk_rqstp;
2480 size_t size = rqst->rq_callsize;
2482 struct rpc_buffer *buf;
2484 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2485 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2490 page = alloc_page(GFP_KERNEL);
2494 buf = page_address(page);
2495 buf->len = PAGE_SIZE;
2497 rqst->rq_buffer = buf->data;
2498 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2503 * Free the space allocated in the bc_alloc routine
2505 static void bc_free(struct rpc_task *task)
2507 void *buffer = task->tk_rqstp->rq_buffer;
2508 struct rpc_buffer *buf;
2510 buf = container_of(buffer, struct rpc_buffer, data);
2511 free_page((unsigned long)buf);
2514 static int bc_sendto(struct rpc_rqst *req)
2516 struct xdr_buf *xdr = &req->rq_snd_buf;
2517 struct sock_xprt *transport =
2518 container_of(req->rq_xprt, struct sock_xprt, xprt);
2519 struct msghdr msg = {
2522 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
2524 unsigned int sent = 0;
2527 req->rq_xtime = ktime_get();
2528 err = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, marker, &sent);
2530 if (err < 0 || sent != (xdr->len + sizeof(marker)))
2536 * bc_send_request - Send a backchannel Call on a TCP socket
2537 * @req: rpc_rqst containing Call message to be sent
2539 * xpt_mutex ensures @rqstp's whole message is written to the socket
2540 * without interruption.
2543 * %0 if the message was sent successfully
2544 * %ENOTCONN if the message was not sent
2546 static int bc_send_request(struct rpc_rqst *req)
2548 struct svc_xprt *xprt;
2552 * Get the server socket associated with this callback xprt
2554 xprt = req->rq_xprt->bc_xprt;
2557 * Grab the mutex to serialize data as the connection is shared
2558 * with the fore channel
2560 mutex_lock(&xprt->xpt_mutex);
2561 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2564 len = bc_sendto(req);
2565 mutex_unlock(&xprt->xpt_mutex);
2574 * The close routine. Since this is client initiated, we do nothing
2577 static void bc_close(struct rpc_xprt *xprt)
2579 xprt_disconnect_done(xprt);
2583 * The xprt destroy routine. Again, because this connection is client
2584 * initiated, we do nothing
2587 static void bc_destroy(struct rpc_xprt *xprt)
2589 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2592 module_put(THIS_MODULE);
2595 static const struct rpc_xprt_ops xs_local_ops = {
2596 .reserve_xprt = xprt_reserve_xprt,
2597 .release_xprt = xprt_release_xprt,
2598 .alloc_slot = xprt_alloc_slot,
2599 .free_slot = xprt_free_slot,
2600 .rpcbind = xs_local_rpcbind,
2601 .set_port = xs_local_set_port,
2602 .connect = xs_local_connect,
2603 .buf_alloc = rpc_malloc,
2604 .buf_free = rpc_free,
2605 .prepare_request = xs_stream_prepare_request,
2606 .send_request = xs_local_send_request,
2607 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2609 .destroy = xs_destroy,
2610 .print_stats = xs_local_print_stats,
2611 .enable_swap = xs_enable_swap,
2612 .disable_swap = xs_disable_swap,
2615 static const struct rpc_xprt_ops xs_udp_ops = {
2616 .set_buffer_size = xs_udp_set_buffer_size,
2617 .reserve_xprt = xprt_reserve_xprt_cong,
2618 .release_xprt = xprt_release_xprt_cong,
2619 .alloc_slot = xprt_alloc_slot,
2620 .free_slot = xprt_free_slot,
2621 .rpcbind = rpcb_getport_async,
2622 .set_port = xs_set_port,
2623 .connect = xs_connect,
2624 .buf_alloc = rpc_malloc,
2625 .buf_free = rpc_free,
2626 .send_request = xs_udp_send_request,
2627 .wait_for_reply_request = xprt_wait_for_reply_request_rtt,
2628 .timer = xs_udp_timer,
2629 .release_request = xprt_release_rqst_cong,
2631 .destroy = xs_destroy,
2632 .print_stats = xs_udp_print_stats,
2633 .enable_swap = xs_enable_swap,
2634 .disable_swap = xs_disable_swap,
2635 .inject_disconnect = xs_inject_disconnect,
2638 static const struct rpc_xprt_ops xs_tcp_ops = {
2639 .reserve_xprt = xprt_reserve_xprt,
2640 .release_xprt = xprt_release_xprt,
2641 .alloc_slot = xprt_alloc_slot,
2642 .free_slot = xprt_free_slot,
2643 .rpcbind = rpcb_getport_async,
2644 .set_port = xs_set_port,
2645 .connect = xs_connect,
2646 .buf_alloc = rpc_malloc,
2647 .buf_free = rpc_free,
2648 .prepare_request = xs_stream_prepare_request,
2649 .send_request = xs_tcp_send_request,
2650 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2651 .close = xs_tcp_shutdown,
2652 .destroy = xs_destroy,
2653 .set_connect_timeout = xs_tcp_set_connect_timeout,
2654 .print_stats = xs_tcp_print_stats,
2655 .enable_swap = xs_enable_swap,
2656 .disable_swap = xs_disable_swap,
2657 .inject_disconnect = xs_inject_disconnect,
2658 #ifdef CONFIG_SUNRPC_BACKCHANNEL
2659 .bc_setup = xprt_setup_bc,
2660 .bc_maxpayload = xs_tcp_bc_maxpayload,
2661 .bc_num_slots = xprt_bc_max_slots,
2662 .bc_free_rqst = xprt_free_bc_rqst,
2663 .bc_destroy = xprt_destroy_bc,
2668 * The rpc_xprt_ops for the server backchannel
2671 static const struct rpc_xprt_ops bc_tcp_ops = {
2672 .reserve_xprt = xprt_reserve_xprt,
2673 .release_xprt = xprt_release_xprt,
2674 .alloc_slot = xprt_alloc_slot,
2675 .free_slot = xprt_free_slot,
2676 .buf_alloc = bc_malloc,
2677 .buf_free = bc_free,
2678 .send_request = bc_send_request,
2679 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2681 .destroy = bc_destroy,
2682 .print_stats = xs_tcp_print_stats,
2683 .enable_swap = xs_enable_swap,
2684 .disable_swap = xs_disable_swap,
2685 .inject_disconnect = xs_inject_disconnect,
2688 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2690 static const struct sockaddr_in sin = {
2691 .sin_family = AF_INET,
2692 .sin_addr.s_addr = htonl(INADDR_ANY),
2694 static const struct sockaddr_in6 sin6 = {
2695 .sin6_family = AF_INET6,
2696 .sin6_addr = IN6ADDR_ANY_INIT,
2703 memcpy(sap, &sin, sizeof(sin));
2706 memcpy(sap, &sin6, sizeof(sin6));
2709 dprintk("RPC: %s: Bad address family\n", __func__);
2710 return -EAFNOSUPPORT;
2715 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2716 unsigned int slot_table_size,
2717 unsigned int max_slot_table_size)
2719 struct rpc_xprt *xprt;
2720 struct sock_xprt *new;
2722 if (args->addrlen > sizeof(xprt->addr)) {
2723 dprintk("RPC: xs_setup_xprt: address too large\n");
2724 return ERR_PTR(-EBADF);
2727 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2728 max_slot_table_size);
2730 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2732 return ERR_PTR(-ENOMEM);
2735 new = container_of(xprt, struct sock_xprt, xprt);
2736 mutex_init(&new->recv_mutex);
2737 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2738 xprt->addrlen = args->addrlen;
2740 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2743 err = xs_init_anyaddr(args->dstaddr->sa_family,
2744 (struct sockaddr *)&new->srcaddr);
2747 return ERR_PTR(err);
2754 static const struct rpc_timeout xs_local_default_timeout = {
2755 .to_initval = 10 * HZ,
2756 .to_maxval = 10 * HZ,
2761 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2762 * @args: rpc transport creation arguments
2764 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2766 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2768 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2769 struct sock_xprt *transport;
2770 struct rpc_xprt *xprt;
2771 struct rpc_xprt *ret;
2773 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2774 xprt_max_tcp_slot_table_entries);
2777 transport = container_of(xprt, struct sock_xprt, xprt);
2780 xprt->xprt_class = &xs_local_transport;
2781 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2783 xprt->bind_timeout = XS_BIND_TO;
2784 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2785 xprt->idle_timeout = XS_IDLE_DISC_TO;
2787 xprt->ops = &xs_local_ops;
2788 xprt->timeout = &xs_local_default_timeout;
2790 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2791 INIT_WORK(&transport->error_worker, xs_error_handle);
2792 INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket);
2794 switch (sun->sun_family) {
2796 if (sun->sun_path[0] != '/') {
2797 dprintk("RPC: bad AF_LOCAL address: %s\n",
2799 ret = ERR_PTR(-EINVAL);
2802 xprt_set_bound(xprt);
2803 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2804 ret = ERR_PTR(xs_local_setup_socket(transport));
2809 ret = ERR_PTR(-EAFNOSUPPORT);
2813 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2814 xprt->address_strings[RPC_DISPLAY_ADDR]);
2816 if (try_module_get(THIS_MODULE))
2818 ret = ERR_PTR(-EINVAL);
2824 static const struct rpc_timeout xs_udp_default_timeout = {
2825 .to_initval = 5 * HZ,
2826 .to_maxval = 30 * HZ,
2827 .to_increment = 5 * HZ,
2832 * xs_setup_udp - Set up transport to use a UDP socket
2833 * @args: rpc transport creation arguments
2836 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2838 struct sockaddr *addr = args->dstaddr;
2839 struct rpc_xprt *xprt;
2840 struct sock_xprt *transport;
2841 struct rpc_xprt *ret;
2843 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2844 xprt_udp_slot_table_entries);
2847 transport = container_of(xprt, struct sock_xprt, xprt);
2849 xprt->prot = IPPROTO_UDP;
2850 xprt->xprt_class = &xs_udp_transport;
2851 /* XXX: header size can vary due to auth type, IPv6, etc. */
2852 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2854 xprt->bind_timeout = XS_BIND_TO;
2855 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2856 xprt->idle_timeout = XS_IDLE_DISC_TO;
2858 xprt->ops = &xs_udp_ops;
2860 xprt->timeout = &xs_udp_default_timeout;
2862 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
2863 INIT_WORK(&transport->error_worker, xs_error_handle);
2864 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2866 switch (addr->sa_family) {
2868 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2869 xprt_set_bound(xprt);
2871 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2874 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2875 xprt_set_bound(xprt);
2877 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2880 ret = ERR_PTR(-EAFNOSUPPORT);
2884 if (xprt_bound(xprt))
2885 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2886 xprt->address_strings[RPC_DISPLAY_ADDR],
2887 xprt->address_strings[RPC_DISPLAY_PORT],
2888 xprt->address_strings[RPC_DISPLAY_PROTO]);
2890 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2891 xprt->address_strings[RPC_DISPLAY_ADDR],
2892 xprt->address_strings[RPC_DISPLAY_PROTO]);
2894 if (try_module_get(THIS_MODULE))
2896 ret = ERR_PTR(-EINVAL);
2902 static const struct rpc_timeout xs_tcp_default_timeout = {
2903 .to_initval = 60 * HZ,
2904 .to_maxval = 60 * HZ,
2909 * xs_setup_tcp - Set up transport to use a TCP socket
2910 * @args: rpc transport creation arguments
2913 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2915 struct sockaddr *addr = args->dstaddr;
2916 struct rpc_xprt *xprt;
2917 struct sock_xprt *transport;
2918 struct rpc_xprt *ret;
2919 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2921 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2922 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2924 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2925 max_slot_table_size);
2928 transport = container_of(xprt, struct sock_xprt, xprt);
2930 xprt->prot = IPPROTO_TCP;
2931 xprt->xprt_class = &xs_tcp_transport;
2932 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2934 xprt->bind_timeout = XS_BIND_TO;
2935 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2936 xprt->idle_timeout = XS_IDLE_DISC_TO;
2938 xprt->ops = &xs_tcp_ops;
2939 xprt->timeout = &xs_tcp_default_timeout;
2941 xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
2942 xprt->connect_timeout = xprt->timeout->to_initval *
2943 (xprt->timeout->to_retries + 1);
2945 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2946 INIT_WORK(&transport->error_worker, xs_error_handle);
2947 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
2949 switch (addr->sa_family) {
2951 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2952 xprt_set_bound(xprt);
2954 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2957 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2958 xprt_set_bound(xprt);
2960 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2963 ret = ERR_PTR(-EAFNOSUPPORT);
2967 if (xprt_bound(xprt))
2968 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2969 xprt->address_strings[RPC_DISPLAY_ADDR],
2970 xprt->address_strings[RPC_DISPLAY_PORT],
2971 xprt->address_strings[RPC_DISPLAY_PROTO]);
2973 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2974 xprt->address_strings[RPC_DISPLAY_ADDR],
2975 xprt->address_strings[RPC_DISPLAY_PROTO]);
2977 if (try_module_get(THIS_MODULE))
2979 ret = ERR_PTR(-EINVAL);
2986 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2987 * @args: rpc transport creation arguments
2990 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2992 struct sockaddr *addr = args->dstaddr;
2993 struct rpc_xprt *xprt;
2994 struct sock_xprt *transport;
2995 struct svc_sock *bc_sock;
2996 struct rpc_xprt *ret;
2998 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2999 xprt_tcp_slot_table_entries);
3002 transport = container_of(xprt, struct sock_xprt, xprt);
3004 xprt->prot = IPPROTO_TCP;
3005 xprt->xprt_class = &xs_bc_tcp_transport;
3006 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3007 xprt->timeout = &xs_tcp_default_timeout;
3010 xprt_set_bound(xprt);
3011 xprt->bind_timeout = 0;
3012 xprt->reestablish_timeout = 0;
3013 xprt->idle_timeout = 0;
3015 xprt->ops = &bc_tcp_ops;
3017 switch (addr->sa_family) {
3019 xs_format_peer_addresses(xprt, "tcp",
3023 xs_format_peer_addresses(xprt, "tcp",
3024 RPCBIND_NETID_TCP6);
3027 ret = ERR_PTR(-EAFNOSUPPORT);
3031 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3032 xprt->address_strings[RPC_DISPLAY_ADDR],
3033 xprt->address_strings[RPC_DISPLAY_PORT],
3034 xprt->address_strings[RPC_DISPLAY_PROTO]);
3037 * Once we've associated a backchannel xprt with a connection,
3038 * we want to keep it around as long as the connection lasts,
3039 * in case we need to start using it for a backchannel again;
3040 * this reference won't be dropped until bc_xprt is destroyed.
3043 args->bc_xprt->xpt_bc_xprt = xprt;
3044 xprt->bc_xprt = args->bc_xprt;
3045 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3046 transport->sock = bc_sock->sk_sock;
3047 transport->inet = bc_sock->sk_sk;
3050 * Since we don't want connections for the backchannel, we set
3051 * the xprt status to connected
3053 xprt_set_connected(xprt);
3055 if (try_module_get(THIS_MODULE))
3058 args->bc_xprt->xpt_bc_xprt = NULL;
3059 args->bc_xprt->xpt_bc_xps = NULL;
3061 ret = ERR_PTR(-EINVAL);
3067 static struct xprt_class xs_local_transport = {
3068 .list = LIST_HEAD_INIT(xs_local_transport.list),
3069 .name = "named UNIX socket",
3070 .owner = THIS_MODULE,
3071 .ident = XPRT_TRANSPORT_LOCAL,
3072 .setup = xs_setup_local,
3076 static struct xprt_class xs_udp_transport = {
3077 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3079 .owner = THIS_MODULE,
3080 .ident = XPRT_TRANSPORT_UDP,
3081 .setup = xs_setup_udp,
3082 .netid = { "udp", "udp6", "" },
3085 static struct xprt_class xs_tcp_transport = {
3086 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3088 .owner = THIS_MODULE,
3089 .ident = XPRT_TRANSPORT_TCP,
3090 .setup = xs_setup_tcp,
3091 .netid = { "tcp", "tcp6", "" },
3094 static struct xprt_class xs_bc_tcp_transport = {
3095 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3096 .name = "tcp NFSv4.1 backchannel",
3097 .owner = THIS_MODULE,
3098 .ident = XPRT_TRANSPORT_BC_TCP,
3099 .setup = xs_setup_bc_tcp,
3104 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3107 int init_socket_xprt(void)
3109 if (!sunrpc_table_header)
3110 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3112 xprt_register_transport(&xs_local_transport);
3113 xprt_register_transport(&xs_udp_transport);
3114 xprt_register_transport(&xs_tcp_transport);
3115 xprt_register_transport(&xs_bc_tcp_transport);
3121 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3124 void cleanup_socket_xprt(void)
3126 if (sunrpc_table_header) {
3127 unregister_sysctl_table(sunrpc_table_header);
3128 sunrpc_table_header = NULL;
3131 xprt_unregister_transport(&xs_local_transport);
3132 xprt_unregister_transport(&xs_udp_transport);
3133 xprt_unregister_transport(&xs_tcp_transport);
3134 xprt_unregister_transport(&xs_bc_tcp_transport);
3137 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3139 return param_set_uint_minmax(val, kp,
3144 static const struct kernel_param_ops param_ops_portnr = {
3145 .set = param_set_portnr,
3146 .get = param_get_uint,
3149 #define param_check_portnr(name, p) \
3150 __param_check(name, p, unsigned int);
3152 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3153 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3155 static int param_set_slot_table_size(const char *val,
3156 const struct kernel_param *kp)
3158 return param_set_uint_minmax(val, kp,
3160 RPC_MAX_SLOT_TABLE);
3163 static const struct kernel_param_ops param_ops_slot_table_size = {
3164 .set = param_set_slot_table_size,
3165 .get = param_get_uint,
3168 #define param_check_slot_table_size(name, p) \
3169 __param_check(name, p, unsigned int);
3171 static int param_set_max_slot_table_size(const char *val,
3172 const struct kernel_param *kp)
3174 return param_set_uint_minmax(val, kp,
3176 RPC_MAX_SLOT_TABLE_LIMIT);
3179 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3180 .set = param_set_max_slot_table_size,
3181 .get = param_get_uint,
3184 #define param_check_max_slot_table_size(name, p) \
3185 __param_check(name, p, unsigned int);
3187 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3188 slot_table_size, 0644);
3189 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3190 max_slot_table_size, 0644);
3191 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3192 slot_table_size, 0644);