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 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_ready - "data ready" callback for AF_LOCAL sockets
913 * @sk: socket with data to read
915 * Currently this assumes we can read the whole reply in a single gulp.
917 static void xs_local_data_ready(struct sock *sk)
919 struct rpc_task *task;
920 struct rpc_xprt *xprt;
921 struct rpc_rqst *rovr;
923 int err, repsize, copied;
927 read_lock_bh(&sk->sk_callback_lock);
928 dprintk("RPC: %s...\n", __func__);
929 xprt = xprt_from_sock(sk);
933 skb = skb_recv_datagram(sk, 0, 1, &err);
937 repsize = skb->len - sizeof(rpc_fraghdr);
939 dprintk("RPC: impossible RPC reply size %d\n", repsize);
943 /* Copy the XID from the skb... */
944 xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
948 /* Look up and lock the request corresponding to the given XID */
949 spin_lock(&xprt->transport_lock);
950 rovr = xprt_lookup_rqst(xprt, *xp);
953 task = rovr->rq_task;
955 copied = rovr->rq_private_buf.buflen;
956 if (copied > repsize)
959 if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
960 dprintk("RPC: sk_buff copy failed\n");
964 xprt_complete_rqst(task, copied);
967 spin_unlock(&xprt->transport_lock);
969 skb_free_datagram(sk, skb);
971 read_unlock_bh(&sk->sk_callback_lock);
975 * xs_udp_data_ready - "data ready" callback for UDP sockets
976 * @sk: socket with data to read
979 static void xs_udp_data_ready(struct sock *sk)
981 struct rpc_task *task;
982 struct rpc_xprt *xprt;
983 struct rpc_rqst *rovr;
985 int err, repsize, copied;
989 read_lock_bh(&sk->sk_callback_lock);
990 dprintk("RPC: xs_udp_data_ready...\n");
991 if (!(xprt = xprt_from_sock(sk)))
994 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
997 repsize = skb->len - sizeof(struct udphdr);
999 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
1003 /* Copy the XID from the skb... */
1004 xp = skb_header_pointer(skb, sizeof(struct udphdr),
1005 sizeof(_xid), &_xid);
1009 /* Look up and lock the request corresponding to the given XID */
1010 spin_lock(&xprt->transport_lock);
1011 rovr = xprt_lookup_rqst(xprt, *xp);
1014 task = rovr->rq_task;
1016 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1019 /* Suck it into the iovec, verify checksum if not done by hw. */
1020 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1021 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
1025 UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
1027 xprt_adjust_cwnd(xprt, task, copied);
1028 xprt_complete_rqst(task, copied);
1031 spin_unlock(&xprt->transport_lock);
1033 skb_free_datagram(sk, skb);
1035 read_unlock_bh(&sk->sk_callback_lock);
1039 * Helper function to force a TCP close if the server is sending
1040 * junk and/or it has put us in CLOSE_WAIT
1042 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1044 xprt_force_disconnect(xprt);
1047 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1049 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1053 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1054 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1055 used = xdr_skb_read_bits(desc, p, len);
1056 transport->tcp_offset += used;
1060 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1061 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1062 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1064 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1065 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1067 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1068 transport->tcp_offset = 0;
1070 /* Sanity check of the record length */
1071 if (unlikely(transport->tcp_reclen < 8)) {
1072 dprintk("RPC: invalid TCP record fragment length\n");
1073 xs_tcp_force_close(xprt);
1076 dprintk("RPC: reading TCP record fragment of length %d\n",
1077 transport->tcp_reclen);
1080 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1082 if (transport->tcp_offset == transport->tcp_reclen) {
1083 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1084 transport->tcp_offset = 0;
1085 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1086 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1087 transport->tcp_flags |= TCP_RCV_COPY_XID;
1088 transport->tcp_copied = 0;
1093 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1098 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1099 dprintk("RPC: reading XID (%Zu bytes)\n", len);
1100 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1101 used = xdr_skb_read_bits(desc, p, len);
1102 transport->tcp_offset += used;
1105 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1106 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1107 transport->tcp_copied = 4;
1108 dprintk("RPC: reading %s XID %08x\n",
1109 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1111 ntohl(transport->tcp_xid));
1112 xs_tcp_check_fraghdr(transport);
1115 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1116 struct xdr_skb_reader *desc)
1123 * We want transport->tcp_offset to be 8 at the end of this routine
1124 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1125 * When this function is called for the first time,
1126 * transport->tcp_offset is 4 (after having already read the xid).
1128 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1129 len = sizeof(transport->tcp_calldir) - offset;
1130 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len);
1131 p = ((char *) &transport->tcp_calldir) + offset;
1132 used = xdr_skb_read_bits(desc, p, len);
1133 transport->tcp_offset += used;
1136 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1138 * We don't yet have the XDR buffer, so we will write the calldir
1139 * out after we get the buffer from the 'struct rpc_rqst'
1141 switch (ntohl(transport->tcp_calldir)) {
1143 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1144 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1145 transport->tcp_flags |= TCP_RPC_REPLY;
1148 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1149 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1150 transport->tcp_flags &= ~TCP_RPC_REPLY;
1153 dprintk("RPC: invalid request message type\n");
1154 xs_tcp_force_close(&transport->xprt);
1156 xs_tcp_check_fraghdr(transport);
1159 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1160 struct xdr_skb_reader *desc,
1161 struct rpc_rqst *req)
1163 struct sock_xprt *transport =
1164 container_of(xprt, struct sock_xprt, xprt);
1165 struct xdr_buf *rcvbuf;
1169 rcvbuf = &req->rq_private_buf;
1171 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1173 * Save the RPC direction in the XDR buffer
1175 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1176 &transport->tcp_calldir,
1177 sizeof(transport->tcp_calldir));
1178 transport->tcp_copied += sizeof(transport->tcp_calldir);
1179 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1183 if (len > transport->tcp_reclen - transport->tcp_offset) {
1184 struct xdr_skb_reader my_desc;
1186 len = transport->tcp_reclen - transport->tcp_offset;
1187 memcpy(&my_desc, desc, sizeof(my_desc));
1188 my_desc.count = len;
1189 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1190 &my_desc, xdr_skb_read_bits);
1194 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1195 desc, xdr_skb_read_bits);
1198 transport->tcp_copied += r;
1199 transport->tcp_offset += r;
1202 /* Error when copying to the receive buffer,
1203 * usually because we weren't able to allocate
1204 * additional buffer pages. All we can do now
1205 * is turn off TCP_RCV_COPY_DATA, so the request
1206 * will not receive any additional updates,
1208 * Any remaining data from this record will
1211 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1212 dprintk("RPC: XID %08x truncated request\n",
1213 ntohl(transport->tcp_xid));
1214 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1215 "tcp_offset = %u, tcp_reclen = %u\n",
1216 xprt, transport->tcp_copied,
1217 transport->tcp_offset, transport->tcp_reclen);
1221 dprintk("RPC: XID %08x read %Zd bytes\n",
1222 ntohl(transport->tcp_xid), r);
1223 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1224 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1225 transport->tcp_offset, transport->tcp_reclen);
1227 if (transport->tcp_copied == req->rq_private_buf.buflen)
1228 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1229 else if (transport->tcp_offset == transport->tcp_reclen) {
1230 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1231 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1236 * Finds the request corresponding to the RPC xid and invokes the common
1237 * tcp read code to read the data.
1239 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1240 struct xdr_skb_reader *desc)
1242 struct sock_xprt *transport =
1243 container_of(xprt, struct sock_xprt, xprt);
1244 struct rpc_rqst *req;
1246 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1248 /* Find and lock the request corresponding to this xid */
1249 spin_lock_bh(&xprt->transport_lock);
1250 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1252 dprintk("RPC: XID %08x request not found!\n",
1253 ntohl(transport->tcp_xid));
1254 spin_unlock_bh(&xprt->transport_lock);
1258 xs_tcp_read_common(xprt, desc, req);
1260 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1261 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1263 spin_unlock_bh(&xprt->transport_lock);
1267 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1269 * Obtains an rpc_rqst previously allocated and invokes the common
1270 * tcp read code to read the data. The result is placed in the callback
1272 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1273 * connection and return -1.
1275 static int xs_tcp_read_callback(struct rpc_xprt *xprt,
1276 struct xdr_skb_reader *desc)
1278 struct sock_xprt *transport =
1279 container_of(xprt, struct sock_xprt, xprt);
1280 struct rpc_rqst *req;
1282 /* Look up and lock the request corresponding to the given XID */
1283 spin_lock_bh(&xprt->transport_lock);
1284 req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
1286 spin_unlock_bh(&xprt->transport_lock);
1287 printk(KERN_WARNING "Callback slot table overflowed\n");
1288 xprt_force_disconnect(xprt);
1292 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1293 xs_tcp_read_common(xprt, desc, req);
1295 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1296 xprt_complete_bc_request(req, transport->tcp_copied);
1297 spin_unlock_bh(&xprt->transport_lock);
1302 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1303 struct xdr_skb_reader *desc)
1305 struct sock_xprt *transport =
1306 container_of(xprt, struct sock_xprt, xprt);
1308 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1309 xs_tcp_read_reply(xprt, desc) :
1310 xs_tcp_read_callback(xprt, desc);
1313 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1314 struct xdr_skb_reader *desc)
1316 return xs_tcp_read_reply(xprt, desc);
1318 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1321 * Read data off the transport. This can be either an RPC_CALL or an
1322 * RPC_REPLY. Relay the processing to helper functions.
1324 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1325 struct xdr_skb_reader *desc)
1327 struct sock_xprt *transport =
1328 container_of(xprt, struct sock_xprt, xprt);
1330 if (_xs_tcp_read_data(xprt, desc) == 0)
1331 xs_tcp_check_fraghdr(transport);
1334 * The transport_lock protects the request handling.
1335 * There's no need to hold it to update the tcp_flags.
1337 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1341 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1345 len = transport->tcp_reclen - transport->tcp_offset;
1346 if (len > desc->count)
1349 desc->offset += len;
1350 transport->tcp_offset += len;
1351 dprintk("RPC: discarded %Zu bytes\n", len);
1352 xs_tcp_check_fraghdr(transport);
1355 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1357 struct rpc_xprt *xprt = rd_desc->arg.data;
1358 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1359 struct xdr_skb_reader desc = {
1365 dprintk("RPC: xs_tcp_data_recv started\n");
1367 trace_xs_tcp_data_recv(transport);
1368 /* Read in a new fragment marker if necessary */
1369 /* Can we ever really expect to get completely empty fragments? */
1370 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1371 xs_tcp_read_fraghdr(xprt, &desc);
1374 /* Read in the xid if necessary */
1375 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1376 xs_tcp_read_xid(transport, &desc);
1379 /* Read in the call/reply flag */
1380 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1381 xs_tcp_read_calldir(transport, &desc);
1384 /* Read in the request data */
1385 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1386 xs_tcp_read_data(xprt, &desc);
1389 /* Skip over any trailing bytes on short reads */
1390 xs_tcp_read_discard(transport, &desc);
1391 } while (desc.count);
1392 trace_xs_tcp_data_recv(transport);
1393 dprintk("RPC: xs_tcp_data_recv done\n");
1394 return len - desc.count;
1397 static void xs_tcp_data_receive(struct sock_xprt *transport)
1399 struct rpc_xprt *xprt = &transport->xprt;
1401 read_descriptor_t rd_desc = {
1402 .count = 2*1024*1024,
1405 unsigned long total = 0;
1408 mutex_lock(&transport->recv_mutex);
1409 sk = transport->inet;
1413 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1416 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1421 rd_desc.count = 65536;
1424 mutex_unlock(&transport->recv_mutex);
1425 trace_xs_tcp_data_ready(xprt, read, total);
1428 static void xs_tcp_data_receive_workfn(struct work_struct *work)
1430 struct sock_xprt *transport =
1431 container_of(work, struct sock_xprt, recv_worker);
1432 xs_tcp_data_receive(transport);
1436 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1437 * @sk: socket with data to read
1440 static void xs_tcp_data_ready(struct sock *sk)
1442 struct sock_xprt *transport;
1443 struct rpc_xprt *xprt;
1445 dprintk("RPC: xs_tcp_data_ready...\n");
1447 read_lock_bh(&sk->sk_callback_lock);
1448 if (!(xprt = xprt_from_sock(sk)))
1450 transport = container_of(xprt, struct sock_xprt, xprt);
1452 /* Any data means we had a useful conversation, so
1453 * the we don't need to delay the next reconnect
1455 if (xprt->reestablish_timeout)
1456 xprt->reestablish_timeout = 0;
1457 queue_work(rpciod_workqueue, &transport->recv_worker);
1460 read_unlock_bh(&sk->sk_callback_lock);
1464 * xs_tcp_state_change - callback to handle TCP socket state changes
1465 * @sk: socket whose state has changed
1468 static void xs_tcp_state_change(struct sock *sk)
1470 struct rpc_xprt *xprt;
1471 struct sock_xprt *transport;
1473 read_lock_bh(&sk->sk_callback_lock);
1474 if (!(xprt = xprt_from_sock(sk)))
1476 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1477 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1478 sk->sk_state, xprt_connected(xprt),
1479 sock_flag(sk, SOCK_DEAD),
1480 sock_flag(sk, SOCK_ZAPPED),
1483 transport = container_of(xprt, struct sock_xprt, xprt);
1484 trace_rpc_socket_state_change(xprt, sk->sk_socket);
1485 switch (sk->sk_state) {
1486 case TCP_ESTABLISHED:
1487 spin_lock(&xprt->transport_lock);
1488 if (!xprt_test_and_set_connected(xprt)) {
1490 /* Reset TCP record info */
1491 transport->tcp_offset = 0;
1492 transport->tcp_reclen = 0;
1493 transport->tcp_copied = 0;
1494 transport->tcp_flags =
1495 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1496 xprt->connect_cookie++;
1497 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1498 xprt_clear_connecting(xprt);
1500 xprt_wake_pending_tasks(xprt, -EAGAIN);
1502 spin_unlock(&xprt->transport_lock);
1505 /* The client initiated a shutdown of the socket */
1506 xprt->connect_cookie++;
1507 xprt->reestablish_timeout = 0;
1508 set_bit(XPRT_CLOSING, &xprt->state);
1509 smp_mb__before_atomic();
1510 clear_bit(XPRT_CONNECTED, &xprt->state);
1511 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1512 smp_mb__after_atomic();
1514 case TCP_CLOSE_WAIT:
1515 /* The server initiated a shutdown of the socket */
1516 xprt->connect_cookie++;
1517 clear_bit(XPRT_CONNECTED, &xprt->state);
1518 xs_tcp_force_close(xprt);
1521 * If the server closed down the connection, make sure that
1522 * we back off before reconnecting
1524 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1525 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1528 set_bit(XPRT_CLOSING, &xprt->state);
1529 smp_mb__before_atomic();
1530 clear_bit(XPRT_CONNECTED, &xprt->state);
1531 smp_mb__after_atomic();
1534 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1535 &transport->sock_state))
1536 xprt_clear_connecting(xprt);
1537 xs_sock_mark_closed(xprt);
1540 read_unlock_bh(&sk->sk_callback_lock);
1543 static void xs_write_space(struct sock *sk)
1545 struct socket *sock;
1546 struct rpc_xprt *xprt;
1548 if (unlikely(!(sock = sk->sk_socket)))
1550 clear_bit(SOCK_NOSPACE, &sock->flags);
1552 if (unlikely(!(xprt = xprt_from_sock(sk))))
1554 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1557 xprt_write_space(xprt);
1561 * xs_udp_write_space - callback invoked when socket buffer space
1563 * @sk: socket whose state has changed
1565 * Called when more output buffer space is available for this socket.
1566 * We try not to wake our writers until they can make "significant"
1567 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1568 * with a bunch of small requests.
1570 static void xs_udp_write_space(struct sock *sk)
1572 read_lock_bh(&sk->sk_callback_lock);
1574 /* from net/core/sock.c:sock_def_write_space */
1575 if (sock_writeable(sk))
1578 read_unlock_bh(&sk->sk_callback_lock);
1582 * xs_tcp_write_space - callback invoked when socket buffer space
1584 * @sk: socket whose state has changed
1586 * Called when more output buffer space is available for this socket.
1587 * We try not to wake our writers until they can make "significant"
1588 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1589 * with a bunch of small requests.
1591 static void xs_tcp_write_space(struct sock *sk)
1593 read_lock_bh(&sk->sk_callback_lock);
1595 /* from net/core/stream.c:sk_stream_write_space */
1596 if (sk_stream_is_writeable(sk))
1599 read_unlock_bh(&sk->sk_callback_lock);
1602 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1604 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1605 struct sock *sk = transport->inet;
1607 if (transport->rcvsize) {
1608 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1609 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1611 if (transport->sndsize) {
1612 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1613 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1614 sk->sk_write_space(sk);
1619 * xs_udp_set_buffer_size - set send and receive limits
1620 * @xprt: generic transport
1621 * @sndsize: requested size of send buffer, in bytes
1622 * @rcvsize: requested size of receive buffer, in bytes
1624 * Set socket send and receive buffer size limits.
1626 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1628 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1630 transport->sndsize = 0;
1632 transport->sndsize = sndsize + 1024;
1633 transport->rcvsize = 0;
1635 transport->rcvsize = rcvsize + 1024;
1637 xs_udp_do_set_buffer_size(xprt);
1641 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1642 * @task: task that timed out
1644 * Adjust the congestion window after a retransmit timeout has occurred.
1646 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1648 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1651 static unsigned short xs_get_random_port(void)
1653 unsigned short range = xprt_max_resvport - xprt_min_resvport;
1654 unsigned short rand = (unsigned short) prandom_u32() % range;
1655 return rand + xprt_min_resvport;
1659 * xs_set_reuseaddr_port - set the socket's port and address reuse options
1662 * Note that this function has to be called on all sockets that share the
1663 * same port, and it must be called before binding.
1665 static void xs_sock_set_reuseport(struct socket *sock)
1669 kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1670 (char *)&opt, sizeof(opt));
1673 static unsigned short xs_sock_getport(struct socket *sock)
1675 struct sockaddr_storage buf;
1677 unsigned short port = 0;
1679 if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0)
1681 switch (buf.ss_family) {
1683 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1686 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1693 * xs_set_port - reset the port number in the remote endpoint address
1694 * @xprt: generic transport
1695 * @port: new port number
1698 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1700 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1702 rpc_set_port(xs_addr(xprt), port);
1703 xs_update_peer_port(xprt);
1706 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1708 if (transport->srcport == 0)
1709 transport->srcport = xs_sock_getport(sock);
1712 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1714 unsigned short port = transport->srcport;
1716 if (port == 0 && transport->xprt.resvport)
1717 port = xs_get_random_port();
1721 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1723 if (transport->srcport != 0)
1724 transport->srcport = 0;
1725 if (!transport->xprt.resvport)
1727 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1728 return xprt_max_resvport;
1731 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1733 struct sockaddr_storage myaddr;
1735 unsigned short port = xs_get_srcport(transport);
1736 unsigned short last;
1739 * If we are asking for any ephemeral port (i.e. port == 0 &&
1740 * transport->xprt.resvport == 0), don't bind. Let the local
1741 * port selection happen implicitly when the socket is used
1742 * (for example at connect time).
1744 * This ensures that we can continue to establish TCP
1745 * connections even when all local ephemeral ports are already
1746 * a part of some TCP connection. This makes no difference
1747 * for UDP sockets, but also doens't harm them.
1749 * If we're asking for any reserved port (i.e. port == 0 &&
1750 * transport->xprt.resvport == 1) xs_get_srcport above will
1751 * ensure that port is non-zero and we will bind as needed.
1756 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1758 rpc_set_port((struct sockaddr *)&myaddr, port);
1759 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1760 transport->xprt.addrlen);
1762 transport->srcport = port;
1766 port = xs_next_srcport(transport, port);
1769 } while (err == -EADDRINUSE && nloop != 2);
1771 if (myaddr.ss_family == AF_INET)
1772 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1773 &((struct sockaddr_in *)&myaddr)->sin_addr,
1774 port, err ? "failed" : "ok", err);
1776 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1777 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1778 port, err ? "failed" : "ok", err);
1783 * We don't support autobind on AF_LOCAL sockets
1785 static void xs_local_rpcbind(struct rpc_task *task)
1788 xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt));
1792 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1796 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1797 static struct lock_class_key xs_key[2];
1798 static struct lock_class_key xs_slock_key[2];
1800 static inline void xs_reclassify_socketu(struct socket *sock)
1802 struct sock *sk = sock->sk;
1804 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1805 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1808 static inline void xs_reclassify_socket4(struct socket *sock)
1810 struct sock *sk = sock->sk;
1812 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1813 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1816 static inline void xs_reclassify_socket6(struct socket *sock)
1818 struct sock *sk = sock->sk;
1820 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1821 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1824 static inline void xs_reclassify_socket(int family, struct socket *sock)
1826 WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1827 if (sock_owned_by_user(sock->sk))
1832 xs_reclassify_socketu(sock);
1835 xs_reclassify_socket4(sock);
1838 xs_reclassify_socket6(sock);
1843 static inline void xs_reclassify_socketu(struct socket *sock)
1847 static inline void xs_reclassify_socket4(struct socket *sock)
1851 static inline void xs_reclassify_socket6(struct socket *sock)
1855 static inline void xs_reclassify_socket(int family, struct socket *sock)
1860 static void xs_dummy_data_receive_workfn(struct work_struct *work)
1864 static void xs_dummy_setup_socket(struct work_struct *work)
1868 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1869 struct sock_xprt *transport, int family, int type,
1870 int protocol, bool reuseport)
1872 struct socket *sock;
1875 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1877 dprintk("RPC: can't create %d transport socket (%d).\n",
1881 xs_reclassify_socket(family, sock);
1884 xs_sock_set_reuseport(sock);
1886 err = xs_bind(transport, sock);
1894 return ERR_PTR(err);
1897 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1898 struct socket *sock)
1900 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1903 if (!transport->inet) {
1904 struct sock *sk = sock->sk;
1906 write_lock_bh(&sk->sk_callback_lock);
1908 xs_save_old_callbacks(transport, sk);
1910 sk->sk_user_data = xprt;
1911 sk->sk_data_ready = xs_local_data_ready;
1912 sk->sk_write_space = xs_udp_write_space;
1913 sk->sk_error_report = xs_error_report;
1914 sk->sk_allocation = GFP_NOIO;
1916 xprt_clear_connected(xprt);
1918 /* Reset to new socket */
1919 transport->sock = sock;
1920 transport->inet = sk;
1922 write_unlock_bh(&sk->sk_callback_lock);
1925 /* Tell the socket layer to start connecting... */
1926 xprt->stat.connect_count++;
1927 xprt->stat.connect_start = jiffies;
1928 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1932 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1933 * @transport: socket transport to connect
1935 static int xs_local_setup_socket(struct sock_xprt *transport)
1937 struct rpc_xprt *xprt = &transport->xprt;
1938 struct socket *sock;
1941 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1942 SOCK_STREAM, 0, &sock, 1);
1944 dprintk("RPC: can't create AF_LOCAL "
1945 "transport socket (%d).\n", -status);
1948 xs_reclassify_socketu(sock);
1950 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1951 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1953 status = xs_local_finish_connecting(xprt, sock);
1954 trace_rpc_socket_connect(xprt, sock, status);
1957 dprintk("RPC: xprt %p connected to %s\n",
1958 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1959 xprt_set_connected(xprt);
1963 dprintk("RPC: xprt %p: socket %s does not exist\n",
1964 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1967 dprintk("RPC: xprt %p: connection refused for %s\n",
1968 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1971 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1973 xprt->address_strings[RPC_DISPLAY_ADDR]);
1977 xprt_clear_connecting(xprt);
1978 xprt_wake_pending_tasks(xprt, status);
1982 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1984 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1987 if (RPC_IS_ASYNC(task)) {
1989 * We want the AF_LOCAL connect to be resolved in the
1990 * filesystem namespace of the process making the rpc
1991 * call. Thus we connect synchronously.
1993 * If we want to support asynchronous AF_LOCAL calls,
1994 * we'll need to figure out how to pass a namespace to
1997 rpc_exit(task, -ENOTCONN);
2000 ret = xs_local_setup_socket(transport);
2001 if (ret && !RPC_IS_SOFTCONN(task))
2002 msleep_interruptible(15000);
2005 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2007 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2008 * know that we have exclusive access to the socket), to guard against
2009 * races with xs_reset_transport.
2011 static void xs_set_memalloc(struct rpc_xprt *xprt)
2013 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2017 * If there's no sock, then we have nothing to set. The
2018 * reconnecting process will get it for us.
2020 if (!transport->inet)
2022 if (atomic_read(&xprt->swapper))
2023 sk_set_memalloc(transport->inet);
2027 * xs_enable_swap - Tag this transport as being used for swap.
2028 * @xprt: transport to tag
2030 * Take a reference to this transport on behalf of the rpc_clnt, and
2031 * optionally mark it for swapping if it wasn't already.
2034 xs_enable_swap(struct rpc_xprt *xprt)
2036 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2038 if (atomic_inc_return(&xprt->swapper) != 1)
2040 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2041 return -ERESTARTSYS;
2043 sk_set_memalloc(xs->inet);
2044 xprt_release_xprt(xprt, NULL);
2049 * xs_disable_swap - Untag this transport as being used for swap.
2050 * @xprt: transport to tag
2052 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2053 * swapper refcount goes to 0, untag the socket as a memalloc socket.
2056 xs_disable_swap(struct rpc_xprt *xprt)
2058 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2060 if (!atomic_dec_and_test(&xprt->swapper))
2062 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2065 sk_clear_memalloc(xs->inet);
2066 xprt_release_xprt(xprt, NULL);
2069 static void xs_set_memalloc(struct rpc_xprt *xprt)
2074 xs_enable_swap(struct rpc_xprt *xprt)
2080 xs_disable_swap(struct rpc_xprt *xprt)
2085 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2087 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2089 if (!transport->inet) {
2090 struct sock *sk = sock->sk;
2092 write_lock_bh(&sk->sk_callback_lock);
2094 xs_save_old_callbacks(transport, sk);
2096 sk->sk_user_data = xprt;
2097 sk->sk_data_ready = xs_udp_data_ready;
2098 sk->sk_write_space = xs_udp_write_space;
2099 sk->sk_allocation = GFP_NOIO;
2101 xprt_set_connected(xprt);
2103 /* Reset to new socket */
2104 transport->sock = sock;
2105 transport->inet = sk;
2107 xs_set_memalloc(xprt);
2109 write_unlock_bh(&sk->sk_callback_lock);
2111 xs_udp_do_set_buffer_size(xprt);
2114 static void xs_udp_setup_socket(struct work_struct *work)
2116 struct sock_xprt *transport =
2117 container_of(work, struct sock_xprt, connect_worker.work);
2118 struct rpc_xprt *xprt = &transport->xprt;
2119 struct socket *sock = transport->sock;
2122 sock = xs_create_sock(xprt, transport,
2123 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2124 IPPROTO_UDP, false);
2128 dprintk("RPC: worker connecting xprt %p via %s to "
2129 "%s (port %s)\n", xprt,
2130 xprt->address_strings[RPC_DISPLAY_PROTO],
2131 xprt->address_strings[RPC_DISPLAY_ADDR],
2132 xprt->address_strings[RPC_DISPLAY_PORT]);
2134 xs_udp_finish_connecting(xprt, sock);
2135 trace_rpc_socket_connect(xprt, sock, 0);
2138 xprt_unlock_connect(xprt, transport);
2139 xprt_clear_connecting(xprt);
2140 xprt_wake_pending_tasks(xprt, status);
2144 * xs_tcp_shutdown - gracefully shut down a TCP socket
2147 * Initiates a graceful shutdown of the TCP socket by calling the
2148 * equivalent of shutdown(SHUT_RDWR);
2150 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2152 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2153 struct socket *sock = transport->sock;
2157 if (xprt_connected(xprt)) {
2158 kernel_sock_shutdown(sock, SHUT_RDWR);
2159 trace_rpc_socket_shutdown(xprt, sock);
2161 xs_reset_transport(transport);
2164 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2166 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2167 int ret = -ENOTCONN;
2169 if (!transport->inet) {
2170 struct sock *sk = sock->sk;
2171 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2172 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2173 unsigned int opt_on = 1;
2176 /* TCP Keepalive options */
2177 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2178 (char *)&opt_on, sizeof(opt_on));
2179 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2180 (char *)&keepidle, sizeof(keepidle));
2181 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2182 (char *)&keepidle, sizeof(keepidle));
2183 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2184 (char *)&keepcnt, sizeof(keepcnt));
2186 /* TCP user timeout (see RFC5482) */
2187 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2188 (xprt->timeout->to_retries + 1);
2189 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2190 (char *)&timeo, sizeof(timeo));
2192 write_lock_bh(&sk->sk_callback_lock);
2194 xs_save_old_callbacks(transport, sk);
2196 sk->sk_user_data = xprt;
2197 sk->sk_data_ready = xs_tcp_data_ready;
2198 sk->sk_state_change = xs_tcp_state_change;
2199 sk->sk_write_space = xs_tcp_write_space;
2200 sk->sk_error_report = xs_error_report;
2201 sk->sk_allocation = GFP_NOIO;
2203 /* socket options */
2204 sock_reset_flag(sk, SOCK_LINGER);
2205 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2207 xprt_clear_connected(xprt);
2209 /* Reset to new socket */
2210 transport->sock = sock;
2211 transport->inet = sk;
2213 write_unlock_bh(&sk->sk_callback_lock);
2216 if (!xprt_bound(xprt))
2219 xs_set_memalloc(xprt);
2221 /* Tell the socket layer to start connecting... */
2222 xprt->stat.connect_count++;
2223 xprt->stat.connect_start = jiffies;
2224 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2225 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2228 xs_set_srcport(transport, sock);
2231 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2232 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2239 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2241 * Invoked by a work queue tasklet.
2243 static void xs_tcp_setup_socket(struct work_struct *work)
2245 struct sock_xprt *transport =
2246 container_of(work, struct sock_xprt, connect_worker.work);
2247 struct socket *sock = transport->sock;
2248 struct rpc_xprt *xprt = &transport->xprt;
2252 sock = xs_create_sock(xprt, transport,
2253 xs_addr(xprt)->sa_family, SOCK_STREAM,
2256 status = PTR_ERR(sock);
2261 dprintk("RPC: worker connecting xprt %p via %s to "
2262 "%s (port %s)\n", xprt,
2263 xprt->address_strings[RPC_DISPLAY_PROTO],
2264 xprt->address_strings[RPC_DISPLAY_ADDR],
2265 xprt->address_strings[RPC_DISPLAY_PORT]);
2267 status = xs_tcp_finish_connecting(xprt, sock);
2268 trace_rpc_socket_connect(xprt, sock, status);
2269 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2270 xprt, -status, xprt_connected(xprt),
2271 sock->sk->sk_state);
2274 printk("%s: connect returned unhandled error %d\n",
2276 case -EADDRNOTAVAIL:
2277 /* We're probably in TIME_WAIT. Get rid of existing socket,
2280 xs_tcp_force_close(xprt);
2285 xprt_unlock_connect(xprt, transport);
2288 /* Happens, for instance, if the user specified a link
2289 * local IPv6 address without a scope-id.
2296 /* retry with existing socket, after a delay */
2297 xs_tcp_force_close(xprt);
2302 xprt_unlock_connect(xprt, transport);
2303 xprt_clear_connecting(xprt);
2304 xprt_wake_pending_tasks(xprt, status);
2308 * xs_connect - connect a socket to a remote endpoint
2309 * @xprt: pointer to transport structure
2310 * @task: address of RPC task that manages state of connect request
2312 * TCP: If the remote end dropped the connection, delay reconnecting.
2314 * UDP socket connects are synchronous, but we use a work queue anyway
2315 * to guarantee that even unprivileged user processes can set up a
2316 * socket on a privileged port.
2318 * If a UDP socket connect fails, the delay behavior here prevents
2319 * retry floods (hard mounts).
2321 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2323 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2325 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2327 if (transport->sock != NULL) {
2328 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2330 xprt, xprt->reestablish_timeout / HZ);
2332 /* Start by resetting any existing state */
2333 xs_reset_transport(transport);
2335 queue_delayed_work(rpciod_workqueue,
2336 &transport->connect_worker,
2337 xprt->reestablish_timeout);
2338 xprt->reestablish_timeout <<= 1;
2339 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2340 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2341 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2342 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2344 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2345 queue_delayed_work(rpciod_workqueue,
2346 &transport->connect_worker, 0);
2351 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2352 * @xprt: rpc_xprt struct containing statistics
2356 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2360 if (xprt_connected(xprt))
2361 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2363 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2364 "%llu %llu %lu %llu %llu\n",
2365 xprt->stat.bind_count,
2366 xprt->stat.connect_count,
2367 xprt->stat.connect_time,
2371 xprt->stat.bad_xids,
2374 xprt->stat.max_slots,
2375 xprt->stat.sending_u,
2376 xprt->stat.pending_u);
2380 * xs_udp_print_stats - display UDP socket-specifc stats
2381 * @xprt: rpc_xprt struct containing statistics
2385 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2387 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2389 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2392 xprt->stat.bind_count,
2395 xprt->stat.bad_xids,
2398 xprt->stat.max_slots,
2399 xprt->stat.sending_u,
2400 xprt->stat.pending_u);
2404 * xs_tcp_print_stats - display TCP socket-specifc stats
2405 * @xprt: rpc_xprt struct containing statistics
2409 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2411 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2414 if (xprt_connected(xprt))
2415 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2417 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2418 "%llu %llu %lu %llu %llu\n",
2420 xprt->stat.bind_count,
2421 xprt->stat.connect_count,
2422 xprt->stat.connect_time,
2426 xprt->stat.bad_xids,
2429 xprt->stat.max_slots,
2430 xprt->stat.sending_u,
2431 xprt->stat.pending_u);
2435 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2436 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2437 * to use the server side send routines.
2439 static void *bc_malloc(struct rpc_task *task, size_t size)
2442 struct rpc_buffer *buf;
2444 WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2445 if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2448 page = alloc_page(GFP_KERNEL);
2452 buf = page_address(page);
2453 buf->len = PAGE_SIZE;
2459 * Free the space allocated in the bc_alloc routine
2461 static void bc_free(void *buffer)
2463 struct rpc_buffer *buf;
2468 buf = container_of(buffer, struct rpc_buffer, data);
2469 free_page((unsigned long)buf);
2473 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2474 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2476 static int bc_sendto(struct rpc_rqst *req)
2479 struct xdr_buf *xbufp = &req->rq_snd_buf;
2480 struct rpc_xprt *xprt = req->rq_xprt;
2481 struct sock_xprt *transport =
2482 container_of(xprt, struct sock_xprt, xprt);
2483 struct socket *sock = transport->sock;
2484 unsigned long headoff;
2485 unsigned long tailoff;
2487 xs_encode_stream_record_marker(xbufp);
2489 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2490 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2491 len = svc_send_common(sock, xbufp,
2492 virt_to_page(xbufp->head[0].iov_base), headoff,
2493 xbufp->tail[0].iov_base, tailoff);
2495 if (len != xbufp->len) {
2496 printk(KERN_NOTICE "Error sending entire callback!\n");
2504 * The send routine. Borrows from svc_send
2506 static int bc_send_request(struct rpc_task *task)
2508 struct rpc_rqst *req = task->tk_rqstp;
2509 struct svc_xprt *xprt;
2512 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2514 * Get the server socket associated with this callback xprt
2516 xprt = req->rq_xprt->bc_xprt;
2519 * Grab the mutex to serialize data as the connection is shared
2520 * with the fore channel
2522 if (!mutex_trylock(&xprt->xpt_mutex)) {
2523 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2524 if (!mutex_trylock(&xprt->xpt_mutex))
2526 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2528 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2531 len = bc_sendto(req);
2532 mutex_unlock(&xprt->xpt_mutex);
2541 * The close routine. Since this is client initiated, we do nothing
2544 static void bc_close(struct rpc_xprt *xprt)
2549 * The xprt destroy routine. Again, because this connection is client
2550 * initiated, we do nothing
2553 static void bc_destroy(struct rpc_xprt *xprt)
2555 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2558 module_put(THIS_MODULE);
2561 static struct rpc_xprt_ops xs_local_ops = {
2562 .reserve_xprt = xprt_reserve_xprt,
2563 .release_xprt = xs_tcp_release_xprt,
2564 .alloc_slot = xprt_alloc_slot,
2565 .rpcbind = xs_local_rpcbind,
2566 .set_port = xs_local_set_port,
2567 .connect = xs_local_connect,
2568 .buf_alloc = rpc_malloc,
2569 .buf_free = rpc_free,
2570 .send_request = xs_local_send_request,
2571 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2573 .destroy = xs_destroy,
2574 .print_stats = xs_local_print_stats,
2575 .enable_swap = xs_enable_swap,
2576 .disable_swap = xs_disable_swap,
2579 static struct rpc_xprt_ops xs_udp_ops = {
2580 .set_buffer_size = xs_udp_set_buffer_size,
2581 .reserve_xprt = xprt_reserve_xprt_cong,
2582 .release_xprt = xprt_release_xprt_cong,
2583 .alloc_slot = xprt_alloc_slot,
2584 .rpcbind = rpcb_getport_async,
2585 .set_port = xs_set_port,
2586 .connect = xs_connect,
2587 .buf_alloc = rpc_malloc,
2588 .buf_free = rpc_free,
2589 .send_request = xs_udp_send_request,
2590 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
2591 .timer = xs_udp_timer,
2592 .release_request = xprt_release_rqst_cong,
2594 .destroy = xs_destroy,
2595 .print_stats = xs_udp_print_stats,
2596 .enable_swap = xs_enable_swap,
2597 .disable_swap = xs_disable_swap,
2598 .inject_disconnect = xs_inject_disconnect,
2601 static struct rpc_xprt_ops xs_tcp_ops = {
2602 .reserve_xprt = xprt_reserve_xprt,
2603 .release_xprt = xs_tcp_release_xprt,
2604 .alloc_slot = xprt_lock_and_alloc_slot,
2605 .rpcbind = rpcb_getport_async,
2606 .set_port = xs_set_port,
2607 .connect = xs_connect,
2608 .buf_alloc = rpc_malloc,
2609 .buf_free = rpc_free,
2610 .send_request = xs_tcp_send_request,
2611 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2612 .close = xs_tcp_shutdown,
2613 .destroy = xs_destroy,
2614 .print_stats = xs_tcp_print_stats,
2615 .enable_swap = xs_enable_swap,
2616 .disable_swap = xs_disable_swap,
2617 .inject_disconnect = xs_inject_disconnect,
2621 * The rpc_xprt_ops for the server backchannel
2624 static struct rpc_xprt_ops bc_tcp_ops = {
2625 .reserve_xprt = xprt_reserve_xprt,
2626 .release_xprt = xprt_release_xprt,
2627 .alloc_slot = xprt_alloc_slot,
2628 .buf_alloc = bc_malloc,
2629 .buf_free = bc_free,
2630 .send_request = bc_send_request,
2631 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2633 .destroy = bc_destroy,
2634 .print_stats = xs_tcp_print_stats,
2635 .enable_swap = xs_enable_swap,
2636 .disable_swap = xs_disable_swap,
2637 .inject_disconnect = xs_inject_disconnect,
2640 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2642 static const struct sockaddr_in sin = {
2643 .sin_family = AF_INET,
2644 .sin_addr.s_addr = htonl(INADDR_ANY),
2646 static const struct sockaddr_in6 sin6 = {
2647 .sin6_family = AF_INET6,
2648 .sin6_addr = IN6ADDR_ANY_INIT,
2655 memcpy(sap, &sin, sizeof(sin));
2658 memcpy(sap, &sin6, sizeof(sin6));
2661 dprintk("RPC: %s: Bad address family\n", __func__);
2662 return -EAFNOSUPPORT;
2667 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2668 unsigned int slot_table_size,
2669 unsigned int max_slot_table_size)
2671 struct rpc_xprt *xprt;
2672 struct sock_xprt *new;
2674 if (args->addrlen > sizeof(xprt->addr)) {
2675 dprintk("RPC: xs_setup_xprt: address too large\n");
2676 return ERR_PTR(-EBADF);
2679 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2680 max_slot_table_size);
2682 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2684 return ERR_PTR(-ENOMEM);
2687 new = container_of(xprt, struct sock_xprt, xprt);
2688 mutex_init(&new->recv_mutex);
2689 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2690 xprt->addrlen = args->addrlen;
2692 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2695 err = xs_init_anyaddr(args->dstaddr->sa_family,
2696 (struct sockaddr *)&new->srcaddr);
2699 return ERR_PTR(err);
2706 static const struct rpc_timeout xs_local_default_timeout = {
2707 .to_initval = 10 * HZ,
2708 .to_maxval = 10 * HZ,
2713 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2714 * @args: rpc transport creation arguments
2716 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2718 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2720 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2721 struct sock_xprt *transport;
2722 struct rpc_xprt *xprt;
2723 struct rpc_xprt *ret;
2725 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2726 xprt_max_tcp_slot_table_entries);
2729 transport = container_of(xprt, struct sock_xprt, xprt);
2732 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2733 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2735 xprt->bind_timeout = XS_BIND_TO;
2736 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2737 xprt->idle_timeout = XS_IDLE_DISC_TO;
2739 xprt->ops = &xs_local_ops;
2740 xprt->timeout = &xs_local_default_timeout;
2742 INIT_WORK(&transport->recv_worker, xs_dummy_data_receive_workfn);
2743 INIT_DELAYED_WORK(&transport->connect_worker,
2744 xs_dummy_setup_socket);
2746 switch (sun->sun_family) {
2748 if (sun->sun_path[0] != '/') {
2749 dprintk("RPC: bad AF_LOCAL address: %s\n",
2751 ret = ERR_PTR(-EINVAL);
2754 xprt_set_bound(xprt);
2755 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2756 ret = ERR_PTR(xs_local_setup_socket(transport));
2761 ret = ERR_PTR(-EAFNOSUPPORT);
2765 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2766 xprt->address_strings[RPC_DISPLAY_ADDR]);
2768 if (try_module_get(THIS_MODULE))
2770 ret = ERR_PTR(-EINVAL);
2776 static const struct rpc_timeout xs_udp_default_timeout = {
2777 .to_initval = 5 * HZ,
2778 .to_maxval = 30 * HZ,
2779 .to_increment = 5 * HZ,
2784 * xs_setup_udp - Set up transport to use a UDP socket
2785 * @args: rpc transport creation arguments
2788 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2790 struct sockaddr *addr = args->dstaddr;
2791 struct rpc_xprt *xprt;
2792 struct sock_xprt *transport;
2793 struct rpc_xprt *ret;
2795 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2796 xprt_udp_slot_table_entries);
2799 transport = container_of(xprt, struct sock_xprt, xprt);
2801 xprt->prot = IPPROTO_UDP;
2803 /* XXX: header size can vary due to auth type, IPv6, etc. */
2804 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2806 xprt->bind_timeout = XS_BIND_TO;
2807 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2808 xprt->idle_timeout = XS_IDLE_DISC_TO;
2810 xprt->ops = &xs_udp_ops;
2812 xprt->timeout = &xs_udp_default_timeout;
2814 INIT_WORK(&transport->recv_worker, xs_dummy_data_receive_workfn);
2815 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2817 switch (addr->sa_family) {
2819 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2820 xprt_set_bound(xprt);
2822 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2825 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2826 xprt_set_bound(xprt);
2828 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2831 ret = ERR_PTR(-EAFNOSUPPORT);
2835 if (xprt_bound(xprt))
2836 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2837 xprt->address_strings[RPC_DISPLAY_ADDR],
2838 xprt->address_strings[RPC_DISPLAY_PORT],
2839 xprt->address_strings[RPC_DISPLAY_PROTO]);
2841 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2842 xprt->address_strings[RPC_DISPLAY_ADDR],
2843 xprt->address_strings[RPC_DISPLAY_PROTO]);
2845 if (try_module_get(THIS_MODULE))
2847 ret = ERR_PTR(-EINVAL);
2853 static const struct rpc_timeout xs_tcp_default_timeout = {
2854 .to_initval = 60 * HZ,
2855 .to_maxval = 60 * HZ,
2860 * xs_setup_tcp - Set up transport to use a TCP socket
2861 * @args: rpc transport creation arguments
2864 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2866 struct sockaddr *addr = args->dstaddr;
2867 struct rpc_xprt *xprt;
2868 struct sock_xprt *transport;
2869 struct rpc_xprt *ret;
2870 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2872 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2873 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2875 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2876 max_slot_table_size);
2879 transport = container_of(xprt, struct sock_xprt, xprt);
2881 xprt->prot = IPPROTO_TCP;
2882 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2883 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2885 xprt->bind_timeout = XS_BIND_TO;
2886 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2887 xprt->idle_timeout = XS_IDLE_DISC_TO;
2889 xprt->ops = &xs_tcp_ops;
2890 xprt->timeout = &xs_tcp_default_timeout;
2892 INIT_WORK(&transport->recv_worker, xs_tcp_data_receive_workfn);
2893 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
2895 switch (addr->sa_family) {
2897 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2898 xprt_set_bound(xprt);
2900 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2903 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2904 xprt_set_bound(xprt);
2906 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2909 ret = ERR_PTR(-EAFNOSUPPORT);
2913 if (xprt_bound(xprt))
2914 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2915 xprt->address_strings[RPC_DISPLAY_ADDR],
2916 xprt->address_strings[RPC_DISPLAY_PORT],
2917 xprt->address_strings[RPC_DISPLAY_PROTO]);
2919 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2920 xprt->address_strings[RPC_DISPLAY_ADDR],
2921 xprt->address_strings[RPC_DISPLAY_PROTO]);
2923 if (try_module_get(THIS_MODULE))
2925 ret = ERR_PTR(-EINVAL);
2932 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2933 * @args: rpc transport creation arguments
2936 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2938 struct sockaddr *addr = args->dstaddr;
2939 struct rpc_xprt *xprt;
2940 struct sock_xprt *transport;
2941 struct svc_sock *bc_sock;
2942 struct rpc_xprt *ret;
2944 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2945 xprt_tcp_slot_table_entries);
2948 transport = container_of(xprt, struct sock_xprt, xprt);
2950 xprt->prot = IPPROTO_TCP;
2951 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2952 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2953 xprt->timeout = &xs_tcp_default_timeout;
2956 xprt_set_bound(xprt);
2957 xprt->bind_timeout = 0;
2958 xprt->reestablish_timeout = 0;
2959 xprt->idle_timeout = 0;
2961 xprt->ops = &bc_tcp_ops;
2963 switch (addr->sa_family) {
2965 xs_format_peer_addresses(xprt, "tcp",
2969 xs_format_peer_addresses(xprt, "tcp",
2970 RPCBIND_NETID_TCP6);
2973 ret = ERR_PTR(-EAFNOSUPPORT);
2977 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2978 xprt->address_strings[RPC_DISPLAY_ADDR],
2979 xprt->address_strings[RPC_DISPLAY_PORT],
2980 xprt->address_strings[RPC_DISPLAY_PROTO]);
2983 * Once we've associated a backchannel xprt with a connection,
2984 * we want to keep it around as long as the connection lasts,
2985 * in case we need to start using it for a backchannel again;
2986 * this reference won't be dropped until bc_xprt is destroyed.
2989 args->bc_xprt->xpt_bc_xprt = xprt;
2990 xprt->bc_xprt = args->bc_xprt;
2991 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2992 transport->sock = bc_sock->sk_sock;
2993 transport->inet = bc_sock->sk_sk;
2996 * Since we don't want connections for the backchannel, we set
2997 * the xprt status to connected
2999 xprt_set_connected(xprt);
3001 if (try_module_get(THIS_MODULE))
3004 args->bc_xprt->xpt_bc_xprt = NULL;
3006 ret = ERR_PTR(-EINVAL);
3012 static struct xprt_class xs_local_transport = {
3013 .list = LIST_HEAD_INIT(xs_local_transport.list),
3014 .name = "named UNIX socket",
3015 .owner = THIS_MODULE,
3016 .ident = XPRT_TRANSPORT_LOCAL,
3017 .setup = xs_setup_local,
3020 static struct xprt_class xs_udp_transport = {
3021 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3023 .owner = THIS_MODULE,
3024 .ident = XPRT_TRANSPORT_UDP,
3025 .setup = xs_setup_udp,
3028 static struct xprt_class xs_tcp_transport = {
3029 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3031 .owner = THIS_MODULE,
3032 .ident = XPRT_TRANSPORT_TCP,
3033 .setup = xs_setup_tcp,
3036 static struct xprt_class xs_bc_tcp_transport = {
3037 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3038 .name = "tcp NFSv4.1 backchannel",
3039 .owner = THIS_MODULE,
3040 .ident = XPRT_TRANSPORT_BC_TCP,
3041 .setup = xs_setup_bc_tcp,
3045 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3048 int init_socket_xprt(void)
3050 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3051 if (!sunrpc_table_header)
3052 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3055 xprt_register_transport(&xs_local_transport);
3056 xprt_register_transport(&xs_udp_transport);
3057 xprt_register_transport(&xs_tcp_transport);
3058 xprt_register_transport(&xs_bc_tcp_transport);
3064 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3067 void cleanup_socket_xprt(void)
3069 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3070 if (sunrpc_table_header) {
3071 unregister_sysctl_table(sunrpc_table_header);
3072 sunrpc_table_header = NULL;
3076 xprt_unregister_transport(&xs_local_transport);
3077 xprt_unregister_transport(&xs_udp_transport);
3078 xprt_unregister_transport(&xs_tcp_transport);
3079 xprt_unregister_transport(&xs_bc_tcp_transport);
3082 static int param_set_uint_minmax(const char *val,
3083 const struct kernel_param *kp,
3084 unsigned int min, unsigned int max)
3091 ret = kstrtouint(val, 0, &num);
3092 if (ret == -EINVAL || num < min || num > max)
3094 *((unsigned int *)kp->arg) = num;
3098 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3100 return param_set_uint_minmax(val, kp,
3105 static const struct kernel_param_ops param_ops_portnr = {
3106 .set = param_set_portnr,
3107 .get = param_get_uint,
3110 #define param_check_portnr(name, p) \
3111 __param_check(name, p, unsigned int);
3113 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3114 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3116 static int param_set_slot_table_size(const char *val,
3117 const struct kernel_param *kp)
3119 return param_set_uint_minmax(val, kp,
3121 RPC_MAX_SLOT_TABLE);
3124 static const struct kernel_param_ops param_ops_slot_table_size = {
3125 .set = param_set_slot_table_size,
3126 .get = param_get_uint,
3129 #define param_check_slot_table_size(name, p) \
3130 __param_check(name, p, unsigned int);
3132 static int param_set_max_slot_table_size(const char *val,
3133 const struct kernel_param *kp)
3135 return param_set_uint_minmax(val, kp,
3137 RPC_MAX_SLOT_TABLE_LIMIT);
3140 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3141 .set = param_set_max_slot_table_size,
3142 .get = param_get_uint,
3145 #define param_check_max_slot_table_size(name, p) \
3146 __param_check(name, p, unsigned int);
3148 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3149 slot_table_size, 0644);
3150 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3151 max_slot_table_size, 0644);
3152 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3153 slot_table_size, 0644);