1 // SPDX-License-Identifier: GPL-2.0
3 * linux/net/sunrpc/xprtsock.c
5 * Client-side transport implementation for sockets.
7 * TCP callback races fixes (C) 1998 Red Hat
8 * TCP send fixes (C) 1998 Red Hat
9 * TCP NFS related read + write fixes
10 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
12 * Rewrite of larges part of the code in order to stabilize TCP stuff.
13 * Fix behaviour when socket buffer is full.
14 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
16 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
18 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
19 * <gilles.quillard@bull.net>
22 #include <linux/types.h>
23 #include <linux/string.h>
24 #include <linux/slab.h>
25 #include <linux/module.h>
26 #include <linux/capability.h>
27 #include <linux/pagemap.h>
28 #include <linux/errno.h>
29 #include <linux/socket.h>
31 #include <linux/net.h>
34 #include <linux/udp.h>
35 #include <linux/tcp.h>
36 #include <linux/sunrpc/clnt.h>
37 #include <linux/sunrpc/addr.h>
38 #include <linux/sunrpc/sched.h>
39 #include <linux/sunrpc/svcsock.h>
40 #include <linux/sunrpc/xprtsock.h>
41 #include <linux/file.h>
42 #ifdef CONFIG_SUNRPC_BACKCHANNEL
43 #include <linux/sunrpc/bc_xprt.h>
47 #include <net/checksum.h>
50 #include <linux/bvec.h>
51 #include <linux/highmem.h>
52 #include <linux/uio.h>
53 #include <linux/sched/mm.h>
55 #include <trace/events/sunrpc.h>
59 static void xs_close(struct rpc_xprt *xprt);
60 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
66 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
67 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
68 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
70 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
71 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
73 #define XS_TCP_LINGER_TO (15U * HZ)
74 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
77 * We can register our own files under /proc/sys/sunrpc by
78 * calling register_sysctl_table() again. The files in that
79 * directory become the union of all files registered there.
81 * We simply need to make sure that we don't collide with
82 * someone else's file names!
85 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
86 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
87 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
88 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
89 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
91 static struct ctl_table_header *sunrpc_table_header;
94 * FIXME: changing the UDP slot table size should also resize the UDP
95 * socket buffers for existing UDP transports
97 static struct ctl_table xs_tunables_table[] = {
99 .procname = "udp_slot_table_entries",
100 .data = &xprt_udp_slot_table_entries,
101 .maxlen = sizeof(unsigned int),
103 .proc_handler = proc_dointvec_minmax,
104 .extra1 = &min_slot_table_size,
105 .extra2 = &max_slot_table_size
108 .procname = "tcp_slot_table_entries",
109 .data = &xprt_tcp_slot_table_entries,
110 .maxlen = sizeof(unsigned int),
112 .proc_handler = proc_dointvec_minmax,
113 .extra1 = &min_slot_table_size,
114 .extra2 = &max_slot_table_size
117 .procname = "tcp_max_slot_table_entries",
118 .data = &xprt_max_tcp_slot_table_entries,
119 .maxlen = sizeof(unsigned int),
121 .proc_handler = proc_dointvec_minmax,
122 .extra1 = &min_slot_table_size,
123 .extra2 = &max_tcp_slot_table_limit
126 .procname = "min_resvport",
127 .data = &xprt_min_resvport,
128 .maxlen = sizeof(unsigned int),
130 .proc_handler = proc_dointvec_minmax,
131 .extra1 = &xprt_min_resvport_limit,
132 .extra2 = &xprt_max_resvport_limit
135 .procname = "max_resvport",
136 .data = &xprt_max_resvport,
137 .maxlen = sizeof(unsigned int),
139 .proc_handler = proc_dointvec_minmax,
140 .extra1 = &xprt_min_resvport_limit,
141 .extra2 = &xprt_max_resvport_limit
144 .procname = "tcp_fin_timeout",
145 .data = &xs_tcp_fin_timeout,
146 .maxlen = sizeof(xs_tcp_fin_timeout),
148 .proc_handler = proc_dointvec_jiffies,
153 static struct ctl_table sunrpc_table[] = {
155 .procname = "sunrpc",
157 .child = xs_tunables_table
163 * Wait duration for a reply from the RPC portmapper.
165 #define XS_BIND_TO (60U * HZ)
168 * Delay if a UDP socket connect error occurs. This is most likely some
169 * kind of resource problem on the local host.
171 #define XS_UDP_REEST_TO (2U * HZ)
174 * The reestablish timeout allows clients to delay for a bit before attempting
175 * to reconnect to a server that just dropped our connection.
177 * We implement an exponential backoff when trying to reestablish a TCP
178 * transport connection with the server. Some servers like to drop a TCP
179 * connection when they are overworked, so we start with a short timeout and
180 * increase over time if the server is down or not responding.
182 #define XS_TCP_INIT_REEST_TO (3U * HZ)
185 * TCP idle timeout; client drops the transport socket if it is idle
186 * for this long. Note that we also timeout UDP sockets to prevent
187 * holding port numbers when there is no RPC traffic.
189 #define XS_IDLE_DISC_TO (5U * 60 * HZ)
191 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
192 # undef RPC_DEBUG_DATA
193 # define RPCDBG_FACILITY RPCDBG_TRANS
196 #ifdef RPC_DEBUG_DATA
197 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
199 u8 *buf = (u8 *) packet;
202 dprintk("RPC: %s\n", msg);
203 for (j = 0; j < count && j < 128; j += 4) {
207 dprintk("0x%04x ", j);
209 dprintk("%02x%02x%02x%02x ",
210 buf[j], buf[j+1], buf[j+2], buf[j+3]);
215 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
221 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
223 return (struct rpc_xprt *) sk->sk_user_data;
226 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
228 return (struct sockaddr *) &xprt->addr;
231 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
233 return (struct sockaddr_un *) &xprt->addr;
236 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
238 return (struct sockaddr_in *) &xprt->addr;
241 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
243 return (struct sockaddr_in6 *) &xprt->addr;
246 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
248 struct sockaddr *sap = xs_addr(xprt);
249 struct sockaddr_in6 *sin6;
250 struct sockaddr_in *sin;
251 struct sockaddr_un *sun;
254 switch (sap->sa_family) {
256 sun = xs_addr_un(xprt);
257 strlcpy(buf, sun->sun_path, sizeof(buf));
258 xprt->address_strings[RPC_DISPLAY_ADDR] =
259 kstrdup(buf, GFP_KERNEL);
262 (void)rpc_ntop(sap, buf, sizeof(buf));
263 xprt->address_strings[RPC_DISPLAY_ADDR] =
264 kstrdup(buf, GFP_KERNEL);
265 sin = xs_addr_in(xprt);
266 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
269 (void)rpc_ntop(sap, buf, sizeof(buf));
270 xprt->address_strings[RPC_DISPLAY_ADDR] =
271 kstrdup(buf, GFP_KERNEL);
272 sin6 = xs_addr_in6(xprt);
273 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
279 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
282 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
284 struct sockaddr *sap = xs_addr(xprt);
287 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
288 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
290 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
291 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
294 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
295 const char *protocol,
298 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
299 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
300 xs_format_common_peer_addresses(xprt);
301 xs_format_common_peer_ports(xprt);
304 static void xs_update_peer_port(struct rpc_xprt *xprt)
306 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
307 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
309 xs_format_common_peer_ports(xprt);
312 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
316 for (i = 0; i < RPC_DISPLAY_MAX; i++)
318 case RPC_DISPLAY_PROTO:
319 case RPC_DISPLAY_NETID:
322 kfree(xprt->address_strings[i]);
327 xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp)
331 if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES))
333 n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT;
334 for (i = 0; i < n; i++) {
337 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp);
338 if (!buf->pages[i]) {
340 return i > buf->page_base ? i - buf->page_base : 0;
347 xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek)
351 iov_iter_advance(&msg->msg_iter, seek);
352 ret = sock_recvmsg(sock, msg, flags);
353 return ret > 0 ? ret + seek : ret;
357 xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags,
358 struct kvec *kvec, size_t count, size_t seek)
360 iov_iter_kvec(&msg->msg_iter, READ, kvec, 1, count);
361 return xs_sock_recvmsg(sock, msg, flags, seek);
365 xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags,
366 struct bio_vec *bvec, unsigned long nr, size_t count,
369 iov_iter_bvec(&msg->msg_iter, READ, bvec, nr, count);
370 return xs_sock_recvmsg(sock, msg, flags, seek);
374 xs_read_discard(struct socket *sock, struct msghdr *msg, int flags,
377 iov_iter_discard(&msg->msg_iter, READ, count);
378 return sock_recvmsg(sock, msg, flags);
381 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
383 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
385 struct bvec_iter bi = {
390 bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
391 for_each_bvec(bv, bvec, bi, bi)
392 flush_dcache_page(bv.bv_page);
396 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
402 xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
403 struct xdr_buf *buf, size_t count, size_t seek, size_t *read)
405 size_t want, seek_init = seek, offset = 0;
408 want = min_t(size_t, count, buf->head[0].iov_len);
410 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek);
414 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
424 want = xs_alloc_sparse_pages(buf,
425 min_t(size_t, count - offset, buf->page_len),
428 ret = xs_read_bvec(sock, msg, flags, buf->bvec,
429 xdr_buf_pagecount(buf),
430 want + buf->page_base,
431 seek + buf->page_base);
434 xs_flush_bvec(buf->bvec, ret, seek + buf->page_base);
435 offset += ret - buf->page_base;
436 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
446 want = min_t(size_t, count - offset, buf->tail[0].iov_len);
448 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek);
452 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
456 } else if (offset < seek_init)
460 *read = offset - seek_init;
468 xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf)
470 if (!transport->recv.copied) {
471 if (buf->head[0].iov_len >= transport->recv.offset)
472 memcpy(buf->head[0].iov_base,
473 &transport->recv.xid,
474 transport->recv.offset);
475 transport->recv.copied = transport->recv.offset;
480 xs_read_stream_request_done(struct sock_xprt *transport)
482 return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT);
486 xs_read_stream_check_eor(struct sock_xprt *transport,
489 if (xs_read_stream_request_done(transport))
490 msg->msg_flags |= MSG_EOR;
494 xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg,
495 int flags, struct rpc_rqst *req)
497 struct xdr_buf *buf = &req->rq_private_buf;
501 xs_read_header(transport, buf);
503 want = transport->recv.len - transport->recv.offset;
505 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf,
506 transport->recv.copied + want,
507 transport->recv.copied,
509 transport->recv.offset += read;
510 transport->recv.copied += read;
513 if (transport->recv.offset == transport->recv.len)
514 xs_read_stream_check_eor(transport, msg);
524 msg->msg_flags |= MSG_TRUNC;
529 return ret < 0 ? ret : read;
533 xs_read_stream_headersize(bool isfrag)
536 return sizeof(__be32);
537 return 3 * sizeof(__be32);
541 xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg,
542 int flags, size_t want, size_t seek)
545 .iov_base = &transport->recv.fraghdr,
548 return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek);
551 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
553 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
555 struct rpc_xprt *xprt = &transport->xprt;
556 struct rpc_rqst *req;
559 /* Look up and lock the request corresponding to the given XID */
560 req = xprt_lookup_bc_request(xprt, transport->recv.xid);
562 printk(KERN_WARNING "Callback slot table overflowed\n");
566 ret = xs_read_stream_request(transport, msg, flags, req);
567 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
568 xprt_complete_bc_request(req, transport->recv.copied);
572 #else /* CONFIG_SUNRPC_BACKCHANNEL */
574 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
578 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
581 xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags)
583 struct rpc_xprt *xprt = &transport->xprt;
584 struct rpc_rqst *req;
587 /* Look up and lock the request corresponding to the given XID */
588 spin_lock(&xprt->queue_lock);
589 req = xprt_lookup_rqst(xprt, transport->recv.xid);
591 msg->msg_flags |= MSG_TRUNC;
595 spin_unlock(&xprt->queue_lock);
597 ret = xs_read_stream_request(transport, msg, flags, req);
599 spin_lock(&xprt->queue_lock);
600 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
601 xprt_complete_rqst(req->rq_task, transport->recv.copied);
602 xprt_unpin_rqst(req);
604 spin_unlock(&xprt->queue_lock);
609 xs_read_stream(struct sock_xprt *transport, int flags)
611 struct msghdr msg = { 0 };
612 size_t want, read = 0;
615 if (transport->recv.len == 0) {
616 want = xs_read_stream_headersize(transport->recv.copied != 0);
617 ret = xs_read_stream_header(transport, &msg, flags, want,
618 transport->recv.offset);
621 transport->recv.offset = ret;
622 if (transport->recv.offset != want)
623 return transport->recv.offset;
624 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) &
625 RPC_FRAGMENT_SIZE_MASK;
626 transport->recv.offset -= sizeof(transport->recv.fraghdr);
630 switch (be32_to_cpu(transport->recv.calldir)) {
632 msg.msg_flags |= MSG_TRUNC;
635 ret = xs_read_stream_call(transport, &msg, flags);
638 ret = xs_read_stream_reply(transport, &msg, flags);
640 if (msg.msg_flags & MSG_TRUNC) {
641 transport->recv.calldir = cpu_to_be32(-1);
642 transport->recv.copied = -1;
647 if (transport->recv.offset < transport->recv.len) {
648 if (!(msg.msg_flags & MSG_TRUNC))
651 ret = xs_read_discard(transport->sock, &msg, flags,
652 transport->recv.len - transport->recv.offset);
655 transport->recv.offset += ret;
657 if (transport->recv.offset != transport->recv.len)
660 if (xs_read_stream_request_done(transport)) {
661 trace_xs_stream_read_request(transport);
662 transport->recv.copied = 0;
664 transport->recv.offset = 0;
665 transport->recv.len = 0;
668 return ret != 0 ? ret : -ESHUTDOWN;
671 static __poll_t xs_poll_socket(struct sock_xprt *transport)
673 return transport->sock->ops->poll(transport->file, transport->sock,
677 static bool xs_poll_socket_readable(struct sock_xprt *transport)
679 __poll_t events = xs_poll_socket(transport);
681 return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP);
684 static void xs_poll_check_readable(struct sock_xprt *transport)
687 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
688 if (!xs_poll_socket_readable(transport))
690 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
691 queue_work(xprtiod_workqueue, &transport->recv_worker);
694 static void xs_stream_data_receive(struct sock_xprt *transport)
699 mutex_lock(&transport->recv_mutex);
700 if (transport->sock == NULL)
703 ret = xs_read_stream(transport, MSG_DONTWAIT);
709 if (ret == -ESHUTDOWN)
710 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
712 xs_poll_check_readable(transport);
714 mutex_unlock(&transport->recv_mutex);
715 trace_xs_stream_read_data(&transport->xprt, ret, read);
718 static void xs_stream_data_receive_workfn(struct work_struct *work)
720 struct sock_xprt *transport =
721 container_of(work, struct sock_xprt, recv_worker);
722 unsigned int pflags = memalloc_nofs_save();
724 xs_stream_data_receive(transport);
725 memalloc_nofs_restore(pflags);
729 xs_stream_reset_connect(struct sock_xprt *transport)
731 transport->recv.offset = 0;
732 transport->recv.len = 0;
733 transport->recv.copied = 0;
734 transport->xmit.offset = 0;
738 xs_stream_start_connect(struct sock_xprt *transport)
740 transport->xprt.stat.connect_count++;
741 transport->xprt.stat.connect_start = jiffies;
744 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
746 static int xs_sendmsg(struct socket *sock, struct msghdr *msg, size_t seek)
749 iov_iter_advance(&msg->msg_iter, seek);
750 return sock_sendmsg(sock, msg);
753 static int xs_send_kvec(struct socket *sock, struct msghdr *msg, struct kvec *vec, size_t seek)
755 iov_iter_kvec(&msg->msg_iter, WRITE, vec, 1, vec->iov_len);
756 return xs_sendmsg(sock, msg, seek);
759 static int xs_send_pagedata(struct socket *sock, struct msghdr *msg, struct xdr_buf *xdr, size_t base)
763 err = xdr_alloc_bvec(xdr, GFP_KERNEL);
767 iov_iter_bvec(&msg->msg_iter, WRITE, xdr->bvec,
768 xdr_buf_pagecount(xdr),
769 xdr->page_len + xdr->page_base);
770 return xs_sendmsg(sock, msg, base + xdr->page_base);
773 #define xs_record_marker_len() sizeof(rpc_fraghdr)
777 * - sending from byte 0 of the message
778 * - the message is wholly contained in @xdr's head iovec
780 static int xs_send_rm_and_kvec(struct socket *sock, struct msghdr *msg,
781 rpc_fraghdr marker, struct kvec *vec, size_t base)
783 struct kvec iov[2] = {
786 .iov_len = sizeof(marker)
790 size_t len = iov[0].iov_len + iov[1].iov_len;
792 iov_iter_kvec(&msg->msg_iter, WRITE, iov, 2, len);
793 return xs_sendmsg(sock, msg, base);
797 * xs_sendpages - write pages directly to a socket
798 * @sock: socket to send on
799 * @addr: UDP only -- address of destination
800 * @addrlen: UDP only -- length of destination address
801 * @xdr: buffer containing this request
802 * @base: starting position in the buffer
803 * @rm: stream record marker field
804 * @sent_p: return the total number of bytes successfully queued for sending
807 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, rpc_fraghdr rm, int *sent_p)
809 struct msghdr msg = {
811 .msg_namelen = addrlen,
812 .msg_flags = XS_SENDMSG_FLAGS | MSG_MORE,
814 unsigned int rmsize = rm ? sizeof(rm) : 0;
815 unsigned int remainder = rmsize + xdr->len - base;
822 want = xdr->head[0].iov_len + rmsize;
824 unsigned int len = want - base;
827 msg.msg_flags &= ~MSG_MORE;
829 err = xs_send_rm_and_kvec(sock, &msg, rm,
830 &xdr->head[0], base);
832 err = xs_send_kvec(sock, &msg, &xdr->head[0], base);
833 if (remainder == 0 || err != len)
840 if (base < xdr->page_len) {
841 unsigned int len = xdr->page_len - base;
844 msg.msg_flags &= ~MSG_MORE;
845 err = xs_send_pagedata(sock, &msg, xdr, base);
846 if (remainder == 0 || err != len)
851 base -= xdr->page_len;
853 if (base >= xdr->tail[0].iov_len)
855 msg.msg_flags &= ~MSG_MORE;
856 err = xs_send_kvec(sock, &msg, &xdr->tail[0], base);
866 * xs_nospace - handle transmit was incomplete
867 * @req: pointer to RPC request
870 static int xs_nospace(struct rpc_rqst *req)
872 struct rpc_xprt *xprt = req->rq_xprt;
873 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
874 struct sock *sk = transport->inet;
877 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
878 req->rq_task->tk_pid,
879 req->rq_slen - transport->xmit.offset,
882 /* Protect against races with write_space */
883 spin_lock_bh(&xprt->transport_lock);
885 /* Don't race with disconnect */
886 if (xprt_connected(xprt)) {
887 /* wait for more buffer space */
888 sk->sk_write_pending++;
889 xprt_wait_for_buffer_space(xprt);
893 spin_unlock_bh(&xprt->transport_lock);
895 /* Race breaker in case memory is freed before above code is called */
896 if (ret == -EAGAIN) {
897 struct socket_wq *wq;
900 wq = rcu_dereference(sk->sk_wq);
901 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
904 sk->sk_write_space(sk);
910 xs_stream_prepare_request(struct rpc_rqst *req)
912 req->rq_task->tk_status = xdr_alloc_bvec(&req->rq_rcv_buf, GFP_KERNEL);
916 * Determine if the previous message in the stream was aborted before it
917 * could complete transmission.
920 xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req)
922 return transport->xmit.offset != 0 && req->rq_bytes_sent == 0;
926 * Return the stream record marker field for a record of length < 2^31-1
929 xs_stream_record_marker(struct xdr_buf *xdr)
933 return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len);
937 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
938 * @req: pointer to RPC request
941 * 0: The request has been sent
942 * EAGAIN: The socket was blocked, please call again later to
943 * complete the request
944 * ENOTCONN: Caller needs to invoke connect logic then call again
945 * other: Some other error occured, the request was not sent
947 static int xs_local_send_request(struct rpc_rqst *req)
949 struct rpc_xprt *xprt = req->rq_xprt;
950 struct sock_xprt *transport =
951 container_of(xprt, struct sock_xprt, xprt);
952 struct xdr_buf *xdr = &req->rq_snd_buf;
956 /* Close the stream if the previous transmission was incomplete */
957 if (xs_send_request_was_aborted(transport, req)) {
962 xs_pktdump("packet data:",
963 req->rq_svec->iov_base, req->rq_svec->iov_len);
965 req->rq_xtime = ktime_get();
966 status = xs_sendpages(transport->sock, NULL, 0, xdr,
967 transport->xmit.offset,
968 xs_stream_record_marker(xdr),
970 dprintk("RPC: %s(%u) = %d\n",
971 __func__, xdr->len - transport->xmit.offset, status);
973 if (status == -EAGAIN && sock_writeable(transport->inet))
976 if (likely(sent > 0) || status == 0) {
977 transport->xmit.offset += sent;
978 req->rq_bytes_sent = transport->xmit.offset;
979 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
980 req->rq_xmit_bytes_sent += transport->xmit.offset;
981 transport->xmit.offset = 0;
991 status = xs_nospace(req);
994 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1006 * xs_udp_send_request - write an RPC request to a UDP socket
1007 * @req: pointer to RPC request
1010 * 0: The request has been sent
1011 * EAGAIN: The socket was blocked, please call again later to
1012 * complete the request
1013 * ENOTCONN: Caller needs to invoke connect logic then call again
1014 * other: Some other error occurred, the request was not sent
1016 static int xs_udp_send_request(struct rpc_rqst *req)
1018 struct rpc_xprt *xprt = req->rq_xprt;
1019 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1020 struct xdr_buf *xdr = &req->rq_snd_buf;
1024 xs_pktdump("packet data:",
1025 req->rq_svec->iov_base,
1026 req->rq_svec->iov_len);
1028 if (!xprt_bound(xprt))
1031 if (!xprt_request_get_cong(xprt, req))
1034 req->rq_xtime = ktime_get();
1035 status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
1038 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
1041 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
1042 if (status == -EPERM)
1043 goto process_status;
1045 if (status == -EAGAIN && sock_writeable(transport->inet))
1048 if (sent > 0 || status == 0) {
1049 req->rq_xmit_bytes_sent += sent;
1050 if (sent >= req->rq_slen)
1052 /* Still some bytes left; set up for a retry later. */
1060 /* Should we call xs_close() here? */
1063 status = xs_nospace(req);
1070 /* When the server has died, an ICMP port unreachable message
1071 * prompts ECONNREFUSED. */
1074 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1082 * xs_tcp_send_request - write an RPC request to a TCP socket
1083 * @req: pointer to RPC request
1086 * 0: The request has been sent
1087 * EAGAIN: The socket was blocked, please call again later to
1088 * complete the request
1089 * ENOTCONN: Caller needs to invoke connect logic then call again
1090 * other: Some other error occurred, the request was not sent
1092 * XXX: In the case of soft timeouts, should we eventually give up
1093 * if sendmsg is not able to make progress?
1095 static int xs_tcp_send_request(struct rpc_rqst *req)
1097 struct rpc_xprt *xprt = req->rq_xprt;
1098 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1099 struct xdr_buf *xdr = &req->rq_snd_buf;
1100 bool vm_wait = false;
1104 /* Close the stream if the previous transmission was incomplete */
1105 if (xs_send_request_was_aborted(transport, req)) {
1106 if (transport->sock != NULL)
1107 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
1111 xs_pktdump("packet data:",
1112 req->rq_svec->iov_base,
1113 req->rq_svec->iov_len);
1115 if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
1116 xs_tcp_set_socket_timeouts(xprt, transport->sock);
1118 /* Continue transmitting the packet/record. We must be careful
1119 * to cope with writespace callbacks arriving _after_ we have
1120 * called sendmsg(). */
1121 req->rq_xtime = ktime_get();
1124 status = xs_sendpages(transport->sock, NULL, 0, xdr,
1125 transport->xmit.offset,
1126 xs_stream_record_marker(xdr),
1129 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
1130 xdr->len - transport->xmit.offset, status);
1132 /* If we've sent the entire packet, immediately
1133 * reset the count of bytes sent. */
1134 transport->xmit.offset += sent;
1135 req->rq_bytes_sent = transport->xmit.offset;
1136 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
1137 req->rq_xmit_bytes_sent += transport->xmit.offset;
1138 transport->xmit.offset = 0;
1142 WARN_ON_ONCE(sent == 0 && status == 0);
1144 if (status == -EAGAIN ) {
1146 * Return EAGAIN if we're sure we're hitting the
1147 * socket send buffer limits.
1149 if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
1152 * Did we hit a memory allocation failure?
1158 /* Retry, knowing now that we're below the
1159 * socket send buffer limit
1173 /* Should we call xs_close() here? */
1176 status = xs_nospace(req);
1186 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1193 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1195 transport->old_data_ready = sk->sk_data_ready;
1196 transport->old_state_change = sk->sk_state_change;
1197 transport->old_write_space = sk->sk_write_space;
1198 transport->old_error_report = sk->sk_error_report;
1201 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1203 sk->sk_data_ready = transport->old_data_ready;
1204 sk->sk_state_change = transport->old_state_change;
1205 sk->sk_write_space = transport->old_write_space;
1206 sk->sk_error_report = transport->old_error_report;
1209 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
1211 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1213 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1216 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1218 smp_mb__before_atomic();
1219 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1220 clear_bit(XPRT_CLOSING, &xprt->state);
1221 xs_sock_reset_state_flags(xprt);
1222 smp_mb__after_atomic();
1226 * xs_error_report - callback to handle TCP socket state errors
1229 * Note: we don't call sock_error() since there may be a rpc_task
1230 * using the socket, and so we don't want to clear sk->sk_err.
1232 static void xs_error_report(struct sock *sk)
1234 struct rpc_xprt *xprt;
1237 read_lock_bh(&sk->sk_callback_lock);
1238 if (!(xprt = xprt_from_sock(sk)))
1244 dprintk("RPC: xs_error_report client %p, error=%d...\n",
1246 trace_rpc_socket_error(xprt, sk->sk_socket, err);
1247 xprt_wake_pending_tasks(xprt, err);
1249 read_unlock_bh(&sk->sk_callback_lock);
1252 static void xs_reset_transport(struct sock_xprt *transport)
1254 struct socket *sock = transport->sock;
1255 struct sock *sk = transport->inet;
1256 struct rpc_xprt *xprt = &transport->xprt;
1257 struct file *filp = transport->file;
1262 if (atomic_read(&transport->xprt.swapper))
1263 sk_clear_memalloc(sk);
1265 kernel_sock_shutdown(sock, SHUT_RDWR);
1267 mutex_lock(&transport->recv_mutex);
1268 write_lock_bh(&sk->sk_callback_lock);
1269 transport->inet = NULL;
1270 transport->sock = NULL;
1271 transport->file = NULL;
1273 sk->sk_user_data = NULL;
1275 xs_restore_old_callbacks(transport, sk);
1276 xprt_clear_connected(xprt);
1277 write_unlock_bh(&sk->sk_callback_lock);
1278 xs_sock_reset_connection_flags(xprt);
1279 /* Reset stream record info */
1280 xs_stream_reset_connect(transport);
1281 mutex_unlock(&transport->recv_mutex);
1283 trace_rpc_socket_close(xprt, sock);
1286 xprt_disconnect_done(xprt);
1290 * xs_close - close a socket
1293 * This is used when all requests are complete; ie, no DRC state remains
1294 * on the server we want to save.
1296 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1297 * xs_reset_transport() zeroing the socket from underneath a writer.
1299 static void xs_close(struct rpc_xprt *xprt)
1301 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1303 dprintk("RPC: xs_close xprt %p\n", xprt);
1305 xs_reset_transport(transport);
1306 xprt->reestablish_timeout = 0;
1309 static void xs_inject_disconnect(struct rpc_xprt *xprt)
1311 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
1313 xprt_disconnect_done(xprt);
1316 static void xs_xprt_free(struct rpc_xprt *xprt)
1318 xs_free_peer_addresses(xprt);
1323 * xs_destroy - prepare to shutdown a transport
1324 * @xprt: doomed transport
1327 static void xs_destroy(struct rpc_xprt *xprt)
1329 struct sock_xprt *transport = container_of(xprt,
1330 struct sock_xprt, xprt);
1331 dprintk("RPC: xs_destroy xprt %p\n", xprt);
1333 cancel_delayed_work_sync(&transport->connect_worker);
1335 cancel_work_sync(&transport->recv_worker);
1337 module_put(THIS_MODULE);
1341 * xs_udp_data_read_skb - receive callback for UDP sockets
1347 static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1349 struct sk_buff *skb)
1351 struct rpc_task *task;
1352 struct rpc_rqst *rovr;
1353 int repsize, copied;
1359 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
1363 /* Copy the XID from the skb... */
1364 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1368 /* Look up and lock the request corresponding to the given XID */
1369 spin_lock(&xprt->queue_lock);
1370 rovr = xprt_lookup_rqst(xprt, *xp);
1373 xprt_pin_rqst(rovr);
1374 xprt_update_rtt(rovr->rq_task);
1375 spin_unlock(&xprt->queue_lock);
1376 task = rovr->rq_task;
1378 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1381 /* Suck it into the iovec, verify checksum if not done by hw. */
1382 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1383 spin_lock(&xprt->queue_lock);
1384 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1389 spin_lock_bh(&xprt->transport_lock);
1390 xprt_adjust_cwnd(xprt, task, copied);
1391 spin_unlock_bh(&xprt->transport_lock);
1392 spin_lock(&xprt->queue_lock);
1393 xprt_complete_rqst(task, copied);
1394 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1396 xprt_unpin_rqst(rovr);
1398 spin_unlock(&xprt->queue_lock);
1401 static void xs_udp_data_receive(struct sock_xprt *transport)
1403 struct sk_buff *skb;
1407 mutex_lock(&transport->recv_mutex);
1408 sk = transport->inet;
1412 skb = skb_recv_udp(sk, 0, 1, &err);
1415 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1419 xs_poll_check_readable(transport);
1421 mutex_unlock(&transport->recv_mutex);
1424 static void xs_udp_data_receive_workfn(struct work_struct *work)
1426 struct sock_xprt *transport =
1427 container_of(work, struct sock_xprt, recv_worker);
1428 unsigned int pflags = memalloc_nofs_save();
1430 xs_udp_data_receive(transport);
1431 memalloc_nofs_restore(pflags);
1435 * xs_data_ready - "data ready" callback for UDP sockets
1436 * @sk: socket with data to read
1439 static void xs_data_ready(struct sock *sk)
1441 struct rpc_xprt *xprt;
1443 read_lock_bh(&sk->sk_callback_lock);
1444 dprintk("RPC: xs_data_ready...\n");
1445 xprt = xprt_from_sock(sk);
1447 struct sock_xprt *transport = container_of(xprt,
1448 struct sock_xprt, xprt);
1449 transport->old_data_ready(sk);
1450 /* Any data means we had a useful conversation, so
1451 * then we don't need to delay the next reconnect
1453 if (xprt->reestablish_timeout)
1454 xprt->reestablish_timeout = 0;
1455 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1456 queue_work(xprtiod_workqueue, &transport->recv_worker);
1458 read_unlock_bh(&sk->sk_callback_lock);
1462 * Helper function to force a TCP close if the server is sending
1463 * junk and/or it has put us in CLOSE_WAIT
1465 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1467 xprt_force_disconnect(xprt);
1470 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1471 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1475 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1478 * xs_tcp_state_change - callback to handle TCP socket state changes
1479 * @sk: socket whose state has changed
1482 static void xs_tcp_state_change(struct sock *sk)
1484 struct rpc_xprt *xprt;
1485 struct sock_xprt *transport;
1487 read_lock_bh(&sk->sk_callback_lock);
1488 if (!(xprt = xprt_from_sock(sk)))
1490 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1491 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1492 sk->sk_state, xprt_connected(xprt),
1493 sock_flag(sk, SOCK_DEAD),
1494 sock_flag(sk, SOCK_ZAPPED),
1497 transport = container_of(xprt, struct sock_xprt, xprt);
1498 trace_rpc_socket_state_change(xprt, sk->sk_socket);
1499 switch (sk->sk_state) {
1500 case TCP_ESTABLISHED:
1501 spin_lock(&xprt->transport_lock);
1502 if (!xprt_test_and_set_connected(xprt)) {
1503 xprt->connect_cookie++;
1504 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1505 xprt_clear_connecting(xprt);
1507 xprt->stat.connect_count++;
1508 xprt->stat.connect_time += (long)jiffies -
1509 xprt->stat.connect_start;
1510 xprt_wake_pending_tasks(xprt, -EAGAIN);
1512 spin_unlock(&xprt->transport_lock);
1515 /* The client initiated a shutdown of the socket */
1516 xprt->connect_cookie++;
1517 xprt->reestablish_timeout = 0;
1518 set_bit(XPRT_CLOSING, &xprt->state);
1519 smp_mb__before_atomic();
1520 clear_bit(XPRT_CONNECTED, &xprt->state);
1521 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1522 smp_mb__after_atomic();
1524 case TCP_CLOSE_WAIT:
1525 /* The server initiated a shutdown of the socket */
1526 xprt->connect_cookie++;
1527 clear_bit(XPRT_CONNECTED, &xprt->state);
1528 xs_tcp_force_close(xprt);
1532 * If the server closed down the connection, make sure that
1533 * we back off before reconnecting
1535 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1536 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1539 set_bit(XPRT_CLOSING, &xprt->state);
1540 smp_mb__before_atomic();
1541 clear_bit(XPRT_CONNECTED, &xprt->state);
1542 smp_mb__after_atomic();
1545 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1546 &transport->sock_state))
1547 xprt_clear_connecting(xprt);
1548 clear_bit(XPRT_CLOSING, &xprt->state);
1549 /* Trigger the socket release */
1550 xs_tcp_force_close(xprt);
1553 read_unlock_bh(&sk->sk_callback_lock);
1556 static void xs_write_space(struct sock *sk)
1558 struct socket_wq *wq;
1559 struct rpc_xprt *xprt;
1563 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1565 if (unlikely(!(xprt = xprt_from_sock(sk))))
1568 wq = rcu_dereference(sk->sk_wq);
1569 if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1572 if (xprt_write_space(xprt))
1573 sk->sk_write_pending--;
1579 * xs_udp_write_space - callback invoked when socket buffer space
1581 * @sk: socket whose state has changed
1583 * Called when more output buffer space is available for this socket.
1584 * We try not to wake our writers until they can make "significant"
1585 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1586 * with a bunch of small requests.
1588 static void xs_udp_write_space(struct sock *sk)
1590 read_lock_bh(&sk->sk_callback_lock);
1592 /* from net/core/sock.c:sock_def_write_space */
1593 if (sock_writeable(sk))
1596 read_unlock_bh(&sk->sk_callback_lock);
1600 * xs_tcp_write_space - callback invoked when socket buffer space
1602 * @sk: socket whose state has changed
1604 * Called when more output buffer space is available for this socket.
1605 * We try not to wake our writers until they can make "significant"
1606 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1607 * with a bunch of small requests.
1609 static void xs_tcp_write_space(struct sock *sk)
1611 read_lock_bh(&sk->sk_callback_lock);
1613 /* from net/core/stream.c:sk_stream_write_space */
1614 if (sk_stream_is_writeable(sk))
1617 read_unlock_bh(&sk->sk_callback_lock);
1620 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1622 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1623 struct sock *sk = transport->inet;
1625 if (transport->rcvsize) {
1626 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1627 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1629 if (transport->sndsize) {
1630 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1631 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1632 sk->sk_write_space(sk);
1637 * xs_udp_set_buffer_size - set send and receive limits
1638 * @xprt: generic transport
1639 * @sndsize: requested size of send buffer, in bytes
1640 * @rcvsize: requested size of receive buffer, in bytes
1642 * Set socket send and receive buffer size limits.
1644 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1646 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1648 transport->sndsize = 0;
1650 transport->sndsize = sndsize + 1024;
1651 transport->rcvsize = 0;
1653 transport->rcvsize = rcvsize + 1024;
1655 xs_udp_do_set_buffer_size(xprt);
1659 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1660 * @xprt: controlling transport
1661 * @task: task that timed out
1663 * Adjust the congestion window after a retransmit timeout has occurred.
1665 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1667 spin_lock_bh(&xprt->transport_lock);
1668 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1669 spin_unlock_bh(&xprt->transport_lock);
1672 static int xs_get_random_port(void)
1674 unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1675 unsigned short range;
1676 unsigned short rand;
1680 range = max - min + 1;
1681 rand = (unsigned short) prandom_u32() % range;
1686 * xs_set_reuseaddr_port - set the socket's port and address reuse options
1689 * Note that this function has to be called on all sockets that share the
1690 * same port, and it must be called before binding.
1692 static void xs_sock_set_reuseport(struct socket *sock)
1696 kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1697 (char *)&opt, sizeof(opt));
1700 static unsigned short xs_sock_getport(struct socket *sock)
1702 struct sockaddr_storage buf;
1703 unsigned short port = 0;
1705 if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1707 switch (buf.ss_family) {
1709 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1712 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1719 * xs_set_port - reset the port number in the remote endpoint address
1720 * @xprt: generic transport
1721 * @port: new port number
1724 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1726 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1728 rpc_set_port(xs_addr(xprt), port);
1729 xs_update_peer_port(xprt);
1732 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1734 if (transport->srcport == 0)
1735 transport->srcport = xs_sock_getport(sock);
1738 static int xs_get_srcport(struct sock_xprt *transport)
1740 int port = transport->srcport;
1742 if (port == 0 && transport->xprt.resvport)
1743 port = xs_get_random_port();
1747 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1749 if (transport->srcport != 0)
1750 transport->srcport = 0;
1751 if (!transport->xprt.resvport)
1753 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1754 return xprt_max_resvport;
1757 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1759 struct sockaddr_storage myaddr;
1761 int port = xs_get_srcport(transport);
1762 unsigned short last;
1765 * If we are asking for any ephemeral port (i.e. port == 0 &&
1766 * transport->xprt.resvport == 0), don't bind. Let the local
1767 * port selection happen implicitly when the socket is used
1768 * (for example at connect time).
1770 * This ensures that we can continue to establish TCP
1771 * connections even when all local ephemeral ports are already
1772 * a part of some TCP connection. This makes no difference
1773 * for UDP sockets, but also doens't harm them.
1775 * If we're asking for any reserved port (i.e. port == 0 &&
1776 * transport->xprt.resvport == 1) xs_get_srcport above will
1777 * ensure that port is non-zero and we will bind as needed.
1782 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1784 rpc_set_port((struct sockaddr *)&myaddr, port);
1785 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1786 transport->xprt.addrlen);
1788 transport->srcport = port;
1792 port = xs_next_srcport(transport, port);
1795 } while (err == -EADDRINUSE && nloop != 2);
1797 if (myaddr.ss_family == AF_INET)
1798 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1799 &((struct sockaddr_in *)&myaddr)->sin_addr,
1800 port, err ? "failed" : "ok", err);
1802 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1803 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1804 port, err ? "failed" : "ok", err);
1809 * We don't support autobind on AF_LOCAL sockets
1811 static void xs_local_rpcbind(struct rpc_task *task)
1813 xprt_set_bound(task->tk_xprt);
1816 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1820 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1821 static struct lock_class_key xs_key[2];
1822 static struct lock_class_key xs_slock_key[2];
1824 static inline void xs_reclassify_socketu(struct socket *sock)
1826 struct sock *sk = sock->sk;
1828 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1829 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1832 static inline void xs_reclassify_socket4(struct socket *sock)
1834 struct sock *sk = sock->sk;
1836 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1837 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1840 static inline void xs_reclassify_socket6(struct socket *sock)
1842 struct sock *sk = sock->sk;
1844 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1845 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1848 static inline void xs_reclassify_socket(int family, struct socket *sock)
1850 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1855 xs_reclassify_socketu(sock);
1858 xs_reclassify_socket4(sock);
1861 xs_reclassify_socket6(sock);
1866 static inline void xs_reclassify_socket(int family, struct socket *sock)
1871 static void xs_dummy_setup_socket(struct work_struct *work)
1875 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1876 struct sock_xprt *transport, int family, int type,
1877 int protocol, bool reuseport)
1880 struct socket *sock;
1883 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1885 dprintk("RPC: can't create %d transport socket (%d).\n",
1889 xs_reclassify_socket(family, sock);
1892 xs_sock_set_reuseport(sock);
1894 err = xs_bind(transport, sock);
1900 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1902 return ERR_CAST(filp);
1903 transport->file = filp;
1907 return ERR_PTR(err);
1910 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1911 struct socket *sock)
1913 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1916 if (!transport->inet) {
1917 struct sock *sk = sock->sk;
1919 write_lock_bh(&sk->sk_callback_lock);
1921 xs_save_old_callbacks(transport, sk);
1923 sk->sk_user_data = xprt;
1924 sk->sk_data_ready = xs_data_ready;
1925 sk->sk_write_space = xs_udp_write_space;
1926 sock_set_flag(sk, SOCK_FASYNC);
1927 sk->sk_error_report = xs_error_report;
1929 xprt_clear_connected(xprt);
1931 /* Reset to new socket */
1932 transport->sock = sock;
1933 transport->inet = sk;
1935 write_unlock_bh(&sk->sk_callback_lock);
1938 xs_stream_start_connect(transport);
1940 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1944 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1945 * @transport: socket transport to connect
1947 static int xs_local_setup_socket(struct sock_xprt *transport)
1949 struct rpc_xprt *xprt = &transport->xprt;
1951 struct socket *sock;
1954 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1955 SOCK_STREAM, 0, &sock, 1);
1957 dprintk("RPC: can't create AF_LOCAL "
1958 "transport socket (%d).\n", -status);
1961 xs_reclassify_socket(AF_LOCAL, sock);
1963 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1965 status = PTR_ERR(filp);
1968 transport->file = filp;
1970 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1971 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1973 status = xs_local_finish_connecting(xprt, sock);
1974 trace_rpc_socket_connect(xprt, sock, status);
1977 dprintk("RPC: xprt %p connected to %s\n",
1978 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1979 xprt->stat.connect_count++;
1980 xprt->stat.connect_time += (long)jiffies -
1981 xprt->stat.connect_start;
1982 xprt_set_connected(xprt);
1986 dprintk("RPC: xprt %p: socket %s does not exist\n",
1987 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1990 dprintk("RPC: xprt %p: connection refused for %s\n",
1991 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1994 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1996 xprt->address_strings[RPC_DISPLAY_ADDR]);
2000 xprt_clear_connecting(xprt);
2001 xprt_wake_pending_tasks(xprt, status);
2005 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2007 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2010 if (RPC_IS_ASYNC(task)) {
2012 * We want the AF_LOCAL connect to be resolved in the
2013 * filesystem namespace of the process making the rpc
2014 * call. Thus we connect synchronously.
2016 * If we want to support asynchronous AF_LOCAL calls,
2017 * we'll need to figure out how to pass a namespace to
2020 rpc_exit(task, -ENOTCONN);
2023 ret = xs_local_setup_socket(transport);
2024 if (ret && !RPC_IS_SOFTCONN(task))
2025 msleep_interruptible(15000);
2028 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2030 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2031 * know that we have exclusive access to the socket), to guard against
2032 * races with xs_reset_transport.
2034 static void xs_set_memalloc(struct rpc_xprt *xprt)
2036 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2040 * If there's no sock, then we have nothing to set. The
2041 * reconnecting process will get it for us.
2043 if (!transport->inet)
2045 if (atomic_read(&xprt->swapper))
2046 sk_set_memalloc(transport->inet);
2050 * xs_enable_swap - Tag this transport as being used for swap.
2051 * @xprt: transport to tag
2053 * Take a reference to this transport on behalf of the rpc_clnt, and
2054 * optionally mark it for swapping if it wasn't already.
2057 xs_enable_swap(struct rpc_xprt *xprt)
2059 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2061 if (atomic_inc_return(&xprt->swapper) != 1)
2063 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2064 return -ERESTARTSYS;
2066 sk_set_memalloc(xs->inet);
2067 xprt_release_xprt(xprt, NULL);
2072 * xs_disable_swap - Untag this transport as being used for swap.
2073 * @xprt: transport to tag
2075 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2076 * swapper refcount goes to 0, untag the socket as a memalloc socket.
2079 xs_disable_swap(struct rpc_xprt *xprt)
2081 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2083 if (!atomic_dec_and_test(&xprt->swapper))
2085 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2088 sk_clear_memalloc(xs->inet);
2089 xprt_release_xprt(xprt, NULL);
2092 static void xs_set_memalloc(struct rpc_xprt *xprt)
2097 xs_enable_swap(struct rpc_xprt *xprt)
2103 xs_disable_swap(struct rpc_xprt *xprt)
2108 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2110 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2112 if (!transport->inet) {
2113 struct sock *sk = sock->sk;
2115 write_lock_bh(&sk->sk_callback_lock);
2117 xs_save_old_callbacks(transport, sk);
2119 sk->sk_user_data = xprt;
2120 sk->sk_data_ready = xs_data_ready;
2121 sk->sk_write_space = xs_udp_write_space;
2122 sock_set_flag(sk, SOCK_FASYNC);
2124 xprt_set_connected(xprt);
2126 /* Reset to new socket */
2127 transport->sock = sock;
2128 transport->inet = sk;
2130 xs_set_memalloc(xprt);
2132 write_unlock_bh(&sk->sk_callback_lock);
2134 xs_udp_do_set_buffer_size(xprt);
2136 xprt->stat.connect_start = jiffies;
2139 static void xs_udp_setup_socket(struct work_struct *work)
2141 struct sock_xprt *transport =
2142 container_of(work, struct sock_xprt, connect_worker.work);
2143 struct rpc_xprt *xprt = &transport->xprt;
2144 struct socket *sock;
2147 sock = xs_create_sock(xprt, transport,
2148 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2149 IPPROTO_UDP, false);
2153 dprintk("RPC: worker connecting xprt %p via %s to "
2154 "%s (port %s)\n", xprt,
2155 xprt->address_strings[RPC_DISPLAY_PROTO],
2156 xprt->address_strings[RPC_DISPLAY_ADDR],
2157 xprt->address_strings[RPC_DISPLAY_PORT]);
2159 xs_udp_finish_connecting(xprt, sock);
2160 trace_rpc_socket_connect(xprt, sock, 0);
2163 xprt_clear_connecting(xprt);
2164 xprt_unlock_connect(xprt, transport);
2165 xprt_wake_pending_tasks(xprt, status);
2169 * xs_tcp_shutdown - gracefully shut down a TCP socket
2172 * Initiates a graceful shutdown of the TCP socket by calling the
2173 * equivalent of shutdown(SHUT_RDWR);
2175 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2177 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2178 struct socket *sock = transport->sock;
2179 int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2185 kernel_sock_shutdown(sock, SHUT_RDWR);
2186 trace_rpc_socket_shutdown(xprt, sock);
2190 xs_reset_transport(transport);
2194 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2195 struct socket *sock)
2197 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2198 unsigned int keepidle;
2199 unsigned int keepcnt;
2200 unsigned int opt_on = 1;
2203 spin_lock_bh(&xprt->transport_lock);
2204 keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2205 keepcnt = xprt->timeout->to_retries + 1;
2206 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2207 (xprt->timeout->to_retries + 1);
2208 clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2209 spin_unlock_bh(&xprt->transport_lock);
2211 /* TCP Keepalive options */
2212 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2213 (char *)&opt_on, sizeof(opt_on));
2214 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2215 (char *)&keepidle, sizeof(keepidle));
2216 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2217 (char *)&keepidle, sizeof(keepidle));
2218 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2219 (char *)&keepcnt, sizeof(keepcnt));
2221 /* TCP user timeout (see RFC5482) */
2222 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2223 (char *)&timeo, sizeof(timeo));
2226 static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2227 unsigned long connect_timeout,
2228 unsigned long reconnect_timeout)
2230 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2231 struct rpc_timeout to;
2232 unsigned long initval;
2234 spin_lock_bh(&xprt->transport_lock);
2235 if (reconnect_timeout < xprt->max_reconnect_timeout)
2236 xprt->max_reconnect_timeout = reconnect_timeout;
2237 if (connect_timeout < xprt->connect_timeout) {
2238 memcpy(&to, xprt->timeout, sizeof(to));
2239 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2240 /* Arbitrary lower limit */
2241 if (initval < XS_TCP_INIT_REEST_TO << 1)
2242 initval = XS_TCP_INIT_REEST_TO << 1;
2243 to.to_initval = initval;
2244 to.to_maxval = initval;
2245 memcpy(&transport->tcp_timeout, &to,
2246 sizeof(transport->tcp_timeout));
2247 xprt->timeout = &transport->tcp_timeout;
2248 xprt->connect_timeout = connect_timeout;
2250 set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2251 spin_unlock_bh(&xprt->transport_lock);
2254 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2256 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2257 int ret = -ENOTCONN;
2259 if (!transport->inet) {
2260 struct sock *sk = sock->sk;
2261 unsigned int addr_pref = IPV6_PREFER_SRC_PUBLIC;
2263 /* Avoid temporary address, they are bad for long-lived
2264 * connections such as NFS mounts.
2265 * RFC4941, section 3.6 suggests that:
2266 * Individual applications, which have specific
2267 * knowledge about the normal duration of connections,
2268 * MAY override this as appropriate.
2270 kernel_setsockopt(sock, SOL_IPV6, IPV6_ADDR_PREFERENCES,
2271 (char *)&addr_pref, sizeof(addr_pref));
2273 xs_tcp_set_socket_timeouts(xprt, sock);
2275 write_lock_bh(&sk->sk_callback_lock);
2277 xs_save_old_callbacks(transport, sk);
2279 sk->sk_user_data = xprt;
2280 sk->sk_data_ready = xs_data_ready;
2281 sk->sk_state_change = xs_tcp_state_change;
2282 sk->sk_write_space = xs_tcp_write_space;
2283 sock_set_flag(sk, SOCK_FASYNC);
2284 sk->sk_error_report = xs_error_report;
2286 /* socket options */
2287 sock_reset_flag(sk, SOCK_LINGER);
2288 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2290 xprt_clear_connected(xprt);
2292 /* Reset to new socket */
2293 transport->sock = sock;
2294 transport->inet = sk;
2296 write_unlock_bh(&sk->sk_callback_lock);
2299 if (!xprt_bound(xprt))
2302 xs_set_memalloc(xprt);
2304 xs_stream_start_connect(transport);
2306 /* Tell the socket layer to start connecting... */
2307 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2308 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2311 xs_set_srcport(transport, sock);
2315 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2316 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2318 case -EADDRNOTAVAIL:
2319 /* Source port number is unavailable. Try a new one! */
2320 transport->srcport = 0;
2327 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2328 * @work: queued work item
2330 * Invoked by a work queue tasklet.
2332 static void xs_tcp_setup_socket(struct work_struct *work)
2334 struct sock_xprt *transport =
2335 container_of(work, struct sock_xprt, connect_worker.work);
2336 struct socket *sock = transport->sock;
2337 struct rpc_xprt *xprt = &transport->xprt;
2341 sock = xs_create_sock(xprt, transport,
2342 xs_addr(xprt)->sa_family, SOCK_STREAM,
2345 status = PTR_ERR(sock);
2350 dprintk("RPC: worker connecting xprt %p via %s to "
2351 "%s (port %s)\n", xprt,
2352 xprt->address_strings[RPC_DISPLAY_PROTO],
2353 xprt->address_strings[RPC_DISPLAY_ADDR],
2354 xprt->address_strings[RPC_DISPLAY_PORT]);
2356 status = xs_tcp_finish_connecting(xprt, sock);
2357 trace_rpc_socket_connect(xprt, sock, status);
2358 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2359 xprt, -status, xprt_connected(xprt),
2360 sock->sk->sk_state);
2363 printk("%s: connect returned unhandled error %d\n",
2366 case -EADDRNOTAVAIL:
2367 /* We're probably in TIME_WAIT. Get rid of existing socket,
2370 xs_tcp_force_close(xprt);
2375 xprt_unlock_connect(xprt, transport);
2378 /* Happens, for instance, if the user specified a link
2379 * local IPv6 address without a scope-id.
2389 * xs_tcp_force_close() wakes tasks with -EIO.
2390 * We need to wake them first to ensure the
2391 * correct error code.
2393 xprt_wake_pending_tasks(xprt, status);
2394 xs_tcp_force_close(xprt);
2399 xprt_clear_connecting(xprt);
2400 xprt_unlock_connect(xprt, transport);
2401 xprt_wake_pending_tasks(xprt, status);
2404 static unsigned long xs_reconnect_delay(const struct rpc_xprt *xprt)
2406 unsigned long start, now = jiffies;
2408 start = xprt->stat.connect_start + xprt->reestablish_timeout;
2409 if (time_after(start, now))
2414 static void xs_reconnect_backoff(struct rpc_xprt *xprt)
2416 xprt->reestablish_timeout <<= 1;
2417 if (xprt->reestablish_timeout > xprt->max_reconnect_timeout)
2418 xprt->reestablish_timeout = xprt->max_reconnect_timeout;
2419 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2420 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2424 * xs_connect - connect a socket to a remote endpoint
2425 * @xprt: pointer to transport structure
2426 * @task: address of RPC task that manages state of connect request
2428 * TCP: If the remote end dropped the connection, delay reconnecting.
2430 * UDP socket connects are synchronous, but we use a work queue anyway
2431 * to guarantee that even unprivileged user processes can set up a
2432 * socket on a privileged port.
2434 * If a UDP socket connect fails, the delay behavior here prevents
2435 * retry floods (hard mounts).
2437 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2439 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2440 unsigned long delay = 0;
2442 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2444 if (transport->sock != NULL) {
2445 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2447 xprt, xprt->reestablish_timeout / HZ);
2449 /* Start by resetting any existing state */
2450 xs_reset_transport(transport);
2452 delay = xs_reconnect_delay(xprt);
2453 xs_reconnect_backoff(xprt);
2456 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2458 queue_delayed_work(xprtiod_workqueue,
2459 &transport->connect_worker,
2464 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2465 * @xprt: rpc_xprt struct containing statistics
2469 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2473 if (xprt_connected(xprt))
2474 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2476 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2477 "%llu %llu %lu %llu %llu\n",
2478 xprt->stat.bind_count,
2479 xprt->stat.connect_count,
2480 xprt->stat.connect_time / HZ,
2484 xprt->stat.bad_xids,
2487 xprt->stat.max_slots,
2488 xprt->stat.sending_u,
2489 xprt->stat.pending_u);
2493 * xs_udp_print_stats - display UDP socket-specifc stats
2494 * @xprt: rpc_xprt struct containing statistics
2498 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2500 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2502 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2505 xprt->stat.bind_count,
2508 xprt->stat.bad_xids,
2511 xprt->stat.max_slots,
2512 xprt->stat.sending_u,
2513 xprt->stat.pending_u);
2517 * xs_tcp_print_stats - display TCP socket-specifc stats
2518 * @xprt: rpc_xprt struct containing statistics
2522 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2524 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2527 if (xprt_connected(xprt))
2528 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2530 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2531 "%llu %llu %lu %llu %llu\n",
2533 xprt->stat.bind_count,
2534 xprt->stat.connect_count,
2535 xprt->stat.connect_time / HZ,
2539 xprt->stat.bad_xids,
2542 xprt->stat.max_slots,
2543 xprt->stat.sending_u,
2544 xprt->stat.pending_u);
2548 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2549 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2550 * to use the server side send routines.
2552 static int bc_malloc(struct rpc_task *task)
2554 struct rpc_rqst *rqst = task->tk_rqstp;
2555 size_t size = rqst->rq_callsize;
2557 struct rpc_buffer *buf;
2559 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2560 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2565 page = alloc_page(GFP_KERNEL);
2569 buf = page_address(page);
2570 buf->len = PAGE_SIZE;
2572 rqst->rq_buffer = buf->data;
2573 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2578 * Free the space allocated in the bc_alloc routine
2580 static void bc_free(struct rpc_task *task)
2582 void *buffer = task->tk_rqstp->rq_buffer;
2583 struct rpc_buffer *buf;
2585 buf = container_of(buffer, struct rpc_buffer, data);
2586 free_page((unsigned long)buf);
2590 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2591 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2593 static int bc_sendto(struct rpc_rqst *req)
2596 struct xdr_buf *xbufp = &req->rq_snd_buf;
2597 struct sock_xprt *transport =
2598 container_of(req->rq_xprt, struct sock_xprt, xprt);
2599 unsigned long headoff;
2600 unsigned long tailoff;
2601 struct page *tailpage;
2602 struct msghdr msg = {
2603 .msg_flags = MSG_MORE
2605 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
2608 .iov_base = &marker,
2609 .iov_len = sizeof(marker),
2612 len = kernel_sendmsg(transport->sock, &msg, &iov, 1, iov.iov_len);
2613 if (len != iov.iov_len)
2617 if (xbufp->tail[0].iov_len)
2618 tailpage = virt_to_page(xbufp->tail[0].iov_base);
2619 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2620 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2621 len = svc_send_common(transport->sock, xbufp,
2622 virt_to_page(xbufp->head[0].iov_base), headoff,
2624 if (len != xbufp->len)
2630 * The send routine. Borrows from svc_send
2632 static int bc_send_request(struct rpc_rqst *req)
2634 struct svc_xprt *xprt;
2637 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2639 * Get the server socket associated with this callback xprt
2641 xprt = req->rq_xprt->bc_xprt;
2644 * Grab the mutex to serialize data as the connection is shared
2645 * with the fore channel
2647 mutex_lock(&xprt->xpt_mutex);
2648 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2651 len = bc_sendto(req);
2652 mutex_unlock(&xprt->xpt_mutex);
2661 * The close routine. Since this is client initiated, we do nothing
2664 static void bc_close(struct rpc_xprt *xprt)
2669 * The xprt destroy routine. Again, because this connection is client
2670 * initiated, we do nothing
2673 static void bc_destroy(struct rpc_xprt *xprt)
2675 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2678 module_put(THIS_MODULE);
2681 static const struct rpc_xprt_ops xs_local_ops = {
2682 .reserve_xprt = xprt_reserve_xprt,
2683 .release_xprt = xprt_release_xprt,
2684 .alloc_slot = xprt_alloc_slot,
2685 .free_slot = xprt_free_slot,
2686 .rpcbind = xs_local_rpcbind,
2687 .set_port = xs_local_set_port,
2688 .connect = xs_local_connect,
2689 .buf_alloc = rpc_malloc,
2690 .buf_free = rpc_free,
2691 .prepare_request = xs_stream_prepare_request,
2692 .send_request = xs_local_send_request,
2693 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2695 .destroy = xs_destroy,
2696 .print_stats = xs_local_print_stats,
2697 .enable_swap = xs_enable_swap,
2698 .disable_swap = xs_disable_swap,
2701 static const struct rpc_xprt_ops xs_udp_ops = {
2702 .set_buffer_size = xs_udp_set_buffer_size,
2703 .reserve_xprt = xprt_reserve_xprt_cong,
2704 .release_xprt = xprt_release_xprt_cong,
2705 .alloc_slot = xprt_alloc_slot,
2706 .free_slot = xprt_free_slot,
2707 .rpcbind = rpcb_getport_async,
2708 .set_port = xs_set_port,
2709 .connect = xs_connect,
2710 .buf_alloc = rpc_malloc,
2711 .buf_free = rpc_free,
2712 .send_request = xs_udp_send_request,
2713 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
2714 .timer = xs_udp_timer,
2715 .release_request = xprt_release_rqst_cong,
2717 .destroy = xs_destroy,
2718 .print_stats = xs_udp_print_stats,
2719 .enable_swap = xs_enable_swap,
2720 .disable_swap = xs_disable_swap,
2721 .inject_disconnect = xs_inject_disconnect,
2724 static const struct rpc_xprt_ops xs_tcp_ops = {
2725 .reserve_xprt = xprt_reserve_xprt,
2726 .release_xprt = xprt_release_xprt,
2727 .alloc_slot = xprt_alloc_slot,
2728 .free_slot = xprt_free_slot,
2729 .rpcbind = rpcb_getport_async,
2730 .set_port = xs_set_port,
2731 .connect = xs_connect,
2732 .buf_alloc = rpc_malloc,
2733 .buf_free = rpc_free,
2734 .prepare_request = xs_stream_prepare_request,
2735 .send_request = xs_tcp_send_request,
2736 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2737 .close = xs_tcp_shutdown,
2738 .destroy = xs_destroy,
2739 .set_connect_timeout = xs_tcp_set_connect_timeout,
2740 .print_stats = xs_tcp_print_stats,
2741 .enable_swap = xs_enable_swap,
2742 .disable_swap = xs_disable_swap,
2743 .inject_disconnect = xs_inject_disconnect,
2744 #ifdef CONFIG_SUNRPC_BACKCHANNEL
2745 .bc_setup = xprt_setup_bc,
2746 .bc_maxpayload = xs_tcp_bc_maxpayload,
2747 .bc_free_rqst = xprt_free_bc_rqst,
2748 .bc_destroy = xprt_destroy_bc,
2753 * The rpc_xprt_ops for the server backchannel
2756 static const struct rpc_xprt_ops bc_tcp_ops = {
2757 .reserve_xprt = xprt_reserve_xprt,
2758 .release_xprt = xprt_release_xprt,
2759 .alloc_slot = xprt_alloc_slot,
2760 .free_slot = xprt_free_slot,
2761 .buf_alloc = bc_malloc,
2762 .buf_free = bc_free,
2763 .send_request = bc_send_request,
2764 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2766 .destroy = bc_destroy,
2767 .print_stats = xs_tcp_print_stats,
2768 .enable_swap = xs_enable_swap,
2769 .disable_swap = xs_disable_swap,
2770 .inject_disconnect = xs_inject_disconnect,
2773 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2775 static const struct sockaddr_in sin = {
2776 .sin_family = AF_INET,
2777 .sin_addr.s_addr = htonl(INADDR_ANY),
2779 static const struct sockaddr_in6 sin6 = {
2780 .sin6_family = AF_INET6,
2781 .sin6_addr = IN6ADDR_ANY_INIT,
2788 memcpy(sap, &sin, sizeof(sin));
2791 memcpy(sap, &sin6, sizeof(sin6));
2794 dprintk("RPC: %s: Bad address family\n", __func__);
2795 return -EAFNOSUPPORT;
2800 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2801 unsigned int slot_table_size,
2802 unsigned int max_slot_table_size)
2804 struct rpc_xprt *xprt;
2805 struct sock_xprt *new;
2807 if (args->addrlen > sizeof(xprt->addr)) {
2808 dprintk("RPC: xs_setup_xprt: address too large\n");
2809 return ERR_PTR(-EBADF);
2812 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2813 max_slot_table_size);
2815 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2817 return ERR_PTR(-ENOMEM);
2820 new = container_of(xprt, struct sock_xprt, xprt);
2821 mutex_init(&new->recv_mutex);
2822 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2823 xprt->addrlen = args->addrlen;
2825 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2828 err = xs_init_anyaddr(args->dstaddr->sa_family,
2829 (struct sockaddr *)&new->srcaddr);
2832 return ERR_PTR(err);
2839 static const struct rpc_timeout xs_local_default_timeout = {
2840 .to_initval = 10 * HZ,
2841 .to_maxval = 10 * HZ,
2846 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2847 * @args: rpc transport creation arguments
2849 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2851 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2853 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2854 struct sock_xprt *transport;
2855 struct rpc_xprt *xprt;
2856 struct rpc_xprt *ret;
2858 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2859 xprt_max_tcp_slot_table_entries);
2862 transport = container_of(xprt, struct sock_xprt, xprt);
2865 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2867 xprt->bind_timeout = XS_BIND_TO;
2868 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2869 xprt->idle_timeout = XS_IDLE_DISC_TO;
2871 xprt->ops = &xs_local_ops;
2872 xprt->timeout = &xs_local_default_timeout;
2874 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2875 INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket);
2877 switch (sun->sun_family) {
2879 if (sun->sun_path[0] != '/') {
2880 dprintk("RPC: bad AF_LOCAL address: %s\n",
2882 ret = ERR_PTR(-EINVAL);
2885 xprt_set_bound(xprt);
2886 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2887 ret = ERR_PTR(xs_local_setup_socket(transport));
2892 ret = ERR_PTR(-EAFNOSUPPORT);
2896 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2897 xprt->address_strings[RPC_DISPLAY_ADDR]);
2899 if (try_module_get(THIS_MODULE))
2901 ret = ERR_PTR(-EINVAL);
2907 static const struct rpc_timeout xs_udp_default_timeout = {
2908 .to_initval = 5 * HZ,
2909 .to_maxval = 30 * HZ,
2910 .to_increment = 5 * HZ,
2915 * xs_setup_udp - Set up transport to use a UDP socket
2916 * @args: rpc transport creation arguments
2919 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2921 struct sockaddr *addr = args->dstaddr;
2922 struct rpc_xprt *xprt;
2923 struct sock_xprt *transport;
2924 struct rpc_xprt *ret;
2926 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2927 xprt_udp_slot_table_entries);
2930 transport = container_of(xprt, struct sock_xprt, xprt);
2932 xprt->prot = IPPROTO_UDP;
2933 /* XXX: header size can vary due to auth type, IPv6, etc. */
2934 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2936 xprt->bind_timeout = XS_BIND_TO;
2937 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2938 xprt->idle_timeout = XS_IDLE_DISC_TO;
2940 xprt->ops = &xs_udp_ops;
2942 xprt->timeout = &xs_udp_default_timeout;
2944 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
2945 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2947 switch (addr->sa_family) {
2949 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2950 xprt_set_bound(xprt);
2952 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2955 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2956 xprt_set_bound(xprt);
2958 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2961 ret = ERR_PTR(-EAFNOSUPPORT);
2965 if (xprt_bound(xprt))
2966 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2967 xprt->address_strings[RPC_DISPLAY_ADDR],
2968 xprt->address_strings[RPC_DISPLAY_PORT],
2969 xprt->address_strings[RPC_DISPLAY_PROTO]);
2971 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2972 xprt->address_strings[RPC_DISPLAY_ADDR],
2973 xprt->address_strings[RPC_DISPLAY_PROTO]);
2975 if (try_module_get(THIS_MODULE))
2977 ret = ERR_PTR(-EINVAL);
2983 static const struct rpc_timeout xs_tcp_default_timeout = {
2984 .to_initval = 60 * HZ,
2985 .to_maxval = 60 * HZ,
2990 * xs_setup_tcp - Set up transport to use a TCP socket
2991 * @args: rpc transport creation arguments
2994 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2996 struct sockaddr *addr = args->dstaddr;
2997 struct rpc_xprt *xprt;
2998 struct sock_xprt *transport;
2999 struct rpc_xprt *ret;
3000 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
3002 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
3003 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
3005 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3006 max_slot_table_size);
3009 transport = container_of(xprt, struct sock_xprt, xprt);
3011 xprt->prot = IPPROTO_TCP;
3012 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3014 xprt->bind_timeout = XS_BIND_TO;
3015 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3016 xprt->idle_timeout = XS_IDLE_DISC_TO;
3018 xprt->ops = &xs_tcp_ops;
3019 xprt->timeout = &xs_tcp_default_timeout;
3021 xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3022 xprt->connect_timeout = xprt->timeout->to_initval *
3023 (xprt->timeout->to_retries + 1);
3025 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3026 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
3028 switch (addr->sa_family) {
3030 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3031 xprt_set_bound(xprt);
3033 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3036 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3037 xprt_set_bound(xprt);
3039 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3042 ret = ERR_PTR(-EAFNOSUPPORT);
3046 if (xprt_bound(xprt))
3047 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3048 xprt->address_strings[RPC_DISPLAY_ADDR],
3049 xprt->address_strings[RPC_DISPLAY_PORT],
3050 xprt->address_strings[RPC_DISPLAY_PROTO]);
3052 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
3053 xprt->address_strings[RPC_DISPLAY_ADDR],
3054 xprt->address_strings[RPC_DISPLAY_PROTO]);
3056 if (try_module_get(THIS_MODULE))
3058 ret = ERR_PTR(-EINVAL);
3065 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3066 * @args: rpc transport creation arguments
3069 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3071 struct sockaddr *addr = args->dstaddr;
3072 struct rpc_xprt *xprt;
3073 struct sock_xprt *transport;
3074 struct svc_sock *bc_sock;
3075 struct rpc_xprt *ret;
3077 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3078 xprt_tcp_slot_table_entries);
3081 transport = container_of(xprt, struct sock_xprt, xprt);
3083 xprt->prot = IPPROTO_TCP;
3084 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3085 xprt->timeout = &xs_tcp_default_timeout;
3088 xprt_set_bound(xprt);
3089 xprt->bind_timeout = 0;
3090 xprt->reestablish_timeout = 0;
3091 xprt->idle_timeout = 0;
3093 xprt->ops = &bc_tcp_ops;
3095 switch (addr->sa_family) {
3097 xs_format_peer_addresses(xprt, "tcp",
3101 xs_format_peer_addresses(xprt, "tcp",
3102 RPCBIND_NETID_TCP6);
3105 ret = ERR_PTR(-EAFNOSUPPORT);
3109 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3110 xprt->address_strings[RPC_DISPLAY_ADDR],
3111 xprt->address_strings[RPC_DISPLAY_PORT],
3112 xprt->address_strings[RPC_DISPLAY_PROTO]);
3115 * Once we've associated a backchannel xprt with a connection,
3116 * we want to keep it around as long as the connection lasts,
3117 * in case we need to start using it for a backchannel again;
3118 * this reference won't be dropped until bc_xprt is destroyed.
3121 args->bc_xprt->xpt_bc_xprt = xprt;
3122 xprt->bc_xprt = args->bc_xprt;
3123 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3124 transport->sock = bc_sock->sk_sock;
3125 transport->inet = bc_sock->sk_sk;
3128 * Since we don't want connections for the backchannel, we set
3129 * the xprt status to connected
3131 xprt_set_connected(xprt);
3133 if (try_module_get(THIS_MODULE))
3136 args->bc_xprt->xpt_bc_xprt = NULL;
3137 args->bc_xprt->xpt_bc_xps = NULL;
3139 ret = ERR_PTR(-EINVAL);
3145 static struct xprt_class xs_local_transport = {
3146 .list = LIST_HEAD_INIT(xs_local_transport.list),
3147 .name = "named UNIX socket",
3148 .owner = THIS_MODULE,
3149 .ident = XPRT_TRANSPORT_LOCAL,
3150 .setup = xs_setup_local,
3153 static struct xprt_class xs_udp_transport = {
3154 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3156 .owner = THIS_MODULE,
3157 .ident = XPRT_TRANSPORT_UDP,
3158 .setup = xs_setup_udp,
3161 static struct xprt_class xs_tcp_transport = {
3162 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3164 .owner = THIS_MODULE,
3165 .ident = XPRT_TRANSPORT_TCP,
3166 .setup = xs_setup_tcp,
3169 static struct xprt_class xs_bc_tcp_transport = {
3170 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3171 .name = "tcp NFSv4.1 backchannel",
3172 .owner = THIS_MODULE,
3173 .ident = XPRT_TRANSPORT_BC_TCP,
3174 .setup = xs_setup_bc_tcp,
3178 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3181 int init_socket_xprt(void)
3183 if (!sunrpc_table_header)
3184 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3186 xprt_register_transport(&xs_local_transport);
3187 xprt_register_transport(&xs_udp_transport);
3188 xprt_register_transport(&xs_tcp_transport);
3189 xprt_register_transport(&xs_bc_tcp_transport);
3195 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3198 void cleanup_socket_xprt(void)
3200 if (sunrpc_table_header) {
3201 unregister_sysctl_table(sunrpc_table_header);
3202 sunrpc_table_header = NULL;
3205 xprt_unregister_transport(&xs_local_transport);
3206 xprt_unregister_transport(&xs_udp_transport);
3207 xprt_unregister_transport(&xs_tcp_transport);
3208 xprt_unregister_transport(&xs_bc_tcp_transport);
3211 static int param_set_uint_minmax(const char *val,
3212 const struct kernel_param *kp,
3213 unsigned int min, unsigned int max)
3220 ret = kstrtouint(val, 0, &num);
3223 if (num < min || num > max)
3225 *((unsigned int *)kp->arg) = num;
3229 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3231 return param_set_uint_minmax(val, kp,
3236 static const struct kernel_param_ops param_ops_portnr = {
3237 .set = param_set_portnr,
3238 .get = param_get_uint,
3241 #define param_check_portnr(name, p) \
3242 __param_check(name, p, unsigned int);
3244 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3245 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3247 static int param_set_slot_table_size(const char *val,
3248 const struct kernel_param *kp)
3250 return param_set_uint_minmax(val, kp,
3252 RPC_MAX_SLOT_TABLE);
3255 static const struct kernel_param_ops param_ops_slot_table_size = {
3256 .set = param_set_slot_table_size,
3257 .get = param_get_uint,
3260 #define param_check_slot_table_size(name, p) \
3261 __param_check(name, p, unsigned int);
3263 static int param_set_max_slot_table_size(const char *val,
3264 const struct kernel_param *kp)
3266 return param_set_uint_minmax(val, kp,
3268 RPC_MAX_SLOT_TABLE_LIMIT);
3271 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3272 .set = param_set_max_slot_table_size,
3273 .get = param_get_uint,
3276 #define param_check_max_slot_table_size(name, p) \
3277 __param_check(name, p, unsigned int);
3279 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3280 slot_table_size, 0644);
3281 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3282 max_slot_table_size, 0644);
3283 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3284 slot_table_size, 0644);