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;
363 if (remainder != 0 || more)
365 err = do_sendpage(sock, *ppage, base, len, flags);
366 if (remainder == 0 || err != len)
380 * xs_sendpages - write pages directly to a socket
381 * @sock: socket to send on
382 * @addr: UDP only -- address of destination
383 * @addrlen: UDP only -- length of destination address
384 * @xdr: buffer containing this request
385 * @base: starting position in the buffer
386 * @zerocopy: true if it is safe to use sendpage()
387 * @sent_p: return the total number of bytes successfully queued for sending
390 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p)
392 unsigned int remainder = xdr->len - base;
399 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
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--;
443 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
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 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
470 * Notify TCP that we're limited by the application
473 set_bit(SOCK_NOSPACE, &transport->sock->flags);
474 sk->sk_write_pending++;
475 /* ...and wait for more buffer space */
476 xprt_wait_for_buffer_space(task, xs_nospace_callback);
479 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
483 spin_unlock_bh(&xprt->transport_lock);
485 /* Race breaker in case memory is freed before above code is called */
486 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);
618 dprintk("RPC: sendmsg returned unrecognized error %d\n",
625 /* When the server has died, an ICMP port unreachable message
626 * prompts ECONNREFUSED. */
627 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
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);
708 dprintk("RPC: sendmsg returned unrecognized error %d\n",
716 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
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 write_lock_bh(&sk->sk_callback_lock);
827 transport->inet = NULL;
828 transport->sock = NULL;
830 sk->sk_user_data = NULL;
832 xs_restore_old_callbacks(transport, sk);
833 xprt_clear_connected(xprt);
834 write_unlock_bh(&sk->sk_callback_lock);
835 xs_sock_reset_connection_flags(xprt);
837 trace_rpc_socket_close(xprt, sock);
842 * xs_close - close a socket
845 * This is used when all requests are complete; ie, no DRC state remains
846 * on the server we want to save.
848 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
849 * xs_reset_transport() zeroing the socket from underneath a writer.
851 static void xs_close(struct rpc_xprt *xprt)
853 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
855 dprintk("RPC: xs_close xprt %p\n", xprt);
857 xs_reset_transport(transport);
858 xprt->reestablish_timeout = 0;
860 xprt_disconnect_done(xprt);
863 static void xs_inject_disconnect(struct rpc_xprt *xprt)
865 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
867 xprt_disconnect_done(xprt);
870 static void xs_xprt_free(struct rpc_xprt *xprt)
872 xs_free_peer_addresses(xprt);
877 * xs_destroy - prepare to shutdown a transport
878 * @xprt: doomed transport
881 static void xs_destroy(struct rpc_xprt *xprt)
883 struct sock_xprt *transport = container_of(xprt,
884 struct sock_xprt, xprt);
885 dprintk("RPC: xs_destroy xprt %p\n", xprt);
887 cancel_delayed_work_sync(&transport->connect_worker);
890 module_put(THIS_MODULE);
893 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
895 struct xdr_skb_reader desc = {
897 .offset = sizeof(rpc_fraghdr),
898 .count = skb->len - sizeof(rpc_fraghdr),
901 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
909 * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets
910 * @sk: socket with data to read
912 * Currently this assumes we can read the whole reply in a single gulp.
914 static void xs_local_data_ready(struct sock *sk)
916 struct rpc_task *task;
917 struct rpc_xprt *xprt;
918 struct rpc_rqst *rovr;
920 int err, repsize, copied;
924 read_lock_bh(&sk->sk_callback_lock);
925 dprintk("RPC: %s...\n", __func__);
926 xprt = xprt_from_sock(sk);
930 skb = skb_recv_datagram(sk, 0, 1, &err);
934 repsize = skb->len - sizeof(rpc_fraghdr);
936 dprintk("RPC: impossible RPC reply size %d\n", repsize);
940 /* Copy the XID from the skb... */
941 xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
945 /* Look up and lock the request corresponding to the given XID */
946 spin_lock(&xprt->transport_lock);
947 rovr = xprt_lookup_rqst(xprt, *xp);
950 task = rovr->rq_task;
952 copied = rovr->rq_private_buf.buflen;
953 if (copied > repsize)
956 if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
957 dprintk("RPC: sk_buff copy failed\n");
961 xprt_complete_rqst(task, copied);
964 spin_unlock(&xprt->transport_lock);
966 skb_free_datagram(sk, skb);
968 read_unlock_bh(&sk->sk_callback_lock);
972 * xs_udp_data_ready - "data ready" callback for UDP sockets
973 * @sk: socket with data to read
976 static void xs_udp_data_ready(struct sock *sk)
978 struct rpc_task *task;
979 struct rpc_xprt *xprt;
980 struct rpc_rqst *rovr;
982 int err, repsize, copied;
986 read_lock_bh(&sk->sk_callback_lock);
987 dprintk("RPC: xs_udp_data_ready...\n");
988 if (!(xprt = xprt_from_sock(sk)))
991 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
994 repsize = skb->len - sizeof(struct udphdr);
996 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
1000 /* Copy the XID from the skb... */
1001 xp = skb_header_pointer(skb, sizeof(struct udphdr),
1002 sizeof(_xid), &_xid);
1006 /* Look up and lock the request corresponding to the given XID */
1007 spin_lock(&xprt->transport_lock);
1008 rovr = xprt_lookup_rqst(xprt, *xp);
1011 task = rovr->rq_task;
1013 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1016 /* Suck it into the iovec, verify checksum if not done by hw. */
1017 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1018 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
1022 UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
1024 xprt_adjust_cwnd(xprt, task, copied);
1025 xprt_complete_rqst(task, copied);
1028 spin_unlock(&xprt->transport_lock);
1030 skb_free_datagram(sk, skb);
1032 read_unlock_bh(&sk->sk_callback_lock);
1036 * Helper function to force a TCP close if the server is sending
1037 * junk and/or it has put us in CLOSE_WAIT
1039 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1041 xprt_force_disconnect(xprt);
1044 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1046 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1050 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1051 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1052 used = xdr_skb_read_bits(desc, p, len);
1053 transport->tcp_offset += used;
1057 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1058 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1059 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1061 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1062 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1064 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1065 transport->tcp_offset = 0;
1067 /* Sanity check of the record length */
1068 if (unlikely(transport->tcp_reclen < 8)) {
1069 dprintk("RPC: invalid TCP record fragment length\n");
1070 xs_tcp_force_close(xprt);
1073 dprintk("RPC: reading TCP record fragment of length %d\n",
1074 transport->tcp_reclen);
1077 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1079 if (transport->tcp_offset == transport->tcp_reclen) {
1080 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1081 transport->tcp_offset = 0;
1082 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1083 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1084 transport->tcp_flags |= TCP_RCV_COPY_XID;
1085 transport->tcp_copied = 0;
1090 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1095 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1096 dprintk("RPC: reading XID (%Zu bytes)\n", len);
1097 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1098 used = xdr_skb_read_bits(desc, p, len);
1099 transport->tcp_offset += used;
1102 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1103 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1104 transport->tcp_copied = 4;
1105 dprintk("RPC: reading %s XID %08x\n",
1106 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1108 ntohl(transport->tcp_xid));
1109 xs_tcp_check_fraghdr(transport);
1112 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1113 struct xdr_skb_reader *desc)
1120 * We want transport->tcp_offset to be 8 at the end of this routine
1121 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1122 * When this function is called for the first time,
1123 * transport->tcp_offset is 4 (after having already read the xid).
1125 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1126 len = sizeof(transport->tcp_calldir) - offset;
1127 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len);
1128 p = ((char *) &transport->tcp_calldir) + offset;
1129 used = xdr_skb_read_bits(desc, p, len);
1130 transport->tcp_offset += used;
1133 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1135 * We don't yet have the XDR buffer, so we will write the calldir
1136 * out after we get the buffer from the 'struct rpc_rqst'
1138 switch (ntohl(transport->tcp_calldir)) {
1140 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1141 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1142 transport->tcp_flags |= TCP_RPC_REPLY;
1145 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1146 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1147 transport->tcp_flags &= ~TCP_RPC_REPLY;
1150 dprintk("RPC: invalid request message type\n");
1151 xs_tcp_force_close(&transport->xprt);
1153 xs_tcp_check_fraghdr(transport);
1156 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1157 struct xdr_skb_reader *desc,
1158 struct rpc_rqst *req)
1160 struct sock_xprt *transport =
1161 container_of(xprt, struct sock_xprt, xprt);
1162 struct xdr_buf *rcvbuf;
1166 rcvbuf = &req->rq_private_buf;
1168 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1170 * Save the RPC direction in the XDR buffer
1172 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1173 &transport->tcp_calldir,
1174 sizeof(transport->tcp_calldir));
1175 transport->tcp_copied += sizeof(transport->tcp_calldir);
1176 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1180 if (len > transport->tcp_reclen - transport->tcp_offset) {
1181 struct xdr_skb_reader my_desc;
1183 len = transport->tcp_reclen - transport->tcp_offset;
1184 memcpy(&my_desc, desc, sizeof(my_desc));
1185 my_desc.count = len;
1186 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1187 &my_desc, xdr_skb_read_bits);
1191 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1192 desc, xdr_skb_read_bits);
1195 transport->tcp_copied += r;
1196 transport->tcp_offset += r;
1199 /* Error when copying to the receive buffer,
1200 * usually because we weren't able to allocate
1201 * additional buffer pages. All we can do now
1202 * is turn off TCP_RCV_COPY_DATA, so the request
1203 * will not receive any additional updates,
1205 * Any remaining data from this record will
1208 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1209 dprintk("RPC: XID %08x truncated request\n",
1210 ntohl(transport->tcp_xid));
1211 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1212 "tcp_offset = %u, tcp_reclen = %u\n",
1213 xprt, transport->tcp_copied,
1214 transport->tcp_offset, transport->tcp_reclen);
1218 dprintk("RPC: XID %08x read %Zd bytes\n",
1219 ntohl(transport->tcp_xid), r);
1220 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1221 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1222 transport->tcp_offset, transport->tcp_reclen);
1224 if (transport->tcp_copied == req->rq_private_buf.buflen)
1225 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1226 else if (transport->tcp_offset == transport->tcp_reclen) {
1227 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1228 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1233 * Finds the request corresponding to the RPC xid and invokes the common
1234 * tcp read code to read the data.
1236 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1237 struct xdr_skb_reader *desc)
1239 struct sock_xprt *transport =
1240 container_of(xprt, struct sock_xprt, xprt);
1241 struct rpc_rqst *req;
1243 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1245 /* Find and lock the request corresponding to this xid */
1246 spin_lock(&xprt->transport_lock);
1247 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1249 dprintk("RPC: XID %08x request not found!\n",
1250 ntohl(transport->tcp_xid));
1251 spin_unlock(&xprt->transport_lock);
1255 xs_tcp_read_common(xprt, desc, req);
1257 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1258 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1260 spin_unlock(&xprt->transport_lock);
1264 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1266 * Obtains an rpc_rqst previously allocated and invokes the common
1267 * tcp read code to read the data. The result is placed in the callback
1269 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1270 * connection and return -1.
1272 static int xs_tcp_read_callback(struct rpc_xprt *xprt,
1273 struct xdr_skb_reader *desc)
1275 struct sock_xprt *transport =
1276 container_of(xprt, struct sock_xprt, xprt);
1277 struct rpc_rqst *req;
1279 /* Look up and lock the request corresponding to the given XID */
1280 spin_lock(&xprt->transport_lock);
1281 req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
1283 spin_unlock(&xprt->transport_lock);
1284 printk(KERN_WARNING "Callback slot table overflowed\n");
1285 xprt_force_disconnect(xprt);
1289 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1290 xs_tcp_read_common(xprt, desc, req);
1292 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1293 xprt_complete_bc_request(req, transport->tcp_copied);
1294 spin_unlock(&xprt->transport_lock);
1299 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1300 struct xdr_skb_reader *desc)
1302 struct sock_xprt *transport =
1303 container_of(xprt, struct sock_xprt, xprt);
1305 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1306 xs_tcp_read_reply(xprt, desc) :
1307 xs_tcp_read_callback(xprt, desc);
1310 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1311 struct xdr_skb_reader *desc)
1313 return xs_tcp_read_reply(xprt, desc);
1315 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1318 * Read data off the transport. This can be either an RPC_CALL or an
1319 * RPC_REPLY. Relay the processing to helper functions.
1321 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1322 struct xdr_skb_reader *desc)
1324 struct sock_xprt *transport =
1325 container_of(xprt, struct sock_xprt, xprt);
1327 if (_xs_tcp_read_data(xprt, desc) == 0)
1328 xs_tcp_check_fraghdr(transport);
1331 * The transport_lock protects the request handling.
1332 * There's no need to hold it to update the tcp_flags.
1334 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1338 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1342 len = transport->tcp_reclen - transport->tcp_offset;
1343 if (len > desc->count)
1346 desc->offset += len;
1347 transport->tcp_offset += len;
1348 dprintk("RPC: discarded %Zu bytes\n", len);
1349 xs_tcp_check_fraghdr(transport);
1352 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1354 struct rpc_xprt *xprt = rd_desc->arg.data;
1355 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1356 struct xdr_skb_reader desc = {
1362 dprintk("RPC: xs_tcp_data_recv started\n");
1364 trace_xs_tcp_data_recv(transport);
1365 /* Read in a new fragment marker if necessary */
1366 /* Can we ever really expect to get completely empty fragments? */
1367 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1368 xs_tcp_read_fraghdr(xprt, &desc);
1371 /* Read in the xid if necessary */
1372 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1373 xs_tcp_read_xid(transport, &desc);
1376 /* Read in the call/reply flag */
1377 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1378 xs_tcp_read_calldir(transport, &desc);
1381 /* Read in the request data */
1382 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1383 xs_tcp_read_data(xprt, &desc);
1386 /* Skip over any trailing bytes on short reads */
1387 xs_tcp_read_discard(transport, &desc);
1388 } while (desc.count);
1389 trace_xs_tcp_data_recv(transport);
1390 dprintk("RPC: xs_tcp_data_recv done\n");
1391 return len - desc.count;
1395 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1396 * @sk: socket with data to read
1399 static void xs_tcp_data_ready(struct sock *sk)
1401 struct rpc_xprt *xprt;
1402 read_descriptor_t rd_desc;
1404 unsigned long total = 0;
1406 dprintk("RPC: xs_tcp_data_ready...\n");
1408 read_lock_bh(&sk->sk_callback_lock);
1409 if (!(xprt = xprt_from_sock(sk))) {
1413 /* Any data means we had a useful conversation, so
1414 * the we don't need to delay the next reconnect
1416 if (xprt->reestablish_timeout)
1417 xprt->reestablish_timeout = 0;
1419 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1420 rd_desc.arg.data = xprt;
1422 rd_desc.count = 65536;
1423 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1428 trace_xs_tcp_data_ready(xprt, read, total);
1429 read_unlock_bh(&sk->sk_callback_lock);
1433 * xs_tcp_state_change - callback to handle TCP socket state changes
1434 * @sk: socket whose state has changed
1437 static void xs_tcp_state_change(struct sock *sk)
1439 struct rpc_xprt *xprt;
1440 struct sock_xprt *transport;
1442 read_lock_bh(&sk->sk_callback_lock);
1443 if (!(xprt = xprt_from_sock(sk)))
1445 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1446 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1447 sk->sk_state, xprt_connected(xprt),
1448 sock_flag(sk, SOCK_DEAD),
1449 sock_flag(sk, SOCK_ZAPPED),
1452 transport = container_of(xprt, struct sock_xprt, xprt);
1453 trace_rpc_socket_state_change(xprt, sk->sk_socket);
1454 switch (sk->sk_state) {
1455 case TCP_ESTABLISHED:
1456 spin_lock(&xprt->transport_lock);
1457 if (!xprt_test_and_set_connected(xprt)) {
1459 /* Reset TCP record info */
1460 transport->tcp_offset = 0;
1461 transport->tcp_reclen = 0;
1462 transport->tcp_copied = 0;
1463 transport->tcp_flags =
1464 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1465 xprt->connect_cookie++;
1466 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1467 xprt_clear_connecting(xprt);
1469 xprt_wake_pending_tasks(xprt, -EAGAIN);
1471 spin_unlock(&xprt->transport_lock);
1474 /* The client initiated a shutdown of the socket */
1475 xprt->connect_cookie++;
1476 xprt->reestablish_timeout = 0;
1477 set_bit(XPRT_CLOSING, &xprt->state);
1478 smp_mb__before_atomic();
1479 clear_bit(XPRT_CONNECTED, &xprt->state);
1480 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1481 smp_mb__after_atomic();
1483 case TCP_CLOSE_WAIT:
1484 /* The server initiated a shutdown of the socket */
1485 xprt->connect_cookie++;
1486 clear_bit(XPRT_CONNECTED, &xprt->state);
1487 xs_tcp_force_close(xprt);
1490 * If the server closed down the connection, make sure that
1491 * we back off before reconnecting
1493 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1494 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1497 set_bit(XPRT_CLOSING, &xprt->state);
1498 smp_mb__before_atomic();
1499 clear_bit(XPRT_CONNECTED, &xprt->state);
1500 smp_mb__after_atomic();
1503 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1504 &transport->sock_state))
1505 xprt_clear_connecting(xprt);
1506 xs_sock_mark_closed(xprt);
1509 read_unlock_bh(&sk->sk_callback_lock);
1512 static void xs_write_space(struct sock *sk)
1514 struct socket *sock;
1515 struct rpc_xprt *xprt;
1517 if (unlikely(!(sock = sk->sk_socket)))
1519 clear_bit(SOCK_NOSPACE, &sock->flags);
1521 if (unlikely(!(xprt = xprt_from_sock(sk))))
1523 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1526 xprt_write_space(xprt);
1530 * xs_udp_write_space - callback invoked when socket buffer space
1532 * @sk: socket whose state has changed
1534 * Called when more output buffer space is available for this socket.
1535 * We try not to wake our writers until they can make "significant"
1536 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1537 * with a bunch of small requests.
1539 static void xs_udp_write_space(struct sock *sk)
1541 read_lock_bh(&sk->sk_callback_lock);
1543 /* from net/core/sock.c:sock_def_write_space */
1544 if (sock_writeable(sk))
1547 read_unlock_bh(&sk->sk_callback_lock);
1551 * xs_tcp_write_space - callback invoked when socket buffer space
1553 * @sk: socket whose state has changed
1555 * Called when more output buffer space is available for this socket.
1556 * We try not to wake our writers until they can make "significant"
1557 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1558 * with a bunch of small requests.
1560 static void xs_tcp_write_space(struct sock *sk)
1562 read_lock_bh(&sk->sk_callback_lock);
1564 /* from net/core/stream.c:sk_stream_write_space */
1565 if (sk_stream_is_writeable(sk))
1568 read_unlock_bh(&sk->sk_callback_lock);
1571 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1573 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1574 struct sock *sk = transport->inet;
1576 if (transport->rcvsize) {
1577 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1578 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1580 if (transport->sndsize) {
1581 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1582 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1583 sk->sk_write_space(sk);
1588 * xs_udp_set_buffer_size - set send and receive limits
1589 * @xprt: generic transport
1590 * @sndsize: requested size of send buffer, in bytes
1591 * @rcvsize: requested size of receive buffer, in bytes
1593 * Set socket send and receive buffer size limits.
1595 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1597 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1599 transport->sndsize = 0;
1601 transport->sndsize = sndsize + 1024;
1602 transport->rcvsize = 0;
1604 transport->rcvsize = rcvsize + 1024;
1606 xs_udp_do_set_buffer_size(xprt);
1610 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1611 * @task: task that timed out
1613 * Adjust the congestion window after a retransmit timeout has occurred.
1615 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1617 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1620 static unsigned short xs_get_random_port(void)
1622 unsigned short range = xprt_max_resvport - xprt_min_resvport;
1623 unsigned short rand = (unsigned short) prandom_u32() % range;
1624 return rand + xprt_min_resvport;
1628 * xs_set_reuseaddr_port - set the socket's port and address reuse options
1631 * Note that this function has to be called on all sockets that share the
1632 * same port, and it must be called before binding.
1634 static void xs_sock_set_reuseport(struct socket *sock)
1638 kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1639 (char *)&opt, sizeof(opt));
1642 static unsigned short xs_sock_getport(struct socket *sock)
1644 struct sockaddr_storage buf;
1646 unsigned short port = 0;
1648 if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0)
1650 switch (buf.ss_family) {
1652 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1655 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1662 * xs_set_port - reset the port number in the remote endpoint address
1663 * @xprt: generic transport
1664 * @port: new port number
1667 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1669 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1671 rpc_set_port(xs_addr(xprt), port);
1672 xs_update_peer_port(xprt);
1675 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1677 if (transport->srcport == 0)
1678 transport->srcport = xs_sock_getport(sock);
1681 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1683 unsigned short port = transport->srcport;
1685 if (port == 0 && transport->xprt.resvport)
1686 port = xs_get_random_port();
1690 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1692 if (transport->srcport != 0)
1693 transport->srcport = 0;
1694 if (!transport->xprt.resvport)
1696 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1697 return xprt_max_resvport;
1700 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1702 struct sockaddr_storage myaddr;
1704 unsigned short port = xs_get_srcport(transport);
1705 unsigned short last;
1708 * If we are asking for any ephemeral port (i.e. port == 0 &&
1709 * transport->xprt.resvport == 0), don't bind. Let the local
1710 * port selection happen implicitly when the socket is used
1711 * (for example at connect time).
1713 * This ensures that we can continue to establish TCP
1714 * connections even when all local ephemeral ports are already
1715 * a part of some TCP connection. This makes no difference
1716 * for UDP sockets, but also doens't harm them.
1718 * If we're asking for any reserved port (i.e. port == 0 &&
1719 * transport->xprt.resvport == 1) xs_get_srcport above will
1720 * ensure that port is non-zero and we will bind as needed.
1725 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1727 rpc_set_port((struct sockaddr *)&myaddr, port);
1728 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1729 transport->xprt.addrlen);
1731 transport->srcport = port;
1735 port = xs_next_srcport(transport, port);
1738 } while (err == -EADDRINUSE && nloop != 2);
1740 if (myaddr.ss_family == AF_INET)
1741 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1742 &((struct sockaddr_in *)&myaddr)->sin_addr,
1743 port, err ? "failed" : "ok", err);
1745 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1746 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1747 port, err ? "failed" : "ok", err);
1752 * We don't support autobind on AF_LOCAL sockets
1754 static void xs_local_rpcbind(struct rpc_task *task)
1757 xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt));
1761 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1765 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1766 static struct lock_class_key xs_key[2];
1767 static struct lock_class_key xs_slock_key[2];
1769 static inline void xs_reclassify_socketu(struct socket *sock)
1771 struct sock *sk = sock->sk;
1773 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1774 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1777 static inline void xs_reclassify_socket4(struct socket *sock)
1779 struct sock *sk = sock->sk;
1781 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1782 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1785 static inline void xs_reclassify_socket6(struct socket *sock)
1787 struct sock *sk = sock->sk;
1789 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1790 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1793 static inline void xs_reclassify_socket(int family, struct socket *sock)
1795 WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1796 if (sock_owned_by_user(sock->sk))
1801 xs_reclassify_socketu(sock);
1804 xs_reclassify_socket4(sock);
1807 xs_reclassify_socket6(sock);
1812 static inline void xs_reclassify_socketu(struct socket *sock)
1816 static inline void xs_reclassify_socket4(struct socket *sock)
1820 static inline void xs_reclassify_socket6(struct socket *sock)
1824 static inline void xs_reclassify_socket(int family, struct socket *sock)
1829 static void xs_dummy_setup_socket(struct work_struct *work)
1833 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1834 struct sock_xprt *transport, int family, int type,
1835 int protocol, bool reuseport)
1837 struct socket *sock;
1840 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1842 dprintk("RPC: can't create %d transport socket (%d).\n",
1846 xs_reclassify_socket(family, sock);
1849 xs_sock_set_reuseport(sock);
1851 err = xs_bind(transport, sock);
1859 return ERR_PTR(err);
1862 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1863 struct socket *sock)
1865 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1868 if (!transport->inet) {
1869 struct sock *sk = sock->sk;
1871 write_lock_bh(&sk->sk_callback_lock);
1873 xs_save_old_callbacks(transport, sk);
1875 sk->sk_user_data = xprt;
1876 sk->sk_data_ready = xs_local_data_ready;
1877 sk->sk_write_space = xs_udp_write_space;
1878 sk->sk_error_report = xs_error_report;
1879 sk->sk_allocation = GFP_NOIO;
1881 xprt_clear_connected(xprt);
1883 /* Reset to new socket */
1884 transport->sock = sock;
1885 transport->inet = sk;
1887 write_unlock_bh(&sk->sk_callback_lock);
1890 /* Tell the socket layer to start connecting... */
1891 xprt->stat.connect_count++;
1892 xprt->stat.connect_start = jiffies;
1893 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1897 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1898 * @transport: socket transport to connect
1900 static int xs_local_setup_socket(struct sock_xprt *transport)
1902 struct rpc_xprt *xprt = &transport->xprt;
1903 struct socket *sock;
1906 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1907 SOCK_STREAM, 0, &sock, 1);
1909 dprintk("RPC: can't create AF_LOCAL "
1910 "transport socket (%d).\n", -status);
1913 xs_reclassify_socketu(sock);
1915 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1916 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1918 status = xs_local_finish_connecting(xprt, sock);
1919 trace_rpc_socket_connect(xprt, sock, status);
1922 dprintk("RPC: xprt %p connected to %s\n",
1923 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1924 xprt_set_connected(xprt);
1928 dprintk("RPC: xprt %p: socket %s does not exist\n",
1929 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1932 dprintk("RPC: xprt %p: connection refused for %s\n",
1933 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1936 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1938 xprt->address_strings[RPC_DISPLAY_ADDR]);
1942 xprt_clear_connecting(xprt);
1943 xprt_wake_pending_tasks(xprt, status);
1947 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1949 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1952 if (RPC_IS_ASYNC(task)) {
1954 * We want the AF_LOCAL connect to be resolved in the
1955 * filesystem namespace of the process making the rpc
1956 * call. Thus we connect synchronously.
1958 * If we want to support asynchronous AF_LOCAL calls,
1959 * we'll need to figure out how to pass a namespace to
1962 rpc_exit(task, -ENOTCONN);
1965 ret = xs_local_setup_socket(transport);
1966 if (ret && !RPC_IS_SOFTCONN(task))
1967 msleep_interruptible(15000);
1970 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
1972 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
1973 * know that we have exclusive access to the socket), to guard against
1974 * races with xs_reset_transport.
1976 static void xs_set_memalloc(struct rpc_xprt *xprt)
1978 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1982 * If there's no sock, then we have nothing to set. The
1983 * reconnecting process will get it for us.
1985 if (!transport->inet)
1987 if (atomic_read(&xprt->swapper))
1988 sk_set_memalloc(transport->inet);
1992 * xs_enable_swap - Tag this transport as being used for swap.
1993 * @xprt: transport to tag
1995 * Take a reference to this transport on behalf of the rpc_clnt, and
1996 * optionally mark it for swapping if it wasn't already.
1999 xs_enable_swap(struct rpc_xprt *xprt)
2001 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2003 if (atomic_inc_return(&xprt->swapper) != 1)
2005 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2006 return -ERESTARTSYS;
2008 sk_set_memalloc(xs->inet);
2009 xprt_release_xprt(xprt, NULL);
2014 * xs_disable_swap - Untag this transport as being used for swap.
2015 * @xprt: transport to tag
2017 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2018 * swapper refcount goes to 0, untag the socket as a memalloc socket.
2021 xs_disable_swap(struct rpc_xprt *xprt)
2023 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2025 if (!atomic_dec_and_test(&xprt->swapper))
2027 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2030 sk_clear_memalloc(xs->inet);
2031 xprt_release_xprt(xprt, NULL);
2034 static void xs_set_memalloc(struct rpc_xprt *xprt)
2039 xs_enable_swap(struct rpc_xprt *xprt)
2045 xs_disable_swap(struct rpc_xprt *xprt)
2050 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2052 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2054 if (!transport->inet) {
2055 struct sock *sk = sock->sk;
2057 write_lock_bh(&sk->sk_callback_lock);
2059 xs_save_old_callbacks(transport, sk);
2061 sk->sk_user_data = xprt;
2062 sk->sk_data_ready = xs_udp_data_ready;
2063 sk->sk_write_space = xs_udp_write_space;
2064 sk->sk_allocation = GFP_NOIO;
2066 xprt_set_connected(xprt);
2068 /* Reset to new socket */
2069 transport->sock = sock;
2070 transport->inet = sk;
2072 xs_set_memalloc(xprt);
2074 write_unlock_bh(&sk->sk_callback_lock);
2076 xs_udp_do_set_buffer_size(xprt);
2079 static void xs_udp_setup_socket(struct work_struct *work)
2081 struct sock_xprt *transport =
2082 container_of(work, struct sock_xprt, connect_worker.work);
2083 struct rpc_xprt *xprt = &transport->xprt;
2084 struct socket *sock = transport->sock;
2087 sock = xs_create_sock(xprt, transport,
2088 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2089 IPPROTO_UDP, false);
2093 dprintk("RPC: worker connecting xprt %p via %s to "
2094 "%s (port %s)\n", xprt,
2095 xprt->address_strings[RPC_DISPLAY_PROTO],
2096 xprt->address_strings[RPC_DISPLAY_ADDR],
2097 xprt->address_strings[RPC_DISPLAY_PORT]);
2099 xs_udp_finish_connecting(xprt, sock);
2100 trace_rpc_socket_connect(xprt, sock, 0);
2103 xprt_unlock_connect(xprt, transport);
2104 xprt_clear_connecting(xprt);
2105 xprt_wake_pending_tasks(xprt, status);
2109 * xs_tcp_shutdown - gracefully shut down a TCP socket
2112 * Initiates a graceful shutdown of the TCP socket by calling the
2113 * equivalent of shutdown(SHUT_RDWR);
2115 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2117 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2118 struct socket *sock = transport->sock;
2122 if (xprt_connected(xprt)) {
2123 kernel_sock_shutdown(sock, SHUT_RDWR);
2124 trace_rpc_socket_shutdown(xprt, sock);
2126 xs_reset_transport(transport);
2129 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2131 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2132 int ret = -ENOTCONN;
2134 if (!transport->inet) {
2135 struct sock *sk = sock->sk;
2136 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2137 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2138 unsigned int opt_on = 1;
2141 /* TCP Keepalive options */
2142 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2143 (char *)&opt_on, sizeof(opt_on));
2144 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2145 (char *)&keepidle, sizeof(keepidle));
2146 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2147 (char *)&keepidle, sizeof(keepidle));
2148 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2149 (char *)&keepcnt, sizeof(keepcnt));
2151 /* TCP user timeout (see RFC5482) */
2152 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2153 (xprt->timeout->to_retries + 1);
2154 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2155 (char *)&timeo, sizeof(timeo));
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_tcp_data_ready;
2163 sk->sk_state_change = xs_tcp_state_change;
2164 sk->sk_write_space = xs_tcp_write_space;
2165 sk->sk_error_report = xs_error_report;
2166 sk->sk_allocation = GFP_NOIO;
2168 /* socket options */
2169 sock_reset_flag(sk, SOCK_LINGER);
2170 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2172 xprt_clear_connected(xprt);
2174 /* Reset to new socket */
2175 transport->sock = sock;
2176 transport->inet = sk;
2178 write_unlock_bh(&sk->sk_callback_lock);
2181 if (!xprt_bound(xprt))
2184 xs_set_memalloc(xprt);
2186 /* Tell the socket layer to start connecting... */
2187 xprt->stat.connect_count++;
2188 xprt->stat.connect_start = jiffies;
2189 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2190 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2193 xs_set_srcport(transport, sock);
2196 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2197 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2204 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2206 * Invoked by a work queue tasklet.
2208 static void xs_tcp_setup_socket(struct work_struct *work)
2210 struct sock_xprt *transport =
2211 container_of(work, struct sock_xprt, connect_worker.work);
2212 struct socket *sock = transport->sock;
2213 struct rpc_xprt *xprt = &transport->xprt;
2217 sock = xs_create_sock(xprt, transport,
2218 xs_addr(xprt)->sa_family, SOCK_STREAM,
2221 status = PTR_ERR(sock);
2226 dprintk("RPC: worker connecting xprt %p via %s to "
2227 "%s (port %s)\n", xprt,
2228 xprt->address_strings[RPC_DISPLAY_PROTO],
2229 xprt->address_strings[RPC_DISPLAY_ADDR],
2230 xprt->address_strings[RPC_DISPLAY_PORT]);
2232 status = xs_tcp_finish_connecting(xprt, sock);
2233 trace_rpc_socket_connect(xprt, sock, status);
2234 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2235 xprt, -status, xprt_connected(xprt),
2236 sock->sk->sk_state);
2239 printk("%s: connect returned unhandled error %d\n",
2241 case -EADDRNOTAVAIL:
2242 /* We're probably in TIME_WAIT. Get rid of existing socket,
2245 xs_tcp_force_close(xprt);
2250 xprt_unlock_connect(xprt, transport);
2253 /* Happens, for instance, if the user specified a link
2254 * local IPv6 address without a scope-id.
2261 /* retry with existing socket, after a delay */
2262 xs_tcp_force_close(xprt);
2267 xprt_unlock_connect(xprt, transport);
2268 xprt_clear_connecting(xprt);
2269 xprt_wake_pending_tasks(xprt, status);
2273 * xs_connect - connect a socket to a remote endpoint
2274 * @xprt: pointer to transport structure
2275 * @task: address of RPC task that manages state of connect request
2277 * TCP: If the remote end dropped the connection, delay reconnecting.
2279 * UDP socket connects are synchronous, but we use a work queue anyway
2280 * to guarantee that even unprivileged user processes can set up a
2281 * socket on a privileged port.
2283 * If a UDP socket connect fails, the delay behavior here prevents
2284 * retry floods (hard mounts).
2286 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2288 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2290 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2292 if (transport->sock != NULL) {
2293 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2295 xprt, xprt->reestablish_timeout / HZ);
2297 /* Start by resetting any existing state */
2298 xs_reset_transport(transport);
2300 queue_delayed_work(rpciod_workqueue,
2301 &transport->connect_worker,
2302 xprt->reestablish_timeout);
2303 xprt->reestablish_timeout <<= 1;
2304 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2305 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2306 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2307 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2309 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2310 queue_delayed_work(rpciod_workqueue,
2311 &transport->connect_worker, 0);
2316 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2317 * @xprt: rpc_xprt struct containing statistics
2321 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2325 if (xprt_connected(xprt))
2326 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2328 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2329 "%llu %llu %lu %llu %llu\n",
2330 xprt->stat.bind_count,
2331 xprt->stat.connect_count,
2332 xprt->stat.connect_time,
2336 xprt->stat.bad_xids,
2339 xprt->stat.max_slots,
2340 xprt->stat.sending_u,
2341 xprt->stat.pending_u);
2345 * xs_udp_print_stats - display UDP socket-specifc stats
2346 * @xprt: rpc_xprt struct containing statistics
2350 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2352 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2354 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2357 xprt->stat.bind_count,
2360 xprt->stat.bad_xids,
2363 xprt->stat.max_slots,
2364 xprt->stat.sending_u,
2365 xprt->stat.pending_u);
2369 * xs_tcp_print_stats - display TCP socket-specifc stats
2370 * @xprt: rpc_xprt struct containing statistics
2374 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2376 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2379 if (xprt_connected(xprt))
2380 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2382 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2383 "%llu %llu %lu %llu %llu\n",
2385 xprt->stat.bind_count,
2386 xprt->stat.connect_count,
2387 xprt->stat.connect_time,
2391 xprt->stat.bad_xids,
2394 xprt->stat.max_slots,
2395 xprt->stat.sending_u,
2396 xprt->stat.pending_u);
2400 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2401 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2402 * to use the server side send routines.
2404 static void *bc_malloc(struct rpc_task *task, size_t size)
2407 struct rpc_buffer *buf;
2409 WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2410 if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2413 page = alloc_page(GFP_KERNEL);
2417 buf = page_address(page);
2418 buf->len = PAGE_SIZE;
2424 * Free the space allocated in the bc_alloc routine
2426 static void bc_free(void *buffer)
2428 struct rpc_buffer *buf;
2433 buf = container_of(buffer, struct rpc_buffer, data);
2434 free_page((unsigned long)buf);
2438 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2439 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2441 static int bc_sendto(struct rpc_rqst *req)
2444 struct xdr_buf *xbufp = &req->rq_snd_buf;
2445 struct rpc_xprt *xprt = req->rq_xprt;
2446 struct sock_xprt *transport =
2447 container_of(xprt, struct sock_xprt, xprt);
2448 struct socket *sock = transport->sock;
2449 unsigned long headoff;
2450 unsigned long tailoff;
2452 xs_encode_stream_record_marker(xbufp);
2454 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2455 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2456 len = svc_send_common(sock, xbufp,
2457 virt_to_page(xbufp->head[0].iov_base), headoff,
2458 xbufp->tail[0].iov_base, tailoff);
2460 if (len != xbufp->len) {
2461 printk(KERN_NOTICE "Error sending entire callback!\n");
2469 * The send routine. Borrows from svc_send
2471 static int bc_send_request(struct rpc_task *task)
2473 struct rpc_rqst *req = task->tk_rqstp;
2474 struct svc_xprt *xprt;
2477 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2479 * Get the server socket associated with this callback xprt
2481 xprt = req->rq_xprt->bc_xprt;
2484 * Grab the mutex to serialize data as the connection is shared
2485 * with the fore channel
2487 if (!mutex_trylock(&xprt->xpt_mutex)) {
2488 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2489 if (!mutex_trylock(&xprt->xpt_mutex))
2491 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2493 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2496 len = bc_sendto(req);
2497 mutex_unlock(&xprt->xpt_mutex);
2506 * The close routine. Since this is client initiated, we do nothing
2509 static void bc_close(struct rpc_xprt *xprt)
2514 * The xprt destroy routine. Again, because this connection is client
2515 * initiated, we do nothing
2518 static void bc_destroy(struct rpc_xprt *xprt)
2520 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2523 module_put(THIS_MODULE);
2526 static struct rpc_xprt_ops xs_local_ops = {
2527 .reserve_xprt = xprt_reserve_xprt,
2528 .release_xprt = xs_tcp_release_xprt,
2529 .alloc_slot = xprt_alloc_slot,
2530 .rpcbind = xs_local_rpcbind,
2531 .set_port = xs_local_set_port,
2532 .connect = xs_local_connect,
2533 .buf_alloc = rpc_malloc,
2534 .buf_free = rpc_free,
2535 .send_request = xs_local_send_request,
2536 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2538 .destroy = xs_destroy,
2539 .print_stats = xs_local_print_stats,
2540 .enable_swap = xs_enable_swap,
2541 .disable_swap = xs_disable_swap,
2544 static struct rpc_xprt_ops xs_udp_ops = {
2545 .set_buffer_size = xs_udp_set_buffer_size,
2546 .reserve_xprt = xprt_reserve_xprt_cong,
2547 .release_xprt = xprt_release_xprt_cong,
2548 .alloc_slot = xprt_alloc_slot,
2549 .rpcbind = rpcb_getport_async,
2550 .set_port = xs_set_port,
2551 .connect = xs_connect,
2552 .buf_alloc = rpc_malloc,
2553 .buf_free = rpc_free,
2554 .send_request = xs_udp_send_request,
2555 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
2556 .timer = xs_udp_timer,
2557 .release_request = xprt_release_rqst_cong,
2559 .destroy = xs_destroy,
2560 .print_stats = xs_udp_print_stats,
2561 .enable_swap = xs_enable_swap,
2562 .disable_swap = xs_disable_swap,
2563 .inject_disconnect = xs_inject_disconnect,
2566 static struct rpc_xprt_ops xs_tcp_ops = {
2567 .reserve_xprt = xprt_reserve_xprt,
2568 .release_xprt = xs_tcp_release_xprt,
2569 .alloc_slot = xprt_lock_and_alloc_slot,
2570 .rpcbind = rpcb_getport_async,
2571 .set_port = xs_set_port,
2572 .connect = xs_connect,
2573 .buf_alloc = rpc_malloc,
2574 .buf_free = rpc_free,
2575 .send_request = xs_tcp_send_request,
2576 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2577 .close = xs_tcp_shutdown,
2578 .destroy = xs_destroy,
2579 .print_stats = xs_tcp_print_stats,
2580 .enable_swap = xs_enable_swap,
2581 .disable_swap = xs_disable_swap,
2582 .inject_disconnect = xs_inject_disconnect,
2586 * The rpc_xprt_ops for the server backchannel
2589 static struct rpc_xprt_ops bc_tcp_ops = {
2590 .reserve_xprt = xprt_reserve_xprt,
2591 .release_xprt = xprt_release_xprt,
2592 .alloc_slot = xprt_alloc_slot,
2593 .buf_alloc = bc_malloc,
2594 .buf_free = bc_free,
2595 .send_request = bc_send_request,
2596 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2598 .destroy = bc_destroy,
2599 .print_stats = xs_tcp_print_stats,
2600 .enable_swap = xs_enable_swap,
2601 .disable_swap = xs_disable_swap,
2602 .inject_disconnect = xs_inject_disconnect,
2605 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2607 static const struct sockaddr_in sin = {
2608 .sin_family = AF_INET,
2609 .sin_addr.s_addr = htonl(INADDR_ANY),
2611 static const struct sockaddr_in6 sin6 = {
2612 .sin6_family = AF_INET6,
2613 .sin6_addr = IN6ADDR_ANY_INIT,
2620 memcpy(sap, &sin, sizeof(sin));
2623 memcpy(sap, &sin6, sizeof(sin6));
2626 dprintk("RPC: %s: Bad address family\n", __func__);
2627 return -EAFNOSUPPORT;
2632 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2633 unsigned int slot_table_size,
2634 unsigned int max_slot_table_size)
2636 struct rpc_xprt *xprt;
2637 struct sock_xprt *new;
2639 if (args->addrlen > sizeof(xprt->addr)) {
2640 dprintk("RPC: xs_setup_xprt: address too large\n");
2641 return ERR_PTR(-EBADF);
2644 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2645 max_slot_table_size);
2647 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2649 return ERR_PTR(-ENOMEM);
2652 new = container_of(xprt, struct sock_xprt, xprt);
2653 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2654 xprt->addrlen = args->addrlen;
2656 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2659 err = xs_init_anyaddr(args->dstaddr->sa_family,
2660 (struct sockaddr *)&new->srcaddr);
2663 return ERR_PTR(err);
2670 static const struct rpc_timeout xs_local_default_timeout = {
2671 .to_initval = 10 * HZ,
2672 .to_maxval = 10 * HZ,
2677 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2678 * @args: rpc transport creation arguments
2680 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2682 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2684 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2685 struct sock_xprt *transport;
2686 struct rpc_xprt *xprt;
2687 struct rpc_xprt *ret;
2689 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2690 xprt_max_tcp_slot_table_entries);
2693 transport = container_of(xprt, struct sock_xprt, xprt);
2696 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2697 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2699 xprt->bind_timeout = XS_BIND_TO;
2700 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2701 xprt->idle_timeout = XS_IDLE_DISC_TO;
2703 xprt->ops = &xs_local_ops;
2704 xprt->timeout = &xs_local_default_timeout;
2706 INIT_DELAYED_WORK(&transport->connect_worker,
2707 xs_dummy_setup_socket);
2709 switch (sun->sun_family) {
2711 if (sun->sun_path[0] != '/') {
2712 dprintk("RPC: bad AF_LOCAL address: %s\n",
2714 ret = ERR_PTR(-EINVAL);
2717 xprt_set_bound(xprt);
2718 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2719 ret = ERR_PTR(xs_local_setup_socket(transport));
2724 ret = ERR_PTR(-EAFNOSUPPORT);
2728 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2729 xprt->address_strings[RPC_DISPLAY_ADDR]);
2731 if (try_module_get(THIS_MODULE))
2733 ret = ERR_PTR(-EINVAL);
2739 static const struct rpc_timeout xs_udp_default_timeout = {
2740 .to_initval = 5 * HZ,
2741 .to_maxval = 30 * HZ,
2742 .to_increment = 5 * HZ,
2747 * xs_setup_udp - Set up transport to use a UDP socket
2748 * @args: rpc transport creation arguments
2751 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2753 struct sockaddr *addr = args->dstaddr;
2754 struct rpc_xprt *xprt;
2755 struct sock_xprt *transport;
2756 struct rpc_xprt *ret;
2758 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2759 xprt_udp_slot_table_entries);
2762 transport = container_of(xprt, struct sock_xprt, xprt);
2764 xprt->prot = IPPROTO_UDP;
2766 /* XXX: header size can vary due to auth type, IPv6, etc. */
2767 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2769 xprt->bind_timeout = XS_BIND_TO;
2770 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2771 xprt->idle_timeout = XS_IDLE_DISC_TO;
2773 xprt->ops = &xs_udp_ops;
2775 xprt->timeout = &xs_udp_default_timeout;
2777 switch (addr->sa_family) {
2779 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2780 xprt_set_bound(xprt);
2782 INIT_DELAYED_WORK(&transport->connect_worker,
2783 xs_udp_setup_socket);
2784 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2787 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2788 xprt_set_bound(xprt);
2790 INIT_DELAYED_WORK(&transport->connect_worker,
2791 xs_udp_setup_socket);
2792 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2795 ret = ERR_PTR(-EAFNOSUPPORT);
2799 if (xprt_bound(xprt))
2800 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2801 xprt->address_strings[RPC_DISPLAY_ADDR],
2802 xprt->address_strings[RPC_DISPLAY_PORT],
2803 xprt->address_strings[RPC_DISPLAY_PROTO]);
2805 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2806 xprt->address_strings[RPC_DISPLAY_ADDR],
2807 xprt->address_strings[RPC_DISPLAY_PROTO]);
2809 if (try_module_get(THIS_MODULE))
2811 ret = ERR_PTR(-EINVAL);
2817 static const struct rpc_timeout xs_tcp_default_timeout = {
2818 .to_initval = 60 * HZ,
2819 .to_maxval = 60 * HZ,
2824 * xs_setup_tcp - Set up transport to use a TCP socket
2825 * @args: rpc transport creation arguments
2828 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2830 struct sockaddr *addr = args->dstaddr;
2831 struct rpc_xprt *xprt;
2832 struct sock_xprt *transport;
2833 struct rpc_xprt *ret;
2834 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2836 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2837 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2839 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2840 max_slot_table_size);
2843 transport = container_of(xprt, struct sock_xprt, xprt);
2845 xprt->prot = IPPROTO_TCP;
2846 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2847 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2849 xprt->bind_timeout = XS_BIND_TO;
2850 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2851 xprt->idle_timeout = XS_IDLE_DISC_TO;
2853 xprt->ops = &xs_tcp_ops;
2854 xprt->timeout = &xs_tcp_default_timeout;
2856 switch (addr->sa_family) {
2858 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2859 xprt_set_bound(xprt);
2861 INIT_DELAYED_WORK(&transport->connect_worker,
2862 xs_tcp_setup_socket);
2863 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2866 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2867 xprt_set_bound(xprt);
2869 INIT_DELAYED_WORK(&transport->connect_worker,
2870 xs_tcp_setup_socket);
2871 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2874 ret = ERR_PTR(-EAFNOSUPPORT);
2878 if (xprt_bound(xprt))
2879 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2880 xprt->address_strings[RPC_DISPLAY_ADDR],
2881 xprt->address_strings[RPC_DISPLAY_PORT],
2882 xprt->address_strings[RPC_DISPLAY_PROTO]);
2884 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2885 xprt->address_strings[RPC_DISPLAY_ADDR],
2886 xprt->address_strings[RPC_DISPLAY_PROTO]);
2888 if (try_module_get(THIS_MODULE))
2890 ret = ERR_PTR(-EINVAL);
2897 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2898 * @args: rpc transport creation arguments
2901 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2903 struct sockaddr *addr = args->dstaddr;
2904 struct rpc_xprt *xprt;
2905 struct sock_xprt *transport;
2906 struct svc_sock *bc_sock;
2907 struct rpc_xprt *ret;
2909 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2910 xprt_tcp_slot_table_entries);
2913 transport = container_of(xprt, struct sock_xprt, xprt);
2915 xprt->prot = IPPROTO_TCP;
2916 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2917 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2918 xprt->timeout = &xs_tcp_default_timeout;
2921 xprt_set_bound(xprt);
2922 xprt->bind_timeout = 0;
2923 xprt->reestablish_timeout = 0;
2924 xprt->idle_timeout = 0;
2926 xprt->ops = &bc_tcp_ops;
2928 switch (addr->sa_family) {
2930 xs_format_peer_addresses(xprt, "tcp",
2934 xs_format_peer_addresses(xprt, "tcp",
2935 RPCBIND_NETID_TCP6);
2938 ret = ERR_PTR(-EAFNOSUPPORT);
2942 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2943 xprt->address_strings[RPC_DISPLAY_ADDR],
2944 xprt->address_strings[RPC_DISPLAY_PORT],
2945 xprt->address_strings[RPC_DISPLAY_PROTO]);
2948 * Once we've associated a backchannel xprt with a connection,
2949 * we want to keep it around as long as the connection lasts,
2950 * in case we need to start using it for a backchannel again;
2951 * this reference won't be dropped until bc_xprt is destroyed.
2954 args->bc_xprt->xpt_bc_xprt = xprt;
2955 xprt->bc_xprt = args->bc_xprt;
2956 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2957 transport->sock = bc_sock->sk_sock;
2958 transport->inet = bc_sock->sk_sk;
2961 * Since we don't want connections for the backchannel, we set
2962 * the xprt status to connected
2964 xprt_set_connected(xprt);
2966 if (try_module_get(THIS_MODULE))
2969 args->bc_xprt->xpt_bc_xprt = NULL;
2971 ret = ERR_PTR(-EINVAL);
2977 static struct xprt_class xs_local_transport = {
2978 .list = LIST_HEAD_INIT(xs_local_transport.list),
2979 .name = "named UNIX socket",
2980 .owner = THIS_MODULE,
2981 .ident = XPRT_TRANSPORT_LOCAL,
2982 .setup = xs_setup_local,
2985 static struct xprt_class xs_udp_transport = {
2986 .list = LIST_HEAD_INIT(xs_udp_transport.list),
2988 .owner = THIS_MODULE,
2989 .ident = XPRT_TRANSPORT_UDP,
2990 .setup = xs_setup_udp,
2993 static struct xprt_class xs_tcp_transport = {
2994 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
2996 .owner = THIS_MODULE,
2997 .ident = XPRT_TRANSPORT_TCP,
2998 .setup = xs_setup_tcp,
3001 static struct xprt_class xs_bc_tcp_transport = {
3002 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3003 .name = "tcp NFSv4.1 backchannel",
3004 .owner = THIS_MODULE,
3005 .ident = XPRT_TRANSPORT_BC_TCP,
3006 .setup = xs_setup_bc_tcp,
3010 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3013 int init_socket_xprt(void)
3015 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3016 if (!sunrpc_table_header)
3017 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3020 xprt_register_transport(&xs_local_transport);
3021 xprt_register_transport(&xs_udp_transport);
3022 xprt_register_transport(&xs_tcp_transport);
3023 xprt_register_transport(&xs_bc_tcp_transport);
3029 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3032 void cleanup_socket_xprt(void)
3034 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3035 if (sunrpc_table_header) {
3036 unregister_sysctl_table(sunrpc_table_header);
3037 sunrpc_table_header = NULL;
3041 xprt_unregister_transport(&xs_local_transport);
3042 xprt_unregister_transport(&xs_udp_transport);
3043 xprt_unregister_transport(&xs_tcp_transport);
3044 xprt_unregister_transport(&xs_bc_tcp_transport);
3047 static int param_set_uint_minmax(const char *val,
3048 const struct kernel_param *kp,
3049 unsigned int min, unsigned int max)
3056 ret = kstrtouint(val, 0, &num);
3057 if (ret == -EINVAL || num < min || num > max)
3059 *((unsigned int *)kp->arg) = num;
3063 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3065 return param_set_uint_minmax(val, kp,
3070 static const struct kernel_param_ops param_ops_portnr = {
3071 .set = param_set_portnr,
3072 .get = param_get_uint,
3075 #define param_check_portnr(name, p) \
3076 __param_check(name, p, unsigned int);
3078 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3079 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3081 static int param_set_slot_table_size(const char *val,
3082 const struct kernel_param *kp)
3084 return param_set_uint_minmax(val, kp,
3086 RPC_MAX_SLOT_TABLE);
3089 static const struct kernel_param_ops param_ops_slot_table_size = {
3090 .set = param_set_slot_table_size,
3091 .get = param_get_uint,
3094 #define param_check_slot_table_size(name, p) \
3095 __param_check(name, p, unsigned int);
3097 static int param_set_max_slot_table_size(const char *val,
3098 const struct kernel_param *kp)
3100 return param_set_uint_minmax(val, kp,
3102 RPC_MAX_SLOT_TABLE_LIMIT);
3105 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3106 .set = param_set_max_slot_table_size,
3107 .get = param_get_uint,
3110 #define param_check_max_slot_table_size(name, p) \
3111 __param_check(name, p, unsigned int);
3113 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3114 slot_table_size, 0644);
3115 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3116 max_slot_table_size, 0644);
3117 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3118 slot_table_size, 0644);