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 #define XS_TCP_LINGER_TO (15U * HZ)
67 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
70 * We can register our own files under /proc/sys/sunrpc by
71 * calling register_sysctl_table() again. The files in that
72 * directory become the union of all files registered there.
74 * We simply need to make sure that we don't collide with
75 * someone else's file names!
78 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
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);
530 if (likely(sent > 0) || status == 0) {
531 req->rq_bytes_sent += sent;
532 req->rq_xmit_bytes_sent += sent;
533 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
534 req->rq_bytes_sent = 0;
543 status = xs_nospace(task);
546 dprintk("RPC: sendmsg returned unrecognized error %d\n",
557 * xs_udp_send_request - write an RPC request to a UDP socket
558 * @task: address of RPC task that manages the state of an RPC request
561 * 0: The request has been sent
562 * EAGAIN: The socket was blocked, please call again later to
563 * complete the request
564 * ENOTCONN: Caller needs to invoke connect logic then call again
565 * other: Some other error occurred, the request was not sent
567 static int xs_udp_send_request(struct rpc_task *task)
569 struct rpc_rqst *req = task->tk_rqstp;
570 struct rpc_xprt *xprt = req->rq_xprt;
571 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
572 struct xdr_buf *xdr = &req->rq_snd_buf;
576 xs_pktdump("packet data:",
577 req->rq_svec->iov_base,
578 req->rq_svec->iov_len);
580 if (!xprt_bound(xprt))
582 status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
583 xdr, req->rq_bytes_sent, true, &sent);
585 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
586 xdr->len - req->rq_bytes_sent, status);
588 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
589 if (status == -EPERM)
592 if (sent > 0 || status == 0) {
593 req->rq_xmit_bytes_sent += sent;
594 if (sent >= req->rq_slen)
596 /* Still some bytes left; set up for a retry later. */
604 /* Should we call xs_close() here? */
607 status = xs_nospace(task);
610 dprintk("RPC: sendmsg returned unrecognized error %d\n",
617 /* When the server has died, an ICMP port unreachable message
618 * prompts ECONNREFUSED. */
619 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
626 * xs_tcp_shutdown - gracefully shut down a TCP socket
629 * Initiates a graceful shutdown of the TCP socket by calling the
630 * equivalent of shutdown(SHUT_WR);
632 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
634 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
635 struct socket *sock = transport->sock;
638 kernel_sock_shutdown(sock, SHUT_WR);
639 trace_rpc_socket_shutdown(xprt, sock);
644 * xs_tcp_send_request - write an RPC request to a TCP socket
645 * @task: address of RPC task that manages the state of an RPC request
648 * 0: The request has been sent
649 * EAGAIN: The socket was blocked, please call again later to
650 * complete the request
651 * ENOTCONN: Caller needs to invoke connect logic then call again
652 * other: Some other error occurred, the request was not sent
654 * XXX: In the case of soft timeouts, should we eventually give up
655 * if sendmsg is not able to make progress?
657 static int xs_tcp_send_request(struct rpc_task *task)
659 struct rpc_rqst *req = task->tk_rqstp;
660 struct rpc_xprt *xprt = req->rq_xprt;
661 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
662 struct xdr_buf *xdr = &req->rq_snd_buf;
663 bool zerocopy = true;
667 xs_encode_stream_record_marker(&req->rq_snd_buf);
669 xs_pktdump("packet data:",
670 req->rq_svec->iov_base,
671 req->rq_svec->iov_len);
672 /* Don't use zero copy if this is a resend. If the RPC call
673 * completes while the socket holds a reference to the pages,
674 * then we may end up resending corrupted data.
676 if (task->tk_flags & RPC_TASK_SENT)
679 /* Continue transmitting the packet/record. We must be careful
680 * to cope with writespace callbacks arriving _after_ we have
681 * called sendmsg(). */
684 status = xs_sendpages(transport->sock, NULL, 0, xdr,
685 req->rq_bytes_sent, zerocopy, &sent);
687 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
688 xdr->len - req->rq_bytes_sent, status);
690 if (unlikely(sent == 0 && status < 0))
693 /* If we've sent the entire packet, immediately
694 * reset the count of bytes sent. */
695 req->rq_bytes_sent += sent;
696 req->rq_xmit_bytes_sent += sent;
697 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
698 req->rq_bytes_sent = 0;
711 /* Should we call xs_close() here? */
715 status = xs_nospace(task);
718 dprintk("RPC: sendmsg returned unrecognized error %d\n",
721 xs_tcp_shutdown(xprt);
725 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
732 * xs_tcp_release_xprt - clean up after a tcp transmission
736 * This cleans up if an error causes us to abort the transmission of a request.
737 * In this case, the socket may need to be reset in order to avoid confusing
740 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
742 struct rpc_rqst *req;
744 if (task != xprt->snd_task)
748 req = task->tk_rqstp;
751 if (req->rq_bytes_sent == 0)
753 if (req->rq_bytes_sent == req->rq_snd_buf.len)
755 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
757 xprt_release_xprt(xprt, task);
760 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
762 transport->old_data_ready = sk->sk_data_ready;
763 transport->old_state_change = sk->sk_state_change;
764 transport->old_write_space = sk->sk_write_space;
765 transport->old_error_report = sk->sk_error_report;
768 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
770 sk->sk_data_ready = transport->old_data_ready;
771 sk->sk_state_change = transport->old_state_change;
772 sk->sk_write_space = transport->old_write_space;
773 sk->sk_error_report = transport->old_error_report;
777 * xs_error_report - callback to handle TCP socket state errors
780 * Note: we don't call sock_error() since there may be a rpc_task
781 * using the socket, and so we don't want to clear sk->sk_err.
783 static void xs_error_report(struct sock *sk)
785 struct rpc_xprt *xprt;
788 read_lock_bh(&sk->sk_callback_lock);
789 if (!(xprt = xprt_from_sock(sk)))
795 dprintk("RPC: xs_error_report client %p, error=%d...\n",
797 trace_rpc_socket_error(xprt, sk->sk_socket, err);
798 if (test_bit(XPRT_CONNECTION_REUSE, &xprt->state))
800 xprt_wake_pending_tasks(xprt, err);
802 read_unlock_bh(&sk->sk_callback_lock);
805 static void xs_reset_transport(struct sock_xprt *transport)
807 struct socket *sock = transport->sock;
808 struct sock *sk = transport->inet;
813 transport->srcport = 0;
815 write_lock_bh(&sk->sk_callback_lock);
816 transport->inet = NULL;
817 transport->sock = NULL;
819 sk->sk_user_data = NULL;
821 xs_restore_old_callbacks(transport, sk);
822 write_unlock_bh(&sk->sk_callback_lock);
824 trace_rpc_socket_close(&transport->xprt, sock);
829 * xs_close - close a socket
832 * This is used when all requests are complete; ie, no DRC state remains
833 * on the server we want to save.
835 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
836 * xs_reset_transport() zeroing the socket from underneath a writer.
838 static void xs_close(struct rpc_xprt *xprt)
840 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
842 dprintk("RPC: xs_close xprt %p\n", xprt);
844 cancel_delayed_work_sync(&transport->connect_worker);
846 xs_reset_transport(transport);
847 xprt->reestablish_timeout = 0;
849 smp_mb__before_atomic();
850 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
851 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
852 clear_bit(XPRT_CLOSING, &xprt->state);
853 smp_mb__after_atomic();
854 xprt_disconnect_done(xprt);
857 static void xs_tcp_close(struct rpc_xprt *xprt)
859 if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state))
862 xs_tcp_shutdown(xprt);
865 static void xs_xprt_free(struct rpc_xprt *xprt)
867 xs_free_peer_addresses(xprt);
872 * xs_destroy - prepare to shutdown a transport
873 * @xprt: doomed transport
876 static void xs_destroy(struct rpc_xprt *xprt)
878 dprintk("RPC: xs_destroy xprt %p\n", xprt);
882 module_put(THIS_MODULE);
885 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
887 struct xdr_skb_reader desc = {
889 .offset = sizeof(rpc_fraghdr),
890 .count = skb->len - sizeof(rpc_fraghdr),
893 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
901 * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets
902 * @sk: socket with data to read
903 * @len: how much data to read
905 * Currently this assumes we can read the whole reply in a single gulp.
907 static void xs_local_data_ready(struct sock *sk)
909 struct rpc_task *task;
910 struct rpc_xprt *xprt;
911 struct rpc_rqst *rovr;
913 int err, repsize, copied;
917 read_lock_bh(&sk->sk_callback_lock);
918 dprintk("RPC: %s...\n", __func__);
919 xprt = xprt_from_sock(sk);
923 skb = skb_recv_datagram(sk, 0, 1, &err);
927 repsize = skb->len - sizeof(rpc_fraghdr);
929 dprintk("RPC: impossible RPC reply size %d\n", repsize);
933 /* Copy the XID from the skb... */
934 xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
938 /* Look up and lock the request corresponding to the given XID */
939 spin_lock(&xprt->transport_lock);
940 rovr = xprt_lookup_rqst(xprt, *xp);
943 task = rovr->rq_task;
945 copied = rovr->rq_private_buf.buflen;
946 if (copied > repsize)
949 if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
950 dprintk("RPC: sk_buff copy failed\n");
954 xprt_complete_rqst(task, copied);
957 spin_unlock(&xprt->transport_lock);
959 skb_free_datagram(sk, skb);
961 read_unlock_bh(&sk->sk_callback_lock);
965 * xs_udp_data_ready - "data ready" callback for UDP sockets
966 * @sk: socket with data to read
967 * @len: how much data to read
970 static void xs_udp_data_ready(struct sock *sk)
972 struct rpc_task *task;
973 struct rpc_xprt *xprt;
974 struct rpc_rqst *rovr;
976 int err, repsize, copied;
980 read_lock_bh(&sk->sk_callback_lock);
981 dprintk("RPC: xs_udp_data_ready...\n");
982 if (!(xprt = xprt_from_sock(sk)))
985 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
988 repsize = skb->len - sizeof(struct udphdr);
990 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
994 /* Copy the XID from the skb... */
995 xp = skb_header_pointer(skb, sizeof(struct udphdr),
996 sizeof(_xid), &_xid);
1000 /* Look up and lock the request corresponding to the given XID */
1001 spin_lock(&xprt->transport_lock);
1002 rovr = xprt_lookup_rqst(xprt, *xp);
1005 task = rovr->rq_task;
1007 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1010 /* Suck it into the iovec, verify checksum if not done by hw. */
1011 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1012 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
1016 UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
1018 xprt_adjust_cwnd(xprt, task, copied);
1019 xprt_complete_rqst(task, copied);
1022 spin_unlock(&xprt->transport_lock);
1024 skb_free_datagram(sk, skb);
1026 read_unlock_bh(&sk->sk_callback_lock);
1030 * Helper function to force a TCP close if the server is sending
1031 * junk and/or it has put us in CLOSE_WAIT
1033 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1035 set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1036 xprt_force_disconnect(xprt);
1039 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1041 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1045 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1046 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1047 used = xdr_skb_read_bits(desc, p, len);
1048 transport->tcp_offset += used;
1052 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1053 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1054 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1056 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1057 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1059 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1060 transport->tcp_offset = 0;
1062 /* Sanity check of the record length */
1063 if (unlikely(transport->tcp_reclen < 8)) {
1064 dprintk("RPC: invalid TCP record fragment length\n");
1065 xs_tcp_force_close(xprt);
1068 dprintk("RPC: reading TCP record fragment of length %d\n",
1069 transport->tcp_reclen);
1072 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1074 if (transport->tcp_offset == transport->tcp_reclen) {
1075 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1076 transport->tcp_offset = 0;
1077 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1078 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1079 transport->tcp_flags |= TCP_RCV_COPY_XID;
1080 transport->tcp_copied = 0;
1085 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1090 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1091 dprintk("RPC: reading XID (%Zu bytes)\n", len);
1092 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1093 used = xdr_skb_read_bits(desc, p, len);
1094 transport->tcp_offset += used;
1097 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1098 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1099 transport->tcp_copied = 4;
1100 dprintk("RPC: reading %s XID %08x\n",
1101 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1103 ntohl(transport->tcp_xid));
1104 xs_tcp_check_fraghdr(transport);
1107 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1108 struct xdr_skb_reader *desc)
1115 * We want transport->tcp_offset to be 8 at the end of this routine
1116 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1117 * When this function is called for the first time,
1118 * transport->tcp_offset is 4 (after having already read the xid).
1120 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1121 len = sizeof(transport->tcp_calldir) - offset;
1122 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len);
1123 p = ((char *) &transport->tcp_calldir) + offset;
1124 used = xdr_skb_read_bits(desc, p, len);
1125 transport->tcp_offset += used;
1128 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1130 * We don't yet have the XDR buffer, so we will write the calldir
1131 * out after we get the buffer from the 'struct rpc_rqst'
1133 switch (ntohl(transport->tcp_calldir)) {
1135 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1136 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1137 transport->tcp_flags |= TCP_RPC_REPLY;
1140 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1141 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1142 transport->tcp_flags &= ~TCP_RPC_REPLY;
1145 dprintk("RPC: invalid request message type\n");
1146 xs_tcp_force_close(&transport->xprt);
1148 xs_tcp_check_fraghdr(transport);
1151 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1152 struct xdr_skb_reader *desc,
1153 struct rpc_rqst *req)
1155 struct sock_xprt *transport =
1156 container_of(xprt, struct sock_xprt, xprt);
1157 struct xdr_buf *rcvbuf;
1161 rcvbuf = &req->rq_private_buf;
1163 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1165 * Save the RPC direction in the XDR buffer
1167 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1168 &transport->tcp_calldir,
1169 sizeof(transport->tcp_calldir));
1170 transport->tcp_copied += sizeof(transport->tcp_calldir);
1171 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1175 if (len > transport->tcp_reclen - transport->tcp_offset) {
1176 struct xdr_skb_reader my_desc;
1178 len = transport->tcp_reclen - transport->tcp_offset;
1179 memcpy(&my_desc, desc, sizeof(my_desc));
1180 my_desc.count = len;
1181 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1182 &my_desc, xdr_skb_read_bits);
1186 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1187 desc, xdr_skb_read_bits);
1190 transport->tcp_copied += r;
1191 transport->tcp_offset += r;
1194 /* Error when copying to the receive buffer,
1195 * usually because we weren't able to allocate
1196 * additional buffer pages. All we can do now
1197 * is turn off TCP_RCV_COPY_DATA, so the request
1198 * will not receive any additional updates,
1200 * Any remaining data from this record will
1203 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1204 dprintk("RPC: XID %08x truncated request\n",
1205 ntohl(transport->tcp_xid));
1206 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1207 "tcp_offset = %u, tcp_reclen = %u\n",
1208 xprt, transport->tcp_copied,
1209 transport->tcp_offset, transport->tcp_reclen);
1213 dprintk("RPC: XID %08x read %Zd bytes\n",
1214 ntohl(transport->tcp_xid), r);
1215 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1216 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1217 transport->tcp_offset, transport->tcp_reclen);
1219 if (transport->tcp_copied == req->rq_private_buf.buflen)
1220 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1221 else if (transport->tcp_offset == transport->tcp_reclen) {
1222 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1223 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1228 * Finds the request corresponding to the RPC xid and invokes the common
1229 * tcp read code to read the data.
1231 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1232 struct xdr_skb_reader *desc)
1234 struct sock_xprt *transport =
1235 container_of(xprt, struct sock_xprt, xprt);
1236 struct rpc_rqst *req;
1238 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1240 /* Find and lock the request corresponding to this xid */
1241 spin_lock(&xprt->transport_lock);
1242 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1244 dprintk("RPC: XID %08x request not found!\n",
1245 ntohl(transport->tcp_xid));
1246 spin_unlock(&xprt->transport_lock);
1250 xs_tcp_read_common(xprt, desc, req);
1252 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1253 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1255 spin_unlock(&xprt->transport_lock);
1259 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1261 * Obtains an rpc_rqst previously allocated and invokes the common
1262 * tcp read code to read the data. The result is placed in the callback
1264 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1265 * connection and return -1.
1267 static int xs_tcp_read_callback(struct rpc_xprt *xprt,
1268 struct xdr_skb_reader *desc)
1270 struct sock_xprt *transport =
1271 container_of(xprt, struct sock_xprt, xprt);
1272 struct rpc_rqst *req;
1274 /* Look up and lock the request corresponding to the given XID */
1275 spin_lock(&xprt->transport_lock);
1276 req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
1278 spin_unlock(&xprt->transport_lock);
1279 printk(KERN_WARNING "Callback slot table overflowed\n");
1280 xprt_force_disconnect(xprt);
1284 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1285 xs_tcp_read_common(xprt, desc, req);
1287 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1288 xprt_complete_bc_request(req, transport->tcp_copied);
1289 spin_unlock(&xprt->transport_lock);
1294 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1295 struct xdr_skb_reader *desc)
1297 struct sock_xprt *transport =
1298 container_of(xprt, struct sock_xprt, xprt);
1300 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1301 xs_tcp_read_reply(xprt, desc) :
1302 xs_tcp_read_callback(xprt, desc);
1305 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1306 struct xdr_skb_reader *desc)
1308 return xs_tcp_read_reply(xprt, desc);
1310 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1313 * Read data off the transport. This can be either an RPC_CALL or an
1314 * RPC_REPLY. Relay the processing to helper functions.
1316 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1317 struct xdr_skb_reader *desc)
1319 struct sock_xprt *transport =
1320 container_of(xprt, struct sock_xprt, xprt);
1322 if (_xs_tcp_read_data(xprt, desc) == 0)
1323 xs_tcp_check_fraghdr(transport);
1326 * The transport_lock protects the request handling.
1327 * There's no need to hold it to update the tcp_flags.
1329 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1333 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1337 len = transport->tcp_reclen - transport->tcp_offset;
1338 if (len > desc->count)
1341 desc->offset += len;
1342 transport->tcp_offset += len;
1343 dprintk("RPC: discarded %Zu bytes\n", len);
1344 xs_tcp_check_fraghdr(transport);
1347 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1349 struct rpc_xprt *xprt = rd_desc->arg.data;
1350 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1351 struct xdr_skb_reader desc = {
1357 dprintk("RPC: xs_tcp_data_recv started\n");
1359 trace_xs_tcp_data_recv(transport);
1360 /* Read in a new fragment marker if necessary */
1361 /* Can we ever really expect to get completely empty fragments? */
1362 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1363 xs_tcp_read_fraghdr(xprt, &desc);
1366 /* Read in the xid if necessary */
1367 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1368 xs_tcp_read_xid(transport, &desc);
1371 /* Read in the call/reply flag */
1372 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1373 xs_tcp_read_calldir(transport, &desc);
1376 /* Read in the request data */
1377 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1378 xs_tcp_read_data(xprt, &desc);
1381 /* Skip over any trailing bytes on short reads */
1382 xs_tcp_read_discard(transport, &desc);
1383 } while (desc.count);
1384 trace_xs_tcp_data_recv(transport);
1385 dprintk("RPC: xs_tcp_data_recv done\n");
1386 return len - desc.count;
1390 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1391 * @sk: socket with data to read
1392 * @bytes: how much data to read
1395 static void xs_tcp_data_ready(struct sock *sk)
1397 struct rpc_xprt *xprt;
1398 read_descriptor_t rd_desc;
1400 unsigned long total = 0;
1402 dprintk("RPC: xs_tcp_data_ready...\n");
1404 read_lock_bh(&sk->sk_callback_lock);
1405 if (!(xprt = xprt_from_sock(sk))) {
1409 /* Any data means we had a useful conversation, so
1410 * the we don't need to delay the next reconnect
1412 if (xprt->reestablish_timeout)
1413 xprt->reestablish_timeout = 0;
1415 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1416 rd_desc.arg.data = xprt;
1418 rd_desc.count = 65536;
1419 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1424 trace_xs_tcp_data_ready(xprt, read, total);
1425 read_unlock_bh(&sk->sk_callback_lock);
1429 * Do the equivalent of linger/linger2 handling for dealing with
1430 * broken servers that don't close the socket in a timely
1433 static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1434 unsigned long timeout)
1436 struct sock_xprt *transport;
1438 if (xprt_test_and_set_connecting(xprt))
1440 set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1441 transport = container_of(xprt, struct sock_xprt, xprt);
1442 queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1446 static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1448 struct sock_xprt *transport;
1450 transport = container_of(xprt, struct sock_xprt, xprt);
1452 if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1453 !cancel_delayed_work(&transport->connect_worker))
1455 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1456 xprt_clear_connecting(xprt);
1459 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1461 smp_mb__before_atomic();
1462 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1463 clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1464 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1465 clear_bit(XPRT_CLOSING, &xprt->state);
1466 smp_mb__after_atomic();
1469 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1471 xs_sock_reset_connection_flags(xprt);
1472 /* Mark transport as closed and wake up all pending tasks */
1473 xprt_disconnect_done(xprt);
1477 * xs_tcp_state_change - callback to handle TCP socket state changes
1478 * @sk: socket whose state has changed
1481 static void xs_tcp_state_change(struct sock *sk)
1483 struct rpc_xprt *xprt;
1485 read_lock_bh(&sk->sk_callback_lock);
1486 if (!(xprt = xprt_from_sock(sk)))
1488 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1489 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1490 sk->sk_state, xprt_connected(xprt),
1491 sock_flag(sk, SOCK_DEAD),
1492 sock_flag(sk, SOCK_ZAPPED),
1495 trace_rpc_socket_state_change(xprt, sk->sk_socket);
1496 switch (sk->sk_state) {
1497 case TCP_ESTABLISHED:
1498 spin_lock(&xprt->transport_lock);
1499 if (!xprt_test_and_set_connected(xprt)) {
1500 struct sock_xprt *transport = container_of(xprt,
1501 struct sock_xprt, xprt);
1503 /* Reset TCP record info */
1504 transport->tcp_offset = 0;
1505 transport->tcp_reclen = 0;
1506 transport->tcp_copied = 0;
1507 transport->tcp_flags =
1508 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1509 xprt->connect_cookie++;
1511 xprt_wake_pending_tasks(xprt, -EAGAIN);
1513 spin_unlock(&xprt->transport_lock);
1516 /* The client initiated a shutdown of the socket */
1517 xprt->connect_cookie++;
1518 xprt->reestablish_timeout = 0;
1519 set_bit(XPRT_CLOSING, &xprt->state);
1520 smp_mb__before_atomic();
1521 clear_bit(XPRT_CONNECTED, &xprt->state);
1522 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1523 smp_mb__after_atomic();
1524 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1526 case TCP_CLOSE_WAIT:
1527 /* The server initiated a shutdown of the socket */
1528 xprt->connect_cookie++;
1529 clear_bit(XPRT_CONNECTED, &xprt->state);
1530 xs_tcp_force_close(xprt);
1533 * If the server closed down the connection, make sure that
1534 * we back off before reconnecting
1536 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1537 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1540 set_bit(XPRT_CLOSING, &xprt->state);
1541 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1542 smp_mb__before_atomic();
1543 clear_bit(XPRT_CONNECTED, &xprt->state);
1544 smp_mb__after_atomic();
1547 xs_tcp_cancel_linger_timeout(xprt);
1548 xs_sock_mark_closed(xprt);
1551 read_unlock_bh(&sk->sk_callback_lock);
1554 static void xs_write_space(struct sock *sk)
1556 struct socket *sock;
1557 struct rpc_xprt *xprt;
1559 if (unlikely(!(sock = sk->sk_socket)))
1561 clear_bit(SOCK_NOSPACE, &sock->flags);
1563 if (unlikely(!(xprt = xprt_from_sock(sk))))
1565 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1568 xprt_write_space(xprt);
1572 * xs_udp_write_space - callback invoked when socket buffer space
1574 * @sk: socket whose state has changed
1576 * Called when more output buffer space is available for this socket.
1577 * We try not to wake our writers until they can make "significant"
1578 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1579 * with a bunch of small requests.
1581 static void xs_udp_write_space(struct sock *sk)
1583 read_lock_bh(&sk->sk_callback_lock);
1585 /* from net/core/sock.c:sock_def_write_space */
1586 if (sock_writeable(sk))
1589 read_unlock_bh(&sk->sk_callback_lock);
1593 * xs_tcp_write_space - callback invoked when socket buffer space
1595 * @sk: socket whose state has changed
1597 * Called when more output buffer space is available for this socket.
1598 * We try not to wake our writers until they can make "significant"
1599 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1600 * with a bunch of small requests.
1602 static void xs_tcp_write_space(struct sock *sk)
1604 read_lock_bh(&sk->sk_callback_lock);
1606 /* from net/core/stream.c:sk_stream_write_space */
1607 if (sk_stream_is_writeable(sk))
1610 read_unlock_bh(&sk->sk_callback_lock);
1613 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1615 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1616 struct sock *sk = transport->inet;
1618 if (transport->rcvsize) {
1619 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1620 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1622 if (transport->sndsize) {
1623 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1624 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1625 sk->sk_write_space(sk);
1630 * xs_udp_set_buffer_size - set send and receive limits
1631 * @xprt: generic transport
1632 * @sndsize: requested size of send buffer, in bytes
1633 * @rcvsize: requested size of receive buffer, in bytes
1635 * Set socket send and receive buffer size limits.
1637 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1639 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1641 transport->sndsize = 0;
1643 transport->sndsize = sndsize + 1024;
1644 transport->rcvsize = 0;
1646 transport->rcvsize = rcvsize + 1024;
1648 xs_udp_do_set_buffer_size(xprt);
1652 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1653 * @task: task that timed out
1655 * Adjust the congestion window after a retransmit timeout has occurred.
1657 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1659 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1662 static unsigned short xs_get_random_port(void)
1664 unsigned short range = xprt_max_resvport - xprt_min_resvport;
1665 unsigned short rand = (unsigned short) prandom_u32() % range;
1666 return rand + xprt_min_resvport;
1670 * xs_set_port - reset the port number in the remote endpoint address
1671 * @xprt: generic transport
1672 * @port: new port number
1675 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1677 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1679 rpc_set_port(xs_addr(xprt), port);
1680 xs_update_peer_port(xprt);
1683 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1685 unsigned short port = transport->srcport;
1687 if (port == 0 && transport->xprt.resvport)
1688 port = xs_get_random_port();
1692 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1694 if (transport->srcport != 0)
1695 transport->srcport = 0;
1696 if (!transport->xprt.resvport)
1698 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1699 return xprt_max_resvport;
1702 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1704 struct sockaddr_storage myaddr;
1706 unsigned short port = xs_get_srcport(transport);
1707 unsigned short last;
1710 * If we are asking for any ephemeral port (i.e. port == 0 &&
1711 * transport->xprt.resvport == 0), don't bind. Let the local
1712 * port selection happen implicitly when the socket is used
1713 * (for example at connect time).
1715 * This ensures that we can continue to establish TCP
1716 * connections even when all local ephemeral ports are already
1717 * a part of some TCP connection. This makes no difference
1718 * for UDP sockets, but also doens't harm them.
1720 * If we're asking for any reserved port (i.e. port == 0 &&
1721 * transport->xprt.resvport == 1) xs_get_srcport above will
1722 * ensure that port is non-zero and we will bind as needed.
1727 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1729 rpc_set_port((struct sockaddr *)&myaddr, port);
1730 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1731 transport->xprt.addrlen);
1733 transport->srcport = port;
1737 port = xs_next_srcport(transport, port);
1740 } while (err == -EADDRINUSE && nloop != 2);
1742 if (myaddr.ss_family == AF_INET)
1743 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1744 &((struct sockaddr_in *)&myaddr)->sin_addr,
1745 port, err ? "failed" : "ok", err);
1747 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1748 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1749 port, err ? "failed" : "ok", err);
1754 * We don't support autobind on AF_LOCAL sockets
1756 static void xs_local_rpcbind(struct rpc_task *task)
1759 xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt));
1763 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1767 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1768 static struct lock_class_key xs_key[2];
1769 static struct lock_class_key xs_slock_key[2];
1771 static inline void xs_reclassify_socketu(struct socket *sock)
1773 struct sock *sk = sock->sk;
1775 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1776 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1779 static inline void xs_reclassify_socket4(struct socket *sock)
1781 struct sock *sk = sock->sk;
1783 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1784 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1787 static inline void xs_reclassify_socket6(struct socket *sock)
1789 struct sock *sk = sock->sk;
1791 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1792 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1795 static inline void xs_reclassify_socket(int family, struct socket *sock)
1797 WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1798 if (sock_owned_by_user(sock->sk))
1803 xs_reclassify_socketu(sock);
1806 xs_reclassify_socket4(sock);
1809 xs_reclassify_socket6(sock);
1814 static inline void xs_reclassify_socketu(struct socket *sock)
1818 static inline void xs_reclassify_socket4(struct socket *sock)
1822 static inline void xs_reclassify_socket6(struct socket *sock)
1826 static inline void xs_reclassify_socket(int family, struct socket *sock)
1831 static void xs_dummy_setup_socket(struct work_struct *work)
1835 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1836 struct sock_xprt *transport, int family, int type, int protocol)
1838 struct socket *sock;
1841 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1843 dprintk("RPC: can't create %d transport socket (%d).\n",
1847 xs_reclassify_socket(family, sock);
1849 err = xs_bind(transport, sock);
1857 return ERR_PTR(err);
1860 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1861 struct socket *sock)
1863 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1866 if (!transport->inet) {
1867 struct sock *sk = sock->sk;
1869 write_lock_bh(&sk->sk_callback_lock);
1871 xs_save_old_callbacks(transport, sk);
1873 sk->sk_user_data = xprt;
1874 sk->sk_data_ready = xs_local_data_ready;
1875 sk->sk_write_space = xs_udp_write_space;
1876 sk->sk_error_report = xs_error_report;
1877 sk->sk_allocation = GFP_ATOMIC;
1879 xprt_clear_connected(xprt);
1881 /* Reset to new socket */
1882 transport->sock = sock;
1883 transport->inet = sk;
1885 write_unlock_bh(&sk->sk_callback_lock);
1888 /* Tell the socket layer to start connecting... */
1889 xprt->stat.connect_count++;
1890 xprt->stat.connect_start = jiffies;
1891 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1895 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1896 * @xprt: RPC transport to connect
1897 * @transport: socket transport to connect
1898 * @create_sock: function to create a socket of the correct type
1900 static int xs_local_setup_socket(struct sock_xprt *transport)
1902 struct rpc_xprt *xprt = &transport->xprt;
1903 struct socket *sock;
1906 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1907 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1908 SOCK_STREAM, 0, &sock, 1);
1910 dprintk("RPC: can't create AF_LOCAL "
1911 "transport socket (%d).\n", -status);
1914 xs_reclassify_socketu(sock);
1916 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1917 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1919 status = xs_local_finish_connecting(xprt, sock);
1920 trace_rpc_socket_connect(xprt, sock, status);
1923 dprintk("RPC: xprt %p connected to %s\n",
1924 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1925 xprt_set_connected(xprt);
1929 dprintk("RPC: xprt %p: socket %s does not exist\n",
1930 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1933 dprintk("RPC: xprt %p: connection refused for %s\n",
1934 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1937 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1939 xprt->address_strings[RPC_DISPLAY_ADDR]);
1943 xprt_clear_connecting(xprt);
1944 xprt_wake_pending_tasks(xprt, status);
1948 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1950 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1953 if (RPC_IS_ASYNC(task)) {
1955 * We want the AF_LOCAL connect to be resolved in the
1956 * filesystem namespace of the process making the rpc
1957 * call. Thus we connect synchronously.
1959 * If we want to support asynchronous AF_LOCAL calls,
1960 * we'll need to figure out how to pass a namespace to
1963 rpc_exit(task, -ENOTCONN);
1966 ret = xs_local_setup_socket(transport);
1967 if (ret && !RPC_IS_SOFTCONN(task))
1968 msleep_interruptible(15000);
1971 #ifdef CONFIG_SUNRPC_SWAP
1972 static void xs_set_memalloc(struct rpc_xprt *xprt)
1974 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1978 sk_set_memalloc(transport->inet);
1982 * xs_swapper - Tag this transport as being used for swap.
1983 * @xprt: transport to tag
1984 * @enable: enable/disable
1987 int xs_swapper(struct rpc_xprt *xprt, int enable)
1989 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1995 xs_set_memalloc(xprt);
1996 } else if (xprt->swapper) {
1998 sk_clear_memalloc(transport->inet);
2003 EXPORT_SYMBOL_GPL(xs_swapper);
2005 static void xs_set_memalloc(struct rpc_xprt *xprt)
2010 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2012 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2014 if (!transport->inet) {
2015 struct sock *sk = sock->sk;
2017 write_lock_bh(&sk->sk_callback_lock);
2019 xs_save_old_callbacks(transport, sk);
2021 sk->sk_user_data = xprt;
2022 sk->sk_data_ready = xs_udp_data_ready;
2023 sk->sk_write_space = xs_udp_write_space;
2024 sk->sk_allocation = GFP_ATOMIC;
2026 xprt_set_connected(xprt);
2028 /* Reset to new socket */
2029 transport->sock = sock;
2030 transport->inet = sk;
2032 xs_set_memalloc(xprt);
2034 write_unlock_bh(&sk->sk_callback_lock);
2036 xs_udp_do_set_buffer_size(xprt);
2039 static void xs_udp_setup_socket(struct work_struct *work)
2041 struct sock_xprt *transport =
2042 container_of(work, struct sock_xprt, connect_worker.work);
2043 struct rpc_xprt *xprt = &transport->xprt;
2044 struct socket *sock = transport->sock;
2047 /* Start by resetting any existing state */
2048 xs_reset_transport(transport);
2049 sock = xs_create_sock(xprt, transport,
2050 xs_addr(xprt)->sa_family, SOCK_DGRAM, IPPROTO_UDP);
2054 dprintk("RPC: worker connecting xprt %p via %s to "
2055 "%s (port %s)\n", xprt,
2056 xprt->address_strings[RPC_DISPLAY_PROTO],
2057 xprt->address_strings[RPC_DISPLAY_ADDR],
2058 xprt->address_strings[RPC_DISPLAY_PORT]);
2060 xs_udp_finish_connecting(xprt, sock);
2061 trace_rpc_socket_connect(xprt, sock, 0);
2064 xprt_clear_connecting(xprt);
2065 xprt_wake_pending_tasks(xprt, status);
2069 * We need to preserve the port number so the reply cache on the server can
2070 * find our cached RPC replies when we get around to reconnecting.
2072 static void xs_abort_connection(struct sock_xprt *transport)
2075 struct sockaddr any;
2077 dprintk("RPC: disconnecting xprt %p to reuse port\n", transport);
2080 * Disconnect the transport socket by doing a connect operation
2081 * with AF_UNSPEC. This should return immediately...
2083 memset(&any, 0, sizeof(any));
2084 any.sa_family = AF_UNSPEC;
2085 result = kernel_connect(transport->sock, &any, sizeof(any), 0);
2086 trace_rpc_socket_reset_connection(&transport->xprt,
2087 transport->sock, result);
2089 xs_sock_reset_connection_flags(&transport->xprt);
2090 dprintk("RPC: AF_UNSPEC connect return code %d\n", result);
2093 static void xs_tcp_reuse_connection(struct sock_xprt *transport)
2095 unsigned int state = transport->inet->sk_state;
2097 if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED) {
2098 /* we don't need to abort the connection if the socket
2099 * hasn't undergone a shutdown
2101 if (transport->inet->sk_shutdown == 0)
2103 dprintk("RPC: %s: TCP_CLOSEd and sk_shutdown set to %d\n",
2104 __func__, transport->inet->sk_shutdown);
2106 if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT)) {
2107 /* we don't need to abort the connection if the socket
2108 * hasn't undergone a shutdown
2110 if (transport->inet->sk_shutdown == 0)
2112 dprintk("RPC: %s: ESTABLISHED/SYN_SENT "
2113 "sk_shutdown set to %d\n",
2114 __func__, transport->inet->sk_shutdown);
2116 xs_abort_connection(transport);
2119 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2121 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2122 int ret = -ENOTCONN;
2124 if (!transport->inet) {
2125 struct sock *sk = sock->sk;
2126 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2127 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2128 unsigned int opt_on = 1;
2130 /* TCP Keepalive options */
2131 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2132 (char *)&opt_on, sizeof(opt_on));
2133 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2134 (char *)&keepidle, sizeof(keepidle));
2135 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2136 (char *)&keepidle, sizeof(keepidle));
2137 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2138 (char *)&keepcnt, sizeof(keepcnt));
2140 write_lock_bh(&sk->sk_callback_lock);
2142 xs_save_old_callbacks(transport, sk);
2144 sk->sk_user_data = xprt;
2145 sk->sk_data_ready = xs_tcp_data_ready;
2146 sk->sk_state_change = xs_tcp_state_change;
2147 sk->sk_write_space = xs_tcp_write_space;
2148 sk->sk_error_report = xs_error_report;
2149 sk->sk_allocation = GFP_ATOMIC;
2151 /* socket options */
2152 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
2153 sock_reset_flag(sk, SOCK_LINGER);
2154 tcp_sk(sk)->linger2 = 0;
2155 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2157 xprt_clear_connected(xprt);
2159 /* Reset to new socket */
2160 transport->sock = sock;
2161 transport->inet = sk;
2163 write_unlock_bh(&sk->sk_callback_lock);
2166 if (!xprt_bound(xprt))
2169 xs_set_memalloc(xprt);
2171 /* Tell the socket layer to start connecting... */
2172 xprt->stat.connect_count++;
2173 xprt->stat.connect_start = jiffies;
2174 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2179 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2180 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2187 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2188 * @xprt: RPC transport to connect
2189 * @transport: socket transport to connect
2190 * @create_sock: function to create a socket of the correct type
2192 * Invoked by a work queue tasklet.
2194 static void xs_tcp_setup_socket(struct work_struct *work)
2196 struct sock_xprt *transport =
2197 container_of(work, struct sock_xprt, connect_worker.work);
2198 struct socket *sock = transport->sock;
2199 struct rpc_xprt *xprt = &transport->xprt;
2203 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
2204 sock = xs_create_sock(xprt, transport,
2205 xs_addr(xprt)->sa_family, SOCK_STREAM, IPPROTO_TCP);
2207 status = PTR_ERR(sock);
2213 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
2215 /* "close" the socket, preserving the local port */
2216 set_bit(XPRT_CONNECTION_REUSE, &xprt->state);
2217 xs_tcp_reuse_connection(transport);
2218 clear_bit(XPRT_CONNECTION_REUSE, &xprt->state);
2224 dprintk("RPC: worker connecting xprt %p via %s to "
2225 "%s (port %s)\n", xprt,
2226 xprt->address_strings[RPC_DISPLAY_PROTO],
2227 xprt->address_strings[RPC_DISPLAY_ADDR],
2228 xprt->address_strings[RPC_DISPLAY_PORT]);
2230 status = xs_tcp_finish_connecting(xprt, sock);
2231 trace_rpc_socket_connect(xprt, sock, status);
2232 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2233 xprt, -status, xprt_connected(xprt),
2234 sock->sk->sk_state);
2237 printk("%s: connect returned unhandled error %d\n",
2239 case -EADDRNOTAVAIL:
2240 /* We're probably in TIME_WAIT. Get rid of existing socket,
2243 xs_tcp_force_close(xprt);
2248 xprt_clear_connecting(xprt);
2251 /* Happens, for instance, if the user specified a link
2252 * local IPv6 address without a scope-id.
2258 /* retry with existing socket, after a delay */
2264 xprt_clear_connecting(xprt);
2265 xprt_wake_pending_tasks(xprt, status);
2269 * xs_connect - connect a socket to a remote endpoint
2270 * @xprt: pointer to transport structure
2271 * @task: address of RPC task that manages state of connect request
2273 * TCP: If the remote end dropped the connection, delay reconnecting.
2275 * UDP socket connects are synchronous, but we use a work queue anyway
2276 * to guarantee that even unprivileged user processes can set up a
2277 * socket on a privileged port.
2279 * If a UDP socket connect fails, the delay behavior here prevents
2280 * retry floods (hard mounts).
2282 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2284 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2286 if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
2287 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2289 xprt, xprt->reestablish_timeout / HZ);
2290 queue_delayed_work(rpciod_workqueue,
2291 &transport->connect_worker,
2292 xprt->reestablish_timeout);
2293 xprt->reestablish_timeout <<= 1;
2294 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2295 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2296 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2297 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2299 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2300 queue_delayed_work(rpciod_workqueue,
2301 &transport->connect_worker, 0);
2306 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2307 * @xprt: rpc_xprt struct containing statistics
2311 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2315 if (xprt_connected(xprt))
2316 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2318 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2319 "%llu %llu %lu %llu %llu\n",
2320 xprt->stat.bind_count,
2321 xprt->stat.connect_count,
2322 xprt->stat.connect_time,
2326 xprt->stat.bad_xids,
2329 xprt->stat.max_slots,
2330 xprt->stat.sending_u,
2331 xprt->stat.pending_u);
2335 * xs_udp_print_stats - display UDP socket-specifc stats
2336 * @xprt: rpc_xprt struct containing statistics
2340 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2342 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2344 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2347 xprt->stat.bind_count,
2350 xprt->stat.bad_xids,
2353 xprt->stat.max_slots,
2354 xprt->stat.sending_u,
2355 xprt->stat.pending_u);
2359 * xs_tcp_print_stats - display TCP socket-specifc stats
2360 * @xprt: rpc_xprt struct containing statistics
2364 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2366 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2369 if (xprt_connected(xprt))
2370 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2372 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2373 "%llu %llu %lu %llu %llu\n",
2375 xprt->stat.bind_count,
2376 xprt->stat.connect_count,
2377 xprt->stat.connect_time,
2381 xprt->stat.bad_xids,
2384 xprt->stat.max_slots,
2385 xprt->stat.sending_u,
2386 xprt->stat.pending_u);
2390 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2391 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2392 * to use the server side send routines.
2394 static void *bc_malloc(struct rpc_task *task, size_t size)
2397 struct rpc_buffer *buf;
2399 WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2400 if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2403 page = alloc_page(GFP_KERNEL);
2407 buf = page_address(page);
2408 buf->len = PAGE_SIZE;
2414 * Free the space allocated in the bc_alloc routine
2416 static void bc_free(void *buffer)
2418 struct rpc_buffer *buf;
2423 buf = container_of(buffer, struct rpc_buffer, data);
2424 free_page((unsigned long)buf);
2428 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2429 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2431 static int bc_sendto(struct rpc_rqst *req)
2434 struct xdr_buf *xbufp = &req->rq_snd_buf;
2435 struct rpc_xprt *xprt = req->rq_xprt;
2436 struct sock_xprt *transport =
2437 container_of(xprt, struct sock_xprt, xprt);
2438 struct socket *sock = transport->sock;
2439 unsigned long headoff;
2440 unsigned long tailoff;
2442 xs_encode_stream_record_marker(xbufp);
2444 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2445 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2446 len = svc_send_common(sock, xbufp,
2447 virt_to_page(xbufp->head[0].iov_base), headoff,
2448 xbufp->tail[0].iov_base, tailoff);
2450 if (len != xbufp->len) {
2451 printk(KERN_NOTICE "Error sending entire callback!\n");
2459 * The send routine. Borrows from svc_send
2461 static int bc_send_request(struct rpc_task *task)
2463 struct rpc_rqst *req = task->tk_rqstp;
2464 struct svc_xprt *xprt;
2467 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2469 * Get the server socket associated with this callback xprt
2471 xprt = req->rq_xprt->bc_xprt;
2474 * Grab the mutex to serialize data as the connection is shared
2475 * with the fore channel
2477 if (!mutex_trylock(&xprt->xpt_mutex)) {
2478 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2479 if (!mutex_trylock(&xprt->xpt_mutex))
2481 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2483 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2486 len = bc_sendto(req);
2487 mutex_unlock(&xprt->xpt_mutex);
2496 * The close routine. Since this is client initiated, we do nothing
2499 static void bc_close(struct rpc_xprt *xprt)
2504 * The xprt destroy routine. Again, because this connection is client
2505 * initiated, we do nothing
2508 static void bc_destroy(struct rpc_xprt *xprt)
2510 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2513 module_put(THIS_MODULE);
2516 static struct rpc_xprt_ops xs_local_ops = {
2517 .reserve_xprt = xprt_reserve_xprt,
2518 .release_xprt = xs_tcp_release_xprt,
2519 .alloc_slot = xprt_alloc_slot,
2520 .rpcbind = xs_local_rpcbind,
2521 .set_port = xs_local_set_port,
2522 .connect = xs_local_connect,
2523 .buf_alloc = rpc_malloc,
2524 .buf_free = rpc_free,
2525 .send_request = xs_local_send_request,
2526 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2528 .destroy = xs_destroy,
2529 .print_stats = xs_local_print_stats,
2532 static struct rpc_xprt_ops xs_udp_ops = {
2533 .set_buffer_size = xs_udp_set_buffer_size,
2534 .reserve_xprt = xprt_reserve_xprt_cong,
2535 .release_xprt = xprt_release_xprt_cong,
2536 .alloc_slot = xprt_alloc_slot,
2537 .rpcbind = rpcb_getport_async,
2538 .set_port = xs_set_port,
2539 .connect = xs_connect,
2540 .buf_alloc = rpc_malloc,
2541 .buf_free = rpc_free,
2542 .send_request = xs_udp_send_request,
2543 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
2544 .timer = xs_udp_timer,
2545 .release_request = xprt_release_rqst_cong,
2547 .destroy = xs_destroy,
2548 .print_stats = xs_udp_print_stats,
2551 static struct rpc_xprt_ops xs_tcp_ops = {
2552 .reserve_xprt = xprt_reserve_xprt,
2553 .release_xprt = xs_tcp_release_xprt,
2554 .alloc_slot = xprt_lock_and_alloc_slot,
2555 .rpcbind = rpcb_getport_async,
2556 .set_port = xs_set_port,
2557 .connect = xs_connect,
2558 .buf_alloc = rpc_malloc,
2559 .buf_free = rpc_free,
2560 .send_request = xs_tcp_send_request,
2561 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2562 .close = xs_tcp_close,
2563 .destroy = xs_destroy,
2564 .print_stats = xs_tcp_print_stats,
2568 * The rpc_xprt_ops for the server backchannel
2571 static struct rpc_xprt_ops bc_tcp_ops = {
2572 .reserve_xprt = xprt_reserve_xprt,
2573 .release_xprt = xprt_release_xprt,
2574 .alloc_slot = xprt_alloc_slot,
2575 .buf_alloc = bc_malloc,
2576 .buf_free = bc_free,
2577 .send_request = bc_send_request,
2578 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2580 .destroy = bc_destroy,
2581 .print_stats = xs_tcp_print_stats,
2584 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2586 static const struct sockaddr_in sin = {
2587 .sin_family = AF_INET,
2588 .sin_addr.s_addr = htonl(INADDR_ANY),
2590 static const struct sockaddr_in6 sin6 = {
2591 .sin6_family = AF_INET6,
2592 .sin6_addr = IN6ADDR_ANY_INIT,
2599 memcpy(sap, &sin, sizeof(sin));
2602 memcpy(sap, &sin6, sizeof(sin6));
2605 dprintk("RPC: %s: Bad address family\n", __func__);
2606 return -EAFNOSUPPORT;
2611 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2612 unsigned int slot_table_size,
2613 unsigned int max_slot_table_size)
2615 struct rpc_xprt *xprt;
2616 struct sock_xprt *new;
2618 if (args->addrlen > sizeof(xprt->addr)) {
2619 dprintk("RPC: xs_setup_xprt: address too large\n");
2620 return ERR_PTR(-EBADF);
2623 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2624 max_slot_table_size);
2626 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2628 return ERR_PTR(-ENOMEM);
2631 new = container_of(xprt, struct sock_xprt, xprt);
2632 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2633 xprt->addrlen = args->addrlen;
2635 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2638 err = xs_init_anyaddr(args->dstaddr->sa_family,
2639 (struct sockaddr *)&new->srcaddr);
2642 return ERR_PTR(err);
2649 static const struct rpc_timeout xs_local_default_timeout = {
2650 .to_initval = 10 * HZ,
2651 .to_maxval = 10 * HZ,
2656 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2657 * @args: rpc transport creation arguments
2659 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2661 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2663 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2664 struct sock_xprt *transport;
2665 struct rpc_xprt *xprt;
2666 struct rpc_xprt *ret;
2668 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2669 xprt_max_tcp_slot_table_entries);
2672 transport = container_of(xprt, struct sock_xprt, xprt);
2675 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2676 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2678 xprt->bind_timeout = XS_BIND_TO;
2679 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2680 xprt->idle_timeout = XS_IDLE_DISC_TO;
2682 xprt->ops = &xs_local_ops;
2683 xprt->timeout = &xs_local_default_timeout;
2685 INIT_DELAYED_WORK(&transport->connect_worker,
2686 xs_dummy_setup_socket);
2688 switch (sun->sun_family) {
2690 if (sun->sun_path[0] != '/') {
2691 dprintk("RPC: bad AF_LOCAL address: %s\n",
2693 ret = ERR_PTR(-EINVAL);
2696 xprt_set_bound(xprt);
2697 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2698 ret = ERR_PTR(xs_local_setup_socket(transport));
2703 ret = ERR_PTR(-EAFNOSUPPORT);
2707 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2708 xprt->address_strings[RPC_DISPLAY_ADDR]);
2710 if (try_module_get(THIS_MODULE))
2712 ret = ERR_PTR(-EINVAL);
2718 static const struct rpc_timeout xs_udp_default_timeout = {
2719 .to_initval = 5 * HZ,
2720 .to_maxval = 30 * HZ,
2721 .to_increment = 5 * HZ,
2726 * xs_setup_udp - Set up transport to use a UDP socket
2727 * @args: rpc transport creation arguments
2730 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2732 struct sockaddr *addr = args->dstaddr;
2733 struct rpc_xprt *xprt;
2734 struct sock_xprt *transport;
2735 struct rpc_xprt *ret;
2737 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2738 xprt_udp_slot_table_entries);
2741 transport = container_of(xprt, struct sock_xprt, xprt);
2743 xprt->prot = IPPROTO_UDP;
2745 /* XXX: header size can vary due to auth type, IPv6, etc. */
2746 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2748 xprt->bind_timeout = XS_BIND_TO;
2749 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2750 xprt->idle_timeout = XS_IDLE_DISC_TO;
2752 xprt->ops = &xs_udp_ops;
2754 xprt->timeout = &xs_udp_default_timeout;
2756 switch (addr->sa_family) {
2758 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2759 xprt_set_bound(xprt);
2761 INIT_DELAYED_WORK(&transport->connect_worker,
2762 xs_udp_setup_socket);
2763 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2766 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2767 xprt_set_bound(xprt);
2769 INIT_DELAYED_WORK(&transport->connect_worker,
2770 xs_udp_setup_socket);
2771 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2774 ret = ERR_PTR(-EAFNOSUPPORT);
2778 if (xprt_bound(xprt))
2779 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2780 xprt->address_strings[RPC_DISPLAY_ADDR],
2781 xprt->address_strings[RPC_DISPLAY_PORT],
2782 xprt->address_strings[RPC_DISPLAY_PROTO]);
2784 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2785 xprt->address_strings[RPC_DISPLAY_ADDR],
2786 xprt->address_strings[RPC_DISPLAY_PROTO]);
2788 if (try_module_get(THIS_MODULE))
2790 ret = ERR_PTR(-EINVAL);
2796 static const struct rpc_timeout xs_tcp_default_timeout = {
2797 .to_initval = 60 * HZ,
2798 .to_maxval = 60 * HZ,
2803 * xs_setup_tcp - Set up transport to use a TCP socket
2804 * @args: rpc transport creation arguments
2807 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2809 struct sockaddr *addr = args->dstaddr;
2810 struct rpc_xprt *xprt;
2811 struct sock_xprt *transport;
2812 struct rpc_xprt *ret;
2813 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2815 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2816 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2818 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2819 max_slot_table_size);
2822 transport = container_of(xprt, struct sock_xprt, xprt);
2824 xprt->prot = IPPROTO_TCP;
2825 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2826 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2828 xprt->bind_timeout = XS_BIND_TO;
2829 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2830 xprt->idle_timeout = XS_IDLE_DISC_TO;
2832 xprt->ops = &xs_tcp_ops;
2833 xprt->timeout = &xs_tcp_default_timeout;
2835 switch (addr->sa_family) {
2837 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2838 xprt_set_bound(xprt);
2840 INIT_DELAYED_WORK(&transport->connect_worker,
2841 xs_tcp_setup_socket);
2842 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2845 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2846 xprt_set_bound(xprt);
2848 INIT_DELAYED_WORK(&transport->connect_worker,
2849 xs_tcp_setup_socket);
2850 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2853 ret = ERR_PTR(-EAFNOSUPPORT);
2857 if (xprt_bound(xprt))
2858 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2859 xprt->address_strings[RPC_DISPLAY_ADDR],
2860 xprt->address_strings[RPC_DISPLAY_PORT],
2861 xprt->address_strings[RPC_DISPLAY_PROTO]);
2863 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2864 xprt->address_strings[RPC_DISPLAY_ADDR],
2865 xprt->address_strings[RPC_DISPLAY_PROTO]);
2867 if (try_module_get(THIS_MODULE))
2869 ret = ERR_PTR(-EINVAL);
2876 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2877 * @args: rpc transport creation arguments
2880 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2882 struct sockaddr *addr = args->dstaddr;
2883 struct rpc_xprt *xprt;
2884 struct sock_xprt *transport;
2885 struct svc_sock *bc_sock;
2886 struct rpc_xprt *ret;
2888 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2889 xprt_tcp_slot_table_entries);
2892 transport = container_of(xprt, struct sock_xprt, xprt);
2894 xprt->prot = IPPROTO_TCP;
2895 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2896 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2897 xprt->timeout = &xs_tcp_default_timeout;
2900 xprt_set_bound(xprt);
2901 xprt->bind_timeout = 0;
2902 xprt->reestablish_timeout = 0;
2903 xprt->idle_timeout = 0;
2905 xprt->ops = &bc_tcp_ops;
2907 switch (addr->sa_family) {
2909 xs_format_peer_addresses(xprt, "tcp",
2913 xs_format_peer_addresses(xprt, "tcp",
2914 RPCBIND_NETID_TCP6);
2917 ret = ERR_PTR(-EAFNOSUPPORT);
2921 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2922 xprt->address_strings[RPC_DISPLAY_ADDR],
2923 xprt->address_strings[RPC_DISPLAY_PORT],
2924 xprt->address_strings[RPC_DISPLAY_PROTO]);
2927 * Once we've associated a backchannel xprt with a connection,
2928 * we want to keep it around as long as the connection lasts,
2929 * in case we need to start using it for a backchannel again;
2930 * this reference won't be dropped until bc_xprt is destroyed.
2933 args->bc_xprt->xpt_bc_xprt = xprt;
2934 xprt->bc_xprt = args->bc_xprt;
2935 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2936 transport->sock = bc_sock->sk_sock;
2937 transport->inet = bc_sock->sk_sk;
2940 * Since we don't want connections for the backchannel, we set
2941 * the xprt status to connected
2943 xprt_set_connected(xprt);
2945 if (try_module_get(THIS_MODULE))
2948 args->bc_xprt->xpt_bc_xprt = NULL;
2950 ret = ERR_PTR(-EINVAL);
2956 static struct xprt_class xs_local_transport = {
2957 .list = LIST_HEAD_INIT(xs_local_transport.list),
2958 .name = "named UNIX socket",
2959 .owner = THIS_MODULE,
2960 .ident = XPRT_TRANSPORT_LOCAL,
2961 .setup = xs_setup_local,
2964 static struct xprt_class xs_udp_transport = {
2965 .list = LIST_HEAD_INIT(xs_udp_transport.list),
2967 .owner = THIS_MODULE,
2968 .ident = XPRT_TRANSPORT_UDP,
2969 .setup = xs_setup_udp,
2972 static struct xprt_class xs_tcp_transport = {
2973 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
2975 .owner = THIS_MODULE,
2976 .ident = XPRT_TRANSPORT_TCP,
2977 .setup = xs_setup_tcp,
2980 static struct xprt_class xs_bc_tcp_transport = {
2981 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
2982 .name = "tcp NFSv4.1 backchannel",
2983 .owner = THIS_MODULE,
2984 .ident = XPRT_TRANSPORT_BC_TCP,
2985 .setup = xs_setup_bc_tcp,
2989 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2992 int init_socket_xprt(void)
2994 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
2995 if (!sunrpc_table_header)
2996 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2999 xprt_register_transport(&xs_local_transport);
3000 xprt_register_transport(&xs_udp_transport);
3001 xprt_register_transport(&xs_tcp_transport);
3002 xprt_register_transport(&xs_bc_tcp_transport);
3008 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3011 void cleanup_socket_xprt(void)
3013 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3014 if (sunrpc_table_header) {
3015 unregister_sysctl_table(sunrpc_table_header);
3016 sunrpc_table_header = NULL;
3020 xprt_unregister_transport(&xs_local_transport);
3021 xprt_unregister_transport(&xs_udp_transport);
3022 xprt_unregister_transport(&xs_tcp_transport);
3023 xprt_unregister_transport(&xs_bc_tcp_transport);
3026 static int param_set_uint_minmax(const char *val,
3027 const struct kernel_param *kp,
3028 unsigned int min, unsigned int max)
3035 ret = kstrtouint(val, 0, &num);
3036 if (ret == -EINVAL || num < min || num > max)
3038 *((unsigned int *)kp->arg) = num;
3042 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3044 return param_set_uint_minmax(val, kp,
3049 static struct kernel_param_ops param_ops_portnr = {
3050 .set = param_set_portnr,
3051 .get = param_get_uint,
3054 #define param_check_portnr(name, p) \
3055 __param_check(name, p, unsigned int);
3057 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3058 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3060 static int param_set_slot_table_size(const char *val,
3061 const struct kernel_param *kp)
3063 return param_set_uint_minmax(val, kp,
3065 RPC_MAX_SLOT_TABLE);
3068 static struct kernel_param_ops param_ops_slot_table_size = {
3069 .set = param_set_slot_table_size,
3070 .get = param_get_uint,
3073 #define param_check_slot_table_size(name, p) \
3074 __param_check(name, p, unsigned int);
3076 static int param_set_max_slot_table_size(const char *val,
3077 const struct kernel_param *kp)
3079 return param_set_uint_minmax(val, kp,
3081 RPC_MAX_SLOT_TABLE_LIMIT);
3084 static struct kernel_param_ops param_ops_max_slot_table_size = {
3085 .set = param_set_max_slot_table_size,
3086 .get = param_get_uint,
3089 #define param_check_max_slot_table_size(name, p) \
3090 __param_check(name, p, unsigned int);
3092 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3093 slot_table_size, 0644);
3094 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3095 max_slot_table_size, 0644);
3096 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3097 slot_table_size, 0644);