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)
180 #define XS_TCP_MAX_REEST_TO (5U * 60 * HZ)
183 * TCP idle timeout; client drops the transport socket if it is idle
184 * for this long. Note that we also timeout UDP sockets to prevent
185 * holding port numbers when there is no RPC traffic.
187 #define XS_IDLE_DISC_TO (5U * 60 * HZ)
189 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
190 # undef RPC_DEBUG_DATA
191 # define RPCDBG_FACILITY RPCDBG_TRANS
194 #ifdef RPC_DEBUG_DATA
195 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
197 u8 *buf = (u8 *) packet;
200 dprintk("RPC: %s\n", msg);
201 for (j = 0; j < count && j < 128; j += 4) {
205 dprintk("0x%04x ", j);
207 dprintk("%02x%02x%02x%02x ",
208 buf[j], buf[j+1], buf[j+2], buf[j+3]);
213 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
219 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
221 return (struct rpc_xprt *) sk->sk_user_data;
224 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
226 return (struct sockaddr *) &xprt->addr;
229 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
231 return (struct sockaddr_un *) &xprt->addr;
234 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
236 return (struct sockaddr_in *) &xprt->addr;
239 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
241 return (struct sockaddr_in6 *) &xprt->addr;
244 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
246 struct sockaddr *sap = xs_addr(xprt);
247 struct sockaddr_in6 *sin6;
248 struct sockaddr_in *sin;
249 struct sockaddr_un *sun;
252 switch (sap->sa_family) {
254 sun = xs_addr_un(xprt);
255 strlcpy(buf, sun->sun_path, sizeof(buf));
256 xprt->address_strings[RPC_DISPLAY_ADDR] =
257 kstrdup(buf, GFP_KERNEL);
260 (void)rpc_ntop(sap, buf, sizeof(buf));
261 xprt->address_strings[RPC_DISPLAY_ADDR] =
262 kstrdup(buf, GFP_KERNEL);
263 sin = xs_addr_in(xprt);
264 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
267 (void)rpc_ntop(sap, buf, sizeof(buf));
268 xprt->address_strings[RPC_DISPLAY_ADDR] =
269 kstrdup(buf, GFP_KERNEL);
270 sin6 = xs_addr_in6(xprt);
271 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
277 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
280 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
282 struct sockaddr *sap = xs_addr(xprt);
285 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
286 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
288 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
289 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
292 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
293 const char *protocol,
296 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
297 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
298 xs_format_common_peer_addresses(xprt);
299 xs_format_common_peer_ports(xprt);
302 static void xs_update_peer_port(struct rpc_xprt *xprt)
304 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
305 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
307 xs_format_common_peer_ports(xprt);
310 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
314 for (i = 0; i < RPC_DISPLAY_MAX; i++)
316 case RPC_DISPLAY_PROTO:
317 case RPC_DISPLAY_NETID:
320 kfree(xprt->address_strings[i]);
324 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
326 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
328 struct msghdr msg = {
330 .msg_namelen = addrlen,
331 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
334 .iov_base = vec->iov_base + base,
335 .iov_len = vec->iov_len - base,
338 if (iov.iov_len != 0)
339 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
340 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
343 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p)
345 ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
346 int offset, size_t size, int flags);
348 unsigned int remainder;
351 remainder = xdr->page_len - base;
352 base += xdr->page_base;
353 ppage = xdr->pages + (base >> PAGE_SHIFT);
355 do_sendpage = sock->ops->sendpage;
357 do_sendpage = sock_no_sendpage;
359 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
360 int flags = XS_SENDMSG_FLAGS;
366 flags |= MSG_SENDPAGE_NOTLAST | MSG_MORE;
367 err = do_sendpage(sock, *ppage, base, len, flags);
368 if (remainder == 0 || err != len)
382 * xs_sendpages - write pages directly to a socket
383 * @sock: socket to send on
384 * @addr: UDP only -- address of destination
385 * @addrlen: UDP only -- length of destination address
386 * @xdr: buffer containing this request
387 * @base: starting position in the buffer
388 * @zerocopy: true if it is safe to use sendpage()
389 * @sent_p: return the total number of bytes successfully queued for sending
392 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p)
394 unsigned int remainder = xdr->len - base;
406 if (base < xdr->head[0].iov_len || addr != NULL) {
407 unsigned int len = xdr->head[0].iov_len - base;
409 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
410 if (remainder == 0 || err != len)
415 base -= xdr->head[0].iov_len;
417 if (base < xdr->page_len) {
418 unsigned int len = xdr->page_len - base;
420 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy, &sent);
422 if (remainder == 0 || sent != len)
426 base -= xdr->page_len;
428 if (base >= xdr->tail[0].iov_len)
430 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
439 static void xs_nospace_callback(struct rpc_task *task)
441 struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
443 transport->inet->sk_write_pending--;
447 * xs_nospace - place task on wait queue if transmit was incomplete
448 * @task: task to put to sleep
451 static int xs_nospace(struct rpc_task *task)
453 struct rpc_rqst *req = task->tk_rqstp;
454 struct rpc_xprt *xprt = req->rq_xprt;
455 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
456 struct sock *sk = transport->inet;
459 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
460 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
463 /* Protect against races with write_space */
464 spin_lock_bh(&xprt->transport_lock);
466 /* Don't race with disconnect */
467 if (xprt_connected(xprt)) {
468 /* wait for more buffer space */
469 sk->sk_write_pending++;
470 xprt_wait_for_buffer_space(task, xs_nospace_callback);
474 spin_unlock_bh(&xprt->transport_lock);
476 /* Race breaker in case memory is freed before above code is called */
477 if (ret == -EAGAIN) {
478 struct socket_wq *wq;
481 wq = rcu_dereference(sk->sk_wq);
482 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
485 sk->sk_write_space(sk);
491 * Construct a stream transport record marker in @buf.
493 static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
495 u32 reclen = buf->len - sizeof(rpc_fraghdr);
496 rpc_fraghdr *base = buf->head[0].iov_base;
497 *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
501 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
502 * @task: RPC task that manages the state of an RPC request
505 * 0: The request has been sent
506 * EAGAIN: The socket was blocked, please call again later to
507 * complete the request
508 * ENOTCONN: Caller needs to invoke connect logic then call again
509 * other: Some other error occured, the request was not sent
511 static int xs_local_send_request(struct rpc_task *task)
513 struct rpc_rqst *req = task->tk_rqstp;
514 struct rpc_xprt *xprt = req->rq_xprt;
515 struct sock_xprt *transport =
516 container_of(xprt, struct sock_xprt, xprt);
517 struct xdr_buf *xdr = &req->rq_snd_buf;
521 xs_encode_stream_record_marker(&req->rq_snd_buf);
523 xs_pktdump("packet data:",
524 req->rq_svec->iov_base, req->rq_svec->iov_len);
526 status = xs_sendpages(transport->sock, NULL, 0, xdr, req->rq_bytes_sent,
528 dprintk("RPC: %s(%u) = %d\n",
529 __func__, xdr->len - req->rq_bytes_sent, status);
531 if (status == -EAGAIN && sock_writeable(transport->inet))
534 if (likely(sent > 0) || status == 0) {
535 req->rq_bytes_sent += sent;
536 req->rq_xmit_bytes_sent += sent;
537 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
538 req->rq_bytes_sent = 0;
548 status = xs_nospace(task);
551 dprintk("RPC: sendmsg returned unrecognized error %d\n",
562 * xs_udp_send_request - write an RPC request to a UDP socket
563 * @task: address of RPC task that manages the state of an RPC request
566 * 0: The request has been sent
567 * EAGAIN: The socket was blocked, please call again later to
568 * complete the request
569 * ENOTCONN: Caller needs to invoke connect logic then call again
570 * other: Some other error occurred, the request was not sent
572 static int xs_udp_send_request(struct rpc_task *task)
574 struct rpc_rqst *req = task->tk_rqstp;
575 struct rpc_xprt *xprt = req->rq_xprt;
576 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
577 struct xdr_buf *xdr = &req->rq_snd_buf;
581 xs_pktdump("packet data:",
582 req->rq_svec->iov_base,
583 req->rq_svec->iov_len);
585 if (!xprt_bound(xprt))
587 status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
588 xdr, req->rq_bytes_sent, true, &sent);
590 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
591 xdr->len - req->rq_bytes_sent, status);
593 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
594 if (status == -EPERM)
597 if (status == -EAGAIN && sock_writeable(transport->inet))
600 if (sent > 0 || status == 0) {
601 req->rq_xmit_bytes_sent += sent;
602 if (sent >= req->rq_slen)
604 /* Still some bytes left; set up for a retry later. */
612 /* Should we call xs_close() here? */
615 status = xs_nospace(task);
622 /* When the server has died, an ICMP port unreachable message
623 * prompts ECONNREFUSED. */
626 dprintk("RPC: sendmsg returned unrecognized error %d\n",
634 * xs_tcp_send_request - write an RPC request to a TCP socket
635 * @task: address of RPC task that manages the state of an RPC request
638 * 0: The request has been sent
639 * EAGAIN: The socket was blocked, please call again later to
640 * complete the request
641 * ENOTCONN: Caller needs to invoke connect logic then call again
642 * other: Some other error occurred, the request was not sent
644 * XXX: In the case of soft timeouts, should we eventually give up
645 * if sendmsg is not able to make progress?
647 static int xs_tcp_send_request(struct rpc_task *task)
649 struct rpc_rqst *req = task->tk_rqstp;
650 struct rpc_xprt *xprt = req->rq_xprt;
651 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
652 struct xdr_buf *xdr = &req->rq_snd_buf;
653 bool zerocopy = true;
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;
696 if (status == -EAGAIN && sk_stream_is_writeable(transport->inet))
702 /* Should we call xs_close() here? */
705 status = xs_nospace(task);
715 dprintk("RPC: sendmsg returned unrecognized error %d\n",
723 * xs_tcp_release_xprt - clean up after a tcp transmission
727 * This cleans up if an error causes us to abort the transmission of a request.
728 * In this case, the socket may need to be reset in order to avoid confusing
731 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
733 struct rpc_rqst *req;
735 if (task != xprt->snd_task)
739 req = task->tk_rqstp;
742 if (req->rq_bytes_sent == 0)
744 if (req->rq_bytes_sent == req->rq_snd_buf.len)
746 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
748 xprt_release_xprt(xprt, task);
751 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
753 transport->old_data_ready = sk->sk_data_ready;
754 transport->old_state_change = sk->sk_state_change;
755 transport->old_write_space = sk->sk_write_space;
756 transport->old_error_report = sk->sk_error_report;
759 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
761 sk->sk_data_ready = transport->old_data_ready;
762 sk->sk_state_change = transport->old_state_change;
763 sk->sk_write_space = transport->old_write_space;
764 sk->sk_error_report = transport->old_error_report;
767 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
769 smp_mb__before_atomic();
770 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
771 clear_bit(XPRT_CLOSING, &xprt->state);
772 smp_mb__after_atomic();
775 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
777 xs_sock_reset_connection_flags(xprt);
778 /* Mark transport as closed and wake up all pending tasks */
779 xprt_disconnect_done(xprt);
783 * xs_error_report - callback to handle TCP socket state errors
786 * Note: we don't call sock_error() since there may be a rpc_task
787 * using the socket, and so we don't want to clear sk->sk_err.
789 static void xs_error_report(struct sock *sk)
791 struct rpc_xprt *xprt;
794 read_lock_bh(&sk->sk_callback_lock);
795 if (!(xprt = xprt_from_sock(sk)))
801 /* Is this a reset event? */
802 if (sk->sk_state == TCP_CLOSE)
803 xs_sock_mark_closed(xprt);
804 dprintk("RPC: xs_error_report client %p, error=%d...\n",
806 trace_rpc_socket_error(xprt, sk->sk_socket, err);
807 xprt_wake_pending_tasks(xprt, err);
809 read_unlock_bh(&sk->sk_callback_lock);
812 static void xs_reset_transport(struct sock_xprt *transport)
814 struct socket *sock = transport->sock;
815 struct sock *sk = transport->inet;
816 struct rpc_xprt *xprt = &transport->xprt;
821 if (atomic_read(&transport->xprt.swapper))
822 sk_clear_memalloc(sk);
824 kernel_sock_shutdown(sock, SHUT_RDWR);
826 mutex_lock(&transport->recv_mutex);
827 write_lock_bh(&sk->sk_callback_lock);
828 transport->inet = NULL;
829 transport->sock = NULL;
831 sk->sk_user_data = NULL;
833 xs_restore_old_callbacks(transport, sk);
834 xprt_clear_connected(xprt);
835 write_unlock_bh(&sk->sk_callback_lock);
836 xs_sock_reset_connection_flags(xprt);
837 mutex_unlock(&transport->recv_mutex);
839 trace_rpc_socket_close(xprt, sock);
844 * xs_close - close a socket
847 * This is used when all requests are complete; ie, no DRC state remains
848 * on the server we want to save.
850 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
851 * xs_reset_transport() zeroing the socket from underneath a writer.
853 static void xs_close(struct rpc_xprt *xprt)
855 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
857 dprintk("RPC: xs_close xprt %p\n", xprt);
859 xs_reset_transport(transport);
860 xprt->reestablish_timeout = 0;
862 xprt_disconnect_done(xprt);
865 static void xs_inject_disconnect(struct rpc_xprt *xprt)
867 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
869 xprt_disconnect_done(xprt);
872 static void xs_xprt_free(struct rpc_xprt *xprt)
874 xs_free_peer_addresses(xprt);
879 * xs_destroy - prepare to shutdown a transport
880 * @xprt: doomed transport
883 static void xs_destroy(struct rpc_xprt *xprt)
885 struct sock_xprt *transport = container_of(xprt,
886 struct sock_xprt, xprt);
887 dprintk("RPC: xs_destroy xprt %p\n", xprt);
889 cancel_delayed_work_sync(&transport->connect_worker);
891 cancel_work_sync(&transport->recv_worker);
893 module_put(THIS_MODULE);
896 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
898 struct xdr_skb_reader desc = {
900 .offset = sizeof(rpc_fraghdr),
901 .count = skb->len - sizeof(rpc_fraghdr),
904 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
912 * xs_local_data_read_skb
917 * Currently this assumes we can read the whole reply in a single gulp.
919 static void xs_local_data_read_skb(struct rpc_xprt *xprt,
923 struct rpc_task *task;
924 struct rpc_rqst *rovr;
929 repsize = skb->len - sizeof(rpc_fraghdr);
931 dprintk("RPC: impossible RPC reply size %d\n", repsize);
935 /* Copy the XID from the skb... */
936 xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
940 /* Look up and lock the request corresponding to the given XID */
941 spin_lock_bh(&xprt->transport_lock);
942 rovr = xprt_lookup_rqst(xprt, *xp);
945 task = rovr->rq_task;
947 copied = rovr->rq_private_buf.buflen;
948 if (copied > repsize)
951 if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
952 dprintk("RPC: sk_buff copy failed\n");
956 xprt_complete_rqst(task, copied);
959 spin_unlock_bh(&xprt->transport_lock);
962 static void xs_local_data_receive(struct sock_xprt *transport)
968 mutex_lock(&transport->recv_mutex);
969 sk = transport->inet;
973 skb = skb_recv_datagram(sk, 0, 1, &err);
976 xs_local_data_read_skb(&transport->xprt, sk, skb);
977 skb_free_datagram(sk, skb);
980 mutex_unlock(&transport->recv_mutex);
983 static void xs_local_data_receive_workfn(struct work_struct *work)
985 struct sock_xprt *transport =
986 container_of(work, struct sock_xprt, recv_worker);
987 xs_local_data_receive(transport);
991 * xs_udp_data_read_skb - receive callback for UDP sockets
997 static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1001 struct rpc_task *task;
1002 struct rpc_rqst *rovr;
1003 int repsize, copied;
1007 repsize = skb->len - sizeof(struct udphdr);
1009 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
1013 /* Copy the XID from the skb... */
1014 xp = skb_header_pointer(skb, sizeof(struct udphdr),
1015 sizeof(_xid), &_xid);
1019 /* Look up and lock the request corresponding to the given XID */
1020 spin_lock_bh(&xprt->transport_lock);
1021 rovr = xprt_lookup_rqst(xprt, *xp);
1024 task = rovr->rq_task;
1026 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1029 /* Suck it into the iovec, verify checksum if not done by hw. */
1030 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1031 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
1035 UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
1037 xprt_adjust_cwnd(xprt, task, copied);
1038 xprt_complete_rqst(task, copied);
1041 spin_unlock_bh(&xprt->transport_lock);
1044 static void xs_udp_data_receive(struct sock_xprt *transport)
1046 struct sk_buff *skb;
1050 mutex_lock(&transport->recv_mutex);
1051 sk = transport->inet;
1055 skb = skb_recv_datagram(sk, 0, 1, &err);
1058 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1059 skb_free_datagram(sk, skb);
1062 mutex_unlock(&transport->recv_mutex);
1065 static void xs_udp_data_receive_workfn(struct work_struct *work)
1067 struct sock_xprt *transport =
1068 container_of(work, struct sock_xprt, recv_worker);
1069 xs_udp_data_receive(transport);
1073 * xs_data_ready - "data ready" callback for UDP sockets
1074 * @sk: socket with data to read
1077 static void xs_data_ready(struct sock *sk)
1079 struct rpc_xprt *xprt;
1081 read_lock_bh(&sk->sk_callback_lock);
1082 dprintk("RPC: xs_data_ready...\n");
1083 xprt = xprt_from_sock(sk);
1085 struct sock_xprt *transport = container_of(xprt,
1086 struct sock_xprt, xprt);
1087 queue_work(rpciod_workqueue, &transport->recv_worker);
1089 read_unlock_bh(&sk->sk_callback_lock);
1093 * Helper function to force a TCP close if the server is sending
1094 * junk and/or it has put us in CLOSE_WAIT
1096 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1098 xprt_force_disconnect(xprt);
1101 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1103 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1107 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1108 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1109 used = xdr_skb_read_bits(desc, p, len);
1110 transport->tcp_offset += used;
1114 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1115 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1116 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1118 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1119 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1121 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1122 transport->tcp_offset = 0;
1124 /* Sanity check of the record length */
1125 if (unlikely(transport->tcp_reclen < 8)) {
1126 dprintk("RPC: invalid TCP record fragment length\n");
1127 xs_tcp_force_close(xprt);
1130 dprintk("RPC: reading TCP record fragment of length %d\n",
1131 transport->tcp_reclen);
1134 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1136 if (transport->tcp_offset == transport->tcp_reclen) {
1137 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1138 transport->tcp_offset = 0;
1139 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1140 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1141 transport->tcp_flags |= TCP_RCV_COPY_XID;
1142 transport->tcp_copied = 0;
1147 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1152 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1153 dprintk("RPC: reading XID (%Zu bytes)\n", len);
1154 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1155 used = xdr_skb_read_bits(desc, p, len);
1156 transport->tcp_offset += used;
1159 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1160 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1161 transport->tcp_copied = 4;
1162 dprintk("RPC: reading %s XID %08x\n",
1163 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1165 ntohl(transport->tcp_xid));
1166 xs_tcp_check_fraghdr(transport);
1169 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1170 struct xdr_skb_reader *desc)
1177 * We want transport->tcp_offset to be 8 at the end of this routine
1178 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1179 * When this function is called for the first time,
1180 * transport->tcp_offset is 4 (after having already read the xid).
1182 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1183 len = sizeof(transport->tcp_calldir) - offset;
1184 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len);
1185 p = ((char *) &transport->tcp_calldir) + offset;
1186 used = xdr_skb_read_bits(desc, p, len);
1187 transport->tcp_offset += used;
1190 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1192 * We don't yet have the XDR buffer, so we will write the calldir
1193 * out after we get the buffer from the 'struct rpc_rqst'
1195 switch (ntohl(transport->tcp_calldir)) {
1197 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1198 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1199 transport->tcp_flags |= TCP_RPC_REPLY;
1202 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1203 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1204 transport->tcp_flags &= ~TCP_RPC_REPLY;
1207 dprintk("RPC: invalid request message type\n");
1208 xs_tcp_force_close(&transport->xprt);
1210 xs_tcp_check_fraghdr(transport);
1213 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1214 struct xdr_skb_reader *desc,
1215 struct rpc_rqst *req)
1217 struct sock_xprt *transport =
1218 container_of(xprt, struct sock_xprt, xprt);
1219 struct xdr_buf *rcvbuf;
1223 rcvbuf = &req->rq_private_buf;
1225 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1227 * Save the RPC direction in the XDR buffer
1229 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1230 &transport->tcp_calldir,
1231 sizeof(transport->tcp_calldir));
1232 transport->tcp_copied += sizeof(transport->tcp_calldir);
1233 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1237 if (len > transport->tcp_reclen - transport->tcp_offset) {
1238 struct xdr_skb_reader my_desc;
1240 len = transport->tcp_reclen - transport->tcp_offset;
1241 memcpy(&my_desc, desc, sizeof(my_desc));
1242 my_desc.count = len;
1243 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1244 &my_desc, xdr_skb_read_bits);
1248 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1249 desc, xdr_skb_read_bits);
1252 transport->tcp_copied += r;
1253 transport->tcp_offset += r;
1256 /* Error when copying to the receive buffer,
1257 * usually because we weren't able to allocate
1258 * additional buffer pages. All we can do now
1259 * is turn off TCP_RCV_COPY_DATA, so the request
1260 * will not receive any additional updates,
1262 * Any remaining data from this record will
1265 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1266 dprintk("RPC: XID %08x truncated request\n",
1267 ntohl(transport->tcp_xid));
1268 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1269 "tcp_offset = %u, tcp_reclen = %u\n",
1270 xprt, transport->tcp_copied,
1271 transport->tcp_offset, transport->tcp_reclen);
1275 dprintk("RPC: XID %08x read %Zd bytes\n",
1276 ntohl(transport->tcp_xid), r);
1277 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1278 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1279 transport->tcp_offset, transport->tcp_reclen);
1281 if (transport->tcp_copied == req->rq_private_buf.buflen)
1282 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1283 else if (transport->tcp_offset == transport->tcp_reclen) {
1284 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1285 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1290 * Finds the request corresponding to the RPC xid and invokes the common
1291 * tcp read code to read the data.
1293 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1294 struct xdr_skb_reader *desc)
1296 struct sock_xprt *transport =
1297 container_of(xprt, struct sock_xprt, xprt);
1298 struct rpc_rqst *req;
1300 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1302 /* Find and lock the request corresponding to this xid */
1303 spin_lock_bh(&xprt->transport_lock);
1304 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1306 dprintk("RPC: XID %08x request not found!\n",
1307 ntohl(transport->tcp_xid));
1308 spin_unlock_bh(&xprt->transport_lock);
1312 xs_tcp_read_common(xprt, desc, req);
1314 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1315 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1317 spin_unlock_bh(&xprt->transport_lock);
1321 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1323 * Obtains an rpc_rqst previously allocated and invokes the common
1324 * tcp read code to read the data. The result is placed in the callback
1326 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1327 * connection and return -1.
1329 static int xs_tcp_read_callback(struct rpc_xprt *xprt,
1330 struct xdr_skb_reader *desc)
1332 struct sock_xprt *transport =
1333 container_of(xprt, struct sock_xprt, xprt);
1334 struct rpc_rqst *req;
1336 /* Look up and lock the request corresponding to the given XID */
1337 spin_lock_bh(&xprt->transport_lock);
1338 req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
1340 spin_unlock_bh(&xprt->transport_lock);
1341 printk(KERN_WARNING "Callback slot table overflowed\n");
1342 xprt_force_disconnect(xprt);
1346 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1347 xs_tcp_read_common(xprt, desc, req);
1349 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1350 xprt_complete_bc_request(req, transport->tcp_copied);
1351 spin_unlock_bh(&xprt->transport_lock);
1356 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1357 struct xdr_skb_reader *desc)
1359 struct sock_xprt *transport =
1360 container_of(xprt, struct sock_xprt, xprt);
1362 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1363 xs_tcp_read_reply(xprt, desc) :
1364 xs_tcp_read_callback(xprt, desc);
1367 static int xs_tcp_bc_up(struct svc_serv *serv, struct net *net)
1371 ret = svc_create_xprt(serv, "tcp-bc", net, PF_INET, 0,
1372 SVC_SOCK_ANONYMOUS);
1378 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1379 struct xdr_skb_reader *desc)
1381 return xs_tcp_read_reply(xprt, desc);
1383 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1386 * Read data off the transport. This can be either an RPC_CALL or an
1387 * RPC_REPLY. Relay the processing to helper functions.
1389 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1390 struct xdr_skb_reader *desc)
1392 struct sock_xprt *transport =
1393 container_of(xprt, struct sock_xprt, xprt);
1395 if (_xs_tcp_read_data(xprt, desc) == 0)
1396 xs_tcp_check_fraghdr(transport);
1399 * The transport_lock protects the request handling.
1400 * There's no need to hold it to update the tcp_flags.
1402 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1406 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1410 len = transport->tcp_reclen - transport->tcp_offset;
1411 if (len > desc->count)
1414 desc->offset += len;
1415 transport->tcp_offset += len;
1416 dprintk("RPC: discarded %Zu bytes\n", len);
1417 xs_tcp_check_fraghdr(transport);
1420 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1422 struct rpc_xprt *xprt = rd_desc->arg.data;
1423 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1424 struct xdr_skb_reader desc = {
1430 dprintk("RPC: xs_tcp_data_recv started\n");
1432 trace_xs_tcp_data_recv(transport);
1433 /* Read in a new fragment marker if necessary */
1434 /* Can we ever really expect to get completely empty fragments? */
1435 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1436 xs_tcp_read_fraghdr(xprt, &desc);
1439 /* Read in the xid if necessary */
1440 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1441 xs_tcp_read_xid(transport, &desc);
1444 /* Read in the call/reply flag */
1445 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1446 xs_tcp_read_calldir(transport, &desc);
1449 /* Read in the request data */
1450 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1451 xs_tcp_read_data(xprt, &desc);
1454 /* Skip over any trailing bytes on short reads */
1455 xs_tcp_read_discard(transport, &desc);
1456 } while (desc.count);
1457 trace_xs_tcp_data_recv(transport);
1458 dprintk("RPC: xs_tcp_data_recv done\n");
1459 return len - desc.count;
1462 static void xs_tcp_data_receive(struct sock_xprt *transport)
1464 struct rpc_xprt *xprt = &transport->xprt;
1466 read_descriptor_t rd_desc = {
1467 .count = 2*1024*1024,
1470 unsigned long total = 0;
1473 mutex_lock(&transport->recv_mutex);
1474 sk = transport->inet;
1478 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1481 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1486 rd_desc.count = 65536;
1489 mutex_unlock(&transport->recv_mutex);
1490 trace_xs_tcp_data_ready(xprt, read, total);
1493 static void xs_tcp_data_receive_workfn(struct work_struct *work)
1495 struct sock_xprt *transport =
1496 container_of(work, struct sock_xprt, recv_worker);
1497 xs_tcp_data_receive(transport);
1501 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1502 * @sk: socket with data to read
1505 static void xs_tcp_data_ready(struct sock *sk)
1507 struct sock_xprt *transport;
1508 struct rpc_xprt *xprt;
1510 dprintk("RPC: xs_tcp_data_ready...\n");
1512 read_lock_bh(&sk->sk_callback_lock);
1513 if (!(xprt = xprt_from_sock(sk)))
1515 transport = container_of(xprt, struct sock_xprt, xprt);
1517 /* Any data means we had a useful conversation, so
1518 * the we don't need to delay the next reconnect
1520 if (xprt->reestablish_timeout)
1521 xprt->reestablish_timeout = 0;
1522 queue_work(rpciod_workqueue, &transport->recv_worker);
1525 read_unlock_bh(&sk->sk_callback_lock);
1529 * xs_tcp_state_change - callback to handle TCP socket state changes
1530 * @sk: socket whose state has changed
1533 static void xs_tcp_state_change(struct sock *sk)
1535 struct rpc_xprt *xprt;
1536 struct sock_xprt *transport;
1538 read_lock_bh(&sk->sk_callback_lock);
1539 if (!(xprt = xprt_from_sock(sk)))
1541 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1542 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1543 sk->sk_state, xprt_connected(xprt),
1544 sock_flag(sk, SOCK_DEAD),
1545 sock_flag(sk, SOCK_ZAPPED),
1548 transport = container_of(xprt, struct sock_xprt, xprt);
1549 trace_rpc_socket_state_change(xprt, sk->sk_socket);
1550 switch (sk->sk_state) {
1551 case TCP_ESTABLISHED:
1552 spin_lock(&xprt->transport_lock);
1553 if (!xprt_test_and_set_connected(xprt)) {
1555 /* Reset TCP record info */
1556 transport->tcp_offset = 0;
1557 transport->tcp_reclen = 0;
1558 transport->tcp_copied = 0;
1559 transport->tcp_flags =
1560 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1561 xprt->connect_cookie++;
1562 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1563 xprt_clear_connecting(xprt);
1565 xprt_wake_pending_tasks(xprt, -EAGAIN);
1567 spin_unlock(&xprt->transport_lock);
1570 /* The client initiated a shutdown of the socket */
1571 xprt->connect_cookie++;
1572 xprt->reestablish_timeout = 0;
1573 set_bit(XPRT_CLOSING, &xprt->state);
1574 smp_mb__before_atomic();
1575 clear_bit(XPRT_CONNECTED, &xprt->state);
1576 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1577 smp_mb__after_atomic();
1579 case TCP_CLOSE_WAIT:
1580 /* The server initiated a shutdown of the socket */
1581 xprt->connect_cookie++;
1582 clear_bit(XPRT_CONNECTED, &xprt->state);
1583 xs_tcp_force_close(xprt);
1586 * If the server closed down the connection, make sure that
1587 * we back off before reconnecting
1589 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1590 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1593 set_bit(XPRT_CLOSING, &xprt->state);
1594 smp_mb__before_atomic();
1595 clear_bit(XPRT_CONNECTED, &xprt->state);
1596 smp_mb__after_atomic();
1599 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1600 &transport->sock_state))
1601 xprt_clear_connecting(xprt);
1602 xs_sock_mark_closed(xprt);
1605 read_unlock_bh(&sk->sk_callback_lock);
1608 static void xs_write_space(struct sock *sk)
1610 struct socket_wq *wq;
1611 struct rpc_xprt *xprt;
1615 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1617 if (unlikely(!(xprt = xprt_from_sock(sk))))
1620 wq = rcu_dereference(sk->sk_wq);
1621 if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1624 xprt_write_space(xprt);
1630 * xs_udp_write_space - callback invoked when socket buffer space
1632 * @sk: socket whose state has changed
1634 * Called when more output buffer space is available for this socket.
1635 * We try not to wake our writers until they can make "significant"
1636 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1637 * with a bunch of small requests.
1639 static void xs_udp_write_space(struct sock *sk)
1641 read_lock_bh(&sk->sk_callback_lock);
1643 /* from net/core/sock.c:sock_def_write_space */
1644 if (sock_writeable(sk))
1647 read_unlock_bh(&sk->sk_callback_lock);
1651 * xs_tcp_write_space - callback invoked when socket buffer space
1653 * @sk: socket whose state has changed
1655 * Called when more output buffer space is available for this socket.
1656 * We try not to wake our writers until they can make "significant"
1657 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1658 * with a bunch of small requests.
1660 static void xs_tcp_write_space(struct sock *sk)
1662 read_lock_bh(&sk->sk_callback_lock);
1664 /* from net/core/stream.c:sk_stream_write_space */
1665 if (sk_stream_is_writeable(sk))
1668 read_unlock_bh(&sk->sk_callback_lock);
1671 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1673 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1674 struct sock *sk = transport->inet;
1676 if (transport->rcvsize) {
1677 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1678 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1680 if (transport->sndsize) {
1681 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1682 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1683 sk->sk_write_space(sk);
1688 * xs_udp_set_buffer_size - set send and receive limits
1689 * @xprt: generic transport
1690 * @sndsize: requested size of send buffer, in bytes
1691 * @rcvsize: requested size of receive buffer, in bytes
1693 * Set socket send and receive buffer size limits.
1695 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1697 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1699 transport->sndsize = 0;
1701 transport->sndsize = sndsize + 1024;
1702 transport->rcvsize = 0;
1704 transport->rcvsize = rcvsize + 1024;
1706 xs_udp_do_set_buffer_size(xprt);
1710 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1711 * @task: task that timed out
1713 * Adjust the congestion window after a retransmit timeout has occurred.
1715 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1717 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1720 static unsigned short xs_get_random_port(void)
1722 unsigned short range = xprt_max_resvport - xprt_min_resvport;
1723 unsigned short rand = (unsigned short) prandom_u32() % range;
1724 return rand + xprt_min_resvport;
1728 * xs_set_reuseaddr_port - set the socket's port and address reuse options
1731 * Note that this function has to be called on all sockets that share the
1732 * same port, and it must be called before binding.
1734 static void xs_sock_set_reuseport(struct socket *sock)
1738 kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1739 (char *)&opt, sizeof(opt));
1742 static unsigned short xs_sock_getport(struct socket *sock)
1744 struct sockaddr_storage buf;
1746 unsigned short port = 0;
1748 if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0)
1750 switch (buf.ss_family) {
1752 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1755 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1762 * xs_set_port - reset the port number in the remote endpoint address
1763 * @xprt: generic transport
1764 * @port: new port number
1767 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1769 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1771 rpc_set_port(xs_addr(xprt), port);
1772 xs_update_peer_port(xprt);
1775 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1777 if (transport->srcport == 0)
1778 transport->srcport = xs_sock_getport(sock);
1781 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1783 unsigned short port = transport->srcport;
1785 if (port == 0 && transport->xprt.resvport)
1786 port = xs_get_random_port();
1790 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1792 if (transport->srcport != 0)
1793 transport->srcport = 0;
1794 if (!transport->xprt.resvport)
1796 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1797 return xprt_max_resvport;
1800 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1802 struct sockaddr_storage myaddr;
1804 unsigned short port = xs_get_srcport(transport);
1805 unsigned short last;
1808 * If we are asking for any ephemeral port (i.e. port == 0 &&
1809 * transport->xprt.resvport == 0), don't bind. Let the local
1810 * port selection happen implicitly when the socket is used
1811 * (for example at connect time).
1813 * This ensures that we can continue to establish TCP
1814 * connections even when all local ephemeral ports are already
1815 * a part of some TCP connection. This makes no difference
1816 * for UDP sockets, but also doens't harm them.
1818 * If we're asking for any reserved port (i.e. port == 0 &&
1819 * transport->xprt.resvport == 1) xs_get_srcport above will
1820 * ensure that port is non-zero and we will bind as needed.
1825 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1827 rpc_set_port((struct sockaddr *)&myaddr, port);
1828 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1829 transport->xprt.addrlen);
1831 transport->srcport = port;
1835 port = xs_next_srcport(transport, port);
1838 } while (err == -EADDRINUSE && nloop != 2);
1840 if (myaddr.ss_family == AF_INET)
1841 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1842 &((struct sockaddr_in *)&myaddr)->sin_addr,
1843 port, err ? "failed" : "ok", err);
1845 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1846 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1847 port, err ? "failed" : "ok", err);
1852 * We don't support autobind on AF_LOCAL sockets
1854 static void xs_local_rpcbind(struct rpc_task *task)
1857 xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt));
1861 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1865 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1866 static struct lock_class_key xs_key[2];
1867 static struct lock_class_key xs_slock_key[2];
1869 static inline void xs_reclassify_socketu(struct socket *sock)
1871 struct sock *sk = sock->sk;
1873 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1874 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1877 static inline void xs_reclassify_socket4(struct socket *sock)
1879 struct sock *sk = sock->sk;
1881 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1882 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1885 static inline void xs_reclassify_socket6(struct socket *sock)
1887 struct sock *sk = sock->sk;
1889 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1890 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1893 static inline void xs_reclassify_socket(int family, struct socket *sock)
1895 WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1896 if (sock_owned_by_user(sock->sk))
1901 xs_reclassify_socketu(sock);
1904 xs_reclassify_socket4(sock);
1907 xs_reclassify_socket6(sock);
1912 static inline void xs_reclassify_socketu(struct socket *sock)
1916 static inline void xs_reclassify_socket4(struct socket *sock)
1920 static inline void xs_reclassify_socket6(struct socket *sock)
1924 static inline void xs_reclassify_socket(int family, struct socket *sock)
1929 static void xs_dummy_setup_socket(struct work_struct *work)
1933 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1934 struct sock_xprt *transport, int family, int type,
1935 int protocol, bool reuseport)
1937 struct socket *sock;
1940 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1942 dprintk("RPC: can't create %d transport socket (%d).\n",
1946 xs_reclassify_socket(family, sock);
1949 xs_sock_set_reuseport(sock);
1951 err = xs_bind(transport, sock);
1959 return ERR_PTR(err);
1962 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1963 struct socket *sock)
1965 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1968 if (!transport->inet) {
1969 struct sock *sk = sock->sk;
1971 write_lock_bh(&sk->sk_callback_lock);
1973 xs_save_old_callbacks(transport, sk);
1975 sk->sk_user_data = xprt;
1976 sk->sk_data_ready = xs_data_ready;
1977 sk->sk_write_space = xs_udp_write_space;
1978 sk->sk_error_report = xs_error_report;
1979 sk->sk_allocation = GFP_NOIO;
1981 xprt_clear_connected(xprt);
1983 /* Reset to new socket */
1984 transport->sock = sock;
1985 transport->inet = sk;
1987 write_unlock_bh(&sk->sk_callback_lock);
1990 /* Tell the socket layer to start connecting... */
1991 xprt->stat.connect_count++;
1992 xprt->stat.connect_start = jiffies;
1993 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1997 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1998 * @transport: socket transport to connect
2000 static int xs_local_setup_socket(struct sock_xprt *transport)
2002 struct rpc_xprt *xprt = &transport->xprt;
2003 struct socket *sock;
2006 status = __sock_create(xprt->xprt_net, AF_LOCAL,
2007 SOCK_STREAM, 0, &sock, 1);
2009 dprintk("RPC: can't create AF_LOCAL "
2010 "transport socket (%d).\n", -status);
2013 xs_reclassify_socketu(sock);
2015 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
2016 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2018 status = xs_local_finish_connecting(xprt, sock);
2019 trace_rpc_socket_connect(xprt, sock, status);
2022 dprintk("RPC: xprt %p connected to %s\n",
2023 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2024 xprt_set_connected(xprt);
2028 dprintk("RPC: xprt %p: socket %s does not exist\n",
2029 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2032 dprintk("RPC: xprt %p: connection refused for %s\n",
2033 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2036 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2038 xprt->address_strings[RPC_DISPLAY_ADDR]);
2042 xprt_clear_connecting(xprt);
2043 xprt_wake_pending_tasks(xprt, status);
2047 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2049 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2052 if (RPC_IS_ASYNC(task)) {
2054 * We want the AF_LOCAL connect to be resolved in the
2055 * filesystem namespace of the process making the rpc
2056 * call. Thus we connect synchronously.
2058 * If we want to support asynchronous AF_LOCAL calls,
2059 * we'll need to figure out how to pass a namespace to
2062 rpc_exit(task, -ENOTCONN);
2065 ret = xs_local_setup_socket(transport);
2066 if (ret && !RPC_IS_SOFTCONN(task))
2067 msleep_interruptible(15000);
2070 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2072 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2073 * know that we have exclusive access to the socket), to guard against
2074 * races with xs_reset_transport.
2076 static void xs_set_memalloc(struct rpc_xprt *xprt)
2078 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2082 * If there's no sock, then we have nothing to set. The
2083 * reconnecting process will get it for us.
2085 if (!transport->inet)
2087 if (atomic_read(&xprt->swapper))
2088 sk_set_memalloc(transport->inet);
2092 * xs_enable_swap - Tag this transport as being used for swap.
2093 * @xprt: transport to tag
2095 * Take a reference to this transport on behalf of the rpc_clnt, and
2096 * optionally mark it for swapping if it wasn't already.
2099 xs_enable_swap(struct rpc_xprt *xprt)
2101 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2103 if (atomic_inc_return(&xprt->swapper) != 1)
2105 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2106 return -ERESTARTSYS;
2108 sk_set_memalloc(xs->inet);
2109 xprt_release_xprt(xprt, NULL);
2114 * xs_disable_swap - Untag this transport as being used for swap.
2115 * @xprt: transport to tag
2117 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2118 * swapper refcount goes to 0, untag the socket as a memalloc socket.
2121 xs_disable_swap(struct rpc_xprt *xprt)
2123 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2125 if (!atomic_dec_and_test(&xprt->swapper))
2127 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2130 sk_clear_memalloc(xs->inet);
2131 xprt_release_xprt(xprt, NULL);
2134 static void xs_set_memalloc(struct rpc_xprt *xprt)
2139 xs_enable_swap(struct rpc_xprt *xprt)
2145 xs_disable_swap(struct rpc_xprt *xprt)
2150 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2152 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2154 if (!transport->inet) {
2155 struct sock *sk = sock->sk;
2157 write_lock_bh(&sk->sk_callback_lock);
2159 xs_save_old_callbacks(transport, sk);
2161 sk->sk_user_data = xprt;
2162 sk->sk_data_ready = xs_data_ready;
2163 sk->sk_write_space = xs_udp_write_space;
2164 sk->sk_allocation = GFP_NOIO;
2166 xprt_set_connected(xprt);
2168 /* Reset to new socket */
2169 transport->sock = sock;
2170 transport->inet = sk;
2172 xs_set_memalloc(xprt);
2174 write_unlock_bh(&sk->sk_callback_lock);
2176 xs_udp_do_set_buffer_size(xprt);
2179 static void xs_udp_setup_socket(struct work_struct *work)
2181 struct sock_xprt *transport =
2182 container_of(work, struct sock_xprt, connect_worker.work);
2183 struct rpc_xprt *xprt = &transport->xprt;
2184 struct socket *sock = transport->sock;
2187 sock = xs_create_sock(xprt, transport,
2188 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2189 IPPROTO_UDP, false);
2193 dprintk("RPC: worker connecting xprt %p via %s to "
2194 "%s (port %s)\n", xprt,
2195 xprt->address_strings[RPC_DISPLAY_PROTO],
2196 xprt->address_strings[RPC_DISPLAY_ADDR],
2197 xprt->address_strings[RPC_DISPLAY_PORT]);
2199 xs_udp_finish_connecting(xprt, sock);
2200 trace_rpc_socket_connect(xprt, sock, 0);
2203 xprt_unlock_connect(xprt, transport);
2204 xprt_clear_connecting(xprt);
2205 xprt_wake_pending_tasks(xprt, status);
2209 * xs_tcp_shutdown - gracefully shut down a TCP socket
2212 * Initiates a graceful shutdown of the TCP socket by calling the
2213 * equivalent of shutdown(SHUT_RDWR);
2215 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2217 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2218 struct socket *sock = transport->sock;
2222 if (xprt_connected(xprt)) {
2223 kernel_sock_shutdown(sock, SHUT_RDWR);
2224 trace_rpc_socket_shutdown(xprt, sock);
2226 xs_reset_transport(transport);
2229 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2231 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2232 int ret = -ENOTCONN;
2234 if (!transport->inet) {
2235 struct sock *sk = sock->sk;
2236 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2237 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2238 unsigned int opt_on = 1;
2241 /* TCP Keepalive options */
2242 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2243 (char *)&opt_on, sizeof(opt_on));
2244 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2245 (char *)&keepidle, sizeof(keepidle));
2246 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2247 (char *)&keepidle, sizeof(keepidle));
2248 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2249 (char *)&keepcnt, sizeof(keepcnt));
2251 /* TCP user timeout (see RFC5482) */
2252 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2253 (xprt->timeout->to_retries + 1);
2254 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2255 (char *)&timeo, sizeof(timeo));
2257 write_lock_bh(&sk->sk_callback_lock);
2259 xs_save_old_callbacks(transport, sk);
2261 sk->sk_user_data = xprt;
2262 sk->sk_data_ready = xs_tcp_data_ready;
2263 sk->sk_state_change = xs_tcp_state_change;
2264 sk->sk_write_space = xs_tcp_write_space;
2265 sk->sk_error_report = xs_error_report;
2266 sk->sk_allocation = GFP_NOIO;
2268 /* socket options */
2269 sock_reset_flag(sk, SOCK_LINGER);
2270 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2272 xprt_clear_connected(xprt);
2274 /* Reset to new socket */
2275 transport->sock = sock;
2276 transport->inet = sk;
2278 write_unlock_bh(&sk->sk_callback_lock);
2281 if (!xprt_bound(xprt))
2284 xs_set_memalloc(xprt);
2286 /* Tell the socket layer to start connecting... */
2287 xprt->stat.connect_count++;
2288 xprt->stat.connect_start = jiffies;
2289 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2290 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2293 xs_set_srcport(transport, sock);
2296 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2297 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2299 case -EADDRNOTAVAIL:
2300 /* Source port number is unavailable. Try a new one! */
2301 transport->srcport = 0;
2308 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2310 * Invoked by a work queue tasklet.
2312 static void xs_tcp_setup_socket(struct work_struct *work)
2314 struct sock_xprt *transport =
2315 container_of(work, struct sock_xprt, connect_worker.work);
2316 struct socket *sock = transport->sock;
2317 struct rpc_xprt *xprt = &transport->xprt;
2321 sock = xs_create_sock(xprt, transport,
2322 xs_addr(xprt)->sa_family, SOCK_STREAM,
2325 status = PTR_ERR(sock);
2330 dprintk("RPC: worker connecting xprt %p via %s to "
2331 "%s (port %s)\n", xprt,
2332 xprt->address_strings[RPC_DISPLAY_PROTO],
2333 xprt->address_strings[RPC_DISPLAY_ADDR],
2334 xprt->address_strings[RPC_DISPLAY_PORT]);
2336 status = xs_tcp_finish_connecting(xprt, sock);
2337 trace_rpc_socket_connect(xprt, sock, status);
2338 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2339 xprt, -status, xprt_connected(xprt),
2340 sock->sk->sk_state);
2343 printk("%s: connect returned unhandled error %d\n",
2345 case -EADDRNOTAVAIL:
2346 /* We're probably in TIME_WAIT. Get rid of existing socket,
2349 xs_tcp_force_close(xprt);
2354 xprt_unlock_connect(xprt, transport);
2357 /* Happens, for instance, if the user specified a link
2358 * local IPv6 address without a scope-id.
2365 /* retry with existing socket, after a delay */
2366 xs_tcp_force_close(xprt);
2371 xprt_unlock_connect(xprt, transport);
2372 xprt_clear_connecting(xprt);
2373 xprt_wake_pending_tasks(xprt, status);
2377 * xs_connect - connect a socket to a remote endpoint
2378 * @xprt: pointer to transport structure
2379 * @task: address of RPC task that manages state of connect request
2381 * TCP: If the remote end dropped the connection, delay reconnecting.
2383 * UDP socket connects are synchronous, but we use a work queue anyway
2384 * to guarantee that even unprivileged user processes can set up a
2385 * socket on a privileged port.
2387 * If a UDP socket connect fails, the delay behavior here prevents
2388 * retry floods (hard mounts).
2390 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2392 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2394 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2396 if (transport->sock != NULL) {
2397 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2399 xprt, xprt->reestablish_timeout / HZ);
2401 /* Start by resetting any existing state */
2402 xs_reset_transport(transport);
2404 queue_delayed_work(rpciod_workqueue,
2405 &transport->connect_worker,
2406 xprt->reestablish_timeout);
2407 xprt->reestablish_timeout <<= 1;
2408 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2409 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2410 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2411 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2413 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2414 queue_delayed_work(rpciod_workqueue,
2415 &transport->connect_worker, 0);
2420 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2421 * @xprt: rpc_xprt struct containing statistics
2425 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2429 if (xprt_connected(xprt))
2430 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2432 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2433 "%llu %llu %lu %llu %llu\n",
2434 xprt->stat.bind_count,
2435 xprt->stat.connect_count,
2436 xprt->stat.connect_time,
2440 xprt->stat.bad_xids,
2443 xprt->stat.max_slots,
2444 xprt->stat.sending_u,
2445 xprt->stat.pending_u);
2449 * xs_udp_print_stats - display UDP socket-specifc stats
2450 * @xprt: rpc_xprt struct containing statistics
2454 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2456 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2458 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2461 xprt->stat.bind_count,
2464 xprt->stat.bad_xids,
2467 xprt->stat.max_slots,
2468 xprt->stat.sending_u,
2469 xprt->stat.pending_u);
2473 * xs_tcp_print_stats - display TCP socket-specifc stats
2474 * @xprt: rpc_xprt struct containing statistics
2478 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2480 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2483 if (xprt_connected(xprt))
2484 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2486 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2487 "%llu %llu %lu %llu %llu\n",
2489 xprt->stat.bind_count,
2490 xprt->stat.connect_count,
2491 xprt->stat.connect_time,
2495 xprt->stat.bad_xids,
2498 xprt->stat.max_slots,
2499 xprt->stat.sending_u,
2500 xprt->stat.pending_u);
2504 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2505 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2506 * to use the server side send routines.
2508 static void *bc_malloc(struct rpc_task *task, size_t size)
2511 struct rpc_buffer *buf;
2513 WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2514 if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2517 page = alloc_page(GFP_KERNEL);
2521 buf = page_address(page);
2522 buf->len = PAGE_SIZE;
2528 * Free the space allocated in the bc_alloc routine
2530 static void bc_free(void *buffer)
2532 struct rpc_buffer *buf;
2537 buf = container_of(buffer, struct rpc_buffer, data);
2538 free_page((unsigned long)buf);
2542 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2543 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2545 static int bc_sendto(struct rpc_rqst *req)
2548 struct xdr_buf *xbufp = &req->rq_snd_buf;
2549 struct rpc_xprt *xprt = req->rq_xprt;
2550 struct sock_xprt *transport =
2551 container_of(xprt, struct sock_xprt, xprt);
2552 struct socket *sock = transport->sock;
2553 unsigned long headoff;
2554 unsigned long tailoff;
2556 xs_encode_stream_record_marker(xbufp);
2558 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2559 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2560 len = svc_send_common(sock, xbufp,
2561 virt_to_page(xbufp->head[0].iov_base), headoff,
2562 xbufp->tail[0].iov_base, tailoff);
2564 if (len != xbufp->len) {
2565 printk(KERN_NOTICE "Error sending entire callback!\n");
2573 * The send routine. Borrows from svc_send
2575 static int bc_send_request(struct rpc_task *task)
2577 struct rpc_rqst *req = task->tk_rqstp;
2578 struct svc_xprt *xprt;
2581 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2583 * Get the server socket associated with this callback xprt
2585 xprt = req->rq_xprt->bc_xprt;
2588 * Grab the mutex to serialize data as the connection is shared
2589 * with the fore channel
2591 if (!mutex_trylock(&xprt->xpt_mutex)) {
2592 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2593 if (!mutex_trylock(&xprt->xpt_mutex))
2595 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2597 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2600 len = bc_sendto(req);
2601 mutex_unlock(&xprt->xpt_mutex);
2610 * The close routine. Since this is client initiated, we do nothing
2613 static void bc_close(struct rpc_xprt *xprt)
2618 * The xprt destroy routine. Again, because this connection is client
2619 * initiated, we do nothing
2622 static void bc_destroy(struct rpc_xprt *xprt)
2624 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2627 module_put(THIS_MODULE);
2630 static struct rpc_xprt_ops xs_local_ops = {
2631 .reserve_xprt = xprt_reserve_xprt,
2632 .release_xprt = xs_tcp_release_xprt,
2633 .alloc_slot = xprt_alloc_slot,
2634 .rpcbind = xs_local_rpcbind,
2635 .set_port = xs_local_set_port,
2636 .connect = xs_local_connect,
2637 .buf_alloc = rpc_malloc,
2638 .buf_free = rpc_free,
2639 .send_request = xs_local_send_request,
2640 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2642 .destroy = xs_destroy,
2643 .print_stats = xs_local_print_stats,
2644 .enable_swap = xs_enable_swap,
2645 .disable_swap = xs_disable_swap,
2648 static struct rpc_xprt_ops xs_udp_ops = {
2649 .set_buffer_size = xs_udp_set_buffer_size,
2650 .reserve_xprt = xprt_reserve_xprt_cong,
2651 .release_xprt = xprt_release_xprt_cong,
2652 .alloc_slot = xprt_alloc_slot,
2653 .rpcbind = rpcb_getport_async,
2654 .set_port = xs_set_port,
2655 .connect = xs_connect,
2656 .buf_alloc = rpc_malloc,
2657 .buf_free = rpc_free,
2658 .send_request = xs_udp_send_request,
2659 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
2660 .timer = xs_udp_timer,
2661 .release_request = xprt_release_rqst_cong,
2663 .destroy = xs_destroy,
2664 .print_stats = xs_udp_print_stats,
2665 .enable_swap = xs_enable_swap,
2666 .disable_swap = xs_disable_swap,
2667 .inject_disconnect = xs_inject_disconnect,
2670 static struct rpc_xprt_ops xs_tcp_ops = {
2671 .reserve_xprt = xprt_reserve_xprt,
2672 .release_xprt = xs_tcp_release_xprt,
2673 .alloc_slot = xprt_lock_and_alloc_slot,
2674 .rpcbind = rpcb_getport_async,
2675 .set_port = xs_set_port,
2676 .connect = xs_connect,
2677 .buf_alloc = rpc_malloc,
2678 .buf_free = rpc_free,
2679 .send_request = xs_tcp_send_request,
2680 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2681 .close = xs_tcp_shutdown,
2682 .destroy = xs_destroy,
2683 .print_stats = xs_tcp_print_stats,
2684 .enable_swap = xs_enable_swap,
2685 .disable_swap = xs_disable_swap,
2686 .inject_disconnect = xs_inject_disconnect,
2687 #ifdef CONFIG_SUNRPC_BACKCHANNEL
2688 .bc_setup = xprt_setup_bc,
2689 .bc_up = xs_tcp_bc_up,
2690 .bc_free_rqst = xprt_free_bc_rqst,
2691 .bc_destroy = xprt_destroy_bc,
2696 * The rpc_xprt_ops for the server backchannel
2699 static struct rpc_xprt_ops bc_tcp_ops = {
2700 .reserve_xprt = xprt_reserve_xprt,
2701 .release_xprt = xprt_release_xprt,
2702 .alloc_slot = xprt_alloc_slot,
2703 .buf_alloc = bc_malloc,
2704 .buf_free = bc_free,
2705 .send_request = bc_send_request,
2706 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2708 .destroy = bc_destroy,
2709 .print_stats = xs_tcp_print_stats,
2710 .enable_swap = xs_enable_swap,
2711 .disable_swap = xs_disable_swap,
2712 .inject_disconnect = xs_inject_disconnect,
2715 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2717 static const struct sockaddr_in sin = {
2718 .sin_family = AF_INET,
2719 .sin_addr.s_addr = htonl(INADDR_ANY),
2721 static const struct sockaddr_in6 sin6 = {
2722 .sin6_family = AF_INET6,
2723 .sin6_addr = IN6ADDR_ANY_INIT,
2730 memcpy(sap, &sin, sizeof(sin));
2733 memcpy(sap, &sin6, sizeof(sin6));
2736 dprintk("RPC: %s: Bad address family\n", __func__);
2737 return -EAFNOSUPPORT;
2742 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2743 unsigned int slot_table_size,
2744 unsigned int max_slot_table_size)
2746 struct rpc_xprt *xprt;
2747 struct sock_xprt *new;
2749 if (args->addrlen > sizeof(xprt->addr)) {
2750 dprintk("RPC: xs_setup_xprt: address too large\n");
2751 return ERR_PTR(-EBADF);
2754 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2755 max_slot_table_size);
2757 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2759 return ERR_PTR(-ENOMEM);
2762 new = container_of(xprt, struct sock_xprt, xprt);
2763 mutex_init(&new->recv_mutex);
2764 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2765 xprt->addrlen = args->addrlen;
2767 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2770 err = xs_init_anyaddr(args->dstaddr->sa_family,
2771 (struct sockaddr *)&new->srcaddr);
2774 return ERR_PTR(err);
2781 static const struct rpc_timeout xs_local_default_timeout = {
2782 .to_initval = 10 * HZ,
2783 .to_maxval = 10 * HZ,
2788 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2789 * @args: rpc transport creation arguments
2791 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2793 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2795 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2796 struct sock_xprt *transport;
2797 struct rpc_xprt *xprt;
2798 struct rpc_xprt *ret;
2800 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2801 xprt_max_tcp_slot_table_entries);
2804 transport = container_of(xprt, struct sock_xprt, xprt);
2807 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2808 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2810 xprt->bind_timeout = XS_BIND_TO;
2811 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2812 xprt->idle_timeout = XS_IDLE_DISC_TO;
2814 xprt->ops = &xs_local_ops;
2815 xprt->timeout = &xs_local_default_timeout;
2817 INIT_WORK(&transport->recv_worker, xs_local_data_receive_workfn);
2818 INIT_DELAYED_WORK(&transport->connect_worker,
2819 xs_dummy_setup_socket);
2821 switch (sun->sun_family) {
2823 if (sun->sun_path[0] != '/') {
2824 dprintk("RPC: bad AF_LOCAL address: %s\n",
2826 ret = ERR_PTR(-EINVAL);
2829 xprt_set_bound(xprt);
2830 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2831 ret = ERR_PTR(xs_local_setup_socket(transport));
2836 ret = ERR_PTR(-EAFNOSUPPORT);
2840 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2841 xprt->address_strings[RPC_DISPLAY_ADDR]);
2843 if (try_module_get(THIS_MODULE))
2845 ret = ERR_PTR(-EINVAL);
2851 static const struct rpc_timeout xs_udp_default_timeout = {
2852 .to_initval = 5 * HZ,
2853 .to_maxval = 30 * HZ,
2854 .to_increment = 5 * HZ,
2859 * xs_setup_udp - Set up transport to use a UDP socket
2860 * @args: rpc transport creation arguments
2863 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2865 struct sockaddr *addr = args->dstaddr;
2866 struct rpc_xprt *xprt;
2867 struct sock_xprt *transport;
2868 struct rpc_xprt *ret;
2870 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2871 xprt_udp_slot_table_entries);
2874 transport = container_of(xprt, struct sock_xprt, xprt);
2876 xprt->prot = IPPROTO_UDP;
2878 /* XXX: header size can vary due to auth type, IPv6, etc. */
2879 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2881 xprt->bind_timeout = XS_BIND_TO;
2882 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2883 xprt->idle_timeout = XS_IDLE_DISC_TO;
2885 xprt->ops = &xs_udp_ops;
2887 xprt->timeout = &xs_udp_default_timeout;
2889 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
2890 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2892 switch (addr->sa_family) {
2894 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2895 xprt_set_bound(xprt);
2897 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2900 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2901 xprt_set_bound(xprt);
2903 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2906 ret = ERR_PTR(-EAFNOSUPPORT);
2910 if (xprt_bound(xprt))
2911 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2912 xprt->address_strings[RPC_DISPLAY_ADDR],
2913 xprt->address_strings[RPC_DISPLAY_PORT],
2914 xprt->address_strings[RPC_DISPLAY_PROTO]);
2916 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2917 xprt->address_strings[RPC_DISPLAY_ADDR],
2918 xprt->address_strings[RPC_DISPLAY_PROTO]);
2920 if (try_module_get(THIS_MODULE))
2922 ret = ERR_PTR(-EINVAL);
2928 static const struct rpc_timeout xs_tcp_default_timeout = {
2929 .to_initval = 60 * HZ,
2930 .to_maxval = 60 * HZ,
2935 * xs_setup_tcp - Set up transport to use a TCP socket
2936 * @args: rpc transport creation arguments
2939 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2941 struct sockaddr *addr = args->dstaddr;
2942 struct rpc_xprt *xprt;
2943 struct sock_xprt *transport;
2944 struct rpc_xprt *ret;
2945 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2947 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2948 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2950 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2951 max_slot_table_size);
2954 transport = container_of(xprt, struct sock_xprt, xprt);
2956 xprt->prot = IPPROTO_TCP;
2957 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2958 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2960 xprt->bind_timeout = XS_BIND_TO;
2961 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2962 xprt->idle_timeout = XS_IDLE_DISC_TO;
2964 xprt->ops = &xs_tcp_ops;
2965 xprt->timeout = &xs_tcp_default_timeout;
2967 INIT_WORK(&transport->recv_worker, xs_tcp_data_receive_workfn);
2968 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
2970 switch (addr->sa_family) {
2972 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2973 xprt_set_bound(xprt);
2975 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2978 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2979 xprt_set_bound(xprt);
2981 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2984 ret = ERR_PTR(-EAFNOSUPPORT);
2988 if (xprt_bound(xprt))
2989 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2990 xprt->address_strings[RPC_DISPLAY_ADDR],
2991 xprt->address_strings[RPC_DISPLAY_PORT],
2992 xprt->address_strings[RPC_DISPLAY_PROTO]);
2994 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2995 xprt->address_strings[RPC_DISPLAY_ADDR],
2996 xprt->address_strings[RPC_DISPLAY_PROTO]);
2998 if (try_module_get(THIS_MODULE))
3000 ret = ERR_PTR(-EINVAL);
3007 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3008 * @args: rpc transport creation arguments
3011 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3013 struct sockaddr *addr = args->dstaddr;
3014 struct rpc_xprt *xprt;
3015 struct sock_xprt *transport;
3016 struct svc_sock *bc_sock;
3017 struct rpc_xprt *ret;
3019 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3020 xprt_tcp_slot_table_entries);
3023 transport = container_of(xprt, struct sock_xprt, xprt);
3025 xprt->prot = IPPROTO_TCP;
3026 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3027 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3028 xprt->timeout = &xs_tcp_default_timeout;
3031 xprt_set_bound(xprt);
3032 xprt->bind_timeout = 0;
3033 xprt->reestablish_timeout = 0;
3034 xprt->idle_timeout = 0;
3036 xprt->ops = &bc_tcp_ops;
3038 switch (addr->sa_family) {
3040 xs_format_peer_addresses(xprt, "tcp",
3044 xs_format_peer_addresses(xprt, "tcp",
3045 RPCBIND_NETID_TCP6);
3048 ret = ERR_PTR(-EAFNOSUPPORT);
3052 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3053 xprt->address_strings[RPC_DISPLAY_ADDR],
3054 xprt->address_strings[RPC_DISPLAY_PORT],
3055 xprt->address_strings[RPC_DISPLAY_PROTO]);
3058 * Once we've associated a backchannel xprt with a connection,
3059 * we want to keep it around as long as the connection lasts,
3060 * in case we need to start using it for a backchannel again;
3061 * this reference won't be dropped until bc_xprt is destroyed.
3064 args->bc_xprt->xpt_bc_xprt = xprt;
3065 xprt->bc_xprt = args->bc_xprt;
3066 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3067 transport->sock = bc_sock->sk_sock;
3068 transport->inet = bc_sock->sk_sk;
3071 * Since we don't want connections for the backchannel, we set
3072 * the xprt status to connected
3074 xprt_set_connected(xprt);
3076 if (try_module_get(THIS_MODULE))
3079 args->bc_xprt->xpt_bc_xprt = NULL;
3081 ret = ERR_PTR(-EINVAL);
3087 static struct xprt_class xs_local_transport = {
3088 .list = LIST_HEAD_INIT(xs_local_transport.list),
3089 .name = "named UNIX socket",
3090 .owner = THIS_MODULE,
3091 .ident = XPRT_TRANSPORT_LOCAL,
3092 .setup = xs_setup_local,
3095 static struct xprt_class xs_udp_transport = {
3096 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3098 .owner = THIS_MODULE,
3099 .ident = XPRT_TRANSPORT_UDP,
3100 .setup = xs_setup_udp,
3103 static struct xprt_class xs_tcp_transport = {
3104 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3106 .owner = THIS_MODULE,
3107 .ident = XPRT_TRANSPORT_TCP,
3108 .setup = xs_setup_tcp,
3111 static struct xprt_class xs_bc_tcp_transport = {
3112 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3113 .name = "tcp NFSv4.1 backchannel",
3114 .owner = THIS_MODULE,
3115 .ident = XPRT_TRANSPORT_BC_TCP,
3116 .setup = xs_setup_bc_tcp,
3120 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3123 int init_socket_xprt(void)
3125 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3126 if (!sunrpc_table_header)
3127 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3130 xprt_register_transport(&xs_local_transport);
3131 xprt_register_transport(&xs_udp_transport);
3132 xprt_register_transport(&xs_tcp_transport);
3133 xprt_register_transport(&xs_bc_tcp_transport);
3139 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3142 void cleanup_socket_xprt(void)
3144 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3145 if (sunrpc_table_header) {
3146 unregister_sysctl_table(sunrpc_table_header);
3147 sunrpc_table_header = NULL;
3151 xprt_unregister_transport(&xs_local_transport);
3152 xprt_unregister_transport(&xs_udp_transport);
3153 xprt_unregister_transport(&xs_tcp_transport);
3154 xprt_unregister_transport(&xs_bc_tcp_transport);
3157 static int param_set_uint_minmax(const char *val,
3158 const struct kernel_param *kp,
3159 unsigned int min, unsigned int max)
3166 ret = kstrtouint(val, 0, &num);
3167 if (ret == -EINVAL || num < min || num > max)
3169 *((unsigned int *)kp->arg) = num;
3173 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3175 return param_set_uint_minmax(val, kp,
3180 static const struct kernel_param_ops param_ops_portnr = {
3181 .set = param_set_portnr,
3182 .get = param_get_uint,
3185 #define param_check_portnr(name, p) \
3186 __param_check(name, p, unsigned int);
3188 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3189 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3191 static int param_set_slot_table_size(const char *val,
3192 const struct kernel_param *kp)
3194 return param_set_uint_minmax(val, kp,
3196 RPC_MAX_SLOT_TABLE);
3199 static const struct kernel_param_ops param_ops_slot_table_size = {
3200 .set = param_set_slot_table_size,
3201 .get = param_get_uint,
3204 #define param_check_slot_table_size(name, p) \
3205 __param_check(name, p, unsigned int);
3207 static int param_set_max_slot_table_size(const char *val,
3208 const struct kernel_param *kp)
3210 return param_set_uint_minmax(val, kp,
3212 RPC_MAX_SLOT_TABLE_LIMIT);
3215 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3216 .set = param_set_max_slot_table_size,
3217 .get = param_get_uint,
3220 #define param_check_max_slot_table_size(name, p) \
3221 __param_check(name, p, unsigned int);
3223 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3224 slot_table_size, 0644);
3225 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3226 max_slot_table_size, 0644);
3227 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3228 slot_table_size, 0644);