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 <net/tls_prot.h>
51 #include <net/handshake.h>
53 #include <linux/bvec.h>
54 #include <linux/highmem.h>
55 #include <linux/uio.h>
56 #include <linux/sched/mm.h>
58 #include <trace/events/sock.h>
59 #include <trace/events/sunrpc.h>
64 static void xs_close(struct rpc_xprt *xprt);
65 static void xs_reset_srcport(struct sock_xprt *transport);
66 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock);
67 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
73 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
74 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
75 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
77 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
78 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
80 #define XS_TCP_LINGER_TO (15U * HZ)
81 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
84 * We can register our own files under /proc/sys/sunrpc by
85 * calling register_sysctl() again. The files in that
86 * directory become the union of all files registered there.
88 * We simply need to make sure that we don't collide with
89 * someone else's file names!
92 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
93 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
94 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
95 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
96 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
98 static struct ctl_table_header *sunrpc_table_header;
100 static struct xprt_class xs_local_transport;
101 static struct xprt_class xs_udp_transport;
102 static struct xprt_class xs_tcp_transport;
103 static struct xprt_class xs_tcp_tls_transport;
104 static struct xprt_class xs_bc_tcp_transport;
107 * FIXME: changing the UDP slot table size should also resize the UDP
108 * socket buffers for existing UDP transports
110 static struct ctl_table xs_tunables_table[] = {
112 .procname = "udp_slot_table_entries",
113 .data = &xprt_udp_slot_table_entries,
114 .maxlen = sizeof(unsigned int),
116 .proc_handler = proc_dointvec_minmax,
117 .extra1 = &min_slot_table_size,
118 .extra2 = &max_slot_table_size
121 .procname = "tcp_slot_table_entries",
122 .data = &xprt_tcp_slot_table_entries,
123 .maxlen = sizeof(unsigned int),
125 .proc_handler = proc_dointvec_minmax,
126 .extra1 = &min_slot_table_size,
127 .extra2 = &max_slot_table_size
130 .procname = "tcp_max_slot_table_entries",
131 .data = &xprt_max_tcp_slot_table_entries,
132 .maxlen = sizeof(unsigned int),
134 .proc_handler = proc_dointvec_minmax,
135 .extra1 = &min_slot_table_size,
136 .extra2 = &max_tcp_slot_table_limit
139 .procname = "min_resvport",
140 .data = &xprt_min_resvport,
141 .maxlen = sizeof(unsigned int),
143 .proc_handler = proc_dointvec_minmax,
144 .extra1 = &xprt_min_resvport_limit,
145 .extra2 = &xprt_max_resvport_limit
148 .procname = "max_resvport",
149 .data = &xprt_max_resvport,
150 .maxlen = sizeof(unsigned int),
152 .proc_handler = proc_dointvec_minmax,
153 .extra1 = &xprt_min_resvport_limit,
154 .extra2 = &xprt_max_resvport_limit
157 .procname = "tcp_fin_timeout",
158 .data = &xs_tcp_fin_timeout,
159 .maxlen = sizeof(xs_tcp_fin_timeout),
161 .proc_handler = proc_dointvec_jiffies,
167 * Wait duration for a reply from the RPC portmapper.
169 #define XS_BIND_TO (60U * HZ)
172 * Delay if a UDP socket connect error occurs. This is most likely some
173 * kind of resource problem on the local host.
175 #define XS_UDP_REEST_TO (2U * HZ)
178 * The reestablish timeout allows clients to delay for a bit before attempting
179 * to reconnect to a server that just dropped our connection.
181 * We implement an exponential backoff when trying to reestablish a TCP
182 * transport connection with the server. Some servers like to drop a TCP
183 * connection when they are overworked, so we start with a short timeout and
184 * increase over time if the server is down or not responding.
186 #define XS_TCP_INIT_REEST_TO (3U * HZ)
189 * TCP idle timeout; client drops the transport socket if it is idle
190 * for this long. Note that we also timeout UDP sockets to prevent
191 * holding port numbers when there is no RPC traffic.
193 #define XS_IDLE_DISC_TO (5U * 60 * HZ)
196 * TLS handshake timeout.
198 #define XS_TLS_HANDSHAKE_TO (10U * HZ)
200 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
201 # undef RPC_DEBUG_DATA
202 # define RPCDBG_FACILITY RPCDBG_TRANS
205 #ifdef RPC_DEBUG_DATA
206 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
208 u8 *buf = (u8 *) packet;
211 dprintk("RPC: %s\n", msg);
212 for (j = 0; j < count && j < 128; j += 4) {
216 dprintk("0x%04x ", j);
218 dprintk("%02x%02x%02x%02x ",
219 buf[j], buf[j+1], buf[j+2], buf[j+3]);
224 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
230 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
232 return (struct rpc_xprt *) sk->sk_user_data;
235 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
237 return (struct sockaddr *) &xprt->addr;
240 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
242 return (struct sockaddr_un *) &xprt->addr;
245 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
247 return (struct sockaddr_in *) &xprt->addr;
250 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
252 return (struct sockaddr_in6 *) &xprt->addr;
255 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
257 struct sockaddr *sap = xs_addr(xprt);
258 struct sockaddr_in6 *sin6;
259 struct sockaddr_in *sin;
260 struct sockaddr_un *sun;
263 switch (sap->sa_family) {
265 sun = xs_addr_un(xprt);
266 if (sun->sun_path[0]) {
267 strscpy(buf, sun->sun_path, sizeof(buf));
270 strscpy(buf+1, sun->sun_path+1, sizeof(buf)-1);
272 xprt->address_strings[RPC_DISPLAY_ADDR] =
273 kstrdup(buf, GFP_KERNEL);
276 (void)rpc_ntop(sap, buf, sizeof(buf));
277 xprt->address_strings[RPC_DISPLAY_ADDR] =
278 kstrdup(buf, GFP_KERNEL);
279 sin = xs_addr_in(xprt);
280 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
283 (void)rpc_ntop(sap, buf, sizeof(buf));
284 xprt->address_strings[RPC_DISPLAY_ADDR] =
285 kstrdup(buf, GFP_KERNEL);
286 sin6 = xs_addr_in6(xprt);
287 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
293 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
296 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
298 struct sockaddr *sap = xs_addr(xprt);
301 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
302 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
304 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
305 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
308 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
309 const char *protocol,
312 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
313 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
314 xs_format_common_peer_addresses(xprt);
315 xs_format_common_peer_ports(xprt);
318 static void xs_update_peer_port(struct rpc_xprt *xprt)
320 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
321 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
323 xs_format_common_peer_ports(xprt);
326 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
330 for (i = 0; i < RPC_DISPLAY_MAX; i++)
332 case RPC_DISPLAY_PROTO:
333 case RPC_DISPLAY_NETID:
336 kfree(xprt->address_strings[i]);
341 xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp)
345 if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES))
347 n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT;
348 for (i = 0; i < n; i++) {
351 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp);
352 if (!buf->pages[i]) {
354 return i > buf->page_base ? i - buf->page_base : 0;
361 xs_sock_process_cmsg(struct socket *sock, struct msghdr *msg,
362 struct cmsghdr *cmsg, int ret)
364 u8 content_type = tls_get_record_type(sock->sk, cmsg);
365 u8 level, description;
367 switch (content_type) {
370 case TLS_RECORD_TYPE_DATA:
371 /* TLS sets EOR at the end of each application data
372 * record, even though there might be more frames
373 * waiting to be decrypted.
375 msg->msg_flags &= ~MSG_EOR;
377 case TLS_RECORD_TYPE_ALERT:
378 tls_alert_recv(sock->sk, msg, &level, &description);
379 ret = (level == TLS_ALERT_LEVEL_FATAL) ?
383 /* discard this record type */
390 xs_sock_recv_cmsg(struct socket *sock, struct msghdr *msg, int flags)
394 u8 buf[CMSG_SPACE(sizeof(u8))];
398 msg->msg_control = &u;
399 msg->msg_controllen = sizeof(u);
400 ret = sock_recvmsg(sock, msg, flags);
401 if (msg->msg_controllen != sizeof(u))
402 ret = xs_sock_process_cmsg(sock, msg, &u.cmsg, ret);
407 xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek)
411 iov_iter_advance(&msg->msg_iter, seek);
412 ret = xs_sock_recv_cmsg(sock, msg, flags);
413 return ret > 0 ? ret + seek : ret;
417 xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags,
418 struct kvec *kvec, size_t count, size_t seek)
420 iov_iter_kvec(&msg->msg_iter, ITER_DEST, kvec, 1, count);
421 return xs_sock_recvmsg(sock, msg, flags, seek);
425 xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags,
426 struct bio_vec *bvec, unsigned long nr, size_t count,
429 iov_iter_bvec(&msg->msg_iter, ITER_DEST, bvec, nr, count);
430 return xs_sock_recvmsg(sock, msg, flags, seek);
434 xs_read_discard(struct socket *sock, struct msghdr *msg, int flags,
437 iov_iter_discard(&msg->msg_iter, ITER_DEST, count);
438 return xs_sock_recv_cmsg(sock, msg, flags);
441 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
443 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
445 struct bvec_iter bi = {
450 bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
451 for_each_bvec(bv, bvec, bi, bi)
452 flush_dcache_page(bv.bv_page);
456 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
462 xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
463 struct xdr_buf *buf, size_t count, size_t seek, size_t *read)
465 size_t want, seek_init = seek, offset = 0;
468 want = min_t(size_t, count, buf->head[0].iov_len);
470 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek);
474 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
484 want = xs_alloc_sparse_pages(
485 buf, min_t(size_t, count - offset, buf->page_len),
486 GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN);
488 ret = xs_read_bvec(sock, msg, flags, buf->bvec,
489 xdr_buf_pagecount(buf),
490 want + buf->page_base,
491 seek + buf->page_base);
494 xs_flush_bvec(buf->bvec, ret, seek + buf->page_base);
495 ret -= buf->page_base;
497 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
507 want = min_t(size_t, count - offset, buf->tail[0].iov_len);
509 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek);
513 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
517 } else if (offset < seek_init)
521 *read = offset - seek_init;
529 xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf)
531 if (!transport->recv.copied) {
532 if (buf->head[0].iov_len >= transport->recv.offset)
533 memcpy(buf->head[0].iov_base,
534 &transport->recv.xid,
535 transport->recv.offset);
536 transport->recv.copied = transport->recv.offset;
541 xs_read_stream_request_done(struct sock_xprt *transport)
543 return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT);
547 xs_read_stream_check_eor(struct sock_xprt *transport,
550 if (xs_read_stream_request_done(transport))
551 msg->msg_flags |= MSG_EOR;
555 xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg,
556 int flags, struct rpc_rqst *req)
558 struct xdr_buf *buf = &req->rq_private_buf;
562 xs_read_header(transport, buf);
564 want = transport->recv.len - transport->recv.offset;
566 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf,
567 transport->recv.copied + want,
568 transport->recv.copied,
570 transport->recv.offset += read;
571 transport->recv.copied += read;
574 if (transport->recv.offset == transport->recv.len)
575 xs_read_stream_check_eor(transport, msg);
585 msg->msg_flags |= MSG_TRUNC;
590 return ret < 0 ? ret : read;
594 xs_read_stream_headersize(bool isfrag)
597 return sizeof(__be32);
598 return 3 * sizeof(__be32);
602 xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg,
603 int flags, size_t want, size_t seek)
606 .iov_base = &transport->recv.fraghdr,
609 return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek);
612 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
614 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
616 struct rpc_xprt *xprt = &transport->xprt;
617 struct rpc_rqst *req;
620 /* Is this transport associated with the backchannel? */
624 /* Look up and lock the request corresponding to the given XID */
625 req = xprt_lookup_bc_request(xprt, transport->recv.xid);
627 printk(KERN_WARNING "Callback slot table overflowed\n");
630 if (transport->recv.copied && !req->rq_private_buf.len)
633 ret = xs_read_stream_request(transport, msg, flags, req);
634 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
635 xprt_complete_bc_request(req, transport->recv.copied);
637 req->rq_private_buf.len = transport->recv.copied;
641 #else /* CONFIG_SUNRPC_BACKCHANNEL */
643 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
647 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
650 xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags)
652 struct rpc_xprt *xprt = &transport->xprt;
653 struct rpc_rqst *req;
656 /* Look up and lock the request corresponding to the given XID */
657 spin_lock(&xprt->queue_lock);
658 req = xprt_lookup_rqst(xprt, transport->recv.xid);
659 if (!req || (transport->recv.copied && !req->rq_private_buf.len)) {
660 msg->msg_flags |= MSG_TRUNC;
664 spin_unlock(&xprt->queue_lock);
666 ret = xs_read_stream_request(transport, msg, flags, req);
668 spin_lock(&xprt->queue_lock);
669 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
670 xprt_complete_rqst(req->rq_task, transport->recv.copied);
672 req->rq_private_buf.len = transport->recv.copied;
673 xprt_unpin_rqst(req);
675 spin_unlock(&xprt->queue_lock);
680 xs_read_stream(struct sock_xprt *transport, int flags)
682 struct msghdr msg = { 0 };
683 size_t want, read = 0;
686 if (transport->recv.len == 0) {
687 want = xs_read_stream_headersize(transport->recv.copied != 0);
688 ret = xs_read_stream_header(transport, &msg, flags, want,
689 transport->recv.offset);
692 transport->recv.offset = ret;
693 if (transport->recv.offset != want)
694 return transport->recv.offset;
695 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) &
696 RPC_FRAGMENT_SIZE_MASK;
697 transport->recv.offset -= sizeof(transport->recv.fraghdr);
701 switch (be32_to_cpu(transport->recv.calldir)) {
703 msg.msg_flags |= MSG_TRUNC;
706 ret = xs_read_stream_call(transport, &msg, flags);
709 ret = xs_read_stream_reply(transport, &msg, flags);
711 if (msg.msg_flags & MSG_TRUNC) {
712 transport->recv.calldir = cpu_to_be32(-1);
713 transport->recv.copied = -1;
718 if (transport->recv.offset < transport->recv.len) {
719 if (!(msg.msg_flags & MSG_TRUNC))
722 ret = xs_read_discard(transport->sock, &msg, flags,
723 transport->recv.len - transport->recv.offset);
726 transport->recv.offset += ret;
728 if (transport->recv.offset != transport->recv.len)
731 if (xs_read_stream_request_done(transport)) {
732 trace_xs_stream_read_request(transport);
733 transport->recv.copied = 0;
735 transport->recv.offset = 0;
736 transport->recv.len = 0;
739 return ret != 0 ? ret : -ESHUTDOWN;
742 static __poll_t xs_poll_socket(struct sock_xprt *transport)
744 return transport->sock->ops->poll(transport->file, transport->sock,
748 static bool xs_poll_socket_readable(struct sock_xprt *transport)
750 __poll_t events = xs_poll_socket(transport);
752 return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP);
755 static void xs_poll_check_readable(struct sock_xprt *transport)
758 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
759 if (test_bit(XPRT_SOCK_IGNORE_RECV, &transport->sock_state))
761 if (!xs_poll_socket_readable(transport))
763 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
764 queue_work(xprtiod_workqueue, &transport->recv_worker);
767 static void xs_stream_data_receive(struct sock_xprt *transport)
772 mutex_lock(&transport->recv_mutex);
773 if (transport->sock == NULL)
776 ret = xs_read_stream(transport, MSG_DONTWAIT);
782 if (ret == -ESHUTDOWN)
783 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
784 else if (ret == -EACCES)
785 xprt_wake_pending_tasks(&transport->xprt, -EACCES);
787 xs_poll_check_readable(transport);
789 mutex_unlock(&transport->recv_mutex);
790 trace_xs_stream_read_data(&transport->xprt, ret, read);
793 static void xs_stream_data_receive_workfn(struct work_struct *work)
795 struct sock_xprt *transport =
796 container_of(work, struct sock_xprt, recv_worker);
797 unsigned int pflags = memalloc_nofs_save();
799 xs_stream_data_receive(transport);
800 memalloc_nofs_restore(pflags);
804 xs_stream_reset_connect(struct sock_xprt *transport)
806 transport->recv.offset = 0;
807 transport->recv.len = 0;
808 transport->recv.copied = 0;
809 transport->xmit.offset = 0;
813 xs_stream_start_connect(struct sock_xprt *transport)
815 transport->xprt.stat.connect_count++;
816 transport->xprt.stat.connect_start = jiffies;
819 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
822 * xs_nospace - handle transmit was incomplete
823 * @req: pointer to RPC request
824 * @transport: pointer to struct sock_xprt
827 static int xs_nospace(struct rpc_rqst *req, struct sock_xprt *transport)
829 struct rpc_xprt *xprt = &transport->xprt;
830 struct sock *sk = transport->inet;
833 trace_rpc_socket_nospace(req, transport);
835 /* Protect against races with write_space */
836 spin_lock(&xprt->transport_lock);
838 /* Don't race with disconnect */
839 if (xprt_connected(xprt)) {
840 /* wait for more buffer space */
841 set_bit(XPRT_SOCK_NOSPACE, &transport->sock_state);
842 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
843 sk->sk_write_pending++;
844 xprt_wait_for_buffer_space(xprt);
848 spin_unlock(&xprt->transport_lock);
852 static int xs_sock_nospace(struct rpc_rqst *req)
854 struct sock_xprt *transport =
855 container_of(req->rq_xprt, struct sock_xprt, xprt);
856 struct sock *sk = transport->inet;
860 if (!sock_writeable(sk))
861 ret = xs_nospace(req, transport);
866 static int xs_stream_nospace(struct rpc_rqst *req, bool vm_wait)
868 struct sock_xprt *transport =
869 container_of(req->rq_xprt, struct sock_xprt, xprt);
870 struct sock *sk = transport->inet;
876 if (!sk_stream_memory_free(sk))
877 ret = xs_nospace(req, transport);
882 static int xs_stream_prepare_request(struct rpc_rqst *req, struct xdr_buf *buf)
884 return xdr_alloc_bvec(buf, rpc_task_gfp_mask());
887 static void xs_stream_abort_send_request(struct rpc_rqst *req)
889 struct rpc_xprt *xprt = req->rq_xprt;
890 struct sock_xprt *transport =
891 container_of(xprt, struct sock_xprt, xprt);
893 if (transport->xmit.offset != 0 &&
894 !test_bit(XPRT_CLOSE_WAIT, &xprt->state))
895 xprt_force_disconnect(xprt);
899 * Determine if the previous message in the stream was aborted before it
900 * could complete transmission.
903 xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req)
905 return transport->xmit.offset != 0 && req->rq_bytes_sent == 0;
909 * Return the stream record marker field for a record of length < 2^31-1
912 xs_stream_record_marker(struct xdr_buf *xdr)
916 return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len);
920 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
921 * @req: pointer to RPC request
924 * 0: The request has been sent
925 * EAGAIN: The socket was blocked, please call again later to
926 * complete the request
927 * ENOTCONN: Caller needs to invoke connect logic then call again
928 * other: Some other error occurred, the request was not sent
930 static int xs_local_send_request(struct rpc_rqst *req)
932 struct rpc_xprt *xprt = req->rq_xprt;
933 struct sock_xprt *transport =
934 container_of(xprt, struct sock_xprt, xprt);
935 struct xdr_buf *xdr = &req->rq_snd_buf;
936 rpc_fraghdr rm = xs_stream_record_marker(xdr);
937 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
938 struct msghdr msg = {
939 .msg_flags = XS_SENDMSG_FLAGS,
945 /* Close the stream if the previous transmission was incomplete */
946 if (xs_send_request_was_aborted(transport, req)) {
947 xprt_force_disconnect(xprt);
951 xs_pktdump("packet data:",
952 req->rq_svec->iov_base, req->rq_svec->iov_len);
954 vm_wait = sk_stream_is_writeable(transport->inet) ? true : false;
956 req->rq_xtime = ktime_get();
957 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
958 transport->xmit.offset, rm, &sent);
959 dprintk("RPC: %s(%u) = %d\n",
960 __func__, xdr->len - transport->xmit.offset, status);
962 if (likely(sent > 0) || status == 0) {
963 transport->xmit.offset += sent;
964 req->rq_bytes_sent = transport->xmit.offset;
965 if (likely(req->rq_bytes_sent >= msglen)) {
966 req->rq_xmit_bytes_sent += transport->xmit.offset;
967 transport->xmit.offset = 0;
976 status = xs_stream_nospace(req, vm_wait);
979 dprintk("RPC: sendmsg returned unrecognized error %d\n",
983 xprt_force_disconnect(xprt);
991 * xs_udp_send_request - write an RPC request to a UDP socket
992 * @req: pointer to RPC request
995 * 0: The request has been sent
996 * EAGAIN: The socket was blocked, please call again later to
997 * complete the request
998 * ENOTCONN: Caller needs to invoke connect logic then call again
999 * other: Some other error occurred, the request was not sent
1001 static int xs_udp_send_request(struct rpc_rqst *req)
1003 struct rpc_xprt *xprt = req->rq_xprt;
1004 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1005 struct xdr_buf *xdr = &req->rq_snd_buf;
1006 struct msghdr msg = {
1007 .msg_name = xs_addr(xprt),
1008 .msg_namelen = xprt->addrlen,
1009 .msg_flags = XS_SENDMSG_FLAGS,
1014 xs_pktdump("packet data:",
1015 req->rq_svec->iov_base,
1016 req->rq_svec->iov_len);
1018 if (!xprt_bound(xprt))
1021 if (!xprt_request_get_cong(xprt, req))
1024 status = xdr_alloc_bvec(xdr, rpc_task_gfp_mask());
1027 req->rq_xtime = ktime_get();
1028 status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, 0, &sent);
1030 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
1033 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
1034 if (status == -EPERM)
1035 goto process_status;
1037 if (status == -EAGAIN && sock_writeable(transport->inet))
1040 if (sent > 0 || status == 0) {
1041 req->rq_xmit_bytes_sent += sent;
1042 if (sent >= req->rq_slen)
1044 /* Still some bytes left; set up for a retry later. */
1052 /* Should we call xs_close() here? */
1055 status = xs_sock_nospace(req);
1062 /* When the server has died, an ICMP port unreachable message
1063 * prompts ECONNREFUSED. */
1066 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1074 * xs_tcp_send_request - write an RPC request to a TCP socket
1075 * @req: pointer to RPC request
1078 * 0: The request has been sent
1079 * EAGAIN: The socket was blocked, please call again later to
1080 * complete the request
1081 * ENOTCONN: Caller needs to invoke connect logic then call again
1082 * other: Some other error occurred, the request was not sent
1084 * XXX: In the case of soft timeouts, should we eventually give up
1085 * if sendmsg is not able to make progress?
1087 static int xs_tcp_send_request(struct rpc_rqst *req)
1089 struct rpc_xprt *xprt = req->rq_xprt;
1090 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1091 struct xdr_buf *xdr = &req->rq_snd_buf;
1092 rpc_fraghdr rm = xs_stream_record_marker(xdr);
1093 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
1094 struct msghdr msg = {
1095 .msg_flags = XS_SENDMSG_FLAGS,
1101 /* Close the stream if the previous transmission was incomplete */
1102 if (xs_send_request_was_aborted(transport, req)) {
1103 if (transport->sock != NULL)
1104 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
1107 if (!transport->inet)
1110 xs_pktdump("packet data:",
1111 req->rq_svec->iov_base,
1112 req->rq_svec->iov_len);
1114 if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
1115 xs_tcp_set_socket_timeouts(xprt, transport->sock);
1117 xs_set_srcport(transport, transport->sock);
1119 /* Continue transmitting the packet/record. We must be careful
1120 * to cope with writespace callbacks arriving _after_ we have
1121 * called sendmsg(). */
1122 req->rq_xtime = ktime_get();
1123 tcp_sock_set_cork(transport->inet, true);
1125 vm_wait = sk_stream_is_writeable(transport->inet) ? true : false;
1128 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
1129 transport->xmit.offset, rm, &sent);
1131 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
1132 xdr->len - transport->xmit.offset, status);
1134 /* If we've sent the entire packet, immediately
1135 * reset the count of bytes sent. */
1136 transport->xmit.offset += sent;
1137 req->rq_bytes_sent = transport->xmit.offset;
1138 if (likely(req->rq_bytes_sent >= msglen)) {
1139 req->rq_xmit_bytes_sent += transport->xmit.offset;
1140 transport->xmit.offset = 0;
1141 if (atomic_long_read(&xprt->xmit_queuelen) == 1)
1142 tcp_sock_set_cork(transport->inet, false);
1146 WARN_ON_ONCE(sent == 0 && status == 0);
1151 } while (status == 0);
1156 /* Should we call xs_close() here? */
1159 status = xs_stream_nospace(req, vm_wait);
1169 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1176 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1178 transport->old_data_ready = sk->sk_data_ready;
1179 transport->old_state_change = sk->sk_state_change;
1180 transport->old_write_space = sk->sk_write_space;
1181 transport->old_error_report = sk->sk_error_report;
1184 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1186 sk->sk_data_ready = transport->old_data_ready;
1187 sk->sk_state_change = transport->old_state_change;
1188 sk->sk_write_space = transport->old_write_space;
1189 sk->sk_error_report = transport->old_error_report;
1192 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
1194 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1196 transport->xprt_err = 0;
1197 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1198 clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state);
1199 clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state);
1200 clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state);
1201 clear_bit(XPRT_SOCK_NOSPACE, &transport->sock_state);
1204 static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr)
1206 set_bit(nr, &transport->sock_state);
1207 queue_work(xprtiod_workqueue, &transport->error_worker);
1210 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1212 xprt->connect_cookie++;
1213 smp_mb__before_atomic();
1214 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1215 clear_bit(XPRT_CLOSING, &xprt->state);
1216 xs_sock_reset_state_flags(xprt);
1217 smp_mb__after_atomic();
1221 * xs_error_report - callback to handle TCP socket state errors
1224 * Note: we don't call sock_error() since there may be a rpc_task
1225 * using the socket, and so we don't want to clear sk->sk_err.
1227 static void xs_error_report(struct sock *sk)
1229 struct sock_xprt *transport;
1230 struct rpc_xprt *xprt;
1232 if (!(xprt = xprt_from_sock(sk)))
1235 transport = container_of(xprt, struct sock_xprt, xprt);
1236 transport->xprt_err = -sk->sk_err;
1237 if (transport->xprt_err == 0)
1239 dprintk("RPC: xs_error_report client %p, error=%d...\n",
1240 xprt, -transport->xprt_err);
1241 trace_rpc_socket_error(xprt, sk->sk_socket, transport->xprt_err);
1243 /* barrier ensures xprt_err is set before XPRT_SOCK_WAKE_ERROR */
1244 smp_mb__before_atomic();
1245 xs_run_error_worker(transport, XPRT_SOCK_WAKE_ERROR);
1248 static void xs_reset_transport(struct sock_xprt *transport)
1250 struct socket *sock = transport->sock;
1251 struct sock *sk = transport->inet;
1252 struct rpc_xprt *xprt = &transport->xprt;
1253 struct file *filp = transport->file;
1258 * Make sure we're calling this in a context from which it is safe
1259 * to call __fput_sync(). In practice that means rpciod and the
1262 if (!(current->flags & PF_WQ_WORKER)) {
1264 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
1268 if (atomic_read(&transport->xprt.swapper))
1269 sk_clear_memalloc(sk);
1271 tls_handshake_cancel(sk);
1273 kernel_sock_shutdown(sock, SHUT_RDWR);
1275 mutex_lock(&transport->recv_mutex);
1277 transport->inet = NULL;
1278 transport->sock = NULL;
1279 transport->file = NULL;
1281 sk->sk_user_data = NULL;
1283 xs_restore_old_callbacks(transport, sk);
1284 xprt_clear_connected(xprt);
1285 xs_sock_reset_connection_flags(xprt);
1286 /* Reset stream record info */
1287 xs_stream_reset_connect(transport);
1289 mutex_unlock(&transport->recv_mutex);
1291 trace_rpc_socket_close(xprt, sock);
1294 xprt_disconnect_done(xprt);
1298 * xs_close - close a socket
1301 * This is used when all requests are complete; ie, no DRC state remains
1302 * on the server we want to save.
1304 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1305 * xs_reset_transport() zeroing the socket from underneath a writer.
1307 static void xs_close(struct rpc_xprt *xprt)
1309 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1311 dprintk("RPC: xs_close xprt %p\n", xprt);
1313 if (transport->sock)
1314 tls_handshake_close(transport->sock);
1315 xs_reset_transport(transport);
1316 xprt->reestablish_timeout = 0;
1319 static void xs_inject_disconnect(struct rpc_xprt *xprt)
1321 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
1323 xprt_disconnect_done(xprt);
1326 static void xs_xprt_free(struct rpc_xprt *xprt)
1328 xs_free_peer_addresses(xprt);
1333 * xs_destroy - prepare to shutdown a transport
1334 * @xprt: doomed transport
1337 static void xs_destroy(struct rpc_xprt *xprt)
1339 struct sock_xprt *transport = container_of(xprt,
1340 struct sock_xprt, xprt);
1341 dprintk("RPC: xs_destroy xprt %p\n", xprt);
1343 cancel_delayed_work_sync(&transport->connect_worker);
1345 cancel_work_sync(&transport->recv_worker);
1346 cancel_work_sync(&transport->error_worker);
1348 module_put(THIS_MODULE);
1352 * xs_udp_data_read_skb - receive callback for UDP sockets
1358 static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1360 struct sk_buff *skb)
1362 struct rpc_task *task;
1363 struct rpc_rqst *rovr;
1364 int repsize, copied;
1370 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
1374 /* Copy the XID from the skb... */
1375 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1379 /* Look up and lock the request corresponding to the given XID */
1380 spin_lock(&xprt->queue_lock);
1381 rovr = xprt_lookup_rqst(xprt, *xp);
1384 xprt_pin_rqst(rovr);
1385 xprt_update_rtt(rovr->rq_task);
1386 spin_unlock(&xprt->queue_lock);
1387 task = rovr->rq_task;
1389 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1392 /* Suck it into the iovec, verify checksum if not done by hw. */
1393 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1394 spin_lock(&xprt->queue_lock);
1395 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1400 spin_lock(&xprt->transport_lock);
1401 xprt_adjust_cwnd(xprt, task, copied);
1402 spin_unlock(&xprt->transport_lock);
1403 spin_lock(&xprt->queue_lock);
1404 xprt_complete_rqst(task, copied);
1405 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1407 xprt_unpin_rqst(rovr);
1409 spin_unlock(&xprt->queue_lock);
1412 static void xs_udp_data_receive(struct sock_xprt *transport)
1414 struct sk_buff *skb;
1418 mutex_lock(&transport->recv_mutex);
1419 sk = transport->inet;
1423 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err);
1426 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1430 xs_poll_check_readable(transport);
1432 mutex_unlock(&transport->recv_mutex);
1435 static void xs_udp_data_receive_workfn(struct work_struct *work)
1437 struct sock_xprt *transport =
1438 container_of(work, struct sock_xprt, recv_worker);
1439 unsigned int pflags = memalloc_nofs_save();
1441 xs_udp_data_receive(transport);
1442 memalloc_nofs_restore(pflags);
1446 * xs_data_ready - "data ready" callback for sockets
1447 * @sk: socket with data to read
1450 static void xs_data_ready(struct sock *sk)
1452 struct rpc_xprt *xprt;
1454 trace_sk_data_ready(sk);
1456 xprt = xprt_from_sock(sk);
1458 struct sock_xprt *transport = container_of(xprt,
1459 struct sock_xprt, xprt);
1461 trace_xs_data_ready(xprt);
1463 transport->old_data_ready(sk);
1465 if (test_bit(XPRT_SOCK_IGNORE_RECV, &transport->sock_state))
1468 /* Any data means we had a useful conversation, so
1469 * then we don't need to delay the next reconnect
1471 if (xprt->reestablish_timeout)
1472 xprt->reestablish_timeout = 0;
1473 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1474 queue_work(xprtiod_workqueue, &transport->recv_worker);
1479 * Helper function to force a TCP close if the server is sending
1480 * junk and/or it has put us in CLOSE_WAIT
1482 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1484 xprt_force_disconnect(xprt);
1487 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1488 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1492 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1495 * xs_local_state_change - callback to handle AF_LOCAL socket state changes
1496 * @sk: socket whose state has changed
1499 static void xs_local_state_change(struct sock *sk)
1501 struct rpc_xprt *xprt;
1502 struct sock_xprt *transport;
1504 if (!(xprt = xprt_from_sock(sk)))
1506 transport = container_of(xprt, struct sock_xprt, xprt);
1507 if (sk->sk_shutdown & SHUTDOWN_MASK) {
1508 clear_bit(XPRT_CONNECTED, &xprt->state);
1509 /* Trigger the socket release */
1510 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1515 * xs_tcp_state_change - callback to handle TCP socket state changes
1516 * @sk: socket whose state has changed
1519 static void xs_tcp_state_change(struct sock *sk)
1521 struct rpc_xprt *xprt;
1522 struct sock_xprt *transport;
1524 if (!(xprt = xprt_from_sock(sk)))
1526 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1527 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1528 sk->sk_state, xprt_connected(xprt),
1529 sock_flag(sk, SOCK_DEAD),
1530 sock_flag(sk, SOCK_ZAPPED),
1533 transport = container_of(xprt, struct sock_xprt, xprt);
1534 trace_rpc_socket_state_change(xprt, sk->sk_socket);
1535 switch (sk->sk_state) {
1536 case TCP_ESTABLISHED:
1537 if (!xprt_test_and_set_connected(xprt)) {
1538 xprt->connect_cookie++;
1539 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1540 xprt_clear_connecting(xprt);
1542 xprt->stat.connect_count++;
1543 xprt->stat.connect_time += (long)jiffies -
1544 xprt->stat.connect_start;
1545 xs_run_error_worker(transport, XPRT_SOCK_WAKE_PENDING);
1549 /* The client initiated a shutdown of the socket */
1550 xprt->connect_cookie++;
1551 xprt->reestablish_timeout = 0;
1552 set_bit(XPRT_CLOSING, &xprt->state);
1553 smp_mb__before_atomic();
1554 clear_bit(XPRT_CONNECTED, &xprt->state);
1555 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1556 smp_mb__after_atomic();
1558 case TCP_CLOSE_WAIT:
1559 /* The server initiated a shutdown of the socket */
1560 xprt->connect_cookie++;
1561 clear_bit(XPRT_CONNECTED, &xprt->state);
1562 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1566 * If the server closed down the connection, make sure that
1567 * we back off before reconnecting
1569 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1570 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1573 set_bit(XPRT_CLOSING, &xprt->state);
1574 smp_mb__before_atomic();
1575 clear_bit(XPRT_CONNECTED, &xprt->state);
1576 smp_mb__after_atomic();
1579 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1580 &transport->sock_state)) {
1581 xs_reset_srcport(transport);
1582 xprt_clear_connecting(xprt);
1584 clear_bit(XPRT_CLOSING, &xprt->state);
1585 /* Trigger the socket release */
1586 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1590 static void xs_write_space(struct sock *sk)
1592 struct sock_xprt *transport;
1593 struct rpc_xprt *xprt;
1597 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1599 if (unlikely(!(xprt = xprt_from_sock(sk))))
1601 transport = container_of(xprt, struct sock_xprt, xprt);
1602 if (!test_and_clear_bit(XPRT_SOCK_NOSPACE, &transport->sock_state))
1604 xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE);
1605 sk->sk_write_pending--;
1609 * xs_udp_write_space - callback invoked when socket buffer space
1611 * @sk: socket whose state has changed
1613 * Called when more output buffer space is available for this socket.
1614 * We try not to wake our writers until they can make "significant"
1615 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1616 * with a bunch of small requests.
1618 static void xs_udp_write_space(struct sock *sk)
1620 /* from net/core/sock.c:sock_def_write_space */
1621 if (sock_writeable(sk))
1626 * xs_tcp_write_space - callback invoked when socket buffer space
1628 * @sk: socket whose state has changed
1630 * Called when more output buffer space is available for this socket.
1631 * We try not to wake our writers until they can make "significant"
1632 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1633 * with a bunch of small requests.
1635 static void xs_tcp_write_space(struct sock *sk)
1637 /* from net/core/stream.c:sk_stream_write_space */
1638 if (sk_stream_is_writeable(sk))
1642 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1644 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1645 struct sock *sk = transport->inet;
1647 if (transport->rcvsize) {
1648 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1649 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1651 if (transport->sndsize) {
1652 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1653 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1654 sk->sk_write_space(sk);
1659 * xs_udp_set_buffer_size - set send and receive limits
1660 * @xprt: generic transport
1661 * @sndsize: requested size of send buffer, in bytes
1662 * @rcvsize: requested size of receive buffer, in bytes
1664 * Set socket send and receive buffer size limits.
1666 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1668 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1670 transport->sndsize = 0;
1672 transport->sndsize = sndsize + 1024;
1673 transport->rcvsize = 0;
1675 transport->rcvsize = rcvsize + 1024;
1677 xs_udp_do_set_buffer_size(xprt);
1681 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1682 * @xprt: controlling transport
1683 * @task: task that timed out
1685 * Adjust the congestion window after a retransmit timeout has occurred.
1687 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1689 spin_lock(&xprt->transport_lock);
1690 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1691 spin_unlock(&xprt->transport_lock);
1694 static int xs_get_random_port(void)
1696 unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1697 unsigned short range;
1698 unsigned short rand;
1702 range = max - min + 1;
1703 rand = get_random_u32_below(range);
1707 static unsigned short xs_sock_getport(struct socket *sock)
1709 struct sockaddr_storage buf;
1710 unsigned short port = 0;
1712 if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1714 switch (buf.ss_family) {
1716 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1719 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1726 * xs_set_port - reset the port number in the remote endpoint address
1727 * @xprt: generic transport
1728 * @port: new port number
1731 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1733 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1735 rpc_set_port(xs_addr(xprt), port);
1736 xs_update_peer_port(xprt);
1739 static void xs_reset_srcport(struct sock_xprt *transport)
1741 transport->srcport = 0;
1744 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1746 if (transport->srcport == 0 && transport->xprt.reuseport)
1747 transport->srcport = xs_sock_getport(sock);
1750 static int xs_get_srcport(struct sock_xprt *transport)
1752 int port = transport->srcport;
1754 if (port == 0 && transport->xprt.resvport)
1755 port = xs_get_random_port();
1759 static unsigned short xs_sock_srcport(struct rpc_xprt *xprt)
1761 struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt);
1762 unsigned short ret = 0;
1763 mutex_lock(&sock->recv_mutex);
1765 ret = xs_sock_getport(sock->sock);
1766 mutex_unlock(&sock->recv_mutex);
1770 static int xs_sock_srcaddr(struct rpc_xprt *xprt, char *buf, size_t buflen)
1772 struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt);
1775 struct sockaddr_storage st;
1777 int ret = -ENOTCONN;
1779 mutex_lock(&sock->recv_mutex);
1781 ret = kernel_getsockname(sock->sock, &saddr.sa);
1783 ret = snprintf(buf, buflen, "%pISc", &saddr.sa);
1785 mutex_unlock(&sock->recv_mutex);
1789 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1791 if (transport->srcport != 0)
1792 transport->srcport = 0;
1793 if (!transport->xprt.resvport)
1795 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1796 return xprt_max_resvport;
1799 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1801 struct sockaddr_storage myaddr;
1803 int port = xs_get_srcport(transport);
1804 unsigned short last;
1807 * If we are asking for any ephemeral port (i.e. port == 0 &&
1808 * transport->xprt.resvport == 0), don't bind. Let the local
1809 * port selection happen implicitly when the socket is used
1810 * (for example at connect time).
1812 * This ensures that we can continue to establish TCP
1813 * connections even when all local ephemeral ports are already
1814 * a part of some TCP connection. This makes no difference
1815 * for UDP sockets, but also doesn't harm them.
1817 * If we're asking for any reserved port (i.e. port == 0 &&
1818 * transport->xprt.resvport == 1) xs_get_srcport above will
1819 * ensure that port is non-zero and we will bind as needed.
1824 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1826 rpc_set_port((struct sockaddr *)&myaddr, port);
1827 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1828 transport->xprt.addrlen);
1830 if (transport->xprt.reuseport)
1831 transport->srcport = port;
1835 port = xs_next_srcport(transport, port);
1838 } while (err == -EADDRINUSE && nloop != 2);
1840 if (myaddr.ss_family == AF_INET)
1841 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1842 &((struct sockaddr_in *)&myaddr)->sin_addr,
1843 port, err ? "failed" : "ok", err);
1845 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1846 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1847 port, err ? "failed" : "ok", err);
1852 * We don't support autobind on AF_LOCAL sockets
1854 static void xs_local_rpcbind(struct rpc_task *task)
1856 xprt_set_bound(task->tk_xprt);
1859 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1863 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1864 static struct lock_class_key xs_key[3];
1865 static struct lock_class_key xs_slock_key[3];
1867 static inline void xs_reclassify_socketu(struct socket *sock)
1869 struct sock *sk = sock->sk;
1871 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1872 &xs_slock_key[0], "sk_lock-AF_LOCAL-RPC", &xs_key[0]);
1875 static inline void xs_reclassify_socket4(struct socket *sock)
1877 struct sock *sk = sock->sk;
1879 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1880 &xs_slock_key[1], "sk_lock-AF_INET-RPC", &xs_key[1]);
1883 static inline void xs_reclassify_socket6(struct socket *sock)
1885 struct sock *sk = sock->sk;
1887 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1888 &xs_slock_key[2], "sk_lock-AF_INET6-RPC", &xs_key[2]);
1891 static inline void xs_reclassify_socket(int family, struct socket *sock)
1893 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1898 xs_reclassify_socketu(sock);
1901 xs_reclassify_socket4(sock);
1904 xs_reclassify_socket6(sock);
1909 static inline void xs_reclassify_socket(int family, struct socket *sock)
1914 static void xs_dummy_setup_socket(struct work_struct *work)
1918 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1919 struct sock_xprt *transport, int family, int type,
1920 int protocol, bool reuseport)
1923 struct socket *sock;
1926 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1928 dprintk("RPC: can't create %d transport socket (%d).\n",
1932 xs_reclassify_socket(family, sock);
1935 sock_set_reuseport(sock->sk);
1937 err = xs_bind(transport, sock);
1943 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1945 return ERR_CAST(filp);
1946 transport->file = filp;
1950 return ERR_PTR(err);
1953 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1954 struct socket *sock)
1956 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1959 if (!transport->inet) {
1960 struct sock *sk = sock->sk;
1964 xs_save_old_callbacks(transport, sk);
1966 sk->sk_user_data = xprt;
1967 sk->sk_data_ready = xs_data_ready;
1968 sk->sk_write_space = xs_udp_write_space;
1969 sk->sk_state_change = xs_local_state_change;
1970 sk->sk_error_report = xs_error_report;
1971 sk->sk_use_task_frag = false;
1973 xprt_clear_connected(xprt);
1975 /* Reset to new socket */
1976 transport->sock = sock;
1977 transport->inet = sk;
1982 xs_stream_start_connect(transport);
1984 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1988 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1989 * @transport: socket transport to connect
1991 static int xs_local_setup_socket(struct sock_xprt *transport)
1993 struct rpc_xprt *xprt = &transport->xprt;
1995 struct socket *sock;
1998 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1999 SOCK_STREAM, 0, &sock, 1);
2001 dprintk("RPC: can't create AF_LOCAL "
2002 "transport socket (%d).\n", -status);
2005 xs_reclassify_socket(AF_LOCAL, sock);
2007 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
2009 status = PTR_ERR(filp);
2012 transport->file = filp;
2014 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
2015 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2017 status = xs_local_finish_connecting(xprt, sock);
2018 trace_rpc_socket_connect(xprt, sock, status);
2021 dprintk("RPC: xprt %p connected to %s\n",
2022 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2023 xprt->stat.connect_count++;
2024 xprt->stat.connect_time += (long)jiffies -
2025 xprt->stat.connect_start;
2026 xprt_set_connected(xprt);
2031 dprintk("RPC: xprt %p: socket %s does not exist\n",
2032 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2035 dprintk("RPC: xprt %p: connection refused for %s\n",
2036 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2039 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2041 xprt->address_strings[RPC_DISPLAY_ADDR]);
2045 xprt_clear_connecting(xprt);
2046 xprt_wake_pending_tasks(xprt, status);
2050 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2052 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2055 if (transport->file)
2056 goto force_disconnect;
2058 if (RPC_IS_ASYNC(task)) {
2060 * We want the AF_LOCAL connect to be resolved in the
2061 * filesystem namespace of the process making the rpc
2062 * call. Thus we connect synchronously.
2064 * If we want to support asynchronous AF_LOCAL calls,
2065 * we'll need to figure out how to pass a namespace to
2068 rpc_task_set_rpc_status(task, -ENOTCONN);
2071 ret = xs_local_setup_socket(transport);
2072 if (ret && !RPC_IS_SOFTCONN(task))
2073 msleep_interruptible(15000);
2076 xprt_force_disconnect(xprt);
2078 xprt_clear_connecting(xprt);
2079 xprt_wake_pending_tasks(xprt, -ENOTCONN);
2082 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2084 * Note that this should be called with XPRT_LOCKED held, or recv_mutex
2085 * held, or when we otherwise know that we have exclusive access to the
2086 * socket, to guard against races with xs_reset_transport.
2088 static void xs_set_memalloc(struct rpc_xprt *xprt)
2090 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2094 * If there's no sock, then we have nothing to set. The
2095 * reconnecting process will get it for us.
2097 if (!transport->inet)
2099 if (atomic_read(&xprt->swapper))
2100 sk_set_memalloc(transport->inet);
2104 * xs_enable_swap - Tag this transport as being used for swap.
2105 * @xprt: transport to tag
2107 * Take a reference to this transport on behalf of the rpc_clnt, and
2108 * optionally mark it for swapping if it wasn't already.
2111 xs_enable_swap(struct rpc_xprt *xprt)
2113 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2115 mutex_lock(&xs->recv_mutex);
2116 if (atomic_inc_return(&xprt->swapper) == 1 &&
2118 sk_set_memalloc(xs->inet);
2119 mutex_unlock(&xs->recv_mutex);
2124 * xs_disable_swap - Untag this transport as being used for swap.
2125 * @xprt: transport to tag
2127 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2128 * swapper refcount goes to 0, untag the socket as a memalloc socket.
2131 xs_disable_swap(struct rpc_xprt *xprt)
2133 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2135 mutex_lock(&xs->recv_mutex);
2136 if (atomic_dec_and_test(&xprt->swapper) &&
2138 sk_clear_memalloc(xs->inet);
2139 mutex_unlock(&xs->recv_mutex);
2142 static void xs_set_memalloc(struct rpc_xprt *xprt)
2147 xs_enable_swap(struct rpc_xprt *xprt)
2153 xs_disable_swap(struct rpc_xprt *xprt)
2158 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2160 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2162 if (!transport->inet) {
2163 struct sock *sk = sock->sk;
2167 xs_save_old_callbacks(transport, sk);
2169 sk->sk_user_data = xprt;
2170 sk->sk_data_ready = xs_data_ready;
2171 sk->sk_write_space = xs_udp_write_space;
2172 sk->sk_use_task_frag = false;
2174 xprt_set_connected(xprt);
2176 /* Reset to new socket */
2177 transport->sock = sock;
2178 transport->inet = sk;
2180 xs_set_memalloc(xprt);
2184 xs_udp_do_set_buffer_size(xprt);
2186 xprt->stat.connect_start = jiffies;
2189 static void xs_udp_setup_socket(struct work_struct *work)
2191 struct sock_xprt *transport =
2192 container_of(work, struct sock_xprt, connect_worker.work);
2193 struct rpc_xprt *xprt = &transport->xprt;
2194 struct socket *sock;
2196 unsigned int pflags = current->flags;
2198 if (atomic_read(&xprt->swapper))
2199 current->flags |= PF_MEMALLOC;
2200 sock = xs_create_sock(xprt, transport,
2201 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2202 IPPROTO_UDP, false);
2206 dprintk("RPC: worker connecting xprt %p via %s to "
2207 "%s (port %s)\n", xprt,
2208 xprt->address_strings[RPC_DISPLAY_PROTO],
2209 xprt->address_strings[RPC_DISPLAY_ADDR],
2210 xprt->address_strings[RPC_DISPLAY_PORT]);
2212 xs_udp_finish_connecting(xprt, sock);
2213 trace_rpc_socket_connect(xprt, sock, 0);
2216 xprt_clear_connecting(xprt);
2217 xprt_unlock_connect(xprt, transport);
2218 xprt_wake_pending_tasks(xprt, status);
2219 current_restore_flags(pflags, PF_MEMALLOC);
2223 * xs_tcp_shutdown - gracefully shut down a TCP socket
2226 * Initiates a graceful shutdown of the TCP socket by calling the
2227 * equivalent of shutdown(SHUT_RDWR);
2229 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2231 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2232 struct socket *sock = transport->sock;
2233 int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2237 if (!xprt->reuseport) {
2246 case TCP_ESTABLISHED:
2247 case TCP_CLOSE_WAIT:
2248 kernel_sock_shutdown(sock, SHUT_RDWR);
2249 trace_rpc_socket_shutdown(xprt, sock);
2252 xs_reset_transport(transport);
2256 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2257 struct socket *sock)
2259 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2260 struct net *net = sock_net(sock->sk);
2261 unsigned long connect_timeout;
2262 unsigned long syn_retries;
2263 unsigned int keepidle;
2264 unsigned int keepcnt;
2268 spin_lock(&xprt->transport_lock);
2269 keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2270 keepcnt = xprt->timeout->to_retries + 1;
2271 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2272 (xprt->timeout->to_retries + 1);
2273 clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2274 spin_unlock(&xprt->transport_lock);
2276 /* TCP Keepalive options */
2277 sock_set_keepalive(sock->sk);
2278 tcp_sock_set_keepidle(sock->sk, keepidle);
2279 tcp_sock_set_keepintvl(sock->sk, keepidle);
2280 tcp_sock_set_keepcnt(sock->sk, keepcnt);
2282 /* TCP user timeout (see RFC5482) */
2283 tcp_sock_set_user_timeout(sock->sk, timeo);
2285 /* Connect timeout */
2286 connect_timeout = max_t(unsigned long,
2287 DIV_ROUND_UP(xprt->connect_timeout, HZ), 1);
2288 syn_retries = max_t(unsigned long,
2289 READ_ONCE(net->ipv4.sysctl_tcp_syn_retries), 1);
2290 for (t = 0; t <= syn_retries && (1UL << t) < connect_timeout; t++)
2292 if (t <= syn_retries)
2293 tcp_sock_set_syncnt(sock->sk, t - 1);
2296 static void xs_tcp_do_set_connect_timeout(struct rpc_xprt *xprt,
2297 unsigned long connect_timeout)
2299 struct sock_xprt *transport =
2300 container_of(xprt, struct sock_xprt, xprt);
2301 struct rpc_timeout to;
2302 unsigned long initval;
2304 memcpy(&to, xprt->timeout, sizeof(to));
2305 /* Arbitrary lower limit */
2306 initval = max_t(unsigned long, connect_timeout, XS_TCP_INIT_REEST_TO);
2307 to.to_initval = initval;
2308 to.to_maxval = initval;
2310 memcpy(&transport->tcp_timeout, &to, sizeof(transport->tcp_timeout));
2311 xprt->timeout = &transport->tcp_timeout;
2312 xprt->connect_timeout = connect_timeout;
2315 static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2316 unsigned long connect_timeout,
2317 unsigned long reconnect_timeout)
2319 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2321 spin_lock(&xprt->transport_lock);
2322 if (reconnect_timeout < xprt->max_reconnect_timeout)
2323 xprt->max_reconnect_timeout = reconnect_timeout;
2324 if (connect_timeout < xprt->connect_timeout)
2325 xs_tcp_do_set_connect_timeout(xprt, connect_timeout);
2326 set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2327 spin_unlock(&xprt->transport_lock);
2330 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2332 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2334 if (!transport->inet) {
2335 struct sock *sk = sock->sk;
2337 /* Avoid temporary address, they are bad for long-lived
2338 * connections such as NFS mounts.
2339 * RFC4941, section 3.6 suggests that:
2340 * Individual applications, which have specific
2341 * knowledge about the normal duration of connections,
2342 * MAY override this as appropriate.
2344 if (xs_addr(xprt)->sa_family == PF_INET6) {
2345 ip6_sock_set_addr_preferences(sk,
2346 IPV6_PREFER_SRC_PUBLIC);
2349 xs_tcp_set_socket_timeouts(xprt, sock);
2350 tcp_sock_set_nodelay(sk);
2354 xs_save_old_callbacks(transport, sk);
2356 sk->sk_user_data = xprt;
2357 sk->sk_data_ready = xs_data_ready;
2358 sk->sk_state_change = xs_tcp_state_change;
2359 sk->sk_write_space = xs_tcp_write_space;
2360 sk->sk_error_report = xs_error_report;
2361 sk->sk_use_task_frag = false;
2363 /* socket options */
2364 sock_reset_flag(sk, SOCK_LINGER);
2366 xprt_clear_connected(xprt);
2368 /* Reset to new socket */
2369 transport->sock = sock;
2370 transport->inet = sk;
2375 if (!xprt_bound(xprt))
2378 xs_set_memalloc(xprt);
2380 xs_stream_start_connect(transport);
2382 /* Tell the socket layer to start connecting... */
2383 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2384 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2388 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2389 * @work: queued work item
2391 * Invoked by a work queue tasklet.
2393 static void xs_tcp_setup_socket(struct work_struct *work)
2395 struct sock_xprt *transport =
2396 container_of(work, struct sock_xprt, connect_worker.work);
2397 struct socket *sock = transport->sock;
2398 struct rpc_xprt *xprt = &transport->xprt;
2400 unsigned int pflags = current->flags;
2402 if (atomic_read(&xprt->swapper))
2403 current->flags |= PF_MEMALLOC;
2405 if (xprt_connected(xprt))
2407 if (test_and_clear_bit(XPRT_SOCK_CONNECT_SENT,
2408 &transport->sock_state) ||
2410 xs_reset_transport(transport);
2411 sock = xs_create_sock(xprt, transport, xs_addr(xprt)->sa_family,
2412 SOCK_STREAM, IPPROTO_TCP, true);
2414 xprt_wake_pending_tasks(xprt, PTR_ERR(sock));
2419 dprintk("RPC: worker connecting xprt %p via %s to "
2420 "%s (port %s)\n", xprt,
2421 xprt->address_strings[RPC_DISPLAY_PROTO],
2422 xprt->address_strings[RPC_DISPLAY_ADDR],
2423 xprt->address_strings[RPC_DISPLAY_PORT]);
2425 status = xs_tcp_finish_connecting(xprt, sock);
2426 trace_rpc_socket_connect(xprt, sock, status);
2427 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2428 xprt, -status, xprt_connected(xprt),
2429 sock->sk->sk_state);
2434 set_bit(XPRT_SOCK_CONNECT_SENT, &transport->sock_state);
2435 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2436 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2440 case -EADDRNOTAVAIL:
2441 /* Source port number is unavailable. Try a new one! */
2442 transport->srcport = 0;
2446 /* Happens, for instance, if the user specified a link
2447 * local IPv6 address without a scope-id.
2458 printk("%s: connect returned unhandled error %d\n",
2463 /* xs_tcp_force_close() wakes tasks with a fixed error code.
2464 * We need to wake them first to ensure the correct error code.
2466 xprt_wake_pending_tasks(xprt, status);
2467 xs_tcp_force_close(xprt);
2469 xprt_clear_connecting(xprt);
2471 xprt_unlock_connect(xprt, transport);
2472 current_restore_flags(pflags, PF_MEMALLOC);
2476 * Transfer the connected socket to @upper_transport, then mark that
2479 static int xs_tcp_tls_finish_connecting(struct rpc_xprt *lower_xprt,
2480 struct sock_xprt *upper_transport)
2482 struct sock_xprt *lower_transport =
2483 container_of(lower_xprt, struct sock_xprt, xprt);
2484 struct rpc_xprt *upper_xprt = &upper_transport->xprt;
2486 if (!upper_transport->inet) {
2487 struct socket *sock = lower_transport->sock;
2488 struct sock *sk = sock->sk;
2490 /* Avoid temporary address, they are bad for long-lived
2491 * connections such as NFS mounts.
2492 * RFC4941, section 3.6 suggests that:
2493 * Individual applications, which have specific
2494 * knowledge about the normal duration of connections,
2495 * MAY override this as appropriate.
2497 if (xs_addr(upper_xprt)->sa_family == PF_INET6)
2498 ip6_sock_set_addr_preferences(sk, IPV6_PREFER_SRC_PUBLIC);
2500 xs_tcp_set_socket_timeouts(upper_xprt, sock);
2501 tcp_sock_set_nodelay(sk);
2505 /* @sk is already connected, so it now has the RPC callbacks.
2506 * Reach into @lower_transport to save the original ones.
2508 upper_transport->old_data_ready = lower_transport->old_data_ready;
2509 upper_transport->old_state_change = lower_transport->old_state_change;
2510 upper_transport->old_write_space = lower_transport->old_write_space;
2511 upper_transport->old_error_report = lower_transport->old_error_report;
2512 sk->sk_user_data = upper_xprt;
2514 /* socket options */
2515 sock_reset_flag(sk, SOCK_LINGER);
2517 xprt_clear_connected(upper_xprt);
2519 upper_transport->sock = sock;
2520 upper_transport->inet = sk;
2521 upper_transport->file = lower_transport->file;
2525 /* Reset lower_transport before shutting down its clnt */
2526 mutex_lock(&lower_transport->recv_mutex);
2527 lower_transport->inet = NULL;
2528 lower_transport->sock = NULL;
2529 lower_transport->file = NULL;
2531 xprt_clear_connected(lower_xprt);
2532 xs_sock_reset_connection_flags(lower_xprt);
2533 xs_stream_reset_connect(lower_transport);
2534 mutex_unlock(&lower_transport->recv_mutex);
2537 if (!xprt_bound(upper_xprt))
2540 xs_set_memalloc(upper_xprt);
2542 if (!xprt_test_and_set_connected(upper_xprt)) {
2543 upper_xprt->connect_cookie++;
2544 clear_bit(XPRT_SOCK_CONNECTING, &upper_transport->sock_state);
2545 xprt_clear_connecting(upper_xprt);
2547 upper_xprt->stat.connect_count++;
2548 upper_xprt->stat.connect_time += (long)jiffies -
2549 upper_xprt->stat.connect_start;
2550 xs_run_error_worker(upper_transport, XPRT_SOCK_WAKE_PENDING);
2556 * xs_tls_handshake_done - TLS handshake completion handler
2557 * @data: address of xprt to wake
2558 * @status: status of handshake
2559 * @peerid: serial number of key containing the remote's identity
2562 static void xs_tls_handshake_done(void *data, int status, key_serial_t peerid)
2564 struct rpc_xprt *lower_xprt = data;
2565 struct sock_xprt *lower_transport =
2566 container_of(lower_xprt, struct sock_xprt, xprt);
2568 lower_transport->xprt_err = status ? -EACCES : 0;
2569 complete(&lower_transport->handshake_done);
2570 xprt_put(lower_xprt);
2573 static int xs_tls_handshake_sync(struct rpc_xprt *lower_xprt, struct xprtsec_parms *xprtsec)
2575 struct sock_xprt *lower_transport =
2576 container_of(lower_xprt, struct sock_xprt, xprt);
2577 struct tls_handshake_args args = {
2578 .ta_sock = lower_transport->sock,
2579 .ta_done = xs_tls_handshake_done,
2580 .ta_data = xprt_get(lower_xprt),
2581 .ta_peername = lower_xprt->servername,
2583 struct sock *sk = lower_transport->inet;
2586 init_completion(&lower_transport->handshake_done);
2587 set_bit(XPRT_SOCK_IGNORE_RECV, &lower_transport->sock_state);
2588 lower_transport->xprt_err = -ETIMEDOUT;
2589 switch (xprtsec->policy) {
2590 case RPC_XPRTSEC_TLS_ANON:
2591 rc = tls_client_hello_anon(&args, GFP_KERNEL);
2595 case RPC_XPRTSEC_TLS_X509:
2596 args.ta_my_cert = xprtsec->cert_serial;
2597 args.ta_my_privkey = xprtsec->privkey_serial;
2598 rc = tls_client_hello_x509(&args, GFP_KERNEL);
2607 rc = wait_for_completion_interruptible_timeout(&lower_transport->handshake_done,
2608 XS_TLS_HANDSHAKE_TO);
2610 if (!tls_handshake_cancel(sk)) {
2617 rc = lower_transport->xprt_err;
2620 xs_stream_reset_connect(lower_transport);
2621 clear_bit(XPRT_SOCK_IGNORE_RECV, &lower_transport->sock_state);
2625 xprt_put(lower_xprt);
2630 * xs_tcp_tls_setup_socket - establish a TLS session on a TCP socket
2631 * @work: queued work item
2633 * Invoked by a work queue tasklet.
2635 * For RPC-with-TLS, there is a two-stage connection process.
2637 * The "upper-layer xprt" is visible to the RPC consumer. Once it has
2638 * been marked connected, the consumer knows that a TCP connection and
2639 * a TLS session have been established.
2641 * A "lower-layer xprt", created in this function, handles the mechanics
2642 * of connecting the TCP socket, performing the RPC_AUTH_TLS probe, and
2643 * then driving the TLS handshake. Once all that is complete, the upper
2644 * layer xprt is marked connected.
2646 static void xs_tcp_tls_setup_socket(struct work_struct *work)
2648 struct sock_xprt *upper_transport =
2649 container_of(work, struct sock_xprt, connect_worker.work);
2650 struct rpc_clnt *upper_clnt = upper_transport->clnt;
2651 struct rpc_xprt *upper_xprt = &upper_transport->xprt;
2652 struct rpc_create_args args = {
2653 .net = upper_xprt->xprt_net,
2654 .protocol = upper_xprt->prot,
2655 .address = (struct sockaddr *)&upper_xprt->addr,
2656 .addrsize = upper_xprt->addrlen,
2657 .timeout = upper_clnt->cl_timeout,
2658 .servername = upper_xprt->servername,
2659 .program = upper_clnt->cl_program,
2660 .prognumber = upper_clnt->cl_prog,
2661 .version = upper_clnt->cl_vers,
2662 .authflavor = RPC_AUTH_TLS,
2663 .cred = upper_clnt->cl_cred,
2665 .policy = RPC_XPRTSEC_NONE,
2668 unsigned int pflags = current->flags;
2669 struct rpc_clnt *lower_clnt;
2670 struct rpc_xprt *lower_xprt;
2673 if (atomic_read(&upper_xprt->swapper))
2674 current->flags |= PF_MEMALLOC;
2676 xs_stream_start_connect(upper_transport);
2678 /* This implicitly sends an RPC_AUTH_TLS probe */
2679 lower_clnt = rpc_create(&args);
2680 if (IS_ERR(lower_clnt)) {
2681 trace_rpc_tls_unavailable(upper_clnt, upper_xprt);
2682 clear_bit(XPRT_SOCK_CONNECTING, &upper_transport->sock_state);
2683 xprt_clear_connecting(upper_xprt);
2684 xprt_wake_pending_tasks(upper_xprt, PTR_ERR(lower_clnt));
2685 xs_run_error_worker(upper_transport, XPRT_SOCK_WAKE_PENDING);
2689 /* RPC_AUTH_TLS probe was successful. Try a TLS handshake on
2693 lower_xprt = rcu_dereference(lower_clnt->cl_xprt);
2696 if (wait_on_bit_lock(&lower_xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2699 status = xs_tls_handshake_sync(lower_xprt, &upper_xprt->xprtsec);
2701 trace_rpc_tls_not_started(upper_clnt, upper_xprt);
2705 status = xs_tcp_tls_finish_connecting(lower_xprt, upper_transport);
2708 xprt_release_write(lower_xprt, NULL);
2710 trace_rpc_socket_connect(upper_xprt, upper_transport->sock, 0);
2711 if (!xprt_test_and_set_connected(upper_xprt)) {
2712 upper_xprt->connect_cookie++;
2713 clear_bit(XPRT_SOCK_CONNECTING, &upper_transport->sock_state);
2714 xprt_clear_connecting(upper_xprt);
2716 upper_xprt->stat.connect_count++;
2717 upper_xprt->stat.connect_time += (long)jiffies -
2718 upper_xprt->stat.connect_start;
2719 xs_run_error_worker(upper_transport, XPRT_SOCK_WAKE_PENDING);
2721 rpc_shutdown_client(lower_clnt);
2724 current_restore_flags(pflags, PF_MEMALLOC);
2725 upper_transport->clnt = NULL;
2726 xprt_unlock_connect(upper_xprt, upper_transport);
2730 xprt_release_write(lower_xprt, NULL);
2731 rpc_shutdown_client(lower_clnt);
2733 /* xprt_force_disconnect() wakes tasks with a fixed tk_status code.
2734 * Wake them first here to ensure they get our tk_status code.
2736 xprt_wake_pending_tasks(upper_xprt, status);
2737 xs_tcp_force_close(upper_xprt);
2738 xprt_clear_connecting(upper_xprt);
2743 * xs_connect - connect a socket to a remote endpoint
2744 * @xprt: pointer to transport structure
2745 * @task: address of RPC task that manages state of connect request
2747 * TCP: If the remote end dropped the connection, delay reconnecting.
2749 * UDP socket connects are synchronous, but we use a work queue anyway
2750 * to guarantee that even unprivileged user processes can set up a
2751 * socket on a privileged port.
2753 * If a UDP socket connect fails, the delay behavior here prevents
2754 * retry floods (hard mounts).
2756 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2758 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2759 unsigned long delay = 0;
2761 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2763 if (transport->sock != NULL) {
2764 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2765 "seconds\n", xprt, xprt->reestablish_timeout / HZ);
2767 delay = xprt_reconnect_delay(xprt);
2768 xprt_reconnect_backoff(xprt, XS_TCP_INIT_REEST_TO);
2771 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2773 transport->clnt = task->tk_client;
2774 queue_delayed_work(xprtiod_workqueue,
2775 &transport->connect_worker,
2779 static void xs_wake_disconnect(struct sock_xprt *transport)
2781 if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state))
2782 xs_tcp_force_close(&transport->xprt);
2785 static void xs_wake_write(struct sock_xprt *transport)
2787 if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state))
2788 xprt_write_space(&transport->xprt);
2791 static void xs_wake_error(struct sock_xprt *transport)
2795 if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2797 sockerr = xchg(&transport->xprt_err, 0);
2799 xprt_wake_pending_tasks(&transport->xprt, sockerr);
2800 xs_tcp_force_close(&transport->xprt);
2804 static void xs_wake_pending(struct sock_xprt *transport)
2806 if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING, &transport->sock_state))
2807 xprt_wake_pending_tasks(&transport->xprt, -EAGAIN);
2810 static void xs_error_handle(struct work_struct *work)
2812 struct sock_xprt *transport = container_of(work,
2813 struct sock_xprt, error_worker);
2815 xs_wake_disconnect(transport);
2816 xs_wake_write(transport);
2817 xs_wake_error(transport);
2818 xs_wake_pending(transport);
2822 * xs_local_print_stats - display AF_LOCAL socket-specific stats
2823 * @xprt: rpc_xprt struct containing statistics
2827 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2831 if (xprt_connected(xprt))
2832 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2834 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2835 "%llu %llu %lu %llu %llu\n",
2836 xprt->stat.bind_count,
2837 xprt->stat.connect_count,
2838 xprt->stat.connect_time / HZ,
2842 xprt->stat.bad_xids,
2845 xprt->stat.max_slots,
2846 xprt->stat.sending_u,
2847 xprt->stat.pending_u);
2851 * xs_udp_print_stats - display UDP socket-specific stats
2852 * @xprt: rpc_xprt struct containing statistics
2856 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2858 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2860 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2863 xprt->stat.bind_count,
2866 xprt->stat.bad_xids,
2869 xprt->stat.max_slots,
2870 xprt->stat.sending_u,
2871 xprt->stat.pending_u);
2875 * xs_tcp_print_stats - display TCP socket-specific stats
2876 * @xprt: rpc_xprt struct containing statistics
2880 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2882 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2885 if (xprt_connected(xprt))
2886 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2888 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2889 "%llu %llu %lu %llu %llu\n",
2891 xprt->stat.bind_count,
2892 xprt->stat.connect_count,
2893 xprt->stat.connect_time / HZ,
2897 xprt->stat.bad_xids,
2900 xprt->stat.max_slots,
2901 xprt->stat.sending_u,
2902 xprt->stat.pending_u);
2906 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2907 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2908 * to use the server side send routines.
2910 static int bc_malloc(struct rpc_task *task)
2912 struct rpc_rqst *rqst = task->tk_rqstp;
2913 size_t size = rqst->rq_callsize;
2915 struct rpc_buffer *buf;
2917 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2918 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2923 page = alloc_page(GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN);
2927 buf = page_address(page);
2928 buf->len = PAGE_SIZE;
2930 rqst->rq_buffer = buf->data;
2931 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2936 * Free the space allocated in the bc_alloc routine
2938 static void bc_free(struct rpc_task *task)
2940 void *buffer = task->tk_rqstp->rq_buffer;
2941 struct rpc_buffer *buf;
2943 buf = container_of(buffer, struct rpc_buffer, data);
2944 free_page((unsigned long)buf);
2947 static int bc_sendto(struct rpc_rqst *req)
2949 struct xdr_buf *xdr = &req->rq_snd_buf;
2950 struct sock_xprt *transport =
2951 container_of(req->rq_xprt, struct sock_xprt, xprt);
2952 struct msghdr msg = {
2955 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
2957 unsigned int sent = 0;
2960 req->rq_xtime = ktime_get();
2961 err = xdr_alloc_bvec(xdr, rpc_task_gfp_mask());
2964 err = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, marker, &sent);
2966 if (err < 0 || sent != (xdr->len + sizeof(marker)))
2972 * bc_send_request - Send a backchannel Call on a TCP socket
2973 * @req: rpc_rqst containing Call message to be sent
2975 * xpt_mutex ensures @rqstp's whole message is written to the socket
2976 * without interruption.
2979 * %0 if the message was sent successfully
2980 * %ENOTCONN if the message was not sent
2982 static int bc_send_request(struct rpc_rqst *req)
2984 struct svc_xprt *xprt;
2988 * Get the server socket associated with this callback xprt
2990 xprt = req->rq_xprt->bc_xprt;
2993 * Grab the mutex to serialize data as the connection is shared
2994 * with the fore channel
2996 mutex_lock(&xprt->xpt_mutex);
2997 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
3000 len = bc_sendto(req);
3001 mutex_unlock(&xprt->xpt_mutex);
3009 static void bc_close(struct rpc_xprt *xprt)
3011 xprt_disconnect_done(xprt);
3014 static void bc_destroy(struct rpc_xprt *xprt)
3016 dprintk("RPC: bc_destroy xprt %p\n", xprt);
3019 module_put(THIS_MODULE);
3022 static const struct rpc_xprt_ops xs_local_ops = {
3023 .reserve_xprt = xprt_reserve_xprt,
3024 .release_xprt = xprt_release_xprt,
3025 .alloc_slot = xprt_alloc_slot,
3026 .free_slot = xprt_free_slot,
3027 .rpcbind = xs_local_rpcbind,
3028 .set_port = xs_local_set_port,
3029 .connect = xs_local_connect,
3030 .buf_alloc = rpc_malloc,
3031 .buf_free = rpc_free,
3032 .prepare_request = xs_stream_prepare_request,
3033 .send_request = xs_local_send_request,
3034 .abort_send_request = xs_stream_abort_send_request,
3035 .wait_for_reply_request = xprt_wait_for_reply_request_def,
3037 .destroy = xs_destroy,
3038 .print_stats = xs_local_print_stats,
3039 .enable_swap = xs_enable_swap,
3040 .disable_swap = xs_disable_swap,
3043 static const struct rpc_xprt_ops xs_udp_ops = {
3044 .set_buffer_size = xs_udp_set_buffer_size,
3045 .reserve_xprt = xprt_reserve_xprt_cong,
3046 .release_xprt = xprt_release_xprt_cong,
3047 .alloc_slot = xprt_alloc_slot,
3048 .free_slot = xprt_free_slot,
3049 .rpcbind = rpcb_getport_async,
3050 .set_port = xs_set_port,
3051 .connect = xs_connect,
3052 .get_srcaddr = xs_sock_srcaddr,
3053 .get_srcport = xs_sock_srcport,
3054 .buf_alloc = rpc_malloc,
3055 .buf_free = rpc_free,
3056 .send_request = xs_udp_send_request,
3057 .wait_for_reply_request = xprt_wait_for_reply_request_rtt,
3058 .timer = xs_udp_timer,
3059 .release_request = xprt_release_rqst_cong,
3061 .destroy = xs_destroy,
3062 .print_stats = xs_udp_print_stats,
3063 .enable_swap = xs_enable_swap,
3064 .disable_swap = xs_disable_swap,
3065 .inject_disconnect = xs_inject_disconnect,
3068 static const struct rpc_xprt_ops xs_tcp_ops = {
3069 .reserve_xprt = xprt_reserve_xprt,
3070 .release_xprt = xprt_release_xprt,
3071 .alloc_slot = xprt_alloc_slot,
3072 .free_slot = xprt_free_slot,
3073 .rpcbind = rpcb_getport_async,
3074 .set_port = xs_set_port,
3075 .connect = xs_connect,
3076 .get_srcaddr = xs_sock_srcaddr,
3077 .get_srcport = xs_sock_srcport,
3078 .buf_alloc = rpc_malloc,
3079 .buf_free = rpc_free,
3080 .prepare_request = xs_stream_prepare_request,
3081 .send_request = xs_tcp_send_request,
3082 .abort_send_request = xs_stream_abort_send_request,
3083 .wait_for_reply_request = xprt_wait_for_reply_request_def,
3084 .close = xs_tcp_shutdown,
3085 .destroy = xs_destroy,
3086 .set_connect_timeout = xs_tcp_set_connect_timeout,
3087 .print_stats = xs_tcp_print_stats,
3088 .enable_swap = xs_enable_swap,
3089 .disable_swap = xs_disable_swap,
3090 .inject_disconnect = xs_inject_disconnect,
3091 #ifdef CONFIG_SUNRPC_BACKCHANNEL
3092 .bc_setup = xprt_setup_bc,
3093 .bc_maxpayload = xs_tcp_bc_maxpayload,
3094 .bc_num_slots = xprt_bc_max_slots,
3095 .bc_free_rqst = xprt_free_bc_rqst,
3096 .bc_destroy = xprt_destroy_bc,
3101 * The rpc_xprt_ops for the server backchannel
3104 static const struct rpc_xprt_ops bc_tcp_ops = {
3105 .reserve_xprt = xprt_reserve_xprt,
3106 .release_xprt = xprt_release_xprt,
3107 .alloc_slot = xprt_alloc_slot,
3108 .free_slot = xprt_free_slot,
3109 .buf_alloc = bc_malloc,
3110 .buf_free = bc_free,
3111 .send_request = bc_send_request,
3112 .wait_for_reply_request = xprt_wait_for_reply_request_def,
3114 .destroy = bc_destroy,
3115 .print_stats = xs_tcp_print_stats,
3116 .enable_swap = xs_enable_swap,
3117 .disable_swap = xs_disable_swap,
3118 .inject_disconnect = xs_inject_disconnect,
3121 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
3123 static const struct sockaddr_in sin = {
3124 .sin_family = AF_INET,
3125 .sin_addr.s_addr = htonl(INADDR_ANY),
3127 static const struct sockaddr_in6 sin6 = {
3128 .sin6_family = AF_INET6,
3129 .sin6_addr = IN6ADDR_ANY_INIT,
3136 memcpy(sap, &sin, sizeof(sin));
3139 memcpy(sap, &sin6, sizeof(sin6));
3142 dprintk("RPC: %s: Bad address family\n", __func__);
3143 return -EAFNOSUPPORT;
3148 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
3149 unsigned int slot_table_size,
3150 unsigned int max_slot_table_size)
3152 struct rpc_xprt *xprt;
3153 struct sock_xprt *new;
3155 if (args->addrlen > sizeof(xprt->addr)) {
3156 dprintk("RPC: xs_setup_xprt: address too large\n");
3157 return ERR_PTR(-EBADF);
3160 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
3161 max_slot_table_size);
3163 dprintk("RPC: xs_setup_xprt: couldn't allocate "
3165 return ERR_PTR(-ENOMEM);
3168 new = container_of(xprt, struct sock_xprt, xprt);
3169 mutex_init(&new->recv_mutex);
3170 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
3171 xprt->addrlen = args->addrlen;
3173 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
3176 err = xs_init_anyaddr(args->dstaddr->sa_family,
3177 (struct sockaddr *)&new->srcaddr);
3180 return ERR_PTR(err);
3187 static const struct rpc_timeout xs_local_default_timeout = {
3188 .to_initval = 10 * HZ,
3189 .to_maxval = 10 * HZ,
3194 * xs_setup_local - Set up transport to use an AF_LOCAL socket
3195 * @args: rpc transport creation arguments
3197 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
3199 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
3201 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
3202 struct sock_xprt *transport;
3203 struct rpc_xprt *xprt;
3204 struct rpc_xprt *ret;
3206 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3207 xprt_max_tcp_slot_table_entries);
3210 transport = container_of(xprt, struct sock_xprt, xprt);
3213 xprt->xprt_class = &xs_local_transport;
3214 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3216 xprt->bind_timeout = XS_BIND_TO;
3217 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3218 xprt->idle_timeout = XS_IDLE_DISC_TO;
3220 xprt->ops = &xs_local_ops;
3221 xprt->timeout = &xs_local_default_timeout;
3223 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3224 INIT_WORK(&transport->error_worker, xs_error_handle);
3225 INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket);
3227 switch (sun->sun_family) {
3229 if (sun->sun_path[0] != '/' && sun->sun_path[0] != '\0') {
3230 dprintk("RPC: bad AF_LOCAL address: %s\n",
3232 ret = ERR_PTR(-EINVAL);
3235 xprt_set_bound(xprt);
3236 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
3239 ret = ERR_PTR(-EAFNOSUPPORT);
3243 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
3244 xprt->address_strings[RPC_DISPLAY_ADDR]);
3246 if (try_module_get(THIS_MODULE))
3248 ret = ERR_PTR(-EINVAL);
3254 static const struct rpc_timeout xs_udp_default_timeout = {
3255 .to_initval = 5 * HZ,
3256 .to_maxval = 30 * HZ,
3257 .to_increment = 5 * HZ,
3262 * xs_setup_udp - Set up transport to use a UDP socket
3263 * @args: rpc transport creation arguments
3266 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
3268 struct sockaddr *addr = args->dstaddr;
3269 struct rpc_xprt *xprt;
3270 struct sock_xprt *transport;
3271 struct rpc_xprt *ret;
3273 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
3274 xprt_udp_slot_table_entries);
3277 transport = container_of(xprt, struct sock_xprt, xprt);
3279 xprt->prot = IPPROTO_UDP;
3280 xprt->xprt_class = &xs_udp_transport;
3281 /* XXX: header size can vary due to auth type, IPv6, etc. */
3282 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
3284 xprt->bind_timeout = XS_BIND_TO;
3285 xprt->reestablish_timeout = XS_UDP_REEST_TO;
3286 xprt->idle_timeout = XS_IDLE_DISC_TO;
3288 xprt->ops = &xs_udp_ops;
3290 xprt->timeout = &xs_udp_default_timeout;
3292 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
3293 INIT_WORK(&transport->error_worker, xs_error_handle);
3294 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
3296 switch (addr->sa_family) {
3298 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3299 xprt_set_bound(xprt);
3301 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
3304 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3305 xprt_set_bound(xprt);
3307 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
3310 ret = ERR_PTR(-EAFNOSUPPORT);
3314 if (xprt_bound(xprt))
3315 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3316 xprt->address_strings[RPC_DISPLAY_ADDR],
3317 xprt->address_strings[RPC_DISPLAY_PORT],
3318 xprt->address_strings[RPC_DISPLAY_PROTO]);
3320 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
3321 xprt->address_strings[RPC_DISPLAY_ADDR],
3322 xprt->address_strings[RPC_DISPLAY_PROTO]);
3324 if (try_module_get(THIS_MODULE))
3326 ret = ERR_PTR(-EINVAL);
3332 static const struct rpc_timeout xs_tcp_default_timeout = {
3333 .to_initval = 60 * HZ,
3334 .to_maxval = 60 * HZ,
3339 * xs_setup_tcp - Set up transport to use a TCP socket
3340 * @args: rpc transport creation arguments
3343 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
3345 struct sockaddr *addr = args->dstaddr;
3346 struct rpc_xprt *xprt;
3347 struct sock_xprt *transport;
3348 struct rpc_xprt *ret;
3349 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
3351 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
3352 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
3354 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3355 max_slot_table_size);
3358 transport = container_of(xprt, struct sock_xprt, xprt);
3360 xprt->prot = IPPROTO_TCP;
3361 xprt->xprt_class = &xs_tcp_transport;
3362 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3364 xprt->bind_timeout = XS_BIND_TO;
3365 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3366 xprt->idle_timeout = XS_IDLE_DISC_TO;
3368 xprt->ops = &xs_tcp_ops;
3369 xprt->timeout = &xs_tcp_default_timeout;
3371 xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3372 if (args->reconnect_timeout)
3373 xprt->max_reconnect_timeout = args->reconnect_timeout;
3375 xprt->connect_timeout = xprt->timeout->to_initval *
3376 (xprt->timeout->to_retries + 1);
3377 if (args->connect_timeout)
3378 xs_tcp_do_set_connect_timeout(xprt, args->connect_timeout);
3380 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3381 INIT_WORK(&transport->error_worker, xs_error_handle);
3382 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
3384 switch (addr->sa_family) {
3386 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3387 xprt_set_bound(xprt);
3389 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3392 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3393 xprt_set_bound(xprt);
3395 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3398 ret = ERR_PTR(-EAFNOSUPPORT);
3402 if (xprt_bound(xprt))
3403 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3404 xprt->address_strings[RPC_DISPLAY_ADDR],
3405 xprt->address_strings[RPC_DISPLAY_PORT],
3406 xprt->address_strings[RPC_DISPLAY_PROTO]);
3408 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
3409 xprt->address_strings[RPC_DISPLAY_ADDR],
3410 xprt->address_strings[RPC_DISPLAY_PROTO]);
3412 if (try_module_get(THIS_MODULE))
3414 ret = ERR_PTR(-EINVAL);
3421 * xs_setup_tcp_tls - Set up transport to use a TCP with TLS
3422 * @args: rpc transport creation arguments
3425 static struct rpc_xprt *xs_setup_tcp_tls(struct xprt_create *args)
3427 struct sockaddr *addr = args->dstaddr;
3428 struct rpc_xprt *xprt;
3429 struct sock_xprt *transport;
3430 struct rpc_xprt *ret;
3431 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
3433 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
3434 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
3436 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3437 max_slot_table_size);
3440 transport = container_of(xprt, struct sock_xprt, xprt);
3442 xprt->prot = IPPROTO_TCP;
3443 xprt->xprt_class = &xs_tcp_transport;
3444 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3446 xprt->bind_timeout = XS_BIND_TO;
3447 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3448 xprt->idle_timeout = XS_IDLE_DISC_TO;
3450 xprt->ops = &xs_tcp_ops;
3451 xprt->timeout = &xs_tcp_default_timeout;
3453 xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3454 xprt->connect_timeout = xprt->timeout->to_initval *
3455 (xprt->timeout->to_retries + 1);
3457 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3458 INIT_WORK(&transport->error_worker, xs_error_handle);
3460 switch (args->xprtsec.policy) {
3461 case RPC_XPRTSEC_TLS_ANON:
3462 case RPC_XPRTSEC_TLS_X509:
3463 xprt->xprtsec = args->xprtsec;
3464 INIT_DELAYED_WORK(&transport->connect_worker,
3465 xs_tcp_tls_setup_socket);
3468 ret = ERR_PTR(-EACCES);
3472 switch (addr->sa_family) {
3474 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3475 xprt_set_bound(xprt);
3477 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3480 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3481 xprt_set_bound(xprt);
3483 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3486 ret = ERR_PTR(-EAFNOSUPPORT);
3490 if (xprt_bound(xprt))
3491 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3492 xprt->address_strings[RPC_DISPLAY_ADDR],
3493 xprt->address_strings[RPC_DISPLAY_PORT],
3494 xprt->address_strings[RPC_DISPLAY_PROTO]);
3496 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
3497 xprt->address_strings[RPC_DISPLAY_ADDR],
3498 xprt->address_strings[RPC_DISPLAY_PROTO]);
3500 if (try_module_get(THIS_MODULE))
3502 ret = ERR_PTR(-EINVAL);
3509 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3510 * @args: rpc transport creation arguments
3513 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3515 struct sockaddr *addr = args->dstaddr;
3516 struct rpc_xprt *xprt;
3517 struct sock_xprt *transport;
3518 struct svc_sock *bc_sock;
3519 struct rpc_xprt *ret;
3521 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3522 xprt_tcp_slot_table_entries);
3525 transport = container_of(xprt, struct sock_xprt, xprt);
3527 xprt->prot = IPPROTO_TCP;
3528 xprt->xprt_class = &xs_bc_tcp_transport;
3529 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3530 xprt->timeout = &xs_tcp_default_timeout;
3533 xprt_set_bound(xprt);
3534 xprt->bind_timeout = 0;
3535 xprt->reestablish_timeout = 0;
3536 xprt->idle_timeout = 0;
3538 xprt->ops = &bc_tcp_ops;
3540 switch (addr->sa_family) {
3542 xs_format_peer_addresses(xprt, "tcp",
3546 xs_format_peer_addresses(xprt, "tcp",
3547 RPCBIND_NETID_TCP6);
3550 ret = ERR_PTR(-EAFNOSUPPORT);
3554 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3555 xprt->address_strings[RPC_DISPLAY_ADDR],
3556 xprt->address_strings[RPC_DISPLAY_PORT],
3557 xprt->address_strings[RPC_DISPLAY_PROTO]);
3560 * Once we've associated a backchannel xprt with a connection,
3561 * we want to keep it around as long as the connection lasts,
3562 * in case we need to start using it for a backchannel again;
3563 * this reference won't be dropped until bc_xprt is destroyed.
3566 args->bc_xprt->xpt_bc_xprt = xprt;
3567 xprt->bc_xprt = args->bc_xprt;
3568 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3569 transport->sock = bc_sock->sk_sock;
3570 transport->inet = bc_sock->sk_sk;
3573 * Since we don't want connections for the backchannel, we set
3574 * the xprt status to connected
3576 xprt_set_connected(xprt);
3578 if (try_module_get(THIS_MODULE))
3581 args->bc_xprt->xpt_bc_xprt = NULL;
3582 args->bc_xprt->xpt_bc_xps = NULL;
3584 ret = ERR_PTR(-EINVAL);
3590 static struct xprt_class xs_local_transport = {
3591 .list = LIST_HEAD_INIT(xs_local_transport.list),
3592 .name = "named UNIX socket",
3593 .owner = THIS_MODULE,
3594 .ident = XPRT_TRANSPORT_LOCAL,
3595 .setup = xs_setup_local,
3599 static struct xprt_class xs_udp_transport = {
3600 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3602 .owner = THIS_MODULE,
3603 .ident = XPRT_TRANSPORT_UDP,
3604 .setup = xs_setup_udp,
3605 .netid = { "udp", "udp6", "" },
3608 static struct xprt_class xs_tcp_transport = {
3609 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3611 .owner = THIS_MODULE,
3612 .ident = XPRT_TRANSPORT_TCP,
3613 .setup = xs_setup_tcp,
3614 .netid = { "tcp", "tcp6", "" },
3617 static struct xprt_class xs_tcp_tls_transport = {
3618 .list = LIST_HEAD_INIT(xs_tcp_tls_transport.list),
3619 .name = "tcp-with-tls",
3620 .owner = THIS_MODULE,
3621 .ident = XPRT_TRANSPORT_TCP_TLS,
3622 .setup = xs_setup_tcp_tls,
3623 .netid = { "tcp", "tcp6", "" },
3626 static struct xprt_class xs_bc_tcp_transport = {
3627 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3628 .name = "tcp NFSv4.1 backchannel",
3629 .owner = THIS_MODULE,
3630 .ident = XPRT_TRANSPORT_BC_TCP,
3631 .setup = xs_setup_bc_tcp,
3636 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3639 int init_socket_xprt(void)
3641 if (!sunrpc_table_header)
3642 sunrpc_table_header = register_sysctl("sunrpc", xs_tunables_table);
3644 xprt_register_transport(&xs_local_transport);
3645 xprt_register_transport(&xs_udp_transport);
3646 xprt_register_transport(&xs_tcp_transport);
3647 xprt_register_transport(&xs_tcp_tls_transport);
3648 xprt_register_transport(&xs_bc_tcp_transport);
3654 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3657 void cleanup_socket_xprt(void)
3659 if (sunrpc_table_header) {
3660 unregister_sysctl_table(sunrpc_table_header);
3661 sunrpc_table_header = NULL;
3664 xprt_unregister_transport(&xs_local_transport);
3665 xprt_unregister_transport(&xs_udp_transport);
3666 xprt_unregister_transport(&xs_tcp_transport);
3667 xprt_unregister_transport(&xs_tcp_tls_transport);
3668 xprt_unregister_transport(&xs_bc_tcp_transport);
3671 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3673 return param_set_uint_minmax(val, kp,
3678 static const struct kernel_param_ops param_ops_portnr = {
3679 .set = param_set_portnr,
3680 .get = param_get_uint,
3683 #define param_check_portnr(name, p) \
3684 __param_check(name, p, unsigned int);
3686 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3687 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3689 static int param_set_slot_table_size(const char *val,
3690 const struct kernel_param *kp)
3692 return param_set_uint_minmax(val, kp,
3694 RPC_MAX_SLOT_TABLE);
3697 static const struct kernel_param_ops param_ops_slot_table_size = {
3698 .set = param_set_slot_table_size,
3699 .get = param_get_uint,
3702 #define param_check_slot_table_size(name, p) \
3703 __param_check(name, p, unsigned int);
3705 static int param_set_max_slot_table_size(const char *val,
3706 const struct kernel_param *kp)
3708 return param_set_uint_minmax(val, kp,
3710 RPC_MAX_SLOT_TABLE_LIMIT);
3713 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3714 .set = param_set_max_slot_table_size,
3715 .get = param_get_uint,
3718 #define param_check_max_slot_table_size(name, p) \
3719 __param_check(name, p, unsigned int);
3721 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3722 slot_table_size, 0644);
3723 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3724 max_slot_table_size, 0644);
3725 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3726 slot_table_size, 0644);