1 // SPDX-License-Identifier: GPL-2.0
3 * linux/net/sunrpc/xprtsock.c
5 * Client-side transport implementation for sockets.
7 * TCP callback races fixes (C) 1998 Red Hat
8 * TCP send fixes (C) 1998 Red Hat
9 * TCP NFS related read + write fixes
10 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
12 * Rewrite of larges part of the code in order to stabilize TCP stuff.
13 * Fix behaviour when socket buffer is full.
14 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
16 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
18 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
19 * <gilles.quillard@bull.net>
22 #include <linux/types.h>
23 #include <linux/string.h>
24 #include <linux/slab.h>
25 #include <linux/module.h>
26 #include <linux/capability.h>
27 #include <linux/pagemap.h>
28 #include <linux/errno.h>
29 #include <linux/socket.h>
31 #include <linux/net.h>
34 #include <linux/udp.h>
35 #include <linux/tcp.h>
36 #include <linux/sunrpc/clnt.h>
37 #include <linux/sunrpc/addr.h>
38 #include <linux/sunrpc/sched.h>
39 #include <linux/sunrpc/svcsock.h>
40 #include <linux/sunrpc/xprtsock.h>
41 #include <linux/file.h>
42 #ifdef CONFIG_SUNRPC_BACKCHANNEL
43 #include <linux/sunrpc/bc_xprt.h>
47 #include <net/checksum.h>
50 #include <linux/bvec.h>
51 #include <linux/uio.h>
53 #include <trace/events/sunrpc.h>
57 static void xs_close(struct rpc_xprt *xprt);
58 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
64 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
65 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
66 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
68 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
69 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
71 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
73 #define XS_TCP_LINGER_TO (15U * HZ)
74 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
77 * We can register our own files under /proc/sys/sunrpc by
78 * calling register_sysctl_table() again. The files in that
79 * directory become the union of all files registered there.
81 * We simply need to make sure that we don't collide with
82 * someone else's file names!
85 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
86 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
87 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
88 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
89 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
91 static struct ctl_table_header *sunrpc_table_header;
94 * FIXME: changing the UDP slot table size should also resize the UDP
95 * socket buffers for existing UDP transports
97 static struct ctl_table xs_tunables_table[] = {
99 .procname = "udp_slot_table_entries",
100 .data = &xprt_udp_slot_table_entries,
101 .maxlen = sizeof(unsigned int),
103 .proc_handler = proc_dointvec_minmax,
104 .extra1 = &min_slot_table_size,
105 .extra2 = &max_slot_table_size
108 .procname = "tcp_slot_table_entries",
109 .data = &xprt_tcp_slot_table_entries,
110 .maxlen = sizeof(unsigned int),
112 .proc_handler = proc_dointvec_minmax,
113 .extra1 = &min_slot_table_size,
114 .extra2 = &max_slot_table_size
117 .procname = "tcp_max_slot_table_entries",
118 .data = &xprt_max_tcp_slot_table_entries,
119 .maxlen = sizeof(unsigned int),
121 .proc_handler = proc_dointvec_minmax,
122 .extra1 = &min_slot_table_size,
123 .extra2 = &max_tcp_slot_table_limit
126 .procname = "min_resvport",
127 .data = &xprt_min_resvport,
128 .maxlen = sizeof(unsigned int),
130 .proc_handler = proc_dointvec_minmax,
131 .extra1 = &xprt_min_resvport_limit,
132 .extra2 = &xprt_max_resvport_limit
135 .procname = "max_resvport",
136 .data = &xprt_max_resvport,
137 .maxlen = sizeof(unsigned int),
139 .proc_handler = proc_dointvec_minmax,
140 .extra1 = &xprt_min_resvport_limit,
141 .extra2 = &xprt_max_resvport_limit
144 .procname = "tcp_fin_timeout",
145 .data = &xs_tcp_fin_timeout,
146 .maxlen = sizeof(xs_tcp_fin_timeout),
148 .proc_handler = proc_dointvec_jiffies,
153 static struct ctl_table sunrpc_table[] = {
155 .procname = "sunrpc",
157 .child = xs_tunables_table
165 * Wait duration for a reply from the RPC portmapper.
167 #define XS_BIND_TO (60U * HZ)
170 * Delay if a UDP socket connect error occurs. This is most likely some
171 * kind of resource problem on the local host.
173 #define XS_UDP_REEST_TO (2U * HZ)
176 * The reestablish timeout allows clients to delay for a bit before attempting
177 * to reconnect to a server that just dropped our connection.
179 * We implement an exponential backoff when trying to reestablish a TCP
180 * transport connection with the server. Some servers like to drop a TCP
181 * connection when they are overworked, so we start with a short timeout and
182 * increase over time if the server is down or not responding.
184 #define XS_TCP_INIT_REEST_TO (3U * HZ)
187 * TCP idle timeout; client drops the transport socket if it is idle
188 * for this long. Note that we also timeout UDP sockets to prevent
189 * holding port numbers when there is no RPC traffic.
191 #define XS_IDLE_DISC_TO (5U * 60 * HZ)
193 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
194 # undef RPC_DEBUG_DATA
195 # define RPCDBG_FACILITY RPCDBG_TRANS
198 #ifdef RPC_DEBUG_DATA
199 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
201 u8 *buf = (u8 *) packet;
204 dprintk("RPC: %s\n", msg);
205 for (j = 0; j < count && j < 128; j += 4) {
209 dprintk("0x%04x ", j);
211 dprintk("%02x%02x%02x%02x ",
212 buf[j], buf[j+1], buf[j+2], buf[j+3]);
217 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
223 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
225 return (struct rpc_xprt *) sk->sk_user_data;
228 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
230 return (struct sockaddr *) &xprt->addr;
233 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
235 return (struct sockaddr_un *) &xprt->addr;
238 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
240 return (struct sockaddr_in *) &xprt->addr;
243 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
245 return (struct sockaddr_in6 *) &xprt->addr;
248 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
250 struct sockaddr *sap = xs_addr(xprt);
251 struct sockaddr_in6 *sin6;
252 struct sockaddr_in *sin;
253 struct sockaddr_un *sun;
256 switch (sap->sa_family) {
258 sun = xs_addr_un(xprt);
259 strlcpy(buf, sun->sun_path, sizeof(buf));
260 xprt->address_strings[RPC_DISPLAY_ADDR] =
261 kstrdup(buf, GFP_KERNEL);
264 (void)rpc_ntop(sap, buf, sizeof(buf));
265 xprt->address_strings[RPC_DISPLAY_ADDR] =
266 kstrdup(buf, GFP_KERNEL);
267 sin = xs_addr_in(xprt);
268 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
271 (void)rpc_ntop(sap, buf, sizeof(buf));
272 xprt->address_strings[RPC_DISPLAY_ADDR] =
273 kstrdup(buf, GFP_KERNEL);
274 sin6 = xs_addr_in6(xprt);
275 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
281 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
284 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
286 struct sockaddr *sap = xs_addr(xprt);
289 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
290 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
292 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
293 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
296 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
297 const char *protocol,
300 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
301 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
302 xs_format_common_peer_addresses(xprt);
303 xs_format_common_peer_ports(xprt);
306 static void xs_update_peer_port(struct rpc_xprt *xprt)
308 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
309 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
311 xs_format_common_peer_ports(xprt);
314 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
318 for (i = 0; i < RPC_DISPLAY_MAX; i++)
320 case RPC_DISPLAY_PROTO:
321 case RPC_DISPLAY_NETID:
324 kfree(xprt->address_strings[i]);
329 xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp)
333 if (!(buf->flags & XDRBUF_SPARSE_PAGES))
335 if (want > buf->page_len)
336 want = buf->page_len;
337 n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT;
338 for (i = 0; i < n; i++) {
341 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp);
342 if (!buf->pages[i]) {
343 buf->page_len = (i * PAGE_SIZE) - buf->page_base;
344 return buf->page_len;
351 xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek)
355 iov_iter_advance(&msg->msg_iter, seek);
356 ret = sock_recvmsg(sock, msg, flags);
357 return ret > 0 ? ret + seek : ret;
361 xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags,
362 struct kvec *kvec, size_t count, size_t seek)
364 iov_iter_kvec(&msg->msg_iter, READ, kvec, 1, count);
365 return xs_sock_recvmsg(sock, msg, flags, seek);
369 xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags,
370 struct bio_vec *bvec, unsigned long nr, size_t count,
373 iov_iter_bvec(&msg->msg_iter, READ, bvec, nr, count);
374 return xs_sock_recvmsg(sock, msg, flags, seek);
378 xs_read_discard(struct socket *sock, struct msghdr *msg, int flags,
381 struct kvec kvec = { 0 };
382 return xs_read_kvec(sock, msg, flags | MSG_TRUNC, &kvec, count, 0);
386 xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
387 struct xdr_buf *buf, size_t count, size_t seek, size_t *read)
389 size_t want, seek_init = seek, offset = 0;
392 if (seek < buf->head[0].iov_len) {
393 want = min_t(size_t, count, buf->head[0].iov_len);
394 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek);
398 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
404 seek -= buf->head[0].iov_len;
405 offset += buf->head[0].iov_len;
407 if (seek < buf->page_len) {
408 want = xs_alloc_sparse_pages(buf,
409 min_t(size_t, count - offset, buf->page_len),
411 ret = xs_read_bvec(sock, msg, flags, buf->bvec,
412 xdr_buf_pagecount(buf),
413 want + buf->page_base,
414 seek + buf->page_base);
417 offset += ret - buf->page_base;
418 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
424 seek -= buf->page_len;
425 offset += buf->page_len;
427 if (seek < buf->tail[0].iov_len) {
428 want = min_t(size_t, count - offset, buf->tail[0].iov_len);
429 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek);
433 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
438 offset += buf->tail[0].iov_len;
440 msg->msg_flags |= MSG_TRUNC;
442 *read = offset - seek_init;
453 xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf)
455 if (!transport->recv.copied) {
456 if (buf->head[0].iov_len >= transport->recv.offset)
457 memcpy(buf->head[0].iov_base,
458 &transport->recv.xid,
459 transport->recv.offset);
460 transport->recv.copied = transport->recv.offset;
465 xs_read_stream_request_done(struct sock_xprt *transport)
467 return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT);
471 xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg,
472 int flags, struct rpc_rqst *req)
474 struct xdr_buf *buf = &req->rq_private_buf;
478 xs_read_header(transport, buf);
480 want = transport->recv.len - transport->recv.offset;
481 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf,
482 transport->recv.copied + want, transport->recv.copied,
484 transport->recv.offset += read;
485 transport->recv.copied += read;
486 if (transport->recv.offset == transport->recv.len) {
487 if (xs_read_stream_request_done(transport))
488 msg->msg_flags |= MSG_EOR;
489 return transport->recv.copied;
494 return transport->recv.copied;
505 xs_read_stream_headersize(bool isfrag)
508 return sizeof(__be32);
509 return 3 * sizeof(__be32);
513 xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg,
514 int flags, size_t want, size_t seek)
517 .iov_base = &transport->recv.fraghdr,
520 return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek);
523 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
525 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
527 struct rpc_xprt *xprt = &transport->xprt;
528 struct rpc_rqst *req;
531 /* Look up and lock the request corresponding to the given XID */
532 req = xprt_lookup_bc_request(xprt, transport->recv.xid);
534 printk(KERN_WARNING "Callback slot table overflowed\n");
538 ret = xs_read_stream_request(transport, msg, flags, req);
539 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
540 xprt_complete_bc_request(req, ret);
544 #else /* CONFIG_SUNRPC_BACKCHANNEL */
546 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
550 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
553 xs_read_stream_reply(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 spin_lock(&xprt->queue_lock);
561 req = xprt_lookup_rqst(xprt, transport->recv.xid);
563 msg->msg_flags |= MSG_TRUNC;
567 spin_unlock(&xprt->queue_lock);
569 ret = xs_read_stream_request(transport, msg, flags, req);
571 spin_lock(&xprt->queue_lock);
572 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
573 xprt_complete_rqst(req->rq_task, ret);
574 xprt_unpin_rqst(req);
576 spin_unlock(&xprt->queue_lock);
581 xs_read_stream(struct sock_xprt *transport, int flags)
583 struct msghdr msg = { 0 };
584 size_t want, read = 0;
587 if (transport->recv.len == 0) {
588 want = xs_read_stream_headersize(transport->recv.copied != 0);
589 ret = xs_read_stream_header(transport, &msg, flags, want,
590 transport->recv.offset);
593 transport->recv.offset = ret;
598 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) &
599 RPC_FRAGMENT_SIZE_MASK;
600 transport->recv.offset -= sizeof(transport->recv.fraghdr);
604 switch (be32_to_cpu(transport->recv.calldir)) {
606 ret = xs_read_stream_call(transport, &msg, flags);
609 ret = xs_read_stream_reply(transport, &msg, flags);
611 if (msg.msg_flags & MSG_TRUNC) {
612 transport->recv.calldir = cpu_to_be32(-1);
613 transport->recv.copied = -1;
618 if (transport->recv.offset < transport->recv.len) {
619 ret = xs_read_discard(transport->sock, &msg, flags,
620 transport->recv.len - transport->recv.offset);
623 transport->recv.offset += ret;
625 if (transport->recv.offset != transport->recv.len)
628 if (xs_read_stream_request_done(transport)) {
629 trace_xs_stream_read_request(transport);
630 transport->recv.copied = 0;
632 transport->recv.offset = 0;
633 transport->recv.len = 0;
639 xprt_force_disconnect(&transport->xprt);
645 static void xs_stream_data_receive(struct sock_xprt *transport)
650 mutex_lock(&transport->recv_mutex);
651 if (transport->sock == NULL)
653 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
655 ret = xs_read_stream(transport, MSG_DONTWAIT);
662 mutex_unlock(&transport->recv_mutex);
663 trace_xs_stream_read_data(&transport->xprt, ret, read);
666 static void xs_stream_data_receive_workfn(struct work_struct *work)
668 struct sock_xprt *transport =
669 container_of(work, struct sock_xprt, recv_worker);
670 xs_stream_data_receive(transport);
674 xs_stream_reset_connect(struct sock_xprt *transport)
676 transport->recv.offset = 0;
677 transport->recv.len = 0;
678 transport->recv.copied = 0;
679 transport->xmit.offset = 0;
680 transport->xprt.stat.connect_count++;
681 transport->xprt.stat.connect_start = jiffies;
684 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
686 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
688 struct msghdr msg = {
690 .msg_namelen = addrlen,
691 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
694 .iov_base = vec->iov_base + base,
695 .iov_len = vec->iov_len - base,
698 if (iov.iov_len != 0)
699 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
700 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
703 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p)
705 ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
706 int offset, size_t size, int flags);
708 unsigned int remainder;
711 remainder = xdr->page_len - base;
712 base += xdr->page_base;
713 ppage = xdr->pages + (base >> PAGE_SHIFT);
715 do_sendpage = sock->ops->sendpage;
717 do_sendpage = sock_no_sendpage;
719 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
720 int flags = XS_SENDMSG_FLAGS;
726 flags |= MSG_SENDPAGE_NOTLAST | MSG_MORE;
727 err = do_sendpage(sock, *ppage, base, len, flags);
728 if (remainder == 0 || err != len)
742 * xs_sendpages - write pages directly to a socket
743 * @sock: socket to send on
744 * @addr: UDP only -- address of destination
745 * @addrlen: UDP only -- length of destination address
746 * @xdr: buffer containing this request
747 * @base: starting position in the buffer
748 * @zerocopy: true if it is safe to use sendpage()
749 * @sent_p: return the total number of bytes successfully queued for sending
752 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p)
754 unsigned int remainder = xdr->len - base;
766 if (base < xdr->head[0].iov_len || addr != NULL) {
767 unsigned int len = xdr->head[0].iov_len - base;
769 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
770 if (remainder == 0 || err != len)
775 base -= xdr->head[0].iov_len;
777 if (base < xdr->page_len) {
778 unsigned int len = xdr->page_len - base;
780 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy, &sent);
782 if (remainder == 0 || sent != len)
786 base -= xdr->page_len;
788 if (base >= xdr->tail[0].iov_len)
790 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
800 * xs_nospace - handle transmit was incomplete
801 * @req: pointer to RPC request
804 static int xs_nospace(struct rpc_rqst *req)
806 struct rpc_xprt *xprt = req->rq_xprt;
807 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
808 struct sock *sk = transport->inet;
811 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
812 req->rq_task->tk_pid,
813 req->rq_slen - transport->xmit.offset,
816 /* Protect against races with write_space */
817 spin_lock_bh(&xprt->transport_lock);
819 /* Don't race with disconnect */
820 if (xprt_connected(xprt)) {
821 /* wait for more buffer space */
822 sk->sk_write_pending++;
823 xprt_wait_for_buffer_space(xprt);
827 spin_unlock_bh(&xprt->transport_lock);
829 /* Race breaker in case memory is freed before above code is called */
830 if (ret == -EAGAIN) {
831 struct socket_wq *wq;
834 wq = rcu_dereference(sk->sk_wq);
835 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
838 sk->sk_write_space(sk);
844 xs_stream_prepare_request(struct rpc_rqst *req)
846 req->rq_task->tk_status = xdr_alloc_bvec(&req->rq_rcv_buf, GFP_NOIO);
850 * Determine if the previous message in the stream was aborted before it
851 * could complete transmission.
854 xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req)
856 return transport->xmit.offset != 0 && req->rq_bytes_sent == 0;
860 * Construct a stream transport record marker in @buf.
862 static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
864 u32 reclen = buf->len - sizeof(rpc_fraghdr);
865 rpc_fraghdr *base = buf->head[0].iov_base;
866 *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
870 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
871 * @req: pointer to RPC request
874 * 0: The request has been sent
875 * EAGAIN: The socket was blocked, please call again later to
876 * complete the request
877 * ENOTCONN: Caller needs to invoke connect logic then call again
878 * other: Some other error occured, the request was not sent
880 static int xs_local_send_request(struct rpc_rqst *req)
882 struct rpc_xprt *xprt = req->rq_xprt;
883 struct sock_xprt *transport =
884 container_of(xprt, struct sock_xprt, xprt);
885 struct xdr_buf *xdr = &req->rq_snd_buf;
889 /* Close the stream if the previous transmission was incomplete */
890 if (xs_send_request_was_aborted(transport, req)) {
895 xs_encode_stream_record_marker(&req->rq_snd_buf);
897 xs_pktdump("packet data:",
898 req->rq_svec->iov_base, req->rq_svec->iov_len);
900 req->rq_xtime = ktime_get();
901 status = xs_sendpages(transport->sock, NULL, 0, xdr,
902 transport->xmit.offset,
904 dprintk("RPC: %s(%u) = %d\n",
905 __func__, xdr->len - transport->xmit.offset, status);
907 if (status == -EAGAIN && sock_writeable(transport->inet))
910 if (likely(sent > 0) || status == 0) {
911 transport->xmit.offset += sent;
912 req->rq_bytes_sent = transport->xmit.offset;
913 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
914 req->rq_xmit_bytes_sent += transport->xmit.offset;
915 req->rq_bytes_sent = 0;
916 transport->xmit.offset = 0;
926 status = xs_nospace(req);
929 dprintk("RPC: sendmsg returned unrecognized error %d\n",
941 * xs_udp_send_request - write an RPC request to a UDP socket
942 * @req: pointer to RPC request
945 * 0: The request has been sent
946 * EAGAIN: The socket was blocked, please call again later to
947 * complete the request
948 * ENOTCONN: Caller needs to invoke connect logic then call again
949 * other: Some other error occurred, the request was not sent
951 static int xs_udp_send_request(struct rpc_rqst *req)
953 struct rpc_xprt *xprt = req->rq_xprt;
954 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
955 struct xdr_buf *xdr = &req->rq_snd_buf;
959 xs_pktdump("packet data:",
960 req->rq_svec->iov_base,
961 req->rq_svec->iov_len);
963 if (!xprt_bound(xprt))
966 if (!xprt_request_get_cong(xprt, req))
969 req->rq_xtime = ktime_get();
970 status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
971 xdr, 0, true, &sent);
973 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
976 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
977 if (status == -EPERM)
980 if (status == -EAGAIN && sock_writeable(transport->inet))
983 if (sent > 0 || status == 0) {
984 req->rq_xmit_bytes_sent += sent;
985 if (sent >= req->rq_slen)
987 /* Still some bytes left; set up for a retry later. */
995 /* Should we call xs_close() here? */
998 status = xs_nospace(req);
1005 /* When the server has died, an ICMP port unreachable message
1006 * prompts ECONNREFUSED. */
1009 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1017 * xs_tcp_send_request - write an RPC request to a TCP socket
1018 * @req: pointer to RPC request
1021 * 0: The request has been sent
1022 * EAGAIN: The socket was blocked, please call again later to
1023 * complete the request
1024 * ENOTCONN: Caller needs to invoke connect logic then call again
1025 * other: Some other error occurred, the request was not sent
1027 * XXX: In the case of soft timeouts, should we eventually give up
1028 * if sendmsg is not able to make progress?
1030 static int xs_tcp_send_request(struct rpc_rqst *req)
1032 struct rpc_xprt *xprt = req->rq_xprt;
1033 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1034 struct xdr_buf *xdr = &req->rq_snd_buf;
1035 bool zerocopy = true;
1036 bool vm_wait = false;
1040 /* Close the stream if the previous transmission was incomplete */
1041 if (xs_send_request_was_aborted(transport, req)) {
1042 if (transport->sock != NULL)
1043 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
1047 xs_encode_stream_record_marker(&req->rq_snd_buf);
1049 xs_pktdump("packet data:",
1050 req->rq_svec->iov_base,
1051 req->rq_svec->iov_len);
1052 /* Don't use zero copy if this is a resend. If the RPC call
1053 * completes while the socket holds a reference to the pages,
1054 * then we may end up resending corrupted data.
1056 if (req->rq_task->tk_flags & RPC_TASK_SENT)
1059 if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
1060 xs_tcp_set_socket_timeouts(xprt, transport->sock);
1062 /* Continue transmitting the packet/record. We must be careful
1063 * to cope with writespace callbacks arriving _after_ we have
1064 * called sendmsg(). */
1065 req->rq_xtime = ktime_get();
1068 status = xs_sendpages(transport->sock, NULL, 0, xdr,
1069 transport->xmit.offset,
1072 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
1073 xdr->len - transport->xmit.offset, status);
1075 /* If we've sent the entire packet, immediately
1076 * reset the count of bytes sent. */
1077 transport->xmit.offset += sent;
1078 req->rq_bytes_sent = transport->xmit.offset;
1079 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
1080 req->rq_xmit_bytes_sent += transport->xmit.offset;
1081 req->rq_bytes_sent = 0;
1082 transport->xmit.offset = 0;
1086 WARN_ON_ONCE(sent == 0 && status == 0);
1088 if (status == -EAGAIN ) {
1090 * Return EAGAIN if we're sure we're hitting the
1091 * socket send buffer limits.
1093 if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
1096 * Did we hit a memory allocation failure?
1102 /* Retry, knowing now that we're below the
1103 * socket send buffer limit
1117 /* Should we call xs_close() here? */
1120 status = xs_nospace(req);
1130 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1137 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1139 transport->old_data_ready = sk->sk_data_ready;
1140 transport->old_state_change = sk->sk_state_change;
1141 transport->old_write_space = sk->sk_write_space;
1142 transport->old_error_report = sk->sk_error_report;
1145 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1147 sk->sk_data_ready = transport->old_data_ready;
1148 sk->sk_state_change = transport->old_state_change;
1149 sk->sk_write_space = transport->old_write_space;
1150 sk->sk_error_report = transport->old_error_report;
1153 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
1155 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1157 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1160 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1162 smp_mb__before_atomic();
1163 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1164 clear_bit(XPRT_CLOSING, &xprt->state);
1165 xs_sock_reset_state_flags(xprt);
1166 smp_mb__after_atomic();
1170 * xs_error_report - callback to handle TCP socket state errors
1173 * Note: we don't call sock_error() since there may be a rpc_task
1174 * using the socket, and so we don't want to clear sk->sk_err.
1176 static void xs_error_report(struct sock *sk)
1178 struct rpc_xprt *xprt;
1181 read_lock_bh(&sk->sk_callback_lock);
1182 if (!(xprt = xprt_from_sock(sk)))
1188 dprintk("RPC: xs_error_report client %p, error=%d...\n",
1190 trace_rpc_socket_error(xprt, sk->sk_socket, err);
1191 xprt_wake_pending_tasks(xprt, err);
1193 read_unlock_bh(&sk->sk_callback_lock);
1196 static void xs_reset_transport(struct sock_xprt *transport)
1198 struct socket *sock = transport->sock;
1199 struct sock *sk = transport->inet;
1200 struct rpc_xprt *xprt = &transport->xprt;
1205 if (atomic_read(&transport->xprt.swapper))
1206 sk_clear_memalloc(sk);
1208 kernel_sock_shutdown(sock, SHUT_RDWR);
1210 mutex_lock(&transport->recv_mutex);
1211 write_lock_bh(&sk->sk_callback_lock);
1212 transport->inet = NULL;
1213 transport->sock = NULL;
1215 sk->sk_user_data = NULL;
1217 xs_restore_old_callbacks(transport, sk);
1218 xprt_clear_connected(xprt);
1219 write_unlock_bh(&sk->sk_callback_lock);
1220 xs_sock_reset_connection_flags(xprt);
1221 mutex_unlock(&transport->recv_mutex);
1223 trace_rpc_socket_close(xprt, sock);
1228 * xs_close - close a socket
1231 * This is used when all requests are complete; ie, no DRC state remains
1232 * on the server we want to save.
1234 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1235 * xs_reset_transport() zeroing the socket from underneath a writer.
1237 static void xs_close(struct rpc_xprt *xprt)
1239 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1241 dprintk("RPC: xs_close xprt %p\n", xprt);
1243 xs_reset_transport(transport);
1244 xprt->reestablish_timeout = 0;
1246 xprt_disconnect_done(xprt);
1249 static void xs_inject_disconnect(struct rpc_xprt *xprt)
1251 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
1253 xprt_disconnect_done(xprt);
1256 static void xs_xprt_free(struct rpc_xprt *xprt)
1258 xs_free_peer_addresses(xprt);
1263 * xs_destroy - prepare to shutdown a transport
1264 * @xprt: doomed transport
1267 static void xs_destroy(struct rpc_xprt *xprt)
1269 struct sock_xprt *transport = container_of(xprt,
1270 struct sock_xprt, xprt);
1271 dprintk("RPC: xs_destroy xprt %p\n", xprt);
1273 cancel_delayed_work_sync(&transport->connect_worker);
1275 cancel_work_sync(&transport->recv_worker);
1277 module_put(THIS_MODULE);
1281 * xs_udp_data_read_skb - receive callback for UDP sockets
1287 static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1289 struct sk_buff *skb)
1291 struct rpc_task *task;
1292 struct rpc_rqst *rovr;
1293 int repsize, copied;
1299 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
1303 /* Copy the XID from the skb... */
1304 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1308 /* Look up and lock the request corresponding to the given XID */
1309 spin_lock(&xprt->queue_lock);
1310 rovr = xprt_lookup_rqst(xprt, *xp);
1313 xprt_pin_rqst(rovr);
1314 xprt_update_rtt(rovr->rq_task);
1315 spin_unlock(&xprt->queue_lock);
1316 task = rovr->rq_task;
1318 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1321 /* Suck it into the iovec, verify checksum if not done by hw. */
1322 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1323 spin_lock(&xprt->queue_lock);
1324 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1329 spin_lock_bh(&xprt->transport_lock);
1330 xprt_adjust_cwnd(xprt, task, copied);
1331 spin_unlock_bh(&xprt->transport_lock);
1332 spin_lock(&xprt->queue_lock);
1333 xprt_complete_rqst(task, copied);
1334 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1336 xprt_unpin_rqst(rovr);
1338 spin_unlock(&xprt->queue_lock);
1341 static void xs_udp_data_receive(struct sock_xprt *transport)
1343 struct sk_buff *skb;
1347 mutex_lock(&transport->recv_mutex);
1348 sk = transport->inet;
1351 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1353 skb = skb_recv_udp(sk, 0, 1, &err);
1356 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1361 mutex_unlock(&transport->recv_mutex);
1364 static void xs_udp_data_receive_workfn(struct work_struct *work)
1366 struct sock_xprt *transport =
1367 container_of(work, struct sock_xprt, recv_worker);
1368 xs_udp_data_receive(transport);
1372 * xs_data_ready - "data ready" callback for UDP sockets
1373 * @sk: socket with data to read
1376 static void xs_data_ready(struct sock *sk)
1378 struct rpc_xprt *xprt;
1380 read_lock_bh(&sk->sk_callback_lock);
1381 dprintk("RPC: xs_data_ready...\n");
1382 xprt = xprt_from_sock(sk);
1384 struct sock_xprt *transport = container_of(xprt,
1385 struct sock_xprt, xprt);
1386 transport->old_data_ready(sk);
1387 /* Any data means we had a useful conversation, so
1388 * then we don't need to delay the next reconnect
1390 if (xprt->reestablish_timeout)
1391 xprt->reestablish_timeout = 0;
1392 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1393 queue_work(xprtiod_workqueue, &transport->recv_worker);
1395 read_unlock_bh(&sk->sk_callback_lock);
1399 * Helper function to force a TCP close if the server is sending
1400 * junk and/or it has put us in CLOSE_WAIT
1402 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1404 xprt_force_disconnect(xprt);
1407 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1408 static int xs_tcp_bc_up(struct svc_serv *serv, struct net *net)
1412 ret = svc_create_xprt(serv, "tcp-bc", net, PF_INET, 0,
1413 SVC_SOCK_ANONYMOUS);
1419 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1423 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1426 * xs_tcp_state_change - callback to handle TCP socket state changes
1427 * @sk: socket whose state has changed
1430 static void xs_tcp_state_change(struct sock *sk)
1432 struct rpc_xprt *xprt;
1433 struct sock_xprt *transport;
1435 read_lock_bh(&sk->sk_callback_lock);
1436 if (!(xprt = xprt_from_sock(sk)))
1438 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1439 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1440 sk->sk_state, xprt_connected(xprt),
1441 sock_flag(sk, SOCK_DEAD),
1442 sock_flag(sk, SOCK_ZAPPED),
1445 transport = container_of(xprt, struct sock_xprt, xprt);
1446 trace_rpc_socket_state_change(xprt, sk->sk_socket);
1447 switch (sk->sk_state) {
1448 case TCP_ESTABLISHED:
1449 spin_lock(&xprt->transport_lock);
1450 if (!xprt_test_and_set_connected(xprt)) {
1451 xprt->connect_cookie++;
1452 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1453 xprt_clear_connecting(xprt);
1455 xprt->stat.connect_count++;
1456 xprt->stat.connect_time += (long)jiffies -
1457 xprt->stat.connect_start;
1458 xprt_wake_pending_tasks(xprt, -EAGAIN);
1460 spin_unlock(&xprt->transport_lock);
1463 /* The client initiated a shutdown of the socket */
1464 xprt->connect_cookie++;
1465 xprt->reestablish_timeout = 0;
1466 set_bit(XPRT_CLOSING, &xprt->state);
1467 smp_mb__before_atomic();
1468 clear_bit(XPRT_CONNECTED, &xprt->state);
1469 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1470 smp_mb__after_atomic();
1472 case TCP_CLOSE_WAIT:
1473 /* The server initiated a shutdown of the socket */
1474 xprt->connect_cookie++;
1475 clear_bit(XPRT_CONNECTED, &xprt->state);
1476 xs_tcp_force_close(xprt);
1480 * If the server closed down the connection, make sure that
1481 * we back off before reconnecting
1483 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1484 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1487 set_bit(XPRT_CLOSING, &xprt->state);
1488 smp_mb__before_atomic();
1489 clear_bit(XPRT_CONNECTED, &xprt->state);
1490 smp_mb__after_atomic();
1493 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1494 &transport->sock_state))
1495 xprt_clear_connecting(xprt);
1496 clear_bit(XPRT_CLOSING, &xprt->state);
1498 xprt_wake_pending_tasks(xprt, -sk->sk_err);
1499 /* Trigger the socket release */
1500 xs_tcp_force_close(xprt);
1503 read_unlock_bh(&sk->sk_callback_lock);
1506 static void xs_write_space(struct sock *sk)
1508 struct socket_wq *wq;
1509 struct rpc_xprt *xprt;
1513 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1515 if (unlikely(!(xprt = xprt_from_sock(sk))))
1518 wq = rcu_dereference(sk->sk_wq);
1519 if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1522 if (xprt_write_space(xprt))
1523 sk->sk_write_pending--;
1529 * xs_udp_write_space - callback invoked when socket buffer space
1531 * @sk: socket whose state has changed
1533 * Called when more output buffer space is available for this socket.
1534 * We try not to wake our writers until they can make "significant"
1535 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1536 * with a bunch of small requests.
1538 static void xs_udp_write_space(struct sock *sk)
1540 read_lock_bh(&sk->sk_callback_lock);
1542 /* from net/core/sock.c:sock_def_write_space */
1543 if (sock_writeable(sk))
1546 read_unlock_bh(&sk->sk_callback_lock);
1550 * xs_tcp_write_space - callback invoked when socket buffer space
1552 * @sk: socket whose state has changed
1554 * Called when more output buffer space is available for this socket.
1555 * We try not to wake our writers until they can make "significant"
1556 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1557 * with a bunch of small requests.
1559 static void xs_tcp_write_space(struct sock *sk)
1561 read_lock_bh(&sk->sk_callback_lock);
1563 /* from net/core/stream.c:sk_stream_write_space */
1564 if (sk_stream_is_writeable(sk))
1567 read_unlock_bh(&sk->sk_callback_lock);
1570 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1572 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1573 struct sock *sk = transport->inet;
1575 if (transport->rcvsize) {
1576 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1577 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1579 if (transport->sndsize) {
1580 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1581 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1582 sk->sk_write_space(sk);
1587 * xs_udp_set_buffer_size - set send and receive limits
1588 * @xprt: generic transport
1589 * @sndsize: requested size of send buffer, in bytes
1590 * @rcvsize: requested size of receive buffer, in bytes
1592 * Set socket send and receive buffer size limits.
1594 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1596 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1598 transport->sndsize = 0;
1600 transport->sndsize = sndsize + 1024;
1601 transport->rcvsize = 0;
1603 transport->rcvsize = rcvsize + 1024;
1605 xs_udp_do_set_buffer_size(xprt);
1609 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1610 * @task: task that timed out
1612 * Adjust the congestion window after a retransmit timeout has occurred.
1614 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1616 spin_lock_bh(&xprt->transport_lock);
1617 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1618 spin_unlock_bh(&xprt->transport_lock);
1621 static int xs_get_random_port(void)
1623 unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1624 unsigned short range;
1625 unsigned short rand;
1629 range = max - min + 1;
1630 rand = (unsigned short) prandom_u32() % range;
1635 * xs_set_reuseaddr_port - set the socket's port and address reuse options
1638 * Note that this function has to be called on all sockets that share the
1639 * same port, and it must be called before binding.
1641 static void xs_sock_set_reuseport(struct socket *sock)
1645 kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1646 (char *)&opt, sizeof(opt));
1649 static unsigned short xs_sock_getport(struct socket *sock)
1651 struct sockaddr_storage buf;
1652 unsigned short port = 0;
1654 if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1656 switch (buf.ss_family) {
1658 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1661 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1668 * xs_set_port - reset the port number in the remote endpoint address
1669 * @xprt: generic transport
1670 * @port: new port number
1673 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1675 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1677 rpc_set_port(xs_addr(xprt), port);
1678 xs_update_peer_port(xprt);
1681 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1683 if (transport->srcport == 0)
1684 transport->srcport = xs_sock_getport(sock);
1687 static int xs_get_srcport(struct sock_xprt *transport)
1689 int port = transport->srcport;
1691 if (port == 0 && transport->xprt.resvport)
1692 port = xs_get_random_port();
1696 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1698 if (transport->srcport != 0)
1699 transport->srcport = 0;
1700 if (!transport->xprt.resvport)
1702 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1703 return xprt_max_resvport;
1706 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1708 struct sockaddr_storage myaddr;
1710 int port = xs_get_srcport(transport);
1711 unsigned short last;
1714 * If we are asking for any ephemeral port (i.e. port == 0 &&
1715 * transport->xprt.resvport == 0), don't bind. Let the local
1716 * port selection happen implicitly when the socket is used
1717 * (for example at connect time).
1719 * This ensures that we can continue to establish TCP
1720 * connections even when all local ephemeral ports are already
1721 * a part of some TCP connection. This makes no difference
1722 * for UDP sockets, but also doens't harm them.
1724 * If we're asking for any reserved port (i.e. port == 0 &&
1725 * transport->xprt.resvport == 1) xs_get_srcport above will
1726 * ensure that port is non-zero and we will bind as needed.
1731 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1733 rpc_set_port((struct sockaddr *)&myaddr, port);
1734 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1735 transport->xprt.addrlen);
1737 transport->srcport = port;
1741 port = xs_next_srcport(transport, port);
1744 } while (err == -EADDRINUSE && nloop != 2);
1746 if (myaddr.ss_family == AF_INET)
1747 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1748 &((struct sockaddr_in *)&myaddr)->sin_addr,
1749 port, err ? "failed" : "ok", err);
1751 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1752 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1753 port, err ? "failed" : "ok", err);
1758 * We don't support autobind on AF_LOCAL sockets
1760 static void xs_local_rpcbind(struct rpc_task *task)
1762 xprt_set_bound(task->tk_xprt);
1765 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1769 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1770 static struct lock_class_key xs_key[2];
1771 static struct lock_class_key xs_slock_key[2];
1773 static inline void xs_reclassify_socketu(struct socket *sock)
1775 struct sock *sk = sock->sk;
1777 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1778 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1781 static inline void xs_reclassify_socket4(struct socket *sock)
1783 struct sock *sk = sock->sk;
1785 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1786 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1789 static inline void xs_reclassify_socket6(struct socket *sock)
1791 struct sock *sk = sock->sk;
1793 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1794 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1797 static inline void xs_reclassify_socket(int family, struct socket *sock)
1799 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1804 xs_reclassify_socketu(sock);
1807 xs_reclassify_socket4(sock);
1810 xs_reclassify_socket6(sock);
1815 static inline void xs_reclassify_socket(int family, struct socket *sock)
1820 static void xs_dummy_setup_socket(struct work_struct *work)
1824 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1825 struct sock_xprt *transport, int family, int type,
1826 int protocol, bool reuseport)
1828 struct socket *sock;
1831 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1833 dprintk("RPC: can't create %d transport socket (%d).\n",
1837 xs_reclassify_socket(family, sock);
1840 xs_sock_set_reuseport(sock);
1842 err = xs_bind(transport, sock);
1850 return ERR_PTR(err);
1853 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1854 struct socket *sock)
1856 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1859 if (!transport->inet) {
1860 struct sock *sk = sock->sk;
1862 write_lock_bh(&sk->sk_callback_lock);
1864 xs_save_old_callbacks(transport, sk);
1866 sk->sk_user_data = xprt;
1867 sk->sk_data_ready = xs_data_ready;
1868 sk->sk_write_space = xs_udp_write_space;
1869 sock_set_flag(sk, SOCK_FASYNC);
1870 sk->sk_error_report = xs_error_report;
1871 sk->sk_allocation = GFP_NOIO;
1873 xprt_clear_connected(xprt);
1875 /* Reset to new socket */
1876 transport->sock = sock;
1877 transport->inet = sk;
1879 write_unlock_bh(&sk->sk_callback_lock);
1882 xs_stream_reset_connect(transport);
1884 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1888 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1889 * @transport: socket transport to connect
1891 static int xs_local_setup_socket(struct sock_xprt *transport)
1893 struct rpc_xprt *xprt = &transport->xprt;
1894 struct socket *sock;
1897 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1898 SOCK_STREAM, 0, &sock, 1);
1900 dprintk("RPC: can't create AF_LOCAL "
1901 "transport socket (%d).\n", -status);
1904 xs_reclassify_socket(AF_LOCAL, sock);
1906 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1907 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1909 status = xs_local_finish_connecting(xprt, sock);
1910 trace_rpc_socket_connect(xprt, sock, status);
1913 dprintk("RPC: xprt %p connected to %s\n",
1914 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1915 xprt->stat.connect_count++;
1916 xprt->stat.connect_time += (long)jiffies -
1917 xprt->stat.connect_start;
1918 xprt_set_connected(xprt);
1922 dprintk("RPC: xprt %p: socket %s does not exist\n",
1923 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1926 dprintk("RPC: xprt %p: connection refused for %s\n",
1927 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1930 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1932 xprt->address_strings[RPC_DISPLAY_ADDR]);
1936 xprt_clear_connecting(xprt);
1937 xprt_wake_pending_tasks(xprt, status);
1941 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1943 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1946 if (RPC_IS_ASYNC(task)) {
1948 * We want the AF_LOCAL connect to be resolved in the
1949 * filesystem namespace of the process making the rpc
1950 * call. Thus we connect synchronously.
1952 * If we want to support asynchronous AF_LOCAL calls,
1953 * we'll need to figure out how to pass a namespace to
1956 rpc_exit(task, -ENOTCONN);
1959 ret = xs_local_setup_socket(transport);
1960 if (ret && !RPC_IS_SOFTCONN(task))
1961 msleep_interruptible(15000);
1964 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
1966 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
1967 * know that we have exclusive access to the socket), to guard against
1968 * races with xs_reset_transport.
1970 static void xs_set_memalloc(struct rpc_xprt *xprt)
1972 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1976 * If there's no sock, then we have nothing to set. The
1977 * reconnecting process will get it for us.
1979 if (!transport->inet)
1981 if (atomic_read(&xprt->swapper))
1982 sk_set_memalloc(transport->inet);
1986 * xs_enable_swap - Tag this transport as being used for swap.
1987 * @xprt: transport to tag
1989 * Take a reference to this transport on behalf of the rpc_clnt, and
1990 * optionally mark it for swapping if it wasn't already.
1993 xs_enable_swap(struct rpc_xprt *xprt)
1995 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
1997 if (atomic_inc_return(&xprt->swapper) != 1)
1999 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2000 return -ERESTARTSYS;
2002 sk_set_memalloc(xs->inet);
2003 xprt_release_xprt(xprt, NULL);
2008 * xs_disable_swap - Untag this transport as being used for swap.
2009 * @xprt: transport to tag
2011 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2012 * swapper refcount goes to 0, untag the socket as a memalloc socket.
2015 xs_disable_swap(struct rpc_xprt *xprt)
2017 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2019 if (!atomic_dec_and_test(&xprt->swapper))
2021 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2024 sk_clear_memalloc(xs->inet);
2025 xprt_release_xprt(xprt, NULL);
2028 static void xs_set_memalloc(struct rpc_xprt *xprt)
2033 xs_enable_swap(struct rpc_xprt *xprt)
2039 xs_disable_swap(struct rpc_xprt *xprt)
2044 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2046 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2048 if (!transport->inet) {
2049 struct sock *sk = sock->sk;
2051 write_lock_bh(&sk->sk_callback_lock);
2053 xs_save_old_callbacks(transport, sk);
2055 sk->sk_user_data = xprt;
2056 sk->sk_data_ready = xs_data_ready;
2057 sk->sk_write_space = xs_udp_write_space;
2058 sock_set_flag(sk, SOCK_FASYNC);
2059 sk->sk_allocation = GFP_NOIO;
2061 xprt_set_connected(xprt);
2063 /* Reset to new socket */
2064 transport->sock = sock;
2065 transport->inet = sk;
2067 xs_set_memalloc(xprt);
2069 write_unlock_bh(&sk->sk_callback_lock);
2071 xs_udp_do_set_buffer_size(xprt);
2073 xprt->stat.connect_start = jiffies;
2076 static void xs_udp_setup_socket(struct work_struct *work)
2078 struct sock_xprt *transport =
2079 container_of(work, struct sock_xprt, connect_worker.work);
2080 struct rpc_xprt *xprt = &transport->xprt;
2081 struct socket *sock;
2084 sock = xs_create_sock(xprt, transport,
2085 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2086 IPPROTO_UDP, false);
2090 dprintk("RPC: worker connecting xprt %p via %s to "
2091 "%s (port %s)\n", xprt,
2092 xprt->address_strings[RPC_DISPLAY_PROTO],
2093 xprt->address_strings[RPC_DISPLAY_ADDR],
2094 xprt->address_strings[RPC_DISPLAY_PORT]);
2096 xs_udp_finish_connecting(xprt, sock);
2097 trace_rpc_socket_connect(xprt, sock, 0);
2100 xprt_unlock_connect(xprt, transport);
2101 xprt_clear_connecting(xprt);
2102 xprt_wake_pending_tasks(xprt, status);
2106 * xs_tcp_shutdown - gracefully shut down a TCP socket
2109 * Initiates a graceful shutdown of the TCP socket by calling the
2110 * equivalent of shutdown(SHUT_RDWR);
2112 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2114 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2115 struct socket *sock = transport->sock;
2116 int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2122 kernel_sock_shutdown(sock, SHUT_RDWR);
2123 trace_rpc_socket_shutdown(xprt, sock);
2127 xs_reset_transport(transport);
2131 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2132 struct socket *sock)
2134 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2135 unsigned int keepidle;
2136 unsigned int keepcnt;
2137 unsigned int opt_on = 1;
2140 spin_lock_bh(&xprt->transport_lock);
2141 keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2142 keepcnt = xprt->timeout->to_retries + 1;
2143 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2144 (xprt->timeout->to_retries + 1);
2145 clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2146 spin_unlock_bh(&xprt->transport_lock);
2148 /* TCP Keepalive options */
2149 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2150 (char *)&opt_on, sizeof(opt_on));
2151 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2152 (char *)&keepidle, sizeof(keepidle));
2153 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2154 (char *)&keepidle, sizeof(keepidle));
2155 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2156 (char *)&keepcnt, sizeof(keepcnt));
2158 /* TCP user timeout (see RFC5482) */
2159 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2160 (char *)&timeo, sizeof(timeo));
2163 static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2164 unsigned long connect_timeout,
2165 unsigned long reconnect_timeout)
2167 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2168 struct rpc_timeout to;
2169 unsigned long initval;
2171 spin_lock_bh(&xprt->transport_lock);
2172 if (reconnect_timeout < xprt->max_reconnect_timeout)
2173 xprt->max_reconnect_timeout = reconnect_timeout;
2174 if (connect_timeout < xprt->connect_timeout) {
2175 memcpy(&to, xprt->timeout, sizeof(to));
2176 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2177 /* Arbitrary lower limit */
2178 if (initval < XS_TCP_INIT_REEST_TO << 1)
2179 initval = XS_TCP_INIT_REEST_TO << 1;
2180 to.to_initval = initval;
2181 to.to_maxval = initval;
2182 memcpy(&transport->tcp_timeout, &to,
2183 sizeof(transport->tcp_timeout));
2184 xprt->timeout = &transport->tcp_timeout;
2185 xprt->connect_timeout = connect_timeout;
2187 set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2188 spin_unlock_bh(&xprt->transport_lock);
2191 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2193 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2194 int ret = -ENOTCONN;
2196 if (!transport->inet) {
2197 struct sock *sk = sock->sk;
2198 unsigned int addr_pref = IPV6_PREFER_SRC_PUBLIC;
2200 /* Avoid temporary address, they are bad for long-lived
2201 * connections such as NFS mounts.
2202 * RFC4941, section 3.6 suggests that:
2203 * Individual applications, which have specific
2204 * knowledge about the normal duration of connections,
2205 * MAY override this as appropriate.
2207 kernel_setsockopt(sock, SOL_IPV6, IPV6_ADDR_PREFERENCES,
2208 (char *)&addr_pref, sizeof(addr_pref));
2210 xs_tcp_set_socket_timeouts(xprt, sock);
2212 write_lock_bh(&sk->sk_callback_lock);
2214 xs_save_old_callbacks(transport, sk);
2216 sk->sk_user_data = xprt;
2217 sk->sk_data_ready = xs_data_ready;
2218 sk->sk_state_change = xs_tcp_state_change;
2219 sk->sk_write_space = xs_tcp_write_space;
2220 sock_set_flag(sk, SOCK_FASYNC);
2221 sk->sk_error_report = xs_error_report;
2222 sk->sk_allocation = GFP_NOIO;
2224 /* socket options */
2225 sock_reset_flag(sk, SOCK_LINGER);
2226 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2228 xprt_clear_connected(xprt);
2230 /* Reset to new socket */
2231 transport->sock = sock;
2232 transport->inet = sk;
2234 write_unlock_bh(&sk->sk_callback_lock);
2237 if (!xprt_bound(xprt))
2240 xs_set_memalloc(xprt);
2242 /* Reset TCP record info */
2243 xs_stream_reset_connect(transport);
2245 /* Tell the socket layer to start connecting... */
2246 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2247 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2250 xs_set_srcport(transport, sock);
2254 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2255 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2257 case -EADDRNOTAVAIL:
2258 /* Source port number is unavailable. Try a new one! */
2259 transport->srcport = 0;
2266 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2268 * Invoked by a work queue tasklet.
2270 static void xs_tcp_setup_socket(struct work_struct *work)
2272 struct sock_xprt *transport =
2273 container_of(work, struct sock_xprt, connect_worker.work);
2274 struct socket *sock = transport->sock;
2275 struct rpc_xprt *xprt = &transport->xprt;
2279 sock = xs_create_sock(xprt, transport,
2280 xs_addr(xprt)->sa_family, SOCK_STREAM,
2283 status = PTR_ERR(sock);
2288 dprintk("RPC: worker connecting xprt %p via %s to "
2289 "%s (port %s)\n", xprt,
2290 xprt->address_strings[RPC_DISPLAY_PROTO],
2291 xprt->address_strings[RPC_DISPLAY_ADDR],
2292 xprt->address_strings[RPC_DISPLAY_PORT]);
2294 status = xs_tcp_finish_connecting(xprt, sock);
2295 trace_rpc_socket_connect(xprt, sock, status);
2296 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2297 xprt, -status, xprt_connected(xprt),
2298 sock->sk->sk_state);
2301 printk("%s: connect returned unhandled error %d\n",
2304 case -EADDRNOTAVAIL:
2305 /* We're probably in TIME_WAIT. Get rid of existing socket,
2308 xs_tcp_force_close(xprt);
2313 xprt_unlock_connect(xprt, transport);
2316 /* Happens, for instance, if the user specified a link
2317 * local IPv6 address without a scope-id.
2327 * xs_tcp_force_close() wakes tasks with -EIO.
2328 * We need to wake them first to ensure the
2329 * correct error code.
2331 xprt_wake_pending_tasks(xprt, status);
2332 xs_tcp_force_close(xprt);
2337 xprt_unlock_connect(xprt, transport);
2338 xprt_clear_connecting(xprt);
2339 xprt_wake_pending_tasks(xprt, status);
2342 static unsigned long xs_reconnect_delay(const struct rpc_xprt *xprt)
2344 unsigned long start, now = jiffies;
2346 start = xprt->stat.connect_start + xprt->reestablish_timeout;
2347 if (time_after(start, now))
2352 static void xs_reconnect_backoff(struct rpc_xprt *xprt)
2354 xprt->reestablish_timeout <<= 1;
2355 if (xprt->reestablish_timeout > xprt->max_reconnect_timeout)
2356 xprt->reestablish_timeout = xprt->max_reconnect_timeout;
2357 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2358 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2362 * xs_connect - connect a socket to a remote endpoint
2363 * @xprt: pointer to transport structure
2364 * @task: address of RPC task that manages state of connect request
2366 * TCP: If the remote end dropped the connection, delay reconnecting.
2368 * UDP socket connects are synchronous, but we use a work queue anyway
2369 * to guarantee that even unprivileged user processes can set up a
2370 * socket on a privileged port.
2372 * If a UDP socket connect fails, the delay behavior here prevents
2373 * retry floods (hard mounts).
2375 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2377 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2378 unsigned long delay = 0;
2380 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2382 if (transport->sock != NULL) {
2383 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2385 xprt, xprt->reestablish_timeout / HZ);
2387 /* Start by resetting any existing state */
2388 xs_reset_transport(transport);
2390 delay = xs_reconnect_delay(xprt);
2391 xs_reconnect_backoff(xprt);
2394 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2396 queue_delayed_work(xprtiod_workqueue,
2397 &transport->connect_worker,
2402 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2403 * @xprt: rpc_xprt struct containing statistics
2407 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2411 if (xprt_connected(xprt))
2412 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2414 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2415 "%llu %llu %lu %llu %llu\n",
2416 xprt->stat.bind_count,
2417 xprt->stat.connect_count,
2418 xprt->stat.connect_time / HZ,
2422 xprt->stat.bad_xids,
2425 xprt->stat.max_slots,
2426 xprt->stat.sending_u,
2427 xprt->stat.pending_u);
2431 * xs_udp_print_stats - display UDP socket-specifc stats
2432 * @xprt: rpc_xprt struct containing statistics
2436 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2438 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2440 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2443 xprt->stat.bind_count,
2446 xprt->stat.bad_xids,
2449 xprt->stat.max_slots,
2450 xprt->stat.sending_u,
2451 xprt->stat.pending_u);
2455 * xs_tcp_print_stats - display TCP socket-specifc stats
2456 * @xprt: rpc_xprt struct containing statistics
2460 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2462 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2465 if (xprt_connected(xprt))
2466 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2468 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2469 "%llu %llu %lu %llu %llu\n",
2471 xprt->stat.bind_count,
2472 xprt->stat.connect_count,
2473 xprt->stat.connect_time / HZ,
2477 xprt->stat.bad_xids,
2480 xprt->stat.max_slots,
2481 xprt->stat.sending_u,
2482 xprt->stat.pending_u);
2486 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2487 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2488 * to use the server side send routines.
2490 static int bc_malloc(struct rpc_task *task)
2492 struct rpc_rqst *rqst = task->tk_rqstp;
2493 size_t size = rqst->rq_callsize;
2495 struct rpc_buffer *buf;
2497 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2498 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2503 page = alloc_page(GFP_KERNEL);
2507 buf = page_address(page);
2508 buf->len = PAGE_SIZE;
2510 rqst->rq_buffer = buf->data;
2511 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2516 * Free the space allocated in the bc_alloc routine
2518 static void bc_free(struct rpc_task *task)
2520 void *buffer = task->tk_rqstp->rq_buffer;
2521 struct rpc_buffer *buf;
2523 buf = container_of(buffer, struct rpc_buffer, data);
2524 free_page((unsigned long)buf);
2528 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2529 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2531 static int bc_sendto(struct rpc_rqst *req)
2534 struct xdr_buf *xbufp = &req->rq_snd_buf;
2535 struct rpc_xprt *xprt = req->rq_xprt;
2536 struct sock_xprt *transport =
2537 container_of(xprt, struct sock_xprt, xprt);
2538 struct socket *sock = transport->sock;
2539 unsigned long headoff;
2540 unsigned long tailoff;
2542 xs_encode_stream_record_marker(xbufp);
2544 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2545 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2546 len = svc_send_common(sock, xbufp,
2547 virt_to_page(xbufp->head[0].iov_base), headoff,
2548 xbufp->tail[0].iov_base, tailoff);
2550 if (len != xbufp->len) {
2551 printk(KERN_NOTICE "Error sending entire callback!\n");
2559 * The send routine. Borrows from svc_send
2561 static int bc_send_request(struct rpc_rqst *req)
2563 struct svc_xprt *xprt;
2566 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2568 * Get the server socket associated with this callback xprt
2570 xprt = req->rq_xprt->bc_xprt;
2573 * Grab the mutex to serialize data as the connection is shared
2574 * with the fore channel
2576 mutex_lock(&xprt->xpt_mutex);
2577 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2580 len = bc_sendto(req);
2581 mutex_unlock(&xprt->xpt_mutex);
2590 * The close routine. Since this is client initiated, we do nothing
2593 static void bc_close(struct rpc_xprt *xprt)
2598 * The xprt destroy routine. Again, because this connection is client
2599 * initiated, we do nothing
2602 static void bc_destroy(struct rpc_xprt *xprt)
2604 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2607 module_put(THIS_MODULE);
2610 static const struct rpc_xprt_ops xs_local_ops = {
2611 .reserve_xprt = xprt_reserve_xprt,
2612 .release_xprt = xprt_release_xprt,
2613 .alloc_slot = xprt_alloc_slot,
2614 .free_slot = xprt_free_slot,
2615 .rpcbind = xs_local_rpcbind,
2616 .set_port = xs_local_set_port,
2617 .connect = xs_local_connect,
2618 .buf_alloc = rpc_malloc,
2619 .buf_free = rpc_free,
2620 .prepare_request = xs_stream_prepare_request,
2621 .send_request = xs_local_send_request,
2622 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2624 .destroy = xs_destroy,
2625 .print_stats = xs_local_print_stats,
2626 .enable_swap = xs_enable_swap,
2627 .disable_swap = xs_disable_swap,
2630 static const struct rpc_xprt_ops xs_udp_ops = {
2631 .set_buffer_size = xs_udp_set_buffer_size,
2632 .reserve_xprt = xprt_reserve_xprt_cong,
2633 .release_xprt = xprt_release_xprt_cong,
2634 .alloc_slot = xprt_alloc_slot,
2635 .free_slot = xprt_free_slot,
2636 .rpcbind = rpcb_getport_async,
2637 .set_port = xs_set_port,
2638 .connect = xs_connect,
2639 .buf_alloc = rpc_malloc,
2640 .buf_free = rpc_free,
2641 .send_request = xs_udp_send_request,
2642 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
2643 .timer = xs_udp_timer,
2644 .release_request = xprt_release_rqst_cong,
2646 .destroy = xs_destroy,
2647 .print_stats = xs_udp_print_stats,
2648 .enable_swap = xs_enable_swap,
2649 .disable_swap = xs_disable_swap,
2650 .inject_disconnect = xs_inject_disconnect,
2653 static const struct rpc_xprt_ops xs_tcp_ops = {
2654 .reserve_xprt = xprt_reserve_xprt,
2655 .release_xprt = xprt_release_xprt,
2656 .alloc_slot = xprt_alloc_slot,
2657 .free_slot = xprt_free_slot,
2658 .rpcbind = rpcb_getport_async,
2659 .set_port = xs_set_port,
2660 .connect = xs_connect,
2661 .buf_alloc = rpc_malloc,
2662 .buf_free = rpc_free,
2663 .prepare_request = xs_stream_prepare_request,
2664 .send_request = xs_tcp_send_request,
2665 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2666 .close = xs_tcp_shutdown,
2667 .destroy = xs_destroy,
2668 .set_connect_timeout = xs_tcp_set_connect_timeout,
2669 .print_stats = xs_tcp_print_stats,
2670 .enable_swap = xs_enable_swap,
2671 .disable_swap = xs_disable_swap,
2672 .inject_disconnect = xs_inject_disconnect,
2673 #ifdef CONFIG_SUNRPC_BACKCHANNEL
2674 .bc_setup = xprt_setup_bc,
2675 .bc_up = xs_tcp_bc_up,
2676 .bc_maxpayload = xs_tcp_bc_maxpayload,
2677 .bc_free_rqst = xprt_free_bc_rqst,
2678 .bc_destroy = xprt_destroy_bc,
2683 * The rpc_xprt_ops for the server backchannel
2686 static const struct rpc_xprt_ops bc_tcp_ops = {
2687 .reserve_xprt = xprt_reserve_xprt,
2688 .release_xprt = xprt_release_xprt,
2689 .alloc_slot = xprt_alloc_slot,
2690 .free_slot = xprt_free_slot,
2691 .buf_alloc = bc_malloc,
2692 .buf_free = bc_free,
2693 .send_request = bc_send_request,
2694 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2696 .destroy = bc_destroy,
2697 .print_stats = xs_tcp_print_stats,
2698 .enable_swap = xs_enable_swap,
2699 .disable_swap = xs_disable_swap,
2700 .inject_disconnect = xs_inject_disconnect,
2703 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2705 static const struct sockaddr_in sin = {
2706 .sin_family = AF_INET,
2707 .sin_addr.s_addr = htonl(INADDR_ANY),
2709 static const struct sockaddr_in6 sin6 = {
2710 .sin6_family = AF_INET6,
2711 .sin6_addr = IN6ADDR_ANY_INIT,
2718 memcpy(sap, &sin, sizeof(sin));
2721 memcpy(sap, &sin6, sizeof(sin6));
2724 dprintk("RPC: %s: Bad address family\n", __func__);
2725 return -EAFNOSUPPORT;
2730 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2731 unsigned int slot_table_size,
2732 unsigned int max_slot_table_size)
2734 struct rpc_xprt *xprt;
2735 struct sock_xprt *new;
2737 if (args->addrlen > sizeof(xprt->addr)) {
2738 dprintk("RPC: xs_setup_xprt: address too large\n");
2739 return ERR_PTR(-EBADF);
2742 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2743 max_slot_table_size);
2745 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2747 return ERR_PTR(-ENOMEM);
2750 new = container_of(xprt, struct sock_xprt, xprt);
2751 mutex_init(&new->recv_mutex);
2752 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2753 xprt->addrlen = args->addrlen;
2755 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2758 err = xs_init_anyaddr(args->dstaddr->sa_family,
2759 (struct sockaddr *)&new->srcaddr);
2762 return ERR_PTR(err);
2769 static const struct rpc_timeout xs_local_default_timeout = {
2770 .to_initval = 10 * HZ,
2771 .to_maxval = 10 * HZ,
2776 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2777 * @args: rpc transport creation arguments
2779 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2781 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2783 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2784 struct sock_xprt *transport;
2785 struct rpc_xprt *xprt;
2786 struct rpc_xprt *ret;
2788 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2789 xprt_max_tcp_slot_table_entries);
2792 transport = container_of(xprt, struct sock_xprt, xprt);
2795 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2796 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2798 xprt->bind_timeout = XS_BIND_TO;
2799 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2800 xprt->idle_timeout = XS_IDLE_DISC_TO;
2802 xprt->ops = &xs_local_ops;
2803 xprt->timeout = &xs_local_default_timeout;
2805 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2806 INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket);
2808 switch (sun->sun_family) {
2810 if (sun->sun_path[0] != '/') {
2811 dprintk("RPC: bad AF_LOCAL address: %s\n",
2813 ret = ERR_PTR(-EINVAL);
2816 xprt_set_bound(xprt);
2817 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2818 ret = ERR_PTR(xs_local_setup_socket(transport));
2823 ret = ERR_PTR(-EAFNOSUPPORT);
2827 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2828 xprt->address_strings[RPC_DISPLAY_ADDR]);
2830 if (try_module_get(THIS_MODULE))
2832 ret = ERR_PTR(-EINVAL);
2838 static const struct rpc_timeout xs_udp_default_timeout = {
2839 .to_initval = 5 * HZ,
2840 .to_maxval = 30 * HZ,
2841 .to_increment = 5 * HZ,
2846 * xs_setup_udp - Set up transport to use a UDP socket
2847 * @args: rpc transport creation arguments
2850 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2852 struct sockaddr *addr = args->dstaddr;
2853 struct rpc_xprt *xprt;
2854 struct sock_xprt *transport;
2855 struct rpc_xprt *ret;
2857 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2858 xprt_udp_slot_table_entries);
2861 transport = container_of(xprt, struct sock_xprt, xprt);
2863 xprt->prot = IPPROTO_UDP;
2865 /* XXX: header size can vary due to auth type, IPv6, etc. */
2866 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2868 xprt->bind_timeout = XS_BIND_TO;
2869 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2870 xprt->idle_timeout = XS_IDLE_DISC_TO;
2872 xprt->ops = &xs_udp_ops;
2874 xprt->timeout = &xs_udp_default_timeout;
2876 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
2877 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2879 switch (addr->sa_family) {
2881 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2882 xprt_set_bound(xprt);
2884 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2887 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2888 xprt_set_bound(xprt);
2890 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2893 ret = ERR_PTR(-EAFNOSUPPORT);
2897 if (xprt_bound(xprt))
2898 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2899 xprt->address_strings[RPC_DISPLAY_ADDR],
2900 xprt->address_strings[RPC_DISPLAY_PORT],
2901 xprt->address_strings[RPC_DISPLAY_PROTO]);
2903 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2904 xprt->address_strings[RPC_DISPLAY_ADDR],
2905 xprt->address_strings[RPC_DISPLAY_PROTO]);
2907 if (try_module_get(THIS_MODULE))
2909 ret = ERR_PTR(-EINVAL);
2915 static const struct rpc_timeout xs_tcp_default_timeout = {
2916 .to_initval = 60 * HZ,
2917 .to_maxval = 60 * HZ,
2922 * xs_setup_tcp - Set up transport to use a TCP socket
2923 * @args: rpc transport creation arguments
2926 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2928 struct sockaddr *addr = args->dstaddr;
2929 struct rpc_xprt *xprt;
2930 struct sock_xprt *transport;
2931 struct rpc_xprt *ret;
2932 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2934 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2935 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2937 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2938 max_slot_table_size);
2941 transport = container_of(xprt, struct sock_xprt, xprt);
2943 xprt->prot = IPPROTO_TCP;
2944 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2945 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2947 xprt->bind_timeout = XS_BIND_TO;
2948 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2949 xprt->idle_timeout = XS_IDLE_DISC_TO;
2951 xprt->ops = &xs_tcp_ops;
2952 xprt->timeout = &xs_tcp_default_timeout;
2954 xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
2955 xprt->connect_timeout = xprt->timeout->to_initval *
2956 (xprt->timeout->to_retries + 1);
2958 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2959 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
2961 switch (addr->sa_family) {
2963 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2964 xprt_set_bound(xprt);
2966 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2969 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2970 xprt_set_bound(xprt);
2972 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2975 ret = ERR_PTR(-EAFNOSUPPORT);
2979 if (xprt_bound(xprt))
2980 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2981 xprt->address_strings[RPC_DISPLAY_ADDR],
2982 xprt->address_strings[RPC_DISPLAY_PORT],
2983 xprt->address_strings[RPC_DISPLAY_PROTO]);
2985 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2986 xprt->address_strings[RPC_DISPLAY_ADDR],
2987 xprt->address_strings[RPC_DISPLAY_PROTO]);
2989 if (try_module_get(THIS_MODULE))
2991 ret = ERR_PTR(-EINVAL);
2998 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2999 * @args: rpc transport creation arguments
3002 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3004 struct sockaddr *addr = args->dstaddr;
3005 struct rpc_xprt *xprt;
3006 struct sock_xprt *transport;
3007 struct svc_sock *bc_sock;
3008 struct rpc_xprt *ret;
3010 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3011 xprt_tcp_slot_table_entries);
3014 transport = container_of(xprt, struct sock_xprt, xprt);
3016 xprt->prot = IPPROTO_TCP;
3017 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3018 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3019 xprt->timeout = &xs_tcp_default_timeout;
3022 xprt_set_bound(xprt);
3023 xprt->bind_timeout = 0;
3024 xprt->reestablish_timeout = 0;
3025 xprt->idle_timeout = 0;
3027 xprt->ops = &bc_tcp_ops;
3029 switch (addr->sa_family) {
3031 xs_format_peer_addresses(xprt, "tcp",
3035 xs_format_peer_addresses(xprt, "tcp",
3036 RPCBIND_NETID_TCP6);
3039 ret = ERR_PTR(-EAFNOSUPPORT);
3043 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3044 xprt->address_strings[RPC_DISPLAY_ADDR],
3045 xprt->address_strings[RPC_DISPLAY_PORT],
3046 xprt->address_strings[RPC_DISPLAY_PROTO]);
3049 * Once we've associated a backchannel xprt with a connection,
3050 * we want to keep it around as long as the connection lasts,
3051 * in case we need to start using it for a backchannel again;
3052 * this reference won't be dropped until bc_xprt is destroyed.
3055 args->bc_xprt->xpt_bc_xprt = xprt;
3056 xprt->bc_xprt = args->bc_xprt;
3057 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3058 transport->sock = bc_sock->sk_sock;
3059 transport->inet = bc_sock->sk_sk;
3062 * Since we don't want connections for the backchannel, we set
3063 * the xprt status to connected
3065 xprt_set_connected(xprt);
3067 if (try_module_get(THIS_MODULE))
3070 args->bc_xprt->xpt_bc_xprt = NULL;
3071 args->bc_xprt->xpt_bc_xps = NULL;
3073 ret = ERR_PTR(-EINVAL);
3079 static struct xprt_class xs_local_transport = {
3080 .list = LIST_HEAD_INIT(xs_local_transport.list),
3081 .name = "named UNIX socket",
3082 .owner = THIS_MODULE,
3083 .ident = XPRT_TRANSPORT_LOCAL,
3084 .setup = xs_setup_local,
3087 static struct xprt_class xs_udp_transport = {
3088 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3090 .owner = THIS_MODULE,
3091 .ident = XPRT_TRANSPORT_UDP,
3092 .setup = xs_setup_udp,
3095 static struct xprt_class xs_tcp_transport = {
3096 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3098 .owner = THIS_MODULE,
3099 .ident = XPRT_TRANSPORT_TCP,
3100 .setup = xs_setup_tcp,
3103 static struct xprt_class xs_bc_tcp_transport = {
3104 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3105 .name = "tcp NFSv4.1 backchannel",
3106 .owner = THIS_MODULE,
3107 .ident = XPRT_TRANSPORT_BC_TCP,
3108 .setup = xs_setup_bc_tcp,
3112 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3115 int init_socket_xprt(void)
3117 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3118 if (!sunrpc_table_header)
3119 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3122 xprt_register_transport(&xs_local_transport);
3123 xprt_register_transport(&xs_udp_transport);
3124 xprt_register_transport(&xs_tcp_transport);
3125 xprt_register_transport(&xs_bc_tcp_transport);
3131 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3134 void cleanup_socket_xprt(void)
3136 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3137 if (sunrpc_table_header) {
3138 unregister_sysctl_table(sunrpc_table_header);
3139 sunrpc_table_header = NULL;
3143 xprt_unregister_transport(&xs_local_transport);
3144 xprt_unregister_transport(&xs_udp_transport);
3145 xprt_unregister_transport(&xs_tcp_transport);
3146 xprt_unregister_transport(&xs_bc_tcp_transport);
3149 static int param_set_uint_minmax(const char *val,
3150 const struct kernel_param *kp,
3151 unsigned int min, unsigned int max)
3158 ret = kstrtouint(val, 0, &num);
3161 if (num < min || num > max)
3163 *((unsigned int *)kp->arg) = num;
3167 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3169 return param_set_uint_minmax(val, kp,
3174 static const struct kernel_param_ops param_ops_portnr = {
3175 .set = param_set_portnr,
3176 .get = param_get_uint,
3179 #define param_check_portnr(name, p) \
3180 __param_check(name, p, unsigned int);
3182 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3183 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3185 static int param_set_slot_table_size(const char *val,
3186 const struct kernel_param *kp)
3188 return param_set_uint_minmax(val, kp,
3190 RPC_MAX_SLOT_TABLE);
3193 static const struct kernel_param_ops param_ops_slot_table_size = {
3194 .set = param_set_slot_table_size,
3195 .get = param_get_uint,
3198 #define param_check_slot_table_size(name, p) \
3199 __param_check(name, p, unsigned int);
3201 static int param_set_max_slot_table_size(const char *val,
3202 const struct kernel_param *kp)
3204 return param_set_uint_minmax(val, kp,
3206 RPC_MAX_SLOT_TABLE_LIMIT);
3209 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3210 .set = param_set_max_slot_table_size,
3211 .get = param_get_uint,
3214 #define param_check_max_slot_table_size(name, p) \
3215 __param_check(name, p, unsigned int);
3217 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3218 slot_table_size, 0644);
3219 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3220 max_slot_table_size, 0644);
3221 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3222 slot_table_size, 0644);