2 * linux/net/sunrpc/xprtsock.c
4 * Client-side transport implementation for sockets.
6 * TCP callback races fixes (C) 1998 Red Hat
7 * TCP send fixes (C) 1998 Red Hat
8 * TCP NFS related read + write fixes
9 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
11 * Rewrite of larges part of the code in order to stabilize TCP stuff.
12 * Fix behaviour when socket buffer is full.
13 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
15 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
17 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
18 * <gilles.quillard@bull.net>
21 #include <linux/types.h>
22 #include <linux/string.h>
23 #include <linux/slab.h>
24 #include <linux/module.h>
25 #include <linux/capability.h>
26 #include <linux/pagemap.h>
27 #include <linux/errno.h>
28 #include <linux/socket.h>
30 #include <linux/net.h>
33 #include <linux/udp.h>
34 #include <linux/tcp.h>
35 #include <linux/sunrpc/clnt.h>
36 #include <linux/sunrpc/addr.h>
37 #include <linux/sunrpc/sched.h>
38 #include <linux/sunrpc/svcsock.h>
39 #include <linux/sunrpc/xprtsock.h>
40 #include <linux/file.h>
41 #ifdef CONFIG_SUNRPC_BACKCHANNEL
42 #include <linux/sunrpc/bc_xprt.h>
46 #include <net/checksum.h>
50 #include <trace/events/sunrpc.h>
54 static void xs_close(struct rpc_xprt *xprt);
59 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
60 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
61 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
63 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
64 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
66 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
68 #define XS_TCP_LINGER_TO (15U * HZ)
69 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
72 * We can register our own files under /proc/sys/sunrpc by
73 * calling register_sysctl_table() again. The files in that
74 * directory become the union of all files registered there.
76 * We simply need to make sure that we don't collide with
77 * someone else's file names!
80 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
81 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
82 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
83 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
84 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
86 static struct ctl_table_header *sunrpc_table_header;
89 * FIXME: changing the UDP slot table size should also resize the UDP
90 * socket buffers for existing UDP transports
92 static struct ctl_table xs_tunables_table[] = {
94 .procname = "udp_slot_table_entries",
95 .data = &xprt_udp_slot_table_entries,
96 .maxlen = sizeof(unsigned int),
98 .proc_handler = proc_dointvec_minmax,
99 .extra1 = &min_slot_table_size,
100 .extra2 = &max_slot_table_size
103 .procname = "tcp_slot_table_entries",
104 .data = &xprt_tcp_slot_table_entries,
105 .maxlen = sizeof(unsigned int),
107 .proc_handler = proc_dointvec_minmax,
108 .extra1 = &min_slot_table_size,
109 .extra2 = &max_slot_table_size
112 .procname = "tcp_max_slot_table_entries",
113 .data = &xprt_max_tcp_slot_table_entries,
114 .maxlen = sizeof(unsigned int),
116 .proc_handler = proc_dointvec_minmax,
117 .extra1 = &min_slot_table_size,
118 .extra2 = &max_tcp_slot_table_limit
121 .procname = "min_resvport",
122 .data = &xprt_min_resvport,
123 .maxlen = sizeof(unsigned int),
125 .proc_handler = proc_dointvec_minmax,
126 .extra1 = &xprt_min_resvport_limit,
127 .extra2 = &xprt_max_resvport_limit
130 .procname = "max_resvport",
131 .data = &xprt_max_resvport,
132 .maxlen = sizeof(unsigned int),
134 .proc_handler = proc_dointvec_minmax,
135 .extra1 = &xprt_min_resvport_limit,
136 .extra2 = &xprt_max_resvport_limit
139 .procname = "tcp_fin_timeout",
140 .data = &xs_tcp_fin_timeout,
141 .maxlen = sizeof(xs_tcp_fin_timeout),
143 .proc_handler = proc_dointvec_jiffies,
148 static struct ctl_table sunrpc_table[] = {
150 .procname = "sunrpc",
152 .child = xs_tunables_table
160 * Wait duration for a reply from the RPC portmapper.
162 #define XS_BIND_TO (60U * HZ)
165 * Delay if a UDP socket connect error occurs. This is most likely some
166 * kind of resource problem on the local host.
168 #define XS_UDP_REEST_TO (2U * HZ)
171 * The reestablish timeout allows clients to delay for a bit before attempting
172 * to reconnect to a server that just dropped our connection.
174 * We implement an exponential backoff when trying to reestablish a TCP
175 * transport connection with the server. Some servers like to drop a TCP
176 * connection when they are overworked, so we start with a short timeout and
177 * increase over time if the server is down or not responding.
179 #define XS_TCP_INIT_REEST_TO (3U * HZ)
182 * TCP idle timeout; client drops the transport socket if it is idle
183 * for this long. Note that we also timeout UDP sockets to prevent
184 * holding port numbers when there is no RPC traffic.
186 #define XS_IDLE_DISC_TO (5U * 60 * HZ)
188 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
189 # undef RPC_DEBUG_DATA
190 # define RPCDBG_FACILITY RPCDBG_TRANS
193 #ifdef RPC_DEBUG_DATA
194 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
196 u8 *buf = (u8 *) packet;
199 dprintk("RPC: %s\n", msg);
200 for (j = 0; j < count && j < 128; j += 4) {
204 dprintk("0x%04x ", j);
206 dprintk("%02x%02x%02x%02x ",
207 buf[j], buf[j+1], buf[j+2], buf[j+3]);
212 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
218 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
220 return (struct rpc_xprt *) sk->sk_user_data;
223 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
225 return (struct sockaddr *) &xprt->addr;
228 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
230 return (struct sockaddr_un *) &xprt->addr;
233 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
235 return (struct sockaddr_in *) &xprt->addr;
238 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
240 return (struct sockaddr_in6 *) &xprt->addr;
243 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
245 struct sockaddr *sap = xs_addr(xprt);
246 struct sockaddr_in6 *sin6;
247 struct sockaddr_in *sin;
248 struct sockaddr_un *sun;
251 switch (sap->sa_family) {
253 sun = xs_addr_un(xprt);
254 strlcpy(buf, sun->sun_path, sizeof(buf));
255 xprt->address_strings[RPC_DISPLAY_ADDR] =
256 kstrdup(buf, GFP_KERNEL);
259 (void)rpc_ntop(sap, buf, sizeof(buf));
260 xprt->address_strings[RPC_DISPLAY_ADDR] =
261 kstrdup(buf, GFP_KERNEL);
262 sin = xs_addr_in(xprt);
263 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
266 (void)rpc_ntop(sap, buf, sizeof(buf));
267 xprt->address_strings[RPC_DISPLAY_ADDR] =
268 kstrdup(buf, GFP_KERNEL);
269 sin6 = xs_addr_in6(xprt);
270 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
276 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
279 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
281 struct sockaddr *sap = xs_addr(xprt);
284 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
285 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
287 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
288 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
291 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
292 const char *protocol,
295 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
296 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
297 xs_format_common_peer_addresses(xprt);
298 xs_format_common_peer_ports(xprt);
301 static void xs_update_peer_port(struct rpc_xprt *xprt)
303 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
304 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
306 xs_format_common_peer_ports(xprt);
309 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
313 for (i = 0; i < RPC_DISPLAY_MAX; i++)
315 case RPC_DISPLAY_PROTO:
316 case RPC_DISPLAY_NETID:
319 kfree(xprt->address_strings[i]);
323 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
325 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
327 struct msghdr msg = {
329 .msg_namelen = addrlen,
330 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
333 .iov_base = vec->iov_base + base,
334 .iov_len = vec->iov_len - base,
337 if (iov.iov_len != 0)
338 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
339 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
342 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p)
344 ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
345 int offset, size_t size, int flags);
347 unsigned int remainder;
350 remainder = xdr->page_len - base;
351 base += xdr->page_base;
352 ppage = xdr->pages + (base >> PAGE_SHIFT);
354 do_sendpage = sock->ops->sendpage;
356 do_sendpage = sock_no_sendpage;
358 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
359 int flags = XS_SENDMSG_FLAGS;
365 flags |= MSG_SENDPAGE_NOTLAST | MSG_MORE;
366 err = do_sendpage(sock, *ppage, base, len, flags);
367 if (remainder == 0 || err != len)
381 * xs_sendpages - write pages directly to a socket
382 * @sock: socket to send on
383 * @addr: UDP only -- address of destination
384 * @addrlen: UDP only -- length of destination address
385 * @xdr: buffer containing this request
386 * @base: starting position in the buffer
387 * @zerocopy: true if it is safe to use sendpage()
388 * @sent_p: return the total number of bytes successfully queued for sending
391 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p)
393 unsigned int remainder = xdr->len - base;
405 if (base < xdr->head[0].iov_len || addr != NULL) {
406 unsigned int len = xdr->head[0].iov_len - base;
408 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
409 if (remainder == 0 || err != len)
414 base -= xdr->head[0].iov_len;
416 if (base < xdr->page_len) {
417 unsigned int len = xdr->page_len - base;
419 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy, &sent);
421 if (remainder == 0 || sent != len)
425 base -= xdr->page_len;
427 if (base >= xdr->tail[0].iov_len)
429 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
438 static void xs_nospace_callback(struct rpc_task *task)
440 struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
442 transport->inet->sk_write_pending--;
446 * xs_nospace - place task on wait queue if transmit was incomplete
447 * @task: task to put to sleep
450 static int xs_nospace(struct rpc_task *task)
452 struct rpc_rqst *req = task->tk_rqstp;
453 struct rpc_xprt *xprt = req->rq_xprt;
454 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
455 struct sock *sk = transport->inet;
458 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
459 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
462 /* Protect against races with write_space */
463 spin_lock_bh(&xprt->transport_lock);
465 /* Don't race with disconnect */
466 if (xprt_connected(xprt)) {
467 /* wait for more buffer space */
468 sk->sk_write_pending++;
469 xprt_wait_for_buffer_space(task, xs_nospace_callback);
473 spin_unlock_bh(&xprt->transport_lock);
475 /* Race breaker in case memory is freed before above code is called */
476 if (ret == -EAGAIN) {
477 struct socket_wq *wq;
480 wq = rcu_dereference(sk->sk_wq);
481 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
484 sk->sk_write_space(sk);
490 * Construct a stream transport record marker in @buf.
492 static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
494 u32 reclen = buf->len - sizeof(rpc_fraghdr);
495 rpc_fraghdr *base = buf->head[0].iov_base;
496 *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
500 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
501 * @task: RPC task that manages the state of an RPC request
504 * 0: The request has been sent
505 * EAGAIN: The socket was blocked, please call again later to
506 * complete the request
507 * ENOTCONN: Caller needs to invoke connect logic then call again
508 * other: Some other error occured, the request was not sent
510 static int xs_local_send_request(struct rpc_task *task)
512 struct rpc_rqst *req = task->tk_rqstp;
513 struct rpc_xprt *xprt = req->rq_xprt;
514 struct sock_xprt *transport =
515 container_of(xprt, struct sock_xprt, xprt);
516 struct xdr_buf *xdr = &req->rq_snd_buf;
520 xs_encode_stream_record_marker(&req->rq_snd_buf);
522 xs_pktdump("packet data:",
523 req->rq_svec->iov_base, req->rq_svec->iov_len);
525 status = xs_sendpages(transport->sock, NULL, 0, xdr, req->rq_bytes_sent,
527 dprintk("RPC: %s(%u) = %d\n",
528 __func__, xdr->len - req->rq_bytes_sent, status);
530 if (status == -EAGAIN && sock_writeable(transport->inet))
533 if (likely(sent > 0) || status == 0) {
534 req->rq_bytes_sent += sent;
535 req->rq_xmit_bytes_sent += sent;
536 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
537 req->rq_bytes_sent = 0;
547 status = xs_nospace(task);
550 dprintk("RPC: sendmsg returned unrecognized error %d\n",
561 * xs_udp_send_request - write an RPC request to a UDP socket
562 * @task: address of RPC task that manages the state of an RPC request
565 * 0: The request has been sent
566 * EAGAIN: The socket was blocked, please call again later to
567 * complete the request
568 * ENOTCONN: Caller needs to invoke connect logic then call again
569 * other: Some other error occurred, the request was not sent
571 static int xs_udp_send_request(struct rpc_task *task)
573 struct rpc_rqst *req = task->tk_rqstp;
574 struct rpc_xprt *xprt = req->rq_xprt;
575 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
576 struct xdr_buf *xdr = &req->rq_snd_buf;
580 xs_pktdump("packet data:",
581 req->rq_svec->iov_base,
582 req->rq_svec->iov_len);
584 if (!xprt_bound(xprt))
586 status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
587 xdr, req->rq_bytes_sent, true, &sent);
589 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
590 xdr->len - req->rq_bytes_sent, status);
592 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
593 if (status == -EPERM)
596 if (status == -EAGAIN && sock_writeable(transport->inet))
599 if (sent > 0 || status == 0) {
600 req->rq_xmit_bytes_sent += sent;
601 if (sent >= req->rq_slen)
603 /* Still some bytes left; set up for a retry later. */
611 /* Should we call xs_close() here? */
614 status = xs_nospace(task);
621 /* When the server has died, an ICMP port unreachable message
622 * prompts ECONNREFUSED. */
625 dprintk("RPC: sendmsg returned unrecognized error %d\n",
633 * xs_tcp_send_request - write an RPC request to a TCP socket
634 * @task: address of RPC task that manages the state of an RPC request
637 * 0: The request has been sent
638 * EAGAIN: The socket was blocked, please call again later to
639 * complete the request
640 * ENOTCONN: Caller needs to invoke connect logic then call again
641 * other: Some other error occurred, the request was not sent
643 * XXX: In the case of soft timeouts, should we eventually give up
644 * if sendmsg is not able to make progress?
646 static int xs_tcp_send_request(struct rpc_task *task)
648 struct rpc_rqst *req = task->tk_rqstp;
649 struct rpc_xprt *xprt = req->rq_xprt;
650 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
651 struct xdr_buf *xdr = &req->rq_snd_buf;
652 bool zerocopy = true;
653 bool vm_wait = false;
657 xs_encode_stream_record_marker(&req->rq_snd_buf);
659 xs_pktdump("packet data:",
660 req->rq_svec->iov_base,
661 req->rq_svec->iov_len);
662 /* Don't use zero copy if this is a resend. If the RPC call
663 * completes while the socket holds a reference to the pages,
664 * then we may end up resending corrupted data.
666 if (task->tk_flags & RPC_TASK_SENT)
669 /* Continue transmitting the packet/record. We must be careful
670 * to cope with writespace callbacks arriving _after_ we have
671 * called sendmsg(). */
674 status = xs_sendpages(transport->sock, NULL, 0, xdr,
675 req->rq_bytes_sent, zerocopy, &sent);
677 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
678 xdr->len - req->rq_bytes_sent, status);
680 /* If we've sent the entire packet, immediately
681 * reset the count of bytes sent. */
682 req->rq_bytes_sent += sent;
683 req->rq_xmit_bytes_sent += sent;
684 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
685 req->rq_bytes_sent = 0;
689 WARN_ON_ONCE(sent == 0 && status == 0);
691 if (status == -EAGAIN ) {
693 * Return EAGAIN if we're sure we're hitting the
694 * socket send buffer limits.
696 if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
699 * Did we hit a memory allocation failure?
705 /* Retry, knowing now that we're below the
706 * socket send buffer limit
720 /* Should we call xs_close() here? */
723 status = xs_nospace(task);
733 dprintk("RPC: sendmsg returned unrecognized error %d\n",
741 * xs_tcp_release_xprt - clean up after a tcp transmission
745 * This cleans up if an error causes us to abort the transmission of a request.
746 * In this case, the socket may need to be reset in order to avoid confusing
749 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
751 struct rpc_rqst *req;
753 if (task != xprt->snd_task)
757 req = task->tk_rqstp;
760 if (req->rq_bytes_sent == 0)
762 if (req->rq_bytes_sent == req->rq_snd_buf.len)
764 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
766 xprt_release_xprt(xprt, task);
769 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
771 transport->old_data_ready = sk->sk_data_ready;
772 transport->old_state_change = sk->sk_state_change;
773 transport->old_write_space = sk->sk_write_space;
774 transport->old_error_report = sk->sk_error_report;
777 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
779 sk->sk_data_ready = transport->old_data_ready;
780 sk->sk_state_change = transport->old_state_change;
781 sk->sk_write_space = transport->old_write_space;
782 sk->sk_error_report = transport->old_error_report;
785 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
787 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
789 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
792 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
794 smp_mb__before_atomic();
795 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
796 clear_bit(XPRT_CLOSING, &xprt->state);
797 xs_sock_reset_state_flags(xprt);
798 smp_mb__after_atomic();
801 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
803 xs_sock_reset_connection_flags(xprt);
804 /* Mark transport as closed and wake up all pending tasks */
805 xprt_disconnect_done(xprt);
809 * xs_error_report - callback to handle TCP socket state errors
812 * Note: we don't call sock_error() since there may be a rpc_task
813 * using the socket, and so we don't want to clear sk->sk_err.
815 static void xs_error_report(struct sock *sk)
817 struct rpc_xprt *xprt;
820 read_lock_bh(&sk->sk_callback_lock);
821 if (!(xprt = xprt_from_sock(sk)))
827 /* Is this a reset event? */
828 if (sk->sk_state == TCP_CLOSE)
829 xs_sock_mark_closed(xprt);
830 dprintk("RPC: xs_error_report client %p, error=%d...\n",
832 trace_rpc_socket_error(xprt, sk->sk_socket, err);
833 xprt_wake_pending_tasks(xprt, err);
835 read_unlock_bh(&sk->sk_callback_lock);
838 static void xs_reset_transport(struct sock_xprt *transport)
840 struct socket *sock = transport->sock;
841 struct sock *sk = transport->inet;
842 struct rpc_xprt *xprt = &transport->xprt;
847 if (atomic_read(&transport->xprt.swapper))
848 sk_clear_memalloc(sk);
850 kernel_sock_shutdown(sock, SHUT_RDWR);
852 mutex_lock(&transport->recv_mutex);
853 write_lock_bh(&sk->sk_callback_lock);
854 transport->inet = NULL;
855 transport->sock = NULL;
857 sk->sk_user_data = NULL;
859 xs_restore_old_callbacks(transport, sk);
860 xprt_clear_connected(xprt);
861 write_unlock_bh(&sk->sk_callback_lock);
862 xs_sock_reset_connection_flags(xprt);
863 mutex_unlock(&transport->recv_mutex);
865 trace_rpc_socket_close(xprt, sock);
870 * xs_close - close a socket
873 * This is used when all requests are complete; ie, no DRC state remains
874 * on the server we want to save.
876 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
877 * xs_reset_transport() zeroing the socket from underneath a writer.
879 static void xs_close(struct rpc_xprt *xprt)
881 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
883 dprintk("RPC: xs_close xprt %p\n", xprt);
885 xs_reset_transport(transport);
886 xprt->reestablish_timeout = 0;
888 xprt_disconnect_done(xprt);
891 static void xs_inject_disconnect(struct rpc_xprt *xprt)
893 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
895 xprt_disconnect_done(xprt);
898 static void xs_xprt_free(struct rpc_xprt *xprt)
900 xs_free_peer_addresses(xprt);
905 * xs_destroy - prepare to shutdown a transport
906 * @xprt: doomed transport
909 static void xs_destroy(struct rpc_xprt *xprt)
911 struct sock_xprt *transport = container_of(xprt,
912 struct sock_xprt, xprt);
913 dprintk("RPC: xs_destroy xprt %p\n", xprt);
915 cancel_delayed_work_sync(&transport->connect_worker);
917 cancel_work_sync(&transport->recv_worker);
919 module_put(THIS_MODULE);
922 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
924 struct xdr_skb_reader desc = {
926 .offset = sizeof(rpc_fraghdr),
927 .count = skb->len - sizeof(rpc_fraghdr),
930 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
938 * xs_local_data_read_skb
943 * Currently this assumes we can read the whole reply in a single gulp.
945 static void xs_local_data_read_skb(struct rpc_xprt *xprt,
949 struct rpc_task *task;
950 struct rpc_rqst *rovr;
955 repsize = skb->len - sizeof(rpc_fraghdr);
957 dprintk("RPC: impossible RPC reply size %d\n", repsize);
961 /* Copy the XID from the skb... */
962 xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
966 /* Look up and lock the request corresponding to the given XID */
967 spin_lock_bh(&xprt->transport_lock);
968 rovr = xprt_lookup_rqst(xprt, *xp);
971 task = rovr->rq_task;
973 copied = rovr->rq_private_buf.buflen;
974 if (copied > repsize)
977 if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
978 dprintk("RPC: sk_buff copy failed\n");
982 xprt_complete_rqst(task, copied);
985 spin_unlock_bh(&xprt->transport_lock);
988 static void xs_local_data_receive(struct sock_xprt *transport)
994 mutex_lock(&transport->recv_mutex);
995 sk = transport->inet;
999 skb = skb_recv_datagram(sk, 0, 1, &err);
1001 xs_local_data_read_skb(&transport->xprt, sk, skb);
1002 skb_free_datagram(sk, skb);
1005 if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1009 mutex_unlock(&transport->recv_mutex);
1012 static void xs_local_data_receive_workfn(struct work_struct *work)
1014 struct sock_xprt *transport =
1015 container_of(work, struct sock_xprt, recv_worker);
1016 xs_local_data_receive(transport);
1020 * xs_udp_data_read_skb - receive callback for UDP sockets
1026 static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1028 struct sk_buff *skb)
1030 struct rpc_task *task;
1031 struct rpc_rqst *rovr;
1032 int repsize, copied;
1038 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
1042 /* Copy the XID from the skb... */
1043 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1047 /* Look up and lock the request corresponding to the given XID */
1048 spin_lock_bh(&xprt->transport_lock);
1049 rovr = xprt_lookup_rqst(xprt, *xp);
1052 task = rovr->rq_task;
1054 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1057 /* Suck it into the iovec, verify checksum if not done by hw. */
1058 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1059 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1063 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1065 xprt_adjust_cwnd(xprt, task, copied);
1066 xprt_complete_rqst(task, copied);
1069 spin_unlock_bh(&xprt->transport_lock);
1072 static void xs_udp_data_receive(struct sock_xprt *transport)
1074 struct sk_buff *skb;
1078 mutex_lock(&transport->recv_mutex);
1079 sk = transport->inet;
1083 skb = skb_recv_datagram(sk, 0, 1, &err);
1085 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1086 skb_free_datagram_locked(sk, skb);
1089 if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1093 mutex_unlock(&transport->recv_mutex);
1096 static void xs_udp_data_receive_workfn(struct work_struct *work)
1098 struct sock_xprt *transport =
1099 container_of(work, struct sock_xprt, recv_worker);
1100 xs_udp_data_receive(transport);
1104 * xs_data_ready - "data ready" callback for UDP sockets
1105 * @sk: socket with data to read
1108 static void xs_data_ready(struct sock *sk)
1110 struct rpc_xprt *xprt;
1112 read_lock_bh(&sk->sk_callback_lock);
1113 dprintk("RPC: xs_data_ready...\n");
1114 xprt = xprt_from_sock(sk);
1116 struct sock_xprt *transport = container_of(xprt,
1117 struct sock_xprt, xprt);
1118 transport->old_data_ready(sk);
1119 /* Any data means we had a useful conversation, so
1120 * then we don't need to delay the next reconnect
1122 if (xprt->reestablish_timeout)
1123 xprt->reestablish_timeout = 0;
1124 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1125 queue_work(xprtiod_workqueue, &transport->recv_worker);
1127 read_unlock_bh(&sk->sk_callback_lock);
1131 * Helper function to force a TCP close if the server is sending
1132 * junk and/or it has put us in CLOSE_WAIT
1134 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1136 xprt_force_disconnect(xprt);
1139 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1141 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1145 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1146 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1147 used = xdr_skb_read_bits(desc, p, len);
1148 transport->tcp_offset += used;
1152 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1153 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1154 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1156 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1157 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1159 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1160 transport->tcp_offset = 0;
1162 /* Sanity check of the record length */
1163 if (unlikely(transport->tcp_reclen < 8)) {
1164 dprintk("RPC: invalid TCP record fragment length\n");
1165 xs_tcp_force_close(xprt);
1168 dprintk("RPC: reading TCP record fragment of length %d\n",
1169 transport->tcp_reclen);
1172 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1174 if (transport->tcp_offset == transport->tcp_reclen) {
1175 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1176 transport->tcp_offset = 0;
1177 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1178 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1179 transport->tcp_flags |= TCP_RCV_COPY_XID;
1180 transport->tcp_copied = 0;
1185 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1190 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1191 dprintk("RPC: reading XID (%Zu bytes)\n", len);
1192 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1193 used = xdr_skb_read_bits(desc, p, len);
1194 transport->tcp_offset += used;
1197 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1198 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1199 transport->tcp_copied = 4;
1200 dprintk("RPC: reading %s XID %08x\n",
1201 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1203 ntohl(transport->tcp_xid));
1204 xs_tcp_check_fraghdr(transport);
1207 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1208 struct xdr_skb_reader *desc)
1215 * We want transport->tcp_offset to be 8 at the end of this routine
1216 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1217 * When this function is called for the first time,
1218 * transport->tcp_offset is 4 (after having already read the xid).
1220 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1221 len = sizeof(transport->tcp_calldir) - offset;
1222 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len);
1223 p = ((char *) &transport->tcp_calldir) + offset;
1224 used = xdr_skb_read_bits(desc, p, len);
1225 transport->tcp_offset += used;
1228 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1230 * We don't yet have the XDR buffer, so we will write the calldir
1231 * out after we get the buffer from the 'struct rpc_rqst'
1233 switch (ntohl(transport->tcp_calldir)) {
1235 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1236 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1237 transport->tcp_flags |= TCP_RPC_REPLY;
1240 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1241 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1242 transport->tcp_flags &= ~TCP_RPC_REPLY;
1245 dprintk("RPC: invalid request message type\n");
1246 xs_tcp_force_close(&transport->xprt);
1248 xs_tcp_check_fraghdr(transport);
1251 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1252 struct xdr_skb_reader *desc,
1253 struct rpc_rqst *req)
1255 struct sock_xprt *transport =
1256 container_of(xprt, struct sock_xprt, xprt);
1257 struct xdr_buf *rcvbuf;
1261 rcvbuf = &req->rq_private_buf;
1263 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1265 * Save the RPC direction in the XDR buffer
1267 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1268 &transport->tcp_calldir,
1269 sizeof(transport->tcp_calldir));
1270 transport->tcp_copied += sizeof(transport->tcp_calldir);
1271 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1275 if (len > transport->tcp_reclen - transport->tcp_offset) {
1276 struct xdr_skb_reader my_desc;
1278 len = transport->tcp_reclen - transport->tcp_offset;
1279 memcpy(&my_desc, desc, sizeof(my_desc));
1280 my_desc.count = len;
1281 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1282 &my_desc, xdr_skb_read_bits);
1286 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1287 desc, xdr_skb_read_bits);
1290 transport->tcp_copied += r;
1291 transport->tcp_offset += r;
1294 /* Error when copying to the receive buffer,
1295 * usually because we weren't able to allocate
1296 * additional buffer pages. All we can do now
1297 * is turn off TCP_RCV_COPY_DATA, so the request
1298 * will not receive any additional updates,
1300 * Any remaining data from this record will
1303 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1304 dprintk("RPC: XID %08x truncated request\n",
1305 ntohl(transport->tcp_xid));
1306 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1307 "tcp_offset = %u, tcp_reclen = %u\n",
1308 xprt, transport->tcp_copied,
1309 transport->tcp_offset, transport->tcp_reclen);
1313 dprintk("RPC: XID %08x read %Zd bytes\n",
1314 ntohl(transport->tcp_xid), r);
1315 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1316 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1317 transport->tcp_offset, transport->tcp_reclen);
1319 if (transport->tcp_copied == req->rq_private_buf.buflen)
1320 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1321 else if (transport->tcp_offset == transport->tcp_reclen) {
1322 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1323 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1328 * Finds the request corresponding to the RPC xid and invokes the common
1329 * tcp read code to read the data.
1331 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1332 struct xdr_skb_reader *desc)
1334 struct sock_xprt *transport =
1335 container_of(xprt, struct sock_xprt, xprt);
1336 struct rpc_rqst *req;
1338 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1340 /* Find and lock the request corresponding to this xid */
1341 spin_lock_bh(&xprt->transport_lock);
1342 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1344 dprintk("RPC: XID %08x request not found!\n",
1345 ntohl(transport->tcp_xid));
1346 spin_unlock_bh(&xprt->transport_lock);
1350 xs_tcp_read_common(xprt, desc, req);
1352 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1353 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1355 spin_unlock_bh(&xprt->transport_lock);
1359 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1361 * Obtains an rpc_rqst previously allocated and invokes the common
1362 * tcp read code to read the data. The result is placed in the callback
1364 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1365 * connection and return -1.
1367 static int xs_tcp_read_callback(struct rpc_xprt *xprt,
1368 struct xdr_skb_reader *desc)
1370 struct sock_xprt *transport =
1371 container_of(xprt, struct sock_xprt, xprt);
1372 struct rpc_rqst *req;
1374 /* Look up and lock the request corresponding to the given XID */
1375 spin_lock_bh(&xprt->transport_lock);
1376 req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
1378 spin_unlock_bh(&xprt->transport_lock);
1379 printk(KERN_WARNING "Callback slot table overflowed\n");
1380 xprt_force_disconnect(xprt);
1384 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1385 xs_tcp_read_common(xprt, desc, req);
1387 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1388 xprt_complete_bc_request(req, transport->tcp_copied);
1389 spin_unlock_bh(&xprt->transport_lock);
1394 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1395 struct xdr_skb_reader *desc)
1397 struct sock_xprt *transport =
1398 container_of(xprt, struct sock_xprt, xprt);
1400 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1401 xs_tcp_read_reply(xprt, desc) :
1402 xs_tcp_read_callback(xprt, desc);
1405 static int xs_tcp_bc_up(struct svc_serv *serv, struct net *net)
1409 ret = svc_create_xprt(serv, "tcp-bc", net, PF_INET, 0,
1410 SVC_SOCK_ANONYMOUS);
1416 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1421 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1422 struct xdr_skb_reader *desc)
1424 return xs_tcp_read_reply(xprt, desc);
1426 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1429 * Read data off the transport. This can be either an RPC_CALL or an
1430 * RPC_REPLY. Relay the processing to helper functions.
1432 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1433 struct xdr_skb_reader *desc)
1435 struct sock_xprt *transport =
1436 container_of(xprt, struct sock_xprt, xprt);
1438 if (_xs_tcp_read_data(xprt, desc) == 0)
1439 xs_tcp_check_fraghdr(transport);
1442 * The transport_lock protects the request handling.
1443 * There's no need to hold it to update the tcp_flags.
1445 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1449 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1453 len = transport->tcp_reclen - transport->tcp_offset;
1454 if (len > desc->count)
1457 desc->offset += len;
1458 transport->tcp_offset += len;
1459 dprintk("RPC: discarded %Zu bytes\n", len);
1460 xs_tcp_check_fraghdr(transport);
1463 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1465 struct rpc_xprt *xprt = rd_desc->arg.data;
1466 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1467 struct xdr_skb_reader desc = {
1473 dprintk("RPC: xs_tcp_data_recv started\n");
1475 trace_xs_tcp_data_recv(transport);
1476 /* Read in a new fragment marker if necessary */
1477 /* Can we ever really expect to get completely empty fragments? */
1478 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1479 xs_tcp_read_fraghdr(xprt, &desc);
1482 /* Read in the xid if necessary */
1483 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1484 xs_tcp_read_xid(transport, &desc);
1487 /* Read in the call/reply flag */
1488 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1489 xs_tcp_read_calldir(transport, &desc);
1492 /* Read in the request data */
1493 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1494 xs_tcp_read_data(xprt, &desc);
1497 /* Skip over any trailing bytes on short reads */
1498 xs_tcp_read_discard(transport, &desc);
1499 } while (desc.count);
1500 trace_xs_tcp_data_recv(transport);
1501 dprintk("RPC: xs_tcp_data_recv done\n");
1502 return len - desc.count;
1505 static void xs_tcp_data_receive(struct sock_xprt *transport)
1507 struct rpc_xprt *xprt = &transport->xprt;
1509 read_descriptor_t rd_desc = {
1510 .count = 2*1024*1024,
1513 unsigned long total = 0;
1516 mutex_lock(&transport->recv_mutex);
1517 sk = transport->inet;
1521 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1524 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1526 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1528 if (!test_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1534 rd_desc.count = 65536;
1537 mutex_unlock(&transport->recv_mutex);
1538 trace_xs_tcp_data_ready(xprt, read, total);
1541 static void xs_tcp_data_receive_workfn(struct work_struct *work)
1543 struct sock_xprt *transport =
1544 container_of(work, struct sock_xprt, recv_worker);
1545 xs_tcp_data_receive(transport);
1549 * xs_tcp_state_change - callback to handle TCP socket state changes
1550 * @sk: socket whose state has changed
1553 static void xs_tcp_state_change(struct sock *sk)
1555 struct rpc_xprt *xprt;
1556 struct sock_xprt *transport;
1558 read_lock_bh(&sk->sk_callback_lock);
1559 if (!(xprt = xprt_from_sock(sk)))
1561 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1562 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1563 sk->sk_state, xprt_connected(xprt),
1564 sock_flag(sk, SOCK_DEAD),
1565 sock_flag(sk, SOCK_ZAPPED),
1568 transport = container_of(xprt, struct sock_xprt, xprt);
1569 trace_rpc_socket_state_change(xprt, sk->sk_socket);
1570 switch (sk->sk_state) {
1571 case TCP_ESTABLISHED:
1572 spin_lock(&xprt->transport_lock);
1573 if (!xprt_test_and_set_connected(xprt)) {
1575 /* Reset TCP record info */
1576 transport->tcp_offset = 0;
1577 transport->tcp_reclen = 0;
1578 transport->tcp_copied = 0;
1579 transport->tcp_flags =
1580 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1581 xprt->connect_cookie++;
1582 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1583 xprt_clear_connecting(xprt);
1585 xprt_wake_pending_tasks(xprt, -EAGAIN);
1587 spin_unlock(&xprt->transport_lock);
1590 /* The client initiated a shutdown of the socket */
1591 xprt->connect_cookie++;
1592 xprt->reestablish_timeout = 0;
1593 set_bit(XPRT_CLOSING, &xprt->state);
1594 smp_mb__before_atomic();
1595 clear_bit(XPRT_CONNECTED, &xprt->state);
1596 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1597 smp_mb__after_atomic();
1599 case TCP_CLOSE_WAIT:
1600 /* The server initiated a shutdown of the socket */
1601 xprt->connect_cookie++;
1602 clear_bit(XPRT_CONNECTED, &xprt->state);
1603 xs_tcp_force_close(xprt);
1606 * If the server closed down the connection, make sure that
1607 * we back off before reconnecting
1609 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1610 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1613 set_bit(XPRT_CLOSING, &xprt->state);
1614 smp_mb__before_atomic();
1615 clear_bit(XPRT_CONNECTED, &xprt->state);
1616 smp_mb__after_atomic();
1619 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1620 &transport->sock_state))
1621 xprt_clear_connecting(xprt);
1622 xs_sock_mark_closed(xprt);
1625 read_unlock_bh(&sk->sk_callback_lock);
1628 static void xs_write_space(struct sock *sk)
1630 struct socket_wq *wq;
1631 struct rpc_xprt *xprt;
1635 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1637 if (unlikely(!(xprt = xprt_from_sock(sk))))
1640 wq = rcu_dereference(sk->sk_wq);
1641 if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1644 xprt_write_space(xprt);
1650 * xs_udp_write_space - callback invoked when socket buffer space
1652 * @sk: socket whose state has changed
1654 * Called when more output buffer space is available for this socket.
1655 * We try not to wake our writers until they can make "significant"
1656 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1657 * with a bunch of small requests.
1659 static void xs_udp_write_space(struct sock *sk)
1661 read_lock_bh(&sk->sk_callback_lock);
1663 /* from net/core/sock.c:sock_def_write_space */
1664 if (sock_writeable(sk))
1667 read_unlock_bh(&sk->sk_callback_lock);
1671 * xs_tcp_write_space - callback invoked when socket buffer space
1673 * @sk: socket whose state has changed
1675 * Called when more output buffer space is available for this socket.
1676 * We try not to wake our writers until they can make "significant"
1677 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1678 * with a bunch of small requests.
1680 static void xs_tcp_write_space(struct sock *sk)
1682 read_lock_bh(&sk->sk_callback_lock);
1684 /* from net/core/stream.c:sk_stream_write_space */
1685 if (sk_stream_is_writeable(sk))
1688 read_unlock_bh(&sk->sk_callback_lock);
1691 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1693 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1694 struct sock *sk = transport->inet;
1696 if (transport->rcvsize) {
1697 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1698 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1700 if (transport->sndsize) {
1701 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1702 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1703 sk->sk_write_space(sk);
1708 * xs_udp_set_buffer_size - set send and receive limits
1709 * @xprt: generic transport
1710 * @sndsize: requested size of send buffer, in bytes
1711 * @rcvsize: requested size of receive buffer, in bytes
1713 * Set socket send and receive buffer size limits.
1715 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1717 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1719 transport->sndsize = 0;
1721 transport->sndsize = sndsize + 1024;
1722 transport->rcvsize = 0;
1724 transport->rcvsize = rcvsize + 1024;
1726 xs_udp_do_set_buffer_size(xprt);
1730 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1731 * @task: task that timed out
1733 * Adjust the congestion window after a retransmit timeout has occurred.
1735 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1737 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1740 static int xs_get_random_port(void)
1742 unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1743 unsigned short range;
1744 unsigned short rand;
1748 range = max - min + 1;
1749 rand = (unsigned short) prandom_u32() % range;
1754 * xs_set_reuseaddr_port - set the socket's port and address reuse options
1757 * Note that this function has to be called on all sockets that share the
1758 * same port, and it must be called before binding.
1760 static void xs_sock_set_reuseport(struct socket *sock)
1764 kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1765 (char *)&opt, sizeof(opt));
1768 static unsigned short xs_sock_getport(struct socket *sock)
1770 struct sockaddr_storage buf;
1772 unsigned short port = 0;
1774 if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0)
1776 switch (buf.ss_family) {
1778 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1781 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1788 * xs_set_port - reset the port number in the remote endpoint address
1789 * @xprt: generic transport
1790 * @port: new port number
1793 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1795 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1797 rpc_set_port(xs_addr(xprt), port);
1798 xs_update_peer_port(xprt);
1801 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1803 if (transport->srcport == 0)
1804 transport->srcport = xs_sock_getport(sock);
1807 static int xs_get_srcport(struct sock_xprt *transport)
1809 int port = transport->srcport;
1811 if (port == 0 && transport->xprt.resvport)
1812 port = xs_get_random_port();
1816 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1818 if (transport->srcport != 0)
1819 transport->srcport = 0;
1820 if (!transport->xprt.resvport)
1822 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1823 return xprt_max_resvport;
1826 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1828 struct sockaddr_storage myaddr;
1830 int port = xs_get_srcport(transport);
1831 unsigned short last;
1834 * If we are asking for any ephemeral port (i.e. port == 0 &&
1835 * transport->xprt.resvport == 0), don't bind. Let the local
1836 * port selection happen implicitly when the socket is used
1837 * (for example at connect time).
1839 * This ensures that we can continue to establish TCP
1840 * connections even when all local ephemeral ports are already
1841 * a part of some TCP connection. This makes no difference
1842 * for UDP sockets, but also doens't harm them.
1844 * If we're asking for any reserved port (i.e. port == 0 &&
1845 * transport->xprt.resvport == 1) xs_get_srcport above will
1846 * ensure that port is non-zero and we will bind as needed.
1851 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1853 rpc_set_port((struct sockaddr *)&myaddr, port);
1854 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1855 transport->xprt.addrlen);
1857 transport->srcport = port;
1861 port = xs_next_srcport(transport, port);
1864 } while (err == -EADDRINUSE && nloop != 2);
1866 if (myaddr.ss_family == AF_INET)
1867 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1868 &((struct sockaddr_in *)&myaddr)->sin_addr,
1869 port, err ? "failed" : "ok", err);
1871 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1872 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1873 port, err ? "failed" : "ok", err);
1878 * We don't support autobind on AF_LOCAL sockets
1880 static void xs_local_rpcbind(struct rpc_task *task)
1882 xprt_set_bound(task->tk_xprt);
1885 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1889 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1890 static struct lock_class_key xs_key[2];
1891 static struct lock_class_key xs_slock_key[2];
1893 static inline void xs_reclassify_socketu(struct socket *sock)
1895 struct sock *sk = sock->sk;
1897 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1898 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1901 static inline void xs_reclassify_socket4(struct socket *sock)
1903 struct sock *sk = sock->sk;
1905 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1906 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1909 static inline void xs_reclassify_socket6(struct socket *sock)
1911 struct sock *sk = sock->sk;
1913 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1914 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1917 static inline void xs_reclassify_socket(int family, struct socket *sock)
1919 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1924 xs_reclassify_socketu(sock);
1927 xs_reclassify_socket4(sock);
1930 xs_reclassify_socket6(sock);
1935 static inline void xs_reclassify_socket(int family, struct socket *sock)
1940 static void xs_dummy_setup_socket(struct work_struct *work)
1944 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1945 struct sock_xprt *transport, int family, int type,
1946 int protocol, bool reuseport)
1948 struct socket *sock;
1951 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1953 dprintk("RPC: can't create %d transport socket (%d).\n",
1957 xs_reclassify_socket(family, sock);
1960 xs_sock_set_reuseport(sock);
1962 err = xs_bind(transport, sock);
1970 return ERR_PTR(err);
1973 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1974 struct socket *sock)
1976 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1979 if (!transport->inet) {
1980 struct sock *sk = sock->sk;
1982 write_lock_bh(&sk->sk_callback_lock);
1984 xs_save_old_callbacks(transport, sk);
1986 sk->sk_user_data = xprt;
1987 sk->sk_data_ready = xs_data_ready;
1988 sk->sk_write_space = xs_udp_write_space;
1989 sock_set_flag(sk, SOCK_FASYNC);
1990 sk->sk_error_report = xs_error_report;
1991 sk->sk_allocation = GFP_NOIO;
1993 xprt_clear_connected(xprt);
1995 /* Reset to new socket */
1996 transport->sock = sock;
1997 transport->inet = sk;
1999 write_unlock_bh(&sk->sk_callback_lock);
2002 /* Tell the socket layer to start connecting... */
2003 xprt->stat.connect_count++;
2004 xprt->stat.connect_start = jiffies;
2005 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
2009 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
2010 * @transport: socket transport to connect
2012 static int xs_local_setup_socket(struct sock_xprt *transport)
2014 struct rpc_xprt *xprt = &transport->xprt;
2015 struct socket *sock;
2018 status = __sock_create(xprt->xprt_net, AF_LOCAL,
2019 SOCK_STREAM, 0, &sock, 1);
2021 dprintk("RPC: can't create AF_LOCAL "
2022 "transport socket (%d).\n", -status);
2025 xs_reclassify_socket(AF_LOCAL, sock);
2027 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
2028 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2030 status = xs_local_finish_connecting(xprt, sock);
2031 trace_rpc_socket_connect(xprt, sock, status);
2034 dprintk("RPC: xprt %p connected to %s\n",
2035 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2036 xprt_set_connected(xprt);
2040 dprintk("RPC: xprt %p: socket %s does not exist\n",
2041 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2044 dprintk("RPC: xprt %p: connection refused for %s\n",
2045 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2048 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2050 xprt->address_strings[RPC_DISPLAY_ADDR]);
2054 xprt_clear_connecting(xprt);
2055 xprt_wake_pending_tasks(xprt, status);
2059 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2061 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2064 if (RPC_IS_ASYNC(task)) {
2066 * We want the AF_LOCAL connect to be resolved in the
2067 * filesystem namespace of the process making the rpc
2068 * call. Thus we connect synchronously.
2070 * If we want to support asynchronous AF_LOCAL calls,
2071 * we'll need to figure out how to pass a namespace to
2074 rpc_exit(task, -ENOTCONN);
2077 ret = xs_local_setup_socket(transport);
2078 if (ret && !RPC_IS_SOFTCONN(task))
2079 msleep_interruptible(15000);
2082 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2084 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2085 * know that we have exclusive access to the socket), to guard against
2086 * races with xs_reset_transport.
2088 static void xs_set_memalloc(struct rpc_xprt *xprt)
2090 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2094 * If there's no sock, then we have nothing to set. The
2095 * reconnecting process will get it for us.
2097 if (!transport->inet)
2099 if (atomic_read(&xprt->swapper))
2100 sk_set_memalloc(transport->inet);
2104 * xs_enable_swap - Tag this transport as being used for swap.
2105 * @xprt: transport to tag
2107 * Take a reference to this transport on behalf of the rpc_clnt, and
2108 * optionally mark it for swapping if it wasn't already.
2111 xs_enable_swap(struct rpc_xprt *xprt)
2113 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2115 if (atomic_inc_return(&xprt->swapper) != 1)
2117 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2118 return -ERESTARTSYS;
2120 sk_set_memalloc(xs->inet);
2121 xprt_release_xprt(xprt, NULL);
2126 * xs_disable_swap - Untag this transport as being used for swap.
2127 * @xprt: transport to tag
2129 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2130 * swapper refcount goes to 0, untag the socket as a memalloc socket.
2133 xs_disable_swap(struct rpc_xprt *xprt)
2135 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2137 if (!atomic_dec_and_test(&xprt->swapper))
2139 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2142 sk_clear_memalloc(xs->inet);
2143 xprt_release_xprt(xprt, NULL);
2146 static void xs_set_memalloc(struct rpc_xprt *xprt)
2151 xs_enable_swap(struct rpc_xprt *xprt)
2157 xs_disable_swap(struct rpc_xprt *xprt)
2162 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2164 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2166 if (!transport->inet) {
2167 struct sock *sk = sock->sk;
2169 write_lock_bh(&sk->sk_callback_lock);
2171 xs_save_old_callbacks(transport, sk);
2173 sk->sk_user_data = xprt;
2174 sk->sk_data_ready = xs_data_ready;
2175 sk->sk_write_space = xs_udp_write_space;
2176 sock_set_flag(sk, SOCK_FASYNC);
2177 sk->sk_allocation = GFP_NOIO;
2179 xprt_set_connected(xprt);
2181 /* Reset to new socket */
2182 transport->sock = sock;
2183 transport->inet = sk;
2185 xs_set_memalloc(xprt);
2187 write_unlock_bh(&sk->sk_callback_lock);
2189 xs_udp_do_set_buffer_size(xprt);
2191 xprt->stat.connect_start = jiffies;
2194 static void xs_udp_setup_socket(struct work_struct *work)
2196 struct sock_xprt *transport =
2197 container_of(work, struct sock_xprt, connect_worker.work);
2198 struct rpc_xprt *xprt = &transport->xprt;
2199 struct socket *sock = transport->sock;
2202 sock = xs_create_sock(xprt, transport,
2203 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2204 IPPROTO_UDP, false);
2208 dprintk("RPC: worker connecting xprt %p via %s to "
2209 "%s (port %s)\n", xprt,
2210 xprt->address_strings[RPC_DISPLAY_PROTO],
2211 xprt->address_strings[RPC_DISPLAY_ADDR],
2212 xprt->address_strings[RPC_DISPLAY_PORT]);
2214 xs_udp_finish_connecting(xprt, sock);
2215 trace_rpc_socket_connect(xprt, sock, 0);
2218 xprt_clear_connecting(xprt);
2219 xprt_unlock_connect(xprt, transport);
2220 xprt_wake_pending_tasks(xprt, status);
2224 * xs_tcp_shutdown - gracefully shut down a TCP socket
2227 * Initiates a graceful shutdown of the TCP socket by calling the
2228 * equivalent of shutdown(SHUT_RDWR);
2230 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2232 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2233 struct socket *sock = transport->sock;
2237 if (xprt_connected(xprt)) {
2238 kernel_sock_shutdown(sock, SHUT_RDWR);
2239 trace_rpc_socket_shutdown(xprt, sock);
2241 xs_reset_transport(transport);
2244 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2246 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2247 int ret = -ENOTCONN;
2249 if (!transport->inet) {
2250 struct sock *sk = sock->sk;
2251 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2252 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2253 unsigned int opt_on = 1;
2255 unsigned int addr_pref = IPV6_PREFER_SRC_PUBLIC;
2257 /* TCP Keepalive options */
2258 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2259 (char *)&opt_on, sizeof(opt_on));
2260 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2261 (char *)&keepidle, sizeof(keepidle));
2262 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2263 (char *)&keepidle, sizeof(keepidle));
2264 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2265 (char *)&keepcnt, sizeof(keepcnt));
2267 /* Avoid temporary address, they are bad for long-lived
2268 * connections such as NFS mounts.
2269 * RFC4941, section 3.6 suggests that:
2270 * Individual applications, which have specific
2271 * knowledge about the normal duration of connections,
2272 * MAY override this as appropriate.
2274 kernel_setsockopt(sock, SOL_IPV6, IPV6_ADDR_PREFERENCES,
2275 (char *)&addr_pref, sizeof(addr_pref));
2277 /* TCP user timeout (see RFC5482) */
2278 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2279 (xprt->timeout->to_retries + 1);
2280 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2281 (char *)&timeo, sizeof(timeo));
2283 write_lock_bh(&sk->sk_callback_lock);
2285 xs_save_old_callbacks(transport, sk);
2287 sk->sk_user_data = xprt;
2288 sk->sk_data_ready = xs_data_ready;
2289 sk->sk_state_change = xs_tcp_state_change;
2290 sk->sk_write_space = xs_tcp_write_space;
2291 sock_set_flag(sk, SOCK_FASYNC);
2292 sk->sk_error_report = xs_error_report;
2293 sk->sk_allocation = GFP_NOIO;
2295 /* socket options */
2296 sock_reset_flag(sk, SOCK_LINGER);
2297 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2299 xprt_clear_connected(xprt);
2301 /* Reset to new socket */
2302 transport->sock = sock;
2303 transport->inet = sk;
2305 write_unlock_bh(&sk->sk_callback_lock);
2308 if (!xprt_bound(xprt))
2311 xs_set_memalloc(xprt);
2313 /* Tell the socket layer to start connecting... */
2314 xprt->stat.connect_count++;
2315 xprt->stat.connect_start = jiffies;
2316 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2317 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2320 xs_set_srcport(transport, sock);
2323 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2324 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2326 case -EADDRNOTAVAIL:
2327 /* Source port number is unavailable. Try a new one! */
2328 transport->srcport = 0;
2335 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2337 * Invoked by a work queue tasklet.
2339 static void xs_tcp_setup_socket(struct work_struct *work)
2341 struct sock_xprt *transport =
2342 container_of(work, struct sock_xprt, connect_worker.work);
2343 struct socket *sock = transport->sock;
2344 struct rpc_xprt *xprt = &transport->xprt;
2348 sock = xs_create_sock(xprt, transport,
2349 xs_addr(xprt)->sa_family, SOCK_STREAM,
2352 status = PTR_ERR(sock);
2357 dprintk("RPC: worker connecting xprt %p via %s to "
2358 "%s (port %s)\n", xprt,
2359 xprt->address_strings[RPC_DISPLAY_PROTO],
2360 xprt->address_strings[RPC_DISPLAY_ADDR],
2361 xprt->address_strings[RPC_DISPLAY_PORT]);
2363 status = xs_tcp_finish_connecting(xprt, sock);
2364 trace_rpc_socket_connect(xprt, sock, status);
2365 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2366 xprt, -status, xprt_connected(xprt),
2367 sock->sk->sk_state);
2370 printk("%s: connect returned unhandled error %d\n",
2372 case -EADDRNOTAVAIL:
2373 /* We're probably in TIME_WAIT. Get rid of existing socket,
2376 xs_tcp_force_close(xprt);
2381 xprt_unlock_connect(xprt, transport);
2384 /* Happens, for instance, if the user specified a link
2385 * local IPv6 address without a scope-id.
2394 * xs_tcp_force_close() wakes tasks with -EIO.
2395 * We need to wake them first to ensure the
2396 * correct error code.
2398 xprt_wake_pending_tasks(xprt, status);
2399 xs_tcp_force_close(xprt);
2404 xprt_clear_connecting(xprt);
2405 xprt_unlock_connect(xprt, transport);
2406 xprt_wake_pending_tasks(xprt, status);
2409 static unsigned long xs_reconnect_delay(const struct rpc_xprt *xprt)
2411 unsigned long start, now = jiffies;
2413 start = xprt->stat.connect_start + xprt->reestablish_timeout;
2414 if (time_after(start, now))
2419 static void xs_reconnect_backoff(struct rpc_xprt *xprt)
2421 xprt->reestablish_timeout <<= 1;
2422 if (xprt->reestablish_timeout > xprt->max_reconnect_timeout)
2423 xprt->reestablish_timeout = xprt->max_reconnect_timeout;
2424 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2425 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2429 * xs_connect - connect a socket to a remote endpoint
2430 * @xprt: pointer to transport structure
2431 * @task: address of RPC task that manages state of connect request
2433 * TCP: If the remote end dropped the connection, delay reconnecting.
2435 * UDP socket connects are synchronous, but we use a work queue anyway
2436 * to guarantee that even unprivileged user processes can set up a
2437 * socket on a privileged port.
2439 * If a UDP socket connect fails, the delay behavior here prevents
2440 * retry floods (hard mounts).
2442 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2444 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2445 unsigned long delay = 0;
2447 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2449 if (transport->sock != NULL) {
2450 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2452 xprt, xprt->reestablish_timeout / HZ);
2454 /* Start by resetting any existing state */
2455 xs_reset_transport(transport);
2457 delay = xs_reconnect_delay(xprt);
2458 xs_reconnect_backoff(xprt);
2461 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2463 queue_delayed_work(xprtiod_workqueue,
2464 &transport->connect_worker,
2469 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2470 * @xprt: rpc_xprt struct containing statistics
2474 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2478 if (xprt_connected(xprt))
2479 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2481 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2482 "%llu %llu %lu %llu %llu\n",
2483 xprt->stat.bind_count,
2484 xprt->stat.connect_count,
2485 xprt->stat.connect_time,
2489 xprt->stat.bad_xids,
2492 xprt->stat.max_slots,
2493 xprt->stat.sending_u,
2494 xprt->stat.pending_u);
2498 * xs_udp_print_stats - display UDP socket-specifc stats
2499 * @xprt: rpc_xprt struct containing statistics
2503 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2505 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2507 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2510 xprt->stat.bind_count,
2513 xprt->stat.bad_xids,
2516 xprt->stat.max_slots,
2517 xprt->stat.sending_u,
2518 xprt->stat.pending_u);
2522 * xs_tcp_print_stats - display TCP socket-specifc stats
2523 * @xprt: rpc_xprt struct containing statistics
2527 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2529 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2532 if (xprt_connected(xprt))
2533 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2535 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2536 "%llu %llu %lu %llu %llu\n",
2538 xprt->stat.bind_count,
2539 xprt->stat.connect_count,
2540 xprt->stat.connect_time,
2544 xprt->stat.bad_xids,
2547 xprt->stat.max_slots,
2548 xprt->stat.sending_u,
2549 xprt->stat.pending_u);
2553 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2554 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2555 * to use the server side send routines.
2557 static int bc_malloc(struct rpc_task *task)
2559 struct rpc_rqst *rqst = task->tk_rqstp;
2560 size_t size = rqst->rq_callsize;
2562 struct rpc_buffer *buf;
2564 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2565 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2570 page = alloc_page(GFP_KERNEL);
2574 buf = page_address(page);
2575 buf->len = PAGE_SIZE;
2577 rqst->rq_buffer = buf->data;
2578 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2583 * Free the space allocated in the bc_alloc routine
2585 static void bc_free(struct rpc_task *task)
2587 void *buffer = task->tk_rqstp->rq_buffer;
2588 struct rpc_buffer *buf;
2590 buf = container_of(buffer, struct rpc_buffer, data);
2591 free_page((unsigned long)buf);
2595 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2596 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2598 static int bc_sendto(struct rpc_rqst *req)
2601 struct xdr_buf *xbufp = &req->rq_snd_buf;
2602 struct rpc_xprt *xprt = req->rq_xprt;
2603 struct sock_xprt *transport =
2604 container_of(xprt, struct sock_xprt, xprt);
2605 struct socket *sock = transport->sock;
2606 unsigned long headoff;
2607 unsigned long tailoff;
2609 xs_encode_stream_record_marker(xbufp);
2611 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2612 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2613 len = svc_send_common(sock, xbufp,
2614 virt_to_page(xbufp->head[0].iov_base), headoff,
2615 xbufp->tail[0].iov_base, tailoff);
2617 if (len != xbufp->len) {
2618 printk(KERN_NOTICE "Error sending entire callback!\n");
2626 * The send routine. Borrows from svc_send
2628 static int bc_send_request(struct rpc_task *task)
2630 struct rpc_rqst *req = task->tk_rqstp;
2631 struct svc_xprt *xprt;
2634 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2636 * Get the server socket associated with this callback xprt
2638 xprt = req->rq_xprt->bc_xprt;
2641 * Grab the mutex to serialize data as the connection is shared
2642 * with the fore channel
2644 if (!mutex_trylock(&xprt->xpt_mutex)) {
2645 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2646 if (!mutex_trylock(&xprt->xpt_mutex))
2648 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2650 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2653 len = bc_sendto(req);
2654 mutex_unlock(&xprt->xpt_mutex);
2663 * The close routine. Since this is client initiated, we do nothing
2666 static void bc_close(struct rpc_xprt *xprt)
2671 * The xprt destroy routine. Again, because this connection is client
2672 * initiated, we do nothing
2675 static void bc_destroy(struct rpc_xprt *xprt)
2677 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2680 module_put(THIS_MODULE);
2683 static struct rpc_xprt_ops xs_local_ops = {
2684 .reserve_xprt = xprt_reserve_xprt,
2685 .release_xprt = xs_tcp_release_xprt,
2686 .alloc_slot = xprt_alloc_slot,
2687 .rpcbind = xs_local_rpcbind,
2688 .set_port = xs_local_set_port,
2689 .connect = xs_local_connect,
2690 .buf_alloc = rpc_malloc,
2691 .buf_free = rpc_free,
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 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 .rpcbind = rpcb_getport_async,
2707 .set_port = xs_set_port,
2708 .connect = xs_connect,
2709 .buf_alloc = rpc_malloc,
2710 .buf_free = rpc_free,
2711 .send_request = xs_udp_send_request,
2712 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
2713 .timer = xs_udp_timer,
2714 .release_request = xprt_release_rqst_cong,
2716 .destroy = xs_destroy,
2717 .print_stats = xs_udp_print_stats,
2718 .enable_swap = xs_enable_swap,
2719 .disable_swap = xs_disable_swap,
2720 .inject_disconnect = xs_inject_disconnect,
2723 static struct rpc_xprt_ops xs_tcp_ops = {
2724 .reserve_xprt = xprt_reserve_xprt,
2725 .release_xprt = xs_tcp_release_xprt,
2726 .alloc_slot = xprt_lock_and_alloc_slot,
2727 .rpcbind = rpcb_getport_async,
2728 .set_port = xs_set_port,
2729 .connect = xs_connect,
2730 .buf_alloc = rpc_malloc,
2731 .buf_free = rpc_free,
2732 .send_request = xs_tcp_send_request,
2733 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2734 .close = xs_tcp_shutdown,
2735 .destroy = xs_destroy,
2736 .print_stats = xs_tcp_print_stats,
2737 .enable_swap = xs_enable_swap,
2738 .disable_swap = xs_disable_swap,
2739 .inject_disconnect = xs_inject_disconnect,
2740 #ifdef CONFIG_SUNRPC_BACKCHANNEL
2741 .bc_setup = xprt_setup_bc,
2742 .bc_up = xs_tcp_bc_up,
2743 .bc_maxpayload = xs_tcp_bc_maxpayload,
2744 .bc_free_rqst = xprt_free_bc_rqst,
2745 .bc_destroy = xprt_destroy_bc,
2750 * The rpc_xprt_ops for the server backchannel
2753 static struct rpc_xprt_ops bc_tcp_ops = {
2754 .reserve_xprt = xprt_reserve_xprt,
2755 .release_xprt = xprt_release_xprt,
2756 .alloc_slot = xprt_alloc_slot,
2757 .buf_alloc = bc_malloc,
2758 .buf_free = bc_free,
2759 .send_request = bc_send_request,
2760 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2762 .destroy = bc_destroy,
2763 .print_stats = xs_tcp_print_stats,
2764 .enable_swap = xs_enable_swap,
2765 .disable_swap = xs_disable_swap,
2766 .inject_disconnect = xs_inject_disconnect,
2769 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2771 static const struct sockaddr_in sin = {
2772 .sin_family = AF_INET,
2773 .sin_addr.s_addr = htonl(INADDR_ANY),
2775 static const struct sockaddr_in6 sin6 = {
2776 .sin6_family = AF_INET6,
2777 .sin6_addr = IN6ADDR_ANY_INIT,
2784 memcpy(sap, &sin, sizeof(sin));
2787 memcpy(sap, &sin6, sizeof(sin6));
2790 dprintk("RPC: %s: Bad address family\n", __func__);
2791 return -EAFNOSUPPORT;
2796 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2797 unsigned int slot_table_size,
2798 unsigned int max_slot_table_size)
2800 struct rpc_xprt *xprt;
2801 struct sock_xprt *new;
2803 if (args->addrlen > sizeof(xprt->addr)) {
2804 dprintk("RPC: xs_setup_xprt: address too large\n");
2805 return ERR_PTR(-EBADF);
2808 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2809 max_slot_table_size);
2811 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2813 return ERR_PTR(-ENOMEM);
2816 new = container_of(xprt, struct sock_xprt, xprt);
2817 mutex_init(&new->recv_mutex);
2818 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2819 xprt->addrlen = args->addrlen;
2821 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2824 err = xs_init_anyaddr(args->dstaddr->sa_family,
2825 (struct sockaddr *)&new->srcaddr);
2828 return ERR_PTR(err);
2835 static const struct rpc_timeout xs_local_default_timeout = {
2836 .to_initval = 10 * HZ,
2837 .to_maxval = 10 * HZ,
2842 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2843 * @args: rpc transport creation arguments
2845 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2847 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2849 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2850 struct sock_xprt *transport;
2851 struct rpc_xprt *xprt;
2852 struct rpc_xprt *ret;
2854 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2855 xprt_max_tcp_slot_table_entries);
2858 transport = container_of(xprt, struct sock_xprt, xprt);
2861 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2862 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2864 xprt->bind_timeout = XS_BIND_TO;
2865 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2866 xprt->idle_timeout = XS_IDLE_DISC_TO;
2868 xprt->ops = &xs_local_ops;
2869 xprt->timeout = &xs_local_default_timeout;
2871 INIT_WORK(&transport->recv_worker, xs_local_data_receive_workfn);
2872 INIT_DELAYED_WORK(&transport->connect_worker,
2873 xs_dummy_setup_socket);
2875 switch (sun->sun_family) {
2877 if (sun->sun_path[0] != '/') {
2878 dprintk("RPC: bad AF_LOCAL address: %s\n",
2880 ret = ERR_PTR(-EINVAL);
2883 xprt_set_bound(xprt);
2884 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2885 ret = ERR_PTR(xs_local_setup_socket(transport));
2890 ret = ERR_PTR(-EAFNOSUPPORT);
2894 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2895 xprt->address_strings[RPC_DISPLAY_ADDR]);
2897 if (try_module_get(THIS_MODULE))
2899 ret = ERR_PTR(-EINVAL);
2905 static const struct rpc_timeout xs_udp_default_timeout = {
2906 .to_initval = 5 * HZ,
2907 .to_maxval = 30 * HZ,
2908 .to_increment = 5 * HZ,
2913 * xs_setup_udp - Set up transport to use a UDP socket
2914 * @args: rpc transport creation arguments
2917 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2919 struct sockaddr *addr = args->dstaddr;
2920 struct rpc_xprt *xprt;
2921 struct sock_xprt *transport;
2922 struct rpc_xprt *ret;
2924 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2925 xprt_udp_slot_table_entries);
2928 transport = container_of(xprt, struct sock_xprt, xprt);
2930 xprt->prot = IPPROTO_UDP;
2932 /* XXX: header size can vary due to auth type, IPv6, etc. */
2933 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2935 xprt->bind_timeout = XS_BIND_TO;
2936 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2937 xprt->idle_timeout = XS_IDLE_DISC_TO;
2939 xprt->ops = &xs_udp_ops;
2941 xprt->timeout = &xs_udp_default_timeout;
2943 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
2944 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2946 switch (addr->sa_family) {
2948 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2949 xprt_set_bound(xprt);
2951 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2954 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2955 xprt_set_bound(xprt);
2957 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2960 ret = ERR_PTR(-EAFNOSUPPORT);
2964 if (xprt_bound(xprt))
2965 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2966 xprt->address_strings[RPC_DISPLAY_ADDR],
2967 xprt->address_strings[RPC_DISPLAY_PORT],
2968 xprt->address_strings[RPC_DISPLAY_PROTO]);
2970 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2971 xprt->address_strings[RPC_DISPLAY_ADDR],
2972 xprt->address_strings[RPC_DISPLAY_PROTO]);
2974 if (try_module_get(THIS_MODULE))
2976 ret = ERR_PTR(-EINVAL);
2982 static const struct rpc_timeout xs_tcp_default_timeout = {
2983 .to_initval = 60 * HZ,
2984 .to_maxval = 60 * HZ,
2989 * xs_setup_tcp - Set up transport to use a TCP socket
2990 * @args: rpc transport creation arguments
2993 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2995 struct sockaddr *addr = args->dstaddr;
2996 struct rpc_xprt *xprt;
2997 struct sock_xprt *transport;
2998 struct rpc_xprt *ret;
2999 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
3001 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
3002 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
3004 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3005 max_slot_table_size);
3008 transport = container_of(xprt, struct sock_xprt, xprt);
3010 xprt->prot = IPPROTO_TCP;
3011 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
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;
3023 INIT_WORK(&transport->recv_worker, xs_tcp_data_receive_workfn);
3024 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
3026 switch (addr->sa_family) {
3028 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3029 xprt_set_bound(xprt);
3031 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3034 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3035 xprt_set_bound(xprt);
3037 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3040 ret = ERR_PTR(-EAFNOSUPPORT);
3044 if (xprt_bound(xprt))
3045 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3046 xprt->address_strings[RPC_DISPLAY_ADDR],
3047 xprt->address_strings[RPC_DISPLAY_PORT],
3048 xprt->address_strings[RPC_DISPLAY_PROTO]);
3050 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
3051 xprt->address_strings[RPC_DISPLAY_ADDR],
3052 xprt->address_strings[RPC_DISPLAY_PROTO]);
3054 if (try_module_get(THIS_MODULE))
3056 ret = ERR_PTR(-EINVAL);
3063 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3064 * @args: rpc transport creation arguments
3067 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3069 struct sockaddr *addr = args->dstaddr;
3070 struct rpc_xprt *xprt;
3071 struct sock_xprt *transport;
3072 struct svc_sock *bc_sock;
3073 struct rpc_xprt *ret;
3075 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3076 xprt_tcp_slot_table_entries);
3079 transport = container_of(xprt, struct sock_xprt, xprt);
3081 xprt->prot = IPPROTO_TCP;
3082 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3083 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3084 xprt->timeout = &xs_tcp_default_timeout;
3087 xprt_set_bound(xprt);
3088 xprt->bind_timeout = 0;
3089 xprt->reestablish_timeout = 0;
3090 xprt->idle_timeout = 0;
3092 xprt->ops = &bc_tcp_ops;
3094 switch (addr->sa_family) {
3096 xs_format_peer_addresses(xprt, "tcp",
3100 xs_format_peer_addresses(xprt, "tcp",
3101 RPCBIND_NETID_TCP6);
3104 ret = ERR_PTR(-EAFNOSUPPORT);
3108 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3109 xprt->address_strings[RPC_DISPLAY_ADDR],
3110 xprt->address_strings[RPC_DISPLAY_PORT],
3111 xprt->address_strings[RPC_DISPLAY_PROTO]);
3114 * Once we've associated a backchannel xprt with a connection,
3115 * we want to keep it around as long as the connection lasts,
3116 * in case we need to start using it for a backchannel again;
3117 * this reference won't be dropped until bc_xprt is destroyed.
3120 args->bc_xprt->xpt_bc_xprt = xprt;
3121 xprt->bc_xprt = args->bc_xprt;
3122 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3123 transport->sock = bc_sock->sk_sock;
3124 transport->inet = bc_sock->sk_sk;
3127 * Since we don't want connections for the backchannel, we set
3128 * the xprt status to connected
3130 xprt_set_connected(xprt);
3132 if (try_module_get(THIS_MODULE))
3135 args->bc_xprt->xpt_bc_xprt = NULL;
3136 args->bc_xprt->xpt_bc_xps = NULL;
3138 ret = ERR_PTR(-EINVAL);
3144 static struct xprt_class xs_local_transport = {
3145 .list = LIST_HEAD_INIT(xs_local_transport.list),
3146 .name = "named UNIX socket",
3147 .owner = THIS_MODULE,
3148 .ident = XPRT_TRANSPORT_LOCAL,
3149 .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,
3159 .netid = { "udp", "udp6", "" },
3162 static struct xprt_class xs_tcp_transport = {
3163 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3165 .owner = THIS_MODULE,
3166 .ident = XPRT_TRANSPORT_TCP,
3167 .setup = xs_setup_tcp,
3168 .netid = { "tcp", "tcp6", "" },
3171 static struct xprt_class xs_bc_tcp_transport = {
3172 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3173 .name = "tcp NFSv4.1 backchannel",
3174 .owner = THIS_MODULE,
3175 .ident = XPRT_TRANSPORT_BC_TCP,
3176 .setup = xs_setup_bc_tcp,
3181 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3184 int init_socket_xprt(void)
3186 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3187 if (!sunrpc_table_header)
3188 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3191 xprt_register_transport(&xs_local_transport);
3192 xprt_register_transport(&xs_udp_transport);
3193 xprt_register_transport(&xs_tcp_transport);
3194 xprt_register_transport(&xs_bc_tcp_transport);
3200 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3203 void cleanup_socket_xprt(void)
3205 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3206 if (sunrpc_table_header) {
3207 unregister_sysctl_table(sunrpc_table_header);
3208 sunrpc_table_header = NULL;
3212 xprt_unregister_transport(&xs_local_transport);
3213 xprt_unregister_transport(&xs_udp_transport);
3214 xprt_unregister_transport(&xs_tcp_transport);
3215 xprt_unregister_transport(&xs_bc_tcp_transport);
3218 static int param_set_uint_minmax(const char *val,
3219 const struct kernel_param *kp,
3220 unsigned int min, unsigned int max)
3227 ret = kstrtouint(val, 0, &num);
3228 if (ret == -EINVAL || num < min || num > max)
3230 *((unsigned int *)kp->arg) = num;
3234 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3236 return param_set_uint_minmax(val, kp,
3241 static const struct kernel_param_ops param_ops_portnr = {
3242 .set = param_set_portnr,
3243 .get = param_get_uint,
3246 #define param_check_portnr(name, p) \
3247 __param_check(name, p, unsigned int);
3249 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3250 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3252 static int param_set_slot_table_size(const char *val,
3253 const struct kernel_param *kp)
3255 return param_set_uint_minmax(val, kp,
3257 RPC_MAX_SLOT_TABLE);
3260 static const struct kernel_param_ops param_ops_slot_table_size = {
3261 .set = param_set_slot_table_size,
3262 .get = param_get_uint,
3265 #define param_check_slot_table_size(name, p) \
3266 __param_check(name, p, unsigned int);
3268 static int param_set_max_slot_table_size(const char *val,
3269 const struct kernel_param *kp)
3271 return param_set_uint_minmax(val, kp,
3273 RPC_MAX_SLOT_TABLE_LIMIT);
3276 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3277 .set = param_set_max_slot_table_size,
3278 .get = param_get_uint,
3281 #define param_check_max_slot_table_size(name, p) \
3282 __param_check(name, p, unsigned int);
3284 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3285 slot_table_size, 0644);
3286 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3287 max_slot_table_size, 0644);
3288 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3289 slot_table_size, 0644);