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_set_srcport(struct sock_xprt *transport, struct socket *sock);
62static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
63 struct socket *sock);
64
65/*
66 * xprtsock tunables
67 */
68static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
69static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
70static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
71
72static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
73static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
74
75#define XS_TCP_LINGER_TO (15U * HZ)
76static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
77
78/*
79 * We can register our own files under /proc/sys/sunrpc by
80 * calling register_sysctl_table() again. The files in that
81 * directory become the union of all files registered there.
82 *
83 * We simply need to make sure that we don't collide with
84 * someone else's file names!
85 */
86
87static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
88static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
89static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
90static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
91static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
92
93static struct ctl_table_header *sunrpc_table_header;
94
95static struct xprt_class xs_local_transport;
96static struct xprt_class xs_udp_transport;
97static struct xprt_class xs_tcp_transport;
98static struct xprt_class xs_bc_tcp_transport;
99
100/*
101 * FIXME: changing the UDP slot table size should also resize the UDP
102 * socket buffers for existing UDP transports
103 */
104static struct ctl_table xs_tunables_table[] = {
105 {
106 .procname = "udp_slot_table_entries",
107 .data = &xprt_udp_slot_table_entries,
108 .maxlen = sizeof(unsigned int),
109 .mode = 0644,
110 .proc_handler = proc_dointvec_minmax,
111 .extra1 = &min_slot_table_size,
112 .extra2 = &max_slot_table_size
113 },
114 {
115 .procname = "tcp_slot_table_entries",
116 .data = &xprt_tcp_slot_table_entries,
117 .maxlen = sizeof(unsigned int),
118 .mode = 0644,
119 .proc_handler = proc_dointvec_minmax,
120 .extra1 = &min_slot_table_size,
121 .extra2 = &max_slot_table_size
122 },
123 {
124 .procname = "tcp_max_slot_table_entries",
125 .data = &xprt_max_tcp_slot_table_entries,
126 .maxlen = sizeof(unsigned int),
127 .mode = 0644,
128 .proc_handler = proc_dointvec_minmax,
129 .extra1 = &min_slot_table_size,
130 .extra2 = &max_tcp_slot_table_limit
131 },
132 {
133 .procname = "min_resvport",
134 .data = &xprt_min_resvport,
135 .maxlen = sizeof(unsigned int),
136 .mode = 0644,
137 .proc_handler = proc_dointvec_minmax,
138 .extra1 = &xprt_min_resvport_limit,
139 .extra2 = &xprt_max_resvport_limit
140 },
141 {
142 .procname = "max_resvport",
143 .data = &xprt_max_resvport,
144 .maxlen = sizeof(unsigned int),
145 .mode = 0644,
146 .proc_handler = proc_dointvec_minmax,
147 .extra1 = &xprt_min_resvport_limit,
148 .extra2 = &xprt_max_resvport_limit
149 },
150 {
151 .procname = "tcp_fin_timeout",
152 .data = &xs_tcp_fin_timeout,
153 .maxlen = sizeof(xs_tcp_fin_timeout),
154 .mode = 0644,
155 .proc_handler = proc_dointvec_jiffies,
156 },
157 { },
158};
159
160static struct ctl_table sunrpc_table[] = {
161 {
162 .procname = "sunrpc",
163 .mode = 0555,
164 .child = xs_tunables_table
165 },
166 { },
167};
168
169/*
170 * Wait duration for a reply from the RPC portmapper.
171 */
172#define XS_BIND_TO (60U * HZ)
173
174/*
175 * Delay if a UDP socket connect error occurs. This is most likely some
176 * kind of resource problem on the local host.
177 */
178#define XS_UDP_REEST_TO (2U * HZ)
179
180/*
181 * The reestablish timeout allows clients to delay for a bit before attempting
182 * to reconnect to a server that just dropped our connection.
183 *
184 * We implement an exponential backoff when trying to reestablish a TCP
185 * transport connection with the server. Some servers like to drop a TCP
186 * connection when they are overworked, so we start with a short timeout and
187 * increase over time if the server is down or not responding.
188 */
189#define XS_TCP_INIT_REEST_TO (3U * HZ)
190
191/*
192 * TCP idle timeout; client drops the transport socket if it is idle
193 * for this long. Note that we also timeout UDP sockets to prevent
194 * holding port numbers when there is no RPC traffic.
195 */
196#define XS_IDLE_DISC_TO (5U * 60 * HZ)
197
198#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
199# undef RPC_DEBUG_DATA
200# define RPCDBG_FACILITY RPCDBG_TRANS
201#endif
202
203#ifdef RPC_DEBUG_DATA
204static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
205{
206 u8 *buf = (u8 *) packet;
207 int j;
208
209 dprintk("RPC: %s\n", msg);
210 for (j = 0; j < count && j < 128; j += 4) {
211 if (!(j & 31)) {
212 if (j)
213 dprintk("\n");
214 dprintk("0x%04x ", j);
215 }
216 dprintk("%02x%02x%02x%02x ",
217 buf[j], buf[j+1], buf[j+2], buf[j+3]);
218 }
219 dprintk("\n");
220}
221#else
222static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
223{
224 /* NOP */
225}
226#endif
227
228static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
229{
230 return (struct rpc_xprt *) sk->sk_user_data;
231}
232
233static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
234{
235 return (struct sockaddr *) &xprt->addr;
236}
237
238static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
239{
240 return (struct sockaddr_un *) &xprt->addr;
241}
242
243static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
244{
245 return (struct sockaddr_in *) &xprt->addr;
246}
247
248static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
249{
250 return (struct sockaddr_in6 *) &xprt->addr;
251}
252
253static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
254{
255 struct sockaddr *sap = xs_addr(xprt);
256 struct sockaddr_in6 *sin6;
257 struct sockaddr_in *sin;
258 struct sockaddr_un *sun;
259 char buf[128];
260
261 switch (sap->sa_family) {
262 case AF_LOCAL:
263 sun = xs_addr_un(xprt);
264 strscpy(buf, sun->sun_path, sizeof(buf));
265 xprt->address_strings[RPC_DISPLAY_ADDR] =
266 kstrdup(buf, GFP_KERNEL);
267 break;
268 case AF_INET:
269 (void)rpc_ntop(sap, buf, sizeof(buf));
270 xprt->address_strings[RPC_DISPLAY_ADDR] =
271 kstrdup(buf, GFP_KERNEL);
272 sin = xs_addr_in(xprt);
273 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
274 break;
275 case AF_INET6:
276 (void)rpc_ntop(sap, buf, sizeof(buf));
277 xprt->address_strings[RPC_DISPLAY_ADDR] =
278 kstrdup(buf, GFP_KERNEL);
279 sin6 = xs_addr_in6(xprt);
280 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
281 break;
282 default:
283 BUG();
284 }
285
286 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
287}
288
289static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
290{
291 struct sockaddr *sap = xs_addr(xprt);
292 char buf[128];
293
294 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
295 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
296
297 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
298 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
299}
300
301static void xs_format_peer_addresses(struct rpc_xprt *xprt,
302 const char *protocol,
303 const char *netid)
304{
305 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
306 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
307 xs_format_common_peer_addresses(xprt);
308 xs_format_common_peer_ports(xprt);
309}
310
311static void xs_update_peer_port(struct rpc_xprt *xprt)
312{
313 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
314 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
315
316 xs_format_common_peer_ports(xprt);
317}
318
319static void xs_free_peer_addresses(struct rpc_xprt *xprt)
320{
321 unsigned int i;
322
323 for (i = 0; i < RPC_DISPLAY_MAX; i++)
324 switch (i) {
325 case RPC_DISPLAY_PROTO:
326 case RPC_DISPLAY_NETID:
327 continue;
328 default:
329 kfree(xprt->address_strings[i]);
330 }
331}
332
333static size_t
334xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp)
335{
336 size_t i,n;
337
338 if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES))
339 return want;
340 n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT;
341 for (i = 0; i < n; i++) {
342 if (buf->pages[i])
343 continue;
344 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp);
345 if (!buf->pages[i]) {
346 i *= PAGE_SIZE;
347 return i > buf->page_base ? i - buf->page_base : 0;
348 }
349 }
350 return want;
351}
352
353static ssize_t
354xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek)
355{
356 ssize_t ret;
357 if (seek != 0)
358 iov_iter_advance(&msg->msg_iter, seek);
359 ret = sock_recvmsg(sock, msg, flags);
360 return ret > 0 ? ret + seek : ret;
361}
362
363static ssize_t
364xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags,
365 struct kvec *kvec, size_t count, size_t seek)
366{
367 iov_iter_kvec(&msg->msg_iter, ITER_DEST, kvec, 1, count);
368 return xs_sock_recvmsg(sock, msg, flags, seek);
369}
370
371static ssize_t
372xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags,
373 struct bio_vec *bvec, unsigned long nr, size_t count,
374 size_t seek)
375{
376 iov_iter_bvec(&msg->msg_iter, ITER_DEST, bvec, nr, count);
377 return xs_sock_recvmsg(sock, msg, flags, seek);
378}
379
380static ssize_t
381xs_read_discard(struct socket *sock, struct msghdr *msg, int flags,
382 size_t count)
383{
384 iov_iter_discard(&msg->msg_iter, ITER_DEST, count);
385 return sock_recvmsg(sock, msg, flags);
386}
387
388#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
389static void
390xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
391{
392 struct bvec_iter bi = {
393 .bi_size = count,
394 };
395 struct bio_vec bv;
396
397 bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
398 for_each_bvec(bv, bvec, bi, bi)
399 flush_dcache_page(bv.bv_page);
400}
401#else
402static inline void
403xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
404{
405}
406#endif
407
408static ssize_t
409xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
410 struct xdr_buf *buf, size_t count, size_t seek, size_t *read)
411{
412 size_t want, seek_init = seek, offset = 0;
413 ssize_t ret;
414
415 want = min_t(size_t, count, buf->head[0].iov_len);
416 if (seek < want) {
417 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek);
418 if (ret <= 0)
419 goto sock_err;
420 offset += ret;
421 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
422 goto out;
423 if (ret != want)
424 goto out;
425 seek = 0;
426 } else {
427 seek -= want;
428 offset += want;
429 }
430
431 want = xs_alloc_sparse_pages(
432 buf, min_t(size_t, count - offset, buf->page_len),
433 GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN);
434 if (seek < want) {
435 ret = xs_read_bvec(sock, msg, flags, buf->bvec,
436 xdr_buf_pagecount(buf),
437 want + buf->page_base,
438 seek + buf->page_base);
439 if (ret <= 0)
440 goto sock_err;
441 xs_flush_bvec(buf->bvec, ret, seek + buf->page_base);
442 ret -= buf->page_base;
443 offset += ret;
444 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
445 goto out;
446 if (ret != want)
447 goto out;
448 seek = 0;
449 } else {
450 seek -= want;
451 offset += want;
452 }
453
454 want = min_t(size_t, count - offset, buf->tail[0].iov_len);
455 if (seek < want) {
456 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek);
457 if (ret <= 0)
458 goto sock_err;
459 offset += ret;
460 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
461 goto out;
462 if (ret != want)
463 goto out;
464 } else if (offset < seek_init)
465 offset = seek_init;
466 ret = -EMSGSIZE;
467out:
468 *read = offset - seek_init;
469 return ret;
470sock_err:
471 offset += seek;
472 goto out;
473}
474
475static void
476xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf)
477{
478 if (!transport->recv.copied) {
479 if (buf->head[0].iov_len >= transport->recv.offset)
480 memcpy(buf->head[0].iov_base,
481 &transport->recv.xid,
482 transport->recv.offset);
483 transport->recv.copied = transport->recv.offset;
484 }
485}
486
487static bool
488xs_read_stream_request_done(struct sock_xprt *transport)
489{
490 return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT);
491}
492
493static void
494xs_read_stream_check_eor(struct sock_xprt *transport,
495 struct msghdr *msg)
496{
497 if (xs_read_stream_request_done(transport))
498 msg->msg_flags |= MSG_EOR;
499}
500
501static ssize_t
502xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg,
503 int flags, struct rpc_rqst *req)
504{
505 struct xdr_buf *buf = &req->rq_private_buf;
506 size_t want, read;
507 ssize_t ret;
508
509 xs_read_header(transport, buf);
510
511 want = transport->recv.len - transport->recv.offset;
512 if (want != 0) {
513 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf,
514 transport->recv.copied + want,
515 transport->recv.copied,
516 &read);
517 transport->recv.offset += read;
518 transport->recv.copied += read;
519 }
520
521 if (transport->recv.offset == transport->recv.len)
522 xs_read_stream_check_eor(transport, msg);
523
524 if (want == 0)
525 return 0;
526
527 switch (ret) {
528 default:
529 break;
530 case -EFAULT:
531 case -EMSGSIZE:
532 msg->msg_flags |= MSG_TRUNC;
533 return read;
534 case 0:
535 return -ESHUTDOWN;
536 }
537 return ret < 0 ? ret : read;
538}
539
540static size_t
541xs_read_stream_headersize(bool isfrag)
542{
543 if (isfrag)
544 return sizeof(__be32);
545 return 3 * sizeof(__be32);
546}
547
548static ssize_t
549xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg,
550 int flags, size_t want, size_t seek)
551{
552 struct kvec kvec = {
553 .iov_base = &transport->recv.fraghdr,
554 .iov_len = want,
555 };
556 return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek);
557}
558
559#if defined(CONFIG_SUNRPC_BACKCHANNEL)
560static ssize_t
561xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
562{
563 struct rpc_xprt *xprt = &transport->xprt;
564 struct rpc_rqst *req;
565 ssize_t ret;
566
567 /* Is this transport associated with the backchannel? */
568 if (!xprt->bc_serv)
569 return -ESHUTDOWN;
570
571 /* Look up and lock the request corresponding to the given XID */
572 req = xprt_lookup_bc_request(xprt, transport->recv.xid);
573 if (!req) {
574 printk(KERN_WARNING "Callback slot table overflowed\n");
575 return -ESHUTDOWN;
576 }
577 if (transport->recv.copied && !req->rq_private_buf.len)
578 return -ESHUTDOWN;
579
580 ret = xs_read_stream_request(transport, msg, flags, req);
581 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
582 xprt_complete_bc_request(req, transport->recv.copied);
583 else
584 req->rq_private_buf.len = transport->recv.copied;
585
586 return ret;
587}
588#else /* CONFIG_SUNRPC_BACKCHANNEL */
589static ssize_t
590xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
591{
592 return -ESHUTDOWN;
593}
594#endif /* CONFIG_SUNRPC_BACKCHANNEL */
595
596static ssize_t
597xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags)
598{
599 struct rpc_xprt *xprt = &transport->xprt;
600 struct rpc_rqst *req;
601 ssize_t ret = 0;
602
603 /* Look up and lock the request corresponding to the given XID */
604 spin_lock(&xprt->queue_lock);
605 req = xprt_lookup_rqst(xprt, transport->recv.xid);
606 if (!req || (transport->recv.copied && !req->rq_private_buf.len)) {
607 msg->msg_flags |= MSG_TRUNC;
608 goto out;
609 }
610 xprt_pin_rqst(req);
611 spin_unlock(&xprt->queue_lock);
612
613 ret = xs_read_stream_request(transport, msg, flags, req);
614
615 spin_lock(&xprt->queue_lock);
616 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
617 xprt_complete_rqst(req->rq_task, transport->recv.copied);
618 else
619 req->rq_private_buf.len = transport->recv.copied;
620 xprt_unpin_rqst(req);
621out:
622 spin_unlock(&xprt->queue_lock);
623 return ret;
624}
625
626static ssize_t
627xs_read_stream(struct sock_xprt *transport, int flags)
628{
629 struct msghdr msg = { 0 };
630 size_t want, read = 0;
631 ssize_t ret = 0;
632
633 if (transport->recv.len == 0) {
634 want = xs_read_stream_headersize(transport->recv.copied != 0);
635 ret = xs_read_stream_header(transport, &msg, flags, want,
636 transport->recv.offset);
637 if (ret <= 0)
638 goto out_err;
639 transport->recv.offset = ret;
640 if (transport->recv.offset != want)
641 return transport->recv.offset;
642 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) &
643 RPC_FRAGMENT_SIZE_MASK;
644 transport->recv.offset -= sizeof(transport->recv.fraghdr);
645 read = ret;
646 }
647
648 switch (be32_to_cpu(transport->recv.calldir)) {
649 default:
650 msg.msg_flags |= MSG_TRUNC;
651 break;
652 case RPC_CALL:
653 ret = xs_read_stream_call(transport, &msg, flags);
654 break;
655 case RPC_REPLY:
656 ret = xs_read_stream_reply(transport, &msg, flags);
657 }
658 if (msg.msg_flags & MSG_TRUNC) {
659 transport->recv.calldir = cpu_to_be32(-1);
660 transport->recv.copied = -1;
661 }
662 if (ret < 0)
663 goto out_err;
664 read += ret;
665 if (transport->recv.offset < transport->recv.len) {
666 if (!(msg.msg_flags & MSG_TRUNC))
667 return read;
668 msg.msg_flags = 0;
669 ret = xs_read_discard(transport->sock, &msg, flags,
670 transport->recv.len - transport->recv.offset);
671 if (ret <= 0)
672 goto out_err;
673 transport->recv.offset += ret;
674 read += ret;
675 if (transport->recv.offset != transport->recv.len)
676 return read;
677 }
678 if (xs_read_stream_request_done(transport)) {
679 trace_xs_stream_read_request(transport);
680 transport->recv.copied = 0;
681 }
682 transport->recv.offset = 0;
683 transport->recv.len = 0;
684 return read;
685out_err:
686 return ret != 0 ? ret : -ESHUTDOWN;
687}
688
689static __poll_t xs_poll_socket(struct sock_xprt *transport)
690{
691 return transport->sock->ops->poll(transport->file, transport->sock,
692 NULL);
693}
694
695static bool xs_poll_socket_readable(struct sock_xprt *transport)
696{
697 __poll_t events = xs_poll_socket(transport);
698
699 return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP);
700}
701
702static void xs_poll_check_readable(struct sock_xprt *transport)
703{
704
705 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
706 if (!xs_poll_socket_readable(transport))
707 return;
708 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
709 queue_work(xprtiod_workqueue, &transport->recv_worker);
710}
711
712static void xs_stream_data_receive(struct sock_xprt *transport)
713{
714 size_t read = 0;
715 ssize_t ret = 0;
716
717 mutex_lock(&transport->recv_mutex);
718 if (transport->sock == NULL)
719 goto out;
720 for (;;) {
721 ret = xs_read_stream(transport, MSG_DONTWAIT);
722 if (ret < 0)
723 break;
724 read += ret;
725 cond_resched();
726 }
727 if (ret == -ESHUTDOWN)
728 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
729 else
730 xs_poll_check_readable(transport);
731out:
732 mutex_unlock(&transport->recv_mutex);
733 trace_xs_stream_read_data(&transport->xprt, ret, read);
734}
735
736static void xs_stream_data_receive_workfn(struct work_struct *work)
737{
738 struct sock_xprt *transport =
739 container_of(work, struct sock_xprt, recv_worker);
740 unsigned int pflags = memalloc_nofs_save();
741
742 xs_stream_data_receive(transport);
743 memalloc_nofs_restore(pflags);
744}
745
746static void
747xs_stream_reset_connect(struct sock_xprt *transport)
748{
749 transport->recv.offset = 0;
750 transport->recv.len = 0;
751 transport->recv.copied = 0;
752 transport->xmit.offset = 0;
753}
754
755static void
756xs_stream_start_connect(struct sock_xprt *transport)
757{
758 transport->xprt.stat.connect_count++;
759 transport->xprt.stat.connect_start = jiffies;
760}
761
762#define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
763
764/**
765 * xs_nospace - handle transmit was incomplete
766 * @req: pointer to RPC request
767 * @transport: pointer to struct sock_xprt
768 *
769 */
770static int xs_nospace(struct rpc_rqst *req, struct sock_xprt *transport)
771{
772 struct rpc_xprt *xprt = &transport->xprt;
773 struct sock *sk = transport->inet;
774 int ret = -EAGAIN;
775
776 trace_rpc_socket_nospace(req, transport);
777
778 /* Protect against races with write_space */
779 spin_lock(&xprt->transport_lock);
780
781 /* Don't race with disconnect */
782 if (xprt_connected(xprt)) {
783 /* wait for more buffer space */
784 set_bit(XPRT_SOCK_NOSPACE, &transport->sock_state);
785 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
786 sk->sk_write_pending++;
787 xprt_wait_for_buffer_space(xprt);
788 } else
789 ret = -ENOTCONN;
790
791 spin_unlock(&xprt->transport_lock);
792 return ret;
793}
794
795static int xs_sock_nospace(struct rpc_rqst *req)
796{
797 struct sock_xprt *transport =
798 container_of(req->rq_xprt, struct sock_xprt, xprt);
799 struct sock *sk = transport->inet;
800 int ret = -EAGAIN;
801
802 lock_sock(sk);
803 if (!sock_writeable(sk))
804 ret = xs_nospace(req, transport);
805 release_sock(sk);
806 return ret;
807}
808
809static int xs_stream_nospace(struct rpc_rqst *req, bool vm_wait)
810{
811 struct sock_xprt *transport =
812 container_of(req->rq_xprt, struct sock_xprt, xprt);
813 struct sock *sk = transport->inet;
814 int ret = -EAGAIN;
815
816 if (vm_wait)
817 return -ENOBUFS;
818 lock_sock(sk);
819 if (!sk_stream_memory_free(sk))
820 ret = xs_nospace(req, transport);
821 release_sock(sk);
822 return ret;
823}
824
825static int xs_stream_prepare_request(struct rpc_rqst *req, struct xdr_buf *buf)
826{
827 return xdr_alloc_bvec(buf, rpc_task_gfp_mask());
828}
829
830/*
831 * Determine if the previous message in the stream was aborted before it
832 * could complete transmission.
833 */
834static bool
835xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req)
836{
837 return transport->xmit.offset != 0 && req->rq_bytes_sent == 0;
838}
839
840/*
841 * Return the stream record marker field for a record of length < 2^31-1
842 */
843static rpc_fraghdr
844xs_stream_record_marker(struct xdr_buf *xdr)
845{
846 if (!xdr->len)
847 return 0;
848 return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len);
849}
850
851/**
852 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
853 * @req: pointer to RPC request
854 *
855 * Return values:
856 * 0: The request has been sent
857 * EAGAIN: The socket was blocked, please call again later to
858 * complete the request
859 * ENOTCONN: Caller needs to invoke connect logic then call again
860 * other: Some other error occurred, the request was not sent
861 */
862static int xs_local_send_request(struct rpc_rqst *req)
863{
864 struct rpc_xprt *xprt = req->rq_xprt;
865 struct sock_xprt *transport =
866 container_of(xprt, struct sock_xprt, xprt);
867 struct xdr_buf *xdr = &req->rq_snd_buf;
868 rpc_fraghdr rm = xs_stream_record_marker(xdr);
869 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
870 struct msghdr msg = {
871 .msg_flags = XS_SENDMSG_FLAGS,
872 };
873 bool vm_wait;
874 unsigned int sent;
875 int status;
876
877 /* Close the stream if the previous transmission was incomplete */
878 if (xs_send_request_was_aborted(transport, req)) {
879 xprt_force_disconnect(xprt);
880 return -ENOTCONN;
881 }
882
883 xs_pktdump("packet data:",
884 req->rq_svec->iov_base, req->rq_svec->iov_len);
885
886 vm_wait = sk_stream_is_writeable(transport->inet) ? true : false;
887
888 req->rq_xtime = ktime_get();
889 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
890 transport->xmit.offset, rm, &sent);
891 dprintk("RPC: %s(%u) = %d\n",
892 __func__, xdr->len - transport->xmit.offset, status);
893
894 if (likely(sent > 0) || status == 0) {
895 transport->xmit.offset += sent;
896 req->rq_bytes_sent = transport->xmit.offset;
897 if (likely(req->rq_bytes_sent >= msglen)) {
898 req->rq_xmit_bytes_sent += transport->xmit.offset;
899 transport->xmit.offset = 0;
900 return 0;
901 }
902 status = -EAGAIN;
903 vm_wait = false;
904 }
905
906 switch (status) {
907 case -EAGAIN:
908 status = xs_stream_nospace(req, vm_wait);
909 break;
910 default:
911 dprintk("RPC: sendmsg returned unrecognized error %d\n",
912 -status);
913 fallthrough;
914 case -EPIPE:
915 xprt_force_disconnect(xprt);
916 status = -ENOTCONN;
917 }
918
919 return status;
920}
921
922/**
923 * xs_udp_send_request - write an RPC request to a UDP socket
924 * @req: pointer to RPC request
925 *
926 * Return values:
927 * 0: The request has been sent
928 * EAGAIN: The socket was blocked, please call again later to
929 * complete the request
930 * ENOTCONN: Caller needs to invoke connect logic then call again
931 * other: Some other error occurred, the request was not sent
932 */
933static int xs_udp_send_request(struct rpc_rqst *req)
934{
935 struct rpc_xprt *xprt = req->rq_xprt;
936 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
937 struct xdr_buf *xdr = &req->rq_snd_buf;
938 struct msghdr msg = {
939 .msg_name = xs_addr(xprt),
940 .msg_namelen = xprt->addrlen,
941 .msg_flags = XS_SENDMSG_FLAGS,
942 };
943 unsigned int sent;
944 int status;
945
946 xs_pktdump("packet data:",
947 req->rq_svec->iov_base,
948 req->rq_svec->iov_len);
949
950 if (!xprt_bound(xprt))
951 return -ENOTCONN;
952
953 if (!xprt_request_get_cong(xprt, req))
954 return -EBADSLT;
955
956 status = xdr_alloc_bvec(xdr, rpc_task_gfp_mask());
957 if (status < 0)
958 return status;
959 req->rq_xtime = ktime_get();
960 status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, 0, &sent);
961
962 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
963 xdr->len, status);
964
965 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
966 if (status == -EPERM)
967 goto process_status;
968
969 if (status == -EAGAIN && sock_writeable(transport->inet))
970 status = -ENOBUFS;
971
972 if (sent > 0 || status == 0) {
973 req->rq_xmit_bytes_sent += sent;
974 if (sent >= req->rq_slen)
975 return 0;
976 /* Still some bytes left; set up for a retry later. */
977 status = -EAGAIN;
978 }
979
980process_status:
981 switch (status) {
982 case -ENOTSOCK:
983 status = -ENOTCONN;
984 /* Should we call xs_close() here? */
985 break;
986 case -EAGAIN:
987 status = xs_sock_nospace(req);
988 break;
989 case -ENETUNREACH:
990 case -ENOBUFS:
991 case -EPIPE:
992 case -ECONNREFUSED:
993 case -EPERM:
994 /* When the server has died, an ICMP port unreachable message
995 * prompts ECONNREFUSED. */
996 break;
997 default:
998 dprintk("RPC: sendmsg returned unrecognized error %d\n",
999 -status);
1000 }
1001
1002 return status;
1003}
1004
1005/**
1006 * xs_tcp_send_request - write an RPC request to a TCP socket
1007 * @req: pointer to RPC request
1008 *
1009 * Return values:
1010 * 0: The request has been sent
1011 * EAGAIN: The socket was blocked, please call again later to
1012 * complete the request
1013 * ENOTCONN: Caller needs to invoke connect logic then call again
1014 * other: Some other error occurred, the request was not sent
1015 *
1016 * XXX: In the case of soft timeouts, should we eventually give up
1017 * if sendmsg is not able to make progress?
1018 */
1019static int xs_tcp_send_request(struct rpc_rqst *req)
1020{
1021 struct rpc_xprt *xprt = req->rq_xprt;
1022 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1023 struct xdr_buf *xdr = &req->rq_snd_buf;
1024 rpc_fraghdr rm = xs_stream_record_marker(xdr);
1025 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
1026 struct msghdr msg = {
1027 .msg_flags = XS_SENDMSG_FLAGS,
1028 };
1029 bool vm_wait;
1030 unsigned int sent;
1031 int status;
1032
1033 /* Close the stream if the previous transmission was incomplete */
1034 if (xs_send_request_was_aborted(transport, req)) {
1035 if (transport->sock != NULL)
1036 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
1037 return -ENOTCONN;
1038 }
1039 if (!transport->inet)
1040 return -ENOTCONN;
1041
1042 xs_pktdump("packet data:",
1043 req->rq_svec->iov_base,
1044 req->rq_svec->iov_len);
1045
1046 if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
1047 xs_tcp_set_socket_timeouts(xprt, transport->sock);
1048
1049 xs_set_srcport(transport, transport->sock);
1050
1051 /* Continue transmitting the packet/record. We must be careful
1052 * to cope with writespace callbacks arriving _after_ we have
1053 * called sendmsg(). */
1054 req->rq_xtime = ktime_get();
1055 tcp_sock_set_cork(transport->inet, true);
1056
1057 vm_wait = sk_stream_is_writeable(transport->inet) ? true : false;
1058
1059 do {
1060 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
1061 transport->xmit.offset, rm, &sent);
1062
1063 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
1064 xdr->len - transport->xmit.offset, status);
1065
1066 /* If we've sent the entire packet, immediately
1067 * reset the count of bytes sent. */
1068 transport->xmit.offset += sent;
1069 req->rq_bytes_sent = transport->xmit.offset;
1070 if (likely(req->rq_bytes_sent >= msglen)) {
1071 req->rq_xmit_bytes_sent += transport->xmit.offset;
1072 transport->xmit.offset = 0;
1073 if (atomic_long_read(&xprt->xmit_queuelen) == 1)
1074 tcp_sock_set_cork(transport->inet, false);
1075 return 0;
1076 }
1077
1078 WARN_ON_ONCE(sent == 0 && status == 0);
1079
1080 if (sent > 0)
1081 vm_wait = false;
1082
1083 } while (status == 0);
1084
1085 switch (status) {
1086 case -ENOTSOCK:
1087 status = -ENOTCONN;
1088 /* Should we call xs_close() here? */
1089 break;
1090 case -EAGAIN:
1091 status = xs_stream_nospace(req, vm_wait);
1092 break;
1093 case -ECONNRESET:
1094 case -ECONNREFUSED:
1095 case -ENOTCONN:
1096 case -EADDRINUSE:
1097 case -ENOBUFS:
1098 case -EPIPE:
1099 break;
1100 default:
1101 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1102 -status);
1103 }
1104
1105 return status;
1106}
1107
1108static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1109{
1110 transport->old_data_ready = sk->sk_data_ready;
1111 transport->old_state_change = sk->sk_state_change;
1112 transport->old_write_space = sk->sk_write_space;
1113 transport->old_error_report = sk->sk_error_report;
1114}
1115
1116static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1117{
1118 sk->sk_data_ready = transport->old_data_ready;
1119 sk->sk_state_change = transport->old_state_change;
1120 sk->sk_write_space = transport->old_write_space;
1121 sk->sk_error_report = transport->old_error_report;
1122}
1123
1124static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
1125{
1126 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1127
1128 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1129 clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state);
1130 clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state);
1131 clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state);
1132 clear_bit(XPRT_SOCK_NOSPACE, &transport->sock_state);
1133}
1134
1135static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr)
1136{
1137 set_bit(nr, &transport->sock_state);
1138 queue_work(xprtiod_workqueue, &transport->error_worker);
1139}
1140
1141static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1142{
1143 xprt->connect_cookie++;
1144 smp_mb__before_atomic();
1145 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1146 clear_bit(XPRT_CLOSING, &xprt->state);
1147 xs_sock_reset_state_flags(xprt);
1148 smp_mb__after_atomic();
1149}
1150
1151/**
1152 * xs_error_report - callback to handle TCP socket state errors
1153 * @sk: socket
1154 *
1155 * Note: we don't call sock_error() since there may be a rpc_task
1156 * using the socket, and so we don't want to clear sk->sk_err.
1157 */
1158static void xs_error_report(struct sock *sk)
1159{
1160 struct sock_xprt *transport;
1161 struct rpc_xprt *xprt;
1162
1163 if (!(xprt = xprt_from_sock(sk)))
1164 return;
1165
1166 transport = container_of(xprt, struct sock_xprt, xprt);
1167 transport->xprt_err = -sk->sk_err;
1168 if (transport->xprt_err == 0)
1169 return;
1170 dprintk("RPC: xs_error_report client %p, error=%d...\n",
1171 xprt, -transport->xprt_err);
1172 trace_rpc_socket_error(xprt, sk->sk_socket, transport->xprt_err);
1173
1174 /* barrier ensures xprt_err is set before XPRT_SOCK_WAKE_ERROR */
1175 smp_mb__before_atomic();
1176 xs_run_error_worker(transport, XPRT_SOCK_WAKE_ERROR);
1177}
1178
1179static void xs_reset_transport(struct sock_xprt *transport)
1180{
1181 struct socket *sock = transport->sock;
1182 struct sock *sk = transport->inet;
1183 struct rpc_xprt *xprt = &transport->xprt;
1184 struct file *filp = transport->file;
1185
1186 if (sk == NULL)
1187 return;
1188 /*
1189 * Make sure we're calling this in a context from which it is safe
1190 * to call __fput_sync(). In practice that means rpciod and the
1191 * system workqueue.
1192 */
1193 if (!(current->flags & PF_WQ_WORKER)) {
1194 WARN_ON_ONCE(1);
1195 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
1196 return;
1197 }
1198
1199 if (atomic_read(&transport->xprt.swapper))
1200 sk_clear_memalloc(sk);
1201
1202 kernel_sock_shutdown(sock, SHUT_RDWR);
1203
1204 mutex_lock(&transport->recv_mutex);
1205 lock_sock(sk);
1206 transport->inet = NULL;
1207 transport->sock = NULL;
1208 transport->file = NULL;
1209
1210 sk->sk_user_data = NULL;
1211
1212 xs_restore_old_callbacks(transport, sk);
1213 xprt_clear_connected(xprt);
1214 xs_sock_reset_connection_flags(xprt);
1215 /* Reset stream record info */
1216 xs_stream_reset_connect(transport);
1217 release_sock(sk);
1218 mutex_unlock(&transport->recv_mutex);
1219
1220 trace_rpc_socket_close(xprt, sock);
1221 __fput_sync(filp);
1222
1223 xprt_disconnect_done(xprt);
1224}
1225
1226/**
1227 * xs_close - close a socket
1228 * @xprt: transport
1229 *
1230 * This is used when all requests are complete; ie, no DRC state remains
1231 * on the server we want to save.
1232 *
1233 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1234 * xs_reset_transport() zeroing the socket from underneath a writer.
1235 */
1236static void xs_close(struct rpc_xprt *xprt)
1237{
1238 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1239
1240 dprintk("RPC: xs_close xprt %p\n", xprt);
1241
1242 xs_reset_transport(transport);
1243 xprt->reestablish_timeout = 0;
1244}
1245
1246static void xs_inject_disconnect(struct rpc_xprt *xprt)
1247{
1248 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
1249 xprt);
1250 xprt_disconnect_done(xprt);
1251}
1252
1253static void xs_xprt_free(struct rpc_xprt *xprt)
1254{
1255 xs_free_peer_addresses(xprt);
1256 xprt_free(xprt);
1257}
1258
1259/**
1260 * xs_destroy - prepare to shutdown a transport
1261 * @xprt: doomed transport
1262 *
1263 */
1264static void xs_destroy(struct rpc_xprt *xprt)
1265{
1266 struct sock_xprt *transport = container_of(xprt,
1267 struct sock_xprt, xprt);
1268 dprintk("RPC: xs_destroy xprt %p\n", xprt);
1269
1270 cancel_delayed_work_sync(&transport->connect_worker);
1271 xs_close(xprt);
1272 cancel_work_sync(&transport->recv_worker);
1273 cancel_work_sync(&transport->error_worker);
1274 xs_xprt_free(xprt);
1275 module_put(THIS_MODULE);
1276}
1277
1278/**
1279 * xs_udp_data_read_skb - receive callback for UDP sockets
1280 * @xprt: transport
1281 * @sk: socket
1282 * @skb: skbuff
1283 *
1284 */
1285static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1286 struct sock *sk,
1287 struct sk_buff *skb)
1288{
1289 struct rpc_task *task;
1290 struct rpc_rqst *rovr;
1291 int repsize, copied;
1292 u32 _xid;
1293 __be32 *xp;
1294
1295 repsize = skb->len;
1296 if (repsize < 4) {
1297 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
1298 return;
1299 }
1300
1301 /* Copy the XID from the skb... */
1302 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1303 if (xp == NULL)
1304 return;
1305
1306 /* Look up and lock the request corresponding to the given XID */
1307 spin_lock(&xprt->queue_lock);
1308 rovr = xprt_lookup_rqst(xprt, *xp);
1309 if (!rovr)
1310 goto out_unlock;
1311 xprt_pin_rqst(rovr);
1312 xprt_update_rtt(rovr->rq_task);
1313 spin_unlock(&xprt->queue_lock);
1314 task = rovr->rq_task;
1315
1316 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1317 copied = repsize;
1318
1319 /* Suck it into the iovec, verify checksum if not done by hw. */
1320 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1321 spin_lock(&xprt->queue_lock);
1322 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1323 goto out_unpin;
1324 }
1325
1326
1327 spin_lock(&xprt->transport_lock);
1328 xprt_adjust_cwnd(xprt, task, copied);
1329 spin_unlock(&xprt->transport_lock);
1330 spin_lock(&xprt->queue_lock);
1331 xprt_complete_rqst(task, copied);
1332 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1333out_unpin:
1334 xprt_unpin_rqst(rovr);
1335 out_unlock:
1336 spin_unlock(&xprt->queue_lock);
1337}
1338
1339static void xs_udp_data_receive(struct sock_xprt *transport)
1340{
1341 struct sk_buff *skb;
1342 struct sock *sk;
1343 int err;
1344
1345 mutex_lock(&transport->recv_mutex);
1346 sk = transport->inet;
1347 if (sk == NULL)
1348 goto out;
1349 for (;;) {
1350 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err);
1351 if (skb == NULL)
1352 break;
1353 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1354 consume_skb(skb);
1355 cond_resched();
1356 }
1357 xs_poll_check_readable(transport);
1358out:
1359 mutex_unlock(&transport->recv_mutex);
1360}
1361
1362static void xs_udp_data_receive_workfn(struct work_struct *work)
1363{
1364 struct sock_xprt *transport =
1365 container_of(work, struct sock_xprt, recv_worker);
1366 unsigned int pflags = memalloc_nofs_save();
1367
1368 xs_udp_data_receive(transport);
1369 memalloc_nofs_restore(pflags);
1370}
1371
1372/**
1373 * xs_data_ready - "data ready" callback for sockets
1374 * @sk: socket with data to read
1375 *
1376 */
1377static void xs_data_ready(struct sock *sk)
1378{
1379 struct rpc_xprt *xprt;
1380
1381 xprt = xprt_from_sock(sk);
1382 if (xprt != NULL) {
1383 struct sock_xprt *transport = container_of(xprt,
1384 struct sock_xprt, xprt);
1385
1386 trace_xs_data_ready(xprt);
1387
1388 transport->old_data_ready(sk);
1389 /* Any data means we had a useful conversation, so
1390 * then we don't need to delay the next reconnect
1391 */
1392 if (xprt->reestablish_timeout)
1393 xprt->reestablish_timeout = 0;
1394 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1395 queue_work(xprtiod_workqueue, &transport->recv_worker);
1396 }
1397}
1398
1399/*
1400 * Helper function to force a TCP close if the server is sending
1401 * junk and/or it has put us in CLOSE_WAIT
1402 */
1403static void xs_tcp_force_close(struct rpc_xprt *xprt)
1404{
1405 xprt_force_disconnect(xprt);
1406}
1407
1408#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1409static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1410{
1411 return PAGE_SIZE;
1412}
1413#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1414
1415/**
1416 * xs_local_state_change - callback to handle AF_LOCAL socket state changes
1417 * @sk: socket whose state has changed
1418 *
1419 */
1420static void xs_local_state_change(struct sock *sk)
1421{
1422 struct rpc_xprt *xprt;
1423 struct sock_xprt *transport;
1424
1425 if (!(xprt = xprt_from_sock(sk)))
1426 return;
1427 transport = container_of(xprt, struct sock_xprt, xprt);
1428 if (sk->sk_shutdown & SHUTDOWN_MASK) {
1429 clear_bit(XPRT_CONNECTED, &xprt->state);
1430 /* Trigger the socket release */
1431 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1432 }
1433}
1434
1435/**
1436 * xs_tcp_state_change - callback to handle TCP socket state changes
1437 * @sk: socket whose state has changed
1438 *
1439 */
1440static void xs_tcp_state_change(struct sock *sk)
1441{
1442 struct rpc_xprt *xprt;
1443 struct sock_xprt *transport;
1444
1445 if (!(xprt = xprt_from_sock(sk)))
1446 return;
1447 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1448 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1449 sk->sk_state, xprt_connected(xprt),
1450 sock_flag(sk, SOCK_DEAD),
1451 sock_flag(sk, SOCK_ZAPPED),
1452 sk->sk_shutdown);
1453
1454 transport = container_of(xprt, struct sock_xprt, xprt);
1455 trace_rpc_socket_state_change(xprt, sk->sk_socket);
1456 switch (sk->sk_state) {
1457 case TCP_ESTABLISHED:
1458 if (!xprt_test_and_set_connected(xprt)) {
1459 xprt->connect_cookie++;
1460 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1461 xprt_clear_connecting(xprt);
1462
1463 xprt->stat.connect_count++;
1464 xprt->stat.connect_time += (long)jiffies -
1465 xprt->stat.connect_start;
1466 xs_run_error_worker(transport, XPRT_SOCK_WAKE_PENDING);
1467 }
1468 break;
1469 case TCP_FIN_WAIT1:
1470 /* The client initiated a shutdown of the socket */
1471 xprt->connect_cookie++;
1472 xprt->reestablish_timeout = 0;
1473 set_bit(XPRT_CLOSING, &xprt->state);
1474 smp_mb__before_atomic();
1475 clear_bit(XPRT_CONNECTED, &xprt->state);
1476 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1477 smp_mb__after_atomic();
1478 break;
1479 case TCP_CLOSE_WAIT:
1480 /* The server initiated a shutdown of the socket */
1481 xprt->connect_cookie++;
1482 clear_bit(XPRT_CONNECTED, &xprt->state);
1483 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1484 fallthrough;
1485 case TCP_CLOSING:
1486 /*
1487 * If the server closed down the connection, make sure that
1488 * we back off before reconnecting
1489 */
1490 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1491 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1492 break;
1493 case TCP_LAST_ACK:
1494 set_bit(XPRT_CLOSING, &xprt->state);
1495 smp_mb__before_atomic();
1496 clear_bit(XPRT_CONNECTED, &xprt->state);
1497 smp_mb__after_atomic();
1498 break;
1499 case TCP_CLOSE:
1500 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1501 &transport->sock_state))
1502 xprt_clear_connecting(xprt);
1503 clear_bit(XPRT_CLOSING, &xprt->state);
1504 /* Trigger the socket release */
1505 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1506 }
1507}
1508
1509static void xs_write_space(struct sock *sk)
1510{
1511 struct sock_xprt *transport;
1512 struct rpc_xprt *xprt;
1513
1514 if (!sk->sk_socket)
1515 return;
1516 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1517
1518 if (unlikely(!(xprt = xprt_from_sock(sk))))
1519 return;
1520 transport = container_of(xprt, struct sock_xprt, xprt);
1521 if (!test_and_clear_bit(XPRT_SOCK_NOSPACE, &transport->sock_state))
1522 return;
1523 xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE);
1524 sk->sk_write_pending--;
1525}
1526
1527/**
1528 * xs_udp_write_space - callback invoked when socket buffer space
1529 * becomes available
1530 * @sk: socket whose state has changed
1531 *
1532 * Called when more output buffer space is available for this socket.
1533 * We try not to wake our writers until they can make "significant"
1534 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1535 * with a bunch of small requests.
1536 */
1537static void xs_udp_write_space(struct sock *sk)
1538{
1539 /* from net/core/sock.c:sock_def_write_space */
1540 if (sock_writeable(sk))
1541 xs_write_space(sk);
1542}
1543
1544/**
1545 * xs_tcp_write_space - callback invoked when socket buffer space
1546 * becomes available
1547 * @sk: socket whose state has changed
1548 *
1549 * Called when more output buffer space is available for this socket.
1550 * We try not to wake our writers until they can make "significant"
1551 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1552 * with a bunch of small requests.
1553 */
1554static void xs_tcp_write_space(struct sock *sk)
1555{
1556 /* from net/core/stream.c:sk_stream_write_space */
1557 if (sk_stream_is_writeable(sk))
1558 xs_write_space(sk);
1559}
1560
1561static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1562{
1563 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1564 struct sock *sk = transport->inet;
1565
1566 if (transport->rcvsize) {
1567 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1568 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1569 }
1570 if (transport->sndsize) {
1571 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1572 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1573 sk->sk_write_space(sk);
1574 }
1575}
1576
1577/**
1578 * xs_udp_set_buffer_size - set send and receive limits
1579 * @xprt: generic transport
1580 * @sndsize: requested size of send buffer, in bytes
1581 * @rcvsize: requested size of receive buffer, in bytes
1582 *
1583 * Set socket send and receive buffer size limits.
1584 */
1585static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1586{
1587 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1588
1589 transport->sndsize = 0;
1590 if (sndsize)
1591 transport->sndsize = sndsize + 1024;
1592 transport->rcvsize = 0;
1593 if (rcvsize)
1594 transport->rcvsize = rcvsize + 1024;
1595
1596 xs_udp_do_set_buffer_size(xprt);
1597}
1598
1599/**
1600 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1601 * @xprt: controlling transport
1602 * @task: task that timed out
1603 *
1604 * Adjust the congestion window after a retransmit timeout has occurred.
1605 */
1606static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1607{
1608 spin_lock(&xprt->transport_lock);
1609 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1610 spin_unlock(&xprt->transport_lock);
1611}
1612
1613static int xs_get_random_port(void)
1614{
1615 unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1616 unsigned short range;
1617 unsigned short rand;
1618
1619 if (max < min)
1620 return -EADDRINUSE;
1621 range = max - min + 1;
1622 rand = get_random_u32_below(range);
1623 return rand + min;
1624}
1625
1626static unsigned short xs_sock_getport(struct socket *sock)
1627{
1628 struct sockaddr_storage buf;
1629 unsigned short port = 0;
1630
1631 if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1632 goto out;
1633 switch (buf.ss_family) {
1634 case AF_INET6:
1635 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1636 break;
1637 case AF_INET:
1638 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1639 }
1640out:
1641 return port;
1642}
1643
1644/**
1645 * xs_set_port - reset the port number in the remote endpoint address
1646 * @xprt: generic transport
1647 * @port: new port number
1648 *
1649 */
1650static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1651{
1652 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1653
1654 rpc_set_port(xs_addr(xprt), port);
1655 xs_update_peer_port(xprt);
1656}
1657
1658static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1659{
1660 if (transport->srcport == 0 && transport->xprt.reuseport)
1661 transport->srcport = xs_sock_getport(sock);
1662}
1663
1664static int xs_get_srcport(struct sock_xprt *transport)
1665{
1666 int port = transport->srcport;
1667
1668 if (port == 0 && transport->xprt.resvport)
1669 port = xs_get_random_port();
1670 return port;
1671}
1672
1673static unsigned short xs_sock_srcport(struct rpc_xprt *xprt)
1674{
1675 struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt);
1676 unsigned short ret = 0;
1677 mutex_lock(&sock->recv_mutex);
1678 if (sock->sock)
1679 ret = xs_sock_getport(sock->sock);
1680 mutex_unlock(&sock->recv_mutex);
1681 return ret;
1682}
1683
1684static int xs_sock_srcaddr(struct rpc_xprt *xprt, char *buf, size_t buflen)
1685{
1686 struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt);
1687 union {
1688 struct sockaddr sa;
1689 struct sockaddr_storage st;
1690 } saddr;
1691 int ret = -ENOTCONN;
1692
1693 mutex_lock(&sock->recv_mutex);
1694 if (sock->sock) {
1695 ret = kernel_getsockname(sock->sock, &saddr.sa);
1696 if (ret >= 0)
1697 ret = snprintf(buf, buflen, "%pISc", &saddr.sa);
1698 }
1699 mutex_unlock(&sock->recv_mutex);
1700 return ret;
1701}
1702
1703static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1704{
1705 if (transport->srcport != 0)
1706 transport->srcport = 0;
1707 if (!transport->xprt.resvport)
1708 return 0;
1709 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1710 return xprt_max_resvport;
1711 return --port;
1712}
1713static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1714{
1715 struct sockaddr_storage myaddr;
1716 int err, nloop = 0;
1717 int port = xs_get_srcport(transport);
1718 unsigned short last;
1719
1720 /*
1721 * If we are asking for any ephemeral port (i.e. port == 0 &&
1722 * transport->xprt.resvport == 0), don't bind. Let the local
1723 * port selection happen implicitly when the socket is used
1724 * (for example at connect time).
1725 *
1726 * This ensures that we can continue to establish TCP
1727 * connections even when all local ephemeral ports are already
1728 * a part of some TCP connection. This makes no difference
1729 * for UDP sockets, but also doesn't harm them.
1730 *
1731 * If we're asking for any reserved port (i.e. port == 0 &&
1732 * transport->xprt.resvport == 1) xs_get_srcport above will
1733 * ensure that port is non-zero and we will bind as needed.
1734 */
1735 if (port <= 0)
1736 return port;
1737
1738 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1739 do {
1740 rpc_set_port((struct sockaddr *)&myaddr, port);
1741 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1742 transport->xprt.addrlen);
1743 if (err == 0) {
1744 if (transport->xprt.reuseport)
1745 transport->srcport = port;
1746 break;
1747 }
1748 last = port;
1749 port = xs_next_srcport(transport, port);
1750 if (port > last)
1751 nloop++;
1752 } while (err == -EADDRINUSE && nloop != 2);
1753
1754 if (myaddr.ss_family == AF_INET)
1755 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1756 &((struct sockaddr_in *)&myaddr)->sin_addr,
1757 port, err ? "failed" : "ok", err);
1758 else
1759 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1760 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1761 port, err ? "failed" : "ok", err);
1762 return err;
1763}
1764
1765/*
1766 * We don't support autobind on AF_LOCAL sockets
1767 */
1768static void xs_local_rpcbind(struct rpc_task *task)
1769{
1770 xprt_set_bound(task->tk_xprt);
1771}
1772
1773static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1774{
1775}
1776
1777#ifdef CONFIG_DEBUG_LOCK_ALLOC
1778static struct lock_class_key xs_key[3];
1779static struct lock_class_key xs_slock_key[3];
1780
1781static inline void xs_reclassify_socketu(struct socket *sock)
1782{
1783 struct sock *sk = sock->sk;
1784
1785 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1786 &xs_slock_key[0], "sk_lock-AF_LOCAL-RPC", &xs_key[0]);
1787}
1788
1789static inline void xs_reclassify_socket4(struct socket *sock)
1790{
1791 struct sock *sk = sock->sk;
1792
1793 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1794 &xs_slock_key[1], "sk_lock-AF_INET-RPC", &xs_key[1]);
1795}
1796
1797static inline void xs_reclassify_socket6(struct socket *sock)
1798{
1799 struct sock *sk = sock->sk;
1800
1801 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1802 &xs_slock_key[2], "sk_lock-AF_INET6-RPC", &xs_key[2]);
1803}
1804
1805static inline void xs_reclassify_socket(int family, struct socket *sock)
1806{
1807 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1808 return;
1809
1810 switch (family) {
1811 case AF_LOCAL:
1812 xs_reclassify_socketu(sock);
1813 break;
1814 case AF_INET:
1815 xs_reclassify_socket4(sock);
1816 break;
1817 case AF_INET6:
1818 xs_reclassify_socket6(sock);
1819 break;
1820 }
1821}
1822#else
1823static inline void xs_reclassify_socket(int family, struct socket *sock)
1824{
1825}
1826#endif
1827
1828static void xs_dummy_setup_socket(struct work_struct *work)
1829{
1830}
1831
1832static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1833 struct sock_xprt *transport, int family, int type,
1834 int protocol, bool reuseport)
1835{
1836 struct file *filp;
1837 struct socket *sock;
1838 int err;
1839
1840 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1841 if (err < 0) {
1842 dprintk("RPC: can't create %d transport socket (%d).\n",
1843 protocol, -err);
1844 goto out;
1845 }
1846 xs_reclassify_socket(family, sock);
1847
1848 if (reuseport)
1849 sock_set_reuseport(sock->sk);
1850
1851 err = xs_bind(transport, sock);
1852 if (err) {
1853 sock_release(sock);
1854 goto out;
1855 }
1856
1857 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1858 if (IS_ERR(filp))
1859 return ERR_CAST(filp);
1860 transport->file = filp;
1861
1862 return sock;
1863out:
1864 return ERR_PTR(err);
1865}
1866
1867static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1868 struct socket *sock)
1869{
1870 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1871 xprt);
1872
1873 if (!transport->inet) {
1874 struct sock *sk = sock->sk;
1875
1876 lock_sock(sk);
1877
1878 xs_save_old_callbacks(transport, sk);
1879
1880 sk->sk_user_data = xprt;
1881 sk->sk_data_ready = xs_data_ready;
1882 sk->sk_write_space = xs_udp_write_space;
1883 sk->sk_state_change = xs_local_state_change;
1884 sk->sk_error_report = xs_error_report;
1885 sk->sk_use_task_frag = false;
1886
1887 xprt_clear_connected(xprt);
1888
1889 /* Reset to new socket */
1890 transport->sock = sock;
1891 transport->inet = sk;
1892
1893 release_sock(sk);
1894 }
1895
1896 xs_stream_start_connect(transport);
1897
1898 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1899}
1900
1901/**
1902 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1903 * @transport: socket transport to connect
1904 */
1905static int xs_local_setup_socket(struct sock_xprt *transport)
1906{
1907 struct rpc_xprt *xprt = &transport->xprt;
1908 struct file *filp;
1909 struct socket *sock;
1910 int status;
1911
1912 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1913 SOCK_STREAM, 0, &sock, 1);
1914 if (status < 0) {
1915 dprintk("RPC: can't create AF_LOCAL "
1916 "transport socket (%d).\n", -status);
1917 goto out;
1918 }
1919 xs_reclassify_socket(AF_LOCAL, sock);
1920
1921 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1922 if (IS_ERR(filp)) {
1923 status = PTR_ERR(filp);
1924 goto out;
1925 }
1926 transport->file = filp;
1927
1928 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1929 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1930
1931 status = xs_local_finish_connecting(xprt, sock);
1932 trace_rpc_socket_connect(xprt, sock, status);
1933 switch (status) {
1934 case 0:
1935 dprintk("RPC: xprt %p connected to %s\n",
1936 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1937 xprt->stat.connect_count++;
1938 xprt->stat.connect_time += (long)jiffies -
1939 xprt->stat.connect_start;
1940 xprt_set_connected(xprt);
1941 break;
1942 case -ENOBUFS:
1943 break;
1944 case -ENOENT:
1945 dprintk("RPC: xprt %p: socket %s does not exist\n",
1946 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1947 break;
1948 case -ECONNREFUSED:
1949 dprintk("RPC: xprt %p: connection refused for %s\n",
1950 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1951 break;
1952 default:
1953 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1954 __func__, -status,
1955 xprt->address_strings[RPC_DISPLAY_ADDR]);
1956 }
1957
1958out:
1959 xprt_clear_connecting(xprt);
1960 xprt_wake_pending_tasks(xprt, status);
1961 return status;
1962}
1963
1964static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1965{
1966 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1967 int ret;
1968
1969 if (transport->file)
1970 goto force_disconnect;
1971
1972 if (RPC_IS_ASYNC(task)) {
1973 /*
1974 * We want the AF_LOCAL connect to be resolved in the
1975 * filesystem namespace of the process making the rpc
1976 * call. Thus we connect synchronously.
1977 *
1978 * If we want to support asynchronous AF_LOCAL calls,
1979 * we'll need to figure out how to pass a namespace to
1980 * connect.
1981 */
1982 rpc_task_set_rpc_status(task, -ENOTCONN);
1983 goto out_wake;
1984 }
1985 ret = xs_local_setup_socket(transport);
1986 if (ret && !RPC_IS_SOFTCONN(task))
1987 msleep_interruptible(15000);
1988 return;
1989force_disconnect:
1990 xprt_force_disconnect(xprt);
1991out_wake:
1992 xprt_clear_connecting(xprt);
1993 xprt_wake_pending_tasks(xprt, -ENOTCONN);
1994}
1995
1996#if IS_ENABLED(CONFIG_SUNRPC_SWAP)
1997/*
1998 * Note that this should be called with XPRT_LOCKED held, or recv_mutex
1999 * held, or when we otherwise know that we have exclusive access to the
2000 * socket, to guard against races with xs_reset_transport.
2001 */
2002static void xs_set_memalloc(struct rpc_xprt *xprt)
2003{
2004 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2005 xprt);
2006
2007 /*
2008 * If there's no sock, then we have nothing to set. The
2009 * reconnecting process will get it for us.
2010 */
2011 if (!transport->inet)
2012 return;
2013 if (atomic_read(&xprt->swapper))
2014 sk_set_memalloc(transport->inet);
2015}
2016
2017/**
2018 * xs_enable_swap - Tag this transport as being used for swap.
2019 * @xprt: transport to tag
2020 *
2021 * Take a reference to this transport on behalf of the rpc_clnt, and
2022 * optionally mark it for swapping if it wasn't already.
2023 */
2024static int
2025xs_enable_swap(struct rpc_xprt *xprt)
2026{
2027 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2028
2029 mutex_lock(&xs->recv_mutex);
2030 if (atomic_inc_return(&xprt->swapper) == 1 &&
2031 xs->inet)
2032 sk_set_memalloc(xs->inet);
2033 mutex_unlock(&xs->recv_mutex);
2034 return 0;
2035}
2036
2037/**
2038 * xs_disable_swap - Untag this transport as being used for swap.
2039 * @xprt: transport to tag
2040 *
2041 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2042 * swapper refcount goes to 0, untag the socket as a memalloc socket.
2043 */
2044static void
2045xs_disable_swap(struct rpc_xprt *xprt)
2046{
2047 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2048
2049 mutex_lock(&xs->recv_mutex);
2050 if (atomic_dec_and_test(&xprt->swapper) &&
2051 xs->inet)
2052 sk_clear_memalloc(xs->inet);
2053 mutex_unlock(&xs->recv_mutex);
2054}
2055#else
2056static void xs_set_memalloc(struct rpc_xprt *xprt)
2057{
2058}
2059
2060static int
2061xs_enable_swap(struct rpc_xprt *xprt)
2062{
2063 return -EINVAL;
2064}
2065
2066static void
2067xs_disable_swap(struct rpc_xprt *xprt)
2068{
2069}
2070#endif
2071
2072static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2073{
2074 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2075
2076 if (!transport->inet) {
2077 struct sock *sk = sock->sk;
2078
2079 lock_sock(sk);
2080
2081 xs_save_old_callbacks(transport, sk);
2082
2083 sk->sk_user_data = xprt;
2084 sk->sk_data_ready = xs_data_ready;
2085 sk->sk_write_space = xs_udp_write_space;
2086 sk->sk_use_task_frag = false;
2087
2088 xprt_set_connected(xprt);
2089
2090 /* Reset to new socket */
2091 transport->sock = sock;
2092 transport->inet = sk;
2093
2094 xs_set_memalloc(xprt);
2095
2096 release_sock(sk);
2097 }
2098 xs_udp_do_set_buffer_size(xprt);
2099
2100 xprt->stat.connect_start = jiffies;
2101}
2102
2103static void xs_udp_setup_socket(struct work_struct *work)
2104{
2105 struct sock_xprt *transport =
2106 container_of(work, struct sock_xprt, connect_worker.work);
2107 struct rpc_xprt *xprt = &transport->xprt;
2108 struct socket *sock;
2109 int status = -EIO;
2110 unsigned int pflags = current->flags;
2111
2112 if (atomic_read(&xprt->swapper))
2113 current->flags |= PF_MEMALLOC;
2114 sock = xs_create_sock(xprt, transport,
2115 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2116 IPPROTO_UDP, false);
2117 if (IS_ERR(sock))
2118 goto out;
2119
2120 dprintk("RPC: worker connecting xprt %p via %s to "
2121 "%s (port %s)\n", xprt,
2122 xprt->address_strings[RPC_DISPLAY_PROTO],
2123 xprt->address_strings[RPC_DISPLAY_ADDR],
2124 xprt->address_strings[RPC_DISPLAY_PORT]);
2125
2126 xs_udp_finish_connecting(xprt, sock);
2127 trace_rpc_socket_connect(xprt, sock, 0);
2128 status = 0;
2129out:
2130 xprt_clear_connecting(xprt);
2131 xprt_unlock_connect(xprt, transport);
2132 xprt_wake_pending_tasks(xprt, status);
2133 current_restore_flags(pflags, PF_MEMALLOC);
2134}
2135
2136/**
2137 * xs_tcp_shutdown - gracefully shut down a TCP socket
2138 * @xprt: transport
2139 *
2140 * Initiates a graceful shutdown of the TCP socket by calling the
2141 * equivalent of shutdown(SHUT_RDWR);
2142 */
2143static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2144{
2145 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2146 struct socket *sock = transport->sock;
2147 int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2148
2149 if (sock == NULL)
2150 return;
2151 if (!xprt->reuseport) {
2152 xs_close(xprt);
2153 return;
2154 }
2155 switch (skst) {
2156 case TCP_FIN_WAIT1:
2157 case TCP_FIN_WAIT2:
2158 break;
2159 case TCP_ESTABLISHED:
2160 case TCP_CLOSE_WAIT:
2161 kernel_sock_shutdown(sock, SHUT_RDWR);
2162 trace_rpc_socket_shutdown(xprt, sock);
2163 break;
2164 default:
2165 xs_reset_transport(transport);
2166 }
2167}
2168
2169static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2170 struct socket *sock)
2171{
2172 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2173 unsigned int keepidle;
2174 unsigned int keepcnt;
2175 unsigned int timeo;
2176
2177 spin_lock(&xprt->transport_lock);
2178 keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2179 keepcnt = xprt->timeout->to_retries + 1;
2180 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2181 (xprt->timeout->to_retries + 1);
2182 clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2183 spin_unlock(&xprt->transport_lock);
2184
2185 /* TCP Keepalive options */
2186 sock_set_keepalive(sock->sk);
2187 tcp_sock_set_keepidle(sock->sk, keepidle);
2188 tcp_sock_set_keepintvl(sock->sk, keepidle);
2189 tcp_sock_set_keepcnt(sock->sk, keepcnt);
2190
2191 /* TCP user timeout (see RFC5482) */
2192 tcp_sock_set_user_timeout(sock->sk, timeo);
2193}
2194
2195static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2196 unsigned long connect_timeout,
2197 unsigned long reconnect_timeout)
2198{
2199 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2200 struct rpc_timeout to;
2201 unsigned long initval;
2202
2203 spin_lock(&xprt->transport_lock);
2204 if (reconnect_timeout < xprt->max_reconnect_timeout)
2205 xprt->max_reconnect_timeout = reconnect_timeout;
2206 if (connect_timeout < xprt->connect_timeout) {
2207 memcpy(&to, xprt->timeout, sizeof(to));
2208 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2209 /* Arbitrary lower limit */
2210 if (initval < XS_TCP_INIT_REEST_TO << 1)
2211 initval = XS_TCP_INIT_REEST_TO << 1;
2212 to.to_initval = initval;
2213 to.to_maxval = initval;
2214 memcpy(&transport->tcp_timeout, &to,
2215 sizeof(transport->tcp_timeout));
2216 xprt->timeout = &transport->tcp_timeout;
2217 xprt->connect_timeout = connect_timeout;
2218 }
2219 set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2220 spin_unlock(&xprt->transport_lock);
2221}
2222
2223static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2224{
2225 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2226
2227 if (!transport->inet) {
2228 struct sock *sk = sock->sk;
2229
2230 /* Avoid temporary address, they are bad for long-lived
2231 * connections such as NFS mounts.
2232 * RFC4941, section 3.6 suggests that:
2233 * Individual applications, which have specific
2234 * knowledge about the normal duration of connections,
2235 * MAY override this as appropriate.
2236 */
2237 if (xs_addr(xprt)->sa_family == PF_INET6) {
2238 ip6_sock_set_addr_preferences(sk,
2239 IPV6_PREFER_SRC_PUBLIC);
2240 }
2241
2242 xs_tcp_set_socket_timeouts(xprt, sock);
2243 tcp_sock_set_nodelay(sk);
2244
2245 lock_sock(sk);
2246
2247 xs_save_old_callbacks(transport, sk);
2248
2249 sk->sk_user_data = xprt;
2250 sk->sk_data_ready = xs_data_ready;
2251 sk->sk_state_change = xs_tcp_state_change;
2252 sk->sk_write_space = xs_tcp_write_space;
2253 sk->sk_error_report = xs_error_report;
2254 sk->sk_use_task_frag = false;
2255
2256 /* socket options */
2257 sock_reset_flag(sk, SOCK_LINGER);
2258
2259 xprt_clear_connected(xprt);
2260
2261 /* Reset to new socket */
2262 transport->sock = sock;
2263 transport->inet = sk;
2264
2265 release_sock(sk);
2266 }
2267
2268 if (!xprt_bound(xprt))
2269 return -ENOTCONN;
2270
2271 xs_set_memalloc(xprt);
2272
2273 xs_stream_start_connect(transport);
2274
2275 /* Tell the socket layer to start connecting... */
2276 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2277 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2278}
2279
2280/**
2281 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2282 * @work: queued work item
2283 *
2284 * Invoked by a work queue tasklet.
2285 */
2286static void xs_tcp_setup_socket(struct work_struct *work)
2287{
2288 struct sock_xprt *transport =
2289 container_of(work, struct sock_xprt, connect_worker.work);
2290 struct socket *sock = transport->sock;
2291 struct rpc_xprt *xprt = &transport->xprt;
2292 int status;
2293 unsigned int pflags = current->flags;
2294
2295 if (atomic_read(&xprt->swapper))
2296 current->flags |= PF_MEMALLOC;
2297
2298 if (xprt_connected(xprt))
2299 goto out;
2300 if (test_and_clear_bit(XPRT_SOCK_CONNECT_SENT,
2301 &transport->sock_state) ||
2302 !sock) {
2303 xs_reset_transport(transport);
2304 sock = xs_create_sock(xprt, transport, xs_addr(xprt)->sa_family,
2305 SOCK_STREAM, IPPROTO_TCP, true);
2306 if (IS_ERR(sock)) {
2307 xprt_wake_pending_tasks(xprt, PTR_ERR(sock));
2308 goto out;
2309 }
2310 }
2311
2312 dprintk("RPC: worker connecting xprt %p via %s to "
2313 "%s (port %s)\n", xprt,
2314 xprt->address_strings[RPC_DISPLAY_PROTO],
2315 xprt->address_strings[RPC_DISPLAY_ADDR],
2316 xprt->address_strings[RPC_DISPLAY_PORT]);
2317
2318 status = xs_tcp_finish_connecting(xprt, sock);
2319 trace_rpc_socket_connect(xprt, sock, status);
2320 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2321 xprt, -status, xprt_connected(xprt),
2322 sock->sk->sk_state);
2323 switch (status) {
2324 case 0:
2325 case -EINPROGRESS:
2326 /* SYN_SENT! */
2327 set_bit(XPRT_SOCK_CONNECT_SENT, &transport->sock_state);
2328 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2329 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2330 fallthrough;
2331 case -EALREADY:
2332 goto out_unlock;
2333 case -EADDRNOTAVAIL:
2334 /* Source port number is unavailable. Try a new one! */
2335 transport->srcport = 0;
2336 status = -EAGAIN;
2337 break;
2338 case -EINVAL:
2339 /* Happens, for instance, if the user specified a link
2340 * local IPv6 address without a scope-id.
2341 */
2342 case -ECONNREFUSED:
2343 case -ECONNRESET:
2344 case -ENETDOWN:
2345 case -ENETUNREACH:
2346 case -EHOSTUNREACH:
2347 case -EADDRINUSE:
2348 case -ENOBUFS:
2349 break;
2350 default:
2351 printk("%s: connect returned unhandled error %d\n",
2352 __func__, status);
2353 status = -EAGAIN;
2354 }
2355
2356 /* xs_tcp_force_close() wakes tasks with a fixed error code.
2357 * We need to wake them first to ensure the correct error code.
2358 */
2359 xprt_wake_pending_tasks(xprt, status);
2360 xs_tcp_force_close(xprt);
2361out:
2362 xprt_clear_connecting(xprt);
2363out_unlock:
2364 xprt_unlock_connect(xprt, transport);
2365 current_restore_flags(pflags, PF_MEMALLOC);
2366}
2367
2368/**
2369 * xs_connect - connect a socket to a remote endpoint
2370 * @xprt: pointer to transport structure
2371 * @task: address of RPC task that manages state of connect request
2372 *
2373 * TCP: If the remote end dropped the connection, delay reconnecting.
2374 *
2375 * UDP socket connects are synchronous, but we use a work queue anyway
2376 * to guarantee that even unprivileged user processes can set up a
2377 * socket on a privileged port.
2378 *
2379 * If a UDP socket connect fails, the delay behavior here prevents
2380 * retry floods (hard mounts).
2381 */
2382static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2383{
2384 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2385 unsigned long delay = 0;
2386
2387 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2388
2389 if (transport->sock != NULL) {
2390 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2391 "seconds\n", xprt, xprt->reestablish_timeout / HZ);
2392
2393 delay = xprt_reconnect_delay(xprt);
2394 xprt_reconnect_backoff(xprt, XS_TCP_INIT_REEST_TO);
2395
2396 } else
2397 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2398
2399 queue_delayed_work(xprtiod_workqueue,
2400 &transport->connect_worker,
2401 delay);
2402}
2403
2404static void xs_wake_disconnect(struct sock_xprt *transport)
2405{
2406 if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state))
2407 xs_tcp_force_close(&transport->xprt);
2408}
2409
2410static void xs_wake_write(struct sock_xprt *transport)
2411{
2412 if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state))
2413 xprt_write_space(&transport->xprt);
2414}
2415
2416static void xs_wake_error(struct sock_xprt *transport)
2417{
2418 int sockerr;
2419
2420 if (!test_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2421 return;
2422 mutex_lock(&transport->recv_mutex);
2423 if (transport->sock == NULL)
2424 goto out;
2425 if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2426 goto out;
2427 sockerr = xchg(&transport->xprt_err, 0);
2428 if (sockerr < 0)
2429 xprt_wake_pending_tasks(&transport->xprt, sockerr);
2430out:
2431 mutex_unlock(&transport->recv_mutex);
2432}
2433
2434static void xs_wake_pending(struct sock_xprt *transport)
2435{
2436 if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING, &transport->sock_state))
2437 xprt_wake_pending_tasks(&transport->xprt, -EAGAIN);
2438}
2439
2440static void xs_error_handle(struct work_struct *work)
2441{
2442 struct sock_xprt *transport = container_of(work,
2443 struct sock_xprt, error_worker);
2444
2445 xs_wake_disconnect(transport);
2446 xs_wake_write(transport);
2447 xs_wake_error(transport);
2448 xs_wake_pending(transport);
2449}
2450
2451/**
2452 * xs_local_print_stats - display AF_LOCAL socket-specific stats
2453 * @xprt: rpc_xprt struct containing statistics
2454 * @seq: output file
2455 *
2456 */
2457static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2458{
2459 long idle_time = 0;
2460
2461 if (xprt_connected(xprt))
2462 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2463
2464 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2465 "%llu %llu %lu %llu %llu\n",
2466 xprt->stat.bind_count,
2467 xprt->stat.connect_count,
2468 xprt->stat.connect_time / HZ,
2469 idle_time,
2470 xprt->stat.sends,
2471 xprt->stat.recvs,
2472 xprt->stat.bad_xids,
2473 xprt->stat.req_u,
2474 xprt->stat.bklog_u,
2475 xprt->stat.max_slots,
2476 xprt->stat.sending_u,
2477 xprt->stat.pending_u);
2478}
2479
2480/**
2481 * xs_udp_print_stats - display UDP socket-specific stats
2482 * @xprt: rpc_xprt struct containing statistics
2483 * @seq: output file
2484 *
2485 */
2486static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2487{
2488 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2489
2490 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2491 "%lu %llu %llu\n",
2492 transport->srcport,
2493 xprt->stat.bind_count,
2494 xprt->stat.sends,
2495 xprt->stat.recvs,
2496 xprt->stat.bad_xids,
2497 xprt->stat.req_u,
2498 xprt->stat.bklog_u,
2499 xprt->stat.max_slots,
2500 xprt->stat.sending_u,
2501 xprt->stat.pending_u);
2502}
2503
2504/**
2505 * xs_tcp_print_stats - display TCP socket-specific stats
2506 * @xprt: rpc_xprt struct containing statistics
2507 * @seq: output file
2508 *
2509 */
2510static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2511{
2512 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2513 long idle_time = 0;
2514
2515 if (xprt_connected(xprt))
2516 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2517
2518 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2519 "%llu %llu %lu %llu %llu\n",
2520 transport->srcport,
2521 xprt->stat.bind_count,
2522 xprt->stat.connect_count,
2523 xprt->stat.connect_time / HZ,
2524 idle_time,
2525 xprt->stat.sends,
2526 xprt->stat.recvs,
2527 xprt->stat.bad_xids,
2528 xprt->stat.req_u,
2529 xprt->stat.bklog_u,
2530 xprt->stat.max_slots,
2531 xprt->stat.sending_u,
2532 xprt->stat.pending_u);
2533}
2534
2535/*
2536 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2537 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2538 * to use the server side send routines.
2539 */
2540static int bc_malloc(struct rpc_task *task)
2541{
2542 struct rpc_rqst *rqst = task->tk_rqstp;
2543 size_t size = rqst->rq_callsize;
2544 struct page *page;
2545 struct rpc_buffer *buf;
2546
2547 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2548 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2549 size);
2550 return -EINVAL;
2551 }
2552
2553 page = alloc_page(GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN);
2554 if (!page)
2555 return -ENOMEM;
2556
2557 buf = page_address(page);
2558 buf->len = PAGE_SIZE;
2559
2560 rqst->rq_buffer = buf->data;
2561 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2562 return 0;
2563}
2564
2565/*
2566 * Free the space allocated in the bc_alloc routine
2567 */
2568static void bc_free(struct rpc_task *task)
2569{
2570 void *buffer = task->tk_rqstp->rq_buffer;
2571 struct rpc_buffer *buf;
2572
2573 buf = container_of(buffer, struct rpc_buffer, data);
2574 free_page((unsigned long)buf);
2575}
2576
2577static int bc_sendto(struct rpc_rqst *req)
2578{
2579 struct xdr_buf *xdr = &req->rq_snd_buf;
2580 struct sock_xprt *transport =
2581 container_of(req->rq_xprt, struct sock_xprt, xprt);
2582 struct msghdr msg = {
2583 .msg_flags = 0,
2584 };
2585 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
2586 (u32)xdr->len);
2587 unsigned int sent = 0;
2588 int err;
2589
2590 req->rq_xtime = ktime_get();
2591 err = xdr_alloc_bvec(xdr, rpc_task_gfp_mask());
2592 if (err < 0)
2593 return err;
2594 err = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, marker, &sent);
2595 xdr_free_bvec(xdr);
2596 if (err < 0 || sent != (xdr->len + sizeof(marker)))
2597 return -EAGAIN;
2598 return sent;
2599}
2600
2601/**
2602 * bc_send_request - Send a backchannel Call on a TCP socket
2603 * @req: rpc_rqst containing Call message to be sent
2604 *
2605 * xpt_mutex ensures @rqstp's whole message is written to the socket
2606 * without interruption.
2607 *
2608 * Return values:
2609 * %0 if the message was sent successfully
2610 * %ENOTCONN if the message was not sent
2611 */
2612static int bc_send_request(struct rpc_rqst *req)
2613{
2614 struct svc_xprt *xprt;
2615 int len;
2616
2617 /*
2618 * Get the server socket associated with this callback xprt
2619 */
2620 xprt = req->rq_xprt->bc_xprt;
2621
2622 /*
2623 * Grab the mutex to serialize data as the connection is shared
2624 * with the fore channel
2625 */
2626 mutex_lock(&xprt->xpt_mutex);
2627 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2628 len = -ENOTCONN;
2629 else
2630 len = bc_sendto(req);
2631 mutex_unlock(&xprt->xpt_mutex);
2632
2633 if (len > 0)
2634 len = 0;
2635
2636 return len;
2637}
2638
2639/*
2640 * The close routine. Since this is client initiated, we do nothing
2641 */
2642
2643static void bc_close(struct rpc_xprt *xprt)
2644{
2645 xprt_disconnect_done(xprt);
2646}
2647
2648/*
2649 * The xprt destroy routine. Again, because this connection is client
2650 * initiated, we do nothing
2651 */
2652
2653static void bc_destroy(struct rpc_xprt *xprt)
2654{
2655 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2656
2657 xs_xprt_free(xprt);
2658 module_put(THIS_MODULE);
2659}
2660
2661static const struct rpc_xprt_ops xs_local_ops = {
2662 .reserve_xprt = xprt_reserve_xprt,
2663 .release_xprt = xprt_release_xprt,
2664 .alloc_slot = xprt_alloc_slot,
2665 .free_slot = xprt_free_slot,
2666 .rpcbind = xs_local_rpcbind,
2667 .set_port = xs_local_set_port,
2668 .connect = xs_local_connect,
2669 .buf_alloc = rpc_malloc,
2670 .buf_free = rpc_free,
2671 .prepare_request = xs_stream_prepare_request,
2672 .send_request = xs_local_send_request,
2673 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2674 .close = xs_close,
2675 .destroy = xs_destroy,
2676 .print_stats = xs_local_print_stats,
2677 .enable_swap = xs_enable_swap,
2678 .disable_swap = xs_disable_swap,
2679};
2680
2681static const struct rpc_xprt_ops xs_udp_ops = {
2682 .set_buffer_size = xs_udp_set_buffer_size,
2683 .reserve_xprt = xprt_reserve_xprt_cong,
2684 .release_xprt = xprt_release_xprt_cong,
2685 .alloc_slot = xprt_alloc_slot,
2686 .free_slot = xprt_free_slot,
2687 .rpcbind = rpcb_getport_async,
2688 .set_port = xs_set_port,
2689 .connect = xs_connect,
2690 .get_srcaddr = xs_sock_srcaddr,
2691 .get_srcport = xs_sock_srcport,
2692 .buf_alloc = rpc_malloc,
2693 .buf_free = rpc_free,
2694 .send_request = xs_udp_send_request,
2695 .wait_for_reply_request = xprt_wait_for_reply_request_rtt,
2696 .timer = xs_udp_timer,
2697 .release_request = xprt_release_rqst_cong,
2698 .close = xs_close,
2699 .destroy = xs_destroy,
2700 .print_stats = xs_udp_print_stats,
2701 .enable_swap = xs_enable_swap,
2702 .disable_swap = xs_disable_swap,
2703 .inject_disconnect = xs_inject_disconnect,
2704};
2705
2706static const struct rpc_xprt_ops xs_tcp_ops = {
2707 .reserve_xprt = xprt_reserve_xprt,
2708 .release_xprt = xprt_release_xprt,
2709 .alloc_slot = xprt_alloc_slot,
2710 .free_slot = xprt_free_slot,
2711 .rpcbind = rpcb_getport_async,
2712 .set_port = xs_set_port,
2713 .connect = xs_connect,
2714 .get_srcaddr = xs_sock_srcaddr,
2715 .get_srcport = xs_sock_srcport,
2716 .buf_alloc = rpc_malloc,
2717 .buf_free = rpc_free,
2718 .prepare_request = xs_stream_prepare_request,
2719 .send_request = xs_tcp_send_request,
2720 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2721 .close = xs_tcp_shutdown,
2722 .destroy = xs_destroy,
2723 .set_connect_timeout = xs_tcp_set_connect_timeout,
2724 .print_stats = xs_tcp_print_stats,
2725 .enable_swap = xs_enable_swap,
2726 .disable_swap = xs_disable_swap,
2727 .inject_disconnect = xs_inject_disconnect,
2728#ifdef CONFIG_SUNRPC_BACKCHANNEL
2729 .bc_setup = xprt_setup_bc,
2730 .bc_maxpayload = xs_tcp_bc_maxpayload,
2731 .bc_num_slots = xprt_bc_max_slots,
2732 .bc_free_rqst = xprt_free_bc_rqst,
2733 .bc_destroy = xprt_destroy_bc,
2734#endif
2735};
2736
2737/*
2738 * The rpc_xprt_ops for the server backchannel
2739 */
2740
2741static const struct rpc_xprt_ops bc_tcp_ops = {
2742 .reserve_xprt = xprt_reserve_xprt,
2743 .release_xprt = xprt_release_xprt,
2744 .alloc_slot = xprt_alloc_slot,
2745 .free_slot = xprt_free_slot,
2746 .buf_alloc = bc_malloc,
2747 .buf_free = bc_free,
2748 .send_request = bc_send_request,
2749 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2750 .close = bc_close,
2751 .destroy = bc_destroy,
2752 .print_stats = xs_tcp_print_stats,
2753 .enable_swap = xs_enable_swap,
2754 .disable_swap = xs_disable_swap,
2755 .inject_disconnect = xs_inject_disconnect,
2756};
2757
2758static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2759{
2760 static const struct sockaddr_in sin = {
2761 .sin_family = AF_INET,
2762 .sin_addr.s_addr = htonl(INADDR_ANY),
2763 };
2764 static const struct sockaddr_in6 sin6 = {
2765 .sin6_family = AF_INET6,
2766 .sin6_addr = IN6ADDR_ANY_INIT,
2767 };
2768
2769 switch (family) {
2770 case AF_LOCAL:
2771 break;
2772 case AF_INET:
2773 memcpy(sap, &sin, sizeof(sin));
2774 break;
2775 case AF_INET6:
2776 memcpy(sap, &sin6, sizeof(sin6));
2777 break;
2778 default:
2779 dprintk("RPC: %s: Bad address family\n", __func__);
2780 return -EAFNOSUPPORT;
2781 }
2782 return 0;
2783}
2784
2785static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2786 unsigned int slot_table_size,
2787 unsigned int max_slot_table_size)
2788{
2789 struct rpc_xprt *xprt;
2790 struct sock_xprt *new;
2791
2792 if (args->addrlen > sizeof(xprt->addr)) {
2793 dprintk("RPC: xs_setup_xprt: address too large\n");
2794 return ERR_PTR(-EBADF);
2795 }
2796
2797 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2798 max_slot_table_size);
2799 if (xprt == NULL) {
2800 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2801 "rpc_xprt\n");
2802 return ERR_PTR(-ENOMEM);
2803 }
2804
2805 new = container_of(xprt, struct sock_xprt, xprt);
2806 mutex_init(&new->recv_mutex);
2807 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2808 xprt->addrlen = args->addrlen;
2809 if (args->srcaddr)
2810 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2811 else {
2812 int err;
2813 err = xs_init_anyaddr(args->dstaddr->sa_family,
2814 (struct sockaddr *)&new->srcaddr);
2815 if (err != 0) {
2816 xprt_free(xprt);
2817 return ERR_PTR(err);
2818 }
2819 }
2820
2821 return xprt;
2822}
2823
2824static const struct rpc_timeout xs_local_default_timeout = {
2825 .to_initval = 10 * HZ,
2826 .to_maxval = 10 * HZ,
2827 .to_retries = 2,
2828};
2829
2830/**
2831 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2832 * @args: rpc transport creation arguments
2833 *
2834 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2835 */
2836static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2837{
2838 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2839 struct sock_xprt *transport;
2840 struct rpc_xprt *xprt;
2841 struct rpc_xprt *ret;
2842
2843 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2844 xprt_max_tcp_slot_table_entries);
2845 if (IS_ERR(xprt))
2846 return xprt;
2847 transport = container_of(xprt, struct sock_xprt, xprt);
2848
2849 xprt->prot = 0;
2850 xprt->xprt_class = &xs_local_transport;
2851 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2852
2853 xprt->bind_timeout = XS_BIND_TO;
2854 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2855 xprt->idle_timeout = XS_IDLE_DISC_TO;
2856
2857 xprt->ops = &xs_local_ops;
2858 xprt->timeout = &xs_local_default_timeout;
2859
2860 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2861 INIT_WORK(&transport->error_worker, xs_error_handle);
2862 INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket);
2863
2864 switch (sun->sun_family) {
2865 case AF_LOCAL:
2866 if (sun->sun_path[0] != '/') {
2867 dprintk("RPC: bad AF_LOCAL address: %s\n",
2868 sun->sun_path);
2869 ret = ERR_PTR(-EINVAL);
2870 goto out_err;
2871 }
2872 xprt_set_bound(xprt);
2873 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2874 break;
2875 default:
2876 ret = ERR_PTR(-EAFNOSUPPORT);
2877 goto out_err;
2878 }
2879
2880 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2881 xprt->address_strings[RPC_DISPLAY_ADDR]);
2882
2883 if (try_module_get(THIS_MODULE))
2884 return xprt;
2885 ret = ERR_PTR(-EINVAL);
2886out_err:
2887 xs_xprt_free(xprt);
2888 return ret;
2889}
2890
2891static const struct rpc_timeout xs_udp_default_timeout = {
2892 .to_initval = 5 * HZ,
2893 .to_maxval = 30 * HZ,
2894 .to_increment = 5 * HZ,
2895 .to_retries = 5,
2896};
2897
2898/**
2899 * xs_setup_udp - Set up transport to use a UDP socket
2900 * @args: rpc transport creation arguments
2901 *
2902 */
2903static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2904{
2905 struct sockaddr *addr = args->dstaddr;
2906 struct rpc_xprt *xprt;
2907 struct sock_xprt *transport;
2908 struct rpc_xprt *ret;
2909
2910 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2911 xprt_udp_slot_table_entries);
2912 if (IS_ERR(xprt))
2913 return xprt;
2914 transport = container_of(xprt, struct sock_xprt, xprt);
2915
2916 xprt->prot = IPPROTO_UDP;
2917 xprt->xprt_class = &xs_udp_transport;
2918 /* XXX: header size can vary due to auth type, IPv6, etc. */
2919 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2920
2921 xprt->bind_timeout = XS_BIND_TO;
2922 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2923 xprt->idle_timeout = XS_IDLE_DISC_TO;
2924
2925 xprt->ops = &xs_udp_ops;
2926
2927 xprt->timeout = &xs_udp_default_timeout;
2928
2929 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
2930 INIT_WORK(&transport->error_worker, xs_error_handle);
2931 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2932
2933 switch (addr->sa_family) {
2934 case AF_INET:
2935 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2936 xprt_set_bound(xprt);
2937
2938 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2939 break;
2940 case AF_INET6:
2941 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2942 xprt_set_bound(xprt);
2943
2944 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2945 break;
2946 default:
2947 ret = ERR_PTR(-EAFNOSUPPORT);
2948 goto out_err;
2949 }
2950
2951 if (xprt_bound(xprt))
2952 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2953 xprt->address_strings[RPC_DISPLAY_ADDR],
2954 xprt->address_strings[RPC_DISPLAY_PORT],
2955 xprt->address_strings[RPC_DISPLAY_PROTO]);
2956 else
2957 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2958 xprt->address_strings[RPC_DISPLAY_ADDR],
2959 xprt->address_strings[RPC_DISPLAY_PROTO]);
2960
2961 if (try_module_get(THIS_MODULE))
2962 return xprt;
2963 ret = ERR_PTR(-EINVAL);
2964out_err:
2965 xs_xprt_free(xprt);
2966 return ret;
2967}
2968
2969static const struct rpc_timeout xs_tcp_default_timeout = {
2970 .to_initval = 60 * HZ,
2971 .to_maxval = 60 * HZ,
2972 .to_retries = 2,
2973};
2974
2975/**
2976 * xs_setup_tcp - Set up transport to use a TCP socket
2977 * @args: rpc transport creation arguments
2978 *
2979 */
2980static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2981{
2982 struct sockaddr *addr = args->dstaddr;
2983 struct rpc_xprt *xprt;
2984 struct sock_xprt *transport;
2985 struct rpc_xprt *ret;
2986 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2987
2988 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2989 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2990
2991 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2992 max_slot_table_size);
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->xprt_class = &xs_tcp_transport;
2999 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3000
3001 xprt->bind_timeout = XS_BIND_TO;
3002 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3003 xprt->idle_timeout = XS_IDLE_DISC_TO;
3004
3005 xprt->ops = &xs_tcp_ops;
3006 xprt->timeout = &xs_tcp_default_timeout;
3007
3008 xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3009 xprt->connect_timeout = xprt->timeout->to_initval *
3010 (xprt->timeout->to_retries + 1);
3011
3012 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3013 INIT_WORK(&transport->error_worker, xs_error_handle);
3014 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
3015
3016 switch (addr->sa_family) {
3017 case AF_INET:
3018 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3019 xprt_set_bound(xprt);
3020
3021 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3022 break;
3023 case AF_INET6:
3024 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3025 xprt_set_bound(xprt);
3026
3027 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3028 break;
3029 default:
3030 ret = ERR_PTR(-EAFNOSUPPORT);
3031 goto out_err;
3032 }
3033
3034 if (xprt_bound(xprt))
3035 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3036 xprt->address_strings[RPC_DISPLAY_ADDR],
3037 xprt->address_strings[RPC_DISPLAY_PORT],
3038 xprt->address_strings[RPC_DISPLAY_PROTO]);
3039 else
3040 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
3041 xprt->address_strings[RPC_DISPLAY_ADDR],
3042 xprt->address_strings[RPC_DISPLAY_PROTO]);
3043
3044 if (try_module_get(THIS_MODULE))
3045 return xprt;
3046 ret = ERR_PTR(-EINVAL);
3047out_err:
3048 xs_xprt_free(xprt);
3049 return ret;
3050}
3051
3052/**
3053 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3054 * @args: rpc transport creation arguments
3055 *
3056 */
3057static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3058{
3059 struct sockaddr *addr = args->dstaddr;
3060 struct rpc_xprt *xprt;
3061 struct sock_xprt *transport;
3062 struct svc_sock *bc_sock;
3063 struct rpc_xprt *ret;
3064
3065 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3066 xprt_tcp_slot_table_entries);
3067 if (IS_ERR(xprt))
3068 return xprt;
3069 transport = container_of(xprt, struct sock_xprt, xprt);
3070
3071 xprt->prot = IPPROTO_TCP;
3072 xprt->xprt_class = &xs_bc_tcp_transport;
3073 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3074 xprt->timeout = &xs_tcp_default_timeout;
3075
3076 /* backchannel */
3077 xprt_set_bound(xprt);
3078 xprt->bind_timeout = 0;
3079 xprt->reestablish_timeout = 0;
3080 xprt->idle_timeout = 0;
3081
3082 xprt->ops = &bc_tcp_ops;
3083
3084 switch (addr->sa_family) {
3085 case AF_INET:
3086 xs_format_peer_addresses(xprt, "tcp",
3087 RPCBIND_NETID_TCP);
3088 break;
3089 case AF_INET6:
3090 xs_format_peer_addresses(xprt, "tcp",
3091 RPCBIND_NETID_TCP6);
3092 break;
3093 default:
3094 ret = ERR_PTR(-EAFNOSUPPORT);
3095 goto out_err;
3096 }
3097
3098 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3099 xprt->address_strings[RPC_DISPLAY_ADDR],
3100 xprt->address_strings[RPC_DISPLAY_PORT],
3101 xprt->address_strings[RPC_DISPLAY_PROTO]);
3102
3103 /*
3104 * Once we've associated a backchannel xprt with a connection,
3105 * we want to keep it around as long as the connection lasts,
3106 * in case we need to start using it for a backchannel again;
3107 * this reference won't be dropped until bc_xprt is destroyed.
3108 */
3109 xprt_get(xprt);
3110 args->bc_xprt->xpt_bc_xprt = xprt;
3111 xprt->bc_xprt = args->bc_xprt;
3112 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3113 transport->sock = bc_sock->sk_sock;
3114 transport->inet = bc_sock->sk_sk;
3115
3116 /*
3117 * Since we don't want connections for the backchannel, we set
3118 * the xprt status to connected
3119 */
3120 xprt_set_connected(xprt);
3121
3122 if (try_module_get(THIS_MODULE))
3123 return xprt;
3124
3125 args->bc_xprt->xpt_bc_xprt = NULL;
3126 args->bc_xprt->xpt_bc_xps = NULL;
3127 xprt_put(xprt);
3128 ret = ERR_PTR(-EINVAL);
3129out_err:
3130 xs_xprt_free(xprt);
3131 return ret;
3132}
3133
3134static struct xprt_class xs_local_transport = {
3135 .list = LIST_HEAD_INIT(xs_local_transport.list),
3136 .name = "named UNIX socket",
3137 .owner = THIS_MODULE,
3138 .ident = XPRT_TRANSPORT_LOCAL,
3139 .setup = xs_setup_local,
3140 .netid = { "" },
3141};
3142
3143static struct xprt_class xs_udp_transport = {
3144 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3145 .name = "udp",
3146 .owner = THIS_MODULE,
3147 .ident = XPRT_TRANSPORT_UDP,
3148 .setup = xs_setup_udp,
3149 .netid = { "udp", "udp6", "" },
3150};
3151
3152static struct xprt_class xs_tcp_transport = {
3153 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3154 .name = "tcp",
3155 .owner = THIS_MODULE,
3156 .ident = XPRT_TRANSPORT_TCP,
3157 .setup = xs_setup_tcp,
3158 .netid = { "tcp", "tcp6", "" },
3159};
3160
3161static struct xprt_class xs_bc_tcp_transport = {
3162 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3163 .name = "tcp NFSv4.1 backchannel",
3164 .owner = THIS_MODULE,
3165 .ident = XPRT_TRANSPORT_BC_TCP,
3166 .setup = xs_setup_bc_tcp,
3167 .netid = { "" },
3168};
3169
3170/**
3171 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3172 *
3173 */
3174int init_socket_xprt(void)
3175{
3176 if (!sunrpc_table_header)
3177 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3178
3179 xprt_register_transport(&xs_local_transport);
3180 xprt_register_transport(&xs_udp_transport);
3181 xprt_register_transport(&xs_tcp_transport);
3182 xprt_register_transport(&xs_bc_tcp_transport);
3183
3184 return 0;
3185}
3186
3187/**
3188 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3189 *
3190 */
3191void cleanup_socket_xprt(void)
3192{
3193 if (sunrpc_table_header) {
3194 unregister_sysctl_table(sunrpc_table_header);
3195 sunrpc_table_header = NULL;
3196 }
3197
3198 xprt_unregister_transport(&xs_local_transport);
3199 xprt_unregister_transport(&xs_udp_transport);
3200 xprt_unregister_transport(&xs_tcp_transport);
3201 xprt_unregister_transport(&xs_bc_tcp_transport);
3202}
3203
3204static int param_set_portnr(const char *val, const struct kernel_param *kp)
3205{
3206 return param_set_uint_minmax(val, kp,
3207 RPC_MIN_RESVPORT,
3208 RPC_MAX_RESVPORT);
3209}
3210
3211static const struct kernel_param_ops param_ops_portnr = {
3212 .set = param_set_portnr,
3213 .get = param_get_uint,
3214};
3215
3216#define param_check_portnr(name, p) \
3217 __param_check(name, p, unsigned int);
3218
3219module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3220module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3221
3222static int param_set_slot_table_size(const char *val,
3223 const struct kernel_param *kp)
3224{
3225 return param_set_uint_minmax(val, kp,
3226 RPC_MIN_SLOT_TABLE,
3227 RPC_MAX_SLOT_TABLE);
3228}
3229
3230static const struct kernel_param_ops param_ops_slot_table_size = {
3231 .set = param_set_slot_table_size,
3232 .get = param_get_uint,
3233};
3234
3235#define param_check_slot_table_size(name, p) \
3236 __param_check(name, p, unsigned int);
3237
3238static int param_set_max_slot_table_size(const char *val,
3239 const struct kernel_param *kp)
3240{
3241 return param_set_uint_minmax(val, kp,
3242 RPC_MIN_SLOT_TABLE,
3243 RPC_MAX_SLOT_TABLE_LIMIT);
3244}
3245
3246static const struct kernel_param_ops param_ops_max_slot_table_size = {
3247 .set = param_set_max_slot_table_size,
3248 .get = param_get_uint,
3249};
3250
3251#define param_check_max_slot_table_size(name, p) \
3252 __param_check(name, p, unsigned int);
3253
3254module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3255 slot_table_size, 0644);
3256module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3257 max_slot_table_size, 0644);
3258module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3259 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/addr.h>
37#include <linux/sunrpc/sched.h>
38#include <linux/sunrpc/svcsock.h>
39#include <linux/sunrpc/xprtsock.h>
40#include <linux/file.h>
41#ifdef CONFIG_SUNRPC_BACKCHANNEL
42#include <linux/sunrpc/bc_xprt.h>
43#endif
44
45#include <net/sock.h>
46#include <net/checksum.h>
47#include <net/udp.h>
48#include <net/tcp.h>
49
50#include <trace/events/sunrpc.h>
51
52#include "sunrpc.h"
53
54static void xs_close(struct rpc_xprt *xprt);
55
56/*
57 * xprtsock tunables
58 */
59static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
60static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
61static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
62
63static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
64static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
65
66#define XS_TCP_LINGER_TO (15U * HZ)
67static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
68
69/*
70 * We can register our own files under /proc/sys/sunrpc by
71 * calling register_sysctl_table() again. The files in that
72 * directory become the union of all files registered there.
73 *
74 * We simply need to make sure that we don't collide with
75 * someone else's file names!
76 */
77
78#ifdef RPC_DEBUG
79
80static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
81static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
82static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
83static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
84static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
85
86static struct ctl_table_header *sunrpc_table_header;
87
88/*
89 * FIXME: changing the UDP slot table size should also resize the UDP
90 * socket buffers for existing UDP transports
91 */
92static struct ctl_table xs_tunables_table[] = {
93 {
94 .procname = "udp_slot_table_entries",
95 .data = &xprt_udp_slot_table_entries,
96 .maxlen = sizeof(unsigned int),
97 .mode = 0644,
98 .proc_handler = proc_dointvec_minmax,
99 .extra1 = &min_slot_table_size,
100 .extra2 = &max_slot_table_size
101 },
102 {
103 .procname = "tcp_slot_table_entries",
104 .data = &xprt_tcp_slot_table_entries,
105 .maxlen = sizeof(unsigned int),
106 .mode = 0644,
107 .proc_handler = proc_dointvec_minmax,
108 .extra1 = &min_slot_table_size,
109 .extra2 = &max_slot_table_size
110 },
111 {
112 .procname = "tcp_max_slot_table_entries",
113 .data = &xprt_max_tcp_slot_table_entries,
114 .maxlen = sizeof(unsigned int),
115 .mode = 0644,
116 .proc_handler = proc_dointvec_minmax,
117 .extra1 = &min_slot_table_size,
118 .extra2 = &max_tcp_slot_table_limit
119 },
120 {
121 .procname = "min_resvport",
122 .data = &xprt_min_resvport,
123 .maxlen = sizeof(unsigned int),
124 .mode = 0644,
125 .proc_handler = proc_dointvec_minmax,
126 .extra1 = &xprt_min_resvport_limit,
127 .extra2 = &xprt_max_resvport_limit
128 },
129 {
130 .procname = "max_resvport",
131 .data = &xprt_max_resvport,
132 .maxlen = sizeof(unsigned int),
133 .mode = 0644,
134 .proc_handler = proc_dointvec_minmax,
135 .extra1 = &xprt_min_resvport_limit,
136 .extra2 = &xprt_max_resvport_limit
137 },
138 {
139 .procname = "tcp_fin_timeout",
140 .data = &xs_tcp_fin_timeout,
141 .maxlen = sizeof(xs_tcp_fin_timeout),
142 .mode = 0644,
143 .proc_handler = proc_dointvec_jiffies,
144 },
145 { },
146};
147
148static struct ctl_table sunrpc_table[] = {
149 {
150 .procname = "sunrpc",
151 .mode = 0555,
152 .child = xs_tunables_table
153 },
154 { },
155};
156
157#endif
158
159/*
160 * Wait duration for a reply from the RPC portmapper.
161 */
162#define XS_BIND_TO (60U * HZ)
163
164/*
165 * Delay if a UDP socket connect error occurs. This is most likely some
166 * kind of resource problem on the local host.
167 */
168#define XS_UDP_REEST_TO (2U * HZ)
169
170/*
171 * The reestablish timeout allows clients to delay for a bit before attempting
172 * to reconnect to a server that just dropped our connection.
173 *
174 * We implement an exponential backoff when trying to reestablish a TCP
175 * transport connection with the server. Some servers like to drop a TCP
176 * connection when they are overworked, so we start with a short timeout and
177 * increase over time if the server is down or not responding.
178 */
179#define XS_TCP_INIT_REEST_TO (3U * HZ)
180#define XS_TCP_MAX_REEST_TO (5U * 60 * HZ)
181
182/*
183 * TCP idle timeout; client drops the transport socket if it is idle
184 * for this long. Note that we also timeout UDP sockets to prevent
185 * holding port numbers when there is no RPC traffic.
186 */
187#define XS_IDLE_DISC_TO (5U * 60 * HZ)
188
189#ifdef RPC_DEBUG
190# undef RPC_DEBUG_DATA
191# define RPCDBG_FACILITY RPCDBG_TRANS
192#endif
193
194#ifdef RPC_DEBUG_DATA
195static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
196{
197 u8 *buf = (u8 *) packet;
198 int j;
199
200 dprintk("RPC: %s\n", msg);
201 for (j = 0; j < count && j < 128; j += 4) {
202 if (!(j & 31)) {
203 if (j)
204 dprintk("\n");
205 dprintk("0x%04x ", j);
206 }
207 dprintk("%02x%02x%02x%02x ",
208 buf[j], buf[j+1], buf[j+2], buf[j+3]);
209 }
210 dprintk("\n");
211}
212#else
213static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
214{
215 /* NOP */
216}
217#endif
218
219struct sock_xprt {
220 struct rpc_xprt xprt;
221
222 /*
223 * Network layer
224 */
225 struct socket * sock;
226 struct sock * inet;
227
228 /*
229 * State of TCP reply receive
230 */
231 __be32 tcp_fraghdr,
232 tcp_xid,
233 tcp_calldir;
234
235 u32 tcp_offset,
236 tcp_reclen;
237
238 unsigned long tcp_copied,
239 tcp_flags;
240
241 /*
242 * Connection of transports
243 */
244 struct delayed_work connect_worker;
245 struct sockaddr_storage srcaddr;
246 unsigned short srcport;
247
248 /*
249 * UDP socket buffer size parameters
250 */
251 size_t rcvsize,
252 sndsize;
253
254 /*
255 * Saved socket callback addresses
256 */
257 void (*old_data_ready)(struct sock *);
258 void (*old_state_change)(struct sock *);
259 void (*old_write_space)(struct sock *);
260 void (*old_error_report)(struct sock *);
261};
262
263/*
264 * TCP receive state flags
265 */
266#define TCP_RCV_LAST_FRAG (1UL << 0)
267#define TCP_RCV_COPY_FRAGHDR (1UL << 1)
268#define TCP_RCV_COPY_XID (1UL << 2)
269#define TCP_RCV_COPY_DATA (1UL << 3)
270#define TCP_RCV_READ_CALLDIR (1UL << 4)
271#define TCP_RCV_COPY_CALLDIR (1UL << 5)
272
273/*
274 * TCP RPC flags
275 */
276#define TCP_RPC_REPLY (1UL << 6)
277
278static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
279{
280 return (struct rpc_xprt *) sk->sk_user_data;
281}
282
283static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
284{
285 return (struct sockaddr *) &xprt->addr;
286}
287
288static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
289{
290 return (struct sockaddr_un *) &xprt->addr;
291}
292
293static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
294{
295 return (struct sockaddr_in *) &xprt->addr;
296}
297
298static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
299{
300 return (struct sockaddr_in6 *) &xprt->addr;
301}
302
303static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
304{
305 struct sockaddr *sap = xs_addr(xprt);
306 struct sockaddr_in6 *sin6;
307 struct sockaddr_in *sin;
308 struct sockaddr_un *sun;
309 char buf[128];
310
311 switch (sap->sa_family) {
312 case AF_LOCAL:
313 sun = xs_addr_un(xprt);
314 strlcpy(buf, sun->sun_path, sizeof(buf));
315 xprt->address_strings[RPC_DISPLAY_ADDR] =
316 kstrdup(buf, GFP_KERNEL);
317 break;
318 case AF_INET:
319 (void)rpc_ntop(sap, buf, sizeof(buf));
320 xprt->address_strings[RPC_DISPLAY_ADDR] =
321 kstrdup(buf, GFP_KERNEL);
322 sin = xs_addr_in(xprt);
323 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
324 break;
325 case AF_INET6:
326 (void)rpc_ntop(sap, buf, sizeof(buf));
327 xprt->address_strings[RPC_DISPLAY_ADDR] =
328 kstrdup(buf, GFP_KERNEL);
329 sin6 = xs_addr_in6(xprt);
330 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
331 break;
332 default:
333 BUG();
334 }
335
336 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
337}
338
339static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
340{
341 struct sockaddr *sap = xs_addr(xprt);
342 char buf[128];
343
344 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
345 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
346
347 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
348 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
349}
350
351static void xs_format_peer_addresses(struct rpc_xprt *xprt,
352 const char *protocol,
353 const char *netid)
354{
355 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
356 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
357 xs_format_common_peer_addresses(xprt);
358 xs_format_common_peer_ports(xprt);
359}
360
361static void xs_update_peer_port(struct rpc_xprt *xprt)
362{
363 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
364 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
365
366 xs_format_common_peer_ports(xprt);
367}
368
369static void xs_free_peer_addresses(struct rpc_xprt *xprt)
370{
371 unsigned int i;
372
373 for (i = 0; i < RPC_DISPLAY_MAX; i++)
374 switch (i) {
375 case RPC_DISPLAY_PROTO:
376 case RPC_DISPLAY_NETID:
377 continue;
378 default:
379 kfree(xprt->address_strings[i]);
380 }
381}
382
383#define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
384
385static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
386{
387 struct msghdr msg = {
388 .msg_name = addr,
389 .msg_namelen = addrlen,
390 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
391 };
392 struct kvec iov = {
393 .iov_base = vec->iov_base + base,
394 .iov_len = vec->iov_len - base,
395 };
396
397 if (iov.iov_len != 0)
398 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
399 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
400}
401
402static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy)
403{
404 ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
405 int offset, size_t size, int flags);
406 struct page **ppage;
407 unsigned int remainder;
408 int err, sent = 0;
409
410 remainder = xdr->page_len - base;
411 base += xdr->page_base;
412 ppage = xdr->pages + (base >> PAGE_SHIFT);
413 base &= ~PAGE_MASK;
414 do_sendpage = sock->ops->sendpage;
415 if (!zerocopy)
416 do_sendpage = sock_no_sendpage;
417 for(;;) {
418 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
419 int flags = XS_SENDMSG_FLAGS;
420
421 remainder -= len;
422 if (remainder != 0 || more)
423 flags |= MSG_MORE;
424 err = do_sendpage(sock, *ppage, base, len, flags);
425 if (remainder == 0 || err != len)
426 break;
427 sent += err;
428 ppage++;
429 base = 0;
430 }
431 if (sent == 0)
432 return err;
433 if (err > 0)
434 sent += err;
435 return sent;
436}
437
438/**
439 * xs_sendpages - write pages directly to a socket
440 * @sock: socket to send on
441 * @addr: UDP only -- address of destination
442 * @addrlen: UDP only -- length of destination address
443 * @xdr: buffer containing this request
444 * @base: starting position in the buffer
445 * @zerocopy: true if it is safe to use sendpage()
446 *
447 */
448static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy)
449{
450 unsigned int remainder = xdr->len - base;
451 int err, sent = 0;
452
453 if (unlikely(!sock))
454 return -ENOTSOCK;
455
456 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
457 if (base != 0) {
458 addr = NULL;
459 addrlen = 0;
460 }
461
462 if (base < xdr->head[0].iov_len || addr != NULL) {
463 unsigned int len = xdr->head[0].iov_len - base;
464 remainder -= len;
465 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
466 if (remainder == 0 || err != len)
467 goto out;
468 sent += err;
469 base = 0;
470 } else
471 base -= xdr->head[0].iov_len;
472
473 if (base < xdr->page_len) {
474 unsigned int len = xdr->page_len - base;
475 remainder -= len;
476 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy);
477 if (remainder == 0 || err != len)
478 goto out;
479 sent += err;
480 base = 0;
481 } else
482 base -= xdr->page_len;
483
484 if (base >= xdr->tail[0].iov_len)
485 return sent;
486 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
487out:
488 if (sent == 0)
489 return err;
490 if (err > 0)
491 sent += err;
492 return sent;
493}
494
495static void xs_nospace_callback(struct rpc_task *task)
496{
497 struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
498
499 transport->inet->sk_write_pending--;
500 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
501}
502
503/**
504 * xs_nospace - place task on wait queue if transmit was incomplete
505 * @task: task to put to sleep
506 *
507 */
508static int xs_nospace(struct rpc_task *task)
509{
510 struct rpc_rqst *req = task->tk_rqstp;
511 struct rpc_xprt *xprt = req->rq_xprt;
512 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
513 struct sock *sk = transport->inet;
514 int ret = -EAGAIN;
515
516 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
517 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
518 req->rq_slen);
519
520 /* Protect against races with write_space */
521 spin_lock_bh(&xprt->transport_lock);
522
523 /* Don't race with disconnect */
524 if (xprt_connected(xprt)) {
525 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
526 /*
527 * Notify TCP that we're limited by the application
528 * window size
529 */
530 set_bit(SOCK_NOSPACE, &transport->sock->flags);
531 sk->sk_write_pending++;
532 /* ...and wait for more buffer space */
533 xprt_wait_for_buffer_space(task, xs_nospace_callback);
534 }
535 } else {
536 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
537 ret = -ENOTCONN;
538 }
539
540 spin_unlock_bh(&xprt->transport_lock);
541
542 /* Race breaker in case memory is freed before above code is called */
543 sk->sk_write_space(sk);
544 return ret;
545}
546
547/*
548 * Construct a stream transport record marker in @buf.
549 */
550static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
551{
552 u32 reclen = buf->len - sizeof(rpc_fraghdr);
553 rpc_fraghdr *base = buf->head[0].iov_base;
554 *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
555}
556
557/**
558 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
559 * @task: RPC task that manages the state of an RPC request
560 *
561 * Return values:
562 * 0: The request has been sent
563 * EAGAIN: The socket was blocked, please call again later to
564 * complete the request
565 * ENOTCONN: Caller needs to invoke connect logic then call again
566 * other: Some other error occured, the request was not sent
567 */
568static int xs_local_send_request(struct rpc_task *task)
569{
570 struct rpc_rqst *req = task->tk_rqstp;
571 struct rpc_xprt *xprt = req->rq_xprt;
572 struct sock_xprt *transport =
573 container_of(xprt, struct sock_xprt, xprt);
574 struct xdr_buf *xdr = &req->rq_snd_buf;
575 int status;
576
577 xs_encode_stream_record_marker(&req->rq_snd_buf);
578
579 xs_pktdump("packet data:",
580 req->rq_svec->iov_base, req->rq_svec->iov_len);
581
582 status = xs_sendpages(transport->sock, NULL, 0,
583 xdr, req->rq_bytes_sent, true);
584 dprintk("RPC: %s(%u) = %d\n",
585 __func__, xdr->len - req->rq_bytes_sent, status);
586 if (likely(status >= 0)) {
587 req->rq_bytes_sent += status;
588 req->rq_xmit_bytes_sent += status;
589 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
590 req->rq_bytes_sent = 0;
591 return 0;
592 }
593 status = -EAGAIN;
594 }
595
596 switch (status) {
597 case -EAGAIN:
598 status = xs_nospace(task);
599 break;
600 default:
601 dprintk("RPC: sendmsg returned unrecognized error %d\n",
602 -status);
603 case -EPIPE:
604 xs_close(xprt);
605 status = -ENOTCONN;
606 }
607
608 return status;
609}
610
611/**
612 * xs_udp_send_request - write an RPC request to a UDP socket
613 * @task: address of RPC task that manages the state of an RPC request
614 *
615 * Return values:
616 * 0: The request has been sent
617 * EAGAIN: The socket was blocked, please call again later to
618 * complete the request
619 * ENOTCONN: Caller needs to invoke connect logic then call again
620 * other: Some other error occurred, the request was not sent
621 */
622static int xs_udp_send_request(struct rpc_task *task)
623{
624 struct rpc_rqst *req = task->tk_rqstp;
625 struct rpc_xprt *xprt = req->rq_xprt;
626 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
627 struct xdr_buf *xdr = &req->rq_snd_buf;
628 int status;
629
630 xs_pktdump("packet data:",
631 req->rq_svec->iov_base,
632 req->rq_svec->iov_len);
633
634 if (!xprt_bound(xprt))
635 return -ENOTCONN;
636 status = xs_sendpages(transport->sock,
637 xs_addr(xprt),
638 xprt->addrlen, xdr,
639 req->rq_bytes_sent, true);
640
641 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
642 xdr->len - req->rq_bytes_sent, status);
643
644 if (status >= 0) {
645 req->rq_xmit_bytes_sent += status;
646 if (status >= req->rq_slen)
647 return 0;
648 /* Still some bytes left; set up for a retry later. */
649 status = -EAGAIN;
650 }
651
652 switch (status) {
653 case -ENOTSOCK:
654 status = -ENOTCONN;
655 /* Should we call xs_close() here? */
656 break;
657 case -EAGAIN:
658 status = xs_nospace(task);
659 break;
660 default:
661 dprintk("RPC: sendmsg returned unrecognized error %d\n",
662 -status);
663 case -ENETUNREACH:
664 case -EPIPE:
665 case -ECONNREFUSED:
666 /* When the server has died, an ICMP port unreachable message
667 * prompts ECONNREFUSED. */
668 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
669 }
670
671 return status;
672}
673
674/**
675 * xs_tcp_shutdown - gracefully shut down a TCP socket
676 * @xprt: transport
677 *
678 * Initiates a graceful shutdown of the TCP socket by calling the
679 * equivalent of shutdown(SHUT_WR);
680 */
681static void xs_tcp_shutdown(struct rpc_xprt *xprt)
682{
683 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
684 struct socket *sock = transport->sock;
685
686 if (sock != NULL) {
687 kernel_sock_shutdown(sock, SHUT_WR);
688 trace_rpc_socket_shutdown(xprt, sock);
689 }
690}
691
692/**
693 * xs_tcp_send_request - write an RPC request to a TCP socket
694 * @task: address of RPC task that manages the state of an RPC request
695 *
696 * Return values:
697 * 0: The request has been sent
698 * EAGAIN: The socket was blocked, please call again later to
699 * complete the request
700 * ENOTCONN: Caller needs to invoke connect logic then call again
701 * other: Some other error occurred, the request was not sent
702 *
703 * XXX: In the case of soft timeouts, should we eventually give up
704 * if sendmsg is not able to make progress?
705 */
706static int xs_tcp_send_request(struct rpc_task *task)
707{
708 struct rpc_rqst *req = task->tk_rqstp;
709 struct rpc_xprt *xprt = req->rq_xprt;
710 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
711 struct xdr_buf *xdr = &req->rq_snd_buf;
712 bool zerocopy = true;
713 int status;
714
715 xs_encode_stream_record_marker(&req->rq_snd_buf);
716
717 xs_pktdump("packet data:",
718 req->rq_svec->iov_base,
719 req->rq_svec->iov_len);
720 /* Don't use zero copy if this is a resend. If the RPC call
721 * completes while the socket holds a reference to the pages,
722 * then we may end up resending corrupted data.
723 */
724 if (task->tk_flags & RPC_TASK_SENT)
725 zerocopy = false;
726
727 /* Continue transmitting the packet/record. We must be careful
728 * to cope with writespace callbacks arriving _after_ we have
729 * called sendmsg(). */
730 while (1) {
731 status = xs_sendpages(transport->sock,
732 NULL, 0, xdr, req->rq_bytes_sent,
733 zerocopy);
734
735 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
736 xdr->len - req->rq_bytes_sent, status);
737
738 if (unlikely(status < 0))
739 break;
740
741 /* If we've sent the entire packet, immediately
742 * reset the count of bytes sent. */
743 req->rq_bytes_sent += status;
744 req->rq_xmit_bytes_sent += status;
745 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
746 req->rq_bytes_sent = 0;
747 return 0;
748 }
749
750 if (status != 0)
751 continue;
752 status = -EAGAIN;
753 break;
754 }
755
756 switch (status) {
757 case -ENOTSOCK:
758 status = -ENOTCONN;
759 /* Should we call xs_close() here? */
760 break;
761 case -EAGAIN:
762 status = xs_nospace(task);
763 break;
764 default:
765 dprintk("RPC: sendmsg returned unrecognized error %d\n",
766 -status);
767 case -ECONNRESET:
768 xs_tcp_shutdown(xprt);
769 case -ECONNREFUSED:
770 case -ENOTCONN:
771 case -EPIPE:
772 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
773 }
774
775 return status;
776}
777
778/**
779 * xs_tcp_release_xprt - clean up after a tcp transmission
780 * @xprt: transport
781 * @task: rpc task
782 *
783 * This cleans up if an error causes us to abort the transmission of a request.
784 * In this case, the socket may need to be reset in order to avoid confusing
785 * the server.
786 */
787static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
788{
789 struct rpc_rqst *req;
790
791 if (task != xprt->snd_task)
792 return;
793 if (task == NULL)
794 goto out_release;
795 req = task->tk_rqstp;
796 if (req == NULL)
797 goto out_release;
798 if (req->rq_bytes_sent == 0)
799 goto out_release;
800 if (req->rq_bytes_sent == req->rq_snd_buf.len)
801 goto out_release;
802 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
803out_release:
804 xprt_release_xprt(xprt, task);
805}
806
807static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
808{
809 transport->old_data_ready = sk->sk_data_ready;
810 transport->old_state_change = sk->sk_state_change;
811 transport->old_write_space = sk->sk_write_space;
812 transport->old_error_report = sk->sk_error_report;
813}
814
815static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
816{
817 sk->sk_data_ready = transport->old_data_ready;
818 sk->sk_state_change = transport->old_state_change;
819 sk->sk_write_space = transport->old_write_space;
820 sk->sk_error_report = transport->old_error_report;
821}
822
823/**
824 * xs_error_report - callback to handle TCP socket state errors
825 * @sk: socket
826 *
827 * Note: we don't call sock_error() since there may be a rpc_task
828 * using the socket, and so we don't want to clear sk->sk_err.
829 */
830static void xs_error_report(struct sock *sk)
831{
832 struct rpc_xprt *xprt;
833 int err;
834
835 read_lock_bh(&sk->sk_callback_lock);
836 if (!(xprt = xprt_from_sock(sk)))
837 goto out;
838
839 err = -sk->sk_err;
840 if (err == 0)
841 goto out;
842 dprintk("RPC: xs_error_report client %p, error=%d...\n",
843 xprt, -err);
844 trace_rpc_socket_error(xprt, sk->sk_socket, err);
845 xprt_wake_pending_tasks(xprt, err);
846 out:
847 read_unlock_bh(&sk->sk_callback_lock);
848}
849
850static void xs_reset_transport(struct sock_xprt *transport)
851{
852 struct socket *sock = transport->sock;
853 struct sock *sk = transport->inet;
854
855 if (sk == NULL)
856 return;
857
858 transport->srcport = 0;
859
860 write_lock_bh(&sk->sk_callback_lock);
861 transport->inet = NULL;
862 transport->sock = NULL;
863
864 sk->sk_user_data = NULL;
865
866 xs_restore_old_callbacks(transport, sk);
867 write_unlock_bh(&sk->sk_callback_lock);
868
869 sk->sk_no_check = 0;
870
871 trace_rpc_socket_close(&transport->xprt, sock);
872 sock_release(sock);
873}
874
875/**
876 * xs_close - close a socket
877 * @xprt: transport
878 *
879 * This is used when all requests are complete; ie, no DRC state remains
880 * on the server we want to save.
881 *
882 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
883 * xs_reset_transport() zeroing the socket from underneath a writer.
884 */
885static void xs_close(struct rpc_xprt *xprt)
886{
887 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
888
889 dprintk("RPC: xs_close xprt %p\n", xprt);
890
891 cancel_delayed_work_sync(&transport->connect_worker);
892
893 xs_reset_transport(transport);
894 xprt->reestablish_timeout = 0;
895
896 smp_mb__before_clear_bit();
897 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
898 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
899 clear_bit(XPRT_CLOSING, &xprt->state);
900 smp_mb__after_clear_bit();
901 xprt_disconnect_done(xprt);
902}
903
904static void xs_tcp_close(struct rpc_xprt *xprt)
905{
906 if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state))
907 xs_close(xprt);
908 else
909 xs_tcp_shutdown(xprt);
910}
911
912static void xs_xprt_free(struct rpc_xprt *xprt)
913{
914 xs_free_peer_addresses(xprt);
915 xprt_free(xprt);
916}
917
918/**
919 * xs_destroy - prepare to shutdown a transport
920 * @xprt: doomed transport
921 *
922 */
923static void xs_destroy(struct rpc_xprt *xprt)
924{
925 dprintk("RPC: xs_destroy xprt %p\n", xprt);
926
927 xs_close(xprt);
928 xs_xprt_free(xprt);
929 module_put(THIS_MODULE);
930}
931
932static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
933{
934 struct xdr_skb_reader desc = {
935 .skb = skb,
936 .offset = sizeof(rpc_fraghdr),
937 .count = skb->len - sizeof(rpc_fraghdr),
938 };
939
940 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
941 return -1;
942 if (desc.count)
943 return -1;
944 return 0;
945}
946
947/**
948 * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets
949 * @sk: socket with data to read
950 * @len: how much data to read
951 *
952 * Currently this assumes we can read the whole reply in a single gulp.
953 */
954static void xs_local_data_ready(struct sock *sk)
955{
956 struct rpc_task *task;
957 struct rpc_xprt *xprt;
958 struct rpc_rqst *rovr;
959 struct sk_buff *skb;
960 int err, repsize, copied;
961 u32 _xid;
962 __be32 *xp;
963
964 read_lock_bh(&sk->sk_callback_lock);
965 dprintk("RPC: %s...\n", __func__);
966 xprt = xprt_from_sock(sk);
967 if (xprt == NULL)
968 goto out;
969
970 skb = skb_recv_datagram(sk, 0, 1, &err);
971 if (skb == NULL)
972 goto out;
973
974 repsize = skb->len - sizeof(rpc_fraghdr);
975 if (repsize < 4) {
976 dprintk("RPC: impossible RPC reply size %d\n", repsize);
977 goto dropit;
978 }
979
980 /* Copy the XID from the skb... */
981 xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
982 if (xp == NULL)
983 goto dropit;
984
985 /* Look up and lock the request corresponding to the given XID */
986 spin_lock(&xprt->transport_lock);
987 rovr = xprt_lookup_rqst(xprt, *xp);
988 if (!rovr)
989 goto out_unlock;
990 task = rovr->rq_task;
991
992 copied = rovr->rq_private_buf.buflen;
993 if (copied > repsize)
994 copied = repsize;
995
996 if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
997 dprintk("RPC: sk_buff copy failed\n");
998 goto out_unlock;
999 }
1000
1001 xprt_complete_rqst(task, copied);
1002
1003 out_unlock:
1004 spin_unlock(&xprt->transport_lock);
1005 dropit:
1006 skb_free_datagram(sk, skb);
1007 out:
1008 read_unlock_bh(&sk->sk_callback_lock);
1009}
1010
1011/**
1012 * xs_udp_data_ready - "data ready" callback for UDP sockets
1013 * @sk: socket with data to read
1014 * @len: how much data to read
1015 *
1016 */
1017static void xs_udp_data_ready(struct sock *sk)
1018{
1019 struct rpc_task *task;
1020 struct rpc_xprt *xprt;
1021 struct rpc_rqst *rovr;
1022 struct sk_buff *skb;
1023 int err, repsize, copied;
1024 u32 _xid;
1025 __be32 *xp;
1026
1027 read_lock_bh(&sk->sk_callback_lock);
1028 dprintk("RPC: xs_udp_data_ready...\n");
1029 if (!(xprt = xprt_from_sock(sk)))
1030 goto out;
1031
1032 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
1033 goto out;
1034
1035 repsize = skb->len - sizeof(struct udphdr);
1036 if (repsize < 4) {
1037 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
1038 goto dropit;
1039 }
1040
1041 /* Copy the XID from the skb... */
1042 xp = skb_header_pointer(skb, sizeof(struct udphdr),
1043 sizeof(_xid), &_xid);
1044 if (xp == NULL)
1045 goto dropit;
1046
1047 /* Look up and lock the request corresponding to the given XID */
1048 spin_lock(&xprt->transport_lock);
1049 rovr = xprt_lookup_rqst(xprt, *xp);
1050 if (!rovr)
1051 goto out_unlock;
1052 task = rovr->rq_task;
1053
1054 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1055 copied = repsize;
1056
1057 /* Suck it into the iovec, verify checksum if not done by hw. */
1058 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1059 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
1060 goto out_unlock;
1061 }
1062
1063 UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
1064
1065 xprt_adjust_cwnd(xprt, task, copied);
1066 xprt_complete_rqst(task, copied);
1067
1068 out_unlock:
1069 spin_unlock(&xprt->transport_lock);
1070 dropit:
1071 skb_free_datagram(sk, skb);
1072 out:
1073 read_unlock_bh(&sk->sk_callback_lock);
1074}
1075
1076/*
1077 * Helper function to force a TCP close if the server is sending
1078 * junk and/or it has put us in CLOSE_WAIT
1079 */
1080static void xs_tcp_force_close(struct rpc_xprt *xprt)
1081{
1082 set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1083 xprt_force_disconnect(xprt);
1084}
1085
1086static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1087{
1088 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1089 size_t len, used;
1090 char *p;
1091
1092 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1093 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1094 used = xdr_skb_read_bits(desc, p, len);
1095 transport->tcp_offset += used;
1096 if (used != len)
1097 return;
1098
1099 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1100 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1101 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1102 else
1103 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1104 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1105
1106 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1107 transport->tcp_offset = 0;
1108
1109 /* Sanity check of the record length */
1110 if (unlikely(transport->tcp_reclen < 8)) {
1111 dprintk("RPC: invalid TCP record fragment length\n");
1112 xs_tcp_force_close(xprt);
1113 return;
1114 }
1115 dprintk("RPC: reading TCP record fragment of length %d\n",
1116 transport->tcp_reclen);
1117}
1118
1119static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1120{
1121 if (transport->tcp_offset == transport->tcp_reclen) {
1122 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1123 transport->tcp_offset = 0;
1124 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1125 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1126 transport->tcp_flags |= TCP_RCV_COPY_XID;
1127 transport->tcp_copied = 0;
1128 }
1129 }
1130}
1131
1132static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1133{
1134 size_t len, used;
1135 char *p;
1136
1137 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1138 dprintk("RPC: reading XID (%Zu bytes)\n", len);
1139 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1140 used = xdr_skb_read_bits(desc, p, len);
1141 transport->tcp_offset += used;
1142 if (used != len)
1143 return;
1144 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1145 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1146 transport->tcp_copied = 4;
1147 dprintk("RPC: reading %s XID %08x\n",
1148 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1149 : "request with",
1150 ntohl(transport->tcp_xid));
1151 xs_tcp_check_fraghdr(transport);
1152}
1153
1154static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1155 struct xdr_skb_reader *desc)
1156{
1157 size_t len, used;
1158 u32 offset;
1159 char *p;
1160
1161 /*
1162 * We want transport->tcp_offset to be 8 at the end of this routine
1163 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1164 * When this function is called for the first time,
1165 * transport->tcp_offset is 4 (after having already read the xid).
1166 */
1167 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1168 len = sizeof(transport->tcp_calldir) - offset;
1169 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len);
1170 p = ((char *) &transport->tcp_calldir) + offset;
1171 used = xdr_skb_read_bits(desc, p, len);
1172 transport->tcp_offset += used;
1173 if (used != len)
1174 return;
1175 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1176 /*
1177 * We don't yet have the XDR buffer, so we will write the calldir
1178 * out after we get the buffer from the 'struct rpc_rqst'
1179 */
1180 switch (ntohl(transport->tcp_calldir)) {
1181 case RPC_REPLY:
1182 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1183 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1184 transport->tcp_flags |= TCP_RPC_REPLY;
1185 break;
1186 case RPC_CALL:
1187 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1188 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1189 transport->tcp_flags &= ~TCP_RPC_REPLY;
1190 break;
1191 default:
1192 dprintk("RPC: invalid request message type\n");
1193 xs_tcp_force_close(&transport->xprt);
1194 }
1195 xs_tcp_check_fraghdr(transport);
1196}
1197
1198static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1199 struct xdr_skb_reader *desc,
1200 struct rpc_rqst *req)
1201{
1202 struct sock_xprt *transport =
1203 container_of(xprt, struct sock_xprt, xprt);
1204 struct xdr_buf *rcvbuf;
1205 size_t len;
1206 ssize_t r;
1207
1208 rcvbuf = &req->rq_private_buf;
1209
1210 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1211 /*
1212 * Save the RPC direction in the XDR buffer
1213 */
1214 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1215 &transport->tcp_calldir,
1216 sizeof(transport->tcp_calldir));
1217 transport->tcp_copied += sizeof(transport->tcp_calldir);
1218 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1219 }
1220
1221 len = desc->count;
1222 if (len > transport->tcp_reclen - transport->tcp_offset) {
1223 struct xdr_skb_reader my_desc;
1224
1225 len = transport->tcp_reclen - transport->tcp_offset;
1226 memcpy(&my_desc, desc, sizeof(my_desc));
1227 my_desc.count = len;
1228 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1229 &my_desc, xdr_skb_read_bits);
1230 desc->count -= r;
1231 desc->offset += r;
1232 } else
1233 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1234 desc, xdr_skb_read_bits);
1235
1236 if (r > 0) {
1237 transport->tcp_copied += r;
1238 transport->tcp_offset += r;
1239 }
1240 if (r != len) {
1241 /* Error when copying to the receive buffer,
1242 * usually because we weren't able to allocate
1243 * additional buffer pages. All we can do now
1244 * is turn off TCP_RCV_COPY_DATA, so the request
1245 * will not receive any additional updates,
1246 * and time out.
1247 * Any remaining data from this record will
1248 * be discarded.
1249 */
1250 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1251 dprintk("RPC: XID %08x truncated request\n",
1252 ntohl(transport->tcp_xid));
1253 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1254 "tcp_offset = %u, tcp_reclen = %u\n",
1255 xprt, transport->tcp_copied,
1256 transport->tcp_offset, transport->tcp_reclen);
1257 return;
1258 }
1259
1260 dprintk("RPC: XID %08x read %Zd bytes\n",
1261 ntohl(transport->tcp_xid), r);
1262 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1263 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1264 transport->tcp_offset, transport->tcp_reclen);
1265
1266 if (transport->tcp_copied == req->rq_private_buf.buflen)
1267 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1268 else if (transport->tcp_offset == transport->tcp_reclen) {
1269 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1270 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1271 }
1272}
1273
1274/*
1275 * Finds the request corresponding to the RPC xid and invokes the common
1276 * tcp read code to read the data.
1277 */
1278static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1279 struct xdr_skb_reader *desc)
1280{
1281 struct sock_xprt *transport =
1282 container_of(xprt, struct sock_xprt, xprt);
1283 struct rpc_rqst *req;
1284
1285 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1286
1287 /* Find and lock the request corresponding to this xid */
1288 spin_lock(&xprt->transport_lock);
1289 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1290 if (!req) {
1291 dprintk("RPC: XID %08x request not found!\n",
1292 ntohl(transport->tcp_xid));
1293 spin_unlock(&xprt->transport_lock);
1294 return -1;
1295 }
1296
1297 xs_tcp_read_common(xprt, desc, req);
1298
1299 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1300 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1301
1302 spin_unlock(&xprt->transport_lock);
1303 return 0;
1304}
1305
1306#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1307/*
1308 * Obtains an rpc_rqst previously allocated and invokes the common
1309 * tcp read code to read the data. The result is placed in the callback
1310 * queue.
1311 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1312 * connection and return -1.
1313 */
1314static int xs_tcp_read_callback(struct rpc_xprt *xprt,
1315 struct xdr_skb_reader *desc)
1316{
1317 struct sock_xprt *transport =
1318 container_of(xprt, struct sock_xprt, xprt);
1319 struct rpc_rqst *req;
1320
1321 /* Look up and lock the request corresponding to the given XID */
1322 spin_lock(&xprt->transport_lock);
1323 req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
1324 if (req == NULL) {
1325 spin_unlock(&xprt->transport_lock);
1326 printk(KERN_WARNING "Callback slot table overflowed\n");
1327 xprt_force_disconnect(xprt);
1328 return -1;
1329 }
1330
1331 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1332 xs_tcp_read_common(xprt, desc, req);
1333
1334 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1335 xprt_complete_bc_request(req, transport->tcp_copied);
1336 spin_unlock(&xprt->transport_lock);
1337
1338 return 0;
1339}
1340
1341static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1342 struct xdr_skb_reader *desc)
1343{
1344 struct sock_xprt *transport =
1345 container_of(xprt, struct sock_xprt, xprt);
1346
1347 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1348 xs_tcp_read_reply(xprt, desc) :
1349 xs_tcp_read_callback(xprt, desc);
1350}
1351#else
1352static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1353 struct xdr_skb_reader *desc)
1354{
1355 return xs_tcp_read_reply(xprt, desc);
1356}
1357#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1358
1359/*
1360 * Read data off the transport. This can be either an RPC_CALL or an
1361 * RPC_REPLY. Relay the processing to helper functions.
1362 */
1363static void xs_tcp_read_data(struct rpc_xprt *xprt,
1364 struct xdr_skb_reader *desc)
1365{
1366 struct sock_xprt *transport =
1367 container_of(xprt, struct sock_xprt, xprt);
1368
1369 if (_xs_tcp_read_data(xprt, desc) == 0)
1370 xs_tcp_check_fraghdr(transport);
1371 else {
1372 /*
1373 * The transport_lock protects the request handling.
1374 * There's no need to hold it to update the tcp_flags.
1375 */
1376 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1377 }
1378}
1379
1380static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1381{
1382 size_t len;
1383
1384 len = transport->tcp_reclen - transport->tcp_offset;
1385 if (len > desc->count)
1386 len = desc->count;
1387 desc->count -= len;
1388 desc->offset += len;
1389 transport->tcp_offset += len;
1390 dprintk("RPC: discarded %Zu bytes\n", len);
1391 xs_tcp_check_fraghdr(transport);
1392}
1393
1394static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1395{
1396 struct rpc_xprt *xprt = rd_desc->arg.data;
1397 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1398 struct xdr_skb_reader desc = {
1399 .skb = skb,
1400 .offset = offset,
1401 .count = len,
1402 };
1403
1404 dprintk("RPC: xs_tcp_data_recv started\n");
1405 do {
1406 /* Read in a new fragment marker if necessary */
1407 /* Can we ever really expect to get completely empty fragments? */
1408 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1409 xs_tcp_read_fraghdr(xprt, &desc);
1410 continue;
1411 }
1412 /* Read in the xid if necessary */
1413 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1414 xs_tcp_read_xid(transport, &desc);
1415 continue;
1416 }
1417 /* Read in the call/reply flag */
1418 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1419 xs_tcp_read_calldir(transport, &desc);
1420 continue;
1421 }
1422 /* Read in the request data */
1423 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1424 xs_tcp_read_data(xprt, &desc);
1425 continue;
1426 }
1427 /* Skip over any trailing bytes on short reads */
1428 xs_tcp_read_discard(transport, &desc);
1429 } while (desc.count);
1430 dprintk("RPC: xs_tcp_data_recv done\n");
1431 return len - desc.count;
1432}
1433
1434/**
1435 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1436 * @sk: socket with data to read
1437 * @bytes: how much data to read
1438 *
1439 */
1440static void xs_tcp_data_ready(struct sock *sk)
1441{
1442 struct rpc_xprt *xprt;
1443 read_descriptor_t rd_desc;
1444 int read;
1445
1446 dprintk("RPC: xs_tcp_data_ready...\n");
1447
1448 read_lock_bh(&sk->sk_callback_lock);
1449 if (!(xprt = xprt_from_sock(sk)))
1450 goto out;
1451 /* Any data means we had a useful conversation, so
1452 * the we don't need to delay the next reconnect
1453 */
1454 if (xprt->reestablish_timeout)
1455 xprt->reestablish_timeout = 0;
1456
1457 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1458 rd_desc.arg.data = xprt;
1459 do {
1460 rd_desc.count = 65536;
1461 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1462 } while (read > 0);
1463out:
1464 read_unlock_bh(&sk->sk_callback_lock);
1465}
1466
1467/*
1468 * Do the equivalent of linger/linger2 handling for dealing with
1469 * broken servers that don't close the socket in a timely
1470 * fashion
1471 */
1472static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1473 unsigned long timeout)
1474{
1475 struct sock_xprt *transport;
1476
1477 if (xprt_test_and_set_connecting(xprt))
1478 return;
1479 set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1480 transport = container_of(xprt, struct sock_xprt, xprt);
1481 queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1482 timeout);
1483}
1484
1485static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1486{
1487 struct sock_xprt *transport;
1488
1489 transport = container_of(xprt, struct sock_xprt, xprt);
1490
1491 if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1492 !cancel_delayed_work(&transport->connect_worker))
1493 return;
1494 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1495 xprt_clear_connecting(xprt);
1496}
1497
1498static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1499{
1500 smp_mb__before_clear_bit();
1501 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1502 clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1503 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1504 clear_bit(XPRT_CLOSING, &xprt->state);
1505 smp_mb__after_clear_bit();
1506}
1507
1508static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1509{
1510 xs_sock_reset_connection_flags(xprt);
1511 /* Mark transport as closed and wake up all pending tasks */
1512 xprt_disconnect_done(xprt);
1513}
1514
1515/**
1516 * xs_tcp_state_change - callback to handle TCP socket state changes
1517 * @sk: socket whose state has changed
1518 *
1519 */
1520static void xs_tcp_state_change(struct sock *sk)
1521{
1522 struct rpc_xprt *xprt;
1523
1524 read_lock_bh(&sk->sk_callback_lock);
1525 if (!(xprt = xprt_from_sock(sk)))
1526 goto out;
1527 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1528 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1529 sk->sk_state, xprt_connected(xprt),
1530 sock_flag(sk, SOCK_DEAD),
1531 sock_flag(sk, SOCK_ZAPPED),
1532 sk->sk_shutdown);
1533
1534 trace_rpc_socket_state_change(xprt, sk->sk_socket);
1535 switch (sk->sk_state) {
1536 case TCP_ESTABLISHED:
1537 spin_lock(&xprt->transport_lock);
1538 if (!xprt_test_and_set_connected(xprt)) {
1539 struct sock_xprt *transport = container_of(xprt,
1540 struct sock_xprt, xprt);
1541
1542 /* Reset TCP record info */
1543 transport->tcp_offset = 0;
1544 transport->tcp_reclen = 0;
1545 transport->tcp_copied = 0;
1546 transport->tcp_flags =
1547 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1548 xprt->connect_cookie++;
1549
1550 xprt_wake_pending_tasks(xprt, -EAGAIN);
1551 }
1552 spin_unlock(&xprt->transport_lock);
1553 break;
1554 case TCP_FIN_WAIT1:
1555 /* The client initiated a shutdown of the socket */
1556 xprt->connect_cookie++;
1557 xprt->reestablish_timeout = 0;
1558 set_bit(XPRT_CLOSING, &xprt->state);
1559 smp_mb__before_clear_bit();
1560 clear_bit(XPRT_CONNECTED, &xprt->state);
1561 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1562 smp_mb__after_clear_bit();
1563 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1564 break;
1565 case TCP_CLOSE_WAIT:
1566 /* The server initiated a shutdown of the socket */
1567 xprt->connect_cookie++;
1568 clear_bit(XPRT_CONNECTED, &xprt->state);
1569 xs_tcp_force_close(xprt);
1570 case TCP_CLOSING:
1571 /*
1572 * If the server closed down the connection, make sure that
1573 * we back off before reconnecting
1574 */
1575 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1576 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1577 break;
1578 case TCP_LAST_ACK:
1579 set_bit(XPRT_CLOSING, &xprt->state);
1580 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1581 smp_mb__before_clear_bit();
1582 clear_bit(XPRT_CONNECTED, &xprt->state);
1583 smp_mb__after_clear_bit();
1584 break;
1585 case TCP_CLOSE:
1586 xs_tcp_cancel_linger_timeout(xprt);
1587 xs_sock_mark_closed(xprt);
1588 }
1589 out:
1590 read_unlock_bh(&sk->sk_callback_lock);
1591}
1592
1593static void xs_write_space(struct sock *sk)
1594{
1595 struct socket *sock;
1596 struct rpc_xprt *xprt;
1597
1598 if (unlikely(!(sock = sk->sk_socket)))
1599 return;
1600 clear_bit(SOCK_NOSPACE, &sock->flags);
1601
1602 if (unlikely(!(xprt = xprt_from_sock(sk))))
1603 return;
1604 if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1605 return;
1606
1607 xprt_write_space(xprt);
1608}
1609
1610/**
1611 * xs_udp_write_space - callback invoked when socket buffer space
1612 * becomes available
1613 * @sk: socket whose state has changed
1614 *
1615 * Called when more output buffer space is available for this socket.
1616 * We try not to wake our writers until they can make "significant"
1617 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1618 * with a bunch of small requests.
1619 */
1620static void xs_udp_write_space(struct sock *sk)
1621{
1622 read_lock_bh(&sk->sk_callback_lock);
1623
1624 /* from net/core/sock.c:sock_def_write_space */
1625 if (sock_writeable(sk))
1626 xs_write_space(sk);
1627
1628 read_unlock_bh(&sk->sk_callback_lock);
1629}
1630
1631/**
1632 * xs_tcp_write_space - callback invoked when socket buffer space
1633 * becomes available
1634 * @sk: socket whose state has changed
1635 *
1636 * Called when more output buffer space is available for this socket.
1637 * We try not to wake our writers until they can make "significant"
1638 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1639 * with a bunch of small requests.
1640 */
1641static void xs_tcp_write_space(struct sock *sk)
1642{
1643 read_lock_bh(&sk->sk_callback_lock);
1644
1645 /* from net/core/stream.c:sk_stream_write_space */
1646 if (sk_stream_is_writeable(sk))
1647 xs_write_space(sk);
1648
1649 read_unlock_bh(&sk->sk_callback_lock);
1650}
1651
1652static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1653{
1654 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1655 struct sock *sk = transport->inet;
1656
1657 if (transport->rcvsize) {
1658 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1659 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1660 }
1661 if (transport->sndsize) {
1662 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1663 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1664 sk->sk_write_space(sk);
1665 }
1666}
1667
1668/**
1669 * xs_udp_set_buffer_size - set send and receive limits
1670 * @xprt: generic transport
1671 * @sndsize: requested size of send buffer, in bytes
1672 * @rcvsize: requested size of receive buffer, in bytes
1673 *
1674 * Set socket send and receive buffer size limits.
1675 */
1676static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1677{
1678 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1679
1680 transport->sndsize = 0;
1681 if (sndsize)
1682 transport->sndsize = sndsize + 1024;
1683 transport->rcvsize = 0;
1684 if (rcvsize)
1685 transport->rcvsize = rcvsize + 1024;
1686
1687 xs_udp_do_set_buffer_size(xprt);
1688}
1689
1690/**
1691 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1692 * @task: task that timed out
1693 *
1694 * Adjust the congestion window after a retransmit timeout has occurred.
1695 */
1696static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1697{
1698 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1699}
1700
1701static unsigned short xs_get_random_port(void)
1702{
1703 unsigned short range = xprt_max_resvport - xprt_min_resvport;
1704 unsigned short rand = (unsigned short) prandom_u32() % range;
1705 return rand + xprt_min_resvport;
1706}
1707
1708/**
1709 * xs_set_port - reset the port number in the remote endpoint address
1710 * @xprt: generic transport
1711 * @port: new port number
1712 *
1713 */
1714static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1715{
1716 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1717
1718 rpc_set_port(xs_addr(xprt), port);
1719 xs_update_peer_port(xprt);
1720}
1721
1722static unsigned short xs_get_srcport(struct sock_xprt *transport)
1723{
1724 unsigned short port = transport->srcport;
1725
1726 if (port == 0 && transport->xprt.resvport)
1727 port = xs_get_random_port();
1728 return port;
1729}
1730
1731static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1732{
1733 if (transport->srcport != 0)
1734 transport->srcport = 0;
1735 if (!transport->xprt.resvport)
1736 return 0;
1737 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1738 return xprt_max_resvport;
1739 return --port;
1740}
1741static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1742{
1743 struct sockaddr_storage myaddr;
1744 int err, nloop = 0;
1745 unsigned short port = xs_get_srcport(transport);
1746 unsigned short last;
1747
1748 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1749 do {
1750 rpc_set_port((struct sockaddr *)&myaddr, port);
1751 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1752 transport->xprt.addrlen);
1753 if (port == 0)
1754 break;
1755 if (err == 0) {
1756 transport->srcport = port;
1757 break;
1758 }
1759 last = port;
1760 port = xs_next_srcport(transport, port);
1761 if (port > last)
1762 nloop++;
1763 } while (err == -EADDRINUSE && nloop != 2);
1764
1765 if (myaddr.ss_family == AF_INET)
1766 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1767 &((struct sockaddr_in *)&myaddr)->sin_addr,
1768 port, err ? "failed" : "ok", err);
1769 else
1770 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1771 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1772 port, err ? "failed" : "ok", err);
1773 return err;
1774}
1775
1776/*
1777 * We don't support autobind on AF_LOCAL sockets
1778 */
1779static void xs_local_rpcbind(struct rpc_task *task)
1780{
1781 rcu_read_lock();
1782 xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt));
1783 rcu_read_unlock();
1784}
1785
1786static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1787{
1788}
1789
1790#ifdef CONFIG_DEBUG_LOCK_ALLOC
1791static struct lock_class_key xs_key[2];
1792static struct lock_class_key xs_slock_key[2];
1793
1794static inline void xs_reclassify_socketu(struct socket *sock)
1795{
1796 struct sock *sk = sock->sk;
1797
1798 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1799 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1800}
1801
1802static inline void xs_reclassify_socket4(struct socket *sock)
1803{
1804 struct sock *sk = sock->sk;
1805
1806 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1807 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1808}
1809
1810static inline void xs_reclassify_socket6(struct socket *sock)
1811{
1812 struct sock *sk = sock->sk;
1813
1814 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1815 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1816}
1817
1818static inline void xs_reclassify_socket(int family, struct socket *sock)
1819{
1820 WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1821 if (sock_owned_by_user(sock->sk))
1822 return;
1823
1824 switch (family) {
1825 case AF_LOCAL:
1826 xs_reclassify_socketu(sock);
1827 break;
1828 case AF_INET:
1829 xs_reclassify_socket4(sock);
1830 break;
1831 case AF_INET6:
1832 xs_reclassify_socket6(sock);
1833 break;
1834 }
1835}
1836#else
1837static inline void xs_reclassify_socketu(struct socket *sock)
1838{
1839}
1840
1841static inline void xs_reclassify_socket4(struct socket *sock)
1842{
1843}
1844
1845static inline void xs_reclassify_socket6(struct socket *sock)
1846{
1847}
1848
1849static inline void xs_reclassify_socket(int family, struct socket *sock)
1850{
1851}
1852#endif
1853
1854static void xs_dummy_setup_socket(struct work_struct *work)
1855{
1856}
1857
1858static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1859 struct sock_xprt *transport, int family, int type, int protocol)
1860{
1861 struct socket *sock;
1862 int err;
1863
1864 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1865 if (err < 0) {
1866 dprintk("RPC: can't create %d transport socket (%d).\n",
1867 protocol, -err);
1868 goto out;
1869 }
1870 xs_reclassify_socket(family, sock);
1871
1872 err = xs_bind(transport, sock);
1873 if (err) {
1874 sock_release(sock);
1875 goto out;
1876 }
1877
1878 return sock;
1879out:
1880 return ERR_PTR(err);
1881}
1882
1883static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1884 struct socket *sock)
1885{
1886 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1887 xprt);
1888
1889 if (!transport->inet) {
1890 struct sock *sk = sock->sk;
1891
1892 write_lock_bh(&sk->sk_callback_lock);
1893
1894 xs_save_old_callbacks(transport, sk);
1895
1896 sk->sk_user_data = xprt;
1897 sk->sk_data_ready = xs_local_data_ready;
1898 sk->sk_write_space = xs_udp_write_space;
1899 sk->sk_error_report = xs_error_report;
1900 sk->sk_allocation = GFP_ATOMIC;
1901
1902 xprt_clear_connected(xprt);
1903
1904 /* Reset to new socket */
1905 transport->sock = sock;
1906 transport->inet = sk;
1907
1908 write_unlock_bh(&sk->sk_callback_lock);
1909 }
1910
1911 /* Tell the socket layer to start connecting... */
1912 xprt->stat.connect_count++;
1913 xprt->stat.connect_start = jiffies;
1914 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1915}
1916
1917/**
1918 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1919 * @xprt: RPC transport to connect
1920 * @transport: socket transport to connect
1921 * @create_sock: function to create a socket of the correct type
1922 */
1923static int xs_local_setup_socket(struct sock_xprt *transport)
1924{
1925 struct rpc_xprt *xprt = &transport->xprt;
1926 struct socket *sock;
1927 int status = -EIO;
1928
1929 current->flags |= PF_FSTRANS;
1930
1931 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1932 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1933 SOCK_STREAM, 0, &sock, 1);
1934 if (status < 0) {
1935 dprintk("RPC: can't create AF_LOCAL "
1936 "transport socket (%d).\n", -status);
1937 goto out;
1938 }
1939 xs_reclassify_socketu(sock);
1940
1941 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1942 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1943
1944 status = xs_local_finish_connecting(xprt, sock);
1945 trace_rpc_socket_connect(xprt, sock, status);
1946 switch (status) {
1947 case 0:
1948 dprintk("RPC: xprt %p connected to %s\n",
1949 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1950 xprt_set_connected(xprt);
1951 break;
1952 case -ENOENT:
1953 dprintk("RPC: xprt %p: socket %s does not exist\n",
1954 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1955 break;
1956 case -ECONNREFUSED:
1957 dprintk("RPC: xprt %p: connection refused for %s\n",
1958 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1959 break;
1960 default:
1961 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1962 __func__, -status,
1963 xprt->address_strings[RPC_DISPLAY_ADDR]);
1964 }
1965
1966out:
1967 xprt_clear_connecting(xprt);
1968 xprt_wake_pending_tasks(xprt, status);
1969 current->flags &= ~PF_FSTRANS;
1970 return status;
1971}
1972
1973static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1974{
1975 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1976 int ret;
1977
1978 if (RPC_IS_ASYNC(task)) {
1979 /*
1980 * We want the AF_LOCAL connect to be resolved in the
1981 * filesystem namespace of the process making the rpc
1982 * call. Thus we connect synchronously.
1983 *
1984 * If we want to support asynchronous AF_LOCAL calls,
1985 * we'll need to figure out how to pass a namespace to
1986 * connect.
1987 */
1988 rpc_exit(task, -ENOTCONN);
1989 return;
1990 }
1991 ret = xs_local_setup_socket(transport);
1992 if (ret && !RPC_IS_SOFTCONN(task))
1993 msleep_interruptible(15000);
1994}
1995
1996#ifdef CONFIG_SUNRPC_SWAP
1997static void xs_set_memalloc(struct rpc_xprt *xprt)
1998{
1999 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2000 xprt);
2001
2002 if (xprt->swapper)
2003 sk_set_memalloc(transport->inet);
2004}
2005
2006/**
2007 * xs_swapper - Tag this transport as being used for swap.
2008 * @xprt: transport to tag
2009 * @enable: enable/disable
2010 *
2011 */
2012int xs_swapper(struct rpc_xprt *xprt, int enable)
2013{
2014 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2015 xprt);
2016 int err = 0;
2017
2018 if (enable) {
2019 xprt->swapper++;
2020 xs_set_memalloc(xprt);
2021 } else if (xprt->swapper) {
2022 xprt->swapper--;
2023 sk_clear_memalloc(transport->inet);
2024 }
2025
2026 return err;
2027}
2028EXPORT_SYMBOL_GPL(xs_swapper);
2029#else
2030static void xs_set_memalloc(struct rpc_xprt *xprt)
2031{
2032}
2033#endif
2034
2035static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2036{
2037 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2038
2039 if (!transport->inet) {
2040 struct sock *sk = sock->sk;
2041
2042 write_lock_bh(&sk->sk_callback_lock);
2043
2044 xs_save_old_callbacks(transport, sk);
2045
2046 sk->sk_user_data = xprt;
2047 sk->sk_data_ready = xs_udp_data_ready;
2048 sk->sk_write_space = xs_udp_write_space;
2049 sk->sk_no_check = UDP_CSUM_NORCV;
2050 sk->sk_allocation = GFP_ATOMIC;
2051
2052 xprt_set_connected(xprt);
2053
2054 /* Reset to new socket */
2055 transport->sock = sock;
2056 transport->inet = sk;
2057
2058 xs_set_memalloc(xprt);
2059
2060 write_unlock_bh(&sk->sk_callback_lock);
2061 }
2062 xs_udp_do_set_buffer_size(xprt);
2063}
2064
2065static void xs_udp_setup_socket(struct work_struct *work)
2066{
2067 struct sock_xprt *transport =
2068 container_of(work, struct sock_xprt, connect_worker.work);
2069 struct rpc_xprt *xprt = &transport->xprt;
2070 struct socket *sock = transport->sock;
2071 int status = -EIO;
2072
2073 current->flags |= PF_FSTRANS;
2074
2075 /* Start by resetting any existing state */
2076 xs_reset_transport(transport);
2077 sock = xs_create_sock(xprt, transport,
2078 xs_addr(xprt)->sa_family, SOCK_DGRAM, IPPROTO_UDP);
2079 if (IS_ERR(sock))
2080 goto out;
2081
2082 dprintk("RPC: worker connecting xprt %p via %s to "
2083 "%s (port %s)\n", xprt,
2084 xprt->address_strings[RPC_DISPLAY_PROTO],
2085 xprt->address_strings[RPC_DISPLAY_ADDR],
2086 xprt->address_strings[RPC_DISPLAY_PORT]);
2087
2088 xs_udp_finish_connecting(xprt, sock);
2089 trace_rpc_socket_connect(xprt, sock, 0);
2090 status = 0;
2091out:
2092 xprt_clear_connecting(xprt);
2093 xprt_wake_pending_tasks(xprt, status);
2094 current->flags &= ~PF_FSTRANS;
2095}
2096
2097/*
2098 * We need to preserve the port number so the reply cache on the server can
2099 * find our cached RPC replies when we get around to reconnecting.
2100 */
2101static void xs_abort_connection(struct sock_xprt *transport)
2102{
2103 int result;
2104 struct sockaddr any;
2105
2106 dprintk("RPC: disconnecting xprt %p to reuse port\n", transport);
2107
2108 /*
2109 * Disconnect the transport socket by doing a connect operation
2110 * with AF_UNSPEC. This should return immediately...
2111 */
2112 memset(&any, 0, sizeof(any));
2113 any.sa_family = AF_UNSPEC;
2114 result = kernel_connect(transport->sock, &any, sizeof(any), 0);
2115 trace_rpc_socket_reset_connection(&transport->xprt,
2116 transport->sock, result);
2117 if (!result)
2118 xs_sock_reset_connection_flags(&transport->xprt);
2119 dprintk("RPC: AF_UNSPEC connect return code %d\n", result);
2120}
2121
2122static void xs_tcp_reuse_connection(struct sock_xprt *transport)
2123{
2124 unsigned int state = transport->inet->sk_state;
2125
2126 if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED) {
2127 /* we don't need to abort the connection if the socket
2128 * hasn't undergone a shutdown
2129 */
2130 if (transport->inet->sk_shutdown == 0)
2131 return;
2132 dprintk("RPC: %s: TCP_CLOSEd and sk_shutdown set to %d\n",
2133 __func__, transport->inet->sk_shutdown);
2134 }
2135 if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT)) {
2136 /* we don't need to abort the connection if the socket
2137 * hasn't undergone a shutdown
2138 */
2139 if (transport->inet->sk_shutdown == 0)
2140 return;
2141 dprintk("RPC: %s: ESTABLISHED/SYN_SENT "
2142 "sk_shutdown set to %d\n",
2143 __func__, transport->inet->sk_shutdown);
2144 }
2145 xs_abort_connection(transport);
2146}
2147
2148static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2149{
2150 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2151 int ret = -ENOTCONN;
2152
2153 if (!transport->inet) {
2154 struct sock *sk = sock->sk;
2155 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2156 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2157 unsigned int opt_on = 1;
2158
2159 /* TCP Keepalive options */
2160 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2161 (char *)&opt_on, sizeof(opt_on));
2162 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2163 (char *)&keepidle, sizeof(keepidle));
2164 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2165 (char *)&keepidle, sizeof(keepidle));
2166 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2167 (char *)&keepcnt, sizeof(keepcnt));
2168
2169 write_lock_bh(&sk->sk_callback_lock);
2170
2171 xs_save_old_callbacks(transport, sk);
2172
2173 sk->sk_user_data = xprt;
2174 sk->sk_data_ready = xs_tcp_data_ready;
2175 sk->sk_state_change = xs_tcp_state_change;
2176 sk->sk_write_space = xs_tcp_write_space;
2177 sk->sk_error_report = xs_error_report;
2178 sk->sk_allocation = GFP_ATOMIC;
2179
2180 /* socket options */
2181 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
2182 sock_reset_flag(sk, SOCK_LINGER);
2183 tcp_sk(sk)->linger2 = 0;
2184 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2185
2186 xprt_clear_connected(xprt);
2187
2188 /* Reset to new socket */
2189 transport->sock = sock;
2190 transport->inet = sk;
2191
2192 write_unlock_bh(&sk->sk_callback_lock);
2193 }
2194
2195 if (!xprt_bound(xprt))
2196 goto out;
2197
2198 xs_set_memalloc(xprt);
2199
2200 /* Tell the socket layer to start connecting... */
2201 xprt->stat.connect_count++;
2202 xprt->stat.connect_start = jiffies;
2203 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2204 switch (ret) {
2205 case 0:
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 }
2211out:
2212 return ret;
2213}
2214
2215/**
2216 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2217 * @xprt: RPC transport to connect
2218 * @transport: socket transport to connect
2219 * @create_sock: function to create a socket of the correct type
2220 *
2221 * Invoked by a work queue tasklet.
2222 */
2223static void xs_tcp_setup_socket(struct work_struct *work)
2224{
2225 struct sock_xprt *transport =
2226 container_of(work, struct sock_xprt, connect_worker.work);
2227 struct socket *sock = transport->sock;
2228 struct rpc_xprt *xprt = &transport->xprt;
2229 int status = -EIO;
2230
2231 current->flags |= PF_FSTRANS;
2232
2233 if (!sock) {
2234 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
2235 sock = xs_create_sock(xprt, transport,
2236 xs_addr(xprt)->sa_family, SOCK_STREAM, IPPROTO_TCP);
2237 if (IS_ERR(sock)) {
2238 status = PTR_ERR(sock);
2239 goto out;
2240 }
2241 } else {
2242 int abort_and_exit;
2243
2244 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
2245 &xprt->state);
2246 /* "close" the socket, preserving the local port */
2247 xs_tcp_reuse_connection(transport);
2248
2249 if (abort_and_exit)
2250 goto out_eagain;
2251 }
2252
2253 dprintk("RPC: worker connecting xprt %p via %s to "
2254 "%s (port %s)\n", xprt,
2255 xprt->address_strings[RPC_DISPLAY_PROTO],
2256 xprt->address_strings[RPC_DISPLAY_ADDR],
2257 xprt->address_strings[RPC_DISPLAY_PORT]);
2258
2259 status = xs_tcp_finish_connecting(xprt, sock);
2260 trace_rpc_socket_connect(xprt, sock, status);
2261 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2262 xprt, -status, xprt_connected(xprt),
2263 sock->sk->sk_state);
2264 switch (status) {
2265 default:
2266 printk("%s: connect returned unhandled error %d\n",
2267 __func__, status);
2268 case -EADDRNOTAVAIL:
2269 /* We're probably in TIME_WAIT. Get rid of existing socket,
2270 * and retry
2271 */
2272 xs_tcp_force_close(xprt);
2273 break;
2274 case 0:
2275 case -EINPROGRESS:
2276 case -EALREADY:
2277 xprt_clear_connecting(xprt);
2278 current->flags &= ~PF_FSTRANS;
2279 return;
2280 case -EINVAL:
2281 /* Happens, for instance, if the user specified a link
2282 * local IPv6 address without a scope-id.
2283 */
2284 case -ECONNREFUSED:
2285 case -ECONNRESET:
2286 case -ENETUNREACH:
2287 /* retry with existing socket, after a delay */
2288 goto out;
2289 }
2290out_eagain:
2291 status = -EAGAIN;
2292out:
2293 xprt_clear_connecting(xprt);
2294 xprt_wake_pending_tasks(xprt, status);
2295 current->flags &= ~PF_FSTRANS;
2296}
2297
2298/**
2299 * xs_connect - connect a socket to a remote endpoint
2300 * @xprt: pointer to transport structure
2301 * @task: address of RPC task that manages state of connect request
2302 *
2303 * TCP: If the remote end dropped the connection, delay reconnecting.
2304 *
2305 * UDP socket connects are synchronous, but we use a work queue anyway
2306 * to guarantee that even unprivileged user processes can set up a
2307 * socket on a privileged port.
2308 *
2309 * If a UDP socket connect fails, the delay behavior here prevents
2310 * retry floods (hard mounts).
2311 */
2312static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2313{
2314 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2315
2316 if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
2317 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2318 "seconds\n",
2319 xprt, xprt->reestablish_timeout / HZ);
2320 queue_delayed_work(rpciod_workqueue,
2321 &transport->connect_worker,
2322 xprt->reestablish_timeout);
2323 xprt->reestablish_timeout <<= 1;
2324 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2325 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2326 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2327 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2328 } else {
2329 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2330 queue_delayed_work(rpciod_workqueue,
2331 &transport->connect_worker, 0);
2332 }
2333}
2334
2335/**
2336 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2337 * @xprt: rpc_xprt struct containing statistics
2338 * @seq: output file
2339 *
2340 */
2341static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2342{
2343 long idle_time = 0;
2344
2345 if (xprt_connected(xprt))
2346 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2347
2348 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2349 "%llu %llu %lu %llu %llu\n",
2350 xprt->stat.bind_count,
2351 xprt->stat.connect_count,
2352 xprt->stat.connect_time,
2353 idle_time,
2354 xprt->stat.sends,
2355 xprt->stat.recvs,
2356 xprt->stat.bad_xids,
2357 xprt->stat.req_u,
2358 xprt->stat.bklog_u,
2359 xprt->stat.max_slots,
2360 xprt->stat.sending_u,
2361 xprt->stat.pending_u);
2362}
2363
2364/**
2365 * xs_udp_print_stats - display UDP socket-specifc stats
2366 * @xprt: rpc_xprt struct containing statistics
2367 * @seq: output file
2368 *
2369 */
2370static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2371{
2372 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2373
2374 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2375 "%lu %llu %llu\n",
2376 transport->srcport,
2377 xprt->stat.bind_count,
2378 xprt->stat.sends,
2379 xprt->stat.recvs,
2380 xprt->stat.bad_xids,
2381 xprt->stat.req_u,
2382 xprt->stat.bklog_u,
2383 xprt->stat.max_slots,
2384 xprt->stat.sending_u,
2385 xprt->stat.pending_u);
2386}
2387
2388/**
2389 * xs_tcp_print_stats - display TCP socket-specifc stats
2390 * @xprt: rpc_xprt struct containing statistics
2391 * @seq: output file
2392 *
2393 */
2394static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2395{
2396 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2397 long idle_time = 0;
2398
2399 if (xprt_connected(xprt))
2400 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2401
2402 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2403 "%llu %llu %lu %llu %llu\n",
2404 transport->srcport,
2405 xprt->stat.bind_count,
2406 xprt->stat.connect_count,
2407 xprt->stat.connect_time,
2408 idle_time,
2409 xprt->stat.sends,
2410 xprt->stat.recvs,
2411 xprt->stat.bad_xids,
2412 xprt->stat.req_u,
2413 xprt->stat.bklog_u,
2414 xprt->stat.max_slots,
2415 xprt->stat.sending_u,
2416 xprt->stat.pending_u);
2417}
2418
2419/*
2420 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2421 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2422 * to use the server side send routines.
2423 */
2424static void *bc_malloc(struct rpc_task *task, size_t size)
2425{
2426 struct page *page;
2427 struct rpc_buffer *buf;
2428
2429 WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2430 if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2431 return NULL;
2432
2433 page = alloc_page(GFP_KERNEL);
2434 if (!page)
2435 return NULL;
2436
2437 buf = page_address(page);
2438 buf->len = PAGE_SIZE;
2439
2440 return buf->data;
2441}
2442
2443/*
2444 * Free the space allocated in the bc_alloc routine
2445 */
2446static void bc_free(void *buffer)
2447{
2448 struct rpc_buffer *buf;
2449
2450 if (!buffer)
2451 return;
2452
2453 buf = container_of(buffer, struct rpc_buffer, data);
2454 free_page((unsigned long)buf);
2455}
2456
2457/*
2458 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2459 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2460 */
2461static int bc_sendto(struct rpc_rqst *req)
2462{
2463 int len;
2464 struct xdr_buf *xbufp = &req->rq_snd_buf;
2465 struct rpc_xprt *xprt = req->rq_xprt;
2466 struct sock_xprt *transport =
2467 container_of(xprt, struct sock_xprt, xprt);
2468 struct socket *sock = transport->sock;
2469 unsigned long headoff;
2470 unsigned long tailoff;
2471
2472 xs_encode_stream_record_marker(xbufp);
2473
2474 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2475 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2476 len = svc_send_common(sock, xbufp,
2477 virt_to_page(xbufp->head[0].iov_base), headoff,
2478 xbufp->tail[0].iov_base, tailoff);
2479
2480 if (len != xbufp->len) {
2481 printk(KERN_NOTICE "Error sending entire callback!\n");
2482 len = -EAGAIN;
2483 }
2484
2485 return len;
2486}
2487
2488/*
2489 * The send routine. Borrows from svc_send
2490 */
2491static int bc_send_request(struct rpc_task *task)
2492{
2493 struct rpc_rqst *req = task->tk_rqstp;
2494 struct svc_xprt *xprt;
2495 u32 len;
2496
2497 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2498 /*
2499 * Get the server socket associated with this callback xprt
2500 */
2501 xprt = req->rq_xprt->bc_xprt;
2502
2503 /*
2504 * Grab the mutex to serialize data as the connection is shared
2505 * with the fore channel
2506 */
2507 if (!mutex_trylock(&xprt->xpt_mutex)) {
2508 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2509 if (!mutex_trylock(&xprt->xpt_mutex))
2510 return -EAGAIN;
2511 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2512 }
2513 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2514 len = -ENOTCONN;
2515 else
2516 len = bc_sendto(req);
2517 mutex_unlock(&xprt->xpt_mutex);
2518
2519 if (len > 0)
2520 len = 0;
2521
2522 return len;
2523}
2524
2525/*
2526 * The close routine. Since this is client initiated, we do nothing
2527 */
2528
2529static void bc_close(struct rpc_xprt *xprt)
2530{
2531}
2532
2533/*
2534 * The xprt destroy routine. Again, because this connection is client
2535 * initiated, we do nothing
2536 */
2537
2538static void bc_destroy(struct rpc_xprt *xprt)
2539{
2540 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2541
2542 xs_xprt_free(xprt);
2543 module_put(THIS_MODULE);
2544}
2545
2546static struct rpc_xprt_ops xs_local_ops = {
2547 .reserve_xprt = xprt_reserve_xprt,
2548 .release_xprt = xs_tcp_release_xprt,
2549 .alloc_slot = xprt_alloc_slot,
2550 .rpcbind = xs_local_rpcbind,
2551 .set_port = xs_local_set_port,
2552 .connect = xs_local_connect,
2553 .buf_alloc = rpc_malloc,
2554 .buf_free = rpc_free,
2555 .send_request = xs_local_send_request,
2556 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2557 .close = xs_close,
2558 .destroy = xs_destroy,
2559 .print_stats = xs_local_print_stats,
2560};
2561
2562static struct rpc_xprt_ops xs_udp_ops = {
2563 .set_buffer_size = xs_udp_set_buffer_size,
2564 .reserve_xprt = xprt_reserve_xprt_cong,
2565 .release_xprt = xprt_release_xprt_cong,
2566 .alloc_slot = xprt_alloc_slot,
2567 .rpcbind = rpcb_getport_async,
2568 .set_port = xs_set_port,
2569 .connect = xs_connect,
2570 .buf_alloc = rpc_malloc,
2571 .buf_free = rpc_free,
2572 .send_request = xs_udp_send_request,
2573 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
2574 .timer = xs_udp_timer,
2575 .release_request = xprt_release_rqst_cong,
2576 .close = xs_close,
2577 .destroy = xs_destroy,
2578 .print_stats = xs_udp_print_stats,
2579};
2580
2581static struct rpc_xprt_ops xs_tcp_ops = {
2582 .reserve_xprt = xprt_reserve_xprt,
2583 .release_xprt = xs_tcp_release_xprt,
2584 .alloc_slot = xprt_lock_and_alloc_slot,
2585 .rpcbind = rpcb_getport_async,
2586 .set_port = xs_set_port,
2587 .connect = xs_connect,
2588 .buf_alloc = rpc_malloc,
2589 .buf_free = rpc_free,
2590 .send_request = xs_tcp_send_request,
2591 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2592 .close = xs_tcp_close,
2593 .destroy = xs_destroy,
2594 .print_stats = xs_tcp_print_stats,
2595};
2596
2597/*
2598 * The rpc_xprt_ops for the server backchannel
2599 */
2600
2601static struct rpc_xprt_ops bc_tcp_ops = {
2602 .reserve_xprt = xprt_reserve_xprt,
2603 .release_xprt = xprt_release_xprt,
2604 .alloc_slot = xprt_alloc_slot,
2605 .buf_alloc = bc_malloc,
2606 .buf_free = bc_free,
2607 .send_request = bc_send_request,
2608 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2609 .close = bc_close,
2610 .destroy = bc_destroy,
2611 .print_stats = xs_tcp_print_stats,
2612};
2613
2614static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2615{
2616 static const struct sockaddr_in sin = {
2617 .sin_family = AF_INET,
2618 .sin_addr.s_addr = htonl(INADDR_ANY),
2619 };
2620 static const struct sockaddr_in6 sin6 = {
2621 .sin6_family = AF_INET6,
2622 .sin6_addr = IN6ADDR_ANY_INIT,
2623 };
2624
2625 switch (family) {
2626 case AF_LOCAL:
2627 break;
2628 case AF_INET:
2629 memcpy(sap, &sin, sizeof(sin));
2630 break;
2631 case AF_INET6:
2632 memcpy(sap, &sin6, sizeof(sin6));
2633 break;
2634 default:
2635 dprintk("RPC: %s: Bad address family\n", __func__);
2636 return -EAFNOSUPPORT;
2637 }
2638 return 0;
2639}
2640
2641static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2642 unsigned int slot_table_size,
2643 unsigned int max_slot_table_size)
2644{
2645 struct rpc_xprt *xprt;
2646 struct sock_xprt *new;
2647
2648 if (args->addrlen > sizeof(xprt->addr)) {
2649 dprintk("RPC: xs_setup_xprt: address too large\n");
2650 return ERR_PTR(-EBADF);
2651 }
2652
2653 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2654 max_slot_table_size);
2655 if (xprt == NULL) {
2656 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2657 "rpc_xprt\n");
2658 return ERR_PTR(-ENOMEM);
2659 }
2660
2661 new = container_of(xprt, struct sock_xprt, xprt);
2662 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2663 xprt->addrlen = args->addrlen;
2664 if (args->srcaddr)
2665 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2666 else {
2667 int err;
2668 err = xs_init_anyaddr(args->dstaddr->sa_family,
2669 (struct sockaddr *)&new->srcaddr);
2670 if (err != 0) {
2671 xprt_free(xprt);
2672 return ERR_PTR(err);
2673 }
2674 }
2675
2676 return xprt;
2677}
2678
2679static const struct rpc_timeout xs_local_default_timeout = {
2680 .to_initval = 10 * HZ,
2681 .to_maxval = 10 * HZ,
2682 .to_retries = 2,
2683};
2684
2685/**
2686 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2687 * @args: rpc transport creation arguments
2688 *
2689 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2690 */
2691static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2692{
2693 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2694 struct sock_xprt *transport;
2695 struct rpc_xprt *xprt;
2696 struct rpc_xprt *ret;
2697
2698 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2699 xprt_max_tcp_slot_table_entries);
2700 if (IS_ERR(xprt))
2701 return xprt;
2702 transport = container_of(xprt, struct sock_xprt, xprt);
2703
2704 xprt->prot = 0;
2705 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2706 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2707
2708 xprt->bind_timeout = XS_BIND_TO;
2709 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2710 xprt->idle_timeout = XS_IDLE_DISC_TO;
2711
2712 xprt->ops = &xs_local_ops;
2713 xprt->timeout = &xs_local_default_timeout;
2714
2715 INIT_DELAYED_WORK(&transport->connect_worker,
2716 xs_dummy_setup_socket);
2717
2718 switch (sun->sun_family) {
2719 case AF_LOCAL:
2720 if (sun->sun_path[0] != '/') {
2721 dprintk("RPC: bad AF_LOCAL address: %s\n",
2722 sun->sun_path);
2723 ret = ERR_PTR(-EINVAL);
2724 goto out_err;
2725 }
2726 xprt_set_bound(xprt);
2727 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2728 ret = ERR_PTR(xs_local_setup_socket(transport));
2729 if (ret)
2730 goto out_err;
2731 break;
2732 default:
2733 ret = ERR_PTR(-EAFNOSUPPORT);
2734 goto out_err;
2735 }
2736
2737 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2738 xprt->address_strings[RPC_DISPLAY_ADDR]);
2739
2740 if (try_module_get(THIS_MODULE))
2741 return xprt;
2742 ret = ERR_PTR(-EINVAL);
2743out_err:
2744 xs_xprt_free(xprt);
2745 return ret;
2746}
2747
2748static const struct rpc_timeout xs_udp_default_timeout = {
2749 .to_initval = 5 * HZ,
2750 .to_maxval = 30 * HZ,
2751 .to_increment = 5 * HZ,
2752 .to_retries = 5,
2753};
2754
2755/**
2756 * xs_setup_udp - Set up transport to use a UDP socket
2757 * @args: rpc transport creation arguments
2758 *
2759 */
2760static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2761{
2762 struct sockaddr *addr = args->dstaddr;
2763 struct rpc_xprt *xprt;
2764 struct sock_xprt *transport;
2765 struct rpc_xprt *ret;
2766
2767 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2768 xprt_udp_slot_table_entries);
2769 if (IS_ERR(xprt))
2770 return xprt;
2771 transport = container_of(xprt, struct sock_xprt, xprt);
2772
2773 xprt->prot = IPPROTO_UDP;
2774 xprt->tsh_size = 0;
2775 /* XXX: header size can vary due to auth type, IPv6, etc. */
2776 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2777
2778 xprt->bind_timeout = XS_BIND_TO;
2779 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2780 xprt->idle_timeout = XS_IDLE_DISC_TO;
2781
2782 xprt->ops = &xs_udp_ops;
2783
2784 xprt->timeout = &xs_udp_default_timeout;
2785
2786 switch (addr->sa_family) {
2787 case AF_INET:
2788 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2789 xprt_set_bound(xprt);
2790
2791 INIT_DELAYED_WORK(&transport->connect_worker,
2792 xs_udp_setup_socket);
2793 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2794 break;
2795 case AF_INET6:
2796 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2797 xprt_set_bound(xprt);
2798
2799 INIT_DELAYED_WORK(&transport->connect_worker,
2800 xs_udp_setup_socket);
2801 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2802 break;
2803 default:
2804 ret = ERR_PTR(-EAFNOSUPPORT);
2805 goto out_err;
2806 }
2807
2808 if (xprt_bound(xprt))
2809 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2810 xprt->address_strings[RPC_DISPLAY_ADDR],
2811 xprt->address_strings[RPC_DISPLAY_PORT],
2812 xprt->address_strings[RPC_DISPLAY_PROTO]);
2813 else
2814 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2815 xprt->address_strings[RPC_DISPLAY_ADDR],
2816 xprt->address_strings[RPC_DISPLAY_PROTO]);
2817
2818 if (try_module_get(THIS_MODULE))
2819 return xprt;
2820 ret = ERR_PTR(-EINVAL);
2821out_err:
2822 xs_xprt_free(xprt);
2823 return ret;
2824}
2825
2826static const struct rpc_timeout xs_tcp_default_timeout = {
2827 .to_initval = 60 * HZ,
2828 .to_maxval = 60 * HZ,
2829 .to_retries = 2,
2830};
2831
2832/**
2833 * xs_setup_tcp - Set up transport to use a TCP socket
2834 * @args: rpc transport creation arguments
2835 *
2836 */
2837static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2838{
2839 struct sockaddr *addr = args->dstaddr;
2840 struct rpc_xprt *xprt;
2841 struct sock_xprt *transport;
2842 struct rpc_xprt *ret;
2843 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2844
2845 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2846 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2847
2848 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2849 max_slot_table_size);
2850 if (IS_ERR(xprt))
2851 return xprt;
2852 transport = container_of(xprt, struct sock_xprt, xprt);
2853
2854 xprt->prot = IPPROTO_TCP;
2855 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2856 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2857
2858 xprt->bind_timeout = XS_BIND_TO;
2859 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2860 xprt->idle_timeout = XS_IDLE_DISC_TO;
2861
2862 xprt->ops = &xs_tcp_ops;
2863 xprt->timeout = &xs_tcp_default_timeout;
2864
2865 switch (addr->sa_family) {
2866 case AF_INET:
2867 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2868 xprt_set_bound(xprt);
2869
2870 INIT_DELAYED_WORK(&transport->connect_worker,
2871 xs_tcp_setup_socket);
2872 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2873 break;
2874 case AF_INET6:
2875 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2876 xprt_set_bound(xprt);
2877
2878 INIT_DELAYED_WORK(&transport->connect_worker,
2879 xs_tcp_setup_socket);
2880 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2881 break;
2882 default:
2883 ret = ERR_PTR(-EAFNOSUPPORT);
2884 goto out_err;
2885 }
2886
2887 if (xprt_bound(xprt))
2888 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2889 xprt->address_strings[RPC_DISPLAY_ADDR],
2890 xprt->address_strings[RPC_DISPLAY_PORT],
2891 xprt->address_strings[RPC_DISPLAY_PROTO]);
2892 else
2893 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2894 xprt->address_strings[RPC_DISPLAY_ADDR],
2895 xprt->address_strings[RPC_DISPLAY_PROTO]);
2896
2897 if (try_module_get(THIS_MODULE))
2898 return xprt;
2899 ret = ERR_PTR(-EINVAL);
2900out_err:
2901 xs_xprt_free(xprt);
2902 return ret;
2903}
2904
2905/**
2906 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2907 * @args: rpc transport creation arguments
2908 *
2909 */
2910static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2911{
2912 struct sockaddr *addr = args->dstaddr;
2913 struct rpc_xprt *xprt;
2914 struct sock_xprt *transport;
2915 struct svc_sock *bc_sock;
2916 struct rpc_xprt *ret;
2917
2918 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2919 xprt_tcp_slot_table_entries);
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->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2926 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2927 xprt->timeout = &xs_tcp_default_timeout;
2928
2929 /* backchannel */
2930 xprt_set_bound(xprt);
2931 xprt->bind_timeout = 0;
2932 xprt->reestablish_timeout = 0;
2933 xprt->idle_timeout = 0;
2934
2935 xprt->ops = &bc_tcp_ops;
2936
2937 switch (addr->sa_family) {
2938 case AF_INET:
2939 xs_format_peer_addresses(xprt, "tcp",
2940 RPCBIND_NETID_TCP);
2941 break;
2942 case AF_INET6:
2943 xs_format_peer_addresses(xprt, "tcp",
2944 RPCBIND_NETID_TCP6);
2945 break;
2946 default:
2947 ret = ERR_PTR(-EAFNOSUPPORT);
2948 goto out_err;
2949 }
2950
2951 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2952 xprt->address_strings[RPC_DISPLAY_ADDR],
2953 xprt->address_strings[RPC_DISPLAY_PORT],
2954 xprt->address_strings[RPC_DISPLAY_PROTO]);
2955
2956 /*
2957 * Once we've associated a backchannel xprt with a connection,
2958 * we want to keep it around as long as the connection lasts,
2959 * in case we need to start using it for a backchannel again;
2960 * this reference won't be dropped until bc_xprt is destroyed.
2961 */
2962 xprt_get(xprt);
2963 args->bc_xprt->xpt_bc_xprt = xprt;
2964 xprt->bc_xprt = args->bc_xprt;
2965 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2966 transport->sock = bc_sock->sk_sock;
2967 transport->inet = bc_sock->sk_sk;
2968
2969 /*
2970 * Since we don't want connections for the backchannel, we set
2971 * the xprt status to connected
2972 */
2973 xprt_set_connected(xprt);
2974
2975 if (try_module_get(THIS_MODULE))
2976 return xprt;
2977
2978 args->bc_xprt->xpt_bc_xprt = NULL;
2979 xprt_put(xprt);
2980 ret = ERR_PTR(-EINVAL);
2981out_err:
2982 xs_xprt_free(xprt);
2983 return ret;
2984}
2985
2986static struct xprt_class xs_local_transport = {
2987 .list = LIST_HEAD_INIT(xs_local_transport.list),
2988 .name = "named UNIX socket",
2989 .owner = THIS_MODULE,
2990 .ident = XPRT_TRANSPORT_LOCAL,
2991 .setup = xs_setup_local,
2992};
2993
2994static struct xprt_class xs_udp_transport = {
2995 .list = LIST_HEAD_INIT(xs_udp_transport.list),
2996 .name = "udp",
2997 .owner = THIS_MODULE,
2998 .ident = XPRT_TRANSPORT_UDP,
2999 .setup = xs_setup_udp,
3000};
3001
3002static struct xprt_class xs_tcp_transport = {
3003 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3004 .name = "tcp",
3005 .owner = THIS_MODULE,
3006 .ident = XPRT_TRANSPORT_TCP,
3007 .setup = xs_setup_tcp,
3008};
3009
3010static struct xprt_class xs_bc_tcp_transport = {
3011 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3012 .name = "tcp NFSv4.1 backchannel",
3013 .owner = THIS_MODULE,
3014 .ident = XPRT_TRANSPORT_BC_TCP,
3015 .setup = xs_setup_bc_tcp,
3016};
3017
3018/**
3019 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3020 *
3021 */
3022int init_socket_xprt(void)
3023{
3024#ifdef RPC_DEBUG
3025 if (!sunrpc_table_header)
3026 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3027#endif
3028
3029 xprt_register_transport(&xs_local_transport);
3030 xprt_register_transport(&xs_udp_transport);
3031 xprt_register_transport(&xs_tcp_transport);
3032 xprt_register_transport(&xs_bc_tcp_transport);
3033
3034 return 0;
3035}
3036
3037/**
3038 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3039 *
3040 */
3041void cleanup_socket_xprt(void)
3042{
3043#ifdef RPC_DEBUG
3044 if (sunrpc_table_header) {
3045 unregister_sysctl_table(sunrpc_table_header);
3046 sunrpc_table_header = NULL;
3047 }
3048#endif
3049
3050 xprt_unregister_transport(&xs_local_transport);
3051 xprt_unregister_transport(&xs_udp_transport);
3052 xprt_unregister_transport(&xs_tcp_transport);
3053 xprt_unregister_transport(&xs_bc_tcp_transport);
3054}
3055
3056static int param_set_uint_minmax(const char *val,
3057 const struct kernel_param *kp,
3058 unsigned int min, unsigned int max)
3059{
3060 unsigned long num;
3061 int ret;
3062
3063 if (!val)
3064 return -EINVAL;
3065 ret = strict_strtoul(val, 0, &num);
3066 if (ret == -EINVAL || num < min || num > max)
3067 return -EINVAL;
3068 *((unsigned int *)kp->arg) = num;
3069 return 0;
3070}
3071
3072static int param_set_portnr(const char *val, const struct kernel_param *kp)
3073{
3074 return param_set_uint_minmax(val, kp,
3075 RPC_MIN_RESVPORT,
3076 RPC_MAX_RESVPORT);
3077}
3078
3079static struct kernel_param_ops param_ops_portnr = {
3080 .set = param_set_portnr,
3081 .get = param_get_uint,
3082};
3083
3084#define param_check_portnr(name, p) \
3085 __param_check(name, p, unsigned int);
3086
3087module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3088module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3089
3090static int param_set_slot_table_size(const char *val,
3091 const struct kernel_param *kp)
3092{
3093 return param_set_uint_minmax(val, kp,
3094 RPC_MIN_SLOT_TABLE,
3095 RPC_MAX_SLOT_TABLE);
3096}
3097
3098static struct kernel_param_ops param_ops_slot_table_size = {
3099 .set = param_set_slot_table_size,
3100 .get = param_get_uint,
3101};
3102
3103#define param_check_slot_table_size(name, p) \
3104 __param_check(name, p, unsigned int);
3105
3106static int param_set_max_slot_table_size(const char *val,
3107 const struct kernel_param *kp)
3108{
3109 return param_set_uint_minmax(val, kp,
3110 RPC_MIN_SLOT_TABLE,
3111 RPC_MAX_SLOT_TABLE_LIMIT);
3112}
3113
3114static struct kernel_param_ops param_ops_max_slot_table_size = {
3115 .set = param_set_max_slot_table_size,
3116 .get = param_get_uint,
3117};
3118
3119#define param_check_max_slot_table_size(name, p) \
3120 __param_check(name, p, unsigned int);
3121
3122module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3123 slot_table_size, 0644);
3124module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3125 max_slot_table_size, 0644);
3126module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3127 slot_table_size, 0644);
3128