PCI: fix pci_resource_alignment prototype
[zen-stable.git] / net / sunrpc / xprtsock.c
blobb6309db5622689deaf1c76a6acb1f6a09a02602c
1 /*
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/slab.h>
23 #include <linux/module.h>
24 #include <linux/capability.h>
25 #include <linux/pagemap.h>
26 #include <linux/errno.h>
27 #include <linux/socket.h>
28 #include <linux/in.h>
29 #include <linux/net.h>
30 #include <linux/mm.h>
31 #include <linux/udp.h>
32 #include <linux/tcp.h>
33 #include <linux/sunrpc/clnt.h>
34 #include <linux/sunrpc/sched.h>
35 #include <linux/sunrpc/svcsock.h>
36 #include <linux/sunrpc/xprtsock.h>
37 #include <linux/file.h>
38 #ifdef CONFIG_NFS_V4_1
39 #include <linux/sunrpc/bc_xprt.h>
40 #endif
42 #include <net/sock.h>
43 #include <net/checksum.h>
44 #include <net/udp.h>
45 #include <net/tcp.h>
47 #include "sunrpc.h"
49 * xprtsock tunables
51 unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
52 unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE;
54 unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
55 unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
57 #define XS_TCP_LINGER_TO (15U * HZ)
58 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
61 * We can register our own files under /proc/sys/sunrpc by
62 * calling register_sysctl_table() again. The files in that
63 * directory become the union of all files registered there.
65 * We simply need to make sure that we don't collide with
66 * someone else's file names!
69 #ifdef RPC_DEBUG
71 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
72 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
73 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
74 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
76 static struct ctl_table_header *sunrpc_table_header;
79 * FIXME: changing the UDP slot table size should also resize the UDP
80 * socket buffers for existing UDP transports
82 static ctl_table xs_tunables_table[] = {
84 .procname = "udp_slot_table_entries",
85 .data = &xprt_udp_slot_table_entries,
86 .maxlen = sizeof(unsigned int),
87 .mode = 0644,
88 .proc_handler = proc_dointvec_minmax,
89 .extra1 = &min_slot_table_size,
90 .extra2 = &max_slot_table_size
93 .procname = "tcp_slot_table_entries",
94 .data = &xprt_tcp_slot_table_entries,
95 .maxlen = sizeof(unsigned int),
96 .mode = 0644,
97 .proc_handler = proc_dointvec_minmax,
98 .extra1 = &min_slot_table_size,
99 .extra2 = &max_slot_table_size
102 .procname = "min_resvport",
103 .data = &xprt_min_resvport,
104 .maxlen = sizeof(unsigned int),
105 .mode = 0644,
106 .proc_handler = proc_dointvec_minmax,
107 .extra1 = &xprt_min_resvport_limit,
108 .extra2 = &xprt_max_resvport_limit
111 .procname = "max_resvport",
112 .data = &xprt_max_resvport,
113 .maxlen = sizeof(unsigned int),
114 .mode = 0644,
115 .proc_handler = proc_dointvec_minmax,
116 .extra1 = &xprt_min_resvport_limit,
117 .extra2 = &xprt_max_resvport_limit
120 .procname = "tcp_fin_timeout",
121 .data = &xs_tcp_fin_timeout,
122 .maxlen = sizeof(xs_tcp_fin_timeout),
123 .mode = 0644,
124 .proc_handler = proc_dointvec_jiffies,
126 { },
129 static ctl_table sunrpc_table[] = {
131 .procname = "sunrpc",
132 .mode = 0555,
133 .child = xs_tunables_table
135 { },
138 #endif
141 * Wait duration for a reply from the RPC portmapper.
143 #define XS_BIND_TO (60U * HZ)
146 * Delay if a UDP socket connect error occurs. This is most likely some
147 * kind of resource problem on the local host.
149 #define XS_UDP_REEST_TO (2U * HZ)
152 * The reestablish timeout allows clients to delay for a bit before attempting
153 * to reconnect to a server that just dropped our connection.
155 * We implement an exponential backoff when trying to reestablish a TCP
156 * transport connection with the server. Some servers like to drop a TCP
157 * connection when they are overworked, so we start with a short timeout and
158 * increase over time if the server is down or not responding.
160 #define XS_TCP_INIT_REEST_TO (3U * HZ)
161 #define XS_TCP_MAX_REEST_TO (5U * 60 * HZ)
164 * TCP idle timeout; client drops the transport socket if it is idle
165 * for this long. Note that we also timeout UDP sockets to prevent
166 * holding port numbers when there is no RPC traffic.
168 #define XS_IDLE_DISC_TO (5U * 60 * HZ)
170 #ifdef RPC_DEBUG
171 # undef RPC_DEBUG_DATA
172 # define RPCDBG_FACILITY RPCDBG_TRANS
173 #endif
175 #ifdef RPC_DEBUG_DATA
176 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
178 u8 *buf = (u8 *) packet;
179 int j;
181 dprintk("RPC: %s\n", msg);
182 for (j = 0; j < count && j < 128; j += 4) {
183 if (!(j & 31)) {
184 if (j)
185 dprintk("\n");
186 dprintk("0x%04x ", j);
188 dprintk("%02x%02x%02x%02x ",
189 buf[j], buf[j+1], buf[j+2], buf[j+3]);
191 dprintk("\n");
193 #else
194 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
196 /* NOP */
198 #endif
200 struct sock_xprt {
201 struct rpc_xprt xprt;
204 * Network layer
206 struct socket * sock;
207 struct sock * inet;
210 * State of TCP reply receive
212 __be32 tcp_fraghdr,
213 tcp_xid,
214 tcp_calldir;
216 u32 tcp_offset,
217 tcp_reclen;
219 unsigned long tcp_copied,
220 tcp_flags;
223 * Connection of transports
225 struct delayed_work connect_worker;
226 struct sockaddr_storage srcaddr;
227 unsigned short srcport;
230 * UDP socket buffer size parameters
232 size_t rcvsize,
233 sndsize;
236 * Saved socket callback addresses
238 void (*old_data_ready)(struct sock *, int);
239 void (*old_state_change)(struct sock *);
240 void (*old_write_space)(struct sock *);
241 void (*old_error_report)(struct sock *);
245 * TCP receive state flags
247 #define TCP_RCV_LAST_FRAG (1UL << 0)
248 #define TCP_RCV_COPY_FRAGHDR (1UL << 1)
249 #define TCP_RCV_COPY_XID (1UL << 2)
250 #define TCP_RCV_COPY_DATA (1UL << 3)
251 #define TCP_RCV_READ_CALLDIR (1UL << 4)
252 #define TCP_RCV_COPY_CALLDIR (1UL << 5)
255 * TCP RPC flags
257 #define TCP_RPC_REPLY (1UL << 6)
259 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
261 return (struct sockaddr *) &xprt->addr;
264 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
266 return (struct sockaddr_in *) &xprt->addr;
269 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
271 return (struct sockaddr_in6 *) &xprt->addr;
274 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
276 struct sockaddr *sap = xs_addr(xprt);
277 struct sockaddr_in6 *sin6;
278 struct sockaddr_in *sin;
279 char buf[128];
281 (void)rpc_ntop(sap, buf, sizeof(buf));
282 xprt->address_strings[RPC_DISPLAY_ADDR] = kstrdup(buf, GFP_KERNEL);
284 switch (sap->sa_family) {
285 case AF_INET:
286 sin = xs_addr_in(xprt);
287 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
288 break;
289 case AF_INET6:
290 sin6 = xs_addr_in6(xprt);
291 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
292 break;
293 default:
294 BUG();
296 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
299 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
301 struct sockaddr *sap = xs_addr(xprt);
302 char buf[128];
304 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
305 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
307 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
308 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
311 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
312 const char *protocol,
313 const char *netid)
315 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
316 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
317 xs_format_common_peer_addresses(xprt);
318 xs_format_common_peer_ports(xprt);
321 static void xs_update_peer_port(struct rpc_xprt *xprt)
323 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
324 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
326 xs_format_common_peer_ports(xprt);
329 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
331 unsigned int i;
333 for (i = 0; i < RPC_DISPLAY_MAX; i++)
334 switch (i) {
335 case RPC_DISPLAY_PROTO:
336 case RPC_DISPLAY_NETID:
337 continue;
338 default:
339 kfree(xprt->address_strings[i]);
343 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
345 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
347 struct msghdr msg = {
348 .msg_name = addr,
349 .msg_namelen = addrlen,
350 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
352 struct kvec iov = {
353 .iov_base = vec->iov_base + base,
354 .iov_len = vec->iov_len - base,
357 if (iov.iov_len != 0)
358 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
359 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
362 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
364 struct page **ppage;
365 unsigned int remainder;
366 int err, sent = 0;
368 remainder = xdr->page_len - base;
369 base += xdr->page_base;
370 ppage = xdr->pages + (base >> PAGE_SHIFT);
371 base &= ~PAGE_MASK;
372 for(;;) {
373 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
374 int flags = XS_SENDMSG_FLAGS;
376 remainder -= len;
377 if (remainder != 0 || more)
378 flags |= MSG_MORE;
379 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
380 if (remainder == 0 || err != len)
381 break;
382 sent += err;
383 ppage++;
384 base = 0;
386 if (sent == 0)
387 return err;
388 if (err > 0)
389 sent += err;
390 return sent;
394 * xs_sendpages - write pages directly to a socket
395 * @sock: socket to send on
396 * @addr: UDP only -- address of destination
397 * @addrlen: UDP only -- length of destination address
398 * @xdr: buffer containing this request
399 * @base: starting position in the buffer
402 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
404 unsigned int remainder = xdr->len - base;
405 int err, sent = 0;
407 if (unlikely(!sock))
408 return -ENOTSOCK;
410 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
411 if (base != 0) {
412 addr = NULL;
413 addrlen = 0;
416 if (base < xdr->head[0].iov_len || addr != NULL) {
417 unsigned int len = xdr->head[0].iov_len - base;
418 remainder -= len;
419 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
420 if (remainder == 0 || err != len)
421 goto out;
422 sent += err;
423 base = 0;
424 } else
425 base -= xdr->head[0].iov_len;
427 if (base < xdr->page_len) {
428 unsigned int len = xdr->page_len - base;
429 remainder -= len;
430 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
431 if (remainder == 0 || err != len)
432 goto out;
433 sent += err;
434 base = 0;
435 } else
436 base -= xdr->page_len;
438 if (base >= xdr->tail[0].iov_len)
439 return sent;
440 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
441 out:
442 if (sent == 0)
443 return err;
444 if (err > 0)
445 sent += err;
446 return sent;
449 static void xs_nospace_callback(struct rpc_task *task)
451 struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
453 transport->inet->sk_write_pending--;
454 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
458 * xs_nospace - place task on wait queue if transmit was incomplete
459 * @task: task to put to sleep
462 static int xs_nospace(struct rpc_task *task)
464 struct rpc_rqst *req = task->tk_rqstp;
465 struct rpc_xprt *xprt = req->rq_xprt;
466 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
467 int ret = 0;
469 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
470 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
471 req->rq_slen);
473 /* Protect against races with write_space */
474 spin_lock_bh(&xprt->transport_lock);
476 /* Don't race with disconnect */
477 if (xprt_connected(xprt)) {
478 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
479 ret = -EAGAIN;
481 * Notify TCP that we're limited by the application
482 * window size
484 set_bit(SOCK_NOSPACE, &transport->sock->flags);
485 transport->inet->sk_write_pending++;
486 /* ...and wait for more buffer space */
487 xprt_wait_for_buffer_space(task, xs_nospace_callback);
489 } else {
490 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
491 ret = -ENOTCONN;
494 spin_unlock_bh(&xprt->transport_lock);
495 return ret;
499 * xs_udp_send_request - write an RPC request to a UDP socket
500 * @task: address of RPC task that manages the state of an RPC request
502 * Return values:
503 * 0: The request has been sent
504 * EAGAIN: The socket was blocked, please call again later to
505 * complete the request
506 * ENOTCONN: Caller needs to invoke connect logic then call again
507 * other: Some other error occured, the request was not sent
509 static int xs_udp_send_request(struct rpc_task *task)
511 struct rpc_rqst *req = task->tk_rqstp;
512 struct rpc_xprt *xprt = req->rq_xprt;
513 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
514 struct xdr_buf *xdr = &req->rq_snd_buf;
515 int status;
517 xs_pktdump("packet data:",
518 req->rq_svec->iov_base,
519 req->rq_svec->iov_len);
521 if (!xprt_bound(xprt))
522 return -ENOTCONN;
523 status = xs_sendpages(transport->sock,
524 xs_addr(xprt),
525 xprt->addrlen, xdr,
526 req->rq_bytes_sent);
528 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
529 xdr->len - req->rq_bytes_sent, status);
531 if (status >= 0) {
532 req->rq_xmit_bytes_sent += status;
533 if (status >= req->rq_slen)
534 return 0;
535 /* Still some bytes left; set up for a retry later. */
536 status = -EAGAIN;
539 switch (status) {
540 case -ENOTSOCK:
541 status = -ENOTCONN;
542 /* Should we call xs_close() here? */
543 break;
544 case -EAGAIN:
545 status = xs_nospace(task);
546 break;
547 default:
548 dprintk("RPC: sendmsg returned unrecognized error %d\n",
549 -status);
550 case -ENETUNREACH:
551 case -EPIPE:
552 case -ECONNREFUSED:
553 /* When the server has died, an ICMP port unreachable message
554 * prompts ECONNREFUSED. */
555 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
558 return status;
562 * xs_tcp_shutdown - gracefully shut down a TCP socket
563 * @xprt: transport
565 * Initiates a graceful shutdown of the TCP socket by calling the
566 * equivalent of shutdown(SHUT_WR);
568 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
570 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
571 struct socket *sock = transport->sock;
573 if (sock != NULL)
574 kernel_sock_shutdown(sock, SHUT_WR);
577 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
579 u32 reclen = buf->len - sizeof(rpc_fraghdr);
580 rpc_fraghdr *base = buf->head[0].iov_base;
581 *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
585 * xs_tcp_send_request - write an RPC request to a TCP socket
586 * @task: address of RPC task that manages the state of an RPC request
588 * Return values:
589 * 0: The request has been sent
590 * EAGAIN: The socket was blocked, please call again later to
591 * complete the request
592 * ENOTCONN: Caller needs to invoke connect logic then call again
593 * other: Some other error occured, the request was not sent
595 * XXX: In the case of soft timeouts, should we eventually give up
596 * if sendmsg is not able to make progress?
598 static int xs_tcp_send_request(struct rpc_task *task)
600 struct rpc_rqst *req = task->tk_rqstp;
601 struct rpc_xprt *xprt = req->rq_xprt;
602 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
603 struct xdr_buf *xdr = &req->rq_snd_buf;
604 int status;
606 xs_encode_tcp_record_marker(&req->rq_snd_buf);
608 xs_pktdump("packet data:",
609 req->rq_svec->iov_base,
610 req->rq_svec->iov_len);
612 /* Continue transmitting the packet/record. We must be careful
613 * to cope with writespace callbacks arriving _after_ we have
614 * called sendmsg(). */
615 while (1) {
616 status = xs_sendpages(transport->sock,
617 NULL, 0, xdr, req->rq_bytes_sent);
619 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
620 xdr->len - req->rq_bytes_sent, status);
622 if (unlikely(status < 0))
623 break;
625 /* If we've sent the entire packet, immediately
626 * reset the count of bytes sent. */
627 req->rq_bytes_sent += status;
628 req->rq_xmit_bytes_sent += status;
629 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
630 req->rq_bytes_sent = 0;
631 return 0;
634 if (status != 0)
635 continue;
636 status = -EAGAIN;
637 break;
640 switch (status) {
641 case -ENOTSOCK:
642 status = -ENOTCONN;
643 /* Should we call xs_close() here? */
644 break;
645 case -EAGAIN:
646 status = xs_nospace(task);
647 break;
648 default:
649 dprintk("RPC: sendmsg returned unrecognized error %d\n",
650 -status);
651 case -ECONNRESET:
652 case -EPIPE:
653 xs_tcp_shutdown(xprt);
654 case -ECONNREFUSED:
655 case -ENOTCONN:
656 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
659 return status;
663 * xs_tcp_release_xprt - clean up after a tcp transmission
664 * @xprt: transport
665 * @task: rpc task
667 * This cleans up if an error causes us to abort the transmission of a request.
668 * In this case, the socket may need to be reset in order to avoid confusing
669 * the server.
671 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
673 struct rpc_rqst *req;
675 if (task != xprt->snd_task)
676 return;
677 if (task == NULL)
678 goto out_release;
679 req = task->tk_rqstp;
680 if (req->rq_bytes_sent == 0)
681 goto out_release;
682 if (req->rq_bytes_sent == req->rq_snd_buf.len)
683 goto out_release;
684 set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
685 out_release:
686 xprt_release_xprt(xprt, task);
689 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
691 transport->old_data_ready = sk->sk_data_ready;
692 transport->old_state_change = sk->sk_state_change;
693 transport->old_write_space = sk->sk_write_space;
694 transport->old_error_report = sk->sk_error_report;
697 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
699 sk->sk_data_ready = transport->old_data_ready;
700 sk->sk_state_change = transport->old_state_change;
701 sk->sk_write_space = transport->old_write_space;
702 sk->sk_error_report = transport->old_error_report;
705 static void xs_reset_transport(struct sock_xprt *transport)
707 struct socket *sock = transport->sock;
708 struct sock *sk = transport->inet;
710 if (sk == NULL)
711 return;
713 write_lock_bh(&sk->sk_callback_lock);
714 transport->inet = NULL;
715 transport->sock = NULL;
717 sk->sk_user_data = NULL;
719 xs_restore_old_callbacks(transport, sk);
720 write_unlock_bh(&sk->sk_callback_lock);
722 sk->sk_no_check = 0;
724 sock_release(sock);
728 * xs_close - close a socket
729 * @xprt: transport
731 * This is used when all requests are complete; ie, no DRC state remains
732 * on the server we want to save.
734 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
735 * xs_reset_transport() zeroing the socket from underneath a writer.
737 static void xs_close(struct rpc_xprt *xprt)
739 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
741 dprintk("RPC: xs_close xprt %p\n", xprt);
743 xs_reset_transport(transport);
744 xprt->reestablish_timeout = 0;
746 smp_mb__before_clear_bit();
747 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
748 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
749 clear_bit(XPRT_CLOSING, &xprt->state);
750 smp_mb__after_clear_bit();
751 xprt_disconnect_done(xprt);
754 static void xs_tcp_close(struct rpc_xprt *xprt)
756 if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state))
757 xs_close(xprt);
758 else
759 xs_tcp_shutdown(xprt);
763 * xs_destroy - prepare to shutdown a transport
764 * @xprt: doomed transport
767 static void xs_destroy(struct rpc_xprt *xprt)
769 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
771 dprintk("RPC: xs_destroy xprt %p\n", xprt);
773 cancel_rearming_delayed_work(&transport->connect_worker);
775 xs_close(xprt);
776 xs_free_peer_addresses(xprt);
777 kfree(xprt->slot);
778 kfree(xprt);
779 module_put(THIS_MODULE);
782 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
784 return (struct rpc_xprt *) sk->sk_user_data;
788 * xs_udp_data_ready - "data ready" callback for UDP sockets
789 * @sk: socket with data to read
790 * @len: how much data to read
793 static void xs_udp_data_ready(struct sock *sk, int len)
795 struct rpc_task *task;
796 struct rpc_xprt *xprt;
797 struct rpc_rqst *rovr;
798 struct sk_buff *skb;
799 int err, repsize, copied;
800 u32 _xid;
801 __be32 *xp;
803 read_lock(&sk->sk_callback_lock);
804 dprintk("RPC: xs_udp_data_ready...\n");
805 if (!(xprt = xprt_from_sock(sk)))
806 goto out;
808 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
809 goto out;
811 if (xprt->shutdown)
812 goto dropit;
814 repsize = skb->len - sizeof(struct udphdr);
815 if (repsize < 4) {
816 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
817 goto dropit;
820 /* Copy the XID from the skb... */
821 xp = skb_header_pointer(skb, sizeof(struct udphdr),
822 sizeof(_xid), &_xid);
823 if (xp == NULL)
824 goto dropit;
826 /* Look up and lock the request corresponding to the given XID */
827 spin_lock(&xprt->transport_lock);
828 rovr = xprt_lookup_rqst(xprt, *xp);
829 if (!rovr)
830 goto out_unlock;
831 task = rovr->rq_task;
833 if ((copied = rovr->rq_private_buf.buflen) > repsize)
834 copied = repsize;
836 /* Suck it into the iovec, verify checksum if not done by hw. */
837 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
838 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
839 goto out_unlock;
842 UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
844 /* Something worked... */
845 dst_confirm(skb_dst(skb));
847 xprt_adjust_cwnd(task, copied);
848 xprt_complete_rqst(task, copied);
850 out_unlock:
851 spin_unlock(&xprt->transport_lock);
852 dropit:
853 skb_free_datagram(sk, skb);
854 out:
855 read_unlock(&sk->sk_callback_lock);
858 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
860 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
861 size_t len, used;
862 char *p;
864 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
865 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
866 used = xdr_skb_read_bits(desc, p, len);
867 transport->tcp_offset += used;
868 if (used != len)
869 return;
871 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
872 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
873 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
874 else
875 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
876 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
878 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
879 transport->tcp_offset = 0;
881 /* Sanity check of the record length */
882 if (unlikely(transport->tcp_reclen < 8)) {
883 dprintk("RPC: invalid TCP record fragment length\n");
884 xprt_force_disconnect(xprt);
885 return;
887 dprintk("RPC: reading TCP record fragment of length %d\n",
888 transport->tcp_reclen);
891 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
893 if (transport->tcp_offset == transport->tcp_reclen) {
894 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
895 transport->tcp_offset = 0;
896 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
897 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
898 transport->tcp_flags |= TCP_RCV_COPY_XID;
899 transport->tcp_copied = 0;
904 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
906 size_t len, used;
907 char *p;
909 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
910 dprintk("RPC: reading XID (%Zu bytes)\n", len);
911 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
912 used = xdr_skb_read_bits(desc, p, len);
913 transport->tcp_offset += used;
914 if (used != len)
915 return;
916 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
917 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
918 transport->tcp_copied = 4;
919 dprintk("RPC: reading %s XID %08x\n",
920 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
921 : "request with",
922 ntohl(transport->tcp_xid));
923 xs_tcp_check_fraghdr(transport);
926 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
927 struct xdr_skb_reader *desc)
929 size_t len, used;
930 u32 offset;
931 char *p;
934 * We want transport->tcp_offset to be 8 at the end of this routine
935 * (4 bytes for the xid and 4 bytes for the call/reply flag).
936 * When this function is called for the first time,
937 * transport->tcp_offset is 4 (after having already read the xid).
939 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
940 len = sizeof(transport->tcp_calldir) - offset;
941 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len);
942 p = ((char *) &transport->tcp_calldir) + offset;
943 used = xdr_skb_read_bits(desc, p, len);
944 transport->tcp_offset += used;
945 if (used != len)
946 return;
947 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
949 * We don't yet have the XDR buffer, so we will write the calldir
950 * out after we get the buffer from the 'struct rpc_rqst'
952 switch (ntohl(transport->tcp_calldir)) {
953 case RPC_REPLY:
954 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
955 transport->tcp_flags |= TCP_RCV_COPY_DATA;
956 transport->tcp_flags |= TCP_RPC_REPLY;
957 break;
958 case RPC_CALL:
959 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
960 transport->tcp_flags |= TCP_RCV_COPY_DATA;
961 transport->tcp_flags &= ~TCP_RPC_REPLY;
962 break;
963 default:
964 dprintk("RPC: invalid request message type\n");
965 xprt_force_disconnect(&transport->xprt);
967 xs_tcp_check_fraghdr(transport);
970 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
971 struct xdr_skb_reader *desc,
972 struct rpc_rqst *req)
974 struct sock_xprt *transport =
975 container_of(xprt, struct sock_xprt, xprt);
976 struct xdr_buf *rcvbuf;
977 size_t len;
978 ssize_t r;
980 rcvbuf = &req->rq_private_buf;
982 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
984 * Save the RPC direction in the XDR buffer
986 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
987 &transport->tcp_calldir,
988 sizeof(transport->tcp_calldir));
989 transport->tcp_copied += sizeof(transport->tcp_calldir);
990 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
993 len = desc->count;
994 if (len > transport->tcp_reclen - transport->tcp_offset) {
995 struct xdr_skb_reader my_desc;
997 len = transport->tcp_reclen - transport->tcp_offset;
998 memcpy(&my_desc, desc, sizeof(my_desc));
999 my_desc.count = len;
1000 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1001 &my_desc, xdr_skb_read_bits);
1002 desc->count -= r;
1003 desc->offset += r;
1004 } else
1005 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1006 desc, xdr_skb_read_bits);
1008 if (r > 0) {
1009 transport->tcp_copied += r;
1010 transport->tcp_offset += r;
1012 if (r != len) {
1013 /* Error when copying to the receive buffer,
1014 * usually because we weren't able to allocate
1015 * additional buffer pages. All we can do now
1016 * is turn off TCP_RCV_COPY_DATA, so the request
1017 * will not receive any additional updates,
1018 * and time out.
1019 * Any remaining data from this record will
1020 * be discarded.
1022 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1023 dprintk("RPC: XID %08x truncated request\n",
1024 ntohl(transport->tcp_xid));
1025 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1026 "tcp_offset = %u, tcp_reclen = %u\n",
1027 xprt, transport->tcp_copied,
1028 transport->tcp_offset, transport->tcp_reclen);
1029 return;
1032 dprintk("RPC: XID %08x read %Zd bytes\n",
1033 ntohl(transport->tcp_xid), r);
1034 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1035 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1036 transport->tcp_offset, transport->tcp_reclen);
1038 if (transport->tcp_copied == req->rq_private_buf.buflen)
1039 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1040 else if (transport->tcp_offset == transport->tcp_reclen) {
1041 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1042 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1047 * Finds the request corresponding to the RPC xid and invokes the common
1048 * tcp read code to read the data.
1050 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1051 struct xdr_skb_reader *desc)
1053 struct sock_xprt *transport =
1054 container_of(xprt, struct sock_xprt, xprt);
1055 struct rpc_rqst *req;
1057 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1059 /* Find and lock the request corresponding to this xid */
1060 spin_lock(&xprt->transport_lock);
1061 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1062 if (!req) {
1063 dprintk("RPC: XID %08x request not found!\n",
1064 ntohl(transport->tcp_xid));
1065 spin_unlock(&xprt->transport_lock);
1066 return -1;
1069 xs_tcp_read_common(xprt, desc, req);
1071 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1072 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1074 spin_unlock(&xprt->transport_lock);
1075 return 0;
1078 #if defined(CONFIG_NFS_V4_1)
1080 * Obtains an rpc_rqst previously allocated and invokes the common
1081 * tcp read code to read the data. The result is placed in the callback
1082 * queue.
1083 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1084 * connection and return -1.
1086 static inline int xs_tcp_read_callback(struct rpc_xprt *xprt,
1087 struct xdr_skb_reader *desc)
1089 struct sock_xprt *transport =
1090 container_of(xprt, struct sock_xprt, xprt);
1091 struct rpc_rqst *req;
1093 req = xprt_alloc_bc_request(xprt);
1094 if (req == NULL) {
1095 printk(KERN_WARNING "Callback slot table overflowed\n");
1096 xprt_force_disconnect(xprt);
1097 return -1;
1100 req->rq_xid = transport->tcp_xid;
1101 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1102 xs_tcp_read_common(xprt, desc, req);
1104 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) {
1105 struct svc_serv *bc_serv = xprt->bc_serv;
1108 * Add callback request to callback list. The callback
1109 * service sleeps on the sv_cb_waitq waiting for new
1110 * requests. Wake it up after adding enqueing the
1111 * request.
1113 dprintk("RPC: add callback request to list\n");
1114 spin_lock(&bc_serv->sv_cb_lock);
1115 list_add(&req->rq_bc_list, &bc_serv->sv_cb_list);
1116 spin_unlock(&bc_serv->sv_cb_lock);
1117 wake_up(&bc_serv->sv_cb_waitq);
1120 req->rq_private_buf.len = transport->tcp_copied;
1122 return 0;
1125 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1126 struct xdr_skb_reader *desc)
1128 struct sock_xprt *transport =
1129 container_of(xprt, struct sock_xprt, xprt);
1131 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1132 xs_tcp_read_reply(xprt, desc) :
1133 xs_tcp_read_callback(xprt, desc);
1135 #else
1136 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1137 struct xdr_skb_reader *desc)
1139 return xs_tcp_read_reply(xprt, desc);
1141 #endif /* CONFIG_NFS_V4_1 */
1144 * Read data off the transport. This can be either an RPC_CALL or an
1145 * RPC_REPLY. Relay the processing to helper functions.
1147 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1148 struct xdr_skb_reader *desc)
1150 struct sock_xprt *transport =
1151 container_of(xprt, struct sock_xprt, xprt);
1153 if (_xs_tcp_read_data(xprt, desc) == 0)
1154 xs_tcp_check_fraghdr(transport);
1155 else {
1157 * The transport_lock protects the request handling.
1158 * There's no need to hold it to update the tcp_flags.
1160 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1164 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1166 size_t len;
1168 len = transport->tcp_reclen - transport->tcp_offset;
1169 if (len > desc->count)
1170 len = desc->count;
1171 desc->count -= len;
1172 desc->offset += len;
1173 transport->tcp_offset += len;
1174 dprintk("RPC: discarded %Zu bytes\n", len);
1175 xs_tcp_check_fraghdr(transport);
1178 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1180 struct rpc_xprt *xprt = rd_desc->arg.data;
1181 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1182 struct xdr_skb_reader desc = {
1183 .skb = skb,
1184 .offset = offset,
1185 .count = len,
1188 dprintk("RPC: xs_tcp_data_recv started\n");
1189 do {
1190 /* Read in a new fragment marker if necessary */
1191 /* Can we ever really expect to get completely empty fragments? */
1192 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1193 xs_tcp_read_fraghdr(xprt, &desc);
1194 continue;
1196 /* Read in the xid if necessary */
1197 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1198 xs_tcp_read_xid(transport, &desc);
1199 continue;
1201 /* Read in the call/reply flag */
1202 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1203 xs_tcp_read_calldir(transport, &desc);
1204 continue;
1206 /* Read in the request data */
1207 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1208 xs_tcp_read_data(xprt, &desc);
1209 continue;
1211 /* Skip over any trailing bytes on short reads */
1212 xs_tcp_read_discard(transport, &desc);
1213 } while (desc.count);
1214 dprintk("RPC: xs_tcp_data_recv done\n");
1215 return len - desc.count;
1219 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1220 * @sk: socket with data to read
1221 * @bytes: how much data to read
1224 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1226 struct rpc_xprt *xprt;
1227 read_descriptor_t rd_desc;
1228 int read;
1230 dprintk("RPC: xs_tcp_data_ready...\n");
1232 read_lock(&sk->sk_callback_lock);
1233 if (!(xprt = xprt_from_sock(sk)))
1234 goto out;
1235 if (xprt->shutdown)
1236 goto out;
1238 /* Any data means we had a useful conversation, so
1239 * the we don't need to delay the next reconnect
1241 if (xprt->reestablish_timeout)
1242 xprt->reestablish_timeout = 0;
1244 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1245 rd_desc.arg.data = xprt;
1246 do {
1247 rd_desc.count = 65536;
1248 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1249 } while (read > 0);
1250 out:
1251 read_unlock(&sk->sk_callback_lock);
1255 * Do the equivalent of linger/linger2 handling for dealing with
1256 * broken servers that don't close the socket in a timely
1257 * fashion
1259 static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1260 unsigned long timeout)
1262 struct sock_xprt *transport;
1264 if (xprt_test_and_set_connecting(xprt))
1265 return;
1266 set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1267 transport = container_of(xprt, struct sock_xprt, xprt);
1268 queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1269 timeout);
1272 static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1274 struct sock_xprt *transport;
1276 transport = container_of(xprt, struct sock_xprt, xprt);
1278 if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1279 !cancel_delayed_work(&transport->connect_worker))
1280 return;
1281 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1282 xprt_clear_connecting(xprt);
1285 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1287 smp_mb__before_clear_bit();
1288 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1289 clear_bit(XPRT_CLOSING, &xprt->state);
1290 smp_mb__after_clear_bit();
1291 /* Mark transport as closed and wake up all pending tasks */
1292 xprt_disconnect_done(xprt);
1296 * xs_tcp_state_change - callback to handle TCP socket state changes
1297 * @sk: socket whose state has changed
1300 static void xs_tcp_state_change(struct sock *sk)
1302 struct rpc_xprt *xprt;
1304 read_lock(&sk->sk_callback_lock);
1305 if (!(xprt = xprt_from_sock(sk)))
1306 goto out;
1307 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1308 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1309 sk->sk_state, xprt_connected(xprt),
1310 sock_flag(sk, SOCK_DEAD),
1311 sock_flag(sk, SOCK_ZAPPED),
1312 sk->sk_shutdown);
1314 switch (sk->sk_state) {
1315 case TCP_ESTABLISHED:
1316 spin_lock_bh(&xprt->transport_lock);
1317 if (!xprt_test_and_set_connected(xprt)) {
1318 struct sock_xprt *transport = container_of(xprt,
1319 struct sock_xprt, xprt);
1321 /* Reset TCP record info */
1322 transport->tcp_offset = 0;
1323 transport->tcp_reclen = 0;
1324 transport->tcp_copied = 0;
1325 transport->tcp_flags =
1326 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1328 xprt_wake_pending_tasks(xprt, -EAGAIN);
1330 spin_unlock_bh(&xprt->transport_lock);
1331 break;
1332 case TCP_FIN_WAIT1:
1333 /* The client initiated a shutdown of the socket */
1334 xprt->connect_cookie++;
1335 xprt->reestablish_timeout = 0;
1336 set_bit(XPRT_CLOSING, &xprt->state);
1337 smp_mb__before_clear_bit();
1338 clear_bit(XPRT_CONNECTED, &xprt->state);
1339 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1340 smp_mb__after_clear_bit();
1341 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1342 break;
1343 case TCP_CLOSE_WAIT:
1344 /* The server initiated a shutdown of the socket */
1345 xprt_force_disconnect(xprt);
1346 case TCP_SYN_SENT:
1347 xprt->connect_cookie++;
1348 case TCP_CLOSING:
1350 * If the server closed down the connection, make sure that
1351 * we back off before reconnecting
1353 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1354 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1355 break;
1356 case TCP_LAST_ACK:
1357 set_bit(XPRT_CLOSING, &xprt->state);
1358 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1359 smp_mb__before_clear_bit();
1360 clear_bit(XPRT_CONNECTED, &xprt->state);
1361 smp_mb__after_clear_bit();
1362 break;
1363 case TCP_CLOSE:
1364 xs_tcp_cancel_linger_timeout(xprt);
1365 xs_sock_mark_closed(xprt);
1367 out:
1368 read_unlock(&sk->sk_callback_lock);
1372 * xs_error_report - callback mainly for catching socket errors
1373 * @sk: socket
1375 static void xs_error_report(struct sock *sk)
1377 struct rpc_xprt *xprt;
1379 read_lock(&sk->sk_callback_lock);
1380 if (!(xprt = xprt_from_sock(sk)))
1381 goto out;
1382 dprintk("RPC: %s client %p...\n"
1383 "RPC: error %d\n",
1384 __func__, xprt, sk->sk_err);
1385 xprt_wake_pending_tasks(xprt, -EAGAIN);
1386 out:
1387 read_unlock(&sk->sk_callback_lock);
1390 static void xs_write_space(struct sock *sk)
1392 struct socket *sock;
1393 struct rpc_xprt *xprt;
1395 if (unlikely(!(sock = sk->sk_socket)))
1396 return;
1397 clear_bit(SOCK_NOSPACE, &sock->flags);
1399 if (unlikely(!(xprt = xprt_from_sock(sk))))
1400 return;
1401 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1402 return;
1404 xprt_write_space(xprt);
1408 * xs_udp_write_space - callback invoked when socket buffer space
1409 * becomes available
1410 * @sk: socket whose state has changed
1412 * Called when more output buffer space is available for this socket.
1413 * We try not to wake our writers until they can make "significant"
1414 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1415 * with a bunch of small requests.
1417 static void xs_udp_write_space(struct sock *sk)
1419 read_lock(&sk->sk_callback_lock);
1421 /* from net/core/sock.c:sock_def_write_space */
1422 if (sock_writeable(sk))
1423 xs_write_space(sk);
1425 read_unlock(&sk->sk_callback_lock);
1429 * xs_tcp_write_space - callback invoked when socket buffer space
1430 * becomes available
1431 * @sk: socket whose state has changed
1433 * Called when more output buffer space is available for this socket.
1434 * We try not to wake our writers until they can make "significant"
1435 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1436 * with a bunch of small requests.
1438 static void xs_tcp_write_space(struct sock *sk)
1440 read_lock(&sk->sk_callback_lock);
1442 /* from net/core/stream.c:sk_stream_write_space */
1443 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk))
1444 xs_write_space(sk);
1446 read_unlock(&sk->sk_callback_lock);
1449 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1451 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1452 struct sock *sk = transport->inet;
1454 if (transport->rcvsize) {
1455 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1456 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1458 if (transport->sndsize) {
1459 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1460 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1461 sk->sk_write_space(sk);
1466 * xs_udp_set_buffer_size - set send and receive limits
1467 * @xprt: generic transport
1468 * @sndsize: requested size of send buffer, in bytes
1469 * @rcvsize: requested size of receive buffer, in bytes
1471 * Set socket send and receive buffer size limits.
1473 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1475 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1477 transport->sndsize = 0;
1478 if (sndsize)
1479 transport->sndsize = sndsize + 1024;
1480 transport->rcvsize = 0;
1481 if (rcvsize)
1482 transport->rcvsize = rcvsize + 1024;
1484 xs_udp_do_set_buffer_size(xprt);
1488 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1489 * @task: task that timed out
1491 * Adjust the congestion window after a retransmit timeout has occurred.
1493 static void xs_udp_timer(struct rpc_task *task)
1495 xprt_adjust_cwnd(task, -ETIMEDOUT);
1498 static unsigned short xs_get_random_port(void)
1500 unsigned short range = xprt_max_resvport - xprt_min_resvport;
1501 unsigned short rand = (unsigned short) net_random() % range;
1502 return rand + xprt_min_resvport;
1506 * xs_set_port - reset the port number in the remote endpoint address
1507 * @xprt: generic transport
1508 * @port: new port number
1511 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1513 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1515 rpc_set_port(xs_addr(xprt), port);
1516 xs_update_peer_port(xprt);
1519 static unsigned short xs_get_srcport(struct sock_xprt *transport, struct socket *sock)
1521 unsigned short port = transport->srcport;
1523 if (port == 0 && transport->xprt.resvport)
1524 port = xs_get_random_port();
1525 return port;
1528 static unsigned short xs_next_srcport(struct sock_xprt *transport, struct socket *sock, unsigned short port)
1530 if (transport->srcport != 0)
1531 transport->srcport = 0;
1532 if (!transport->xprt.resvport)
1533 return 0;
1534 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1535 return xprt_max_resvport;
1536 return --port;
1539 static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
1541 struct sockaddr_in myaddr = {
1542 .sin_family = AF_INET,
1544 struct sockaddr_in *sa;
1545 int err, nloop = 0;
1546 unsigned short port = xs_get_srcport(transport, sock);
1547 unsigned short last;
1549 sa = (struct sockaddr_in *)&transport->srcaddr;
1550 myaddr.sin_addr = sa->sin_addr;
1551 do {
1552 myaddr.sin_port = htons(port);
1553 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1554 sizeof(myaddr));
1555 if (port == 0)
1556 break;
1557 if (err == 0) {
1558 transport->srcport = port;
1559 break;
1561 last = port;
1562 port = xs_next_srcport(transport, sock, port);
1563 if (port > last)
1564 nloop++;
1565 } while (err == -EADDRINUSE && nloop != 2);
1566 dprintk("RPC: %s %pI4:%u: %s (%d)\n",
1567 __func__, &myaddr.sin_addr,
1568 port, err ? "failed" : "ok", err);
1569 return err;
1572 static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1574 struct sockaddr_in6 myaddr = {
1575 .sin6_family = AF_INET6,
1577 struct sockaddr_in6 *sa;
1578 int err, nloop = 0;
1579 unsigned short port = xs_get_srcport(transport, sock);
1580 unsigned short last;
1582 sa = (struct sockaddr_in6 *)&transport->srcaddr;
1583 myaddr.sin6_addr = sa->sin6_addr;
1584 do {
1585 myaddr.sin6_port = htons(port);
1586 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1587 sizeof(myaddr));
1588 if (port == 0)
1589 break;
1590 if (err == 0) {
1591 transport->srcport = port;
1592 break;
1594 last = port;
1595 port = xs_next_srcport(transport, sock, port);
1596 if (port > last)
1597 nloop++;
1598 } while (err == -EADDRINUSE && nloop != 2);
1599 dprintk("RPC: xs_bind6 %pI6:%u: %s (%d)\n",
1600 &myaddr.sin6_addr, port, err ? "failed" : "ok", err);
1601 return err;
1604 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1605 static struct lock_class_key xs_key[2];
1606 static struct lock_class_key xs_slock_key[2];
1608 static inline void xs_reclassify_socket4(struct socket *sock)
1610 struct sock *sk = sock->sk;
1612 BUG_ON(sock_owned_by_user(sk));
1613 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1614 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1617 static inline void xs_reclassify_socket6(struct socket *sock)
1619 struct sock *sk = sock->sk;
1621 BUG_ON(sock_owned_by_user(sk));
1622 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1623 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1625 #else
1626 static inline void xs_reclassify_socket4(struct socket *sock)
1630 static inline void xs_reclassify_socket6(struct socket *sock)
1633 #endif
1635 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1637 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1639 if (!transport->inet) {
1640 struct sock *sk = sock->sk;
1642 write_lock_bh(&sk->sk_callback_lock);
1644 xs_save_old_callbacks(transport, sk);
1646 sk->sk_user_data = xprt;
1647 sk->sk_data_ready = xs_udp_data_ready;
1648 sk->sk_write_space = xs_udp_write_space;
1649 sk->sk_error_report = xs_error_report;
1650 sk->sk_no_check = UDP_CSUM_NORCV;
1651 sk->sk_allocation = GFP_ATOMIC;
1653 xprt_set_connected(xprt);
1655 /* Reset to new socket */
1656 transport->sock = sock;
1657 transport->inet = sk;
1659 write_unlock_bh(&sk->sk_callback_lock);
1661 xs_udp_do_set_buffer_size(xprt);
1665 * xs_udp_connect_worker4 - set up a UDP socket
1666 * @work: RPC transport to connect
1668 * Invoked by a work queue tasklet.
1670 static void xs_udp_connect_worker4(struct work_struct *work)
1672 struct sock_xprt *transport =
1673 container_of(work, struct sock_xprt, connect_worker.work);
1674 struct rpc_xprt *xprt = &transport->xprt;
1675 struct socket *sock = transport->sock;
1676 int err, status = -EIO;
1678 if (xprt->shutdown)
1679 goto out;
1681 /* Start by resetting any existing state */
1682 xs_reset_transport(transport);
1684 err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
1685 if (err < 0) {
1686 dprintk("RPC: can't create UDP transport socket (%d).\n", -err);
1687 goto out;
1689 xs_reclassify_socket4(sock);
1691 if (xs_bind4(transport, sock)) {
1692 sock_release(sock);
1693 goto out;
1696 dprintk("RPC: worker connecting xprt %p via %s to "
1697 "%s (port %s)\n", xprt,
1698 xprt->address_strings[RPC_DISPLAY_PROTO],
1699 xprt->address_strings[RPC_DISPLAY_ADDR],
1700 xprt->address_strings[RPC_DISPLAY_PORT]);
1702 xs_udp_finish_connecting(xprt, sock);
1703 status = 0;
1704 out:
1705 xprt_clear_connecting(xprt);
1706 xprt_wake_pending_tasks(xprt, status);
1710 * xs_udp_connect_worker6 - set up a UDP socket
1711 * @work: RPC transport to connect
1713 * Invoked by a work queue tasklet.
1715 static void xs_udp_connect_worker6(struct work_struct *work)
1717 struct sock_xprt *transport =
1718 container_of(work, struct sock_xprt, connect_worker.work);
1719 struct rpc_xprt *xprt = &transport->xprt;
1720 struct socket *sock = transport->sock;
1721 int err, status = -EIO;
1723 if (xprt->shutdown)
1724 goto out;
1726 /* Start by resetting any existing state */
1727 xs_reset_transport(transport);
1729 err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock);
1730 if (err < 0) {
1731 dprintk("RPC: can't create UDP transport socket (%d).\n", -err);
1732 goto out;
1734 xs_reclassify_socket6(sock);
1736 if (xs_bind6(transport, sock) < 0) {
1737 sock_release(sock);
1738 goto out;
1741 dprintk("RPC: worker connecting xprt %p via %s to "
1742 "%s (port %s)\n", xprt,
1743 xprt->address_strings[RPC_DISPLAY_PROTO],
1744 xprt->address_strings[RPC_DISPLAY_ADDR],
1745 xprt->address_strings[RPC_DISPLAY_PORT]);
1747 xs_udp_finish_connecting(xprt, sock);
1748 status = 0;
1749 out:
1750 xprt_clear_connecting(xprt);
1751 xprt_wake_pending_tasks(xprt, status);
1755 * We need to preserve the port number so the reply cache on the server can
1756 * find our cached RPC replies when we get around to reconnecting.
1758 static void xs_abort_connection(struct rpc_xprt *xprt, struct sock_xprt *transport)
1760 int result;
1761 struct sockaddr any;
1763 dprintk("RPC: disconnecting xprt %p to reuse port\n", xprt);
1766 * Disconnect the transport socket by doing a connect operation
1767 * with AF_UNSPEC. This should return immediately...
1769 memset(&any, 0, sizeof(any));
1770 any.sa_family = AF_UNSPEC;
1771 result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1772 if (!result)
1773 xs_sock_mark_closed(xprt);
1774 else
1775 dprintk("RPC: AF_UNSPEC connect return code %d\n",
1776 result);
1779 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt, struct sock_xprt *transport)
1781 unsigned int state = transport->inet->sk_state;
1783 if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED) {
1784 /* we don't need to abort the connection if the socket
1785 * hasn't undergone a shutdown
1787 if (transport->inet->sk_shutdown == 0)
1788 return;
1789 dprintk("RPC: %s: TCP_CLOSEd and sk_shutdown set to %d\n",
1790 __func__, transport->inet->sk_shutdown);
1792 if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT)) {
1793 /* we don't need to abort the connection if the socket
1794 * hasn't undergone a shutdown
1796 if (transport->inet->sk_shutdown == 0)
1797 return;
1798 dprintk("RPC: %s: ESTABLISHED/SYN_SENT "
1799 "sk_shutdown set to %d\n",
1800 __func__, transport->inet->sk_shutdown);
1802 xs_abort_connection(xprt, transport);
1805 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1807 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1809 if (!transport->inet) {
1810 struct sock *sk = sock->sk;
1812 write_lock_bh(&sk->sk_callback_lock);
1814 xs_save_old_callbacks(transport, sk);
1816 sk->sk_user_data = xprt;
1817 sk->sk_data_ready = xs_tcp_data_ready;
1818 sk->sk_state_change = xs_tcp_state_change;
1819 sk->sk_write_space = xs_tcp_write_space;
1820 sk->sk_error_report = xs_error_report;
1821 sk->sk_allocation = GFP_ATOMIC;
1823 /* socket options */
1824 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1825 sock_reset_flag(sk, SOCK_LINGER);
1826 tcp_sk(sk)->linger2 = 0;
1827 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1829 xprt_clear_connected(xprt);
1831 /* Reset to new socket */
1832 transport->sock = sock;
1833 transport->inet = sk;
1835 write_unlock_bh(&sk->sk_callback_lock);
1838 if (!xprt_bound(xprt))
1839 return -ENOTCONN;
1841 /* Tell the socket layer to start connecting... */
1842 xprt->stat.connect_count++;
1843 xprt->stat.connect_start = jiffies;
1844 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
1848 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
1849 * @xprt: RPC transport to connect
1850 * @transport: socket transport to connect
1851 * @create_sock: function to create a socket of the correct type
1853 * Invoked by a work queue tasklet.
1855 static void xs_tcp_setup_socket(struct rpc_xprt *xprt,
1856 struct sock_xprt *transport,
1857 struct socket *(*create_sock)(struct rpc_xprt *,
1858 struct sock_xprt *))
1860 struct socket *sock = transport->sock;
1861 int status = -EIO;
1863 if (xprt->shutdown)
1864 goto out;
1866 if (!sock) {
1867 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1868 sock = create_sock(xprt, transport);
1869 if (IS_ERR(sock)) {
1870 status = PTR_ERR(sock);
1871 goto out;
1873 } else {
1874 int abort_and_exit;
1876 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
1877 &xprt->state);
1878 /* "close" the socket, preserving the local port */
1879 xs_tcp_reuse_connection(xprt, transport);
1881 if (abort_and_exit)
1882 goto out_eagain;
1885 dprintk("RPC: worker connecting xprt %p via %s to "
1886 "%s (port %s)\n", xprt,
1887 xprt->address_strings[RPC_DISPLAY_PROTO],
1888 xprt->address_strings[RPC_DISPLAY_ADDR],
1889 xprt->address_strings[RPC_DISPLAY_PORT]);
1891 status = xs_tcp_finish_connecting(xprt, sock);
1892 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
1893 xprt, -status, xprt_connected(xprt),
1894 sock->sk->sk_state);
1895 switch (status) {
1896 default:
1897 printk("%s: connect returned unhandled error %d\n",
1898 __func__, status);
1899 case -EADDRNOTAVAIL:
1900 /* We're probably in TIME_WAIT. Get rid of existing socket,
1901 * and retry
1903 set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1904 xprt_force_disconnect(xprt);
1905 break;
1906 case -ECONNREFUSED:
1907 case -ECONNRESET:
1908 case -ENETUNREACH:
1909 /* retry with existing socket, after a delay */
1910 case 0:
1911 case -EINPROGRESS:
1912 case -EALREADY:
1913 xprt_clear_connecting(xprt);
1914 return;
1915 case -EINVAL:
1916 /* Happens, for instance, if the user specified a link
1917 * local IPv6 address without a scope-id.
1919 goto out;
1921 out_eagain:
1922 status = -EAGAIN;
1923 out:
1924 xprt_clear_connecting(xprt);
1925 xprt_wake_pending_tasks(xprt, status);
1928 static struct socket *xs_create_tcp_sock4(struct rpc_xprt *xprt,
1929 struct sock_xprt *transport)
1931 struct socket *sock;
1932 int err;
1934 /* start from scratch */
1935 err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock);
1936 if (err < 0) {
1937 dprintk("RPC: can't create TCP transport socket (%d).\n",
1938 -err);
1939 goto out_err;
1941 xs_reclassify_socket4(sock);
1943 if (xs_bind4(transport, sock) < 0) {
1944 sock_release(sock);
1945 goto out_err;
1947 return sock;
1948 out_err:
1949 return ERR_PTR(-EIO);
1953 * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1954 * @work: RPC transport to connect
1956 * Invoked by a work queue tasklet.
1958 static void xs_tcp_connect_worker4(struct work_struct *work)
1960 struct sock_xprt *transport =
1961 container_of(work, struct sock_xprt, connect_worker.work);
1962 struct rpc_xprt *xprt = &transport->xprt;
1964 xs_tcp_setup_socket(xprt, transport, xs_create_tcp_sock4);
1967 static struct socket *xs_create_tcp_sock6(struct rpc_xprt *xprt,
1968 struct sock_xprt *transport)
1970 struct socket *sock;
1971 int err;
1973 /* start from scratch */
1974 err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock);
1975 if (err < 0) {
1976 dprintk("RPC: can't create TCP transport socket (%d).\n",
1977 -err);
1978 goto out_err;
1980 xs_reclassify_socket6(sock);
1982 if (xs_bind6(transport, sock) < 0) {
1983 sock_release(sock);
1984 goto out_err;
1986 return sock;
1987 out_err:
1988 return ERR_PTR(-EIO);
1992 * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
1993 * @work: RPC transport to connect
1995 * Invoked by a work queue tasklet.
1997 static void xs_tcp_connect_worker6(struct work_struct *work)
1999 struct sock_xprt *transport =
2000 container_of(work, struct sock_xprt, connect_worker.work);
2001 struct rpc_xprt *xprt = &transport->xprt;
2003 xs_tcp_setup_socket(xprt, transport, xs_create_tcp_sock6);
2007 * xs_connect - connect a socket to a remote endpoint
2008 * @task: address of RPC task that manages state of connect request
2010 * TCP: If the remote end dropped the connection, delay reconnecting.
2012 * UDP socket connects are synchronous, but we use a work queue anyway
2013 * to guarantee that even unprivileged user processes can set up a
2014 * socket on a privileged port.
2016 * If a UDP socket connect fails, the delay behavior here prevents
2017 * retry floods (hard mounts).
2019 static void xs_connect(struct rpc_task *task)
2021 struct rpc_xprt *xprt = task->tk_xprt;
2022 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2024 if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
2025 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2026 "seconds\n",
2027 xprt, xprt->reestablish_timeout / HZ);
2028 queue_delayed_work(rpciod_workqueue,
2029 &transport->connect_worker,
2030 xprt->reestablish_timeout);
2031 xprt->reestablish_timeout <<= 1;
2032 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2033 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2034 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2035 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2036 } else {
2037 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2038 queue_delayed_work(rpciod_workqueue,
2039 &transport->connect_worker, 0);
2044 * xs_udp_print_stats - display UDP socket-specifc stats
2045 * @xprt: rpc_xprt struct containing statistics
2046 * @seq: output file
2049 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2051 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2053 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
2054 transport->srcport,
2055 xprt->stat.bind_count,
2056 xprt->stat.sends,
2057 xprt->stat.recvs,
2058 xprt->stat.bad_xids,
2059 xprt->stat.req_u,
2060 xprt->stat.bklog_u);
2064 * xs_tcp_print_stats - display TCP socket-specifc stats
2065 * @xprt: rpc_xprt struct containing statistics
2066 * @seq: output file
2069 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2071 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2072 long idle_time = 0;
2074 if (xprt_connected(xprt))
2075 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2077 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
2078 transport->srcport,
2079 xprt->stat.bind_count,
2080 xprt->stat.connect_count,
2081 xprt->stat.connect_time,
2082 idle_time,
2083 xprt->stat.sends,
2084 xprt->stat.recvs,
2085 xprt->stat.bad_xids,
2086 xprt->stat.req_u,
2087 xprt->stat.bklog_u);
2091 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2092 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2093 * to use the server side send routines.
2095 static void *bc_malloc(struct rpc_task *task, size_t size)
2097 struct page *page;
2098 struct rpc_buffer *buf;
2100 BUG_ON(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2101 page = alloc_page(GFP_KERNEL);
2103 if (!page)
2104 return NULL;
2106 buf = page_address(page);
2107 buf->len = PAGE_SIZE;
2109 return buf->data;
2113 * Free the space allocated in the bc_alloc routine
2115 static void bc_free(void *buffer)
2117 struct rpc_buffer *buf;
2119 if (!buffer)
2120 return;
2122 buf = container_of(buffer, struct rpc_buffer, data);
2123 free_page((unsigned long)buf);
2127 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2128 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2130 static int bc_sendto(struct rpc_rqst *req)
2132 int len;
2133 struct xdr_buf *xbufp = &req->rq_snd_buf;
2134 struct rpc_xprt *xprt = req->rq_xprt;
2135 struct sock_xprt *transport =
2136 container_of(xprt, struct sock_xprt, xprt);
2137 struct socket *sock = transport->sock;
2138 unsigned long headoff;
2139 unsigned long tailoff;
2142 * Set up the rpc header and record marker stuff
2144 xs_encode_tcp_record_marker(xbufp);
2146 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2147 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2148 len = svc_send_common(sock, xbufp,
2149 virt_to_page(xbufp->head[0].iov_base), headoff,
2150 xbufp->tail[0].iov_base, tailoff);
2152 if (len != xbufp->len) {
2153 printk(KERN_NOTICE "Error sending entire callback!\n");
2154 len = -EAGAIN;
2157 return len;
2161 * The send routine. Borrows from svc_send
2163 static int bc_send_request(struct rpc_task *task)
2165 struct rpc_rqst *req = task->tk_rqstp;
2166 struct svc_xprt *xprt;
2167 struct svc_sock *svsk;
2168 u32 len;
2170 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2172 * Get the server socket associated with this callback xprt
2174 xprt = req->rq_xprt->bc_xprt;
2175 svsk = container_of(xprt, struct svc_sock, sk_xprt);
2178 * Grab the mutex to serialize data as the connection is shared
2179 * with the fore channel
2181 if (!mutex_trylock(&xprt->xpt_mutex)) {
2182 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2183 if (!mutex_trylock(&xprt->xpt_mutex))
2184 return -EAGAIN;
2185 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2187 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2188 len = -ENOTCONN;
2189 else
2190 len = bc_sendto(req);
2191 mutex_unlock(&xprt->xpt_mutex);
2193 if (len > 0)
2194 len = 0;
2196 return len;
2200 * The close routine. Since this is client initiated, we do nothing
2203 static void bc_close(struct rpc_xprt *xprt)
2208 * The xprt destroy routine. Again, because this connection is client
2209 * initiated, we do nothing
2212 static void bc_destroy(struct rpc_xprt *xprt)
2216 static struct rpc_xprt_ops xs_udp_ops = {
2217 .set_buffer_size = xs_udp_set_buffer_size,
2218 .reserve_xprt = xprt_reserve_xprt_cong,
2219 .release_xprt = xprt_release_xprt_cong,
2220 .rpcbind = rpcb_getport_async,
2221 .set_port = xs_set_port,
2222 .connect = xs_connect,
2223 .buf_alloc = rpc_malloc,
2224 .buf_free = rpc_free,
2225 .send_request = xs_udp_send_request,
2226 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
2227 .timer = xs_udp_timer,
2228 .release_request = xprt_release_rqst_cong,
2229 .close = xs_close,
2230 .destroy = xs_destroy,
2231 .print_stats = xs_udp_print_stats,
2234 static struct rpc_xprt_ops xs_tcp_ops = {
2235 .reserve_xprt = xprt_reserve_xprt,
2236 .release_xprt = xs_tcp_release_xprt,
2237 .rpcbind = rpcb_getport_async,
2238 .set_port = xs_set_port,
2239 .connect = xs_connect,
2240 .buf_alloc = rpc_malloc,
2241 .buf_free = rpc_free,
2242 .send_request = xs_tcp_send_request,
2243 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2244 .close = xs_tcp_close,
2245 .destroy = xs_destroy,
2246 .print_stats = xs_tcp_print_stats,
2250 * The rpc_xprt_ops for the server backchannel
2253 static struct rpc_xprt_ops bc_tcp_ops = {
2254 .reserve_xprt = xprt_reserve_xprt,
2255 .release_xprt = xprt_release_xprt,
2256 .buf_alloc = bc_malloc,
2257 .buf_free = bc_free,
2258 .send_request = bc_send_request,
2259 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2260 .close = bc_close,
2261 .destroy = bc_destroy,
2262 .print_stats = xs_tcp_print_stats,
2265 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2266 unsigned int slot_table_size)
2268 struct rpc_xprt *xprt;
2269 struct sock_xprt *new;
2271 if (args->addrlen > sizeof(xprt->addr)) {
2272 dprintk("RPC: xs_setup_xprt: address too large\n");
2273 return ERR_PTR(-EBADF);
2276 new = kzalloc(sizeof(*new), GFP_KERNEL);
2277 if (new == NULL) {
2278 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2279 "rpc_xprt\n");
2280 return ERR_PTR(-ENOMEM);
2282 xprt = &new->xprt;
2284 xprt->max_reqs = slot_table_size;
2285 xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
2286 if (xprt->slot == NULL) {
2287 kfree(xprt);
2288 dprintk("RPC: xs_setup_xprt: couldn't allocate slot "
2289 "table\n");
2290 return ERR_PTR(-ENOMEM);
2293 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2294 xprt->addrlen = args->addrlen;
2295 if (args->srcaddr)
2296 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2298 return xprt;
2301 static const struct rpc_timeout xs_udp_default_timeout = {
2302 .to_initval = 5 * HZ,
2303 .to_maxval = 30 * HZ,
2304 .to_increment = 5 * HZ,
2305 .to_retries = 5,
2309 * xs_setup_udp - Set up transport to use a UDP socket
2310 * @args: rpc transport creation arguments
2313 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2315 struct sockaddr *addr = args->dstaddr;
2316 struct rpc_xprt *xprt;
2317 struct sock_xprt *transport;
2318 struct rpc_xprt *ret;
2320 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
2321 if (IS_ERR(xprt))
2322 return xprt;
2323 transport = container_of(xprt, struct sock_xprt, xprt);
2325 xprt->prot = IPPROTO_UDP;
2326 xprt->tsh_size = 0;
2327 /* XXX: header size can vary due to auth type, IPv6, etc. */
2328 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2330 xprt->bind_timeout = XS_BIND_TO;
2331 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2332 xprt->idle_timeout = XS_IDLE_DISC_TO;
2334 xprt->ops = &xs_udp_ops;
2336 xprt->timeout = &xs_udp_default_timeout;
2338 switch (addr->sa_family) {
2339 case AF_INET:
2340 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2341 xprt_set_bound(xprt);
2343 INIT_DELAYED_WORK(&transport->connect_worker,
2344 xs_udp_connect_worker4);
2345 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2346 break;
2347 case AF_INET6:
2348 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2349 xprt_set_bound(xprt);
2351 INIT_DELAYED_WORK(&transport->connect_worker,
2352 xs_udp_connect_worker6);
2353 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2354 break;
2355 default:
2356 ret = ERR_PTR(-EAFNOSUPPORT);
2357 goto out_err;
2360 if (xprt_bound(xprt))
2361 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2362 xprt->address_strings[RPC_DISPLAY_ADDR],
2363 xprt->address_strings[RPC_DISPLAY_PORT],
2364 xprt->address_strings[RPC_DISPLAY_PROTO]);
2365 else
2366 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2367 xprt->address_strings[RPC_DISPLAY_ADDR],
2368 xprt->address_strings[RPC_DISPLAY_PROTO]);
2370 if (try_module_get(THIS_MODULE))
2371 return xprt;
2372 ret = ERR_PTR(-EINVAL);
2373 out_err:
2374 kfree(xprt->slot);
2375 kfree(xprt);
2376 return ret;
2379 static const struct rpc_timeout xs_tcp_default_timeout = {
2380 .to_initval = 60 * HZ,
2381 .to_maxval = 60 * HZ,
2382 .to_retries = 2,
2386 * xs_setup_tcp - Set up transport to use a TCP socket
2387 * @args: rpc transport creation arguments
2390 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2392 struct sockaddr *addr = args->dstaddr;
2393 struct rpc_xprt *xprt;
2394 struct sock_xprt *transport;
2395 struct rpc_xprt *ret;
2397 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
2398 if (IS_ERR(xprt))
2399 return xprt;
2400 transport = container_of(xprt, struct sock_xprt, xprt);
2402 xprt->prot = IPPROTO_TCP;
2403 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2404 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2406 xprt->bind_timeout = XS_BIND_TO;
2407 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2408 xprt->idle_timeout = XS_IDLE_DISC_TO;
2410 xprt->ops = &xs_tcp_ops;
2411 xprt->timeout = &xs_tcp_default_timeout;
2413 switch (addr->sa_family) {
2414 case AF_INET:
2415 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2416 xprt_set_bound(xprt);
2418 INIT_DELAYED_WORK(&transport->connect_worker,
2419 xs_tcp_connect_worker4);
2420 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2421 break;
2422 case AF_INET6:
2423 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2424 xprt_set_bound(xprt);
2426 INIT_DELAYED_WORK(&transport->connect_worker,
2427 xs_tcp_connect_worker6);
2428 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2429 break;
2430 default:
2431 ret = ERR_PTR(-EAFNOSUPPORT);
2432 goto out_err;
2435 if (xprt_bound(xprt))
2436 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2437 xprt->address_strings[RPC_DISPLAY_ADDR],
2438 xprt->address_strings[RPC_DISPLAY_PORT],
2439 xprt->address_strings[RPC_DISPLAY_PROTO]);
2440 else
2441 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2442 xprt->address_strings[RPC_DISPLAY_ADDR],
2443 xprt->address_strings[RPC_DISPLAY_PROTO]);
2446 if (try_module_get(THIS_MODULE))
2447 return xprt;
2448 ret = ERR_PTR(-EINVAL);
2449 out_err:
2450 kfree(xprt->slot);
2451 kfree(xprt);
2452 return ret;
2456 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2457 * @args: rpc transport creation arguments
2460 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2462 struct sockaddr *addr = args->dstaddr;
2463 struct rpc_xprt *xprt;
2464 struct sock_xprt *transport;
2465 struct svc_sock *bc_sock;
2466 struct rpc_xprt *ret;
2468 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
2469 if (IS_ERR(xprt))
2470 return xprt;
2471 transport = container_of(xprt, struct sock_xprt, xprt);
2473 xprt->prot = IPPROTO_TCP;
2474 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2475 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2476 xprt->timeout = &xs_tcp_default_timeout;
2478 /* backchannel */
2479 xprt_set_bound(xprt);
2480 xprt->bind_timeout = 0;
2481 xprt->reestablish_timeout = 0;
2482 xprt->idle_timeout = 0;
2485 * The backchannel uses the same socket connection as the
2486 * forechannel
2488 xprt->bc_xprt = args->bc_xprt;
2489 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2490 bc_sock->sk_bc_xprt = xprt;
2491 transport->sock = bc_sock->sk_sock;
2492 transport->inet = bc_sock->sk_sk;
2494 xprt->ops = &bc_tcp_ops;
2496 switch (addr->sa_family) {
2497 case AF_INET:
2498 xs_format_peer_addresses(xprt, "tcp",
2499 RPCBIND_NETID_TCP);
2500 break;
2501 case AF_INET6:
2502 xs_format_peer_addresses(xprt, "tcp",
2503 RPCBIND_NETID_TCP6);
2504 break;
2505 default:
2506 ret = ERR_PTR(-EAFNOSUPPORT);
2507 goto out_err;
2510 if (xprt_bound(xprt))
2511 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2512 xprt->address_strings[RPC_DISPLAY_ADDR],
2513 xprt->address_strings[RPC_DISPLAY_PORT],
2514 xprt->address_strings[RPC_DISPLAY_PROTO]);
2515 else
2516 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2517 xprt->address_strings[RPC_DISPLAY_ADDR],
2518 xprt->address_strings[RPC_DISPLAY_PROTO]);
2521 * Since we don't want connections for the backchannel, we set
2522 * the xprt status to connected
2524 xprt_set_connected(xprt);
2527 if (try_module_get(THIS_MODULE))
2528 return xprt;
2529 ret = ERR_PTR(-EINVAL);
2530 out_err:
2531 kfree(xprt->slot);
2532 kfree(xprt);
2533 return ret;
2536 static struct xprt_class xs_udp_transport = {
2537 .list = LIST_HEAD_INIT(xs_udp_transport.list),
2538 .name = "udp",
2539 .owner = THIS_MODULE,
2540 .ident = XPRT_TRANSPORT_UDP,
2541 .setup = xs_setup_udp,
2544 static struct xprt_class xs_tcp_transport = {
2545 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
2546 .name = "tcp",
2547 .owner = THIS_MODULE,
2548 .ident = XPRT_TRANSPORT_TCP,
2549 .setup = xs_setup_tcp,
2552 static struct xprt_class xs_bc_tcp_transport = {
2553 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
2554 .name = "tcp NFSv4.1 backchannel",
2555 .owner = THIS_MODULE,
2556 .ident = XPRT_TRANSPORT_BC_TCP,
2557 .setup = xs_setup_bc_tcp,
2561 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2564 int init_socket_xprt(void)
2566 #ifdef RPC_DEBUG
2567 if (!sunrpc_table_header)
2568 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2569 #endif
2571 xprt_register_transport(&xs_udp_transport);
2572 xprt_register_transport(&xs_tcp_transport);
2573 xprt_register_transport(&xs_bc_tcp_transport);
2575 return 0;
2579 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2582 void cleanup_socket_xprt(void)
2584 #ifdef RPC_DEBUG
2585 if (sunrpc_table_header) {
2586 unregister_sysctl_table(sunrpc_table_header);
2587 sunrpc_table_header = NULL;
2589 #endif
2591 xprt_unregister_transport(&xs_udp_transport);
2592 xprt_unregister_transport(&xs_tcp_transport);
2593 xprt_unregister_transport(&xs_bc_tcp_transport);
2596 static int param_set_uint_minmax(const char *val,
2597 const struct kernel_param *kp,
2598 unsigned int min, unsigned int max)
2600 unsigned long num;
2601 int ret;
2603 if (!val)
2604 return -EINVAL;
2605 ret = strict_strtoul(val, 0, &num);
2606 if (ret == -EINVAL || num < min || num > max)
2607 return -EINVAL;
2608 *((unsigned int *)kp->arg) = num;
2609 return 0;
2612 static int param_set_portnr(const char *val, const struct kernel_param *kp)
2614 return param_set_uint_minmax(val, kp,
2615 RPC_MIN_RESVPORT,
2616 RPC_MAX_RESVPORT);
2619 static struct kernel_param_ops param_ops_portnr = {
2620 .set = param_set_portnr,
2621 .get = param_get_uint,
2624 #define param_check_portnr(name, p) \
2625 __param_check(name, p, unsigned int);
2627 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
2628 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
2630 static int param_set_slot_table_size(const char *val,
2631 const struct kernel_param *kp)
2633 return param_set_uint_minmax(val, kp,
2634 RPC_MIN_SLOT_TABLE,
2635 RPC_MAX_SLOT_TABLE);
2638 static struct kernel_param_ops param_ops_slot_table_size = {
2639 .set = param_set_slot_table_size,
2640 .get = param_get_uint,
2643 #define param_check_slot_table_size(name, p) \
2644 __param_check(name, p, unsigned int);
2646 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
2647 slot_table_size, 0644);
2648 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
2649 slot_table_size, 0644);