Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/net/sunrpc/xprtsock.c
4 *
5 * Client-side transport implementation for sockets.
6 *
7 * TCP callback races fixes (C) 1998 Red Hat
8 * TCP send fixes (C) 1998 Red Hat
9 * TCP NFS related read + write fixes
10 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
11 *
12 * Rewrite of larges part of the code in order to stabilize TCP stuff.
13 * Fix behaviour when socket buffer is full.
14 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
15 *
16 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
17 *
18 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
19 * <gilles.quillard@bull.net>
20 */
21
22#include <linux/types.h>
23#include <linux/string.h>
24#include <linux/slab.h>
25#include <linux/module.h>
26#include <linux/capability.h>
27#include <linux/pagemap.h>
28#include <linux/errno.h>
29#include <linux/socket.h>
30#include <linux/in.h>
31#include <linux/net.h>
32#include <linux/mm.h>
33#include <linux/un.h>
34#include <linux/udp.h>
35#include <linux/tcp.h>
36#include <linux/sunrpc/clnt.h>
37#include <linux/sunrpc/addr.h>
38#include <linux/sunrpc/sched.h>
39#include <linux/sunrpc/svcsock.h>
40#include <linux/sunrpc/xprtsock.h>
41#include <linux/file.h>
42#ifdef CONFIG_SUNRPC_BACKCHANNEL
43#include <linux/sunrpc/bc_xprt.h>
44#endif
45
46#include <net/sock.h>
47#include <net/checksum.h>
48#include <net/udp.h>
49#include <net/tcp.h>
50#include <linux/bvec.h>
51#include <linux/highmem.h>
52#include <linux/uio.h>
53#include <linux/sched/mm.h>
54
55#include <trace/events/sunrpc.h>
56
57#include "socklib.h"
58#include "sunrpc.h"
59
60static void xs_close(struct rpc_xprt *xprt);
61static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
62 struct socket *sock);
63
64/*
65 * xprtsock tunables
66 */
67static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
68static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
69static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
70
71static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
72static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
73
74#define XS_TCP_LINGER_TO (15U * HZ)
75static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
76
77/*
78 * We can register our own files under /proc/sys/sunrpc by
79 * calling register_sysctl_table() again. The files in that
80 * directory become the union of all files registered there.
81 *
82 * We simply need to make sure that we don't collide with
83 * someone else's file names!
84 */
85
86static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
87static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
88static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
89static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
90static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
91
92static struct ctl_table_header *sunrpc_table_header;
93
94/*
95 * FIXME: changing the UDP slot table size should also resize the UDP
96 * socket buffers for existing UDP transports
97 */
98static struct ctl_table xs_tunables_table[] = {
99 {
100 .procname = "udp_slot_table_entries",
101 .data = &xprt_udp_slot_table_entries,
102 .maxlen = sizeof(unsigned int),
103 .mode = 0644,
104 .proc_handler = proc_dointvec_minmax,
105 .extra1 = &min_slot_table_size,
106 .extra2 = &max_slot_table_size
107 },
108 {
109 .procname = "tcp_slot_table_entries",
110 .data = &xprt_tcp_slot_table_entries,
111 .maxlen = sizeof(unsigned int),
112 .mode = 0644,
113 .proc_handler = proc_dointvec_minmax,
114 .extra1 = &min_slot_table_size,
115 .extra2 = &max_slot_table_size
116 },
117 {
118 .procname = "tcp_max_slot_table_entries",
119 .data = &xprt_max_tcp_slot_table_entries,
120 .maxlen = sizeof(unsigned int),
121 .mode = 0644,
122 .proc_handler = proc_dointvec_minmax,
123 .extra1 = &min_slot_table_size,
124 .extra2 = &max_tcp_slot_table_limit
125 },
126 {
127 .procname = "min_resvport",
128 .data = &xprt_min_resvport,
129 .maxlen = sizeof(unsigned int),
130 .mode = 0644,
131 .proc_handler = proc_dointvec_minmax,
132 .extra1 = &xprt_min_resvport_limit,
133 .extra2 = &xprt_max_resvport_limit
134 },
135 {
136 .procname = "max_resvport",
137 .data = &xprt_max_resvport,
138 .maxlen = sizeof(unsigned int),
139 .mode = 0644,
140 .proc_handler = proc_dointvec_minmax,
141 .extra1 = &xprt_min_resvport_limit,
142 .extra2 = &xprt_max_resvport_limit
143 },
144 {
145 .procname = "tcp_fin_timeout",
146 .data = &xs_tcp_fin_timeout,
147 .maxlen = sizeof(xs_tcp_fin_timeout),
148 .mode = 0644,
149 .proc_handler = proc_dointvec_jiffies,
150 },
151 { },
152};
153
154static struct ctl_table sunrpc_table[] = {
155 {
156 .procname = "sunrpc",
157 .mode = 0555,
158 .child = xs_tunables_table
159 },
160 { },
161};
162
163/*
164 * Wait duration for a reply from the RPC portmapper.
165 */
166#define XS_BIND_TO (60U * HZ)
167
168/*
169 * Delay if a UDP socket connect error occurs. This is most likely some
170 * kind of resource problem on the local host.
171 */
172#define XS_UDP_REEST_TO (2U * HZ)
173
174/*
175 * The reestablish timeout allows clients to delay for a bit before attempting
176 * to reconnect to a server that just dropped our connection.
177 *
178 * We implement an exponential backoff when trying to reestablish a TCP
179 * transport connection with the server. Some servers like to drop a TCP
180 * connection when they are overworked, so we start with a short timeout and
181 * increase over time if the server is down or not responding.
182 */
183#define XS_TCP_INIT_REEST_TO (3U * HZ)
184
185/*
186 * TCP idle timeout; client drops the transport socket if it is idle
187 * for this long. Note that we also timeout UDP sockets to prevent
188 * holding port numbers when there is no RPC traffic.
189 */
190#define XS_IDLE_DISC_TO (5U * 60 * HZ)
191
192#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
193# undef RPC_DEBUG_DATA
194# define RPCDBG_FACILITY RPCDBG_TRANS
195#endif
196
197#ifdef RPC_DEBUG_DATA
198static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
199{
200 u8 *buf = (u8 *) packet;
201 int j;
202
203 dprintk("RPC: %s\n", msg);
204 for (j = 0; j < count && j < 128; j += 4) {
205 if (!(j & 31)) {
206 if (j)
207 dprintk("\n");
208 dprintk("0x%04x ", j);
209 }
210 dprintk("%02x%02x%02x%02x ",
211 buf[j], buf[j+1], buf[j+2], buf[j+3]);
212 }
213 dprintk("\n");
214}
215#else
216static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
217{
218 /* NOP */
219}
220#endif
221
222static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
223{
224 return (struct rpc_xprt *) sk->sk_user_data;
225}
226
227static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
228{
229 return (struct sockaddr *) &xprt->addr;
230}
231
232static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
233{
234 return (struct sockaddr_un *) &xprt->addr;
235}
236
237static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
238{
239 return (struct sockaddr_in *) &xprt->addr;
240}
241
242static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
243{
244 return (struct sockaddr_in6 *) &xprt->addr;
245}
246
247static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
248{
249 struct sockaddr *sap = xs_addr(xprt);
250 struct sockaddr_in6 *sin6;
251 struct sockaddr_in *sin;
252 struct sockaddr_un *sun;
253 char buf[128];
254
255 switch (sap->sa_family) {
256 case AF_LOCAL:
257 sun = xs_addr_un(xprt);
258 strlcpy(buf, sun->sun_path, sizeof(buf));
259 xprt->address_strings[RPC_DISPLAY_ADDR] =
260 kstrdup(buf, GFP_KERNEL);
261 break;
262 case AF_INET:
263 (void)rpc_ntop(sap, buf, sizeof(buf));
264 xprt->address_strings[RPC_DISPLAY_ADDR] =
265 kstrdup(buf, GFP_KERNEL);
266 sin = xs_addr_in(xprt);
267 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
268 break;
269 case AF_INET6:
270 (void)rpc_ntop(sap, buf, sizeof(buf));
271 xprt->address_strings[RPC_DISPLAY_ADDR] =
272 kstrdup(buf, GFP_KERNEL);
273 sin6 = xs_addr_in6(xprt);
274 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
275 break;
276 default:
277 BUG();
278 }
279
280 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
281}
282
283static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
284{
285 struct sockaddr *sap = xs_addr(xprt);
286 char buf[128];
287
288 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
289 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
290
291 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
292 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
293}
294
295static void xs_format_peer_addresses(struct rpc_xprt *xprt,
296 const char *protocol,
297 const char *netid)
298{
299 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
300 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
301 xs_format_common_peer_addresses(xprt);
302 xs_format_common_peer_ports(xprt);
303}
304
305static void xs_update_peer_port(struct rpc_xprt *xprt)
306{
307 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
308 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
309
310 xs_format_common_peer_ports(xprt);
311}
312
313static void xs_free_peer_addresses(struct rpc_xprt *xprt)
314{
315 unsigned int i;
316
317 for (i = 0; i < RPC_DISPLAY_MAX; i++)
318 switch (i) {
319 case RPC_DISPLAY_PROTO:
320 case RPC_DISPLAY_NETID:
321 continue;
322 default:
323 kfree(xprt->address_strings[i]);
324 }
325}
326
327static size_t
328xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp)
329{
330 size_t i,n;
331
332 if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES))
333 return want;
334 n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT;
335 for (i = 0; i < n; i++) {
336 if (buf->pages[i])
337 continue;
338 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp);
339 if (!buf->pages[i]) {
340 i *= PAGE_SIZE;
341 return i > buf->page_base ? i - buf->page_base : 0;
342 }
343 }
344 return want;
345}
346
347static ssize_t
348xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek)
349{
350 ssize_t ret;
351 if (seek != 0)
352 iov_iter_advance(&msg->msg_iter, seek);
353 ret = sock_recvmsg(sock, msg, flags);
354 return ret > 0 ? ret + seek : ret;
355}
356
357static ssize_t
358xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags,
359 struct kvec *kvec, size_t count, size_t seek)
360{
361 iov_iter_kvec(&msg->msg_iter, READ, kvec, 1, count);
362 return xs_sock_recvmsg(sock, msg, flags, seek);
363}
364
365static ssize_t
366xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags,
367 struct bio_vec *bvec, unsigned long nr, size_t count,
368 size_t seek)
369{
370 iov_iter_bvec(&msg->msg_iter, READ, bvec, nr, count);
371 return xs_sock_recvmsg(sock, msg, flags, seek);
372}
373
374static ssize_t
375xs_read_discard(struct socket *sock, struct msghdr *msg, int flags,
376 size_t count)
377{
378 iov_iter_discard(&msg->msg_iter, READ, count);
379 return sock_recvmsg(sock, msg, flags);
380}
381
382#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
383static void
384xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
385{
386 struct bvec_iter bi = {
387 .bi_size = count,
388 };
389 struct bio_vec bv;
390
391 bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
392 for_each_bvec(bv, bvec, bi, bi)
393 flush_dcache_page(bv.bv_page);
394}
395#else
396static inline void
397xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
398{
399}
400#endif
401
402static ssize_t
403xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
404 struct xdr_buf *buf, size_t count, size_t seek, size_t *read)
405{
406 size_t want, seek_init = seek, offset = 0;
407 ssize_t ret;
408
409 want = min_t(size_t, count, buf->head[0].iov_len);
410 if (seek < want) {
411 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek);
412 if (ret <= 0)
413 goto sock_err;
414 offset += ret;
415 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
416 goto out;
417 if (ret != want)
418 goto out;
419 seek = 0;
420 } else {
421 seek -= want;
422 offset += want;
423 }
424
425 want = xs_alloc_sparse_pages(buf,
426 min_t(size_t, count - offset, buf->page_len),
427 GFP_KERNEL);
428 if (seek < want) {
429 ret = xs_read_bvec(sock, msg, flags, buf->bvec,
430 xdr_buf_pagecount(buf),
431 want + buf->page_base,
432 seek + buf->page_base);
433 if (ret <= 0)
434 goto sock_err;
435 xs_flush_bvec(buf->bvec, ret, seek + buf->page_base);
436 offset += ret - buf->page_base;
437 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
438 goto out;
439 if (ret != want)
440 goto out;
441 seek = 0;
442 } else {
443 seek -= want;
444 offset += want;
445 }
446
447 want = min_t(size_t, count - offset, buf->tail[0].iov_len);
448 if (seek < want) {
449 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek);
450 if (ret <= 0)
451 goto sock_err;
452 offset += ret;
453 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
454 goto out;
455 if (ret != want)
456 goto out;
457 } else if (offset < seek_init)
458 offset = seek_init;
459 ret = -EMSGSIZE;
460out:
461 *read = offset - seek_init;
462 return ret;
463sock_err:
464 offset += seek;
465 goto out;
466}
467
468static void
469xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf)
470{
471 if (!transport->recv.copied) {
472 if (buf->head[0].iov_len >= transport->recv.offset)
473 memcpy(buf->head[0].iov_base,
474 &transport->recv.xid,
475 transport->recv.offset);
476 transport->recv.copied = transport->recv.offset;
477 }
478}
479
480static bool
481xs_read_stream_request_done(struct sock_xprt *transport)
482{
483 return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT);
484}
485
486static void
487xs_read_stream_check_eor(struct sock_xprt *transport,
488 struct msghdr *msg)
489{
490 if (xs_read_stream_request_done(transport))
491 msg->msg_flags |= MSG_EOR;
492}
493
494static ssize_t
495xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg,
496 int flags, struct rpc_rqst *req)
497{
498 struct xdr_buf *buf = &req->rq_private_buf;
499 size_t want, read;
500 ssize_t ret;
501
502 xs_read_header(transport, buf);
503
504 want = transport->recv.len - transport->recv.offset;
505 if (want != 0) {
506 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf,
507 transport->recv.copied + want,
508 transport->recv.copied,
509 &read);
510 transport->recv.offset += read;
511 transport->recv.copied += read;
512 }
513
514 if (transport->recv.offset == transport->recv.len)
515 xs_read_stream_check_eor(transport, msg);
516
517 if (want == 0)
518 return 0;
519
520 switch (ret) {
521 default:
522 break;
523 case -EFAULT:
524 case -EMSGSIZE:
525 msg->msg_flags |= MSG_TRUNC;
526 return read;
527 case 0:
528 return -ESHUTDOWN;
529 }
530 return ret < 0 ? ret : read;
531}
532
533static size_t
534xs_read_stream_headersize(bool isfrag)
535{
536 if (isfrag)
537 return sizeof(__be32);
538 return 3 * sizeof(__be32);
539}
540
541static ssize_t
542xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg,
543 int flags, size_t want, size_t seek)
544{
545 struct kvec kvec = {
546 .iov_base = &transport->recv.fraghdr,
547 .iov_len = want,
548 };
549 return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek);
550}
551
552#if defined(CONFIG_SUNRPC_BACKCHANNEL)
553static ssize_t
554xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
555{
556 struct rpc_xprt *xprt = &transport->xprt;
557 struct rpc_rqst *req;
558 ssize_t ret;
559
560 /* Look up and lock the request corresponding to the given XID */
561 req = xprt_lookup_bc_request(xprt, transport->recv.xid);
562 if (!req) {
563 printk(KERN_WARNING "Callback slot table overflowed\n");
564 return -ESHUTDOWN;
565 }
566 if (transport->recv.copied && !req->rq_private_buf.len)
567 return -ESHUTDOWN;
568
569 ret = xs_read_stream_request(transport, msg, flags, req);
570 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
571 xprt_complete_bc_request(req, transport->recv.copied);
572 else
573 req->rq_private_buf.len = transport->recv.copied;
574
575 return ret;
576}
577#else /* CONFIG_SUNRPC_BACKCHANNEL */
578static ssize_t
579xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
580{
581 return -ESHUTDOWN;
582}
583#endif /* CONFIG_SUNRPC_BACKCHANNEL */
584
585static ssize_t
586xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags)
587{
588 struct rpc_xprt *xprt = &transport->xprt;
589 struct rpc_rqst *req;
590 ssize_t ret = 0;
591
592 /* Look up and lock the request corresponding to the given XID */
593 spin_lock(&xprt->queue_lock);
594 req = xprt_lookup_rqst(xprt, transport->recv.xid);
595 if (!req || (transport->recv.copied && !req->rq_private_buf.len)) {
596 msg->msg_flags |= MSG_TRUNC;
597 goto out;
598 }
599 xprt_pin_rqst(req);
600 spin_unlock(&xprt->queue_lock);
601
602 ret = xs_read_stream_request(transport, msg, flags, req);
603
604 spin_lock(&xprt->queue_lock);
605 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
606 xprt_complete_rqst(req->rq_task, transport->recv.copied);
607 else
608 req->rq_private_buf.len = transport->recv.copied;
609 xprt_unpin_rqst(req);
610out:
611 spin_unlock(&xprt->queue_lock);
612 return ret;
613}
614
615static ssize_t
616xs_read_stream(struct sock_xprt *transport, int flags)
617{
618 struct msghdr msg = { 0 };
619 size_t want, read = 0;
620 ssize_t ret = 0;
621
622 if (transport->recv.len == 0) {
623 want = xs_read_stream_headersize(transport->recv.copied != 0);
624 ret = xs_read_stream_header(transport, &msg, flags, want,
625 transport->recv.offset);
626 if (ret <= 0)
627 goto out_err;
628 transport->recv.offset = ret;
629 if (transport->recv.offset != want)
630 return transport->recv.offset;
631 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) &
632 RPC_FRAGMENT_SIZE_MASK;
633 transport->recv.offset -= sizeof(transport->recv.fraghdr);
634 read = ret;
635 }
636
637 switch (be32_to_cpu(transport->recv.calldir)) {
638 default:
639 msg.msg_flags |= MSG_TRUNC;
640 break;
641 case RPC_CALL:
642 ret = xs_read_stream_call(transport, &msg, flags);
643 break;
644 case RPC_REPLY:
645 ret = xs_read_stream_reply(transport, &msg, flags);
646 }
647 if (msg.msg_flags & MSG_TRUNC) {
648 transport->recv.calldir = cpu_to_be32(-1);
649 transport->recv.copied = -1;
650 }
651 if (ret < 0)
652 goto out_err;
653 read += ret;
654 if (transport->recv.offset < transport->recv.len) {
655 if (!(msg.msg_flags & MSG_TRUNC))
656 return read;
657 msg.msg_flags = 0;
658 ret = xs_read_discard(transport->sock, &msg, flags,
659 transport->recv.len - transport->recv.offset);
660 if (ret <= 0)
661 goto out_err;
662 transport->recv.offset += ret;
663 read += ret;
664 if (transport->recv.offset != transport->recv.len)
665 return read;
666 }
667 if (xs_read_stream_request_done(transport)) {
668 trace_xs_stream_read_request(transport);
669 transport->recv.copied = 0;
670 }
671 transport->recv.offset = 0;
672 transport->recv.len = 0;
673 return read;
674out_err:
675 return ret != 0 ? ret : -ESHUTDOWN;
676}
677
678static __poll_t xs_poll_socket(struct sock_xprt *transport)
679{
680 return transport->sock->ops->poll(transport->file, transport->sock,
681 NULL);
682}
683
684static bool xs_poll_socket_readable(struct sock_xprt *transport)
685{
686 __poll_t events = xs_poll_socket(transport);
687
688 return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP);
689}
690
691static void xs_poll_check_readable(struct sock_xprt *transport)
692{
693
694 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
695 if (!xs_poll_socket_readable(transport))
696 return;
697 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
698 queue_work(xprtiod_workqueue, &transport->recv_worker);
699}
700
701static void xs_stream_data_receive(struct sock_xprt *transport)
702{
703 size_t read = 0;
704 ssize_t ret = 0;
705
706 mutex_lock(&transport->recv_mutex);
707 if (transport->sock == NULL)
708 goto out;
709 for (;;) {
710 ret = xs_read_stream(transport, MSG_DONTWAIT);
711 if (ret < 0)
712 break;
713 read += ret;
714 cond_resched();
715 }
716 if (ret == -ESHUTDOWN)
717 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
718 else
719 xs_poll_check_readable(transport);
720out:
721 mutex_unlock(&transport->recv_mutex);
722 trace_xs_stream_read_data(&transport->xprt, ret, read);
723}
724
725static void xs_stream_data_receive_workfn(struct work_struct *work)
726{
727 struct sock_xprt *transport =
728 container_of(work, struct sock_xprt, recv_worker);
729 unsigned int pflags = memalloc_nofs_save();
730
731 xs_stream_data_receive(transport);
732 memalloc_nofs_restore(pflags);
733}
734
735static void
736xs_stream_reset_connect(struct sock_xprt *transport)
737{
738 transport->recv.offset = 0;
739 transport->recv.len = 0;
740 transport->recv.copied = 0;
741 transport->xmit.offset = 0;
742}
743
744static void
745xs_stream_start_connect(struct sock_xprt *transport)
746{
747 transport->xprt.stat.connect_count++;
748 transport->xprt.stat.connect_start = jiffies;
749}
750
751#define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
752
753/**
754 * xs_nospace - handle transmit was incomplete
755 * @req: pointer to RPC request
756 *
757 */
758static int xs_nospace(struct rpc_rqst *req)
759{
760 struct rpc_xprt *xprt = req->rq_xprt;
761 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
762 struct sock *sk = transport->inet;
763 int ret = -EAGAIN;
764
765 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
766 req->rq_task->tk_pid,
767 req->rq_slen - transport->xmit.offset,
768 req->rq_slen);
769
770 /* Protect against races with write_space */
771 spin_lock(&xprt->transport_lock);
772
773 /* Don't race with disconnect */
774 if (xprt_connected(xprt)) {
775 /* wait for more buffer space */
776 sk->sk_write_pending++;
777 xprt_wait_for_buffer_space(xprt);
778 } else
779 ret = -ENOTCONN;
780
781 spin_unlock(&xprt->transport_lock);
782
783 /* Race breaker in case memory is freed before above code is called */
784 if (ret == -EAGAIN) {
785 struct socket_wq *wq;
786
787 rcu_read_lock();
788 wq = rcu_dereference(sk->sk_wq);
789 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
790 rcu_read_unlock();
791
792 sk->sk_write_space(sk);
793 }
794 return ret;
795}
796
797static void
798xs_stream_prepare_request(struct rpc_rqst *req)
799{
800 xdr_free_bvec(&req->rq_rcv_buf);
801 req->rq_task->tk_status = xdr_alloc_bvec(&req->rq_rcv_buf, GFP_KERNEL);
802}
803
804/*
805 * Determine if the previous message in the stream was aborted before it
806 * could complete transmission.
807 */
808static bool
809xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req)
810{
811 return transport->xmit.offset != 0 && req->rq_bytes_sent == 0;
812}
813
814/*
815 * Return the stream record marker field for a record of length < 2^31-1
816 */
817static rpc_fraghdr
818xs_stream_record_marker(struct xdr_buf *xdr)
819{
820 if (!xdr->len)
821 return 0;
822 return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len);
823}
824
825/**
826 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
827 * @req: pointer to RPC request
828 *
829 * Return values:
830 * 0: The request has been sent
831 * EAGAIN: The socket was blocked, please call again later to
832 * complete the request
833 * ENOTCONN: Caller needs to invoke connect logic then call again
834 * other: Some other error occured, the request was not sent
835 */
836static int xs_local_send_request(struct rpc_rqst *req)
837{
838 struct rpc_xprt *xprt = req->rq_xprt;
839 struct sock_xprt *transport =
840 container_of(xprt, struct sock_xprt, xprt);
841 struct xdr_buf *xdr = &req->rq_snd_buf;
842 rpc_fraghdr rm = xs_stream_record_marker(xdr);
843 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
844 struct msghdr msg = {
845 .msg_flags = XS_SENDMSG_FLAGS,
846 };
847 unsigned int sent;
848 int status;
849
850 /* Close the stream if the previous transmission was incomplete */
851 if (xs_send_request_was_aborted(transport, req)) {
852 xs_close(xprt);
853 return -ENOTCONN;
854 }
855
856 xs_pktdump("packet data:",
857 req->rq_svec->iov_base, req->rq_svec->iov_len);
858
859 req->rq_xtime = ktime_get();
860 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
861 transport->xmit.offset, rm, &sent);
862 dprintk("RPC: %s(%u) = %d\n",
863 __func__, xdr->len - transport->xmit.offset, status);
864
865 if (status == -EAGAIN && sock_writeable(transport->inet))
866 status = -ENOBUFS;
867
868 if (likely(sent > 0) || status == 0) {
869 transport->xmit.offset += sent;
870 req->rq_bytes_sent = transport->xmit.offset;
871 if (likely(req->rq_bytes_sent >= msglen)) {
872 req->rq_xmit_bytes_sent += transport->xmit.offset;
873 transport->xmit.offset = 0;
874 return 0;
875 }
876 status = -EAGAIN;
877 }
878
879 switch (status) {
880 case -ENOBUFS:
881 break;
882 case -EAGAIN:
883 status = xs_nospace(req);
884 break;
885 default:
886 dprintk("RPC: sendmsg returned unrecognized error %d\n",
887 -status);
888 fallthrough;
889 case -EPIPE:
890 xs_close(xprt);
891 status = -ENOTCONN;
892 }
893
894 return status;
895}
896
897/**
898 * xs_udp_send_request - write an RPC request to a UDP socket
899 * @req: pointer to RPC request
900 *
901 * Return values:
902 * 0: The request has been sent
903 * EAGAIN: The socket was blocked, please call again later to
904 * complete the request
905 * ENOTCONN: Caller needs to invoke connect logic then call again
906 * other: Some other error occurred, the request was not sent
907 */
908static int xs_udp_send_request(struct rpc_rqst *req)
909{
910 struct rpc_xprt *xprt = req->rq_xprt;
911 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
912 struct xdr_buf *xdr = &req->rq_snd_buf;
913 struct msghdr msg = {
914 .msg_name = xs_addr(xprt),
915 .msg_namelen = xprt->addrlen,
916 .msg_flags = XS_SENDMSG_FLAGS,
917 };
918 unsigned int sent;
919 int status;
920
921 xs_pktdump("packet data:",
922 req->rq_svec->iov_base,
923 req->rq_svec->iov_len);
924
925 if (!xprt_bound(xprt))
926 return -ENOTCONN;
927
928 if (!xprt_request_get_cong(xprt, req))
929 return -EBADSLT;
930
931 req->rq_xtime = ktime_get();
932 status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, 0, &sent);
933
934 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
935 xdr->len, status);
936
937 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
938 if (status == -EPERM)
939 goto process_status;
940
941 if (status == -EAGAIN && sock_writeable(transport->inet))
942 status = -ENOBUFS;
943
944 if (sent > 0 || status == 0) {
945 req->rq_xmit_bytes_sent += sent;
946 if (sent >= req->rq_slen)
947 return 0;
948 /* Still some bytes left; set up for a retry later. */
949 status = -EAGAIN;
950 }
951
952process_status:
953 switch (status) {
954 case -ENOTSOCK:
955 status = -ENOTCONN;
956 /* Should we call xs_close() here? */
957 break;
958 case -EAGAIN:
959 status = xs_nospace(req);
960 break;
961 case -ENETUNREACH:
962 case -ENOBUFS:
963 case -EPIPE:
964 case -ECONNREFUSED:
965 case -EPERM:
966 /* When the server has died, an ICMP port unreachable message
967 * prompts ECONNREFUSED. */
968 break;
969 default:
970 dprintk("RPC: sendmsg returned unrecognized error %d\n",
971 -status);
972 }
973
974 return status;
975}
976
977/**
978 * xs_tcp_send_request - write an RPC request to a TCP socket
979 * @req: pointer to RPC request
980 *
981 * Return values:
982 * 0: The request has been sent
983 * EAGAIN: The socket was blocked, please call again later to
984 * complete the request
985 * ENOTCONN: Caller needs to invoke connect logic then call again
986 * other: Some other error occurred, the request was not sent
987 *
988 * XXX: In the case of soft timeouts, should we eventually give up
989 * if sendmsg is not able to make progress?
990 */
991static int xs_tcp_send_request(struct rpc_rqst *req)
992{
993 struct rpc_xprt *xprt = req->rq_xprt;
994 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
995 struct xdr_buf *xdr = &req->rq_snd_buf;
996 rpc_fraghdr rm = xs_stream_record_marker(xdr);
997 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
998 struct msghdr msg = {
999 .msg_flags = XS_SENDMSG_FLAGS,
1000 };
1001 bool vm_wait = false;
1002 unsigned int sent;
1003 int status;
1004
1005 /* Close the stream if the previous transmission was incomplete */
1006 if (xs_send_request_was_aborted(transport, req)) {
1007 if (transport->sock != NULL)
1008 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
1009 return -ENOTCONN;
1010 }
1011
1012 xs_pktdump("packet data:",
1013 req->rq_svec->iov_base,
1014 req->rq_svec->iov_len);
1015
1016 if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
1017 xs_tcp_set_socket_timeouts(xprt, transport->sock);
1018
1019 /* Continue transmitting the packet/record. We must be careful
1020 * to cope with writespace callbacks arriving _after_ we have
1021 * called sendmsg(). */
1022 req->rq_xtime = ktime_get();
1023 while (1) {
1024 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
1025 transport->xmit.offset, rm, &sent);
1026
1027 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
1028 xdr->len - transport->xmit.offset, status);
1029
1030 /* If we've sent the entire packet, immediately
1031 * reset the count of bytes sent. */
1032 transport->xmit.offset += sent;
1033 req->rq_bytes_sent = transport->xmit.offset;
1034 if (likely(req->rq_bytes_sent >= msglen)) {
1035 req->rq_xmit_bytes_sent += transport->xmit.offset;
1036 transport->xmit.offset = 0;
1037 return 0;
1038 }
1039
1040 WARN_ON_ONCE(sent == 0 && status == 0);
1041
1042 if (status == -EAGAIN ) {
1043 /*
1044 * Return EAGAIN if we're sure we're hitting the
1045 * socket send buffer limits.
1046 */
1047 if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
1048 break;
1049 /*
1050 * Did we hit a memory allocation failure?
1051 */
1052 if (sent == 0) {
1053 status = -ENOBUFS;
1054 if (vm_wait)
1055 break;
1056 /* Retry, knowing now that we're below the
1057 * socket send buffer limit
1058 */
1059 vm_wait = true;
1060 }
1061 continue;
1062 }
1063 if (status < 0)
1064 break;
1065 vm_wait = false;
1066 }
1067
1068 switch (status) {
1069 case -ENOTSOCK:
1070 status = -ENOTCONN;
1071 /* Should we call xs_close() here? */
1072 break;
1073 case -EAGAIN:
1074 status = xs_nospace(req);
1075 break;
1076 case -ECONNRESET:
1077 case -ECONNREFUSED:
1078 case -ENOTCONN:
1079 case -EADDRINUSE:
1080 case -ENOBUFS:
1081 case -EPIPE:
1082 break;
1083 default:
1084 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1085 -status);
1086 }
1087
1088 return status;
1089}
1090
1091static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1092{
1093 transport->old_data_ready = sk->sk_data_ready;
1094 transport->old_state_change = sk->sk_state_change;
1095 transport->old_write_space = sk->sk_write_space;
1096 transport->old_error_report = sk->sk_error_report;
1097}
1098
1099static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1100{
1101 sk->sk_data_ready = transport->old_data_ready;
1102 sk->sk_state_change = transport->old_state_change;
1103 sk->sk_write_space = transport->old_write_space;
1104 sk->sk_error_report = transport->old_error_report;
1105}
1106
1107static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
1108{
1109 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1110
1111 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1112 clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state);
1113 clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state);
1114 clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state);
1115}
1116
1117static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr)
1118{
1119 set_bit(nr, &transport->sock_state);
1120 queue_work(xprtiod_workqueue, &transport->error_worker);
1121}
1122
1123static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1124{
1125 smp_mb__before_atomic();
1126 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1127 clear_bit(XPRT_CLOSING, &xprt->state);
1128 xs_sock_reset_state_flags(xprt);
1129 smp_mb__after_atomic();
1130}
1131
1132/**
1133 * xs_error_report - callback to handle TCP socket state errors
1134 * @sk: socket
1135 *
1136 * Note: we don't call sock_error() since there may be a rpc_task
1137 * using the socket, and so we don't want to clear sk->sk_err.
1138 */
1139static void xs_error_report(struct sock *sk)
1140{
1141 struct sock_xprt *transport;
1142 struct rpc_xprt *xprt;
1143
1144 read_lock_bh(&sk->sk_callback_lock);
1145 if (!(xprt = xprt_from_sock(sk)))
1146 goto out;
1147
1148 transport = container_of(xprt, struct sock_xprt, xprt);
1149 transport->xprt_err = -sk->sk_err;
1150 if (transport->xprt_err == 0)
1151 goto out;
1152 dprintk("RPC: xs_error_report client %p, error=%d...\n",
1153 xprt, -transport->xprt_err);
1154 trace_rpc_socket_error(xprt, sk->sk_socket, transport->xprt_err);
1155
1156 /* barrier ensures xprt_err is set before XPRT_SOCK_WAKE_ERROR */
1157 smp_mb__before_atomic();
1158 xs_run_error_worker(transport, XPRT_SOCK_WAKE_ERROR);
1159 out:
1160 read_unlock_bh(&sk->sk_callback_lock);
1161}
1162
1163static void xs_reset_transport(struct sock_xprt *transport)
1164{
1165 struct socket *sock = transport->sock;
1166 struct sock *sk = transport->inet;
1167 struct rpc_xprt *xprt = &transport->xprt;
1168 struct file *filp = transport->file;
1169
1170 if (sk == NULL)
1171 return;
1172
1173 if (atomic_read(&transport->xprt.swapper))
1174 sk_clear_memalloc(sk);
1175
1176 kernel_sock_shutdown(sock, SHUT_RDWR);
1177
1178 mutex_lock(&transport->recv_mutex);
1179 write_lock_bh(&sk->sk_callback_lock);
1180 transport->inet = NULL;
1181 transport->sock = NULL;
1182 transport->file = NULL;
1183
1184 sk->sk_user_data = NULL;
1185
1186 xs_restore_old_callbacks(transport, sk);
1187 xprt_clear_connected(xprt);
1188 write_unlock_bh(&sk->sk_callback_lock);
1189 xs_sock_reset_connection_flags(xprt);
1190 /* Reset stream record info */
1191 xs_stream_reset_connect(transport);
1192 mutex_unlock(&transport->recv_mutex);
1193
1194 trace_rpc_socket_close(xprt, sock);
1195 fput(filp);
1196
1197 xprt_disconnect_done(xprt);
1198}
1199
1200/**
1201 * xs_close - close a socket
1202 * @xprt: transport
1203 *
1204 * This is used when all requests are complete; ie, no DRC state remains
1205 * on the server we want to save.
1206 *
1207 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1208 * xs_reset_transport() zeroing the socket from underneath a writer.
1209 */
1210static void xs_close(struct rpc_xprt *xprt)
1211{
1212 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1213
1214 dprintk("RPC: xs_close xprt %p\n", xprt);
1215
1216 xs_reset_transport(transport);
1217 xprt->reestablish_timeout = 0;
1218}
1219
1220static void xs_inject_disconnect(struct rpc_xprt *xprt)
1221{
1222 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
1223 xprt);
1224 xprt_disconnect_done(xprt);
1225}
1226
1227static void xs_xprt_free(struct rpc_xprt *xprt)
1228{
1229 xs_free_peer_addresses(xprt);
1230 xprt_free(xprt);
1231}
1232
1233/**
1234 * xs_destroy - prepare to shutdown a transport
1235 * @xprt: doomed transport
1236 *
1237 */
1238static void xs_destroy(struct rpc_xprt *xprt)
1239{
1240 struct sock_xprt *transport = container_of(xprt,
1241 struct sock_xprt, xprt);
1242 dprintk("RPC: xs_destroy xprt %p\n", xprt);
1243
1244 cancel_delayed_work_sync(&transport->connect_worker);
1245 xs_close(xprt);
1246 cancel_work_sync(&transport->recv_worker);
1247 cancel_work_sync(&transport->error_worker);
1248 xs_xprt_free(xprt);
1249 module_put(THIS_MODULE);
1250}
1251
1252/**
1253 * xs_udp_data_read_skb - receive callback for UDP sockets
1254 * @xprt: transport
1255 * @sk: socket
1256 * @skb: skbuff
1257 *
1258 */
1259static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1260 struct sock *sk,
1261 struct sk_buff *skb)
1262{
1263 struct rpc_task *task;
1264 struct rpc_rqst *rovr;
1265 int repsize, copied;
1266 u32 _xid;
1267 __be32 *xp;
1268
1269 repsize = skb->len;
1270 if (repsize < 4) {
1271 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
1272 return;
1273 }
1274
1275 /* Copy the XID from the skb... */
1276 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1277 if (xp == NULL)
1278 return;
1279
1280 /* Look up and lock the request corresponding to the given XID */
1281 spin_lock(&xprt->queue_lock);
1282 rovr = xprt_lookup_rqst(xprt, *xp);
1283 if (!rovr)
1284 goto out_unlock;
1285 xprt_pin_rqst(rovr);
1286 xprt_update_rtt(rovr->rq_task);
1287 spin_unlock(&xprt->queue_lock);
1288 task = rovr->rq_task;
1289
1290 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1291 copied = repsize;
1292
1293 /* Suck it into the iovec, verify checksum if not done by hw. */
1294 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1295 spin_lock(&xprt->queue_lock);
1296 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1297 goto out_unpin;
1298 }
1299
1300
1301 spin_lock(&xprt->transport_lock);
1302 xprt_adjust_cwnd(xprt, task, copied);
1303 spin_unlock(&xprt->transport_lock);
1304 spin_lock(&xprt->queue_lock);
1305 xprt_complete_rqst(task, copied);
1306 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1307out_unpin:
1308 xprt_unpin_rqst(rovr);
1309 out_unlock:
1310 spin_unlock(&xprt->queue_lock);
1311}
1312
1313static void xs_udp_data_receive(struct sock_xprt *transport)
1314{
1315 struct sk_buff *skb;
1316 struct sock *sk;
1317 int err;
1318
1319 mutex_lock(&transport->recv_mutex);
1320 sk = transport->inet;
1321 if (sk == NULL)
1322 goto out;
1323 for (;;) {
1324 skb = skb_recv_udp(sk, 0, 1, &err);
1325 if (skb == NULL)
1326 break;
1327 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1328 consume_skb(skb);
1329 cond_resched();
1330 }
1331 xs_poll_check_readable(transport);
1332out:
1333 mutex_unlock(&transport->recv_mutex);
1334}
1335
1336static void xs_udp_data_receive_workfn(struct work_struct *work)
1337{
1338 struct sock_xprt *transport =
1339 container_of(work, struct sock_xprt, recv_worker);
1340 unsigned int pflags = memalloc_nofs_save();
1341
1342 xs_udp_data_receive(transport);
1343 memalloc_nofs_restore(pflags);
1344}
1345
1346/**
1347 * xs_data_ready - "data ready" callback for UDP sockets
1348 * @sk: socket with data to read
1349 *
1350 */
1351static void xs_data_ready(struct sock *sk)
1352{
1353 struct rpc_xprt *xprt;
1354
1355 read_lock_bh(&sk->sk_callback_lock);
1356 dprintk("RPC: xs_data_ready...\n");
1357 xprt = xprt_from_sock(sk);
1358 if (xprt != NULL) {
1359 struct sock_xprt *transport = container_of(xprt,
1360 struct sock_xprt, xprt);
1361 transport->old_data_ready(sk);
1362 /* Any data means we had a useful conversation, so
1363 * then we don't need to delay the next reconnect
1364 */
1365 if (xprt->reestablish_timeout)
1366 xprt->reestablish_timeout = 0;
1367 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1368 queue_work(xprtiod_workqueue, &transport->recv_worker);
1369 }
1370 read_unlock_bh(&sk->sk_callback_lock);
1371}
1372
1373/*
1374 * Helper function to force a TCP close if the server is sending
1375 * junk and/or it has put us in CLOSE_WAIT
1376 */
1377static void xs_tcp_force_close(struct rpc_xprt *xprt)
1378{
1379 xprt_force_disconnect(xprt);
1380}
1381
1382#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1383static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1384{
1385 return PAGE_SIZE;
1386}
1387#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1388
1389/**
1390 * xs_tcp_state_change - callback to handle TCP socket state changes
1391 * @sk: socket whose state has changed
1392 *
1393 */
1394static void xs_tcp_state_change(struct sock *sk)
1395{
1396 struct rpc_xprt *xprt;
1397 struct sock_xprt *transport;
1398
1399 read_lock_bh(&sk->sk_callback_lock);
1400 if (!(xprt = xprt_from_sock(sk)))
1401 goto out;
1402 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1403 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1404 sk->sk_state, xprt_connected(xprt),
1405 sock_flag(sk, SOCK_DEAD),
1406 sock_flag(sk, SOCK_ZAPPED),
1407 sk->sk_shutdown);
1408
1409 transport = container_of(xprt, struct sock_xprt, xprt);
1410 trace_rpc_socket_state_change(xprt, sk->sk_socket);
1411 switch (sk->sk_state) {
1412 case TCP_ESTABLISHED:
1413 if (!xprt_test_and_set_connected(xprt)) {
1414 xprt->connect_cookie++;
1415 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1416 xprt_clear_connecting(xprt);
1417
1418 xprt->stat.connect_count++;
1419 xprt->stat.connect_time += (long)jiffies -
1420 xprt->stat.connect_start;
1421 xs_run_error_worker(transport, XPRT_SOCK_WAKE_PENDING);
1422 }
1423 break;
1424 case TCP_FIN_WAIT1:
1425 /* The client initiated a shutdown of the socket */
1426 xprt->connect_cookie++;
1427 xprt->reestablish_timeout = 0;
1428 set_bit(XPRT_CLOSING, &xprt->state);
1429 smp_mb__before_atomic();
1430 clear_bit(XPRT_CONNECTED, &xprt->state);
1431 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1432 smp_mb__after_atomic();
1433 break;
1434 case TCP_CLOSE_WAIT:
1435 /* The server initiated a shutdown of the socket */
1436 xprt->connect_cookie++;
1437 clear_bit(XPRT_CONNECTED, &xprt->state);
1438 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1439 fallthrough;
1440 case TCP_CLOSING:
1441 /*
1442 * If the server closed down the connection, make sure that
1443 * we back off before reconnecting
1444 */
1445 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1446 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1447 break;
1448 case TCP_LAST_ACK:
1449 set_bit(XPRT_CLOSING, &xprt->state);
1450 smp_mb__before_atomic();
1451 clear_bit(XPRT_CONNECTED, &xprt->state);
1452 smp_mb__after_atomic();
1453 break;
1454 case TCP_CLOSE:
1455 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1456 &transport->sock_state))
1457 xprt_clear_connecting(xprt);
1458 clear_bit(XPRT_CLOSING, &xprt->state);
1459 /* Trigger the socket release */
1460 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1461 }
1462 out:
1463 read_unlock_bh(&sk->sk_callback_lock);
1464}
1465
1466static void xs_write_space(struct sock *sk)
1467{
1468 struct socket_wq *wq;
1469 struct sock_xprt *transport;
1470 struct rpc_xprt *xprt;
1471
1472 if (!sk->sk_socket)
1473 return;
1474 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1475
1476 if (unlikely(!(xprt = xprt_from_sock(sk))))
1477 return;
1478 transport = container_of(xprt, struct sock_xprt, xprt);
1479 rcu_read_lock();
1480 wq = rcu_dereference(sk->sk_wq);
1481 if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1482 goto out;
1483
1484 xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE);
1485 sk->sk_write_pending--;
1486out:
1487 rcu_read_unlock();
1488}
1489
1490/**
1491 * xs_udp_write_space - callback invoked when socket buffer space
1492 * becomes available
1493 * @sk: socket whose state has changed
1494 *
1495 * Called when more output buffer space is available for this socket.
1496 * We try not to wake our writers until they can make "significant"
1497 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1498 * with a bunch of small requests.
1499 */
1500static void xs_udp_write_space(struct sock *sk)
1501{
1502 read_lock_bh(&sk->sk_callback_lock);
1503
1504 /* from net/core/sock.c:sock_def_write_space */
1505 if (sock_writeable(sk))
1506 xs_write_space(sk);
1507
1508 read_unlock_bh(&sk->sk_callback_lock);
1509}
1510
1511/**
1512 * xs_tcp_write_space - callback invoked when socket buffer space
1513 * becomes available
1514 * @sk: socket whose state has changed
1515 *
1516 * Called when more output buffer space is available for this socket.
1517 * We try not to wake our writers until they can make "significant"
1518 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1519 * with a bunch of small requests.
1520 */
1521static void xs_tcp_write_space(struct sock *sk)
1522{
1523 read_lock_bh(&sk->sk_callback_lock);
1524
1525 /* from net/core/stream.c:sk_stream_write_space */
1526 if (sk_stream_is_writeable(sk))
1527 xs_write_space(sk);
1528
1529 read_unlock_bh(&sk->sk_callback_lock);
1530}
1531
1532static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1533{
1534 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1535 struct sock *sk = transport->inet;
1536
1537 if (transport->rcvsize) {
1538 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1539 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1540 }
1541 if (transport->sndsize) {
1542 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1543 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1544 sk->sk_write_space(sk);
1545 }
1546}
1547
1548/**
1549 * xs_udp_set_buffer_size - set send and receive limits
1550 * @xprt: generic transport
1551 * @sndsize: requested size of send buffer, in bytes
1552 * @rcvsize: requested size of receive buffer, in bytes
1553 *
1554 * Set socket send and receive buffer size limits.
1555 */
1556static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1557{
1558 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1559
1560 transport->sndsize = 0;
1561 if (sndsize)
1562 transport->sndsize = sndsize + 1024;
1563 transport->rcvsize = 0;
1564 if (rcvsize)
1565 transport->rcvsize = rcvsize + 1024;
1566
1567 xs_udp_do_set_buffer_size(xprt);
1568}
1569
1570/**
1571 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1572 * @xprt: controlling transport
1573 * @task: task that timed out
1574 *
1575 * Adjust the congestion window after a retransmit timeout has occurred.
1576 */
1577static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1578{
1579 spin_lock(&xprt->transport_lock);
1580 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1581 spin_unlock(&xprt->transport_lock);
1582}
1583
1584static int xs_get_random_port(void)
1585{
1586 unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1587 unsigned short range;
1588 unsigned short rand;
1589
1590 if (max < min)
1591 return -EADDRINUSE;
1592 range = max - min + 1;
1593 rand = (unsigned short) prandom_u32() % range;
1594 return rand + min;
1595}
1596
1597static unsigned short xs_sock_getport(struct socket *sock)
1598{
1599 struct sockaddr_storage buf;
1600 unsigned short port = 0;
1601
1602 if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1603 goto out;
1604 switch (buf.ss_family) {
1605 case AF_INET6:
1606 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1607 break;
1608 case AF_INET:
1609 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1610 }
1611out:
1612 return port;
1613}
1614
1615/**
1616 * xs_set_port - reset the port number in the remote endpoint address
1617 * @xprt: generic transport
1618 * @port: new port number
1619 *
1620 */
1621static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1622{
1623 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1624
1625 rpc_set_port(xs_addr(xprt), port);
1626 xs_update_peer_port(xprt);
1627}
1628
1629static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1630{
1631 if (transport->srcport == 0 && transport->xprt.reuseport)
1632 transport->srcport = xs_sock_getport(sock);
1633}
1634
1635static int xs_get_srcport(struct sock_xprt *transport)
1636{
1637 int port = transport->srcport;
1638
1639 if (port == 0 && transport->xprt.resvport)
1640 port = xs_get_random_port();
1641 return port;
1642}
1643
1644static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1645{
1646 if (transport->srcport != 0)
1647 transport->srcport = 0;
1648 if (!transport->xprt.resvport)
1649 return 0;
1650 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1651 return xprt_max_resvport;
1652 return --port;
1653}
1654static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1655{
1656 struct sockaddr_storage myaddr;
1657 int err, nloop = 0;
1658 int port = xs_get_srcport(transport);
1659 unsigned short last;
1660
1661 /*
1662 * If we are asking for any ephemeral port (i.e. port == 0 &&
1663 * transport->xprt.resvport == 0), don't bind. Let the local
1664 * port selection happen implicitly when the socket is used
1665 * (for example at connect time).
1666 *
1667 * This ensures that we can continue to establish TCP
1668 * connections even when all local ephemeral ports are already
1669 * a part of some TCP connection. This makes no difference
1670 * for UDP sockets, but also doens't harm them.
1671 *
1672 * If we're asking for any reserved port (i.e. port == 0 &&
1673 * transport->xprt.resvport == 1) xs_get_srcport above will
1674 * ensure that port is non-zero and we will bind as needed.
1675 */
1676 if (port <= 0)
1677 return port;
1678
1679 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1680 do {
1681 rpc_set_port((struct sockaddr *)&myaddr, port);
1682 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1683 transport->xprt.addrlen);
1684 if (err == 0) {
1685 transport->srcport = port;
1686 break;
1687 }
1688 last = port;
1689 port = xs_next_srcport(transport, port);
1690 if (port > last)
1691 nloop++;
1692 } while (err == -EADDRINUSE && nloop != 2);
1693
1694 if (myaddr.ss_family == AF_INET)
1695 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1696 &((struct sockaddr_in *)&myaddr)->sin_addr,
1697 port, err ? "failed" : "ok", err);
1698 else
1699 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1700 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1701 port, err ? "failed" : "ok", err);
1702 return err;
1703}
1704
1705/*
1706 * We don't support autobind on AF_LOCAL sockets
1707 */
1708static void xs_local_rpcbind(struct rpc_task *task)
1709{
1710 xprt_set_bound(task->tk_xprt);
1711}
1712
1713static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1714{
1715}
1716
1717#ifdef CONFIG_DEBUG_LOCK_ALLOC
1718static struct lock_class_key xs_key[2];
1719static struct lock_class_key xs_slock_key[2];
1720
1721static inline void xs_reclassify_socketu(struct socket *sock)
1722{
1723 struct sock *sk = sock->sk;
1724
1725 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1726 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1727}
1728
1729static inline void xs_reclassify_socket4(struct socket *sock)
1730{
1731 struct sock *sk = sock->sk;
1732
1733 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1734 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1735}
1736
1737static inline void xs_reclassify_socket6(struct socket *sock)
1738{
1739 struct sock *sk = sock->sk;
1740
1741 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1742 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1743}
1744
1745static inline void xs_reclassify_socket(int family, struct socket *sock)
1746{
1747 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1748 return;
1749
1750 switch (family) {
1751 case AF_LOCAL:
1752 xs_reclassify_socketu(sock);
1753 break;
1754 case AF_INET:
1755 xs_reclassify_socket4(sock);
1756 break;
1757 case AF_INET6:
1758 xs_reclassify_socket6(sock);
1759 break;
1760 }
1761}
1762#else
1763static inline void xs_reclassify_socket(int family, struct socket *sock)
1764{
1765}
1766#endif
1767
1768static void xs_dummy_setup_socket(struct work_struct *work)
1769{
1770}
1771
1772static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1773 struct sock_xprt *transport, int family, int type,
1774 int protocol, bool reuseport)
1775{
1776 struct file *filp;
1777 struct socket *sock;
1778 int err;
1779
1780 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1781 if (err < 0) {
1782 dprintk("RPC: can't create %d transport socket (%d).\n",
1783 protocol, -err);
1784 goto out;
1785 }
1786 xs_reclassify_socket(family, sock);
1787
1788 if (reuseport)
1789 sock_set_reuseport(sock->sk);
1790
1791 err = xs_bind(transport, sock);
1792 if (err) {
1793 sock_release(sock);
1794 goto out;
1795 }
1796
1797 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1798 if (IS_ERR(filp))
1799 return ERR_CAST(filp);
1800 transport->file = filp;
1801
1802 return sock;
1803out:
1804 return ERR_PTR(err);
1805}
1806
1807static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1808 struct socket *sock)
1809{
1810 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1811 xprt);
1812
1813 if (!transport->inet) {
1814 struct sock *sk = sock->sk;
1815
1816 write_lock_bh(&sk->sk_callback_lock);
1817
1818 xs_save_old_callbacks(transport, sk);
1819
1820 sk->sk_user_data = xprt;
1821 sk->sk_data_ready = xs_data_ready;
1822 sk->sk_write_space = xs_udp_write_space;
1823 sock_set_flag(sk, SOCK_FASYNC);
1824 sk->sk_error_report = xs_error_report;
1825
1826 xprt_clear_connected(xprt);
1827
1828 /* Reset to new socket */
1829 transport->sock = sock;
1830 transport->inet = sk;
1831
1832 write_unlock_bh(&sk->sk_callback_lock);
1833 }
1834
1835 xs_stream_start_connect(transport);
1836
1837 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1838}
1839
1840/**
1841 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1842 * @transport: socket transport to connect
1843 */
1844static int xs_local_setup_socket(struct sock_xprt *transport)
1845{
1846 struct rpc_xprt *xprt = &transport->xprt;
1847 struct file *filp;
1848 struct socket *sock;
1849 int status;
1850
1851 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1852 SOCK_STREAM, 0, &sock, 1);
1853 if (status < 0) {
1854 dprintk("RPC: can't create AF_LOCAL "
1855 "transport socket (%d).\n", -status);
1856 goto out;
1857 }
1858 xs_reclassify_socket(AF_LOCAL, sock);
1859
1860 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1861 if (IS_ERR(filp)) {
1862 status = PTR_ERR(filp);
1863 goto out;
1864 }
1865 transport->file = filp;
1866
1867 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1868 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1869
1870 status = xs_local_finish_connecting(xprt, sock);
1871 trace_rpc_socket_connect(xprt, sock, status);
1872 switch (status) {
1873 case 0:
1874 dprintk("RPC: xprt %p connected to %s\n",
1875 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1876 xprt->stat.connect_count++;
1877 xprt->stat.connect_time += (long)jiffies -
1878 xprt->stat.connect_start;
1879 xprt_set_connected(xprt);
1880 case -ENOBUFS:
1881 break;
1882 case -ENOENT:
1883 dprintk("RPC: xprt %p: socket %s does not exist\n",
1884 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1885 break;
1886 case -ECONNREFUSED:
1887 dprintk("RPC: xprt %p: connection refused for %s\n",
1888 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1889 break;
1890 default:
1891 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1892 __func__, -status,
1893 xprt->address_strings[RPC_DISPLAY_ADDR]);
1894 }
1895
1896out:
1897 xprt_clear_connecting(xprt);
1898 xprt_wake_pending_tasks(xprt, status);
1899 return status;
1900}
1901
1902static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1903{
1904 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1905 int ret;
1906
1907 if (RPC_IS_ASYNC(task)) {
1908 /*
1909 * We want the AF_LOCAL connect to be resolved in the
1910 * filesystem namespace of the process making the rpc
1911 * call. Thus we connect synchronously.
1912 *
1913 * If we want to support asynchronous AF_LOCAL calls,
1914 * we'll need to figure out how to pass a namespace to
1915 * connect.
1916 */
1917 task->tk_rpc_status = -ENOTCONN;
1918 rpc_exit(task, -ENOTCONN);
1919 return;
1920 }
1921 ret = xs_local_setup_socket(transport);
1922 if (ret && !RPC_IS_SOFTCONN(task))
1923 msleep_interruptible(15000);
1924}
1925
1926#if IS_ENABLED(CONFIG_SUNRPC_SWAP)
1927/*
1928 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
1929 * know that we have exclusive access to the socket), to guard against
1930 * races with xs_reset_transport.
1931 */
1932static void xs_set_memalloc(struct rpc_xprt *xprt)
1933{
1934 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1935 xprt);
1936
1937 /*
1938 * If there's no sock, then we have nothing to set. The
1939 * reconnecting process will get it for us.
1940 */
1941 if (!transport->inet)
1942 return;
1943 if (atomic_read(&xprt->swapper))
1944 sk_set_memalloc(transport->inet);
1945}
1946
1947/**
1948 * xs_enable_swap - Tag this transport as being used for swap.
1949 * @xprt: transport to tag
1950 *
1951 * Take a reference to this transport on behalf of the rpc_clnt, and
1952 * optionally mark it for swapping if it wasn't already.
1953 */
1954static int
1955xs_enable_swap(struct rpc_xprt *xprt)
1956{
1957 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
1958
1959 if (atomic_inc_return(&xprt->swapper) != 1)
1960 return 0;
1961 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
1962 return -ERESTARTSYS;
1963 if (xs->inet)
1964 sk_set_memalloc(xs->inet);
1965 xprt_release_xprt(xprt, NULL);
1966 return 0;
1967}
1968
1969/**
1970 * xs_disable_swap - Untag this transport as being used for swap.
1971 * @xprt: transport to tag
1972 *
1973 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
1974 * swapper refcount goes to 0, untag the socket as a memalloc socket.
1975 */
1976static void
1977xs_disable_swap(struct rpc_xprt *xprt)
1978{
1979 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
1980
1981 if (!atomic_dec_and_test(&xprt->swapper))
1982 return;
1983 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
1984 return;
1985 if (xs->inet)
1986 sk_clear_memalloc(xs->inet);
1987 xprt_release_xprt(xprt, NULL);
1988}
1989#else
1990static void xs_set_memalloc(struct rpc_xprt *xprt)
1991{
1992}
1993
1994static int
1995xs_enable_swap(struct rpc_xprt *xprt)
1996{
1997 return -EINVAL;
1998}
1999
2000static void
2001xs_disable_swap(struct rpc_xprt *xprt)
2002{
2003}
2004#endif
2005
2006static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2007{
2008 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2009
2010 if (!transport->inet) {
2011 struct sock *sk = sock->sk;
2012
2013 write_lock_bh(&sk->sk_callback_lock);
2014
2015 xs_save_old_callbacks(transport, sk);
2016
2017 sk->sk_user_data = xprt;
2018 sk->sk_data_ready = xs_data_ready;
2019 sk->sk_write_space = xs_udp_write_space;
2020 sock_set_flag(sk, SOCK_FASYNC);
2021
2022 xprt_set_connected(xprt);
2023
2024 /* Reset to new socket */
2025 transport->sock = sock;
2026 transport->inet = sk;
2027
2028 xs_set_memalloc(xprt);
2029
2030 write_unlock_bh(&sk->sk_callback_lock);
2031 }
2032 xs_udp_do_set_buffer_size(xprt);
2033
2034 xprt->stat.connect_start = jiffies;
2035}
2036
2037static void xs_udp_setup_socket(struct work_struct *work)
2038{
2039 struct sock_xprt *transport =
2040 container_of(work, struct sock_xprt, connect_worker.work);
2041 struct rpc_xprt *xprt = &transport->xprt;
2042 struct socket *sock;
2043 int status = -EIO;
2044
2045 sock = xs_create_sock(xprt, transport,
2046 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2047 IPPROTO_UDP, false);
2048 if (IS_ERR(sock))
2049 goto out;
2050
2051 dprintk("RPC: worker connecting xprt %p via %s to "
2052 "%s (port %s)\n", xprt,
2053 xprt->address_strings[RPC_DISPLAY_PROTO],
2054 xprt->address_strings[RPC_DISPLAY_ADDR],
2055 xprt->address_strings[RPC_DISPLAY_PORT]);
2056
2057 xs_udp_finish_connecting(xprt, sock);
2058 trace_rpc_socket_connect(xprt, sock, 0);
2059 status = 0;
2060out:
2061 xprt_clear_connecting(xprt);
2062 xprt_unlock_connect(xprt, transport);
2063 xprt_wake_pending_tasks(xprt, status);
2064}
2065
2066/**
2067 * xs_tcp_shutdown - gracefully shut down a TCP socket
2068 * @xprt: transport
2069 *
2070 * Initiates a graceful shutdown of the TCP socket by calling the
2071 * equivalent of shutdown(SHUT_RDWR);
2072 */
2073static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2074{
2075 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2076 struct socket *sock = transport->sock;
2077 int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2078
2079 if (sock == NULL)
2080 return;
2081 switch (skst) {
2082 default:
2083 kernel_sock_shutdown(sock, SHUT_RDWR);
2084 trace_rpc_socket_shutdown(xprt, sock);
2085 break;
2086 case TCP_CLOSE:
2087 case TCP_TIME_WAIT:
2088 xs_reset_transport(transport);
2089 }
2090}
2091
2092static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2093 struct socket *sock)
2094{
2095 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2096 unsigned int keepidle;
2097 unsigned int keepcnt;
2098 unsigned int timeo;
2099
2100 spin_lock(&xprt->transport_lock);
2101 keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2102 keepcnt = xprt->timeout->to_retries + 1;
2103 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2104 (xprt->timeout->to_retries + 1);
2105 clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2106 spin_unlock(&xprt->transport_lock);
2107
2108 /* TCP Keepalive options */
2109 sock_set_keepalive(sock->sk);
2110 tcp_sock_set_keepidle(sock->sk, keepidle);
2111 tcp_sock_set_keepintvl(sock->sk, keepidle);
2112 tcp_sock_set_keepcnt(sock->sk, keepcnt);
2113
2114 /* TCP user timeout (see RFC5482) */
2115 tcp_sock_set_user_timeout(sock->sk, timeo);
2116}
2117
2118static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2119 unsigned long connect_timeout,
2120 unsigned long reconnect_timeout)
2121{
2122 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2123 struct rpc_timeout to;
2124 unsigned long initval;
2125
2126 spin_lock(&xprt->transport_lock);
2127 if (reconnect_timeout < xprt->max_reconnect_timeout)
2128 xprt->max_reconnect_timeout = reconnect_timeout;
2129 if (connect_timeout < xprt->connect_timeout) {
2130 memcpy(&to, xprt->timeout, sizeof(to));
2131 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2132 /* Arbitrary lower limit */
2133 if (initval < XS_TCP_INIT_REEST_TO << 1)
2134 initval = XS_TCP_INIT_REEST_TO << 1;
2135 to.to_initval = initval;
2136 to.to_maxval = initval;
2137 memcpy(&transport->tcp_timeout, &to,
2138 sizeof(transport->tcp_timeout));
2139 xprt->timeout = &transport->tcp_timeout;
2140 xprt->connect_timeout = connect_timeout;
2141 }
2142 set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2143 spin_unlock(&xprt->transport_lock);
2144}
2145
2146static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2147{
2148 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2149 int ret = -ENOTCONN;
2150
2151 if (!transport->inet) {
2152 struct sock *sk = sock->sk;
2153
2154 /* Avoid temporary address, they are bad for long-lived
2155 * connections such as NFS mounts.
2156 * RFC4941, section 3.6 suggests that:
2157 * Individual applications, which have specific
2158 * knowledge about the normal duration of connections,
2159 * MAY override this as appropriate.
2160 */
2161 if (xs_addr(xprt)->sa_family == PF_INET6) {
2162 ip6_sock_set_addr_preferences(sk,
2163 IPV6_PREFER_SRC_PUBLIC);
2164 }
2165
2166 xs_tcp_set_socket_timeouts(xprt, sock);
2167
2168 write_lock_bh(&sk->sk_callback_lock);
2169
2170 xs_save_old_callbacks(transport, sk);
2171
2172 sk->sk_user_data = xprt;
2173 sk->sk_data_ready = xs_data_ready;
2174 sk->sk_state_change = xs_tcp_state_change;
2175 sk->sk_write_space = xs_tcp_write_space;
2176 sock_set_flag(sk, SOCK_FASYNC);
2177 sk->sk_error_report = xs_error_report;
2178
2179 /* socket options */
2180 sock_reset_flag(sk, SOCK_LINGER);
2181 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2182
2183 xprt_clear_connected(xprt);
2184
2185 /* Reset to new socket */
2186 transport->sock = sock;
2187 transport->inet = sk;
2188
2189 write_unlock_bh(&sk->sk_callback_lock);
2190 }
2191
2192 if (!xprt_bound(xprt))
2193 goto out;
2194
2195 xs_set_memalloc(xprt);
2196
2197 xs_stream_start_connect(transport);
2198
2199 /* Tell the socket layer to start connecting... */
2200 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2201 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2202 switch (ret) {
2203 case 0:
2204 xs_set_srcport(transport, sock);
2205 fallthrough;
2206 case -EINPROGRESS:
2207 /* SYN_SENT! */
2208 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2209 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2210 break;
2211 case -EADDRNOTAVAIL:
2212 /* Source port number is unavailable. Try a new one! */
2213 transport->srcport = 0;
2214 }
2215out:
2216 return ret;
2217}
2218
2219/**
2220 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2221 * @work: queued work item
2222 *
2223 * Invoked by a work queue tasklet.
2224 */
2225static void xs_tcp_setup_socket(struct work_struct *work)
2226{
2227 struct sock_xprt *transport =
2228 container_of(work, struct sock_xprt, connect_worker.work);
2229 struct socket *sock = transport->sock;
2230 struct rpc_xprt *xprt = &transport->xprt;
2231 int status = -EIO;
2232
2233 if (!sock) {
2234 sock = xs_create_sock(xprt, transport,
2235 xs_addr(xprt)->sa_family, SOCK_STREAM,
2236 IPPROTO_TCP, true);
2237 if (IS_ERR(sock)) {
2238 status = PTR_ERR(sock);
2239 goto out;
2240 }
2241 }
2242
2243 dprintk("RPC: worker connecting xprt %p via %s to "
2244 "%s (port %s)\n", xprt,
2245 xprt->address_strings[RPC_DISPLAY_PROTO],
2246 xprt->address_strings[RPC_DISPLAY_ADDR],
2247 xprt->address_strings[RPC_DISPLAY_PORT]);
2248
2249 status = xs_tcp_finish_connecting(xprt, sock);
2250 trace_rpc_socket_connect(xprt, sock, status);
2251 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2252 xprt, -status, xprt_connected(xprt),
2253 sock->sk->sk_state);
2254 switch (status) {
2255 default:
2256 printk("%s: connect returned unhandled error %d\n",
2257 __func__, status);
2258 fallthrough;
2259 case -EADDRNOTAVAIL:
2260 /* We're probably in TIME_WAIT. Get rid of existing socket,
2261 * and retry
2262 */
2263 xs_tcp_force_close(xprt);
2264 break;
2265 case 0:
2266 case -EINPROGRESS:
2267 case -EALREADY:
2268 xprt_unlock_connect(xprt, transport);
2269 return;
2270 case -EINVAL:
2271 /* Happens, for instance, if the user specified a link
2272 * local IPv6 address without a scope-id.
2273 */
2274 case -ECONNREFUSED:
2275 case -ECONNRESET:
2276 case -ENETDOWN:
2277 case -ENETUNREACH:
2278 case -EHOSTUNREACH:
2279 case -EADDRINUSE:
2280 case -ENOBUFS:
2281 /*
2282 * xs_tcp_force_close() wakes tasks with -EIO.
2283 * We need to wake them first to ensure the
2284 * correct error code.
2285 */
2286 xprt_wake_pending_tasks(xprt, status);
2287 xs_tcp_force_close(xprt);
2288 goto out;
2289 }
2290 status = -EAGAIN;
2291out:
2292 xprt_clear_connecting(xprt);
2293 xprt_unlock_connect(xprt, transport);
2294 xprt_wake_pending_tasks(xprt, status);
2295}
2296
2297/**
2298 * xs_connect - connect a socket to a remote endpoint
2299 * @xprt: pointer to transport structure
2300 * @task: address of RPC task that manages state of connect request
2301 *
2302 * TCP: If the remote end dropped the connection, delay reconnecting.
2303 *
2304 * UDP socket connects are synchronous, but we use a work queue anyway
2305 * to guarantee that even unprivileged user processes can set up a
2306 * socket on a privileged port.
2307 *
2308 * If a UDP socket connect fails, the delay behavior here prevents
2309 * retry floods (hard mounts).
2310 */
2311static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2312{
2313 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2314 unsigned long delay = 0;
2315
2316 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2317
2318 if (transport->sock != NULL) {
2319 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2320 "seconds\n",
2321 xprt, xprt->reestablish_timeout / HZ);
2322
2323 /* Start by resetting any existing state */
2324 xs_reset_transport(transport);
2325
2326 delay = xprt_reconnect_delay(xprt);
2327 xprt_reconnect_backoff(xprt, XS_TCP_INIT_REEST_TO);
2328
2329 } else
2330 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2331
2332 queue_delayed_work(xprtiod_workqueue,
2333 &transport->connect_worker,
2334 delay);
2335}
2336
2337static void xs_wake_disconnect(struct sock_xprt *transport)
2338{
2339 if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state))
2340 xs_tcp_force_close(&transport->xprt);
2341}
2342
2343static void xs_wake_write(struct sock_xprt *transport)
2344{
2345 if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state))
2346 xprt_write_space(&transport->xprt);
2347}
2348
2349static void xs_wake_error(struct sock_xprt *transport)
2350{
2351 int sockerr;
2352
2353 if (!test_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2354 return;
2355 mutex_lock(&transport->recv_mutex);
2356 if (transport->sock == NULL)
2357 goto out;
2358 if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2359 goto out;
2360 sockerr = xchg(&transport->xprt_err, 0);
2361 if (sockerr < 0)
2362 xprt_wake_pending_tasks(&transport->xprt, sockerr);
2363out:
2364 mutex_unlock(&transport->recv_mutex);
2365}
2366
2367static void xs_wake_pending(struct sock_xprt *transport)
2368{
2369 if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING, &transport->sock_state))
2370 xprt_wake_pending_tasks(&transport->xprt, -EAGAIN);
2371}
2372
2373static void xs_error_handle(struct work_struct *work)
2374{
2375 struct sock_xprt *transport = container_of(work,
2376 struct sock_xprt, error_worker);
2377
2378 xs_wake_disconnect(transport);
2379 xs_wake_write(transport);
2380 xs_wake_error(transport);
2381 xs_wake_pending(transport);
2382}
2383
2384/**
2385 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2386 * @xprt: rpc_xprt struct containing statistics
2387 * @seq: output file
2388 *
2389 */
2390static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2391{
2392 long idle_time = 0;
2393
2394 if (xprt_connected(xprt))
2395 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2396
2397 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2398 "%llu %llu %lu %llu %llu\n",
2399 xprt->stat.bind_count,
2400 xprt->stat.connect_count,
2401 xprt->stat.connect_time / HZ,
2402 idle_time,
2403 xprt->stat.sends,
2404 xprt->stat.recvs,
2405 xprt->stat.bad_xids,
2406 xprt->stat.req_u,
2407 xprt->stat.bklog_u,
2408 xprt->stat.max_slots,
2409 xprt->stat.sending_u,
2410 xprt->stat.pending_u);
2411}
2412
2413/**
2414 * xs_udp_print_stats - display UDP socket-specifc stats
2415 * @xprt: rpc_xprt struct containing statistics
2416 * @seq: output file
2417 *
2418 */
2419static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2420{
2421 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2422
2423 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2424 "%lu %llu %llu\n",
2425 transport->srcport,
2426 xprt->stat.bind_count,
2427 xprt->stat.sends,
2428 xprt->stat.recvs,
2429 xprt->stat.bad_xids,
2430 xprt->stat.req_u,
2431 xprt->stat.bklog_u,
2432 xprt->stat.max_slots,
2433 xprt->stat.sending_u,
2434 xprt->stat.pending_u);
2435}
2436
2437/**
2438 * xs_tcp_print_stats - display TCP socket-specifc stats
2439 * @xprt: rpc_xprt struct containing statistics
2440 * @seq: output file
2441 *
2442 */
2443static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2444{
2445 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2446 long idle_time = 0;
2447
2448 if (xprt_connected(xprt))
2449 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2450
2451 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2452 "%llu %llu %lu %llu %llu\n",
2453 transport->srcport,
2454 xprt->stat.bind_count,
2455 xprt->stat.connect_count,
2456 xprt->stat.connect_time / HZ,
2457 idle_time,
2458 xprt->stat.sends,
2459 xprt->stat.recvs,
2460 xprt->stat.bad_xids,
2461 xprt->stat.req_u,
2462 xprt->stat.bklog_u,
2463 xprt->stat.max_slots,
2464 xprt->stat.sending_u,
2465 xprt->stat.pending_u);
2466}
2467
2468/*
2469 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2470 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2471 * to use the server side send routines.
2472 */
2473static int bc_malloc(struct rpc_task *task)
2474{
2475 struct rpc_rqst *rqst = task->tk_rqstp;
2476 size_t size = rqst->rq_callsize;
2477 struct page *page;
2478 struct rpc_buffer *buf;
2479
2480 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2481 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2482 size);
2483 return -EINVAL;
2484 }
2485
2486 page = alloc_page(GFP_KERNEL);
2487 if (!page)
2488 return -ENOMEM;
2489
2490 buf = page_address(page);
2491 buf->len = PAGE_SIZE;
2492
2493 rqst->rq_buffer = buf->data;
2494 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2495 return 0;
2496}
2497
2498/*
2499 * Free the space allocated in the bc_alloc routine
2500 */
2501static void bc_free(struct rpc_task *task)
2502{
2503 void *buffer = task->tk_rqstp->rq_buffer;
2504 struct rpc_buffer *buf;
2505
2506 buf = container_of(buffer, struct rpc_buffer, data);
2507 free_page((unsigned long)buf);
2508}
2509
2510static int bc_sendto(struct rpc_rqst *req)
2511{
2512 struct xdr_buf *xdr = &req->rq_snd_buf;
2513 struct sock_xprt *transport =
2514 container_of(req->rq_xprt, struct sock_xprt, xprt);
2515 struct msghdr msg = {
2516 .msg_flags = 0,
2517 };
2518 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
2519 (u32)xdr->len);
2520 unsigned int sent = 0;
2521 int err;
2522
2523 req->rq_xtime = ktime_get();
2524 err = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, marker, &sent);
2525 xdr_free_bvec(xdr);
2526 if (err < 0 || sent != (xdr->len + sizeof(marker)))
2527 return -EAGAIN;
2528 return sent;
2529}
2530
2531/**
2532 * bc_send_request - Send a backchannel Call on a TCP socket
2533 * @req: rpc_rqst containing Call message to be sent
2534 *
2535 * xpt_mutex ensures @rqstp's whole message is written to the socket
2536 * without interruption.
2537 *
2538 * Return values:
2539 * %0 if the message was sent successfully
2540 * %ENOTCONN if the message was not sent
2541 */
2542static int bc_send_request(struct rpc_rqst *req)
2543{
2544 struct svc_xprt *xprt;
2545 int len;
2546
2547 /*
2548 * Get the server socket associated with this callback xprt
2549 */
2550 xprt = req->rq_xprt->bc_xprt;
2551
2552 /*
2553 * Grab the mutex to serialize data as the connection is shared
2554 * with the fore channel
2555 */
2556 mutex_lock(&xprt->xpt_mutex);
2557 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2558 len = -ENOTCONN;
2559 else
2560 len = bc_sendto(req);
2561 mutex_unlock(&xprt->xpt_mutex);
2562
2563 if (len > 0)
2564 len = 0;
2565
2566 return len;
2567}
2568
2569/*
2570 * The close routine. Since this is client initiated, we do nothing
2571 */
2572
2573static void bc_close(struct rpc_xprt *xprt)
2574{
2575 xprt_disconnect_done(xprt);
2576}
2577
2578/*
2579 * The xprt destroy routine. Again, because this connection is client
2580 * initiated, we do nothing
2581 */
2582
2583static void bc_destroy(struct rpc_xprt *xprt)
2584{
2585 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2586
2587 xs_xprt_free(xprt);
2588 module_put(THIS_MODULE);
2589}
2590
2591static const struct rpc_xprt_ops xs_local_ops = {
2592 .reserve_xprt = xprt_reserve_xprt,
2593 .release_xprt = xprt_release_xprt,
2594 .alloc_slot = xprt_alloc_slot,
2595 .free_slot = xprt_free_slot,
2596 .rpcbind = xs_local_rpcbind,
2597 .set_port = xs_local_set_port,
2598 .connect = xs_local_connect,
2599 .buf_alloc = rpc_malloc,
2600 .buf_free = rpc_free,
2601 .prepare_request = xs_stream_prepare_request,
2602 .send_request = xs_local_send_request,
2603 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2604 .close = xs_close,
2605 .destroy = xs_destroy,
2606 .print_stats = xs_local_print_stats,
2607 .enable_swap = xs_enable_swap,
2608 .disable_swap = xs_disable_swap,
2609};
2610
2611static const struct rpc_xprt_ops xs_udp_ops = {
2612 .set_buffer_size = xs_udp_set_buffer_size,
2613 .reserve_xprt = xprt_reserve_xprt_cong,
2614 .release_xprt = xprt_release_xprt_cong,
2615 .alloc_slot = xprt_alloc_slot,
2616 .free_slot = xprt_free_slot,
2617 .rpcbind = rpcb_getport_async,
2618 .set_port = xs_set_port,
2619 .connect = xs_connect,
2620 .buf_alloc = rpc_malloc,
2621 .buf_free = rpc_free,
2622 .send_request = xs_udp_send_request,
2623 .wait_for_reply_request = xprt_wait_for_reply_request_rtt,
2624 .timer = xs_udp_timer,
2625 .release_request = xprt_release_rqst_cong,
2626 .close = xs_close,
2627 .destroy = xs_destroy,
2628 .print_stats = xs_udp_print_stats,
2629 .enable_swap = xs_enable_swap,
2630 .disable_swap = xs_disable_swap,
2631 .inject_disconnect = xs_inject_disconnect,
2632};
2633
2634static const struct rpc_xprt_ops xs_tcp_ops = {
2635 .reserve_xprt = xprt_reserve_xprt,
2636 .release_xprt = xprt_release_xprt,
2637 .alloc_slot = xprt_alloc_slot,
2638 .free_slot = xprt_free_slot,
2639 .rpcbind = rpcb_getport_async,
2640 .set_port = xs_set_port,
2641 .connect = xs_connect,
2642 .buf_alloc = rpc_malloc,
2643 .buf_free = rpc_free,
2644 .prepare_request = xs_stream_prepare_request,
2645 .send_request = xs_tcp_send_request,
2646 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2647 .close = xs_tcp_shutdown,
2648 .destroy = xs_destroy,
2649 .set_connect_timeout = xs_tcp_set_connect_timeout,
2650 .print_stats = xs_tcp_print_stats,
2651 .enable_swap = xs_enable_swap,
2652 .disable_swap = xs_disable_swap,
2653 .inject_disconnect = xs_inject_disconnect,
2654#ifdef CONFIG_SUNRPC_BACKCHANNEL
2655 .bc_setup = xprt_setup_bc,
2656 .bc_maxpayload = xs_tcp_bc_maxpayload,
2657 .bc_num_slots = xprt_bc_max_slots,
2658 .bc_free_rqst = xprt_free_bc_rqst,
2659 .bc_destroy = xprt_destroy_bc,
2660#endif
2661};
2662
2663/*
2664 * The rpc_xprt_ops for the server backchannel
2665 */
2666
2667static const struct rpc_xprt_ops bc_tcp_ops = {
2668 .reserve_xprt = xprt_reserve_xprt,
2669 .release_xprt = xprt_release_xprt,
2670 .alloc_slot = xprt_alloc_slot,
2671 .free_slot = xprt_free_slot,
2672 .buf_alloc = bc_malloc,
2673 .buf_free = bc_free,
2674 .send_request = bc_send_request,
2675 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2676 .close = bc_close,
2677 .destroy = bc_destroy,
2678 .print_stats = xs_tcp_print_stats,
2679 .enable_swap = xs_enable_swap,
2680 .disable_swap = xs_disable_swap,
2681 .inject_disconnect = xs_inject_disconnect,
2682};
2683
2684static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2685{
2686 static const struct sockaddr_in sin = {
2687 .sin_family = AF_INET,
2688 .sin_addr.s_addr = htonl(INADDR_ANY),
2689 };
2690 static const struct sockaddr_in6 sin6 = {
2691 .sin6_family = AF_INET6,
2692 .sin6_addr = IN6ADDR_ANY_INIT,
2693 };
2694
2695 switch (family) {
2696 case AF_LOCAL:
2697 break;
2698 case AF_INET:
2699 memcpy(sap, &sin, sizeof(sin));
2700 break;
2701 case AF_INET6:
2702 memcpy(sap, &sin6, sizeof(sin6));
2703 break;
2704 default:
2705 dprintk("RPC: %s: Bad address family\n", __func__);
2706 return -EAFNOSUPPORT;
2707 }
2708 return 0;
2709}
2710
2711static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2712 unsigned int slot_table_size,
2713 unsigned int max_slot_table_size)
2714{
2715 struct rpc_xprt *xprt;
2716 struct sock_xprt *new;
2717
2718 if (args->addrlen > sizeof(xprt->addr)) {
2719 dprintk("RPC: xs_setup_xprt: address too large\n");
2720 return ERR_PTR(-EBADF);
2721 }
2722
2723 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2724 max_slot_table_size);
2725 if (xprt == NULL) {
2726 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2727 "rpc_xprt\n");
2728 return ERR_PTR(-ENOMEM);
2729 }
2730
2731 new = container_of(xprt, struct sock_xprt, xprt);
2732 mutex_init(&new->recv_mutex);
2733 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2734 xprt->addrlen = args->addrlen;
2735 if (args->srcaddr)
2736 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2737 else {
2738 int err;
2739 err = xs_init_anyaddr(args->dstaddr->sa_family,
2740 (struct sockaddr *)&new->srcaddr);
2741 if (err != 0) {
2742 xprt_free(xprt);
2743 return ERR_PTR(err);
2744 }
2745 }
2746
2747 return xprt;
2748}
2749
2750static const struct rpc_timeout xs_local_default_timeout = {
2751 .to_initval = 10 * HZ,
2752 .to_maxval = 10 * HZ,
2753 .to_retries = 2,
2754};
2755
2756/**
2757 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2758 * @args: rpc transport creation arguments
2759 *
2760 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2761 */
2762static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2763{
2764 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2765 struct sock_xprt *transport;
2766 struct rpc_xprt *xprt;
2767 struct rpc_xprt *ret;
2768
2769 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2770 xprt_max_tcp_slot_table_entries);
2771 if (IS_ERR(xprt))
2772 return xprt;
2773 transport = container_of(xprt, struct sock_xprt, xprt);
2774
2775 xprt->prot = 0;
2776 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2777
2778 xprt->bind_timeout = XS_BIND_TO;
2779 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2780 xprt->idle_timeout = XS_IDLE_DISC_TO;
2781
2782 xprt->ops = &xs_local_ops;
2783 xprt->timeout = &xs_local_default_timeout;
2784
2785 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2786 INIT_WORK(&transport->error_worker, xs_error_handle);
2787 INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket);
2788
2789 switch (sun->sun_family) {
2790 case AF_LOCAL:
2791 if (sun->sun_path[0] != '/') {
2792 dprintk("RPC: bad AF_LOCAL address: %s\n",
2793 sun->sun_path);
2794 ret = ERR_PTR(-EINVAL);
2795 goto out_err;
2796 }
2797 xprt_set_bound(xprt);
2798 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2799 ret = ERR_PTR(xs_local_setup_socket(transport));
2800 if (ret)
2801 goto out_err;
2802 break;
2803 default:
2804 ret = ERR_PTR(-EAFNOSUPPORT);
2805 goto out_err;
2806 }
2807
2808 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2809 xprt->address_strings[RPC_DISPLAY_ADDR]);
2810
2811 if (try_module_get(THIS_MODULE))
2812 return xprt;
2813 ret = ERR_PTR(-EINVAL);
2814out_err:
2815 xs_xprt_free(xprt);
2816 return ret;
2817}
2818
2819static const struct rpc_timeout xs_udp_default_timeout = {
2820 .to_initval = 5 * HZ,
2821 .to_maxval = 30 * HZ,
2822 .to_increment = 5 * HZ,
2823 .to_retries = 5,
2824};
2825
2826/**
2827 * xs_setup_udp - Set up transport to use a UDP socket
2828 * @args: rpc transport creation arguments
2829 *
2830 */
2831static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2832{
2833 struct sockaddr *addr = args->dstaddr;
2834 struct rpc_xprt *xprt;
2835 struct sock_xprt *transport;
2836 struct rpc_xprt *ret;
2837
2838 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2839 xprt_udp_slot_table_entries);
2840 if (IS_ERR(xprt))
2841 return xprt;
2842 transport = container_of(xprt, struct sock_xprt, xprt);
2843
2844 xprt->prot = IPPROTO_UDP;
2845 /* XXX: header size can vary due to auth type, IPv6, etc. */
2846 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2847
2848 xprt->bind_timeout = XS_BIND_TO;
2849 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2850 xprt->idle_timeout = XS_IDLE_DISC_TO;
2851
2852 xprt->ops = &xs_udp_ops;
2853
2854 xprt->timeout = &xs_udp_default_timeout;
2855
2856 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
2857 INIT_WORK(&transport->error_worker, xs_error_handle);
2858 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2859
2860 switch (addr->sa_family) {
2861 case AF_INET:
2862 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2863 xprt_set_bound(xprt);
2864
2865 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2866 break;
2867 case AF_INET6:
2868 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2869 xprt_set_bound(xprt);
2870
2871 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2872 break;
2873 default:
2874 ret = ERR_PTR(-EAFNOSUPPORT);
2875 goto out_err;
2876 }
2877
2878 if (xprt_bound(xprt))
2879 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2880 xprt->address_strings[RPC_DISPLAY_ADDR],
2881 xprt->address_strings[RPC_DISPLAY_PORT],
2882 xprt->address_strings[RPC_DISPLAY_PROTO]);
2883 else
2884 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2885 xprt->address_strings[RPC_DISPLAY_ADDR],
2886 xprt->address_strings[RPC_DISPLAY_PROTO]);
2887
2888 if (try_module_get(THIS_MODULE))
2889 return xprt;
2890 ret = ERR_PTR(-EINVAL);
2891out_err:
2892 xs_xprt_free(xprt);
2893 return ret;
2894}
2895
2896static const struct rpc_timeout xs_tcp_default_timeout = {
2897 .to_initval = 60 * HZ,
2898 .to_maxval = 60 * HZ,
2899 .to_retries = 2,
2900};
2901
2902/**
2903 * xs_setup_tcp - Set up transport to use a TCP socket
2904 * @args: rpc transport creation arguments
2905 *
2906 */
2907static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2908{
2909 struct sockaddr *addr = args->dstaddr;
2910 struct rpc_xprt *xprt;
2911 struct sock_xprt *transport;
2912 struct rpc_xprt *ret;
2913 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2914
2915 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2916 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2917
2918 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2919 max_slot_table_size);
2920 if (IS_ERR(xprt))
2921 return xprt;
2922 transport = container_of(xprt, struct sock_xprt, xprt);
2923
2924 xprt->prot = IPPROTO_TCP;
2925 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2926
2927 xprt->bind_timeout = XS_BIND_TO;
2928 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2929 xprt->idle_timeout = XS_IDLE_DISC_TO;
2930
2931 xprt->ops = &xs_tcp_ops;
2932 xprt->timeout = &xs_tcp_default_timeout;
2933
2934 xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
2935 xprt->connect_timeout = xprt->timeout->to_initval *
2936 (xprt->timeout->to_retries + 1);
2937
2938 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2939 INIT_WORK(&transport->error_worker, xs_error_handle);
2940 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
2941
2942 switch (addr->sa_family) {
2943 case AF_INET:
2944 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2945 xprt_set_bound(xprt);
2946
2947 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2948 break;
2949 case AF_INET6:
2950 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2951 xprt_set_bound(xprt);
2952
2953 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2954 break;
2955 default:
2956 ret = ERR_PTR(-EAFNOSUPPORT);
2957 goto out_err;
2958 }
2959
2960 if (xprt_bound(xprt))
2961 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2962 xprt->address_strings[RPC_DISPLAY_ADDR],
2963 xprt->address_strings[RPC_DISPLAY_PORT],
2964 xprt->address_strings[RPC_DISPLAY_PROTO]);
2965 else
2966 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2967 xprt->address_strings[RPC_DISPLAY_ADDR],
2968 xprt->address_strings[RPC_DISPLAY_PROTO]);
2969
2970 if (try_module_get(THIS_MODULE))
2971 return xprt;
2972 ret = ERR_PTR(-EINVAL);
2973out_err:
2974 xs_xprt_free(xprt);
2975 return ret;
2976}
2977
2978/**
2979 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2980 * @args: rpc transport creation arguments
2981 *
2982 */
2983static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2984{
2985 struct sockaddr *addr = args->dstaddr;
2986 struct rpc_xprt *xprt;
2987 struct sock_xprt *transport;
2988 struct svc_sock *bc_sock;
2989 struct rpc_xprt *ret;
2990
2991 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2992 xprt_tcp_slot_table_entries);
2993 if (IS_ERR(xprt))
2994 return xprt;
2995 transport = container_of(xprt, struct sock_xprt, xprt);
2996
2997 xprt->prot = IPPROTO_TCP;
2998 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2999 xprt->timeout = &xs_tcp_default_timeout;
3000
3001 /* backchannel */
3002 xprt_set_bound(xprt);
3003 xprt->bind_timeout = 0;
3004 xprt->reestablish_timeout = 0;
3005 xprt->idle_timeout = 0;
3006
3007 xprt->ops = &bc_tcp_ops;
3008
3009 switch (addr->sa_family) {
3010 case AF_INET:
3011 xs_format_peer_addresses(xprt, "tcp",
3012 RPCBIND_NETID_TCP);
3013 break;
3014 case AF_INET6:
3015 xs_format_peer_addresses(xprt, "tcp",
3016 RPCBIND_NETID_TCP6);
3017 break;
3018 default:
3019 ret = ERR_PTR(-EAFNOSUPPORT);
3020 goto out_err;
3021 }
3022
3023 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3024 xprt->address_strings[RPC_DISPLAY_ADDR],
3025 xprt->address_strings[RPC_DISPLAY_PORT],
3026 xprt->address_strings[RPC_DISPLAY_PROTO]);
3027
3028 /*
3029 * Once we've associated a backchannel xprt with a connection,
3030 * we want to keep it around as long as the connection lasts,
3031 * in case we need to start using it for a backchannel again;
3032 * this reference won't be dropped until bc_xprt is destroyed.
3033 */
3034 xprt_get(xprt);
3035 args->bc_xprt->xpt_bc_xprt = xprt;
3036 xprt->bc_xprt = args->bc_xprt;
3037 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3038 transport->sock = bc_sock->sk_sock;
3039 transport->inet = bc_sock->sk_sk;
3040
3041 /*
3042 * Since we don't want connections for the backchannel, we set
3043 * the xprt status to connected
3044 */
3045 xprt_set_connected(xprt);
3046
3047 if (try_module_get(THIS_MODULE))
3048 return xprt;
3049
3050 args->bc_xprt->xpt_bc_xprt = NULL;
3051 args->bc_xprt->xpt_bc_xps = NULL;
3052 xprt_put(xprt);
3053 ret = ERR_PTR(-EINVAL);
3054out_err:
3055 xs_xprt_free(xprt);
3056 return ret;
3057}
3058
3059static struct xprt_class xs_local_transport = {
3060 .list = LIST_HEAD_INIT(xs_local_transport.list),
3061 .name = "named UNIX socket",
3062 .owner = THIS_MODULE,
3063 .ident = XPRT_TRANSPORT_LOCAL,
3064 .setup = xs_setup_local,
3065};
3066
3067static struct xprt_class xs_udp_transport = {
3068 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3069 .name = "udp",
3070 .owner = THIS_MODULE,
3071 .ident = XPRT_TRANSPORT_UDP,
3072 .setup = xs_setup_udp,
3073};
3074
3075static struct xprt_class xs_tcp_transport = {
3076 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3077 .name = "tcp",
3078 .owner = THIS_MODULE,
3079 .ident = XPRT_TRANSPORT_TCP,
3080 .setup = xs_setup_tcp,
3081};
3082
3083static struct xprt_class xs_bc_tcp_transport = {
3084 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3085 .name = "tcp NFSv4.1 backchannel",
3086 .owner = THIS_MODULE,
3087 .ident = XPRT_TRANSPORT_BC_TCP,
3088 .setup = xs_setup_bc_tcp,
3089};
3090
3091/**
3092 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3093 *
3094 */
3095int init_socket_xprt(void)
3096{
3097 if (!sunrpc_table_header)
3098 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3099
3100 xprt_register_transport(&xs_local_transport);
3101 xprt_register_transport(&xs_udp_transport);
3102 xprt_register_transport(&xs_tcp_transport);
3103 xprt_register_transport(&xs_bc_tcp_transport);
3104
3105 return 0;
3106}
3107
3108/**
3109 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3110 *
3111 */
3112void cleanup_socket_xprt(void)
3113{
3114 if (sunrpc_table_header) {
3115 unregister_sysctl_table(sunrpc_table_header);
3116 sunrpc_table_header = NULL;
3117 }
3118
3119 xprt_unregister_transport(&xs_local_transport);
3120 xprt_unregister_transport(&xs_udp_transport);
3121 xprt_unregister_transport(&xs_tcp_transport);
3122 xprt_unregister_transport(&xs_bc_tcp_transport);
3123}
3124
3125static int param_set_uint_minmax(const char *val,
3126 const struct kernel_param *kp,
3127 unsigned int min, unsigned int max)
3128{
3129 unsigned int num;
3130 int ret;
3131
3132 if (!val)
3133 return -EINVAL;
3134 ret = kstrtouint(val, 0, &num);
3135 if (ret)
3136 return ret;
3137 if (num < min || num > max)
3138 return -EINVAL;
3139 *((unsigned int *)kp->arg) = num;
3140 return 0;
3141}
3142
3143static int param_set_portnr(const char *val, const struct kernel_param *kp)
3144{
3145 return param_set_uint_minmax(val, kp,
3146 RPC_MIN_RESVPORT,
3147 RPC_MAX_RESVPORT);
3148}
3149
3150static const struct kernel_param_ops param_ops_portnr = {
3151 .set = param_set_portnr,
3152 .get = param_get_uint,
3153};
3154
3155#define param_check_portnr(name, p) \
3156 __param_check(name, p, unsigned int);
3157
3158module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3159module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3160
3161static int param_set_slot_table_size(const char *val,
3162 const struct kernel_param *kp)
3163{
3164 return param_set_uint_minmax(val, kp,
3165 RPC_MIN_SLOT_TABLE,
3166 RPC_MAX_SLOT_TABLE);
3167}
3168
3169static const struct kernel_param_ops param_ops_slot_table_size = {
3170 .set = param_set_slot_table_size,
3171 .get = param_get_uint,
3172};
3173
3174#define param_check_slot_table_size(name, p) \
3175 __param_check(name, p, unsigned int);
3176
3177static int param_set_max_slot_table_size(const char *val,
3178 const struct kernel_param *kp)
3179{
3180 return param_set_uint_minmax(val, kp,
3181 RPC_MIN_SLOT_TABLE,
3182 RPC_MAX_SLOT_TABLE_LIMIT);
3183}
3184
3185static const struct kernel_param_ops param_ops_max_slot_table_size = {
3186 .set = param_set_max_slot_table_size,
3187 .get = param_get_uint,
3188};
3189
3190#define param_check_max_slot_table_size(name, p) \
3191 __param_check(name, p, unsigned int);
3192
3193module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3194 slot_table_size, 0644);
3195module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3196 max_slot_table_size, 0644);
3197module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3198 slot_table_size, 0644);
1/*
2 * linux/net/sunrpc/xprtsock.c
3 *
4 * Client-side transport implementation for sockets.
5 *
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>
10 *
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>
14 *
15 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
16 *
17 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
18 * <gilles.quillard@bull.net>
19 */
20
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>
29#include <linux/in.h>
30#include <linux/net.h>
31#include <linux/mm.h>
32#include <linux/un.h>
33#include <linux/udp.h>
34#include <linux/tcp.h>
35#include <linux/sunrpc/clnt.h>
36#include <linux/sunrpc/sched.h>
37#include <linux/sunrpc/svcsock.h>
38#include <linux/sunrpc/xprtsock.h>
39#include <linux/file.h>
40#ifdef CONFIG_SUNRPC_BACKCHANNEL
41#include <linux/sunrpc/bc_xprt.h>
42#endif
43
44#include <net/sock.h>
45#include <net/checksum.h>
46#include <net/udp.h>
47#include <net/tcp.h>
48
49#include "sunrpc.h"
50
51static void xs_close(struct rpc_xprt *xprt);
52
53/*
54 * xprtsock tunables
55 */
56static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
57static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
58static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
59
60static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
61static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
62
63#define XS_TCP_LINGER_TO (15U * HZ)
64static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
65
66/*
67 * We can register our own files under /proc/sys/sunrpc by
68 * calling register_sysctl_table() again. The files in that
69 * directory become the union of all files registered there.
70 *
71 * We simply need to make sure that we don't collide with
72 * someone else's file names!
73 */
74
75#ifdef RPC_DEBUG
76
77static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
78static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
79static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
80static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
81static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
82
83static struct ctl_table_header *sunrpc_table_header;
84
85/*
86 * FIXME: changing the UDP slot table size should also resize the UDP
87 * socket buffers for existing UDP transports
88 */
89static ctl_table xs_tunables_table[] = {
90 {
91 .procname = "udp_slot_table_entries",
92 .data = &xprt_udp_slot_table_entries,
93 .maxlen = sizeof(unsigned int),
94 .mode = 0644,
95 .proc_handler = proc_dointvec_minmax,
96 .extra1 = &min_slot_table_size,
97 .extra2 = &max_slot_table_size
98 },
99 {
100 .procname = "tcp_slot_table_entries",
101 .data = &xprt_tcp_slot_table_entries,
102 .maxlen = sizeof(unsigned int),
103 .mode = 0644,
104 .proc_handler = proc_dointvec_minmax,
105 .extra1 = &min_slot_table_size,
106 .extra2 = &max_slot_table_size
107 },
108 {
109 .procname = "tcp_max_slot_table_entries",
110 .data = &xprt_max_tcp_slot_table_entries,
111 .maxlen = sizeof(unsigned int),
112 .mode = 0644,
113 .proc_handler = proc_dointvec_minmax,
114 .extra1 = &min_slot_table_size,
115 .extra2 = &max_tcp_slot_table_limit
116 },
117 {
118 .procname = "min_resvport",
119 .data = &xprt_min_resvport,
120 .maxlen = sizeof(unsigned int),
121 .mode = 0644,
122 .proc_handler = proc_dointvec_minmax,
123 .extra1 = &xprt_min_resvport_limit,
124 .extra2 = &xprt_max_resvport_limit
125 },
126 {
127 .procname = "max_resvport",
128 .data = &xprt_max_resvport,
129 .maxlen = sizeof(unsigned int),
130 .mode = 0644,
131 .proc_handler = proc_dointvec_minmax,
132 .extra1 = &xprt_min_resvport_limit,
133 .extra2 = &xprt_max_resvport_limit
134 },
135 {
136 .procname = "tcp_fin_timeout",
137 .data = &xs_tcp_fin_timeout,
138 .maxlen = sizeof(xs_tcp_fin_timeout),
139 .mode = 0644,
140 .proc_handler = proc_dointvec_jiffies,
141 },
142 { },
143};
144
145static ctl_table sunrpc_table[] = {
146 {
147 .procname = "sunrpc",
148 .mode = 0555,
149 .child = xs_tunables_table
150 },
151 { },
152};
153
154#endif
155
156/*
157 * Wait duration for a reply from the RPC portmapper.
158 */
159#define XS_BIND_TO (60U * HZ)
160
161/*
162 * Delay if a UDP socket connect error occurs. This is most likely some
163 * kind of resource problem on the local host.
164 */
165#define XS_UDP_REEST_TO (2U * HZ)
166
167/*
168 * The reestablish timeout allows clients to delay for a bit before attempting
169 * to reconnect to a server that just dropped our connection.
170 *
171 * We implement an exponential backoff when trying to reestablish a TCP
172 * transport connection with the server. Some servers like to drop a TCP
173 * connection when they are overworked, so we start with a short timeout and
174 * increase over time if the server is down or not responding.
175 */
176#define XS_TCP_INIT_REEST_TO (3U * HZ)
177#define XS_TCP_MAX_REEST_TO (5U * 60 * HZ)
178
179/*
180 * TCP idle timeout; client drops the transport socket if it is idle
181 * for this long. Note that we also timeout UDP sockets to prevent
182 * holding port numbers when there is no RPC traffic.
183 */
184#define XS_IDLE_DISC_TO (5U * 60 * HZ)
185
186#ifdef RPC_DEBUG
187# undef RPC_DEBUG_DATA
188# define RPCDBG_FACILITY RPCDBG_TRANS
189#endif
190
191#ifdef RPC_DEBUG_DATA
192static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
193{
194 u8 *buf = (u8 *) packet;
195 int j;
196
197 dprintk("RPC: %s\n", msg);
198 for (j = 0; j < count && j < 128; j += 4) {
199 if (!(j & 31)) {
200 if (j)
201 dprintk("\n");
202 dprintk("0x%04x ", j);
203 }
204 dprintk("%02x%02x%02x%02x ",
205 buf[j], buf[j+1], buf[j+2], buf[j+3]);
206 }
207 dprintk("\n");
208}
209#else
210static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
211{
212 /* NOP */
213}
214#endif
215
216struct sock_xprt {
217 struct rpc_xprt xprt;
218
219 /*
220 * Network layer
221 */
222 struct socket * sock;
223 struct sock * inet;
224
225 /*
226 * State of TCP reply receive
227 */
228 __be32 tcp_fraghdr,
229 tcp_xid,
230 tcp_calldir;
231
232 u32 tcp_offset,
233 tcp_reclen;
234
235 unsigned long tcp_copied,
236 tcp_flags;
237
238 /*
239 * Connection of transports
240 */
241 struct delayed_work connect_worker;
242 struct sockaddr_storage srcaddr;
243 unsigned short srcport;
244
245 /*
246 * UDP socket buffer size parameters
247 */
248 size_t rcvsize,
249 sndsize;
250
251 /*
252 * Saved socket callback addresses
253 */
254 void (*old_data_ready)(struct sock *, int);
255 void (*old_state_change)(struct sock *);
256 void (*old_write_space)(struct sock *);
257 void (*old_error_report)(struct sock *);
258};
259
260/*
261 * TCP receive state flags
262 */
263#define TCP_RCV_LAST_FRAG (1UL << 0)
264#define TCP_RCV_COPY_FRAGHDR (1UL << 1)
265#define TCP_RCV_COPY_XID (1UL << 2)
266#define TCP_RCV_COPY_DATA (1UL << 3)
267#define TCP_RCV_READ_CALLDIR (1UL << 4)
268#define TCP_RCV_COPY_CALLDIR (1UL << 5)
269
270/*
271 * TCP RPC flags
272 */
273#define TCP_RPC_REPLY (1UL << 6)
274
275static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
276{
277 return (struct sockaddr *) &xprt->addr;
278}
279
280static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
281{
282 return (struct sockaddr_un *) &xprt->addr;
283}
284
285static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
286{
287 return (struct sockaddr_in *) &xprt->addr;
288}
289
290static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
291{
292 return (struct sockaddr_in6 *) &xprt->addr;
293}
294
295static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
296{
297 struct sockaddr *sap = xs_addr(xprt);
298 struct sockaddr_in6 *sin6;
299 struct sockaddr_in *sin;
300 struct sockaddr_un *sun;
301 char buf[128];
302
303 switch (sap->sa_family) {
304 case AF_LOCAL:
305 sun = xs_addr_un(xprt);
306 strlcpy(buf, sun->sun_path, sizeof(buf));
307 xprt->address_strings[RPC_DISPLAY_ADDR] =
308 kstrdup(buf, GFP_KERNEL);
309 break;
310 case AF_INET:
311 (void)rpc_ntop(sap, buf, sizeof(buf));
312 xprt->address_strings[RPC_DISPLAY_ADDR] =
313 kstrdup(buf, GFP_KERNEL);
314 sin = xs_addr_in(xprt);
315 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
316 break;
317 case AF_INET6:
318 (void)rpc_ntop(sap, buf, sizeof(buf));
319 xprt->address_strings[RPC_DISPLAY_ADDR] =
320 kstrdup(buf, GFP_KERNEL);
321 sin6 = xs_addr_in6(xprt);
322 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
323 break;
324 default:
325 BUG();
326 }
327
328 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
329}
330
331static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
332{
333 struct sockaddr *sap = xs_addr(xprt);
334 char buf[128];
335
336 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
337 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
338
339 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
340 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
341}
342
343static void xs_format_peer_addresses(struct rpc_xprt *xprt,
344 const char *protocol,
345 const char *netid)
346{
347 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
348 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
349 xs_format_common_peer_addresses(xprt);
350 xs_format_common_peer_ports(xprt);
351}
352
353static void xs_update_peer_port(struct rpc_xprt *xprt)
354{
355 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
356 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
357
358 xs_format_common_peer_ports(xprt);
359}
360
361static void xs_free_peer_addresses(struct rpc_xprt *xprt)
362{
363 unsigned int i;
364
365 for (i = 0; i < RPC_DISPLAY_MAX; i++)
366 switch (i) {
367 case RPC_DISPLAY_PROTO:
368 case RPC_DISPLAY_NETID:
369 continue;
370 default:
371 kfree(xprt->address_strings[i]);
372 }
373}
374
375#define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
376
377static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
378{
379 struct msghdr msg = {
380 .msg_name = addr,
381 .msg_namelen = addrlen,
382 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
383 };
384 struct kvec iov = {
385 .iov_base = vec->iov_base + base,
386 .iov_len = vec->iov_len - base,
387 };
388
389 if (iov.iov_len != 0)
390 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
391 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
392}
393
394static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
395{
396 struct page **ppage;
397 unsigned int remainder;
398 int err, sent = 0;
399
400 remainder = xdr->page_len - base;
401 base += xdr->page_base;
402 ppage = xdr->pages + (base >> PAGE_SHIFT);
403 base &= ~PAGE_MASK;
404 for(;;) {
405 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
406 int flags = XS_SENDMSG_FLAGS;
407
408 remainder -= len;
409 if (remainder != 0 || more)
410 flags |= MSG_MORE;
411 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
412 if (remainder == 0 || err != len)
413 break;
414 sent += err;
415 ppage++;
416 base = 0;
417 }
418 if (sent == 0)
419 return err;
420 if (err > 0)
421 sent += err;
422 return sent;
423}
424
425/**
426 * xs_sendpages - write pages directly to a socket
427 * @sock: socket to send on
428 * @addr: UDP only -- address of destination
429 * @addrlen: UDP only -- length of destination address
430 * @xdr: buffer containing this request
431 * @base: starting position in the buffer
432 *
433 */
434static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
435{
436 unsigned int remainder = xdr->len - base;
437 int err, sent = 0;
438
439 if (unlikely(!sock))
440 return -ENOTSOCK;
441
442 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
443 if (base != 0) {
444 addr = NULL;
445 addrlen = 0;
446 }
447
448 if (base < xdr->head[0].iov_len || addr != NULL) {
449 unsigned int len = xdr->head[0].iov_len - base;
450 remainder -= len;
451 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
452 if (remainder == 0 || err != len)
453 goto out;
454 sent += err;
455 base = 0;
456 } else
457 base -= xdr->head[0].iov_len;
458
459 if (base < xdr->page_len) {
460 unsigned int len = xdr->page_len - base;
461 remainder -= len;
462 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
463 if (remainder == 0 || err != len)
464 goto out;
465 sent += err;
466 base = 0;
467 } else
468 base -= xdr->page_len;
469
470 if (base >= xdr->tail[0].iov_len)
471 return sent;
472 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
473out:
474 if (sent == 0)
475 return err;
476 if (err > 0)
477 sent += err;
478 return sent;
479}
480
481static void xs_nospace_callback(struct rpc_task *task)
482{
483 struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
484
485 transport->inet->sk_write_pending--;
486 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
487}
488
489/**
490 * xs_nospace - place task on wait queue if transmit was incomplete
491 * @task: task to put to sleep
492 *
493 */
494static int xs_nospace(struct rpc_task *task)
495{
496 struct rpc_rqst *req = task->tk_rqstp;
497 struct rpc_xprt *xprt = req->rq_xprt;
498 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
499 int ret = -EAGAIN;
500
501 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
502 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
503 req->rq_slen);
504
505 /* Protect against races with write_space */
506 spin_lock_bh(&xprt->transport_lock);
507
508 /* Don't race with disconnect */
509 if (xprt_connected(xprt)) {
510 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
511 /*
512 * Notify TCP that we're limited by the application
513 * window size
514 */
515 set_bit(SOCK_NOSPACE, &transport->sock->flags);
516 transport->inet->sk_write_pending++;
517 /* ...and wait for more buffer space */
518 xprt_wait_for_buffer_space(task, xs_nospace_callback);
519 }
520 } else {
521 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
522 ret = -ENOTCONN;
523 }
524
525 spin_unlock_bh(&xprt->transport_lock);
526 return ret;
527}
528
529/*
530 * Construct a stream transport record marker in @buf.
531 */
532static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
533{
534 u32 reclen = buf->len - sizeof(rpc_fraghdr);
535 rpc_fraghdr *base = buf->head[0].iov_base;
536 *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
537}
538
539/**
540 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
541 * @task: RPC task that manages the state of an RPC request
542 *
543 * Return values:
544 * 0: The request has been sent
545 * EAGAIN: The socket was blocked, please call again later to
546 * complete the request
547 * ENOTCONN: Caller needs to invoke connect logic then call again
548 * other: Some other error occured, the request was not sent
549 */
550static int xs_local_send_request(struct rpc_task *task)
551{
552 struct rpc_rqst *req = task->tk_rqstp;
553 struct rpc_xprt *xprt = req->rq_xprt;
554 struct sock_xprt *transport =
555 container_of(xprt, struct sock_xprt, xprt);
556 struct xdr_buf *xdr = &req->rq_snd_buf;
557 int status;
558
559 xs_encode_stream_record_marker(&req->rq_snd_buf);
560
561 xs_pktdump("packet data:",
562 req->rq_svec->iov_base, req->rq_svec->iov_len);
563
564 status = xs_sendpages(transport->sock, NULL, 0,
565 xdr, req->rq_bytes_sent);
566 dprintk("RPC: %s(%u) = %d\n",
567 __func__, xdr->len - req->rq_bytes_sent, status);
568 if (likely(status >= 0)) {
569 req->rq_bytes_sent += status;
570 req->rq_xmit_bytes_sent += status;
571 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
572 req->rq_bytes_sent = 0;
573 return 0;
574 }
575 status = -EAGAIN;
576 }
577
578 switch (status) {
579 case -EAGAIN:
580 status = xs_nospace(task);
581 break;
582 default:
583 dprintk("RPC: sendmsg returned unrecognized error %d\n",
584 -status);
585 case -EPIPE:
586 xs_close(xprt);
587 status = -ENOTCONN;
588 }
589
590 return status;
591}
592
593/**
594 * xs_udp_send_request - write an RPC request to a UDP socket
595 * @task: address of RPC task that manages the state of an RPC request
596 *
597 * Return values:
598 * 0: The request has been sent
599 * EAGAIN: The socket was blocked, please call again later to
600 * complete the request
601 * ENOTCONN: Caller needs to invoke connect logic then call again
602 * other: Some other error occurred, the request was not sent
603 */
604static int xs_udp_send_request(struct rpc_task *task)
605{
606 struct rpc_rqst *req = task->tk_rqstp;
607 struct rpc_xprt *xprt = req->rq_xprt;
608 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
609 struct xdr_buf *xdr = &req->rq_snd_buf;
610 int status;
611
612 xs_pktdump("packet data:",
613 req->rq_svec->iov_base,
614 req->rq_svec->iov_len);
615
616 if (!xprt_bound(xprt))
617 return -ENOTCONN;
618 status = xs_sendpages(transport->sock,
619 xs_addr(xprt),
620 xprt->addrlen, xdr,
621 req->rq_bytes_sent);
622
623 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
624 xdr->len - req->rq_bytes_sent, status);
625
626 if (status >= 0) {
627 req->rq_xmit_bytes_sent += status;
628 if (status >= req->rq_slen)
629 return 0;
630 /* Still some bytes left; set up for a retry later. */
631 status = -EAGAIN;
632 }
633
634 switch (status) {
635 case -ENOTSOCK:
636 status = -ENOTCONN;
637 /* Should we call xs_close() here? */
638 break;
639 case -EAGAIN:
640 status = xs_nospace(task);
641 break;
642 default:
643 dprintk("RPC: sendmsg returned unrecognized error %d\n",
644 -status);
645 case -ENETUNREACH:
646 case -EPIPE:
647 case -ECONNREFUSED:
648 /* When the server has died, an ICMP port unreachable message
649 * prompts ECONNREFUSED. */
650 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
651 }
652
653 return status;
654}
655
656/**
657 * xs_tcp_shutdown - gracefully shut down a TCP socket
658 * @xprt: transport
659 *
660 * Initiates a graceful shutdown of the TCP socket by calling the
661 * equivalent of shutdown(SHUT_WR);
662 */
663static void xs_tcp_shutdown(struct rpc_xprt *xprt)
664{
665 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
666 struct socket *sock = transport->sock;
667
668 if (sock != NULL)
669 kernel_sock_shutdown(sock, SHUT_WR);
670}
671
672/**
673 * xs_tcp_send_request - write an RPC request to a TCP socket
674 * @task: address of RPC task that manages the state of an RPC request
675 *
676 * Return values:
677 * 0: The request has been sent
678 * EAGAIN: The socket was blocked, please call again later to
679 * complete the request
680 * ENOTCONN: Caller needs to invoke connect logic then call again
681 * other: Some other error occurred, the request was not sent
682 *
683 * XXX: In the case of soft timeouts, should we eventually give up
684 * if sendmsg is not able to make progress?
685 */
686static int xs_tcp_send_request(struct rpc_task *task)
687{
688 struct rpc_rqst *req = task->tk_rqstp;
689 struct rpc_xprt *xprt = req->rq_xprt;
690 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
691 struct xdr_buf *xdr = &req->rq_snd_buf;
692 int status;
693
694 xs_encode_stream_record_marker(&req->rq_snd_buf);
695
696 xs_pktdump("packet data:",
697 req->rq_svec->iov_base,
698 req->rq_svec->iov_len);
699
700 /* Continue transmitting the packet/record. We must be careful
701 * to cope with writespace callbacks arriving _after_ we have
702 * called sendmsg(). */
703 while (1) {
704 status = xs_sendpages(transport->sock,
705 NULL, 0, xdr, req->rq_bytes_sent);
706
707 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
708 xdr->len - req->rq_bytes_sent, status);
709
710 if (unlikely(status < 0))
711 break;
712
713 /* If we've sent the entire packet, immediately
714 * reset the count of bytes sent. */
715 req->rq_bytes_sent += status;
716 req->rq_xmit_bytes_sent += status;
717 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
718 req->rq_bytes_sent = 0;
719 return 0;
720 }
721
722 if (status != 0)
723 continue;
724 status = -EAGAIN;
725 break;
726 }
727
728 switch (status) {
729 case -ENOTSOCK:
730 status = -ENOTCONN;
731 /* Should we call xs_close() here? */
732 break;
733 case -EAGAIN:
734 status = xs_nospace(task);
735 break;
736 default:
737 dprintk("RPC: sendmsg returned unrecognized error %d\n",
738 -status);
739 case -ECONNRESET:
740 case -EPIPE:
741 xs_tcp_shutdown(xprt);
742 case -ECONNREFUSED:
743 case -ENOTCONN:
744 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
745 }
746
747 return status;
748}
749
750/**
751 * xs_tcp_release_xprt - clean up after a tcp transmission
752 * @xprt: transport
753 * @task: rpc task
754 *
755 * This cleans up if an error causes us to abort the transmission of a request.
756 * In this case, the socket may need to be reset in order to avoid confusing
757 * the server.
758 */
759static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
760{
761 struct rpc_rqst *req;
762
763 if (task != xprt->snd_task)
764 return;
765 if (task == NULL)
766 goto out_release;
767 req = task->tk_rqstp;
768 if (req == NULL)
769 goto out_release;
770 if (req->rq_bytes_sent == 0)
771 goto out_release;
772 if (req->rq_bytes_sent == req->rq_snd_buf.len)
773 goto out_release;
774 set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
775out_release:
776 xprt_release_xprt(xprt, task);
777}
778
779static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
780{
781 transport->old_data_ready = sk->sk_data_ready;
782 transport->old_state_change = sk->sk_state_change;
783 transport->old_write_space = sk->sk_write_space;
784 transport->old_error_report = sk->sk_error_report;
785}
786
787static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
788{
789 sk->sk_data_ready = transport->old_data_ready;
790 sk->sk_state_change = transport->old_state_change;
791 sk->sk_write_space = transport->old_write_space;
792 sk->sk_error_report = transport->old_error_report;
793}
794
795static void xs_reset_transport(struct sock_xprt *transport)
796{
797 struct socket *sock = transport->sock;
798 struct sock *sk = transport->inet;
799
800 if (sk == NULL)
801 return;
802
803 transport->srcport = 0;
804
805 write_lock_bh(&sk->sk_callback_lock);
806 transport->inet = NULL;
807 transport->sock = NULL;
808
809 sk->sk_user_data = NULL;
810
811 xs_restore_old_callbacks(transport, sk);
812 write_unlock_bh(&sk->sk_callback_lock);
813
814 sk->sk_no_check = 0;
815
816 sock_release(sock);
817}
818
819/**
820 * xs_close - close a socket
821 * @xprt: transport
822 *
823 * This is used when all requests are complete; ie, no DRC state remains
824 * on the server we want to save.
825 *
826 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
827 * xs_reset_transport() zeroing the socket from underneath a writer.
828 */
829static void xs_close(struct rpc_xprt *xprt)
830{
831 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
832
833 dprintk("RPC: xs_close xprt %p\n", xprt);
834
835 xs_reset_transport(transport);
836 xprt->reestablish_timeout = 0;
837
838 smp_mb__before_clear_bit();
839 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
840 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
841 clear_bit(XPRT_CLOSING, &xprt->state);
842 smp_mb__after_clear_bit();
843 xprt_disconnect_done(xprt);
844}
845
846static void xs_tcp_close(struct rpc_xprt *xprt)
847{
848 if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state))
849 xs_close(xprt);
850 else
851 xs_tcp_shutdown(xprt);
852}
853
854/**
855 * xs_destroy - prepare to shutdown a transport
856 * @xprt: doomed transport
857 *
858 */
859static void xs_destroy(struct rpc_xprt *xprt)
860{
861 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
862
863 dprintk("RPC: xs_destroy xprt %p\n", xprt);
864
865 cancel_delayed_work_sync(&transport->connect_worker);
866
867 xs_close(xprt);
868 xs_free_peer_addresses(xprt);
869 xprt_free(xprt);
870 module_put(THIS_MODULE);
871}
872
873static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
874{
875 return (struct rpc_xprt *) sk->sk_user_data;
876}
877
878static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
879{
880 struct xdr_skb_reader desc = {
881 .skb = skb,
882 .offset = sizeof(rpc_fraghdr),
883 .count = skb->len - sizeof(rpc_fraghdr),
884 };
885
886 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
887 return -1;
888 if (desc.count)
889 return -1;
890 return 0;
891}
892
893/**
894 * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets
895 * @sk: socket with data to read
896 * @len: how much data to read
897 *
898 * Currently this assumes we can read the whole reply in a single gulp.
899 */
900static void xs_local_data_ready(struct sock *sk, int len)
901{
902 struct rpc_task *task;
903 struct rpc_xprt *xprt;
904 struct rpc_rqst *rovr;
905 struct sk_buff *skb;
906 int err, repsize, copied;
907 u32 _xid;
908 __be32 *xp;
909
910 read_lock_bh(&sk->sk_callback_lock);
911 dprintk("RPC: %s...\n", __func__);
912 xprt = xprt_from_sock(sk);
913 if (xprt == NULL)
914 goto out;
915
916 skb = skb_recv_datagram(sk, 0, 1, &err);
917 if (skb == NULL)
918 goto out;
919
920 if (xprt->shutdown)
921 goto dropit;
922
923 repsize = skb->len - sizeof(rpc_fraghdr);
924 if (repsize < 4) {
925 dprintk("RPC: impossible RPC reply size %d\n", repsize);
926 goto dropit;
927 }
928
929 /* Copy the XID from the skb... */
930 xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
931 if (xp == NULL)
932 goto dropit;
933
934 /* Look up and lock the request corresponding to the given XID */
935 spin_lock(&xprt->transport_lock);
936 rovr = xprt_lookup_rqst(xprt, *xp);
937 if (!rovr)
938 goto out_unlock;
939 task = rovr->rq_task;
940
941 copied = rovr->rq_private_buf.buflen;
942 if (copied > repsize)
943 copied = repsize;
944
945 if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
946 dprintk("RPC: sk_buff copy failed\n");
947 goto out_unlock;
948 }
949
950 xprt_complete_rqst(task, copied);
951
952 out_unlock:
953 spin_unlock(&xprt->transport_lock);
954 dropit:
955 skb_free_datagram(sk, skb);
956 out:
957 read_unlock_bh(&sk->sk_callback_lock);
958}
959
960/**
961 * xs_udp_data_ready - "data ready" callback for UDP sockets
962 * @sk: socket with data to read
963 * @len: how much data to read
964 *
965 */
966static void xs_udp_data_ready(struct sock *sk, int len)
967{
968 struct rpc_task *task;
969 struct rpc_xprt *xprt;
970 struct rpc_rqst *rovr;
971 struct sk_buff *skb;
972 int err, repsize, copied;
973 u32 _xid;
974 __be32 *xp;
975
976 read_lock_bh(&sk->sk_callback_lock);
977 dprintk("RPC: xs_udp_data_ready...\n");
978 if (!(xprt = xprt_from_sock(sk)))
979 goto out;
980
981 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
982 goto out;
983
984 if (xprt->shutdown)
985 goto dropit;
986
987 repsize = skb->len - sizeof(struct udphdr);
988 if (repsize < 4) {
989 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
990 goto dropit;
991 }
992
993 /* Copy the XID from the skb... */
994 xp = skb_header_pointer(skb, sizeof(struct udphdr),
995 sizeof(_xid), &_xid);
996 if (xp == NULL)
997 goto dropit;
998
999 /* Look up and lock the request corresponding to the given XID */
1000 spin_lock(&xprt->transport_lock);
1001 rovr = xprt_lookup_rqst(xprt, *xp);
1002 if (!rovr)
1003 goto out_unlock;
1004 task = rovr->rq_task;
1005
1006 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1007 copied = repsize;
1008
1009 /* Suck it into the iovec, verify checksum if not done by hw. */
1010 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1011 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
1012 goto out_unlock;
1013 }
1014
1015 UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
1016
1017 /* Something worked... */
1018 dst_confirm(skb_dst(skb));
1019
1020 xprt_adjust_cwnd(task, copied);
1021 xprt_complete_rqst(task, copied);
1022
1023 out_unlock:
1024 spin_unlock(&xprt->transport_lock);
1025 dropit:
1026 skb_free_datagram(sk, skb);
1027 out:
1028 read_unlock_bh(&sk->sk_callback_lock);
1029}
1030
1031static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1032{
1033 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1034 size_t len, used;
1035 char *p;
1036
1037 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1038 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1039 used = xdr_skb_read_bits(desc, p, len);
1040 transport->tcp_offset += used;
1041 if (used != len)
1042 return;
1043
1044 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1045 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1046 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1047 else
1048 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1049 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1050
1051 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1052 transport->tcp_offset = 0;
1053
1054 /* Sanity check of the record length */
1055 if (unlikely(transport->tcp_reclen < 8)) {
1056 dprintk("RPC: invalid TCP record fragment length\n");
1057 xprt_force_disconnect(xprt);
1058 return;
1059 }
1060 dprintk("RPC: reading TCP record fragment of length %d\n",
1061 transport->tcp_reclen);
1062}
1063
1064static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1065{
1066 if (transport->tcp_offset == transport->tcp_reclen) {
1067 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1068 transport->tcp_offset = 0;
1069 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1070 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1071 transport->tcp_flags |= TCP_RCV_COPY_XID;
1072 transport->tcp_copied = 0;
1073 }
1074 }
1075}
1076
1077static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1078{
1079 size_t len, used;
1080 char *p;
1081
1082 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1083 dprintk("RPC: reading XID (%Zu bytes)\n", len);
1084 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1085 used = xdr_skb_read_bits(desc, p, len);
1086 transport->tcp_offset += used;
1087 if (used != len)
1088 return;
1089 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1090 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1091 transport->tcp_copied = 4;
1092 dprintk("RPC: reading %s XID %08x\n",
1093 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1094 : "request with",
1095 ntohl(transport->tcp_xid));
1096 xs_tcp_check_fraghdr(transport);
1097}
1098
1099static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1100 struct xdr_skb_reader *desc)
1101{
1102 size_t len, used;
1103 u32 offset;
1104 char *p;
1105
1106 /*
1107 * We want transport->tcp_offset to be 8 at the end of this routine
1108 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1109 * When this function is called for the first time,
1110 * transport->tcp_offset is 4 (after having already read the xid).
1111 */
1112 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1113 len = sizeof(transport->tcp_calldir) - offset;
1114 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len);
1115 p = ((char *) &transport->tcp_calldir) + offset;
1116 used = xdr_skb_read_bits(desc, p, len);
1117 transport->tcp_offset += used;
1118 if (used != len)
1119 return;
1120 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1121 /*
1122 * We don't yet have the XDR buffer, so we will write the calldir
1123 * out after we get the buffer from the 'struct rpc_rqst'
1124 */
1125 switch (ntohl(transport->tcp_calldir)) {
1126 case RPC_REPLY:
1127 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1128 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1129 transport->tcp_flags |= TCP_RPC_REPLY;
1130 break;
1131 case RPC_CALL:
1132 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1133 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1134 transport->tcp_flags &= ~TCP_RPC_REPLY;
1135 break;
1136 default:
1137 dprintk("RPC: invalid request message type\n");
1138 xprt_force_disconnect(&transport->xprt);
1139 }
1140 xs_tcp_check_fraghdr(transport);
1141}
1142
1143static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1144 struct xdr_skb_reader *desc,
1145 struct rpc_rqst *req)
1146{
1147 struct sock_xprt *transport =
1148 container_of(xprt, struct sock_xprt, xprt);
1149 struct xdr_buf *rcvbuf;
1150 size_t len;
1151 ssize_t r;
1152
1153 rcvbuf = &req->rq_private_buf;
1154
1155 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1156 /*
1157 * Save the RPC direction in the XDR buffer
1158 */
1159 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1160 &transport->tcp_calldir,
1161 sizeof(transport->tcp_calldir));
1162 transport->tcp_copied += sizeof(transport->tcp_calldir);
1163 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1164 }
1165
1166 len = desc->count;
1167 if (len > transport->tcp_reclen - transport->tcp_offset) {
1168 struct xdr_skb_reader my_desc;
1169
1170 len = transport->tcp_reclen - transport->tcp_offset;
1171 memcpy(&my_desc, desc, sizeof(my_desc));
1172 my_desc.count = len;
1173 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1174 &my_desc, xdr_skb_read_bits);
1175 desc->count -= r;
1176 desc->offset += r;
1177 } else
1178 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1179 desc, xdr_skb_read_bits);
1180
1181 if (r > 0) {
1182 transport->tcp_copied += r;
1183 transport->tcp_offset += r;
1184 }
1185 if (r != len) {
1186 /* Error when copying to the receive buffer,
1187 * usually because we weren't able to allocate
1188 * additional buffer pages. All we can do now
1189 * is turn off TCP_RCV_COPY_DATA, so the request
1190 * will not receive any additional updates,
1191 * and time out.
1192 * Any remaining data from this record will
1193 * be discarded.
1194 */
1195 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1196 dprintk("RPC: XID %08x truncated request\n",
1197 ntohl(transport->tcp_xid));
1198 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1199 "tcp_offset = %u, tcp_reclen = %u\n",
1200 xprt, transport->tcp_copied,
1201 transport->tcp_offset, transport->tcp_reclen);
1202 return;
1203 }
1204
1205 dprintk("RPC: XID %08x read %Zd bytes\n",
1206 ntohl(transport->tcp_xid), r);
1207 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1208 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1209 transport->tcp_offset, transport->tcp_reclen);
1210
1211 if (transport->tcp_copied == req->rq_private_buf.buflen)
1212 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1213 else if (transport->tcp_offset == transport->tcp_reclen) {
1214 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1215 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1216 }
1217}
1218
1219/*
1220 * Finds the request corresponding to the RPC xid and invokes the common
1221 * tcp read code to read the data.
1222 */
1223static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1224 struct xdr_skb_reader *desc)
1225{
1226 struct sock_xprt *transport =
1227 container_of(xprt, struct sock_xprt, xprt);
1228 struct rpc_rqst *req;
1229
1230 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1231
1232 /* Find and lock the request corresponding to this xid */
1233 spin_lock(&xprt->transport_lock);
1234 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1235 if (!req) {
1236 dprintk("RPC: XID %08x request not found!\n",
1237 ntohl(transport->tcp_xid));
1238 spin_unlock(&xprt->transport_lock);
1239 return -1;
1240 }
1241
1242 xs_tcp_read_common(xprt, desc, req);
1243
1244 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1245 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1246
1247 spin_unlock(&xprt->transport_lock);
1248 return 0;
1249}
1250
1251#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1252/*
1253 * Obtains an rpc_rqst previously allocated and invokes the common
1254 * tcp read code to read the data. The result is placed in the callback
1255 * queue.
1256 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1257 * connection and return -1.
1258 */
1259static inline int xs_tcp_read_callback(struct rpc_xprt *xprt,
1260 struct xdr_skb_reader *desc)
1261{
1262 struct sock_xprt *transport =
1263 container_of(xprt, struct sock_xprt, xprt);
1264 struct rpc_rqst *req;
1265
1266 req = xprt_alloc_bc_request(xprt);
1267 if (req == NULL) {
1268 printk(KERN_WARNING "Callback slot table overflowed\n");
1269 xprt_force_disconnect(xprt);
1270 return -1;
1271 }
1272
1273 req->rq_xid = transport->tcp_xid;
1274 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1275 xs_tcp_read_common(xprt, desc, req);
1276
1277 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) {
1278 struct svc_serv *bc_serv = xprt->bc_serv;
1279
1280 /*
1281 * Add callback request to callback list. The callback
1282 * service sleeps on the sv_cb_waitq waiting for new
1283 * requests. Wake it up after adding enqueing the
1284 * request.
1285 */
1286 dprintk("RPC: add callback request to list\n");
1287 spin_lock(&bc_serv->sv_cb_lock);
1288 list_add(&req->rq_bc_list, &bc_serv->sv_cb_list);
1289 spin_unlock(&bc_serv->sv_cb_lock);
1290 wake_up(&bc_serv->sv_cb_waitq);
1291 }
1292
1293 req->rq_private_buf.len = transport->tcp_copied;
1294
1295 return 0;
1296}
1297
1298static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1299 struct xdr_skb_reader *desc)
1300{
1301 struct sock_xprt *transport =
1302 container_of(xprt, struct sock_xprt, xprt);
1303
1304 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1305 xs_tcp_read_reply(xprt, desc) :
1306 xs_tcp_read_callback(xprt, desc);
1307}
1308#else
1309static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1310 struct xdr_skb_reader *desc)
1311{
1312 return xs_tcp_read_reply(xprt, desc);
1313}
1314#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1315
1316/*
1317 * Read data off the transport. This can be either an RPC_CALL or an
1318 * RPC_REPLY. Relay the processing to helper functions.
1319 */
1320static void xs_tcp_read_data(struct rpc_xprt *xprt,
1321 struct xdr_skb_reader *desc)
1322{
1323 struct sock_xprt *transport =
1324 container_of(xprt, struct sock_xprt, xprt);
1325
1326 if (_xs_tcp_read_data(xprt, desc) == 0)
1327 xs_tcp_check_fraghdr(transport);
1328 else {
1329 /*
1330 * The transport_lock protects the request handling.
1331 * There's no need to hold it to update the tcp_flags.
1332 */
1333 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1334 }
1335}
1336
1337static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1338{
1339 size_t len;
1340
1341 len = transport->tcp_reclen - transport->tcp_offset;
1342 if (len > desc->count)
1343 len = desc->count;
1344 desc->count -= len;
1345 desc->offset += len;
1346 transport->tcp_offset += len;
1347 dprintk("RPC: discarded %Zu bytes\n", len);
1348 xs_tcp_check_fraghdr(transport);
1349}
1350
1351static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1352{
1353 struct rpc_xprt *xprt = rd_desc->arg.data;
1354 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1355 struct xdr_skb_reader desc = {
1356 .skb = skb,
1357 .offset = offset,
1358 .count = len,
1359 };
1360
1361 dprintk("RPC: xs_tcp_data_recv started\n");
1362 do {
1363 /* Read in a new fragment marker if necessary */
1364 /* Can we ever really expect to get completely empty fragments? */
1365 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1366 xs_tcp_read_fraghdr(xprt, &desc);
1367 continue;
1368 }
1369 /* Read in the xid if necessary */
1370 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1371 xs_tcp_read_xid(transport, &desc);
1372 continue;
1373 }
1374 /* Read in the call/reply flag */
1375 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1376 xs_tcp_read_calldir(transport, &desc);
1377 continue;
1378 }
1379 /* Read in the request data */
1380 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1381 xs_tcp_read_data(xprt, &desc);
1382 continue;
1383 }
1384 /* Skip over any trailing bytes on short reads */
1385 xs_tcp_read_discard(transport, &desc);
1386 } while (desc.count);
1387 dprintk("RPC: xs_tcp_data_recv done\n");
1388 return len - desc.count;
1389}
1390
1391/**
1392 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1393 * @sk: socket with data to read
1394 * @bytes: how much data to read
1395 *
1396 */
1397static void xs_tcp_data_ready(struct sock *sk, int bytes)
1398{
1399 struct rpc_xprt *xprt;
1400 read_descriptor_t rd_desc;
1401 int read;
1402
1403 dprintk("RPC: xs_tcp_data_ready...\n");
1404
1405 read_lock_bh(&sk->sk_callback_lock);
1406 if (!(xprt = xprt_from_sock(sk)))
1407 goto out;
1408 if (xprt->shutdown)
1409 goto out;
1410
1411 /* Any data means we had a useful conversation, so
1412 * the we don't need to delay the next reconnect
1413 */
1414 if (xprt->reestablish_timeout)
1415 xprt->reestablish_timeout = 0;
1416
1417 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1418 rd_desc.arg.data = xprt;
1419 do {
1420 rd_desc.count = 65536;
1421 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1422 } while (read > 0);
1423out:
1424 read_unlock_bh(&sk->sk_callback_lock);
1425}
1426
1427/*
1428 * Do the equivalent of linger/linger2 handling for dealing with
1429 * broken servers that don't close the socket in a timely
1430 * fashion
1431 */
1432static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1433 unsigned long timeout)
1434{
1435 struct sock_xprt *transport;
1436
1437 if (xprt_test_and_set_connecting(xprt))
1438 return;
1439 set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1440 transport = container_of(xprt, struct sock_xprt, xprt);
1441 queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1442 timeout);
1443}
1444
1445static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1446{
1447 struct sock_xprt *transport;
1448
1449 transport = container_of(xprt, struct sock_xprt, xprt);
1450
1451 if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1452 !cancel_delayed_work(&transport->connect_worker))
1453 return;
1454 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1455 xprt_clear_connecting(xprt);
1456}
1457
1458static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1459{
1460 smp_mb__before_clear_bit();
1461 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1462 clear_bit(XPRT_CLOSING, &xprt->state);
1463 smp_mb__after_clear_bit();
1464 /* Mark transport as closed and wake up all pending tasks */
1465 xprt_disconnect_done(xprt);
1466}
1467
1468/**
1469 * xs_tcp_state_change - callback to handle TCP socket state changes
1470 * @sk: socket whose state has changed
1471 *
1472 */
1473static void xs_tcp_state_change(struct sock *sk)
1474{
1475 struct rpc_xprt *xprt;
1476
1477 read_lock_bh(&sk->sk_callback_lock);
1478 if (!(xprt = xprt_from_sock(sk)))
1479 goto out;
1480 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1481 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1482 sk->sk_state, xprt_connected(xprt),
1483 sock_flag(sk, SOCK_DEAD),
1484 sock_flag(sk, SOCK_ZAPPED),
1485 sk->sk_shutdown);
1486
1487 switch (sk->sk_state) {
1488 case TCP_ESTABLISHED:
1489 spin_lock(&xprt->transport_lock);
1490 if (!xprt_test_and_set_connected(xprt)) {
1491 struct sock_xprt *transport = container_of(xprt,
1492 struct sock_xprt, xprt);
1493
1494 /* Reset TCP record info */
1495 transport->tcp_offset = 0;
1496 transport->tcp_reclen = 0;
1497 transport->tcp_copied = 0;
1498 transport->tcp_flags =
1499 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1500
1501 xprt_wake_pending_tasks(xprt, -EAGAIN);
1502 }
1503 spin_unlock(&xprt->transport_lock);
1504 break;
1505 case TCP_FIN_WAIT1:
1506 /* The client initiated a shutdown of the socket */
1507 xprt->connect_cookie++;
1508 xprt->reestablish_timeout = 0;
1509 set_bit(XPRT_CLOSING, &xprt->state);
1510 smp_mb__before_clear_bit();
1511 clear_bit(XPRT_CONNECTED, &xprt->state);
1512 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1513 smp_mb__after_clear_bit();
1514 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1515 break;
1516 case TCP_CLOSE_WAIT:
1517 /* The server initiated a shutdown of the socket */
1518 xprt_force_disconnect(xprt);
1519 xprt->connect_cookie++;
1520 case TCP_CLOSING:
1521 /*
1522 * If the server closed down the connection, make sure that
1523 * we back off before reconnecting
1524 */
1525 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1526 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1527 break;
1528 case TCP_LAST_ACK:
1529 set_bit(XPRT_CLOSING, &xprt->state);
1530 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1531 smp_mb__before_clear_bit();
1532 clear_bit(XPRT_CONNECTED, &xprt->state);
1533 smp_mb__after_clear_bit();
1534 break;
1535 case TCP_CLOSE:
1536 xs_tcp_cancel_linger_timeout(xprt);
1537 xs_sock_mark_closed(xprt);
1538 }
1539 out:
1540 read_unlock_bh(&sk->sk_callback_lock);
1541}
1542
1543/**
1544 * xs_error_report - callback mainly for catching socket errors
1545 * @sk: socket
1546 */
1547static void xs_error_report(struct sock *sk)
1548{
1549 struct rpc_xprt *xprt;
1550
1551 read_lock_bh(&sk->sk_callback_lock);
1552 if (!(xprt = xprt_from_sock(sk)))
1553 goto out;
1554 dprintk("RPC: %s client %p...\n"
1555 "RPC: error %d\n",
1556 __func__, xprt, sk->sk_err);
1557 xprt_wake_pending_tasks(xprt, -EAGAIN);
1558out:
1559 read_unlock_bh(&sk->sk_callback_lock);
1560}
1561
1562static void xs_write_space(struct sock *sk)
1563{
1564 struct socket *sock;
1565 struct rpc_xprt *xprt;
1566
1567 if (unlikely(!(sock = sk->sk_socket)))
1568 return;
1569 clear_bit(SOCK_NOSPACE, &sock->flags);
1570
1571 if (unlikely(!(xprt = xprt_from_sock(sk))))
1572 return;
1573 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1574 return;
1575
1576 xprt_write_space(xprt);
1577}
1578
1579/**
1580 * xs_udp_write_space - callback invoked when socket buffer space
1581 * becomes available
1582 * @sk: socket whose state has changed
1583 *
1584 * Called when more output buffer space is available for this socket.
1585 * We try not to wake our writers until they can make "significant"
1586 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1587 * with a bunch of small requests.
1588 */
1589static void xs_udp_write_space(struct sock *sk)
1590{
1591 read_lock_bh(&sk->sk_callback_lock);
1592
1593 /* from net/core/sock.c:sock_def_write_space */
1594 if (sock_writeable(sk))
1595 xs_write_space(sk);
1596
1597 read_unlock_bh(&sk->sk_callback_lock);
1598}
1599
1600/**
1601 * xs_tcp_write_space - callback invoked when socket buffer space
1602 * becomes available
1603 * @sk: socket whose state has changed
1604 *
1605 * Called when more output buffer space is available for this socket.
1606 * We try not to wake our writers until they can make "significant"
1607 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1608 * with a bunch of small requests.
1609 */
1610static void xs_tcp_write_space(struct sock *sk)
1611{
1612 read_lock_bh(&sk->sk_callback_lock);
1613
1614 /* from net/core/stream.c:sk_stream_write_space */
1615 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk))
1616 xs_write_space(sk);
1617
1618 read_unlock_bh(&sk->sk_callback_lock);
1619}
1620
1621static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1622{
1623 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1624 struct sock *sk = transport->inet;
1625
1626 if (transport->rcvsize) {
1627 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1628 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1629 }
1630 if (transport->sndsize) {
1631 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1632 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1633 sk->sk_write_space(sk);
1634 }
1635}
1636
1637/**
1638 * xs_udp_set_buffer_size - set send and receive limits
1639 * @xprt: generic transport
1640 * @sndsize: requested size of send buffer, in bytes
1641 * @rcvsize: requested size of receive buffer, in bytes
1642 *
1643 * Set socket send and receive buffer size limits.
1644 */
1645static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1646{
1647 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1648
1649 transport->sndsize = 0;
1650 if (sndsize)
1651 transport->sndsize = sndsize + 1024;
1652 transport->rcvsize = 0;
1653 if (rcvsize)
1654 transport->rcvsize = rcvsize + 1024;
1655
1656 xs_udp_do_set_buffer_size(xprt);
1657}
1658
1659/**
1660 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1661 * @task: task that timed out
1662 *
1663 * Adjust the congestion window after a retransmit timeout has occurred.
1664 */
1665static void xs_udp_timer(struct rpc_task *task)
1666{
1667 xprt_adjust_cwnd(task, -ETIMEDOUT);
1668}
1669
1670static unsigned short xs_get_random_port(void)
1671{
1672 unsigned short range = xprt_max_resvport - xprt_min_resvport;
1673 unsigned short rand = (unsigned short) net_random() % range;
1674 return rand + xprt_min_resvport;
1675}
1676
1677/**
1678 * xs_set_port - reset the port number in the remote endpoint address
1679 * @xprt: generic transport
1680 * @port: new port number
1681 *
1682 */
1683static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1684{
1685 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1686
1687 rpc_set_port(xs_addr(xprt), port);
1688 xs_update_peer_port(xprt);
1689}
1690
1691static unsigned short xs_get_srcport(struct sock_xprt *transport)
1692{
1693 unsigned short port = transport->srcport;
1694
1695 if (port == 0 && transport->xprt.resvport)
1696 port = xs_get_random_port();
1697 return port;
1698}
1699
1700static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1701{
1702 if (transport->srcport != 0)
1703 transport->srcport = 0;
1704 if (!transport->xprt.resvport)
1705 return 0;
1706 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1707 return xprt_max_resvport;
1708 return --port;
1709}
1710static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1711{
1712 struct sockaddr_storage myaddr;
1713 int err, nloop = 0;
1714 unsigned short port = xs_get_srcport(transport);
1715 unsigned short last;
1716
1717 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1718 do {
1719 rpc_set_port((struct sockaddr *)&myaddr, port);
1720 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1721 transport->xprt.addrlen);
1722 if (port == 0)
1723 break;
1724 if (err == 0) {
1725 transport->srcport = port;
1726 break;
1727 }
1728 last = port;
1729 port = xs_next_srcport(transport, port);
1730 if (port > last)
1731 nloop++;
1732 } while (err == -EADDRINUSE && nloop != 2);
1733
1734 if (myaddr.ss_family == AF_INET)
1735 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1736 &((struct sockaddr_in *)&myaddr)->sin_addr,
1737 port, err ? "failed" : "ok", err);
1738 else
1739 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1740 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1741 port, err ? "failed" : "ok", err);
1742 return err;
1743}
1744
1745/*
1746 * We don't support autobind on AF_LOCAL sockets
1747 */
1748static void xs_local_rpcbind(struct rpc_task *task)
1749{
1750 xprt_set_bound(task->tk_xprt);
1751}
1752
1753static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1754{
1755}
1756
1757#ifdef CONFIG_DEBUG_LOCK_ALLOC
1758static struct lock_class_key xs_key[2];
1759static struct lock_class_key xs_slock_key[2];
1760
1761static inline void xs_reclassify_socketu(struct socket *sock)
1762{
1763 struct sock *sk = sock->sk;
1764
1765 BUG_ON(sock_owned_by_user(sk));
1766 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1767 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1768}
1769
1770static inline void xs_reclassify_socket4(struct socket *sock)
1771{
1772 struct sock *sk = sock->sk;
1773
1774 BUG_ON(sock_owned_by_user(sk));
1775 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1776 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1777}
1778
1779static inline void xs_reclassify_socket6(struct socket *sock)
1780{
1781 struct sock *sk = sock->sk;
1782
1783 BUG_ON(sock_owned_by_user(sk));
1784 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1785 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1786}
1787
1788static inline void xs_reclassify_socket(int family, struct socket *sock)
1789{
1790 switch (family) {
1791 case AF_LOCAL:
1792 xs_reclassify_socketu(sock);
1793 break;
1794 case AF_INET:
1795 xs_reclassify_socket4(sock);
1796 break;
1797 case AF_INET6:
1798 xs_reclassify_socket6(sock);
1799 break;
1800 }
1801}
1802#else
1803static inline void xs_reclassify_socketu(struct socket *sock)
1804{
1805}
1806
1807static inline void xs_reclassify_socket4(struct socket *sock)
1808{
1809}
1810
1811static inline void xs_reclassify_socket6(struct socket *sock)
1812{
1813}
1814
1815static inline void xs_reclassify_socket(int family, struct socket *sock)
1816{
1817}
1818#endif
1819
1820static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1821 struct sock_xprt *transport, int family, int type, int protocol)
1822{
1823 struct socket *sock;
1824 int err;
1825
1826 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1827 if (err < 0) {
1828 dprintk("RPC: can't create %d transport socket (%d).\n",
1829 protocol, -err);
1830 goto out;
1831 }
1832 xs_reclassify_socket(family, sock);
1833
1834 err = xs_bind(transport, sock);
1835 if (err) {
1836 sock_release(sock);
1837 goto out;
1838 }
1839
1840 return sock;
1841out:
1842 return ERR_PTR(err);
1843}
1844
1845static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1846 struct socket *sock)
1847{
1848 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1849 xprt);
1850
1851 if (!transport->inet) {
1852 struct sock *sk = sock->sk;
1853
1854 write_lock_bh(&sk->sk_callback_lock);
1855
1856 xs_save_old_callbacks(transport, sk);
1857
1858 sk->sk_user_data = xprt;
1859 sk->sk_data_ready = xs_local_data_ready;
1860 sk->sk_write_space = xs_udp_write_space;
1861 sk->sk_error_report = xs_error_report;
1862 sk->sk_allocation = GFP_ATOMIC;
1863
1864 xprt_clear_connected(xprt);
1865
1866 /* Reset to new socket */
1867 transport->sock = sock;
1868 transport->inet = sk;
1869
1870 write_unlock_bh(&sk->sk_callback_lock);
1871 }
1872
1873 /* Tell the socket layer to start connecting... */
1874 xprt->stat.connect_count++;
1875 xprt->stat.connect_start = jiffies;
1876 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1877}
1878
1879/**
1880 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1881 * @xprt: RPC transport to connect
1882 * @transport: socket transport to connect
1883 * @create_sock: function to create a socket of the correct type
1884 *
1885 * Invoked by a work queue tasklet.
1886 */
1887static void xs_local_setup_socket(struct work_struct *work)
1888{
1889 struct sock_xprt *transport =
1890 container_of(work, struct sock_xprt, connect_worker.work);
1891 struct rpc_xprt *xprt = &transport->xprt;
1892 struct socket *sock;
1893 int status = -EIO;
1894
1895 if (xprt->shutdown)
1896 goto out;
1897
1898 current->flags |= PF_FSTRANS;
1899
1900 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1901 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1902 SOCK_STREAM, 0, &sock, 1);
1903 if (status < 0) {
1904 dprintk("RPC: can't create AF_LOCAL "
1905 "transport socket (%d).\n", -status);
1906 goto out;
1907 }
1908 xs_reclassify_socketu(sock);
1909
1910 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1911 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1912
1913 status = xs_local_finish_connecting(xprt, sock);
1914 switch (status) {
1915 case 0:
1916 dprintk("RPC: xprt %p connected to %s\n",
1917 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1918 xprt_set_connected(xprt);
1919 break;
1920 case -ENOENT:
1921 dprintk("RPC: xprt %p: socket %s does not exist\n",
1922 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1923 break;
1924 default:
1925 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1926 __func__, -status,
1927 xprt->address_strings[RPC_DISPLAY_ADDR]);
1928 }
1929
1930out:
1931 xprt_clear_connecting(xprt);
1932 xprt_wake_pending_tasks(xprt, status);
1933 current->flags &= ~PF_FSTRANS;
1934}
1935
1936static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1937{
1938 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1939
1940 if (!transport->inet) {
1941 struct sock *sk = sock->sk;
1942
1943 write_lock_bh(&sk->sk_callback_lock);
1944
1945 xs_save_old_callbacks(transport, sk);
1946
1947 sk->sk_user_data = xprt;
1948 sk->sk_data_ready = xs_udp_data_ready;
1949 sk->sk_write_space = xs_udp_write_space;
1950 sk->sk_error_report = xs_error_report;
1951 sk->sk_no_check = UDP_CSUM_NORCV;
1952 sk->sk_allocation = GFP_ATOMIC;
1953
1954 xprt_set_connected(xprt);
1955
1956 /* Reset to new socket */
1957 transport->sock = sock;
1958 transport->inet = sk;
1959
1960 write_unlock_bh(&sk->sk_callback_lock);
1961 }
1962 xs_udp_do_set_buffer_size(xprt);
1963}
1964
1965static void xs_udp_setup_socket(struct work_struct *work)
1966{
1967 struct sock_xprt *transport =
1968 container_of(work, struct sock_xprt, connect_worker.work);
1969 struct rpc_xprt *xprt = &transport->xprt;
1970 struct socket *sock = transport->sock;
1971 int status = -EIO;
1972
1973 if (xprt->shutdown)
1974 goto out;
1975
1976 current->flags |= PF_FSTRANS;
1977
1978 /* Start by resetting any existing state */
1979 xs_reset_transport(transport);
1980 sock = xs_create_sock(xprt, transport,
1981 xs_addr(xprt)->sa_family, SOCK_DGRAM, IPPROTO_UDP);
1982 if (IS_ERR(sock))
1983 goto out;
1984
1985 dprintk("RPC: worker connecting xprt %p via %s to "
1986 "%s (port %s)\n", xprt,
1987 xprt->address_strings[RPC_DISPLAY_PROTO],
1988 xprt->address_strings[RPC_DISPLAY_ADDR],
1989 xprt->address_strings[RPC_DISPLAY_PORT]);
1990
1991 xs_udp_finish_connecting(xprt, sock);
1992 status = 0;
1993out:
1994 xprt_clear_connecting(xprt);
1995 xprt_wake_pending_tasks(xprt, status);
1996 current->flags &= ~PF_FSTRANS;
1997}
1998
1999/*
2000 * We need to preserve the port number so the reply cache on the server can
2001 * find our cached RPC replies when we get around to reconnecting.
2002 */
2003static void xs_abort_connection(struct sock_xprt *transport)
2004{
2005 int result;
2006 struct sockaddr any;
2007
2008 dprintk("RPC: disconnecting xprt %p to reuse port\n", transport);
2009
2010 /*
2011 * Disconnect the transport socket by doing a connect operation
2012 * with AF_UNSPEC. This should return immediately...
2013 */
2014 memset(&any, 0, sizeof(any));
2015 any.sa_family = AF_UNSPEC;
2016 result = kernel_connect(transport->sock, &any, sizeof(any), 0);
2017 if (!result)
2018 xs_sock_mark_closed(&transport->xprt);
2019 else
2020 dprintk("RPC: AF_UNSPEC connect return code %d\n",
2021 result);
2022}
2023
2024static void xs_tcp_reuse_connection(struct sock_xprt *transport)
2025{
2026 unsigned int state = transport->inet->sk_state;
2027
2028 if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED) {
2029 /* we don't need to abort the connection if the socket
2030 * hasn't undergone a shutdown
2031 */
2032 if (transport->inet->sk_shutdown == 0)
2033 return;
2034 dprintk("RPC: %s: TCP_CLOSEd and sk_shutdown set to %d\n",
2035 __func__, transport->inet->sk_shutdown);
2036 }
2037 if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT)) {
2038 /* we don't need to abort the connection if the socket
2039 * hasn't undergone a shutdown
2040 */
2041 if (transport->inet->sk_shutdown == 0)
2042 return;
2043 dprintk("RPC: %s: ESTABLISHED/SYN_SENT "
2044 "sk_shutdown set to %d\n",
2045 __func__, transport->inet->sk_shutdown);
2046 }
2047 xs_abort_connection(transport);
2048}
2049
2050static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2051{
2052 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2053 int ret = -ENOTCONN;
2054
2055 if (!transport->inet) {
2056 struct sock *sk = sock->sk;
2057
2058 write_lock_bh(&sk->sk_callback_lock);
2059
2060 xs_save_old_callbacks(transport, sk);
2061
2062 sk->sk_user_data = xprt;
2063 sk->sk_data_ready = xs_tcp_data_ready;
2064 sk->sk_state_change = xs_tcp_state_change;
2065 sk->sk_write_space = xs_tcp_write_space;
2066 sk->sk_error_report = xs_error_report;
2067 sk->sk_allocation = GFP_ATOMIC;
2068
2069 /* socket options */
2070 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
2071 sock_reset_flag(sk, SOCK_LINGER);
2072 tcp_sk(sk)->linger2 = 0;
2073 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2074
2075 xprt_clear_connected(xprt);
2076
2077 /* Reset to new socket */
2078 transport->sock = sock;
2079 transport->inet = sk;
2080
2081 write_unlock_bh(&sk->sk_callback_lock);
2082 }
2083
2084 if (!xprt_bound(xprt))
2085 goto out;
2086
2087 /* Tell the socket layer to start connecting... */
2088 xprt->stat.connect_count++;
2089 xprt->stat.connect_start = jiffies;
2090 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2091 switch (ret) {
2092 case 0:
2093 case -EINPROGRESS:
2094 /* SYN_SENT! */
2095 xprt->connect_cookie++;
2096 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2097 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2098 }
2099out:
2100 return ret;
2101}
2102
2103/**
2104 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2105 * @xprt: RPC transport to connect
2106 * @transport: socket transport to connect
2107 * @create_sock: function to create a socket of the correct type
2108 *
2109 * Invoked by a work queue tasklet.
2110 */
2111static void xs_tcp_setup_socket(struct work_struct *work)
2112{
2113 struct sock_xprt *transport =
2114 container_of(work, struct sock_xprt, connect_worker.work);
2115 struct socket *sock = transport->sock;
2116 struct rpc_xprt *xprt = &transport->xprt;
2117 int status = -EIO;
2118
2119 if (xprt->shutdown)
2120 goto out;
2121
2122 current->flags |= PF_FSTRANS;
2123
2124 if (!sock) {
2125 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
2126 sock = xs_create_sock(xprt, transport,
2127 xs_addr(xprt)->sa_family, SOCK_STREAM, IPPROTO_TCP);
2128 if (IS_ERR(sock)) {
2129 status = PTR_ERR(sock);
2130 goto out;
2131 }
2132 } else {
2133 int abort_and_exit;
2134
2135 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
2136 &xprt->state);
2137 /* "close" the socket, preserving the local port */
2138 xs_tcp_reuse_connection(transport);
2139
2140 if (abort_and_exit)
2141 goto out_eagain;
2142 }
2143
2144 dprintk("RPC: worker connecting xprt %p via %s to "
2145 "%s (port %s)\n", xprt,
2146 xprt->address_strings[RPC_DISPLAY_PROTO],
2147 xprt->address_strings[RPC_DISPLAY_ADDR],
2148 xprt->address_strings[RPC_DISPLAY_PORT]);
2149
2150 status = xs_tcp_finish_connecting(xprt, sock);
2151 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2152 xprt, -status, xprt_connected(xprt),
2153 sock->sk->sk_state);
2154 switch (status) {
2155 default:
2156 printk("%s: connect returned unhandled error %d\n",
2157 __func__, status);
2158 case -EADDRNOTAVAIL:
2159 /* We're probably in TIME_WAIT. Get rid of existing socket,
2160 * and retry
2161 */
2162 set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
2163 xprt_force_disconnect(xprt);
2164 break;
2165 case -ECONNREFUSED:
2166 case -ECONNRESET:
2167 case -ENETUNREACH:
2168 /* retry with existing socket, after a delay */
2169 case 0:
2170 case -EINPROGRESS:
2171 case -EALREADY:
2172 xprt_clear_connecting(xprt);
2173 current->flags &= ~PF_FSTRANS;
2174 return;
2175 case -EINVAL:
2176 /* Happens, for instance, if the user specified a link
2177 * local IPv6 address without a scope-id.
2178 */
2179 goto out;
2180 }
2181out_eagain:
2182 status = -EAGAIN;
2183out:
2184 xprt_clear_connecting(xprt);
2185 xprt_wake_pending_tasks(xprt, status);
2186 current->flags &= ~PF_FSTRANS;
2187}
2188
2189/**
2190 * xs_connect - connect a socket to a remote endpoint
2191 * @task: address of RPC task that manages state of connect request
2192 *
2193 * TCP: If the remote end dropped the connection, delay reconnecting.
2194 *
2195 * UDP socket connects are synchronous, but we use a work queue anyway
2196 * to guarantee that even unprivileged user processes can set up a
2197 * socket on a privileged port.
2198 *
2199 * If a UDP socket connect fails, the delay behavior here prevents
2200 * retry floods (hard mounts).
2201 */
2202static void xs_connect(struct rpc_task *task)
2203{
2204 struct rpc_xprt *xprt = task->tk_xprt;
2205 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2206
2207 if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
2208 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2209 "seconds\n",
2210 xprt, xprt->reestablish_timeout / HZ);
2211 queue_delayed_work(rpciod_workqueue,
2212 &transport->connect_worker,
2213 xprt->reestablish_timeout);
2214 xprt->reestablish_timeout <<= 1;
2215 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2216 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2217 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2218 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2219 } else {
2220 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2221 queue_delayed_work(rpciod_workqueue,
2222 &transport->connect_worker, 0);
2223 }
2224}
2225
2226/**
2227 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2228 * @xprt: rpc_xprt struct containing statistics
2229 * @seq: output file
2230 *
2231 */
2232static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2233{
2234 long idle_time = 0;
2235
2236 if (xprt_connected(xprt))
2237 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2238
2239 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2240 "%llu %llu %lu %llu %llu\n",
2241 xprt->stat.bind_count,
2242 xprt->stat.connect_count,
2243 xprt->stat.connect_time,
2244 idle_time,
2245 xprt->stat.sends,
2246 xprt->stat.recvs,
2247 xprt->stat.bad_xids,
2248 xprt->stat.req_u,
2249 xprt->stat.bklog_u,
2250 xprt->stat.max_slots,
2251 xprt->stat.sending_u,
2252 xprt->stat.pending_u);
2253}
2254
2255/**
2256 * xs_udp_print_stats - display UDP socket-specifc stats
2257 * @xprt: rpc_xprt struct containing statistics
2258 * @seq: output file
2259 *
2260 */
2261static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2262{
2263 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2264
2265 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2266 "%lu %llu %llu\n",
2267 transport->srcport,
2268 xprt->stat.bind_count,
2269 xprt->stat.sends,
2270 xprt->stat.recvs,
2271 xprt->stat.bad_xids,
2272 xprt->stat.req_u,
2273 xprt->stat.bklog_u,
2274 xprt->stat.max_slots,
2275 xprt->stat.sending_u,
2276 xprt->stat.pending_u);
2277}
2278
2279/**
2280 * xs_tcp_print_stats - display TCP socket-specifc stats
2281 * @xprt: rpc_xprt struct containing statistics
2282 * @seq: output file
2283 *
2284 */
2285static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2286{
2287 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2288 long idle_time = 0;
2289
2290 if (xprt_connected(xprt))
2291 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2292
2293 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2294 "%llu %llu %lu %llu %llu\n",
2295 transport->srcport,
2296 xprt->stat.bind_count,
2297 xprt->stat.connect_count,
2298 xprt->stat.connect_time,
2299 idle_time,
2300 xprt->stat.sends,
2301 xprt->stat.recvs,
2302 xprt->stat.bad_xids,
2303 xprt->stat.req_u,
2304 xprt->stat.bklog_u,
2305 xprt->stat.max_slots,
2306 xprt->stat.sending_u,
2307 xprt->stat.pending_u);
2308}
2309
2310/*
2311 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2312 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2313 * to use the server side send routines.
2314 */
2315static void *bc_malloc(struct rpc_task *task, size_t size)
2316{
2317 struct page *page;
2318 struct rpc_buffer *buf;
2319
2320 BUG_ON(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2321 page = alloc_page(GFP_KERNEL);
2322
2323 if (!page)
2324 return NULL;
2325
2326 buf = page_address(page);
2327 buf->len = PAGE_SIZE;
2328
2329 return buf->data;
2330}
2331
2332/*
2333 * Free the space allocated in the bc_alloc routine
2334 */
2335static void bc_free(void *buffer)
2336{
2337 struct rpc_buffer *buf;
2338
2339 if (!buffer)
2340 return;
2341
2342 buf = container_of(buffer, struct rpc_buffer, data);
2343 free_page((unsigned long)buf);
2344}
2345
2346/*
2347 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2348 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2349 */
2350static int bc_sendto(struct rpc_rqst *req)
2351{
2352 int len;
2353 struct xdr_buf *xbufp = &req->rq_snd_buf;
2354 struct rpc_xprt *xprt = req->rq_xprt;
2355 struct sock_xprt *transport =
2356 container_of(xprt, struct sock_xprt, xprt);
2357 struct socket *sock = transport->sock;
2358 unsigned long headoff;
2359 unsigned long tailoff;
2360
2361 xs_encode_stream_record_marker(xbufp);
2362
2363 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2364 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2365 len = svc_send_common(sock, xbufp,
2366 virt_to_page(xbufp->head[0].iov_base), headoff,
2367 xbufp->tail[0].iov_base, tailoff);
2368
2369 if (len != xbufp->len) {
2370 printk(KERN_NOTICE "Error sending entire callback!\n");
2371 len = -EAGAIN;
2372 }
2373
2374 return len;
2375}
2376
2377/*
2378 * The send routine. Borrows from svc_send
2379 */
2380static int bc_send_request(struct rpc_task *task)
2381{
2382 struct rpc_rqst *req = task->tk_rqstp;
2383 struct svc_xprt *xprt;
2384 struct svc_sock *svsk;
2385 u32 len;
2386
2387 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2388 /*
2389 * Get the server socket associated with this callback xprt
2390 */
2391 xprt = req->rq_xprt->bc_xprt;
2392 svsk = container_of(xprt, struct svc_sock, sk_xprt);
2393
2394 /*
2395 * Grab the mutex to serialize data as the connection is shared
2396 * with the fore channel
2397 */
2398 if (!mutex_trylock(&xprt->xpt_mutex)) {
2399 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2400 if (!mutex_trylock(&xprt->xpt_mutex))
2401 return -EAGAIN;
2402 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2403 }
2404 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2405 len = -ENOTCONN;
2406 else
2407 len = bc_sendto(req);
2408 mutex_unlock(&xprt->xpt_mutex);
2409
2410 if (len > 0)
2411 len = 0;
2412
2413 return len;
2414}
2415
2416/*
2417 * The close routine. Since this is client initiated, we do nothing
2418 */
2419
2420static void bc_close(struct rpc_xprt *xprt)
2421{
2422}
2423
2424/*
2425 * The xprt destroy routine. Again, because this connection is client
2426 * initiated, we do nothing
2427 */
2428
2429static void bc_destroy(struct rpc_xprt *xprt)
2430{
2431}
2432
2433static struct rpc_xprt_ops xs_local_ops = {
2434 .reserve_xprt = xprt_reserve_xprt,
2435 .release_xprt = xs_tcp_release_xprt,
2436 .rpcbind = xs_local_rpcbind,
2437 .set_port = xs_local_set_port,
2438 .connect = xs_connect,
2439 .buf_alloc = rpc_malloc,
2440 .buf_free = rpc_free,
2441 .send_request = xs_local_send_request,
2442 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2443 .close = xs_close,
2444 .destroy = xs_destroy,
2445 .print_stats = xs_local_print_stats,
2446};
2447
2448static struct rpc_xprt_ops xs_udp_ops = {
2449 .set_buffer_size = xs_udp_set_buffer_size,
2450 .reserve_xprt = xprt_reserve_xprt_cong,
2451 .release_xprt = xprt_release_xprt_cong,
2452 .rpcbind = rpcb_getport_async,
2453 .set_port = xs_set_port,
2454 .connect = xs_connect,
2455 .buf_alloc = rpc_malloc,
2456 .buf_free = rpc_free,
2457 .send_request = xs_udp_send_request,
2458 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
2459 .timer = xs_udp_timer,
2460 .release_request = xprt_release_rqst_cong,
2461 .close = xs_close,
2462 .destroy = xs_destroy,
2463 .print_stats = xs_udp_print_stats,
2464};
2465
2466static struct rpc_xprt_ops xs_tcp_ops = {
2467 .reserve_xprt = xprt_reserve_xprt,
2468 .release_xprt = xs_tcp_release_xprt,
2469 .rpcbind = rpcb_getport_async,
2470 .set_port = xs_set_port,
2471 .connect = xs_connect,
2472 .buf_alloc = rpc_malloc,
2473 .buf_free = rpc_free,
2474 .send_request = xs_tcp_send_request,
2475 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2476 .close = xs_tcp_close,
2477 .destroy = xs_destroy,
2478 .print_stats = xs_tcp_print_stats,
2479};
2480
2481/*
2482 * The rpc_xprt_ops for the server backchannel
2483 */
2484
2485static struct rpc_xprt_ops bc_tcp_ops = {
2486 .reserve_xprt = xprt_reserve_xprt,
2487 .release_xprt = xprt_release_xprt,
2488 .rpcbind = xs_local_rpcbind,
2489 .buf_alloc = bc_malloc,
2490 .buf_free = bc_free,
2491 .send_request = bc_send_request,
2492 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2493 .close = bc_close,
2494 .destroy = bc_destroy,
2495 .print_stats = xs_tcp_print_stats,
2496};
2497
2498static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2499{
2500 static const struct sockaddr_in sin = {
2501 .sin_family = AF_INET,
2502 .sin_addr.s_addr = htonl(INADDR_ANY),
2503 };
2504 static const struct sockaddr_in6 sin6 = {
2505 .sin6_family = AF_INET6,
2506 .sin6_addr = IN6ADDR_ANY_INIT,
2507 };
2508
2509 switch (family) {
2510 case AF_LOCAL:
2511 break;
2512 case AF_INET:
2513 memcpy(sap, &sin, sizeof(sin));
2514 break;
2515 case AF_INET6:
2516 memcpy(sap, &sin6, sizeof(sin6));
2517 break;
2518 default:
2519 dprintk("RPC: %s: Bad address family\n", __func__);
2520 return -EAFNOSUPPORT;
2521 }
2522 return 0;
2523}
2524
2525static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2526 unsigned int slot_table_size,
2527 unsigned int max_slot_table_size)
2528{
2529 struct rpc_xprt *xprt;
2530 struct sock_xprt *new;
2531
2532 if (args->addrlen > sizeof(xprt->addr)) {
2533 dprintk("RPC: xs_setup_xprt: address too large\n");
2534 return ERR_PTR(-EBADF);
2535 }
2536
2537 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2538 max_slot_table_size);
2539 if (xprt == NULL) {
2540 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2541 "rpc_xprt\n");
2542 return ERR_PTR(-ENOMEM);
2543 }
2544
2545 new = container_of(xprt, struct sock_xprt, xprt);
2546 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2547 xprt->addrlen = args->addrlen;
2548 if (args->srcaddr)
2549 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2550 else {
2551 int err;
2552 err = xs_init_anyaddr(args->dstaddr->sa_family,
2553 (struct sockaddr *)&new->srcaddr);
2554 if (err != 0) {
2555 xprt_free(xprt);
2556 return ERR_PTR(err);
2557 }
2558 }
2559
2560 return xprt;
2561}
2562
2563static const struct rpc_timeout xs_local_default_timeout = {
2564 .to_initval = 10 * HZ,
2565 .to_maxval = 10 * HZ,
2566 .to_retries = 2,
2567};
2568
2569/**
2570 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2571 * @args: rpc transport creation arguments
2572 *
2573 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2574 */
2575static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2576{
2577 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2578 struct sock_xprt *transport;
2579 struct rpc_xprt *xprt;
2580 struct rpc_xprt *ret;
2581
2582 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2583 xprt_max_tcp_slot_table_entries);
2584 if (IS_ERR(xprt))
2585 return xprt;
2586 transport = container_of(xprt, struct sock_xprt, xprt);
2587
2588 xprt->prot = 0;
2589 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2590 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2591
2592 xprt->bind_timeout = XS_BIND_TO;
2593 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2594 xprt->idle_timeout = XS_IDLE_DISC_TO;
2595
2596 xprt->ops = &xs_local_ops;
2597 xprt->timeout = &xs_local_default_timeout;
2598
2599 switch (sun->sun_family) {
2600 case AF_LOCAL:
2601 if (sun->sun_path[0] != '/') {
2602 dprintk("RPC: bad AF_LOCAL address: %s\n",
2603 sun->sun_path);
2604 ret = ERR_PTR(-EINVAL);
2605 goto out_err;
2606 }
2607 xprt_set_bound(xprt);
2608 INIT_DELAYED_WORK(&transport->connect_worker,
2609 xs_local_setup_socket);
2610 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2611 break;
2612 default:
2613 ret = ERR_PTR(-EAFNOSUPPORT);
2614 goto out_err;
2615 }
2616
2617 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2618 xprt->address_strings[RPC_DISPLAY_ADDR]);
2619
2620 if (try_module_get(THIS_MODULE))
2621 return xprt;
2622 ret = ERR_PTR(-EINVAL);
2623out_err:
2624 xprt_free(xprt);
2625 return ret;
2626}
2627
2628static const struct rpc_timeout xs_udp_default_timeout = {
2629 .to_initval = 5 * HZ,
2630 .to_maxval = 30 * HZ,
2631 .to_increment = 5 * HZ,
2632 .to_retries = 5,
2633};
2634
2635/**
2636 * xs_setup_udp - Set up transport to use a UDP socket
2637 * @args: rpc transport creation arguments
2638 *
2639 */
2640static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2641{
2642 struct sockaddr *addr = args->dstaddr;
2643 struct rpc_xprt *xprt;
2644 struct sock_xprt *transport;
2645 struct rpc_xprt *ret;
2646
2647 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2648 xprt_udp_slot_table_entries);
2649 if (IS_ERR(xprt))
2650 return xprt;
2651 transport = container_of(xprt, struct sock_xprt, xprt);
2652
2653 xprt->prot = IPPROTO_UDP;
2654 xprt->tsh_size = 0;
2655 /* XXX: header size can vary due to auth type, IPv6, etc. */
2656 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2657
2658 xprt->bind_timeout = XS_BIND_TO;
2659 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2660 xprt->idle_timeout = XS_IDLE_DISC_TO;
2661
2662 xprt->ops = &xs_udp_ops;
2663
2664 xprt->timeout = &xs_udp_default_timeout;
2665
2666 switch (addr->sa_family) {
2667 case AF_INET:
2668 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2669 xprt_set_bound(xprt);
2670
2671 INIT_DELAYED_WORK(&transport->connect_worker,
2672 xs_udp_setup_socket);
2673 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2674 break;
2675 case AF_INET6:
2676 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2677 xprt_set_bound(xprt);
2678
2679 INIT_DELAYED_WORK(&transport->connect_worker,
2680 xs_udp_setup_socket);
2681 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2682 break;
2683 default:
2684 ret = ERR_PTR(-EAFNOSUPPORT);
2685 goto out_err;
2686 }
2687
2688 if (xprt_bound(xprt))
2689 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2690 xprt->address_strings[RPC_DISPLAY_ADDR],
2691 xprt->address_strings[RPC_DISPLAY_PORT],
2692 xprt->address_strings[RPC_DISPLAY_PROTO]);
2693 else
2694 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2695 xprt->address_strings[RPC_DISPLAY_ADDR],
2696 xprt->address_strings[RPC_DISPLAY_PROTO]);
2697
2698 if (try_module_get(THIS_MODULE))
2699 return xprt;
2700 ret = ERR_PTR(-EINVAL);
2701out_err:
2702 xprt_free(xprt);
2703 return ret;
2704}
2705
2706static const struct rpc_timeout xs_tcp_default_timeout = {
2707 .to_initval = 60 * HZ,
2708 .to_maxval = 60 * HZ,
2709 .to_retries = 2,
2710};
2711
2712/**
2713 * xs_setup_tcp - Set up transport to use a TCP socket
2714 * @args: rpc transport creation arguments
2715 *
2716 */
2717static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2718{
2719 struct sockaddr *addr = args->dstaddr;
2720 struct rpc_xprt *xprt;
2721 struct sock_xprt *transport;
2722 struct rpc_xprt *ret;
2723
2724 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2725 xprt_max_tcp_slot_table_entries);
2726 if (IS_ERR(xprt))
2727 return xprt;
2728 transport = container_of(xprt, struct sock_xprt, xprt);
2729
2730 xprt->prot = IPPROTO_TCP;
2731 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2732 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2733
2734 xprt->bind_timeout = XS_BIND_TO;
2735 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2736 xprt->idle_timeout = XS_IDLE_DISC_TO;
2737
2738 xprt->ops = &xs_tcp_ops;
2739 xprt->timeout = &xs_tcp_default_timeout;
2740
2741 switch (addr->sa_family) {
2742 case AF_INET:
2743 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2744 xprt_set_bound(xprt);
2745
2746 INIT_DELAYED_WORK(&transport->connect_worker,
2747 xs_tcp_setup_socket);
2748 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2749 break;
2750 case AF_INET6:
2751 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2752 xprt_set_bound(xprt);
2753
2754 INIT_DELAYED_WORK(&transport->connect_worker,
2755 xs_tcp_setup_socket);
2756 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2757 break;
2758 default:
2759 ret = ERR_PTR(-EAFNOSUPPORT);
2760 goto out_err;
2761 }
2762
2763 if (xprt_bound(xprt))
2764 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2765 xprt->address_strings[RPC_DISPLAY_ADDR],
2766 xprt->address_strings[RPC_DISPLAY_PORT],
2767 xprt->address_strings[RPC_DISPLAY_PROTO]);
2768 else
2769 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2770 xprt->address_strings[RPC_DISPLAY_ADDR],
2771 xprt->address_strings[RPC_DISPLAY_PROTO]);
2772
2773
2774 if (try_module_get(THIS_MODULE))
2775 return xprt;
2776 ret = ERR_PTR(-EINVAL);
2777out_err:
2778 xprt_free(xprt);
2779 return ret;
2780}
2781
2782/**
2783 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2784 * @args: rpc transport creation arguments
2785 *
2786 */
2787static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2788{
2789 struct sockaddr *addr = args->dstaddr;
2790 struct rpc_xprt *xprt;
2791 struct sock_xprt *transport;
2792 struct svc_sock *bc_sock;
2793 struct rpc_xprt *ret;
2794
2795 if (args->bc_xprt->xpt_bc_xprt) {
2796 /*
2797 * This server connection already has a backchannel
2798 * export; we can't create a new one, as we wouldn't be
2799 * able to match replies based on xid any more. So,
2800 * reuse the already-existing one:
2801 */
2802 return args->bc_xprt->xpt_bc_xprt;
2803 }
2804 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2805 xprt_tcp_slot_table_entries);
2806 if (IS_ERR(xprt))
2807 return xprt;
2808 transport = container_of(xprt, struct sock_xprt, xprt);
2809
2810 xprt->prot = IPPROTO_TCP;
2811 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2812 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2813 xprt->timeout = &xs_tcp_default_timeout;
2814
2815 /* backchannel */
2816 xprt_set_bound(xprt);
2817 xprt->bind_timeout = 0;
2818 xprt->reestablish_timeout = 0;
2819 xprt->idle_timeout = 0;
2820
2821 xprt->ops = &bc_tcp_ops;
2822
2823 switch (addr->sa_family) {
2824 case AF_INET:
2825 xs_format_peer_addresses(xprt, "tcp",
2826 RPCBIND_NETID_TCP);
2827 break;
2828 case AF_INET6:
2829 xs_format_peer_addresses(xprt, "tcp",
2830 RPCBIND_NETID_TCP6);
2831 break;
2832 default:
2833 ret = ERR_PTR(-EAFNOSUPPORT);
2834 goto out_err;
2835 }
2836
2837 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2838 xprt->address_strings[RPC_DISPLAY_ADDR],
2839 xprt->address_strings[RPC_DISPLAY_PORT],
2840 xprt->address_strings[RPC_DISPLAY_PROTO]);
2841
2842 /*
2843 * Once we've associated a backchannel xprt with a connection,
2844 * we want to keep it around as long as long as the connection
2845 * lasts, in case we need to start using it for a backchannel
2846 * again; this reference won't be dropped until bc_xprt is
2847 * destroyed.
2848 */
2849 xprt_get(xprt);
2850 args->bc_xprt->xpt_bc_xprt = xprt;
2851 xprt->bc_xprt = args->bc_xprt;
2852 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2853 transport->sock = bc_sock->sk_sock;
2854 transport->inet = bc_sock->sk_sk;
2855
2856 /*
2857 * Since we don't want connections for the backchannel, we set
2858 * the xprt status to connected
2859 */
2860 xprt_set_connected(xprt);
2861
2862
2863 if (try_module_get(THIS_MODULE))
2864 return xprt;
2865 xprt_put(xprt);
2866 ret = ERR_PTR(-EINVAL);
2867out_err:
2868 xprt_free(xprt);
2869 return ret;
2870}
2871
2872static struct xprt_class xs_local_transport = {
2873 .list = LIST_HEAD_INIT(xs_local_transport.list),
2874 .name = "named UNIX socket",
2875 .owner = THIS_MODULE,
2876 .ident = XPRT_TRANSPORT_LOCAL,
2877 .setup = xs_setup_local,
2878};
2879
2880static struct xprt_class xs_udp_transport = {
2881 .list = LIST_HEAD_INIT(xs_udp_transport.list),
2882 .name = "udp",
2883 .owner = THIS_MODULE,
2884 .ident = XPRT_TRANSPORT_UDP,
2885 .setup = xs_setup_udp,
2886};
2887
2888static struct xprt_class xs_tcp_transport = {
2889 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
2890 .name = "tcp",
2891 .owner = THIS_MODULE,
2892 .ident = XPRT_TRANSPORT_TCP,
2893 .setup = xs_setup_tcp,
2894};
2895
2896static struct xprt_class xs_bc_tcp_transport = {
2897 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
2898 .name = "tcp NFSv4.1 backchannel",
2899 .owner = THIS_MODULE,
2900 .ident = XPRT_TRANSPORT_BC_TCP,
2901 .setup = xs_setup_bc_tcp,
2902};
2903
2904/**
2905 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2906 *
2907 */
2908int init_socket_xprt(void)
2909{
2910#ifdef RPC_DEBUG
2911 if (!sunrpc_table_header)
2912 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2913#endif
2914
2915 xprt_register_transport(&xs_local_transport);
2916 xprt_register_transport(&xs_udp_transport);
2917 xprt_register_transport(&xs_tcp_transport);
2918 xprt_register_transport(&xs_bc_tcp_transport);
2919
2920 return 0;
2921}
2922
2923/**
2924 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2925 *
2926 */
2927void cleanup_socket_xprt(void)
2928{
2929#ifdef RPC_DEBUG
2930 if (sunrpc_table_header) {
2931 unregister_sysctl_table(sunrpc_table_header);
2932 sunrpc_table_header = NULL;
2933 }
2934#endif
2935
2936 xprt_unregister_transport(&xs_local_transport);
2937 xprt_unregister_transport(&xs_udp_transport);
2938 xprt_unregister_transport(&xs_tcp_transport);
2939 xprt_unregister_transport(&xs_bc_tcp_transport);
2940}
2941
2942static int param_set_uint_minmax(const char *val,
2943 const struct kernel_param *kp,
2944 unsigned int min, unsigned int max)
2945{
2946 unsigned long num;
2947 int ret;
2948
2949 if (!val)
2950 return -EINVAL;
2951 ret = strict_strtoul(val, 0, &num);
2952 if (ret == -EINVAL || num < min || num > max)
2953 return -EINVAL;
2954 *((unsigned int *)kp->arg) = num;
2955 return 0;
2956}
2957
2958static int param_set_portnr(const char *val, const struct kernel_param *kp)
2959{
2960 return param_set_uint_minmax(val, kp,
2961 RPC_MIN_RESVPORT,
2962 RPC_MAX_RESVPORT);
2963}
2964
2965static struct kernel_param_ops param_ops_portnr = {
2966 .set = param_set_portnr,
2967 .get = param_get_uint,
2968};
2969
2970#define param_check_portnr(name, p) \
2971 __param_check(name, p, unsigned int);
2972
2973module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
2974module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
2975
2976static int param_set_slot_table_size(const char *val,
2977 const struct kernel_param *kp)
2978{
2979 return param_set_uint_minmax(val, kp,
2980 RPC_MIN_SLOT_TABLE,
2981 RPC_MAX_SLOT_TABLE);
2982}
2983
2984static struct kernel_param_ops param_ops_slot_table_size = {
2985 .set = param_set_slot_table_size,
2986 .get = param_get_uint,
2987};
2988
2989#define param_check_slot_table_size(name, p) \
2990 __param_check(name, p, unsigned int);
2991
2992static int param_set_max_slot_table_size(const char *val,
2993 const struct kernel_param *kp)
2994{
2995 return param_set_uint_minmax(val, kp,
2996 RPC_MIN_SLOT_TABLE,
2997 RPC_MAX_SLOT_TABLE_LIMIT);
2998}
2999
3000static struct kernel_param_ops param_ops_max_slot_table_size = {
3001 .set = param_set_max_slot_table_size,
3002 .get = param_get_uint,
3003};
3004
3005#define param_check_max_slot_table_size(name, p) \
3006 __param_check(name, p, unsigned int);
3007
3008module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3009 slot_table_size, 0644);
3010module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3011 max_slot_table_size, 0644);
3012module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3013 slot_table_size, 0644);
3014