Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/net/sunrpc/xprtsock.c
4 *
5 * Client-side transport implementation for sockets.
6 *
7 * TCP callback races fixes (C) 1998 Red Hat
8 * TCP send fixes (C) 1998 Red Hat
9 * TCP NFS related read + write fixes
10 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
11 *
12 * Rewrite of larges part of the code in order to stabilize TCP stuff.
13 * Fix behaviour when socket buffer is full.
14 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
15 *
16 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
17 *
18 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
19 * <gilles.quillard@bull.net>
20 */
21
22#include <linux/types.h>
23#include <linux/string.h>
24#include <linux/slab.h>
25#include <linux/module.h>
26#include <linux/capability.h>
27#include <linux/pagemap.h>
28#include <linux/errno.h>
29#include <linux/socket.h>
30#include <linux/in.h>
31#include <linux/net.h>
32#include <linux/mm.h>
33#include <linux/un.h>
34#include <linux/udp.h>
35#include <linux/tcp.h>
36#include <linux/sunrpc/clnt.h>
37#include <linux/sunrpc/addr.h>
38#include <linux/sunrpc/sched.h>
39#include <linux/sunrpc/svcsock.h>
40#include <linux/sunrpc/xprtsock.h>
41#include <linux/file.h>
42#ifdef CONFIG_SUNRPC_BACKCHANNEL
43#include <linux/sunrpc/bc_xprt.h>
44#endif
45
46#include <net/sock.h>
47#include <net/checksum.h>
48#include <net/udp.h>
49#include <net/tcp.h>
50#include <linux/bvec.h>
51#include <linux/highmem.h>
52#include <linux/uio.h>
53#include <linux/sched/mm.h>
54
55#include <trace/events/sunrpc.h>
56
57#include "socklib.h"
58#include "sunrpc.h"
59
60static void xs_close(struct rpc_xprt *xprt);
61static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
62 struct socket *sock);
63
64/*
65 * xprtsock tunables
66 */
67static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
68static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
69static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
70
71static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
72static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
73
74#define XS_TCP_LINGER_TO (15U * HZ)
75static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
76
77/*
78 * We can register our own files under /proc/sys/sunrpc by
79 * calling register_sysctl_table() again. The files in that
80 * directory become the union of all files registered there.
81 *
82 * We simply need to make sure that we don't collide with
83 * someone else's file names!
84 */
85
86static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
87static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
88static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
89static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
90static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
91
92static struct ctl_table_header *sunrpc_table_header;
93
94/*
95 * FIXME: changing the UDP slot table size should also resize the UDP
96 * socket buffers for existing UDP transports
97 */
98static struct ctl_table xs_tunables_table[] = {
99 {
100 .procname = "udp_slot_table_entries",
101 .data = &xprt_udp_slot_table_entries,
102 .maxlen = sizeof(unsigned int),
103 .mode = 0644,
104 .proc_handler = proc_dointvec_minmax,
105 .extra1 = &min_slot_table_size,
106 .extra2 = &max_slot_table_size
107 },
108 {
109 .procname = "tcp_slot_table_entries",
110 .data = &xprt_tcp_slot_table_entries,
111 .maxlen = sizeof(unsigned int),
112 .mode = 0644,
113 .proc_handler = proc_dointvec_minmax,
114 .extra1 = &min_slot_table_size,
115 .extra2 = &max_slot_table_size
116 },
117 {
118 .procname = "tcp_max_slot_table_entries",
119 .data = &xprt_max_tcp_slot_table_entries,
120 .maxlen = sizeof(unsigned int),
121 .mode = 0644,
122 .proc_handler = proc_dointvec_minmax,
123 .extra1 = &min_slot_table_size,
124 .extra2 = &max_tcp_slot_table_limit
125 },
126 {
127 .procname = "min_resvport",
128 .data = &xprt_min_resvport,
129 .maxlen = sizeof(unsigned int),
130 .mode = 0644,
131 .proc_handler = proc_dointvec_minmax,
132 .extra1 = &xprt_min_resvport_limit,
133 .extra2 = &xprt_max_resvport_limit
134 },
135 {
136 .procname = "max_resvport",
137 .data = &xprt_max_resvport,
138 .maxlen = sizeof(unsigned int),
139 .mode = 0644,
140 .proc_handler = proc_dointvec_minmax,
141 .extra1 = &xprt_min_resvport_limit,
142 .extra2 = &xprt_max_resvport_limit
143 },
144 {
145 .procname = "tcp_fin_timeout",
146 .data = &xs_tcp_fin_timeout,
147 .maxlen = sizeof(xs_tcp_fin_timeout),
148 .mode = 0644,
149 .proc_handler = proc_dointvec_jiffies,
150 },
151 { },
152};
153
154static struct ctl_table sunrpc_table[] = {
155 {
156 .procname = "sunrpc",
157 .mode = 0555,
158 .child = xs_tunables_table
159 },
160 { },
161};
162
163/*
164 * Wait duration for a reply from the RPC portmapper.
165 */
166#define XS_BIND_TO (60U * HZ)
167
168/*
169 * Delay if a UDP socket connect error occurs. This is most likely some
170 * kind of resource problem on the local host.
171 */
172#define XS_UDP_REEST_TO (2U * HZ)
173
174/*
175 * The reestablish timeout allows clients to delay for a bit before attempting
176 * to reconnect to a server that just dropped our connection.
177 *
178 * We implement an exponential backoff when trying to reestablish a TCP
179 * transport connection with the server. Some servers like to drop a TCP
180 * connection when they are overworked, so we start with a short timeout and
181 * increase over time if the server is down or not responding.
182 */
183#define XS_TCP_INIT_REEST_TO (3U * HZ)
184
185/*
186 * TCP idle timeout; client drops the transport socket if it is idle
187 * for this long. Note that we also timeout UDP sockets to prevent
188 * holding port numbers when there is no RPC traffic.
189 */
190#define XS_IDLE_DISC_TO (5U * 60 * HZ)
191
192#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
193# undef RPC_DEBUG_DATA
194# define RPCDBG_FACILITY RPCDBG_TRANS
195#endif
196
197#ifdef RPC_DEBUG_DATA
198static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
199{
200 u8 *buf = (u8 *) packet;
201 int j;
202
203 dprintk("RPC: %s\n", msg);
204 for (j = 0; j < count && j < 128; j += 4) {
205 if (!(j & 31)) {
206 if (j)
207 dprintk("\n");
208 dprintk("0x%04x ", j);
209 }
210 dprintk("%02x%02x%02x%02x ",
211 buf[j], buf[j+1], buf[j+2], buf[j+3]);
212 }
213 dprintk("\n");
214}
215#else
216static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
217{
218 /* NOP */
219}
220#endif
221
222static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
223{
224 return (struct rpc_xprt *) sk->sk_user_data;
225}
226
227static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
228{
229 return (struct sockaddr *) &xprt->addr;
230}
231
232static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
233{
234 return (struct sockaddr_un *) &xprt->addr;
235}
236
237static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
238{
239 return (struct sockaddr_in *) &xprt->addr;
240}
241
242static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
243{
244 return (struct sockaddr_in6 *) &xprt->addr;
245}
246
247static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
248{
249 struct sockaddr *sap = xs_addr(xprt);
250 struct sockaddr_in6 *sin6;
251 struct sockaddr_in *sin;
252 struct sockaddr_un *sun;
253 char buf[128];
254
255 switch (sap->sa_family) {
256 case AF_LOCAL:
257 sun = xs_addr_un(xprt);
258 strlcpy(buf, sun->sun_path, sizeof(buf));
259 xprt->address_strings[RPC_DISPLAY_ADDR] =
260 kstrdup(buf, GFP_KERNEL);
261 break;
262 case AF_INET:
263 (void)rpc_ntop(sap, buf, sizeof(buf));
264 xprt->address_strings[RPC_DISPLAY_ADDR] =
265 kstrdup(buf, GFP_KERNEL);
266 sin = xs_addr_in(xprt);
267 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
268 break;
269 case AF_INET6:
270 (void)rpc_ntop(sap, buf, sizeof(buf));
271 xprt->address_strings[RPC_DISPLAY_ADDR] =
272 kstrdup(buf, GFP_KERNEL);
273 sin6 = xs_addr_in6(xprt);
274 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
275 break;
276 default:
277 BUG();
278 }
279
280 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
281}
282
283static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
284{
285 struct sockaddr *sap = xs_addr(xprt);
286 char buf[128];
287
288 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
289 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
290
291 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
292 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
293}
294
295static void xs_format_peer_addresses(struct rpc_xprt *xprt,
296 const char *protocol,
297 const char *netid)
298{
299 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
300 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
301 xs_format_common_peer_addresses(xprt);
302 xs_format_common_peer_ports(xprt);
303}
304
305static void xs_update_peer_port(struct rpc_xprt *xprt)
306{
307 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
308 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
309
310 xs_format_common_peer_ports(xprt);
311}
312
313static void xs_free_peer_addresses(struct rpc_xprt *xprt)
314{
315 unsigned int i;
316
317 for (i = 0; i < RPC_DISPLAY_MAX; i++)
318 switch (i) {
319 case RPC_DISPLAY_PROTO:
320 case RPC_DISPLAY_NETID:
321 continue;
322 default:
323 kfree(xprt->address_strings[i]);
324 }
325}
326
327static size_t
328xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp)
329{
330 size_t i,n;
331
332 if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES))
333 return want;
334 n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT;
335 for (i = 0; i < n; i++) {
336 if (buf->pages[i])
337 continue;
338 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp);
339 if (!buf->pages[i]) {
340 i *= PAGE_SIZE;
341 return i > buf->page_base ? i - buf->page_base : 0;
342 }
343 }
344 return want;
345}
346
347static ssize_t
348xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek)
349{
350 ssize_t ret;
351 if (seek != 0)
352 iov_iter_advance(&msg->msg_iter, seek);
353 ret = sock_recvmsg(sock, msg, flags);
354 return ret > 0 ? ret + seek : ret;
355}
356
357static ssize_t
358xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags,
359 struct kvec *kvec, size_t count, size_t seek)
360{
361 iov_iter_kvec(&msg->msg_iter, READ, kvec, 1, count);
362 return xs_sock_recvmsg(sock, msg, flags, seek);
363}
364
365static ssize_t
366xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags,
367 struct bio_vec *bvec, unsigned long nr, size_t count,
368 size_t seek)
369{
370 iov_iter_bvec(&msg->msg_iter, READ, bvec, nr, count);
371 return xs_sock_recvmsg(sock, msg, flags, seek);
372}
373
374static ssize_t
375xs_read_discard(struct socket *sock, struct msghdr *msg, int flags,
376 size_t count)
377{
378 iov_iter_discard(&msg->msg_iter, READ, count);
379 return sock_recvmsg(sock, msg, flags);
380}
381
382#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
383static void
384xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
385{
386 struct bvec_iter bi = {
387 .bi_size = count,
388 };
389 struct bio_vec bv;
390
391 bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
392 for_each_bvec(bv, bvec, bi, bi)
393 flush_dcache_page(bv.bv_page);
394}
395#else
396static inline void
397xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
398{
399}
400#endif
401
402static ssize_t
403xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
404 struct xdr_buf *buf, size_t count, size_t seek, size_t *read)
405{
406 size_t want, seek_init = seek, offset = 0;
407 ssize_t ret;
408
409 want = min_t(size_t, count, buf->head[0].iov_len);
410 if (seek < want) {
411 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek);
412 if (ret <= 0)
413 goto sock_err;
414 offset += ret;
415 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
416 goto out;
417 if (ret != want)
418 goto out;
419 seek = 0;
420 } else {
421 seek -= want;
422 offset += want;
423 }
424
425 want = xs_alloc_sparse_pages(buf,
426 min_t(size_t, count - offset, buf->page_len),
427 GFP_KERNEL);
428 if (seek < want) {
429 ret = xs_read_bvec(sock, msg, flags, buf->bvec,
430 xdr_buf_pagecount(buf),
431 want + buf->page_base,
432 seek + buf->page_base);
433 if (ret <= 0)
434 goto sock_err;
435 xs_flush_bvec(buf->bvec, ret, seek + buf->page_base);
436 offset += ret - buf->page_base;
437 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
438 goto out;
439 if (ret != want)
440 goto out;
441 seek = 0;
442 } else {
443 seek -= want;
444 offset += want;
445 }
446
447 want = min_t(size_t, count - offset, buf->tail[0].iov_len);
448 if (seek < want) {
449 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek);
450 if (ret <= 0)
451 goto sock_err;
452 offset += ret;
453 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
454 goto out;
455 if (ret != want)
456 goto out;
457 } else if (offset < seek_init)
458 offset = seek_init;
459 ret = -EMSGSIZE;
460out:
461 *read = offset - seek_init;
462 return ret;
463sock_err:
464 offset += seek;
465 goto out;
466}
467
468static void
469xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf)
470{
471 if (!transport->recv.copied) {
472 if (buf->head[0].iov_len >= transport->recv.offset)
473 memcpy(buf->head[0].iov_base,
474 &transport->recv.xid,
475 transport->recv.offset);
476 transport->recv.copied = transport->recv.offset;
477 }
478}
479
480static bool
481xs_read_stream_request_done(struct sock_xprt *transport)
482{
483 return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT);
484}
485
486static void
487xs_read_stream_check_eor(struct sock_xprt *transport,
488 struct msghdr *msg)
489{
490 if (xs_read_stream_request_done(transport))
491 msg->msg_flags |= MSG_EOR;
492}
493
494static ssize_t
495xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg,
496 int flags, struct rpc_rqst *req)
497{
498 struct xdr_buf *buf = &req->rq_private_buf;
499 size_t want, read;
500 ssize_t ret;
501
502 xs_read_header(transport, buf);
503
504 want = transport->recv.len - transport->recv.offset;
505 if (want != 0) {
506 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf,
507 transport->recv.copied + want,
508 transport->recv.copied,
509 &read);
510 transport->recv.offset += read;
511 transport->recv.copied += read;
512 }
513
514 if (transport->recv.offset == transport->recv.len)
515 xs_read_stream_check_eor(transport, msg);
516
517 if (want == 0)
518 return 0;
519
520 switch (ret) {
521 default:
522 break;
523 case -EFAULT:
524 case -EMSGSIZE:
525 msg->msg_flags |= MSG_TRUNC;
526 return read;
527 case 0:
528 return -ESHUTDOWN;
529 }
530 return ret < 0 ? ret : read;
531}
532
533static size_t
534xs_read_stream_headersize(bool isfrag)
535{
536 if (isfrag)
537 return sizeof(__be32);
538 return 3 * sizeof(__be32);
539}
540
541static ssize_t
542xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg,
543 int flags, size_t want, size_t seek)
544{
545 struct kvec kvec = {
546 .iov_base = &transport->recv.fraghdr,
547 .iov_len = want,
548 };
549 return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek);
550}
551
552#if defined(CONFIG_SUNRPC_BACKCHANNEL)
553static ssize_t
554xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
555{
556 struct rpc_xprt *xprt = &transport->xprt;
557 struct rpc_rqst *req;
558 ssize_t ret;
559
560 /* Look up and lock the request corresponding to the given XID */
561 req = xprt_lookup_bc_request(xprt, transport->recv.xid);
562 if (!req) {
563 printk(KERN_WARNING "Callback slot table overflowed\n");
564 return -ESHUTDOWN;
565 }
566 if (transport->recv.copied && !req->rq_private_buf.len)
567 return -ESHUTDOWN;
568
569 ret = xs_read_stream_request(transport, msg, flags, req);
570 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
571 xprt_complete_bc_request(req, transport->recv.copied);
572 else
573 req->rq_private_buf.len = transport->recv.copied;
574
575 return ret;
576}
577#else /* CONFIG_SUNRPC_BACKCHANNEL */
578static ssize_t
579xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
580{
581 return -ESHUTDOWN;
582}
583#endif /* CONFIG_SUNRPC_BACKCHANNEL */
584
585static ssize_t
586xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags)
587{
588 struct rpc_xprt *xprt = &transport->xprt;
589 struct rpc_rqst *req;
590 ssize_t ret = 0;
591
592 /* Look up and lock the request corresponding to the given XID */
593 spin_lock(&xprt->queue_lock);
594 req = xprt_lookup_rqst(xprt, transport->recv.xid);
595 if (!req || (transport->recv.copied && !req->rq_private_buf.len)) {
596 msg->msg_flags |= MSG_TRUNC;
597 goto out;
598 }
599 xprt_pin_rqst(req);
600 spin_unlock(&xprt->queue_lock);
601
602 ret = xs_read_stream_request(transport, msg, flags, req);
603
604 spin_lock(&xprt->queue_lock);
605 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
606 xprt_complete_rqst(req->rq_task, transport->recv.copied);
607 else
608 req->rq_private_buf.len = transport->recv.copied;
609 xprt_unpin_rqst(req);
610out:
611 spin_unlock(&xprt->queue_lock);
612 return ret;
613}
614
615static ssize_t
616xs_read_stream(struct sock_xprt *transport, int flags)
617{
618 struct msghdr msg = { 0 };
619 size_t want, read = 0;
620 ssize_t ret = 0;
621
622 if (transport->recv.len == 0) {
623 want = xs_read_stream_headersize(transport->recv.copied != 0);
624 ret = xs_read_stream_header(transport, &msg, flags, want,
625 transport->recv.offset);
626 if (ret <= 0)
627 goto out_err;
628 transport->recv.offset = ret;
629 if (transport->recv.offset != want)
630 return transport->recv.offset;
631 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) &
632 RPC_FRAGMENT_SIZE_MASK;
633 transport->recv.offset -= sizeof(transport->recv.fraghdr);
634 read = ret;
635 }
636
637 switch (be32_to_cpu(transport->recv.calldir)) {
638 default:
639 msg.msg_flags |= MSG_TRUNC;
640 break;
641 case RPC_CALL:
642 ret = xs_read_stream_call(transport, &msg, flags);
643 break;
644 case RPC_REPLY:
645 ret = xs_read_stream_reply(transport, &msg, flags);
646 }
647 if (msg.msg_flags & MSG_TRUNC) {
648 transport->recv.calldir = cpu_to_be32(-1);
649 transport->recv.copied = -1;
650 }
651 if (ret < 0)
652 goto out_err;
653 read += ret;
654 if (transport->recv.offset < transport->recv.len) {
655 if (!(msg.msg_flags & MSG_TRUNC))
656 return read;
657 msg.msg_flags = 0;
658 ret = xs_read_discard(transport->sock, &msg, flags,
659 transport->recv.len - transport->recv.offset);
660 if (ret <= 0)
661 goto out_err;
662 transport->recv.offset += ret;
663 read += ret;
664 if (transport->recv.offset != transport->recv.len)
665 return read;
666 }
667 if (xs_read_stream_request_done(transport)) {
668 trace_xs_stream_read_request(transport);
669 transport->recv.copied = 0;
670 }
671 transport->recv.offset = 0;
672 transport->recv.len = 0;
673 return read;
674out_err:
675 return ret != 0 ? ret : -ESHUTDOWN;
676}
677
678static __poll_t xs_poll_socket(struct sock_xprt *transport)
679{
680 return transport->sock->ops->poll(transport->file, transport->sock,
681 NULL);
682}
683
684static bool xs_poll_socket_readable(struct sock_xprt *transport)
685{
686 __poll_t events = xs_poll_socket(transport);
687
688 return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP);
689}
690
691static void xs_poll_check_readable(struct sock_xprt *transport)
692{
693
694 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
695 if (!xs_poll_socket_readable(transport))
696 return;
697 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
698 queue_work(xprtiod_workqueue, &transport->recv_worker);
699}
700
701static void xs_stream_data_receive(struct sock_xprt *transport)
702{
703 size_t read = 0;
704 ssize_t ret = 0;
705
706 mutex_lock(&transport->recv_mutex);
707 if (transport->sock == NULL)
708 goto out;
709 for (;;) {
710 ret = xs_read_stream(transport, MSG_DONTWAIT);
711 if (ret < 0)
712 break;
713 read += ret;
714 cond_resched();
715 }
716 if (ret == -ESHUTDOWN)
717 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
718 else
719 xs_poll_check_readable(transport);
720out:
721 mutex_unlock(&transport->recv_mutex);
722 trace_xs_stream_read_data(&transport->xprt, ret, read);
723}
724
725static void xs_stream_data_receive_workfn(struct work_struct *work)
726{
727 struct sock_xprt *transport =
728 container_of(work, struct sock_xprt, recv_worker);
729 unsigned int pflags = memalloc_nofs_save();
730
731 xs_stream_data_receive(transport);
732 memalloc_nofs_restore(pflags);
733}
734
735static void
736xs_stream_reset_connect(struct sock_xprt *transport)
737{
738 transport->recv.offset = 0;
739 transport->recv.len = 0;
740 transport->recv.copied = 0;
741 transport->xmit.offset = 0;
742}
743
744static void
745xs_stream_start_connect(struct sock_xprt *transport)
746{
747 transport->xprt.stat.connect_count++;
748 transport->xprt.stat.connect_start = jiffies;
749}
750
751#define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
752
753/**
754 * xs_nospace - handle transmit was incomplete
755 * @req: pointer to RPC request
756 *
757 */
758static int xs_nospace(struct rpc_rqst *req)
759{
760 struct rpc_xprt *xprt = req->rq_xprt;
761 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
762 struct sock *sk = transport->inet;
763 int ret = -EAGAIN;
764
765 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
766 req->rq_task->tk_pid,
767 req->rq_slen - transport->xmit.offset,
768 req->rq_slen);
769
770 /* Protect against races with write_space */
771 spin_lock(&xprt->transport_lock);
772
773 /* Don't race with disconnect */
774 if (xprt_connected(xprt)) {
775 /* wait for more buffer space */
776 sk->sk_write_pending++;
777 xprt_wait_for_buffer_space(xprt);
778 } else
779 ret = -ENOTCONN;
780
781 spin_unlock(&xprt->transport_lock);
782
783 /* Race breaker in case memory is freed before above code is called */
784 if (ret == -EAGAIN) {
785 struct socket_wq *wq;
786
787 rcu_read_lock();
788 wq = rcu_dereference(sk->sk_wq);
789 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
790 rcu_read_unlock();
791
792 sk->sk_write_space(sk);
793 }
794 return ret;
795}
796
797static void
798xs_stream_prepare_request(struct rpc_rqst *req)
799{
800 xdr_free_bvec(&req->rq_rcv_buf);
801 req->rq_task->tk_status = xdr_alloc_bvec(&req->rq_rcv_buf, GFP_KERNEL);
802}
803
804/*
805 * Determine if the previous message in the stream was aborted before it
806 * could complete transmission.
807 */
808static bool
809xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req)
810{
811 return transport->xmit.offset != 0 && req->rq_bytes_sent == 0;
812}
813
814/*
815 * Return the stream record marker field for a record of length < 2^31-1
816 */
817static rpc_fraghdr
818xs_stream_record_marker(struct xdr_buf *xdr)
819{
820 if (!xdr->len)
821 return 0;
822 return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len);
823}
824
825/**
826 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
827 * @req: pointer to RPC request
828 *
829 * Return values:
830 * 0: The request has been sent
831 * EAGAIN: The socket was blocked, please call again later to
832 * complete the request
833 * ENOTCONN: Caller needs to invoke connect logic then call again
834 * other: Some other error occured, the request was not sent
835 */
836static int xs_local_send_request(struct rpc_rqst *req)
837{
838 struct rpc_xprt *xprt = req->rq_xprt;
839 struct sock_xprt *transport =
840 container_of(xprt, struct sock_xprt, xprt);
841 struct xdr_buf *xdr = &req->rq_snd_buf;
842 rpc_fraghdr rm = xs_stream_record_marker(xdr);
843 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
844 struct msghdr msg = {
845 .msg_flags = XS_SENDMSG_FLAGS,
846 };
847 unsigned int sent;
848 int status;
849
850 /* Close the stream if the previous transmission was incomplete */
851 if (xs_send_request_was_aborted(transport, req)) {
852 xs_close(xprt);
853 return -ENOTCONN;
854 }
855
856 xs_pktdump("packet data:",
857 req->rq_svec->iov_base, req->rq_svec->iov_len);
858
859 req->rq_xtime = ktime_get();
860 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
861 transport->xmit.offset, rm, &sent);
862 dprintk("RPC: %s(%u) = %d\n",
863 __func__, xdr->len - transport->xmit.offset, status);
864
865 if (status == -EAGAIN && sock_writeable(transport->inet))
866 status = -ENOBUFS;
867
868 if (likely(sent > 0) || status == 0) {
869 transport->xmit.offset += sent;
870 req->rq_bytes_sent = transport->xmit.offset;
871 if (likely(req->rq_bytes_sent >= msglen)) {
872 req->rq_xmit_bytes_sent += transport->xmit.offset;
873 transport->xmit.offset = 0;
874 return 0;
875 }
876 status = -EAGAIN;
877 }
878
879 switch (status) {
880 case -ENOBUFS:
881 break;
882 case -EAGAIN:
883 status = xs_nospace(req);
884 break;
885 default:
886 dprintk("RPC: sendmsg returned unrecognized error %d\n",
887 -status);
888 fallthrough;
889 case -EPIPE:
890 xs_close(xprt);
891 status = -ENOTCONN;
892 }
893
894 return status;
895}
896
897/**
898 * xs_udp_send_request - write an RPC request to a UDP socket
899 * @req: pointer to RPC request
900 *
901 * Return values:
902 * 0: The request has been sent
903 * EAGAIN: The socket was blocked, please call again later to
904 * complete the request
905 * ENOTCONN: Caller needs to invoke connect logic then call again
906 * other: Some other error occurred, the request was not sent
907 */
908static int xs_udp_send_request(struct rpc_rqst *req)
909{
910 struct rpc_xprt *xprt = req->rq_xprt;
911 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
912 struct xdr_buf *xdr = &req->rq_snd_buf;
913 struct msghdr msg = {
914 .msg_name = xs_addr(xprt),
915 .msg_namelen = xprt->addrlen,
916 .msg_flags = XS_SENDMSG_FLAGS,
917 };
918 unsigned int sent;
919 int status;
920
921 xs_pktdump("packet data:",
922 req->rq_svec->iov_base,
923 req->rq_svec->iov_len);
924
925 if (!xprt_bound(xprt))
926 return -ENOTCONN;
927
928 if (!xprt_request_get_cong(xprt, req))
929 return -EBADSLT;
930
931 req->rq_xtime = ktime_get();
932 status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, 0, &sent);
933
934 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
935 xdr->len, status);
936
937 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
938 if (status == -EPERM)
939 goto process_status;
940
941 if (status == -EAGAIN && sock_writeable(transport->inet))
942 status = -ENOBUFS;
943
944 if (sent > 0 || status == 0) {
945 req->rq_xmit_bytes_sent += sent;
946 if (sent >= req->rq_slen)
947 return 0;
948 /* Still some bytes left; set up for a retry later. */
949 status = -EAGAIN;
950 }
951
952process_status:
953 switch (status) {
954 case -ENOTSOCK:
955 status = -ENOTCONN;
956 /* Should we call xs_close() here? */
957 break;
958 case -EAGAIN:
959 status = xs_nospace(req);
960 break;
961 case -ENETUNREACH:
962 case -ENOBUFS:
963 case -EPIPE:
964 case -ECONNREFUSED:
965 case -EPERM:
966 /* When the server has died, an ICMP port unreachable message
967 * prompts ECONNREFUSED. */
968 break;
969 default:
970 dprintk("RPC: sendmsg returned unrecognized error %d\n",
971 -status);
972 }
973
974 return status;
975}
976
977/**
978 * xs_tcp_send_request - write an RPC request to a TCP socket
979 * @req: pointer to RPC request
980 *
981 * Return values:
982 * 0: The request has been sent
983 * EAGAIN: The socket was blocked, please call again later to
984 * complete the request
985 * ENOTCONN: Caller needs to invoke connect logic then call again
986 * other: Some other error occurred, the request was not sent
987 *
988 * XXX: In the case of soft timeouts, should we eventually give up
989 * if sendmsg is not able to make progress?
990 */
991static int xs_tcp_send_request(struct rpc_rqst *req)
992{
993 struct rpc_xprt *xprt = req->rq_xprt;
994 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
995 struct xdr_buf *xdr = &req->rq_snd_buf;
996 rpc_fraghdr rm = xs_stream_record_marker(xdr);
997 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
998 struct msghdr msg = {
999 .msg_flags = XS_SENDMSG_FLAGS,
1000 };
1001 bool vm_wait = false;
1002 unsigned int sent;
1003 int status;
1004
1005 /* Close the stream if the previous transmission was incomplete */
1006 if (xs_send_request_was_aborted(transport, req)) {
1007 if (transport->sock != NULL)
1008 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
1009 return -ENOTCONN;
1010 }
1011
1012 xs_pktdump("packet data:",
1013 req->rq_svec->iov_base,
1014 req->rq_svec->iov_len);
1015
1016 if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
1017 xs_tcp_set_socket_timeouts(xprt, transport->sock);
1018
1019 /* Continue transmitting the packet/record. We must be careful
1020 * to cope with writespace callbacks arriving _after_ we have
1021 * called sendmsg(). */
1022 req->rq_xtime = ktime_get();
1023 while (1) {
1024 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
1025 transport->xmit.offset, rm, &sent);
1026
1027 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
1028 xdr->len - transport->xmit.offset, status);
1029
1030 /* If we've sent the entire packet, immediately
1031 * reset the count of bytes sent. */
1032 transport->xmit.offset += sent;
1033 req->rq_bytes_sent = transport->xmit.offset;
1034 if (likely(req->rq_bytes_sent >= msglen)) {
1035 req->rq_xmit_bytes_sent += transport->xmit.offset;
1036 transport->xmit.offset = 0;
1037 return 0;
1038 }
1039
1040 WARN_ON_ONCE(sent == 0 && status == 0);
1041
1042 if (status == -EAGAIN ) {
1043 /*
1044 * Return EAGAIN if we're sure we're hitting the
1045 * socket send buffer limits.
1046 */
1047 if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
1048 break;
1049 /*
1050 * Did we hit a memory allocation failure?
1051 */
1052 if (sent == 0) {
1053 status = -ENOBUFS;
1054 if (vm_wait)
1055 break;
1056 /* Retry, knowing now that we're below the
1057 * socket send buffer limit
1058 */
1059 vm_wait = true;
1060 }
1061 continue;
1062 }
1063 if (status < 0)
1064 break;
1065 vm_wait = false;
1066 }
1067
1068 switch (status) {
1069 case -ENOTSOCK:
1070 status = -ENOTCONN;
1071 /* Should we call xs_close() here? */
1072 break;
1073 case -EAGAIN:
1074 status = xs_nospace(req);
1075 break;
1076 case -ECONNRESET:
1077 case -ECONNREFUSED:
1078 case -ENOTCONN:
1079 case -EADDRINUSE:
1080 case -ENOBUFS:
1081 case -EPIPE:
1082 break;
1083 default:
1084 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1085 -status);
1086 }
1087
1088 return status;
1089}
1090
1091static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1092{
1093 transport->old_data_ready = sk->sk_data_ready;
1094 transport->old_state_change = sk->sk_state_change;
1095 transport->old_write_space = sk->sk_write_space;
1096 transport->old_error_report = sk->sk_error_report;
1097}
1098
1099static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1100{
1101 sk->sk_data_ready = transport->old_data_ready;
1102 sk->sk_state_change = transport->old_state_change;
1103 sk->sk_write_space = transport->old_write_space;
1104 sk->sk_error_report = transport->old_error_report;
1105}
1106
1107static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
1108{
1109 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1110
1111 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1112 clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state);
1113 clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state);
1114 clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state);
1115}
1116
1117static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr)
1118{
1119 set_bit(nr, &transport->sock_state);
1120 queue_work(xprtiod_workqueue, &transport->error_worker);
1121}
1122
1123static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1124{
1125 smp_mb__before_atomic();
1126 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1127 clear_bit(XPRT_CLOSING, &xprt->state);
1128 xs_sock_reset_state_flags(xprt);
1129 smp_mb__after_atomic();
1130}
1131
1132/**
1133 * xs_error_report - callback to handle TCP socket state errors
1134 * @sk: socket
1135 *
1136 * Note: we don't call sock_error() since there may be a rpc_task
1137 * using the socket, and so we don't want to clear sk->sk_err.
1138 */
1139static void xs_error_report(struct sock *sk)
1140{
1141 struct sock_xprt *transport;
1142 struct rpc_xprt *xprt;
1143
1144 read_lock_bh(&sk->sk_callback_lock);
1145 if (!(xprt = xprt_from_sock(sk)))
1146 goto out;
1147
1148 transport = container_of(xprt, struct sock_xprt, xprt);
1149 transport->xprt_err = -sk->sk_err;
1150 if (transport->xprt_err == 0)
1151 goto out;
1152 dprintk("RPC: xs_error_report client %p, error=%d...\n",
1153 xprt, -transport->xprt_err);
1154 trace_rpc_socket_error(xprt, sk->sk_socket, transport->xprt_err);
1155
1156 /* barrier ensures xprt_err is set before XPRT_SOCK_WAKE_ERROR */
1157 smp_mb__before_atomic();
1158 xs_run_error_worker(transport, XPRT_SOCK_WAKE_ERROR);
1159 out:
1160 read_unlock_bh(&sk->sk_callback_lock);
1161}
1162
1163static void xs_reset_transport(struct sock_xprt *transport)
1164{
1165 struct socket *sock = transport->sock;
1166 struct sock *sk = transport->inet;
1167 struct rpc_xprt *xprt = &transport->xprt;
1168 struct file *filp = transport->file;
1169
1170 if (sk == NULL)
1171 return;
1172
1173 if (atomic_read(&transport->xprt.swapper))
1174 sk_clear_memalloc(sk);
1175
1176 kernel_sock_shutdown(sock, SHUT_RDWR);
1177
1178 mutex_lock(&transport->recv_mutex);
1179 write_lock_bh(&sk->sk_callback_lock);
1180 transport->inet = NULL;
1181 transport->sock = NULL;
1182 transport->file = NULL;
1183
1184 sk->sk_user_data = NULL;
1185
1186 xs_restore_old_callbacks(transport, sk);
1187 xprt_clear_connected(xprt);
1188 write_unlock_bh(&sk->sk_callback_lock);
1189 xs_sock_reset_connection_flags(xprt);
1190 /* Reset stream record info */
1191 xs_stream_reset_connect(transport);
1192 mutex_unlock(&transport->recv_mutex);
1193
1194 trace_rpc_socket_close(xprt, sock);
1195 fput(filp);
1196
1197 xprt_disconnect_done(xprt);
1198}
1199
1200/**
1201 * xs_close - close a socket
1202 * @xprt: transport
1203 *
1204 * This is used when all requests are complete; ie, no DRC state remains
1205 * on the server we want to save.
1206 *
1207 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1208 * xs_reset_transport() zeroing the socket from underneath a writer.
1209 */
1210static void xs_close(struct rpc_xprt *xprt)
1211{
1212 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1213
1214 dprintk("RPC: xs_close xprt %p\n", xprt);
1215
1216 xs_reset_transport(transport);
1217 xprt->reestablish_timeout = 0;
1218}
1219
1220static void xs_inject_disconnect(struct rpc_xprt *xprt)
1221{
1222 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
1223 xprt);
1224 xprt_disconnect_done(xprt);
1225}
1226
1227static void xs_xprt_free(struct rpc_xprt *xprt)
1228{
1229 xs_free_peer_addresses(xprt);
1230 xprt_free(xprt);
1231}
1232
1233/**
1234 * xs_destroy - prepare to shutdown a transport
1235 * @xprt: doomed transport
1236 *
1237 */
1238static void xs_destroy(struct rpc_xprt *xprt)
1239{
1240 struct sock_xprt *transport = container_of(xprt,
1241 struct sock_xprt, xprt);
1242 dprintk("RPC: xs_destroy xprt %p\n", xprt);
1243
1244 cancel_delayed_work_sync(&transport->connect_worker);
1245 xs_close(xprt);
1246 cancel_work_sync(&transport->recv_worker);
1247 cancel_work_sync(&transport->error_worker);
1248 xs_xprt_free(xprt);
1249 module_put(THIS_MODULE);
1250}
1251
1252/**
1253 * xs_udp_data_read_skb - receive callback for UDP sockets
1254 * @xprt: transport
1255 * @sk: socket
1256 * @skb: skbuff
1257 *
1258 */
1259static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1260 struct sock *sk,
1261 struct sk_buff *skb)
1262{
1263 struct rpc_task *task;
1264 struct rpc_rqst *rovr;
1265 int repsize, copied;
1266 u32 _xid;
1267 __be32 *xp;
1268
1269 repsize = skb->len;
1270 if (repsize < 4) {
1271 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
1272 return;
1273 }
1274
1275 /* Copy the XID from the skb... */
1276 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1277 if (xp == NULL)
1278 return;
1279
1280 /* Look up and lock the request corresponding to the given XID */
1281 spin_lock(&xprt->queue_lock);
1282 rovr = xprt_lookup_rqst(xprt, *xp);
1283 if (!rovr)
1284 goto out_unlock;
1285 xprt_pin_rqst(rovr);
1286 xprt_update_rtt(rovr->rq_task);
1287 spin_unlock(&xprt->queue_lock);
1288 task = rovr->rq_task;
1289
1290 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1291 copied = repsize;
1292
1293 /* Suck it into the iovec, verify checksum if not done by hw. */
1294 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1295 spin_lock(&xprt->queue_lock);
1296 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1297 goto out_unpin;
1298 }
1299
1300
1301 spin_lock(&xprt->transport_lock);
1302 xprt_adjust_cwnd(xprt, task, copied);
1303 spin_unlock(&xprt->transport_lock);
1304 spin_lock(&xprt->queue_lock);
1305 xprt_complete_rqst(task, copied);
1306 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1307out_unpin:
1308 xprt_unpin_rqst(rovr);
1309 out_unlock:
1310 spin_unlock(&xprt->queue_lock);
1311}
1312
1313static void xs_udp_data_receive(struct sock_xprt *transport)
1314{
1315 struct sk_buff *skb;
1316 struct sock *sk;
1317 int err;
1318
1319 mutex_lock(&transport->recv_mutex);
1320 sk = transport->inet;
1321 if (sk == NULL)
1322 goto out;
1323 for (;;) {
1324 skb = skb_recv_udp(sk, 0, 1, &err);
1325 if (skb == NULL)
1326 break;
1327 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1328 consume_skb(skb);
1329 cond_resched();
1330 }
1331 xs_poll_check_readable(transport);
1332out:
1333 mutex_unlock(&transport->recv_mutex);
1334}
1335
1336static void xs_udp_data_receive_workfn(struct work_struct *work)
1337{
1338 struct sock_xprt *transport =
1339 container_of(work, struct sock_xprt, recv_worker);
1340 unsigned int pflags = memalloc_nofs_save();
1341
1342 xs_udp_data_receive(transport);
1343 memalloc_nofs_restore(pflags);
1344}
1345
1346/**
1347 * xs_data_ready - "data ready" callback for UDP sockets
1348 * @sk: socket with data to read
1349 *
1350 */
1351static void xs_data_ready(struct sock *sk)
1352{
1353 struct rpc_xprt *xprt;
1354
1355 read_lock_bh(&sk->sk_callback_lock);
1356 dprintk("RPC: xs_data_ready...\n");
1357 xprt = xprt_from_sock(sk);
1358 if (xprt != NULL) {
1359 struct sock_xprt *transport = container_of(xprt,
1360 struct sock_xprt, xprt);
1361 transport->old_data_ready(sk);
1362 /* Any data means we had a useful conversation, so
1363 * then we don't need to delay the next reconnect
1364 */
1365 if (xprt->reestablish_timeout)
1366 xprt->reestablish_timeout = 0;
1367 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1368 queue_work(xprtiod_workqueue, &transport->recv_worker);
1369 }
1370 read_unlock_bh(&sk->sk_callback_lock);
1371}
1372
1373/*
1374 * Helper function to force a TCP close if the server is sending
1375 * junk and/or it has put us in CLOSE_WAIT
1376 */
1377static void xs_tcp_force_close(struct rpc_xprt *xprt)
1378{
1379 xprt_force_disconnect(xprt);
1380}
1381
1382#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1383static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1384{
1385 return PAGE_SIZE;
1386}
1387#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1388
1389/**
1390 * xs_tcp_state_change - callback to handle TCP socket state changes
1391 * @sk: socket whose state has changed
1392 *
1393 */
1394static void xs_tcp_state_change(struct sock *sk)
1395{
1396 struct rpc_xprt *xprt;
1397 struct sock_xprt *transport;
1398
1399 read_lock_bh(&sk->sk_callback_lock);
1400 if (!(xprt = xprt_from_sock(sk)))
1401 goto out;
1402 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1403 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1404 sk->sk_state, xprt_connected(xprt),
1405 sock_flag(sk, SOCK_DEAD),
1406 sock_flag(sk, SOCK_ZAPPED),
1407 sk->sk_shutdown);
1408
1409 transport = container_of(xprt, struct sock_xprt, xprt);
1410 trace_rpc_socket_state_change(xprt, sk->sk_socket);
1411 switch (sk->sk_state) {
1412 case TCP_ESTABLISHED:
1413 if (!xprt_test_and_set_connected(xprt)) {
1414 xprt->connect_cookie++;
1415 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1416 xprt_clear_connecting(xprt);
1417
1418 xprt->stat.connect_count++;
1419 xprt->stat.connect_time += (long)jiffies -
1420 xprt->stat.connect_start;
1421 xs_run_error_worker(transport, XPRT_SOCK_WAKE_PENDING);
1422 }
1423 break;
1424 case TCP_FIN_WAIT1:
1425 /* The client initiated a shutdown of the socket */
1426 xprt->connect_cookie++;
1427 xprt->reestablish_timeout = 0;
1428 set_bit(XPRT_CLOSING, &xprt->state);
1429 smp_mb__before_atomic();
1430 clear_bit(XPRT_CONNECTED, &xprt->state);
1431 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1432 smp_mb__after_atomic();
1433 break;
1434 case TCP_CLOSE_WAIT:
1435 /* The server initiated a shutdown of the socket */
1436 xprt->connect_cookie++;
1437 clear_bit(XPRT_CONNECTED, &xprt->state);
1438 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1439 fallthrough;
1440 case TCP_CLOSING:
1441 /*
1442 * If the server closed down the connection, make sure that
1443 * we back off before reconnecting
1444 */
1445 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1446 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1447 break;
1448 case TCP_LAST_ACK:
1449 set_bit(XPRT_CLOSING, &xprt->state);
1450 smp_mb__before_atomic();
1451 clear_bit(XPRT_CONNECTED, &xprt->state);
1452 smp_mb__after_atomic();
1453 break;
1454 case TCP_CLOSE:
1455 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1456 &transport->sock_state))
1457 xprt_clear_connecting(xprt);
1458 clear_bit(XPRT_CLOSING, &xprt->state);
1459 /* Trigger the socket release */
1460 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1461 }
1462 out:
1463 read_unlock_bh(&sk->sk_callback_lock);
1464}
1465
1466static void xs_write_space(struct sock *sk)
1467{
1468 struct socket_wq *wq;
1469 struct sock_xprt *transport;
1470 struct rpc_xprt *xprt;
1471
1472 if (!sk->sk_socket)
1473 return;
1474 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1475
1476 if (unlikely(!(xprt = xprt_from_sock(sk))))
1477 return;
1478 transport = container_of(xprt, struct sock_xprt, xprt);
1479 rcu_read_lock();
1480 wq = rcu_dereference(sk->sk_wq);
1481 if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1482 goto out;
1483
1484 xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE);
1485 sk->sk_write_pending--;
1486out:
1487 rcu_read_unlock();
1488}
1489
1490/**
1491 * xs_udp_write_space - callback invoked when socket buffer space
1492 * becomes available
1493 * @sk: socket whose state has changed
1494 *
1495 * Called when more output buffer space is available for this socket.
1496 * We try not to wake our writers until they can make "significant"
1497 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1498 * with a bunch of small requests.
1499 */
1500static void xs_udp_write_space(struct sock *sk)
1501{
1502 read_lock_bh(&sk->sk_callback_lock);
1503
1504 /* from net/core/sock.c:sock_def_write_space */
1505 if (sock_writeable(sk))
1506 xs_write_space(sk);
1507
1508 read_unlock_bh(&sk->sk_callback_lock);
1509}
1510
1511/**
1512 * xs_tcp_write_space - callback invoked when socket buffer space
1513 * becomes available
1514 * @sk: socket whose state has changed
1515 *
1516 * Called when more output buffer space is available for this socket.
1517 * We try not to wake our writers until they can make "significant"
1518 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1519 * with a bunch of small requests.
1520 */
1521static void xs_tcp_write_space(struct sock *sk)
1522{
1523 read_lock_bh(&sk->sk_callback_lock);
1524
1525 /* from net/core/stream.c:sk_stream_write_space */
1526 if (sk_stream_is_writeable(sk))
1527 xs_write_space(sk);
1528
1529 read_unlock_bh(&sk->sk_callback_lock);
1530}
1531
1532static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1533{
1534 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1535 struct sock *sk = transport->inet;
1536
1537 if (transport->rcvsize) {
1538 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1539 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1540 }
1541 if (transport->sndsize) {
1542 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1543 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1544 sk->sk_write_space(sk);
1545 }
1546}
1547
1548/**
1549 * xs_udp_set_buffer_size - set send and receive limits
1550 * @xprt: generic transport
1551 * @sndsize: requested size of send buffer, in bytes
1552 * @rcvsize: requested size of receive buffer, in bytes
1553 *
1554 * Set socket send and receive buffer size limits.
1555 */
1556static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1557{
1558 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1559
1560 transport->sndsize = 0;
1561 if (sndsize)
1562 transport->sndsize = sndsize + 1024;
1563 transport->rcvsize = 0;
1564 if (rcvsize)
1565 transport->rcvsize = rcvsize + 1024;
1566
1567 xs_udp_do_set_buffer_size(xprt);
1568}
1569
1570/**
1571 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1572 * @xprt: controlling transport
1573 * @task: task that timed out
1574 *
1575 * Adjust the congestion window after a retransmit timeout has occurred.
1576 */
1577static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1578{
1579 spin_lock(&xprt->transport_lock);
1580 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1581 spin_unlock(&xprt->transport_lock);
1582}
1583
1584static int xs_get_random_port(void)
1585{
1586 unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1587 unsigned short range;
1588 unsigned short rand;
1589
1590 if (max < min)
1591 return -EADDRINUSE;
1592 range = max - min + 1;
1593 rand = (unsigned short) prandom_u32() % range;
1594 return rand + min;
1595}
1596
1597static unsigned short xs_sock_getport(struct socket *sock)
1598{
1599 struct sockaddr_storage buf;
1600 unsigned short port = 0;
1601
1602 if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1603 goto out;
1604 switch (buf.ss_family) {
1605 case AF_INET6:
1606 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1607 break;
1608 case AF_INET:
1609 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1610 }
1611out:
1612 return port;
1613}
1614
1615/**
1616 * xs_set_port - reset the port number in the remote endpoint address
1617 * @xprt: generic transport
1618 * @port: new port number
1619 *
1620 */
1621static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1622{
1623 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1624
1625 rpc_set_port(xs_addr(xprt), port);
1626 xs_update_peer_port(xprt);
1627}
1628
1629static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1630{
1631 if (transport->srcport == 0 && transport->xprt.reuseport)
1632 transport->srcport = xs_sock_getport(sock);
1633}
1634
1635static int xs_get_srcport(struct sock_xprt *transport)
1636{
1637 int port = transport->srcport;
1638
1639 if (port == 0 && transport->xprt.resvport)
1640 port = xs_get_random_port();
1641 return port;
1642}
1643
1644static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1645{
1646 if (transport->srcport != 0)
1647 transport->srcport = 0;
1648 if (!transport->xprt.resvport)
1649 return 0;
1650 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1651 return xprt_max_resvport;
1652 return --port;
1653}
1654static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1655{
1656 struct sockaddr_storage myaddr;
1657 int err, nloop = 0;
1658 int port = xs_get_srcport(transport);
1659 unsigned short last;
1660
1661 /*
1662 * If we are asking for any ephemeral port (i.e. port == 0 &&
1663 * transport->xprt.resvport == 0), don't bind. Let the local
1664 * port selection happen implicitly when the socket is used
1665 * (for example at connect time).
1666 *
1667 * This ensures that we can continue to establish TCP
1668 * connections even when all local ephemeral ports are already
1669 * a part of some TCP connection. This makes no difference
1670 * for UDP sockets, but also doens't harm them.
1671 *
1672 * If we're asking for any reserved port (i.e. port == 0 &&
1673 * transport->xprt.resvport == 1) xs_get_srcport above will
1674 * ensure that port is non-zero and we will bind as needed.
1675 */
1676 if (port <= 0)
1677 return port;
1678
1679 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1680 do {
1681 rpc_set_port((struct sockaddr *)&myaddr, port);
1682 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1683 transport->xprt.addrlen);
1684 if (err == 0) {
1685 transport->srcport = port;
1686 break;
1687 }
1688 last = port;
1689 port = xs_next_srcport(transport, port);
1690 if (port > last)
1691 nloop++;
1692 } while (err == -EADDRINUSE && nloop != 2);
1693
1694 if (myaddr.ss_family == AF_INET)
1695 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1696 &((struct sockaddr_in *)&myaddr)->sin_addr,
1697 port, err ? "failed" : "ok", err);
1698 else
1699 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1700 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1701 port, err ? "failed" : "ok", err);
1702 return err;
1703}
1704
1705/*
1706 * We don't support autobind on AF_LOCAL sockets
1707 */
1708static void xs_local_rpcbind(struct rpc_task *task)
1709{
1710 xprt_set_bound(task->tk_xprt);
1711}
1712
1713static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1714{
1715}
1716
1717#ifdef CONFIG_DEBUG_LOCK_ALLOC
1718static struct lock_class_key xs_key[2];
1719static struct lock_class_key xs_slock_key[2];
1720
1721static inline void xs_reclassify_socketu(struct socket *sock)
1722{
1723 struct sock *sk = sock->sk;
1724
1725 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1726 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1727}
1728
1729static inline void xs_reclassify_socket4(struct socket *sock)
1730{
1731 struct sock *sk = sock->sk;
1732
1733 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1734 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1735}
1736
1737static inline void xs_reclassify_socket6(struct socket *sock)
1738{
1739 struct sock *sk = sock->sk;
1740
1741 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1742 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1743}
1744
1745static inline void xs_reclassify_socket(int family, struct socket *sock)
1746{
1747 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1748 return;
1749
1750 switch (family) {
1751 case AF_LOCAL:
1752 xs_reclassify_socketu(sock);
1753 break;
1754 case AF_INET:
1755 xs_reclassify_socket4(sock);
1756 break;
1757 case AF_INET6:
1758 xs_reclassify_socket6(sock);
1759 break;
1760 }
1761}
1762#else
1763static inline void xs_reclassify_socket(int family, struct socket *sock)
1764{
1765}
1766#endif
1767
1768static void xs_dummy_setup_socket(struct work_struct *work)
1769{
1770}
1771
1772static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1773 struct sock_xprt *transport, int family, int type,
1774 int protocol, bool reuseport)
1775{
1776 struct file *filp;
1777 struct socket *sock;
1778 int err;
1779
1780 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1781 if (err < 0) {
1782 dprintk("RPC: can't create %d transport socket (%d).\n",
1783 protocol, -err);
1784 goto out;
1785 }
1786 xs_reclassify_socket(family, sock);
1787
1788 if (reuseport)
1789 sock_set_reuseport(sock->sk);
1790
1791 err = xs_bind(transport, sock);
1792 if (err) {
1793 sock_release(sock);
1794 goto out;
1795 }
1796
1797 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1798 if (IS_ERR(filp))
1799 return ERR_CAST(filp);
1800 transport->file = filp;
1801
1802 return sock;
1803out:
1804 return ERR_PTR(err);
1805}
1806
1807static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1808 struct socket *sock)
1809{
1810 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1811 xprt);
1812
1813 if (!transport->inet) {
1814 struct sock *sk = sock->sk;
1815
1816 write_lock_bh(&sk->sk_callback_lock);
1817
1818 xs_save_old_callbacks(transport, sk);
1819
1820 sk->sk_user_data = xprt;
1821 sk->sk_data_ready = xs_data_ready;
1822 sk->sk_write_space = xs_udp_write_space;
1823 sock_set_flag(sk, SOCK_FASYNC);
1824 sk->sk_error_report = xs_error_report;
1825
1826 xprt_clear_connected(xprt);
1827
1828 /* Reset to new socket */
1829 transport->sock = sock;
1830 transport->inet = sk;
1831
1832 write_unlock_bh(&sk->sk_callback_lock);
1833 }
1834
1835 xs_stream_start_connect(transport);
1836
1837 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1838}
1839
1840/**
1841 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1842 * @transport: socket transport to connect
1843 */
1844static int xs_local_setup_socket(struct sock_xprt *transport)
1845{
1846 struct rpc_xprt *xprt = &transport->xprt;
1847 struct file *filp;
1848 struct socket *sock;
1849 int status;
1850
1851 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1852 SOCK_STREAM, 0, &sock, 1);
1853 if (status < 0) {
1854 dprintk("RPC: can't create AF_LOCAL "
1855 "transport socket (%d).\n", -status);
1856 goto out;
1857 }
1858 xs_reclassify_socket(AF_LOCAL, sock);
1859
1860 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1861 if (IS_ERR(filp)) {
1862 status = PTR_ERR(filp);
1863 goto out;
1864 }
1865 transport->file = filp;
1866
1867 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1868 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1869
1870 status = xs_local_finish_connecting(xprt, sock);
1871 trace_rpc_socket_connect(xprt, sock, status);
1872 switch (status) {
1873 case 0:
1874 dprintk("RPC: xprt %p connected to %s\n",
1875 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1876 xprt->stat.connect_count++;
1877 xprt->stat.connect_time += (long)jiffies -
1878 xprt->stat.connect_start;
1879 xprt_set_connected(xprt);
1880 case -ENOBUFS:
1881 break;
1882 case -ENOENT:
1883 dprintk("RPC: xprt %p: socket %s does not exist\n",
1884 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1885 break;
1886 case -ECONNREFUSED:
1887 dprintk("RPC: xprt %p: connection refused for %s\n",
1888 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1889 break;
1890 default:
1891 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1892 __func__, -status,
1893 xprt->address_strings[RPC_DISPLAY_ADDR]);
1894 }
1895
1896out:
1897 xprt_clear_connecting(xprt);
1898 xprt_wake_pending_tasks(xprt, status);
1899 return status;
1900}
1901
1902static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1903{
1904 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1905 int ret;
1906
1907 if (RPC_IS_ASYNC(task)) {
1908 /*
1909 * We want the AF_LOCAL connect to be resolved in the
1910 * filesystem namespace of the process making the rpc
1911 * call. Thus we connect synchronously.
1912 *
1913 * If we want to support asynchronous AF_LOCAL calls,
1914 * we'll need to figure out how to pass a namespace to
1915 * connect.
1916 */
1917 task->tk_rpc_status = -ENOTCONN;
1918 rpc_exit(task, -ENOTCONN);
1919 return;
1920 }
1921 ret = xs_local_setup_socket(transport);
1922 if (ret && !RPC_IS_SOFTCONN(task))
1923 msleep_interruptible(15000);
1924}
1925
1926#if IS_ENABLED(CONFIG_SUNRPC_SWAP)
1927/*
1928 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
1929 * know that we have exclusive access to the socket), to guard against
1930 * races with xs_reset_transport.
1931 */
1932static void xs_set_memalloc(struct rpc_xprt *xprt)
1933{
1934 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1935 xprt);
1936
1937 /*
1938 * If there's no sock, then we have nothing to set. The
1939 * reconnecting process will get it for us.
1940 */
1941 if (!transport->inet)
1942 return;
1943 if (atomic_read(&xprt->swapper))
1944 sk_set_memalloc(transport->inet);
1945}
1946
1947/**
1948 * xs_enable_swap - Tag this transport as being used for swap.
1949 * @xprt: transport to tag
1950 *
1951 * Take a reference to this transport on behalf of the rpc_clnt, and
1952 * optionally mark it for swapping if it wasn't already.
1953 */
1954static int
1955xs_enable_swap(struct rpc_xprt *xprt)
1956{
1957 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
1958
1959 if (atomic_inc_return(&xprt->swapper) != 1)
1960 return 0;
1961 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
1962 return -ERESTARTSYS;
1963 if (xs->inet)
1964 sk_set_memalloc(xs->inet);
1965 xprt_release_xprt(xprt, NULL);
1966 return 0;
1967}
1968
1969/**
1970 * xs_disable_swap - Untag this transport as being used for swap.
1971 * @xprt: transport to tag
1972 *
1973 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
1974 * swapper refcount goes to 0, untag the socket as a memalloc socket.
1975 */
1976static void
1977xs_disable_swap(struct rpc_xprt *xprt)
1978{
1979 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
1980
1981 if (!atomic_dec_and_test(&xprt->swapper))
1982 return;
1983 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
1984 return;
1985 if (xs->inet)
1986 sk_clear_memalloc(xs->inet);
1987 xprt_release_xprt(xprt, NULL);
1988}
1989#else
1990static void xs_set_memalloc(struct rpc_xprt *xprt)
1991{
1992}
1993
1994static int
1995xs_enable_swap(struct rpc_xprt *xprt)
1996{
1997 return -EINVAL;
1998}
1999
2000static void
2001xs_disable_swap(struct rpc_xprt *xprt)
2002{
2003}
2004#endif
2005
2006static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2007{
2008 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2009
2010 if (!transport->inet) {
2011 struct sock *sk = sock->sk;
2012
2013 write_lock_bh(&sk->sk_callback_lock);
2014
2015 xs_save_old_callbacks(transport, sk);
2016
2017 sk->sk_user_data = xprt;
2018 sk->sk_data_ready = xs_data_ready;
2019 sk->sk_write_space = xs_udp_write_space;
2020 sock_set_flag(sk, SOCK_FASYNC);
2021
2022 xprt_set_connected(xprt);
2023
2024 /* Reset to new socket */
2025 transport->sock = sock;
2026 transport->inet = sk;
2027
2028 xs_set_memalloc(xprt);
2029
2030 write_unlock_bh(&sk->sk_callback_lock);
2031 }
2032 xs_udp_do_set_buffer_size(xprt);
2033
2034 xprt->stat.connect_start = jiffies;
2035}
2036
2037static void xs_udp_setup_socket(struct work_struct *work)
2038{
2039 struct sock_xprt *transport =
2040 container_of(work, struct sock_xprt, connect_worker.work);
2041 struct rpc_xprt *xprt = &transport->xprt;
2042 struct socket *sock;
2043 int status = -EIO;
2044
2045 sock = xs_create_sock(xprt, transport,
2046 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2047 IPPROTO_UDP, false);
2048 if (IS_ERR(sock))
2049 goto out;
2050
2051 dprintk("RPC: worker connecting xprt %p via %s to "
2052 "%s (port %s)\n", xprt,
2053 xprt->address_strings[RPC_DISPLAY_PROTO],
2054 xprt->address_strings[RPC_DISPLAY_ADDR],
2055 xprt->address_strings[RPC_DISPLAY_PORT]);
2056
2057 xs_udp_finish_connecting(xprt, sock);
2058 trace_rpc_socket_connect(xprt, sock, 0);
2059 status = 0;
2060out:
2061 xprt_clear_connecting(xprt);
2062 xprt_unlock_connect(xprt, transport);
2063 xprt_wake_pending_tasks(xprt, status);
2064}
2065
2066/**
2067 * xs_tcp_shutdown - gracefully shut down a TCP socket
2068 * @xprt: transport
2069 *
2070 * Initiates a graceful shutdown of the TCP socket by calling the
2071 * equivalent of shutdown(SHUT_RDWR);
2072 */
2073static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2074{
2075 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2076 struct socket *sock = transport->sock;
2077 int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2078
2079 if (sock == NULL)
2080 return;
2081 switch (skst) {
2082 default:
2083 kernel_sock_shutdown(sock, SHUT_RDWR);
2084 trace_rpc_socket_shutdown(xprt, sock);
2085 break;
2086 case TCP_CLOSE:
2087 case TCP_TIME_WAIT:
2088 xs_reset_transport(transport);
2089 }
2090}
2091
2092static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2093 struct socket *sock)
2094{
2095 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2096 unsigned int keepidle;
2097 unsigned int keepcnt;
2098 unsigned int timeo;
2099
2100 spin_lock(&xprt->transport_lock);
2101 keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2102 keepcnt = xprt->timeout->to_retries + 1;
2103 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2104 (xprt->timeout->to_retries + 1);
2105 clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2106 spin_unlock(&xprt->transport_lock);
2107
2108 /* TCP Keepalive options */
2109 sock_set_keepalive(sock->sk);
2110 tcp_sock_set_keepidle(sock->sk, keepidle);
2111 tcp_sock_set_keepintvl(sock->sk, keepidle);
2112 tcp_sock_set_keepcnt(sock->sk, keepcnt);
2113
2114 /* TCP user timeout (see RFC5482) */
2115 tcp_sock_set_user_timeout(sock->sk, timeo);
2116}
2117
2118static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2119 unsigned long connect_timeout,
2120 unsigned long reconnect_timeout)
2121{
2122 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2123 struct rpc_timeout to;
2124 unsigned long initval;
2125
2126 spin_lock(&xprt->transport_lock);
2127 if (reconnect_timeout < xprt->max_reconnect_timeout)
2128 xprt->max_reconnect_timeout = reconnect_timeout;
2129 if (connect_timeout < xprt->connect_timeout) {
2130 memcpy(&to, xprt->timeout, sizeof(to));
2131 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2132 /* Arbitrary lower limit */
2133 if (initval < XS_TCP_INIT_REEST_TO << 1)
2134 initval = XS_TCP_INIT_REEST_TO << 1;
2135 to.to_initval = initval;
2136 to.to_maxval = initval;
2137 memcpy(&transport->tcp_timeout, &to,
2138 sizeof(transport->tcp_timeout));
2139 xprt->timeout = &transport->tcp_timeout;
2140 xprt->connect_timeout = connect_timeout;
2141 }
2142 set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2143 spin_unlock(&xprt->transport_lock);
2144}
2145
2146static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2147{
2148 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2149 int ret = -ENOTCONN;
2150
2151 if (!transport->inet) {
2152 struct sock *sk = sock->sk;
2153
2154 /* Avoid temporary address, they are bad for long-lived
2155 * connections such as NFS mounts.
2156 * RFC4941, section 3.6 suggests that:
2157 * Individual applications, which have specific
2158 * knowledge about the normal duration of connections,
2159 * MAY override this as appropriate.
2160 */
2161 if (xs_addr(xprt)->sa_family == PF_INET6) {
2162 ip6_sock_set_addr_preferences(sk,
2163 IPV6_PREFER_SRC_PUBLIC);
2164 }
2165
2166 xs_tcp_set_socket_timeouts(xprt, sock);
2167
2168 write_lock_bh(&sk->sk_callback_lock);
2169
2170 xs_save_old_callbacks(transport, sk);
2171
2172 sk->sk_user_data = xprt;
2173 sk->sk_data_ready = xs_data_ready;
2174 sk->sk_state_change = xs_tcp_state_change;
2175 sk->sk_write_space = xs_tcp_write_space;
2176 sock_set_flag(sk, SOCK_FASYNC);
2177 sk->sk_error_report = xs_error_report;
2178
2179 /* socket options */
2180 sock_reset_flag(sk, SOCK_LINGER);
2181 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2182
2183 xprt_clear_connected(xprt);
2184
2185 /* Reset to new socket */
2186 transport->sock = sock;
2187 transport->inet = sk;
2188
2189 write_unlock_bh(&sk->sk_callback_lock);
2190 }
2191
2192 if (!xprt_bound(xprt))
2193 goto out;
2194
2195 xs_set_memalloc(xprt);
2196
2197 xs_stream_start_connect(transport);
2198
2199 /* Tell the socket layer to start connecting... */
2200 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2201 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2202 switch (ret) {
2203 case 0:
2204 xs_set_srcport(transport, sock);
2205 fallthrough;
2206 case -EINPROGRESS:
2207 /* SYN_SENT! */
2208 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2209 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2210 break;
2211 case -EADDRNOTAVAIL:
2212 /* Source port number is unavailable. Try a new one! */
2213 transport->srcport = 0;
2214 }
2215out:
2216 return ret;
2217}
2218
2219/**
2220 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2221 * @work: queued work item
2222 *
2223 * Invoked by a work queue tasklet.
2224 */
2225static void xs_tcp_setup_socket(struct work_struct *work)
2226{
2227 struct sock_xprt *transport =
2228 container_of(work, struct sock_xprt, connect_worker.work);
2229 struct socket *sock = transport->sock;
2230 struct rpc_xprt *xprt = &transport->xprt;
2231 int status = -EIO;
2232
2233 if (!sock) {
2234 sock = xs_create_sock(xprt, transport,
2235 xs_addr(xprt)->sa_family, SOCK_STREAM,
2236 IPPROTO_TCP, true);
2237 if (IS_ERR(sock)) {
2238 status = PTR_ERR(sock);
2239 goto out;
2240 }
2241 }
2242
2243 dprintk("RPC: worker connecting xprt %p via %s to "
2244 "%s (port %s)\n", xprt,
2245 xprt->address_strings[RPC_DISPLAY_PROTO],
2246 xprt->address_strings[RPC_DISPLAY_ADDR],
2247 xprt->address_strings[RPC_DISPLAY_PORT]);
2248
2249 status = xs_tcp_finish_connecting(xprt, sock);
2250 trace_rpc_socket_connect(xprt, sock, status);
2251 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2252 xprt, -status, xprt_connected(xprt),
2253 sock->sk->sk_state);
2254 switch (status) {
2255 default:
2256 printk("%s: connect returned unhandled error %d\n",
2257 __func__, status);
2258 fallthrough;
2259 case -EADDRNOTAVAIL:
2260 /* We're probably in TIME_WAIT. Get rid of existing socket,
2261 * and retry
2262 */
2263 xs_tcp_force_close(xprt);
2264 break;
2265 case 0:
2266 case -EINPROGRESS:
2267 case -EALREADY:
2268 xprt_unlock_connect(xprt, transport);
2269 return;
2270 case -EINVAL:
2271 /* Happens, for instance, if the user specified a link
2272 * local IPv6 address without a scope-id.
2273 */
2274 case -ECONNREFUSED:
2275 case -ECONNRESET:
2276 case -ENETDOWN:
2277 case -ENETUNREACH:
2278 case -EHOSTUNREACH:
2279 case -EADDRINUSE:
2280 case -ENOBUFS:
2281 /*
2282 * xs_tcp_force_close() wakes tasks with -EIO.
2283 * We need to wake them first to ensure the
2284 * correct error code.
2285 */
2286 xprt_wake_pending_tasks(xprt, status);
2287 xs_tcp_force_close(xprt);
2288 goto out;
2289 }
2290 status = -EAGAIN;
2291out:
2292 xprt_clear_connecting(xprt);
2293 xprt_unlock_connect(xprt, transport);
2294 xprt_wake_pending_tasks(xprt, status);
2295}
2296
2297/**
2298 * xs_connect - connect a socket to a remote endpoint
2299 * @xprt: pointer to transport structure
2300 * @task: address of RPC task that manages state of connect request
2301 *
2302 * TCP: If the remote end dropped the connection, delay reconnecting.
2303 *
2304 * UDP socket connects are synchronous, but we use a work queue anyway
2305 * to guarantee that even unprivileged user processes can set up a
2306 * socket on a privileged port.
2307 *
2308 * If a UDP socket connect fails, the delay behavior here prevents
2309 * retry floods (hard mounts).
2310 */
2311static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2312{
2313 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2314 unsigned long delay = 0;
2315
2316 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2317
2318 if (transport->sock != NULL) {
2319 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2320 "seconds\n",
2321 xprt, xprt->reestablish_timeout / HZ);
2322
2323 /* Start by resetting any existing state */
2324 xs_reset_transport(transport);
2325
2326 delay = xprt_reconnect_delay(xprt);
2327 xprt_reconnect_backoff(xprt, XS_TCP_INIT_REEST_TO);
2328
2329 } else
2330 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2331
2332 queue_delayed_work(xprtiod_workqueue,
2333 &transport->connect_worker,
2334 delay);
2335}
2336
2337static void xs_wake_disconnect(struct sock_xprt *transport)
2338{
2339 if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state))
2340 xs_tcp_force_close(&transport->xprt);
2341}
2342
2343static void xs_wake_write(struct sock_xprt *transport)
2344{
2345 if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state))
2346 xprt_write_space(&transport->xprt);
2347}
2348
2349static void xs_wake_error(struct sock_xprt *transport)
2350{
2351 int sockerr;
2352
2353 if (!test_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2354 return;
2355 mutex_lock(&transport->recv_mutex);
2356 if (transport->sock == NULL)
2357 goto out;
2358 if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2359 goto out;
2360 sockerr = xchg(&transport->xprt_err, 0);
2361 if (sockerr < 0)
2362 xprt_wake_pending_tasks(&transport->xprt, sockerr);
2363out:
2364 mutex_unlock(&transport->recv_mutex);
2365}
2366
2367static void xs_wake_pending(struct sock_xprt *transport)
2368{
2369 if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING, &transport->sock_state))
2370 xprt_wake_pending_tasks(&transport->xprt, -EAGAIN);
2371}
2372
2373static void xs_error_handle(struct work_struct *work)
2374{
2375 struct sock_xprt *transport = container_of(work,
2376 struct sock_xprt, error_worker);
2377
2378 xs_wake_disconnect(transport);
2379 xs_wake_write(transport);
2380 xs_wake_error(transport);
2381 xs_wake_pending(transport);
2382}
2383
2384/**
2385 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2386 * @xprt: rpc_xprt struct containing statistics
2387 * @seq: output file
2388 *
2389 */
2390static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2391{
2392 long idle_time = 0;
2393
2394 if (xprt_connected(xprt))
2395 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2396
2397 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2398 "%llu %llu %lu %llu %llu\n",
2399 xprt->stat.bind_count,
2400 xprt->stat.connect_count,
2401 xprt->stat.connect_time / HZ,
2402 idle_time,
2403 xprt->stat.sends,
2404 xprt->stat.recvs,
2405 xprt->stat.bad_xids,
2406 xprt->stat.req_u,
2407 xprt->stat.bklog_u,
2408 xprt->stat.max_slots,
2409 xprt->stat.sending_u,
2410 xprt->stat.pending_u);
2411}
2412
2413/**
2414 * xs_udp_print_stats - display UDP socket-specifc stats
2415 * @xprt: rpc_xprt struct containing statistics
2416 * @seq: output file
2417 *
2418 */
2419static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2420{
2421 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2422
2423 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2424 "%lu %llu %llu\n",
2425 transport->srcport,
2426 xprt->stat.bind_count,
2427 xprt->stat.sends,
2428 xprt->stat.recvs,
2429 xprt->stat.bad_xids,
2430 xprt->stat.req_u,
2431 xprt->stat.bklog_u,
2432 xprt->stat.max_slots,
2433 xprt->stat.sending_u,
2434 xprt->stat.pending_u);
2435}
2436
2437/**
2438 * xs_tcp_print_stats - display TCP socket-specifc stats
2439 * @xprt: rpc_xprt struct containing statistics
2440 * @seq: output file
2441 *
2442 */
2443static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2444{
2445 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2446 long idle_time = 0;
2447
2448 if (xprt_connected(xprt))
2449 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2450
2451 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2452 "%llu %llu %lu %llu %llu\n",
2453 transport->srcport,
2454 xprt->stat.bind_count,
2455 xprt->stat.connect_count,
2456 xprt->stat.connect_time / HZ,
2457 idle_time,
2458 xprt->stat.sends,
2459 xprt->stat.recvs,
2460 xprt->stat.bad_xids,
2461 xprt->stat.req_u,
2462 xprt->stat.bklog_u,
2463 xprt->stat.max_slots,
2464 xprt->stat.sending_u,
2465 xprt->stat.pending_u);
2466}
2467
2468/*
2469 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2470 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2471 * to use the server side send routines.
2472 */
2473static int bc_malloc(struct rpc_task *task)
2474{
2475 struct rpc_rqst *rqst = task->tk_rqstp;
2476 size_t size = rqst->rq_callsize;
2477 struct page *page;
2478 struct rpc_buffer *buf;
2479
2480 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2481 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2482 size);
2483 return -EINVAL;
2484 }
2485
2486 page = alloc_page(GFP_KERNEL);
2487 if (!page)
2488 return -ENOMEM;
2489
2490 buf = page_address(page);
2491 buf->len = PAGE_SIZE;
2492
2493 rqst->rq_buffer = buf->data;
2494 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2495 return 0;
2496}
2497
2498/*
2499 * Free the space allocated in the bc_alloc routine
2500 */
2501static void bc_free(struct rpc_task *task)
2502{
2503 void *buffer = task->tk_rqstp->rq_buffer;
2504 struct rpc_buffer *buf;
2505
2506 buf = container_of(buffer, struct rpc_buffer, data);
2507 free_page((unsigned long)buf);
2508}
2509
2510static int bc_sendto(struct rpc_rqst *req)
2511{
2512 struct xdr_buf *xdr = &req->rq_snd_buf;
2513 struct sock_xprt *transport =
2514 container_of(req->rq_xprt, struct sock_xprt, xprt);
2515 struct msghdr msg = {
2516 .msg_flags = 0,
2517 };
2518 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
2519 (u32)xdr->len);
2520 unsigned int sent = 0;
2521 int err;
2522
2523 req->rq_xtime = ktime_get();
2524 err = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, marker, &sent);
2525 xdr_free_bvec(xdr);
2526 if (err < 0 || sent != (xdr->len + sizeof(marker)))
2527 return -EAGAIN;
2528 return sent;
2529}
2530
2531/**
2532 * bc_send_request - Send a backchannel Call on a TCP socket
2533 * @req: rpc_rqst containing Call message to be sent
2534 *
2535 * xpt_mutex ensures @rqstp's whole message is written to the socket
2536 * without interruption.
2537 *
2538 * Return values:
2539 * %0 if the message was sent successfully
2540 * %ENOTCONN if the message was not sent
2541 */
2542static int bc_send_request(struct rpc_rqst *req)
2543{
2544 struct svc_xprt *xprt;
2545 int len;
2546
2547 /*
2548 * Get the server socket associated with this callback xprt
2549 */
2550 xprt = req->rq_xprt->bc_xprt;
2551
2552 /*
2553 * Grab the mutex to serialize data as the connection is shared
2554 * with the fore channel
2555 */
2556 mutex_lock(&xprt->xpt_mutex);
2557 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2558 len = -ENOTCONN;
2559 else
2560 len = bc_sendto(req);
2561 mutex_unlock(&xprt->xpt_mutex);
2562
2563 if (len > 0)
2564 len = 0;
2565
2566 return len;
2567}
2568
2569/*
2570 * The close routine. Since this is client initiated, we do nothing
2571 */
2572
2573static void bc_close(struct rpc_xprt *xprt)
2574{
2575 xprt_disconnect_done(xprt);
2576}
2577
2578/*
2579 * The xprt destroy routine. Again, because this connection is client
2580 * initiated, we do nothing
2581 */
2582
2583static void bc_destroy(struct rpc_xprt *xprt)
2584{
2585 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2586
2587 xs_xprt_free(xprt);
2588 module_put(THIS_MODULE);
2589}
2590
2591static const struct rpc_xprt_ops xs_local_ops = {
2592 .reserve_xprt = xprt_reserve_xprt,
2593 .release_xprt = xprt_release_xprt,
2594 .alloc_slot = xprt_alloc_slot,
2595 .free_slot = xprt_free_slot,
2596 .rpcbind = xs_local_rpcbind,
2597 .set_port = xs_local_set_port,
2598 .connect = xs_local_connect,
2599 .buf_alloc = rpc_malloc,
2600 .buf_free = rpc_free,
2601 .prepare_request = xs_stream_prepare_request,
2602 .send_request = xs_local_send_request,
2603 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2604 .close = xs_close,
2605 .destroy = xs_destroy,
2606 .print_stats = xs_local_print_stats,
2607 .enable_swap = xs_enable_swap,
2608 .disable_swap = xs_disable_swap,
2609};
2610
2611static const struct rpc_xprt_ops xs_udp_ops = {
2612 .set_buffer_size = xs_udp_set_buffer_size,
2613 .reserve_xprt = xprt_reserve_xprt_cong,
2614 .release_xprt = xprt_release_xprt_cong,
2615 .alloc_slot = xprt_alloc_slot,
2616 .free_slot = xprt_free_slot,
2617 .rpcbind = rpcb_getport_async,
2618 .set_port = xs_set_port,
2619 .connect = xs_connect,
2620 .buf_alloc = rpc_malloc,
2621 .buf_free = rpc_free,
2622 .send_request = xs_udp_send_request,
2623 .wait_for_reply_request = xprt_wait_for_reply_request_rtt,
2624 .timer = xs_udp_timer,
2625 .release_request = xprt_release_rqst_cong,
2626 .close = xs_close,
2627 .destroy = xs_destroy,
2628 .print_stats = xs_udp_print_stats,
2629 .enable_swap = xs_enable_swap,
2630 .disable_swap = xs_disable_swap,
2631 .inject_disconnect = xs_inject_disconnect,
2632};
2633
2634static const struct rpc_xprt_ops xs_tcp_ops = {
2635 .reserve_xprt = xprt_reserve_xprt,
2636 .release_xprt = xprt_release_xprt,
2637 .alloc_slot = xprt_alloc_slot,
2638 .free_slot = xprt_free_slot,
2639 .rpcbind = rpcb_getport_async,
2640 .set_port = xs_set_port,
2641 .connect = xs_connect,
2642 .buf_alloc = rpc_malloc,
2643 .buf_free = rpc_free,
2644 .prepare_request = xs_stream_prepare_request,
2645 .send_request = xs_tcp_send_request,
2646 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2647 .close = xs_tcp_shutdown,
2648 .destroy = xs_destroy,
2649 .set_connect_timeout = xs_tcp_set_connect_timeout,
2650 .print_stats = xs_tcp_print_stats,
2651 .enable_swap = xs_enable_swap,
2652 .disable_swap = xs_disable_swap,
2653 .inject_disconnect = xs_inject_disconnect,
2654#ifdef CONFIG_SUNRPC_BACKCHANNEL
2655 .bc_setup = xprt_setup_bc,
2656 .bc_maxpayload = xs_tcp_bc_maxpayload,
2657 .bc_num_slots = xprt_bc_max_slots,
2658 .bc_free_rqst = xprt_free_bc_rqst,
2659 .bc_destroy = xprt_destroy_bc,
2660#endif
2661};
2662
2663/*
2664 * The rpc_xprt_ops for the server backchannel
2665 */
2666
2667static const struct rpc_xprt_ops bc_tcp_ops = {
2668 .reserve_xprt = xprt_reserve_xprt,
2669 .release_xprt = xprt_release_xprt,
2670 .alloc_slot = xprt_alloc_slot,
2671 .free_slot = xprt_free_slot,
2672 .buf_alloc = bc_malloc,
2673 .buf_free = bc_free,
2674 .send_request = bc_send_request,
2675 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2676 .close = bc_close,
2677 .destroy = bc_destroy,
2678 .print_stats = xs_tcp_print_stats,
2679 .enable_swap = xs_enable_swap,
2680 .disable_swap = xs_disable_swap,
2681 .inject_disconnect = xs_inject_disconnect,
2682};
2683
2684static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2685{
2686 static const struct sockaddr_in sin = {
2687 .sin_family = AF_INET,
2688 .sin_addr.s_addr = htonl(INADDR_ANY),
2689 };
2690 static const struct sockaddr_in6 sin6 = {
2691 .sin6_family = AF_INET6,
2692 .sin6_addr = IN6ADDR_ANY_INIT,
2693 };
2694
2695 switch (family) {
2696 case AF_LOCAL:
2697 break;
2698 case AF_INET:
2699 memcpy(sap, &sin, sizeof(sin));
2700 break;
2701 case AF_INET6:
2702 memcpy(sap, &sin6, sizeof(sin6));
2703 break;
2704 default:
2705 dprintk("RPC: %s: Bad address family\n", __func__);
2706 return -EAFNOSUPPORT;
2707 }
2708 return 0;
2709}
2710
2711static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2712 unsigned int slot_table_size,
2713 unsigned int max_slot_table_size)
2714{
2715 struct rpc_xprt *xprt;
2716 struct sock_xprt *new;
2717
2718 if (args->addrlen > sizeof(xprt->addr)) {
2719 dprintk("RPC: xs_setup_xprt: address too large\n");
2720 return ERR_PTR(-EBADF);
2721 }
2722
2723 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2724 max_slot_table_size);
2725 if (xprt == NULL) {
2726 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2727 "rpc_xprt\n");
2728 return ERR_PTR(-ENOMEM);
2729 }
2730
2731 new = container_of(xprt, struct sock_xprt, xprt);
2732 mutex_init(&new->recv_mutex);
2733 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2734 xprt->addrlen = args->addrlen;
2735 if (args->srcaddr)
2736 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2737 else {
2738 int err;
2739 err = xs_init_anyaddr(args->dstaddr->sa_family,
2740 (struct sockaddr *)&new->srcaddr);
2741 if (err != 0) {
2742 xprt_free(xprt);
2743 return ERR_PTR(err);
2744 }
2745 }
2746
2747 return xprt;
2748}
2749
2750static const struct rpc_timeout xs_local_default_timeout = {
2751 .to_initval = 10 * HZ,
2752 .to_maxval = 10 * HZ,
2753 .to_retries = 2,
2754};
2755
2756/**
2757 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2758 * @args: rpc transport creation arguments
2759 *
2760 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2761 */
2762static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2763{
2764 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2765 struct sock_xprt *transport;
2766 struct rpc_xprt *xprt;
2767 struct rpc_xprt *ret;
2768
2769 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2770 xprt_max_tcp_slot_table_entries);
2771 if (IS_ERR(xprt))
2772 return xprt;
2773 transport = container_of(xprt, struct sock_xprt, xprt);
2774
2775 xprt->prot = 0;
2776 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2777
2778 xprt->bind_timeout = XS_BIND_TO;
2779 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2780 xprt->idle_timeout = XS_IDLE_DISC_TO;
2781
2782 xprt->ops = &xs_local_ops;
2783 xprt->timeout = &xs_local_default_timeout;
2784
2785 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2786 INIT_WORK(&transport->error_worker, xs_error_handle);
2787 INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket);
2788
2789 switch (sun->sun_family) {
2790 case AF_LOCAL:
2791 if (sun->sun_path[0] != '/') {
2792 dprintk("RPC: bad AF_LOCAL address: %s\n",
2793 sun->sun_path);
2794 ret = ERR_PTR(-EINVAL);
2795 goto out_err;
2796 }
2797 xprt_set_bound(xprt);
2798 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2799 ret = ERR_PTR(xs_local_setup_socket(transport));
2800 if (ret)
2801 goto out_err;
2802 break;
2803 default:
2804 ret = ERR_PTR(-EAFNOSUPPORT);
2805 goto out_err;
2806 }
2807
2808 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2809 xprt->address_strings[RPC_DISPLAY_ADDR]);
2810
2811 if (try_module_get(THIS_MODULE))
2812 return xprt;
2813 ret = ERR_PTR(-EINVAL);
2814out_err:
2815 xs_xprt_free(xprt);
2816 return ret;
2817}
2818
2819static const struct rpc_timeout xs_udp_default_timeout = {
2820 .to_initval = 5 * HZ,
2821 .to_maxval = 30 * HZ,
2822 .to_increment = 5 * HZ,
2823 .to_retries = 5,
2824};
2825
2826/**
2827 * xs_setup_udp - Set up transport to use a UDP socket
2828 * @args: rpc transport creation arguments
2829 *
2830 */
2831static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2832{
2833 struct sockaddr *addr = args->dstaddr;
2834 struct rpc_xprt *xprt;
2835 struct sock_xprt *transport;
2836 struct rpc_xprt *ret;
2837
2838 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2839 xprt_udp_slot_table_entries);
2840 if (IS_ERR(xprt))
2841 return xprt;
2842 transport = container_of(xprt, struct sock_xprt, xprt);
2843
2844 xprt->prot = IPPROTO_UDP;
2845 /* XXX: header size can vary due to auth type, IPv6, etc. */
2846 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2847
2848 xprt->bind_timeout = XS_BIND_TO;
2849 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2850 xprt->idle_timeout = XS_IDLE_DISC_TO;
2851
2852 xprt->ops = &xs_udp_ops;
2853
2854 xprt->timeout = &xs_udp_default_timeout;
2855
2856 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
2857 INIT_WORK(&transport->error_worker, xs_error_handle);
2858 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2859
2860 switch (addr->sa_family) {
2861 case AF_INET:
2862 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2863 xprt_set_bound(xprt);
2864
2865 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2866 break;
2867 case AF_INET6:
2868 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2869 xprt_set_bound(xprt);
2870
2871 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2872 break;
2873 default:
2874 ret = ERR_PTR(-EAFNOSUPPORT);
2875 goto out_err;
2876 }
2877
2878 if (xprt_bound(xprt))
2879 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2880 xprt->address_strings[RPC_DISPLAY_ADDR],
2881 xprt->address_strings[RPC_DISPLAY_PORT],
2882 xprt->address_strings[RPC_DISPLAY_PROTO]);
2883 else
2884 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2885 xprt->address_strings[RPC_DISPLAY_ADDR],
2886 xprt->address_strings[RPC_DISPLAY_PROTO]);
2887
2888 if (try_module_get(THIS_MODULE))
2889 return xprt;
2890 ret = ERR_PTR(-EINVAL);
2891out_err:
2892 xs_xprt_free(xprt);
2893 return ret;
2894}
2895
2896static const struct rpc_timeout xs_tcp_default_timeout = {
2897 .to_initval = 60 * HZ,
2898 .to_maxval = 60 * HZ,
2899 .to_retries = 2,
2900};
2901
2902/**
2903 * xs_setup_tcp - Set up transport to use a TCP socket
2904 * @args: rpc transport creation arguments
2905 *
2906 */
2907static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2908{
2909 struct sockaddr *addr = args->dstaddr;
2910 struct rpc_xprt *xprt;
2911 struct sock_xprt *transport;
2912 struct rpc_xprt *ret;
2913 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2914
2915 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2916 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2917
2918 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2919 max_slot_table_size);
2920 if (IS_ERR(xprt))
2921 return xprt;
2922 transport = container_of(xprt, struct sock_xprt, xprt);
2923
2924 xprt->prot = IPPROTO_TCP;
2925 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2926
2927 xprt->bind_timeout = XS_BIND_TO;
2928 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2929 xprt->idle_timeout = XS_IDLE_DISC_TO;
2930
2931 xprt->ops = &xs_tcp_ops;
2932 xprt->timeout = &xs_tcp_default_timeout;
2933
2934 xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
2935 xprt->connect_timeout = xprt->timeout->to_initval *
2936 (xprt->timeout->to_retries + 1);
2937
2938 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2939 INIT_WORK(&transport->error_worker, xs_error_handle);
2940 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
2941
2942 switch (addr->sa_family) {
2943 case AF_INET:
2944 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2945 xprt_set_bound(xprt);
2946
2947 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2948 break;
2949 case AF_INET6:
2950 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2951 xprt_set_bound(xprt);
2952
2953 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2954 break;
2955 default:
2956 ret = ERR_PTR(-EAFNOSUPPORT);
2957 goto out_err;
2958 }
2959
2960 if (xprt_bound(xprt))
2961 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2962 xprt->address_strings[RPC_DISPLAY_ADDR],
2963 xprt->address_strings[RPC_DISPLAY_PORT],
2964 xprt->address_strings[RPC_DISPLAY_PROTO]);
2965 else
2966 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2967 xprt->address_strings[RPC_DISPLAY_ADDR],
2968 xprt->address_strings[RPC_DISPLAY_PROTO]);
2969
2970 if (try_module_get(THIS_MODULE))
2971 return xprt;
2972 ret = ERR_PTR(-EINVAL);
2973out_err:
2974 xs_xprt_free(xprt);
2975 return ret;
2976}
2977
2978/**
2979 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2980 * @args: rpc transport creation arguments
2981 *
2982 */
2983static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2984{
2985 struct sockaddr *addr = args->dstaddr;
2986 struct rpc_xprt *xprt;
2987 struct sock_xprt *transport;
2988 struct svc_sock *bc_sock;
2989 struct rpc_xprt *ret;
2990
2991 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2992 xprt_tcp_slot_table_entries);
2993 if (IS_ERR(xprt))
2994 return xprt;
2995 transport = container_of(xprt, struct sock_xprt, xprt);
2996
2997 xprt->prot = IPPROTO_TCP;
2998 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2999 xprt->timeout = &xs_tcp_default_timeout;
3000
3001 /* backchannel */
3002 xprt_set_bound(xprt);
3003 xprt->bind_timeout = 0;
3004 xprt->reestablish_timeout = 0;
3005 xprt->idle_timeout = 0;
3006
3007 xprt->ops = &bc_tcp_ops;
3008
3009 switch (addr->sa_family) {
3010 case AF_INET:
3011 xs_format_peer_addresses(xprt, "tcp",
3012 RPCBIND_NETID_TCP);
3013 break;
3014 case AF_INET6:
3015 xs_format_peer_addresses(xprt, "tcp",
3016 RPCBIND_NETID_TCP6);
3017 break;
3018 default:
3019 ret = ERR_PTR(-EAFNOSUPPORT);
3020 goto out_err;
3021 }
3022
3023 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3024 xprt->address_strings[RPC_DISPLAY_ADDR],
3025 xprt->address_strings[RPC_DISPLAY_PORT],
3026 xprt->address_strings[RPC_DISPLAY_PROTO]);
3027
3028 /*
3029 * Once we've associated a backchannel xprt with a connection,
3030 * we want to keep it around as long as the connection lasts,
3031 * in case we need to start using it for a backchannel again;
3032 * this reference won't be dropped until bc_xprt is destroyed.
3033 */
3034 xprt_get(xprt);
3035 args->bc_xprt->xpt_bc_xprt = xprt;
3036 xprt->bc_xprt = args->bc_xprt;
3037 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3038 transport->sock = bc_sock->sk_sock;
3039 transport->inet = bc_sock->sk_sk;
3040
3041 /*
3042 * Since we don't want connections for the backchannel, we set
3043 * the xprt status to connected
3044 */
3045 xprt_set_connected(xprt);
3046
3047 if (try_module_get(THIS_MODULE))
3048 return xprt;
3049
3050 args->bc_xprt->xpt_bc_xprt = NULL;
3051 args->bc_xprt->xpt_bc_xps = NULL;
3052 xprt_put(xprt);
3053 ret = ERR_PTR(-EINVAL);
3054out_err:
3055 xs_xprt_free(xprt);
3056 return ret;
3057}
3058
3059static struct xprt_class xs_local_transport = {
3060 .list = LIST_HEAD_INIT(xs_local_transport.list),
3061 .name = "named UNIX socket",
3062 .owner = THIS_MODULE,
3063 .ident = XPRT_TRANSPORT_LOCAL,
3064 .setup = xs_setup_local,
3065};
3066
3067static struct xprt_class xs_udp_transport = {
3068 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3069 .name = "udp",
3070 .owner = THIS_MODULE,
3071 .ident = XPRT_TRANSPORT_UDP,
3072 .setup = xs_setup_udp,
3073};
3074
3075static struct xprt_class xs_tcp_transport = {
3076 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3077 .name = "tcp",
3078 .owner = THIS_MODULE,
3079 .ident = XPRT_TRANSPORT_TCP,
3080 .setup = xs_setup_tcp,
3081};
3082
3083static struct xprt_class xs_bc_tcp_transport = {
3084 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3085 .name = "tcp NFSv4.1 backchannel",
3086 .owner = THIS_MODULE,
3087 .ident = XPRT_TRANSPORT_BC_TCP,
3088 .setup = xs_setup_bc_tcp,
3089};
3090
3091/**
3092 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3093 *
3094 */
3095int init_socket_xprt(void)
3096{
3097 if (!sunrpc_table_header)
3098 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3099
3100 xprt_register_transport(&xs_local_transport);
3101 xprt_register_transport(&xs_udp_transport);
3102 xprt_register_transport(&xs_tcp_transport);
3103 xprt_register_transport(&xs_bc_tcp_transport);
3104
3105 return 0;
3106}
3107
3108/**
3109 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3110 *
3111 */
3112void cleanup_socket_xprt(void)
3113{
3114 if (sunrpc_table_header) {
3115 unregister_sysctl_table(sunrpc_table_header);
3116 sunrpc_table_header = NULL;
3117 }
3118
3119 xprt_unregister_transport(&xs_local_transport);
3120 xprt_unregister_transport(&xs_udp_transport);
3121 xprt_unregister_transport(&xs_tcp_transport);
3122 xprt_unregister_transport(&xs_bc_tcp_transport);
3123}
3124
3125static int param_set_uint_minmax(const char *val,
3126 const struct kernel_param *kp,
3127 unsigned int min, unsigned int max)
3128{
3129 unsigned int num;
3130 int ret;
3131
3132 if (!val)
3133 return -EINVAL;
3134 ret = kstrtouint(val, 0, &num);
3135 if (ret)
3136 return ret;
3137 if (num < min || num > max)
3138 return -EINVAL;
3139 *((unsigned int *)kp->arg) = num;
3140 return 0;
3141}
3142
3143static int param_set_portnr(const char *val, const struct kernel_param *kp)
3144{
3145 return param_set_uint_minmax(val, kp,
3146 RPC_MIN_RESVPORT,
3147 RPC_MAX_RESVPORT);
3148}
3149
3150static const struct kernel_param_ops param_ops_portnr = {
3151 .set = param_set_portnr,
3152 .get = param_get_uint,
3153};
3154
3155#define param_check_portnr(name, p) \
3156 __param_check(name, p, unsigned int);
3157
3158module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3159module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3160
3161static int param_set_slot_table_size(const char *val,
3162 const struct kernel_param *kp)
3163{
3164 return param_set_uint_minmax(val, kp,
3165 RPC_MIN_SLOT_TABLE,
3166 RPC_MAX_SLOT_TABLE);
3167}
3168
3169static const struct kernel_param_ops param_ops_slot_table_size = {
3170 .set = param_set_slot_table_size,
3171 .get = param_get_uint,
3172};
3173
3174#define param_check_slot_table_size(name, p) \
3175 __param_check(name, p, unsigned int);
3176
3177static int param_set_max_slot_table_size(const char *val,
3178 const struct kernel_param *kp)
3179{
3180 return param_set_uint_minmax(val, kp,
3181 RPC_MIN_SLOT_TABLE,
3182 RPC_MAX_SLOT_TABLE_LIMIT);
3183}
3184
3185static const struct kernel_param_ops param_ops_max_slot_table_size = {
3186 .set = param_set_max_slot_table_size,
3187 .get = param_get_uint,
3188};
3189
3190#define param_check_max_slot_table_size(name, p) \
3191 __param_check(name, p, unsigned int);
3192
3193module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3194 slot_table_size, 0644);
3195module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3196 max_slot_table_size, 0644);
3197module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3198 slot_table_size, 0644);
1/*
2 * linux/net/sunrpc/xprtsock.c
3 *
4 * Client-side transport implementation for sockets.
5 *
6 * TCP callback races fixes (C) 1998 Red Hat
7 * TCP send fixes (C) 1998 Red Hat
8 * TCP NFS related read + write fixes
9 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
10 *
11 * Rewrite of larges part of the code in order to stabilize TCP stuff.
12 * Fix behaviour when socket buffer is full.
13 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
14 *
15 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
16 *
17 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
18 * <gilles.quillard@bull.net>
19 */
20
21#include <linux/types.h>
22#include <linux/string.h>
23#include <linux/slab.h>
24#include <linux/module.h>
25#include <linux/capability.h>
26#include <linux/pagemap.h>
27#include <linux/errno.h>
28#include <linux/socket.h>
29#include <linux/in.h>
30#include <linux/net.h>
31#include <linux/mm.h>
32#include <linux/un.h>
33#include <linux/udp.h>
34#include <linux/tcp.h>
35#include <linux/sunrpc/clnt.h>
36#include <linux/sunrpc/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#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
67
68#define XS_TCP_LINGER_TO (15U * HZ)
69static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
70
71/*
72 * We can register our own files under /proc/sys/sunrpc by
73 * calling register_sysctl_table() again. The files in that
74 * directory become the union of all files registered there.
75 *
76 * We simply need to make sure that we don't collide with
77 * someone else's file names!
78 */
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
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,
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
181/*
182 * TCP idle timeout; client drops the transport socket if it is idle
183 * for this long. Note that we also timeout UDP sockets to prevent
184 * holding port numbers when there is no RPC traffic.
185 */
186#define XS_IDLE_DISC_TO (5U * 60 * HZ)
187
188#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
189# undef RPC_DEBUG_DATA
190# define RPCDBG_FACILITY RPCDBG_TRANS
191#endif
192
193#ifdef RPC_DEBUG_DATA
194static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
195{
196 u8 *buf = (u8 *) packet;
197 int j;
198
199 dprintk("RPC: %s\n", msg);
200 for (j = 0; j < count && j < 128; j += 4) {
201 if (!(j & 31)) {
202 if (j)
203 dprintk("\n");
204 dprintk("0x%04x ", j);
205 }
206 dprintk("%02x%02x%02x%02x ",
207 buf[j], buf[j+1], buf[j+2], buf[j+3]);
208 }
209 dprintk("\n");
210}
211#else
212static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
213{
214 /* NOP */
215}
216#endif
217
218static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
219{
220 return (struct rpc_xprt *) sk->sk_user_data;
221}
222
223static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
224{
225 return (struct sockaddr *) &xprt->addr;
226}
227
228static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
229{
230 return (struct sockaddr_un *) &xprt->addr;
231}
232
233static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
234{
235 return (struct sockaddr_in *) &xprt->addr;
236}
237
238static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
239{
240 return (struct sockaddr_in6 *) &xprt->addr;
241}
242
243static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
244{
245 struct sockaddr *sap = xs_addr(xprt);
246 struct sockaddr_in6 *sin6;
247 struct sockaddr_in *sin;
248 struct sockaddr_un *sun;
249 char buf[128];
250
251 switch (sap->sa_family) {
252 case AF_LOCAL:
253 sun = xs_addr_un(xprt);
254 strlcpy(buf, sun->sun_path, sizeof(buf));
255 xprt->address_strings[RPC_DISPLAY_ADDR] =
256 kstrdup(buf, GFP_KERNEL);
257 break;
258 case AF_INET:
259 (void)rpc_ntop(sap, buf, sizeof(buf));
260 xprt->address_strings[RPC_DISPLAY_ADDR] =
261 kstrdup(buf, GFP_KERNEL);
262 sin = xs_addr_in(xprt);
263 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
264 break;
265 case AF_INET6:
266 (void)rpc_ntop(sap, buf, sizeof(buf));
267 xprt->address_strings[RPC_DISPLAY_ADDR] =
268 kstrdup(buf, GFP_KERNEL);
269 sin6 = xs_addr_in6(xprt);
270 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
271 break;
272 default:
273 BUG();
274 }
275
276 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
277}
278
279static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
280{
281 struct sockaddr *sap = xs_addr(xprt);
282 char buf[128];
283
284 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
285 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
286
287 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
288 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
289}
290
291static void xs_format_peer_addresses(struct rpc_xprt *xprt,
292 const char *protocol,
293 const char *netid)
294{
295 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
296 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
297 xs_format_common_peer_addresses(xprt);
298 xs_format_common_peer_ports(xprt);
299}
300
301static void xs_update_peer_port(struct rpc_xprt *xprt)
302{
303 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
304 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
305
306 xs_format_common_peer_ports(xprt);
307}
308
309static void xs_free_peer_addresses(struct rpc_xprt *xprt)
310{
311 unsigned int i;
312
313 for (i = 0; i < RPC_DISPLAY_MAX; i++)
314 switch (i) {
315 case RPC_DISPLAY_PROTO:
316 case RPC_DISPLAY_NETID:
317 continue;
318 default:
319 kfree(xprt->address_strings[i]);
320 }
321}
322
323#define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
324
325static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
326{
327 struct msghdr msg = {
328 .msg_name = addr,
329 .msg_namelen = addrlen,
330 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
331 };
332 struct kvec iov = {
333 .iov_base = vec->iov_base + base,
334 .iov_len = vec->iov_len - base,
335 };
336
337 if (iov.iov_len != 0)
338 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
339 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
340}
341
342static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p)
343{
344 ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
345 int offset, size_t size, int flags);
346 struct page **ppage;
347 unsigned int remainder;
348 int err;
349
350 remainder = xdr->page_len - base;
351 base += xdr->page_base;
352 ppage = xdr->pages + (base >> PAGE_SHIFT);
353 base &= ~PAGE_MASK;
354 do_sendpage = sock->ops->sendpage;
355 if (!zerocopy)
356 do_sendpage = sock_no_sendpage;
357 for(;;) {
358 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
359 int flags = XS_SENDMSG_FLAGS;
360
361 remainder -= len;
362 if (more)
363 flags |= MSG_MORE;
364 if (remainder != 0)
365 flags |= MSG_SENDPAGE_NOTLAST | MSG_MORE;
366 err = do_sendpage(sock, *ppage, base, len, flags);
367 if (remainder == 0 || err != len)
368 break;
369 *sent_p += err;
370 ppage++;
371 base = 0;
372 }
373 if (err > 0) {
374 *sent_p += err;
375 err = 0;
376 }
377 return err;
378}
379
380/**
381 * xs_sendpages - write pages directly to a socket
382 * @sock: socket to send on
383 * @addr: UDP only -- address of destination
384 * @addrlen: UDP only -- length of destination address
385 * @xdr: buffer containing this request
386 * @base: starting position in the buffer
387 * @zerocopy: true if it is safe to use sendpage()
388 * @sent_p: return the total number of bytes successfully queued for sending
389 *
390 */
391static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p)
392{
393 unsigned int remainder = xdr->len - base;
394 int err = 0;
395 int sent = 0;
396
397 if (unlikely(!sock))
398 return -ENOTSOCK;
399
400 if (base != 0) {
401 addr = NULL;
402 addrlen = 0;
403 }
404
405 if (base < xdr->head[0].iov_len || addr != NULL) {
406 unsigned int len = xdr->head[0].iov_len - base;
407 remainder -= len;
408 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
409 if (remainder == 0 || err != len)
410 goto out;
411 *sent_p += err;
412 base = 0;
413 } else
414 base -= xdr->head[0].iov_len;
415
416 if (base < xdr->page_len) {
417 unsigned int len = xdr->page_len - base;
418 remainder -= len;
419 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy, &sent);
420 *sent_p += sent;
421 if (remainder == 0 || sent != len)
422 goto out;
423 base = 0;
424 } else
425 base -= xdr->page_len;
426
427 if (base >= xdr->tail[0].iov_len)
428 return 0;
429 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
430out:
431 if (err > 0) {
432 *sent_p += err;
433 err = 0;
434 }
435 return err;
436}
437
438static void xs_nospace_callback(struct rpc_task *task)
439{
440 struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
441
442 transport->inet->sk_write_pending--;
443}
444
445/**
446 * xs_nospace - place task on wait queue if transmit was incomplete
447 * @task: task to put to sleep
448 *
449 */
450static int xs_nospace(struct rpc_task *task)
451{
452 struct rpc_rqst *req = task->tk_rqstp;
453 struct rpc_xprt *xprt = req->rq_xprt;
454 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
455 struct sock *sk = transport->inet;
456 int ret = -EAGAIN;
457
458 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
459 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
460 req->rq_slen);
461
462 /* Protect against races with write_space */
463 spin_lock_bh(&xprt->transport_lock);
464
465 /* Don't race with disconnect */
466 if (xprt_connected(xprt)) {
467 /* wait for more buffer space */
468 sk->sk_write_pending++;
469 xprt_wait_for_buffer_space(task, xs_nospace_callback);
470 } else
471 ret = -ENOTCONN;
472
473 spin_unlock_bh(&xprt->transport_lock);
474
475 /* Race breaker in case memory is freed before above code is called */
476 if (ret == -EAGAIN) {
477 struct socket_wq *wq;
478
479 rcu_read_lock();
480 wq = rcu_dereference(sk->sk_wq);
481 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
482 rcu_read_unlock();
483
484 sk->sk_write_space(sk);
485 }
486 return ret;
487}
488
489/*
490 * Construct a stream transport record marker in @buf.
491 */
492static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
493{
494 u32 reclen = buf->len - sizeof(rpc_fraghdr);
495 rpc_fraghdr *base = buf->head[0].iov_base;
496 *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
497}
498
499/**
500 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
501 * @task: RPC task that manages the state of an RPC request
502 *
503 * Return values:
504 * 0: The request has been sent
505 * EAGAIN: The socket was blocked, please call again later to
506 * complete the request
507 * ENOTCONN: Caller needs to invoke connect logic then call again
508 * other: Some other error occured, the request was not sent
509 */
510static int xs_local_send_request(struct rpc_task *task)
511{
512 struct rpc_rqst *req = task->tk_rqstp;
513 struct rpc_xprt *xprt = req->rq_xprt;
514 struct sock_xprt *transport =
515 container_of(xprt, struct sock_xprt, xprt);
516 struct xdr_buf *xdr = &req->rq_snd_buf;
517 int status;
518 int sent = 0;
519
520 xs_encode_stream_record_marker(&req->rq_snd_buf);
521
522 xs_pktdump("packet data:",
523 req->rq_svec->iov_base, req->rq_svec->iov_len);
524
525 status = xs_sendpages(transport->sock, NULL, 0, xdr, req->rq_bytes_sent,
526 true, &sent);
527 dprintk("RPC: %s(%u) = %d\n",
528 __func__, xdr->len - req->rq_bytes_sent, status);
529
530 if (status == -EAGAIN && sock_writeable(transport->inet))
531 status = -ENOBUFS;
532
533 if (likely(sent > 0) || status == 0) {
534 req->rq_bytes_sent += sent;
535 req->rq_xmit_bytes_sent += sent;
536 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
537 req->rq_bytes_sent = 0;
538 return 0;
539 }
540 status = -EAGAIN;
541 }
542
543 switch (status) {
544 case -ENOBUFS:
545 break;
546 case -EAGAIN:
547 status = xs_nospace(task);
548 break;
549 default:
550 dprintk("RPC: sendmsg returned unrecognized error %d\n",
551 -status);
552 case -EPIPE:
553 xs_close(xprt);
554 status = -ENOTCONN;
555 }
556
557 return status;
558}
559
560/**
561 * xs_udp_send_request - write an RPC request to a UDP socket
562 * @task: address of RPC task that manages the state of an RPC request
563 *
564 * Return values:
565 * 0: The request has been sent
566 * EAGAIN: The socket was blocked, please call again later to
567 * complete the request
568 * ENOTCONN: Caller needs to invoke connect logic then call again
569 * other: Some other error occurred, the request was not sent
570 */
571static int xs_udp_send_request(struct rpc_task *task)
572{
573 struct rpc_rqst *req = task->tk_rqstp;
574 struct rpc_xprt *xprt = req->rq_xprt;
575 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
576 struct xdr_buf *xdr = &req->rq_snd_buf;
577 int sent = 0;
578 int status;
579
580 xs_pktdump("packet data:",
581 req->rq_svec->iov_base,
582 req->rq_svec->iov_len);
583
584 if (!xprt_bound(xprt))
585 return -ENOTCONN;
586 status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
587 xdr, req->rq_bytes_sent, true, &sent);
588
589 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
590 xdr->len - req->rq_bytes_sent, status);
591
592 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
593 if (status == -EPERM)
594 goto process_status;
595
596 if (status == -EAGAIN && sock_writeable(transport->inet))
597 status = -ENOBUFS;
598
599 if (sent > 0 || status == 0) {
600 req->rq_xmit_bytes_sent += sent;
601 if (sent >= req->rq_slen)
602 return 0;
603 /* Still some bytes left; set up for a retry later. */
604 status = -EAGAIN;
605 }
606
607process_status:
608 switch (status) {
609 case -ENOTSOCK:
610 status = -ENOTCONN;
611 /* Should we call xs_close() here? */
612 break;
613 case -EAGAIN:
614 status = xs_nospace(task);
615 break;
616 case -ENETUNREACH:
617 case -ENOBUFS:
618 case -EPIPE:
619 case -ECONNREFUSED:
620 case -EPERM:
621 /* When the server has died, an ICMP port unreachable message
622 * prompts ECONNREFUSED. */
623 break;
624 default:
625 dprintk("RPC: sendmsg returned unrecognized error %d\n",
626 -status);
627 }
628
629 return status;
630}
631
632/**
633 * xs_tcp_send_request - write an RPC request to a TCP socket
634 * @task: address of RPC task that manages the state of an RPC request
635 *
636 * Return values:
637 * 0: The request has been sent
638 * EAGAIN: The socket was blocked, please call again later to
639 * complete the request
640 * ENOTCONN: Caller needs to invoke connect logic then call again
641 * other: Some other error occurred, the request was not sent
642 *
643 * XXX: In the case of soft timeouts, should we eventually give up
644 * if sendmsg is not able to make progress?
645 */
646static int xs_tcp_send_request(struct rpc_task *task)
647{
648 struct rpc_rqst *req = task->tk_rqstp;
649 struct rpc_xprt *xprt = req->rq_xprt;
650 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
651 struct xdr_buf *xdr = &req->rq_snd_buf;
652 bool zerocopy = true;
653 bool vm_wait = false;
654 int status;
655 int sent;
656
657 xs_encode_stream_record_marker(&req->rq_snd_buf);
658
659 xs_pktdump("packet data:",
660 req->rq_svec->iov_base,
661 req->rq_svec->iov_len);
662 /* Don't use zero copy if this is a resend. If the RPC call
663 * completes while the socket holds a reference to the pages,
664 * then we may end up resending corrupted data.
665 */
666 if (task->tk_flags & RPC_TASK_SENT)
667 zerocopy = false;
668
669 /* Continue transmitting the packet/record. We must be careful
670 * to cope with writespace callbacks arriving _after_ we have
671 * called sendmsg(). */
672 while (1) {
673 sent = 0;
674 status = xs_sendpages(transport->sock, NULL, 0, xdr,
675 req->rq_bytes_sent, zerocopy, &sent);
676
677 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
678 xdr->len - req->rq_bytes_sent, status);
679
680 /* If we've sent the entire packet, immediately
681 * reset the count of bytes sent. */
682 req->rq_bytes_sent += sent;
683 req->rq_xmit_bytes_sent += sent;
684 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
685 req->rq_bytes_sent = 0;
686 return 0;
687 }
688
689 WARN_ON_ONCE(sent == 0 && status == 0);
690
691 if (status == -EAGAIN ) {
692 /*
693 * Return EAGAIN if we're sure we're hitting the
694 * socket send buffer limits.
695 */
696 if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
697 break;
698 /*
699 * Did we hit a memory allocation failure?
700 */
701 if (sent == 0) {
702 status = -ENOBUFS;
703 if (vm_wait)
704 break;
705 /* Retry, knowing now that we're below the
706 * socket send buffer limit
707 */
708 vm_wait = true;
709 }
710 continue;
711 }
712 if (status < 0)
713 break;
714 vm_wait = false;
715 }
716
717 switch (status) {
718 case -ENOTSOCK:
719 status = -ENOTCONN;
720 /* Should we call xs_close() here? */
721 break;
722 case -EAGAIN:
723 status = xs_nospace(task);
724 break;
725 case -ECONNRESET:
726 case -ECONNREFUSED:
727 case -ENOTCONN:
728 case -EADDRINUSE:
729 case -ENOBUFS:
730 case -EPIPE:
731 break;
732 default:
733 dprintk("RPC: sendmsg returned unrecognized error %d\n",
734 -status);
735 }
736
737 return status;
738}
739
740/**
741 * xs_tcp_release_xprt - clean up after a tcp transmission
742 * @xprt: transport
743 * @task: rpc task
744 *
745 * This cleans up if an error causes us to abort the transmission of a request.
746 * In this case, the socket may need to be reset in order to avoid confusing
747 * the server.
748 */
749static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
750{
751 struct rpc_rqst *req;
752
753 if (task != xprt->snd_task)
754 return;
755 if (task == NULL)
756 goto out_release;
757 req = task->tk_rqstp;
758 if (req == NULL)
759 goto out_release;
760 if (req->rq_bytes_sent == 0)
761 goto out_release;
762 if (req->rq_bytes_sent == req->rq_snd_buf.len)
763 goto out_release;
764 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
765out_release:
766 xprt_release_xprt(xprt, task);
767}
768
769static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
770{
771 transport->old_data_ready = sk->sk_data_ready;
772 transport->old_state_change = sk->sk_state_change;
773 transport->old_write_space = sk->sk_write_space;
774 transport->old_error_report = sk->sk_error_report;
775}
776
777static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
778{
779 sk->sk_data_ready = transport->old_data_ready;
780 sk->sk_state_change = transport->old_state_change;
781 sk->sk_write_space = transport->old_write_space;
782 sk->sk_error_report = transport->old_error_report;
783}
784
785static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
786{
787 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
788
789 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
790}
791
792static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
793{
794 smp_mb__before_atomic();
795 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
796 clear_bit(XPRT_CLOSING, &xprt->state);
797 xs_sock_reset_state_flags(xprt);
798 smp_mb__after_atomic();
799}
800
801static void xs_sock_mark_closed(struct rpc_xprt *xprt)
802{
803 xs_sock_reset_connection_flags(xprt);
804 /* Mark transport as closed and wake up all pending tasks */
805 xprt_disconnect_done(xprt);
806}
807
808/**
809 * xs_error_report - callback to handle TCP socket state errors
810 * @sk: socket
811 *
812 * Note: we don't call sock_error() since there may be a rpc_task
813 * using the socket, and so we don't want to clear sk->sk_err.
814 */
815static void xs_error_report(struct sock *sk)
816{
817 struct rpc_xprt *xprt;
818 int err;
819
820 read_lock_bh(&sk->sk_callback_lock);
821 if (!(xprt = xprt_from_sock(sk)))
822 goto out;
823
824 err = -sk->sk_err;
825 if (err == 0)
826 goto out;
827 /* Is this a reset event? */
828 if (sk->sk_state == TCP_CLOSE)
829 xs_sock_mark_closed(xprt);
830 dprintk("RPC: xs_error_report client %p, error=%d...\n",
831 xprt, -err);
832 trace_rpc_socket_error(xprt, sk->sk_socket, err);
833 xprt_wake_pending_tasks(xprt, err);
834 out:
835 read_unlock_bh(&sk->sk_callback_lock);
836}
837
838static void xs_reset_transport(struct sock_xprt *transport)
839{
840 struct socket *sock = transport->sock;
841 struct sock *sk = transport->inet;
842 struct rpc_xprt *xprt = &transport->xprt;
843
844 if (sk == NULL)
845 return;
846
847 if (atomic_read(&transport->xprt.swapper))
848 sk_clear_memalloc(sk);
849
850 kernel_sock_shutdown(sock, SHUT_RDWR);
851
852 mutex_lock(&transport->recv_mutex);
853 write_lock_bh(&sk->sk_callback_lock);
854 transport->inet = NULL;
855 transport->sock = NULL;
856
857 sk->sk_user_data = NULL;
858
859 xs_restore_old_callbacks(transport, sk);
860 xprt_clear_connected(xprt);
861 write_unlock_bh(&sk->sk_callback_lock);
862 xs_sock_reset_connection_flags(xprt);
863 mutex_unlock(&transport->recv_mutex);
864
865 trace_rpc_socket_close(xprt, sock);
866 sock_release(sock);
867}
868
869/**
870 * xs_close - close a socket
871 * @xprt: transport
872 *
873 * This is used when all requests are complete; ie, no DRC state remains
874 * on the server we want to save.
875 *
876 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
877 * xs_reset_transport() zeroing the socket from underneath a writer.
878 */
879static void xs_close(struct rpc_xprt *xprt)
880{
881 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
882
883 dprintk("RPC: xs_close xprt %p\n", xprt);
884
885 xs_reset_transport(transport);
886 xprt->reestablish_timeout = 0;
887
888 xprt_disconnect_done(xprt);
889}
890
891static void xs_inject_disconnect(struct rpc_xprt *xprt)
892{
893 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
894 xprt);
895 xprt_disconnect_done(xprt);
896}
897
898static void xs_xprt_free(struct rpc_xprt *xprt)
899{
900 xs_free_peer_addresses(xprt);
901 xprt_free(xprt);
902}
903
904/**
905 * xs_destroy - prepare to shutdown a transport
906 * @xprt: doomed transport
907 *
908 */
909static void xs_destroy(struct rpc_xprt *xprt)
910{
911 struct sock_xprt *transport = container_of(xprt,
912 struct sock_xprt, xprt);
913 dprintk("RPC: xs_destroy xprt %p\n", xprt);
914
915 cancel_delayed_work_sync(&transport->connect_worker);
916 xs_close(xprt);
917 cancel_work_sync(&transport->recv_worker);
918 xs_xprt_free(xprt);
919 module_put(THIS_MODULE);
920}
921
922static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
923{
924 struct xdr_skb_reader desc = {
925 .skb = skb,
926 .offset = sizeof(rpc_fraghdr),
927 .count = skb->len - sizeof(rpc_fraghdr),
928 };
929
930 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
931 return -1;
932 if (desc.count)
933 return -1;
934 return 0;
935}
936
937/**
938 * xs_local_data_read_skb
939 * @xprt: transport
940 * @sk: socket
941 * @skb: skbuff
942 *
943 * Currently this assumes we can read the whole reply in a single gulp.
944 */
945static void xs_local_data_read_skb(struct rpc_xprt *xprt,
946 struct sock *sk,
947 struct sk_buff *skb)
948{
949 struct rpc_task *task;
950 struct rpc_rqst *rovr;
951 int repsize, copied;
952 u32 _xid;
953 __be32 *xp;
954
955 repsize = skb->len - sizeof(rpc_fraghdr);
956 if (repsize < 4) {
957 dprintk("RPC: impossible RPC reply size %d\n", repsize);
958 return;
959 }
960
961 /* Copy the XID from the skb... */
962 xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
963 if (xp == NULL)
964 return;
965
966 /* Look up and lock the request corresponding to the given XID */
967 spin_lock_bh(&xprt->transport_lock);
968 rovr = xprt_lookup_rqst(xprt, *xp);
969 if (!rovr)
970 goto out_unlock;
971 task = rovr->rq_task;
972
973 copied = rovr->rq_private_buf.buflen;
974 if (copied > repsize)
975 copied = repsize;
976
977 if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
978 dprintk("RPC: sk_buff copy failed\n");
979 goto out_unlock;
980 }
981
982 xprt_complete_rqst(task, copied);
983
984 out_unlock:
985 spin_unlock_bh(&xprt->transport_lock);
986}
987
988static void xs_local_data_receive(struct sock_xprt *transport)
989{
990 struct sk_buff *skb;
991 struct sock *sk;
992 int err;
993
994 mutex_lock(&transport->recv_mutex);
995 sk = transport->inet;
996 if (sk == NULL)
997 goto out;
998 for (;;) {
999 skb = skb_recv_datagram(sk, 0, 1, &err);
1000 if (skb != NULL) {
1001 xs_local_data_read_skb(&transport->xprt, sk, skb);
1002 skb_free_datagram(sk, skb);
1003 continue;
1004 }
1005 if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1006 break;
1007 }
1008out:
1009 mutex_unlock(&transport->recv_mutex);
1010}
1011
1012static void xs_local_data_receive_workfn(struct work_struct *work)
1013{
1014 struct sock_xprt *transport =
1015 container_of(work, struct sock_xprt, recv_worker);
1016 xs_local_data_receive(transport);
1017}
1018
1019/**
1020 * xs_udp_data_read_skb - receive callback for UDP sockets
1021 * @xprt: transport
1022 * @sk: socket
1023 * @skb: skbuff
1024 *
1025 */
1026static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1027 struct sock *sk,
1028 struct sk_buff *skb)
1029{
1030 struct rpc_task *task;
1031 struct rpc_rqst *rovr;
1032 int repsize, copied;
1033 u32 _xid;
1034 __be32 *xp;
1035
1036 repsize = skb->len;
1037 if (repsize < 4) {
1038 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
1039 return;
1040 }
1041
1042 /* Copy the XID from the skb... */
1043 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1044 if (xp == NULL)
1045 return;
1046
1047 /* Look up and lock the request corresponding to the given XID */
1048 spin_lock_bh(&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(sk, UDP_MIB_INERRORS);
1060 goto out_unlock;
1061 }
1062
1063 __UDPX_INC_STATS(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_bh(&xprt->transport_lock);
1070}
1071
1072static void xs_udp_data_receive(struct sock_xprt *transport)
1073{
1074 struct sk_buff *skb;
1075 struct sock *sk;
1076 int err;
1077
1078 mutex_lock(&transport->recv_mutex);
1079 sk = transport->inet;
1080 if (sk == NULL)
1081 goto out;
1082 for (;;) {
1083 skb = skb_recv_udp(sk, 0, 1, &err);
1084 if (skb != NULL) {
1085 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1086 consume_skb(skb);
1087 continue;
1088 }
1089 if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1090 break;
1091 }
1092out:
1093 mutex_unlock(&transport->recv_mutex);
1094}
1095
1096static void xs_udp_data_receive_workfn(struct work_struct *work)
1097{
1098 struct sock_xprt *transport =
1099 container_of(work, struct sock_xprt, recv_worker);
1100 xs_udp_data_receive(transport);
1101}
1102
1103/**
1104 * xs_data_ready - "data ready" callback for UDP sockets
1105 * @sk: socket with data to read
1106 *
1107 */
1108static void xs_data_ready(struct sock *sk)
1109{
1110 struct rpc_xprt *xprt;
1111
1112 read_lock_bh(&sk->sk_callback_lock);
1113 dprintk("RPC: xs_data_ready...\n");
1114 xprt = xprt_from_sock(sk);
1115 if (xprt != NULL) {
1116 struct sock_xprt *transport = container_of(xprt,
1117 struct sock_xprt, xprt);
1118 transport->old_data_ready(sk);
1119 /* Any data means we had a useful conversation, so
1120 * then we don't need to delay the next reconnect
1121 */
1122 if (xprt->reestablish_timeout)
1123 xprt->reestablish_timeout = 0;
1124 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1125 queue_work(xprtiod_workqueue, &transport->recv_worker);
1126 }
1127 read_unlock_bh(&sk->sk_callback_lock);
1128}
1129
1130/*
1131 * Helper function to force a TCP close if the server is sending
1132 * junk and/or it has put us in CLOSE_WAIT
1133 */
1134static void xs_tcp_force_close(struct rpc_xprt *xprt)
1135{
1136 xprt_force_disconnect(xprt);
1137}
1138
1139static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1140{
1141 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1142 size_t len, used;
1143 char *p;
1144
1145 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1146 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1147 used = xdr_skb_read_bits(desc, p, len);
1148 transport->tcp_offset += used;
1149 if (used != len)
1150 return;
1151
1152 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1153 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1154 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1155 else
1156 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1157 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1158
1159 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1160 transport->tcp_offset = 0;
1161
1162 /* Sanity check of the record length */
1163 if (unlikely(transport->tcp_reclen < 8)) {
1164 dprintk("RPC: invalid TCP record fragment length\n");
1165 xs_tcp_force_close(xprt);
1166 return;
1167 }
1168 dprintk("RPC: reading TCP record fragment of length %d\n",
1169 transport->tcp_reclen);
1170}
1171
1172static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1173{
1174 if (transport->tcp_offset == transport->tcp_reclen) {
1175 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1176 transport->tcp_offset = 0;
1177 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1178 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1179 transport->tcp_flags |= TCP_RCV_COPY_XID;
1180 transport->tcp_copied = 0;
1181 }
1182 }
1183}
1184
1185static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1186{
1187 size_t len, used;
1188 char *p;
1189
1190 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1191 dprintk("RPC: reading XID (%Zu bytes)\n", len);
1192 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1193 used = xdr_skb_read_bits(desc, p, len);
1194 transport->tcp_offset += used;
1195 if (used != len)
1196 return;
1197 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1198 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1199 transport->tcp_copied = 4;
1200 dprintk("RPC: reading %s XID %08x\n",
1201 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1202 : "request with",
1203 ntohl(transport->tcp_xid));
1204 xs_tcp_check_fraghdr(transport);
1205}
1206
1207static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1208 struct xdr_skb_reader *desc)
1209{
1210 size_t len, used;
1211 u32 offset;
1212 char *p;
1213
1214 /*
1215 * We want transport->tcp_offset to be 8 at the end of this routine
1216 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1217 * When this function is called for the first time,
1218 * transport->tcp_offset is 4 (after having already read the xid).
1219 */
1220 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1221 len = sizeof(transport->tcp_calldir) - offset;
1222 dprintk("RPC: reading CALL/REPLY flag (%Zu bytes)\n", len);
1223 p = ((char *) &transport->tcp_calldir) + offset;
1224 used = xdr_skb_read_bits(desc, p, len);
1225 transport->tcp_offset += used;
1226 if (used != len)
1227 return;
1228 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1229 /*
1230 * We don't yet have the XDR buffer, so we will write the calldir
1231 * out after we get the buffer from the 'struct rpc_rqst'
1232 */
1233 switch (ntohl(transport->tcp_calldir)) {
1234 case RPC_REPLY:
1235 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1236 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1237 transport->tcp_flags |= TCP_RPC_REPLY;
1238 break;
1239 case RPC_CALL:
1240 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1241 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1242 transport->tcp_flags &= ~TCP_RPC_REPLY;
1243 break;
1244 default:
1245 dprintk("RPC: invalid request message type\n");
1246 xs_tcp_force_close(&transport->xprt);
1247 }
1248 xs_tcp_check_fraghdr(transport);
1249}
1250
1251static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1252 struct xdr_skb_reader *desc,
1253 struct rpc_rqst *req)
1254{
1255 struct sock_xprt *transport =
1256 container_of(xprt, struct sock_xprt, xprt);
1257 struct xdr_buf *rcvbuf;
1258 size_t len;
1259 ssize_t r;
1260
1261 rcvbuf = &req->rq_private_buf;
1262
1263 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1264 /*
1265 * Save the RPC direction in the XDR buffer
1266 */
1267 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1268 &transport->tcp_calldir,
1269 sizeof(transport->tcp_calldir));
1270 transport->tcp_copied += sizeof(transport->tcp_calldir);
1271 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1272 }
1273
1274 len = desc->count;
1275 if (len > transport->tcp_reclen - transport->tcp_offset) {
1276 struct xdr_skb_reader my_desc;
1277
1278 len = transport->tcp_reclen - transport->tcp_offset;
1279 memcpy(&my_desc, desc, sizeof(my_desc));
1280 my_desc.count = len;
1281 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1282 &my_desc, xdr_skb_read_bits);
1283 desc->count -= r;
1284 desc->offset += r;
1285 } else
1286 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1287 desc, xdr_skb_read_bits);
1288
1289 if (r > 0) {
1290 transport->tcp_copied += r;
1291 transport->tcp_offset += r;
1292 }
1293 if (r != len) {
1294 /* Error when copying to the receive buffer,
1295 * usually because we weren't able to allocate
1296 * additional buffer pages. All we can do now
1297 * is turn off TCP_RCV_COPY_DATA, so the request
1298 * will not receive any additional updates,
1299 * and time out.
1300 * Any remaining data from this record will
1301 * be discarded.
1302 */
1303 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1304 dprintk("RPC: XID %08x truncated request\n",
1305 ntohl(transport->tcp_xid));
1306 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1307 "tcp_offset = %u, tcp_reclen = %u\n",
1308 xprt, transport->tcp_copied,
1309 transport->tcp_offset, transport->tcp_reclen);
1310 return;
1311 }
1312
1313 dprintk("RPC: XID %08x read %Zd bytes\n",
1314 ntohl(transport->tcp_xid), r);
1315 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1316 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1317 transport->tcp_offset, transport->tcp_reclen);
1318
1319 if (transport->tcp_copied == req->rq_private_buf.buflen)
1320 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1321 else if (transport->tcp_offset == transport->tcp_reclen) {
1322 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1323 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1324 }
1325}
1326
1327/*
1328 * Finds the request corresponding to the RPC xid and invokes the common
1329 * tcp read code to read the data.
1330 */
1331static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1332 struct xdr_skb_reader *desc)
1333{
1334 struct sock_xprt *transport =
1335 container_of(xprt, struct sock_xprt, xprt);
1336 struct rpc_rqst *req;
1337
1338 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1339
1340 /* Find and lock the request corresponding to this xid */
1341 spin_lock_bh(&xprt->transport_lock);
1342 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1343 if (!req) {
1344 dprintk("RPC: XID %08x request not found!\n",
1345 ntohl(transport->tcp_xid));
1346 spin_unlock_bh(&xprt->transport_lock);
1347 return -1;
1348 }
1349
1350 xs_tcp_read_common(xprt, desc, req);
1351
1352 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1353 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1354
1355 spin_unlock_bh(&xprt->transport_lock);
1356 return 0;
1357}
1358
1359#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1360/*
1361 * Obtains an rpc_rqst previously allocated and invokes the common
1362 * tcp read code to read the data. The result is placed in the callback
1363 * queue.
1364 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1365 * connection and return -1.
1366 */
1367static int xs_tcp_read_callback(struct rpc_xprt *xprt,
1368 struct xdr_skb_reader *desc)
1369{
1370 struct sock_xprt *transport =
1371 container_of(xprt, struct sock_xprt, xprt);
1372 struct rpc_rqst *req;
1373
1374 /* Look up and lock the request corresponding to the given XID */
1375 spin_lock_bh(&xprt->transport_lock);
1376 req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
1377 if (req == NULL) {
1378 spin_unlock_bh(&xprt->transport_lock);
1379 printk(KERN_WARNING "Callback slot table overflowed\n");
1380 xprt_force_disconnect(xprt);
1381 return -1;
1382 }
1383
1384 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1385 xs_tcp_read_common(xprt, desc, req);
1386
1387 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1388 xprt_complete_bc_request(req, transport->tcp_copied);
1389 spin_unlock_bh(&xprt->transport_lock);
1390
1391 return 0;
1392}
1393
1394static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1395 struct xdr_skb_reader *desc)
1396{
1397 struct sock_xprt *transport =
1398 container_of(xprt, struct sock_xprt, xprt);
1399
1400 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1401 xs_tcp_read_reply(xprt, desc) :
1402 xs_tcp_read_callback(xprt, desc);
1403}
1404
1405static int xs_tcp_bc_up(struct svc_serv *serv, struct net *net)
1406{
1407 int ret;
1408
1409 ret = svc_create_xprt(serv, "tcp-bc", net, PF_INET, 0,
1410 SVC_SOCK_ANONYMOUS);
1411 if (ret < 0)
1412 return ret;
1413 return 0;
1414}
1415
1416static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1417{
1418 return PAGE_SIZE;
1419}
1420#else
1421static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1422 struct xdr_skb_reader *desc)
1423{
1424 return xs_tcp_read_reply(xprt, desc);
1425}
1426#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1427
1428/*
1429 * Read data off the transport. This can be either an RPC_CALL or an
1430 * RPC_REPLY. Relay the processing to helper functions.
1431 */
1432static void xs_tcp_read_data(struct rpc_xprt *xprt,
1433 struct xdr_skb_reader *desc)
1434{
1435 struct sock_xprt *transport =
1436 container_of(xprt, struct sock_xprt, xprt);
1437
1438 if (_xs_tcp_read_data(xprt, desc) == 0)
1439 xs_tcp_check_fraghdr(transport);
1440 else {
1441 /*
1442 * The transport_lock protects the request handling.
1443 * There's no need to hold it to update the tcp_flags.
1444 */
1445 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1446 }
1447}
1448
1449static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1450{
1451 size_t len;
1452
1453 len = transport->tcp_reclen - transport->tcp_offset;
1454 if (len > desc->count)
1455 len = desc->count;
1456 desc->count -= len;
1457 desc->offset += len;
1458 transport->tcp_offset += len;
1459 dprintk("RPC: discarded %Zu bytes\n", len);
1460 xs_tcp_check_fraghdr(transport);
1461}
1462
1463static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1464{
1465 struct rpc_xprt *xprt = rd_desc->arg.data;
1466 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1467 struct xdr_skb_reader desc = {
1468 .skb = skb,
1469 .offset = offset,
1470 .count = len,
1471 };
1472
1473 dprintk("RPC: xs_tcp_data_recv started\n");
1474 do {
1475 trace_xs_tcp_data_recv(transport);
1476 /* Read in a new fragment marker if necessary */
1477 /* Can we ever really expect to get completely empty fragments? */
1478 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1479 xs_tcp_read_fraghdr(xprt, &desc);
1480 continue;
1481 }
1482 /* Read in the xid if necessary */
1483 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1484 xs_tcp_read_xid(transport, &desc);
1485 continue;
1486 }
1487 /* Read in the call/reply flag */
1488 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1489 xs_tcp_read_calldir(transport, &desc);
1490 continue;
1491 }
1492 /* Read in the request data */
1493 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1494 xs_tcp_read_data(xprt, &desc);
1495 continue;
1496 }
1497 /* Skip over any trailing bytes on short reads */
1498 xs_tcp_read_discard(transport, &desc);
1499 } while (desc.count);
1500 trace_xs_tcp_data_recv(transport);
1501 dprintk("RPC: xs_tcp_data_recv done\n");
1502 return len - desc.count;
1503}
1504
1505static void xs_tcp_data_receive(struct sock_xprt *transport)
1506{
1507 struct rpc_xprt *xprt = &transport->xprt;
1508 struct sock *sk;
1509 read_descriptor_t rd_desc = {
1510 .count = 2*1024*1024,
1511 .arg.data = xprt,
1512 };
1513 unsigned long total = 0;
1514 int read = 0;
1515
1516 mutex_lock(&transport->recv_mutex);
1517 sk = transport->inet;
1518 if (sk == NULL)
1519 goto out;
1520
1521 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1522 for (;;) {
1523 lock_sock(sk);
1524 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1525 if (read <= 0) {
1526 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1527 release_sock(sk);
1528 if (!test_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1529 break;
1530 } else {
1531 release_sock(sk);
1532 total += read;
1533 }
1534 rd_desc.count = 65536;
1535 }
1536out:
1537 mutex_unlock(&transport->recv_mutex);
1538 trace_xs_tcp_data_ready(xprt, read, total);
1539}
1540
1541static void xs_tcp_data_receive_workfn(struct work_struct *work)
1542{
1543 struct sock_xprt *transport =
1544 container_of(work, struct sock_xprt, recv_worker);
1545 xs_tcp_data_receive(transport);
1546}
1547
1548/**
1549 * xs_tcp_state_change - callback to handle TCP socket state changes
1550 * @sk: socket whose state has changed
1551 *
1552 */
1553static void xs_tcp_state_change(struct sock *sk)
1554{
1555 struct rpc_xprt *xprt;
1556 struct sock_xprt *transport;
1557
1558 read_lock_bh(&sk->sk_callback_lock);
1559 if (!(xprt = xprt_from_sock(sk)))
1560 goto out;
1561 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1562 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1563 sk->sk_state, xprt_connected(xprt),
1564 sock_flag(sk, SOCK_DEAD),
1565 sock_flag(sk, SOCK_ZAPPED),
1566 sk->sk_shutdown);
1567
1568 transport = container_of(xprt, struct sock_xprt, xprt);
1569 trace_rpc_socket_state_change(xprt, sk->sk_socket);
1570 switch (sk->sk_state) {
1571 case TCP_ESTABLISHED:
1572 spin_lock(&xprt->transport_lock);
1573 if (!xprt_test_and_set_connected(xprt)) {
1574
1575 /* Reset TCP record info */
1576 transport->tcp_offset = 0;
1577 transport->tcp_reclen = 0;
1578 transport->tcp_copied = 0;
1579 transport->tcp_flags =
1580 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1581 xprt->connect_cookie++;
1582 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1583 xprt_clear_connecting(xprt);
1584
1585 xprt_wake_pending_tasks(xprt, -EAGAIN);
1586 }
1587 spin_unlock(&xprt->transport_lock);
1588 break;
1589 case TCP_FIN_WAIT1:
1590 /* The client initiated a shutdown of the socket */
1591 xprt->connect_cookie++;
1592 xprt->reestablish_timeout = 0;
1593 set_bit(XPRT_CLOSING, &xprt->state);
1594 smp_mb__before_atomic();
1595 clear_bit(XPRT_CONNECTED, &xprt->state);
1596 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1597 smp_mb__after_atomic();
1598 break;
1599 case TCP_CLOSE_WAIT:
1600 /* The server initiated a shutdown of the socket */
1601 xprt->connect_cookie++;
1602 clear_bit(XPRT_CONNECTED, &xprt->state);
1603 xs_tcp_force_close(xprt);
1604 case TCP_CLOSING:
1605 /*
1606 * If the server closed down the connection, make sure that
1607 * we back off before reconnecting
1608 */
1609 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1610 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1611 break;
1612 case TCP_LAST_ACK:
1613 set_bit(XPRT_CLOSING, &xprt->state);
1614 smp_mb__before_atomic();
1615 clear_bit(XPRT_CONNECTED, &xprt->state);
1616 smp_mb__after_atomic();
1617 break;
1618 case TCP_CLOSE:
1619 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1620 &transport->sock_state))
1621 xprt_clear_connecting(xprt);
1622 xs_sock_mark_closed(xprt);
1623 }
1624 out:
1625 read_unlock_bh(&sk->sk_callback_lock);
1626}
1627
1628static void xs_write_space(struct sock *sk)
1629{
1630 struct socket_wq *wq;
1631 struct rpc_xprt *xprt;
1632
1633 if (!sk->sk_socket)
1634 return;
1635 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1636
1637 if (unlikely(!(xprt = xprt_from_sock(sk))))
1638 return;
1639 rcu_read_lock();
1640 wq = rcu_dereference(sk->sk_wq);
1641 if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1642 goto out;
1643
1644 xprt_write_space(xprt);
1645out:
1646 rcu_read_unlock();
1647}
1648
1649/**
1650 * xs_udp_write_space - callback invoked when socket buffer space
1651 * becomes available
1652 * @sk: socket whose state has changed
1653 *
1654 * Called when more output buffer space is available for this socket.
1655 * We try not to wake our writers until they can make "significant"
1656 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1657 * with a bunch of small requests.
1658 */
1659static void xs_udp_write_space(struct sock *sk)
1660{
1661 read_lock_bh(&sk->sk_callback_lock);
1662
1663 /* from net/core/sock.c:sock_def_write_space */
1664 if (sock_writeable(sk))
1665 xs_write_space(sk);
1666
1667 read_unlock_bh(&sk->sk_callback_lock);
1668}
1669
1670/**
1671 * xs_tcp_write_space - callback invoked when socket buffer space
1672 * becomes available
1673 * @sk: socket whose state has changed
1674 *
1675 * Called when more output buffer space is available for this socket.
1676 * We try not to wake our writers until they can make "significant"
1677 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1678 * with a bunch of small requests.
1679 */
1680static void xs_tcp_write_space(struct sock *sk)
1681{
1682 read_lock_bh(&sk->sk_callback_lock);
1683
1684 /* from net/core/stream.c:sk_stream_write_space */
1685 if (sk_stream_is_writeable(sk))
1686 xs_write_space(sk);
1687
1688 read_unlock_bh(&sk->sk_callback_lock);
1689}
1690
1691static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1692{
1693 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1694 struct sock *sk = transport->inet;
1695
1696 if (transport->rcvsize) {
1697 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1698 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1699 }
1700 if (transport->sndsize) {
1701 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1702 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1703 sk->sk_write_space(sk);
1704 }
1705}
1706
1707/**
1708 * xs_udp_set_buffer_size - set send and receive limits
1709 * @xprt: generic transport
1710 * @sndsize: requested size of send buffer, in bytes
1711 * @rcvsize: requested size of receive buffer, in bytes
1712 *
1713 * Set socket send and receive buffer size limits.
1714 */
1715static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1716{
1717 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1718
1719 transport->sndsize = 0;
1720 if (sndsize)
1721 transport->sndsize = sndsize + 1024;
1722 transport->rcvsize = 0;
1723 if (rcvsize)
1724 transport->rcvsize = rcvsize + 1024;
1725
1726 xs_udp_do_set_buffer_size(xprt);
1727}
1728
1729/**
1730 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1731 * @task: task that timed out
1732 *
1733 * Adjust the congestion window after a retransmit timeout has occurred.
1734 */
1735static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1736{
1737 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1738}
1739
1740static unsigned short xs_get_random_port(void)
1741{
1742 unsigned short range = xprt_max_resvport - xprt_min_resvport + 1;
1743 unsigned short rand = (unsigned short) prandom_u32() % range;
1744 return rand + xprt_min_resvport;
1745}
1746
1747/**
1748 * xs_set_reuseaddr_port - set the socket's port and address reuse options
1749 * @sock: socket
1750 *
1751 * Note that this function has to be called on all sockets that share the
1752 * same port, and it must be called before binding.
1753 */
1754static void xs_sock_set_reuseport(struct socket *sock)
1755{
1756 int opt = 1;
1757
1758 kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1759 (char *)&opt, sizeof(opt));
1760}
1761
1762static unsigned short xs_sock_getport(struct socket *sock)
1763{
1764 struct sockaddr_storage buf;
1765 int buflen;
1766 unsigned short port = 0;
1767
1768 if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0)
1769 goto out;
1770 switch (buf.ss_family) {
1771 case AF_INET6:
1772 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1773 break;
1774 case AF_INET:
1775 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1776 }
1777out:
1778 return port;
1779}
1780
1781/**
1782 * xs_set_port - reset the port number in the remote endpoint address
1783 * @xprt: generic transport
1784 * @port: new port number
1785 *
1786 */
1787static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1788{
1789 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1790
1791 rpc_set_port(xs_addr(xprt), port);
1792 xs_update_peer_port(xprt);
1793}
1794
1795static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1796{
1797 if (transport->srcport == 0)
1798 transport->srcport = xs_sock_getport(sock);
1799}
1800
1801static unsigned short xs_get_srcport(struct sock_xprt *transport)
1802{
1803 unsigned short port = transport->srcport;
1804
1805 if (port == 0 && transport->xprt.resvport)
1806 port = xs_get_random_port();
1807 return port;
1808}
1809
1810static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1811{
1812 if (transport->srcport != 0)
1813 transport->srcport = 0;
1814 if (!transport->xprt.resvport)
1815 return 0;
1816 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1817 return xprt_max_resvport;
1818 return --port;
1819}
1820static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1821{
1822 struct sockaddr_storage myaddr;
1823 int err, nloop = 0;
1824 unsigned short port = xs_get_srcport(transport);
1825 unsigned short last;
1826
1827 /*
1828 * If we are asking for any ephemeral port (i.e. port == 0 &&
1829 * transport->xprt.resvport == 0), don't bind. Let the local
1830 * port selection happen implicitly when the socket is used
1831 * (for example at connect time).
1832 *
1833 * This ensures that we can continue to establish TCP
1834 * connections even when all local ephemeral ports are already
1835 * a part of some TCP connection. This makes no difference
1836 * for UDP sockets, but also doens't harm them.
1837 *
1838 * If we're asking for any reserved port (i.e. port == 0 &&
1839 * transport->xprt.resvport == 1) xs_get_srcport above will
1840 * ensure that port is non-zero and we will bind as needed.
1841 */
1842 if (port == 0)
1843 return 0;
1844
1845 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1846 do {
1847 rpc_set_port((struct sockaddr *)&myaddr, port);
1848 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1849 transport->xprt.addrlen);
1850 if (err == 0) {
1851 transport->srcport = port;
1852 break;
1853 }
1854 last = port;
1855 port = xs_next_srcport(transport, port);
1856 if (port > last)
1857 nloop++;
1858 } while (err == -EADDRINUSE && nloop != 2);
1859
1860 if (myaddr.ss_family == AF_INET)
1861 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1862 &((struct sockaddr_in *)&myaddr)->sin_addr,
1863 port, err ? "failed" : "ok", err);
1864 else
1865 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1866 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1867 port, err ? "failed" : "ok", err);
1868 return err;
1869}
1870
1871/*
1872 * We don't support autobind on AF_LOCAL sockets
1873 */
1874static void xs_local_rpcbind(struct rpc_task *task)
1875{
1876 xprt_set_bound(task->tk_xprt);
1877}
1878
1879static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1880{
1881}
1882
1883#ifdef CONFIG_DEBUG_LOCK_ALLOC
1884static struct lock_class_key xs_key[2];
1885static struct lock_class_key xs_slock_key[2];
1886
1887static inline void xs_reclassify_socketu(struct socket *sock)
1888{
1889 struct sock *sk = sock->sk;
1890
1891 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1892 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1893}
1894
1895static inline void xs_reclassify_socket4(struct socket *sock)
1896{
1897 struct sock *sk = sock->sk;
1898
1899 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1900 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1901}
1902
1903static inline void xs_reclassify_socket6(struct socket *sock)
1904{
1905 struct sock *sk = sock->sk;
1906
1907 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1908 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1909}
1910
1911static inline void xs_reclassify_socket(int family, struct socket *sock)
1912{
1913 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1914 return;
1915
1916 switch (family) {
1917 case AF_LOCAL:
1918 xs_reclassify_socketu(sock);
1919 break;
1920 case AF_INET:
1921 xs_reclassify_socket4(sock);
1922 break;
1923 case AF_INET6:
1924 xs_reclassify_socket6(sock);
1925 break;
1926 }
1927}
1928#else
1929static inline void xs_reclassify_socket(int family, struct socket *sock)
1930{
1931}
1932#endif
1933
1934static void xs_dummy_setup_socket(struct work_struct *work)
1935{
1936}
1937
1938static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1939 struct sock_xprt *transport, int family, int type,
1940 int protocol, bool reuseport)
1941{
1942 struct socket *sock;
1943 int err;
1944
1945 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1946 if (err < 0) {
1947 dprintk("RPC: can't create %d transport socket (%d).\n",
1948 protocol, -err);
1949 goto out;
1950 }
1951 xs_reclassify_socket(family, sock);
1952
1953 if (reuseport)
1954 xs_sock_set_reuseport(sock);
1955
1956 err = xs_bind(transport, sock);
1957 if (err) {
1958 sock_release(sock);
1959 goto out;
1960 }
1961
1962 return sock;
1963out:
1964 return ERR_PTR(err);
1965}
1966
1967static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1968 struct socket *sock)
1969{
1970 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1971 xprt);
1972
1973 if (!transport->inet) {
1974 struct sock *sk = sock->sk;
1975
1976 write_lock_bh(&sk->sk_callback_lock);
1977
1978 xs_save_old_callbacks(transport, sk);
1979
1980 sk->sk_user_data = xprt;
1981 sk->sk_data_ready = xs_data_ready;
1982 sk->sk_write_space = xs_udp_write_space;
1983 sock_set_flag(sk, SOCK_FASYNC);
1984 sk->sk_error_report = xs_error_report;
1985 sk->sk_allocation = GFP_NOIO;
1986
1987 xprt_clear_connected(xprt);
1988
1989 /* Reset to new socket */
1990 transport->sock = sock;
1991 transport->inet = sk;
1992
1993 write_unlock_bh(&sk->sk_callback_lock);
1994 }
1995
1996 /* Tell the socket layer to start connecting... */
1997 xprt->stat.connect_count++;
1998 xprt->stat.connect_start = jiffies;
1999 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
2000}
2001
2002/**
2003 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
2004 * @transport: socket transport to connect
2005 */
2006static int xs_local_setup_socket(struct sock_xprt *transport)
2007{
2008 struct rpc_xprt *xprt = &transport->xprt;
2009 struct socket *sock;
2010 int status = -EIO;
2011
2012 status = __sock_create(xprt->xprt_net, AF_LOCAL,
2013 SOCK_STREAM, 0, &sock, 1);
2014 if (status < 0) {
2015 dprintk("RPC: can't create AF_LOCAL "
2016 "transport socket (%d).\n", -status);
2017 goto out;
2018 }
2019 xs_reclassify_socket(AF_LOCAL, sock);
2020
2021 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
2022 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2023
2024 status = xs_local_finish_connecting(xprt, sock);
2025 trace_rpc_socket_connect(xprt, sock, status);
2026 switch (status) {
2027 case 0:
2028 dprintk("RPC: xprt %p connected to %s\n",
2029 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2030 xprt_set_connected(xprt);
2031 case -ENOBUFS:
2032 break;
2033 case -ENOENT:
2034 dprintk("RPC: xprt %p: socket %s does not exist\n",
2035 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2036 break;
2037 case -ECONNREFUSED:
2038 dprintk("RPC: xprt %p: connection refused for %s\n",
2039 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2040 break;
2041 default:
2042 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2043 __func__, -status,
2044 xprt->address_strings[RPC_DISPLAY_ADDR]);
2045 }
2046
2047out:
2048 xprt_clear_connecting(xprt);
2049 xprt_wake_pending_tasks(xprt, status);
2050 return status;
2051}
2052
2053static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2054{
2055 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2056 int ret;
2057
2058 if (RPC_IS_ASYNC(task)) {
2059 /*
2060 * We want the AF_LOCAL connect to be resolved in the
2061 * filesystem namespace of the process making the rpc
2062 * call. Thus we connect synchronously.
2063 *
2064 * If we want to support asynchronous AF_LOCAL calls,
2065 * we'll need to figure out how to pass a namespace to
2066 * connect.
2067 */
2068 rpc_exit(task, -ENOTCONN);
2069 return;
2070 }
2071 ret = xs_local_setup_socket(transport);
2072 if (ret && !RPC_IS_SOFTCONN(task))
2073 msleep_interruptible(15000);
2074}
2075
2076#if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2077/*
2078 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2079 * know that we have exclusive access to the socket), to guard against
2080 * races with xs_reset_transport.
2081 */
2082static void xs_set_memalloc(struct rpc_xprt *xprt)
2083{
2084 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2085 xprt);
2086
2087 /*
2088 * If there's no sock, then we have nothing to set. The
2089 * reconnecting process will get it for us.
2090 */
2091 if (!transport->inet)
2092 return;
2093 if (atomic_read(&xprt->swapper))
2094 sk_set_memalloc(transport->inet);
2095}
2096
2097/**
2098 * xs_enable_swap - Tag this transport as being used for swap.
2099 * @xprt: transport to tag
2100 *
2101 * Take a reference to this transport on behalf of the rpc_clnt, and
2102 * optionally mark it for swapping if it wasn't already.
2103 */
2104static int
2105xs_enable_swap(struct rpc_xprt *xprt)
2106{
2107 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2108
2109 if (atomic_inc_return(&xprt->swapper) != 1)
2110 return 0;
2111 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2112 return -ERESTARTSYS;
2113 if (xs->inet)
2114 sk_set_memalloc(xs->inet);
2115 xprt_release_xprt(xprt, NULL);
2116 return 0;
2117}
2118
2119/**
2120 * xs_disable_swap - Untag this transport as being used for swap.
2121 * @xprt: transport to tag
2122 *
2123 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2124 * swapper refcount goes to 0, untag the socket as a memalloc socket.
2125 */
2126static void
2127xs_disable_swap(struct rpc_xprt *xprt)
2128{
2129 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2130
2131 if (!atomic_dec_and_test(&xprt->swapper))
2132 return;
2133 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2134 return;
2135 if (xs->inet)
2136 sk_clear_memalloc(xs->inet);
2137 xprt_release_xprt(xprt, NULL);
2138}
2139#else
2140static void xs_set_memalloc(struct rpc_xprt *xprt)
2141{
2142}
2143
2144static int
2145xs_enable_swap(struct rpc_xprt *xprt)
2146{
2147 return -EINVAL;
2148}
2149
2150static void
2151xs_disable_swap(struct rpc_xprt *xprt)
2152{
2153}
2154#endif
2155
2156static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2157{
2158 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2159
2160 if (!transport->inet) {
2161 struct sock *sk = sock->sk;
2162
2163 write_lock_bh(&sk->sk_callback_lock);
2164
2165 xs_save_old_callbacks(transport, sk);
2166
2167 sk->sk_user_data = xprt;
2168 sk->sk_data_ready = xs_data_ready;
2169 sk->sk_write_space = xs_udp_write_space;
2170 sock_set_flag(sk, SOCK_FASYNC);
2171 sk->sk_allocation = GFP_NOIO;
2172
2173 xprt_set_connected(xprt);
2174
2175 /* Reset to new socket */
2176 transport->sock = sock;
2177 transport->inet = sk;
2178
2179 xs_set_memalloc(xprt);
2180
2181 write_unlock_bh(&sk->sk_callback_lock);
2182 }
2183 xs_udp_do_set_buffer_size(xprt);
2184
2185 xprt->stat.connect_start = jiffies;
2186}
2187
2188static void xs_udp_setup_socket(struct work_struct *work)
2189{
2190 struct sock_xprt *transport =
2191 container_of(work, struct sock_xprt, connect_worker.work);
2192 struct rpc_xprt *xprt = &transport->xprt;
2193 struct socket *sock = transport->sock;
2194 int status = -EIO;
2195
2196 sock = xs_create_sock(xprt, transport,
2197 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2198 IPPROTO_UDP, false);
2199 if (IS_ERR(sock))
2200 goto out;
2201
2202 dprintk("RPC: worker connecting xprt %p via %s to "
2203 "%s (port %s)\n", xprt,
2204 xprt->address_strings[RPC_DISPLAY_PROTO],
2205 xprt->address_strings[RPC_DISPLAY_ADDR],
2206 xprt->address_strings[RPC_DISPLAY_PORT]);
2207
2208 xs_udp_finish_connecting(xprt, sock);
2209 trace_rpc_socket_connect(xprt, sock, 0);
2210 status = 0;
2211out:
2212 xprt_unlock_connect(xprt, transport);
2213 xprt_clear_connecting(xprt);
2214 xprt_wake_pending_tasks(xprt, status);
2215}
2216
2217/**
2218 * xs_tcp_shutdown - gracefully shut down a TCP socket
2219 * @xprt: transport
2220 *
2221 * Initiates a graceful shutdown of the TCP socket by calling the
2222 * equivalent of shutdown(SHUT_RDWR);
2223 */
2224static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2225{
2226 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2227 struct socket *sock = transport->sock;
2228
2229 if (sock == NULL)
2230 return;
2231 if (xprt_connected(xprt)) {
2232 kernel_sock_shutdown(sock, SHUT_RDWR);
2233 trace_rpc_socket_shutdown(xprt, sock);
2234 } else
2235 xs_reset_transport(transport);
2236}
2237
2238static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2239{
2240 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2241 int ret = -ENOTCONN;
2242
2243 if (!transport->inet) {
2244 struct sock *sk = sock->sk;
2245 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2246 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2247 unsigned int opt_on = 1;
2248 unsigned int timeo;
2249 unsigned int addr_pref = IPV6_PREFER_SRC_PUBLIC;
2250
2251 /* TCP Keepalive options */
2252 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2253 (char *)&opt_on, sizeof(opt_on));
2254 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2255 (char *)&keepidle, sizeof(keepidle));
2256 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2257 (char *)&keepidle, sizeof(keepidle));
2258 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2259 (char *)&keepcnt, sizeof(keepcnt));
2260
2261 /* Avoid temporary address, they are bad for long-lived
2262 * connections such as NFS mounts.
2263 * RFC4941, section 3.6 suggests that:
2264 * Individual applications, which have specific
2265 * knowledge about the normal duration of connections,
2266 * MAY override this as appropriate.
2267 */
2268 kernel_setsockopt(sock, SOL_IPV6, IPV6_ADDR_PREFERENCES,
2269 (char *)&addr_pref, sizeof(addr_pref));
2270
2271 /* TCP user timeout (see RFC5482) */
2272 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2273 (xprt->timeout->to_retries + 1);
2274 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2275 (char *)&timeo, sizeof(timeo));
2276
2277 write_lock_bh(&sk->sk_callback_lock);
2278
2279 xs_save_old_callbacks(transport, sk);
2280
2281 sk->sk_user_data = xprt;
2282 sk->sk_data_ready = xs_data_ready;
2283 sk->sk_state_change = xs_tcp_state_change;
2284 sk->sk_write_space = xs_tcp_write_space;
2285 sock_set_flag(sk, SOCK_FASYNC);
2286 sk->sk_error_report = xs_error_report;
2287 sk->sk_allocation = GFP_NOIO;
2288
2289 /* socket options */
2290 sock_reset_flag(sk, SOCK_LINGER);
2291 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2292
2293 xprt_clear_connected(xprt);
2294
2295 /* Reset to new socket */
2296 transport->sock = sock;
2297 transport->inet = sk;
2298
2299 write_unlock_bh(&sk->sk_callback_lock);
2300 }
2301
2302 if (!xprt_bound(xprt))
2303 goto out;
2304
2305 xs_set_memalloc(xprt);
2306
2307 /* Tell the socket layer to start connecting... */
2308 xprt->stat.connect_count++;
2309 xprt->stat.connect_start = jiffies;
2310 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2311 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2312 switch (ret) {
2313 case 0:
2314 xs_set_srcport(transport, sock);
2315 case -EINPROGRESS:
2316 /* SYN_SENT! */
2317 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2318 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2319 break;
2320 case -EADDRNOTAVAIL:
2321 /* Source port number is unavailable. Try a new one! */
2322 transport->srcport = 0;
2323 }
2324out:
2325 return ret;
2326}
2327
2328/**
2329 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2330 *
2331 * Invoked by a work queue tasklet.
2332 */
2333static void xs_tcp_setup_socket(struct work_struct *work)
2334{
2335 struct sock_xprt *transport =
2336 container_of(work, struct sock_xprt, connect_worker.work);
2337 struct socket *sock = transport->sock;
2338 struct rpc_xprt *xprt = &transport->xprt;
2339 int status = -EIO;
2340
2341 if (!sock) {
2342 sock = xs_create_sock(xprt, transport,
2343 xs_addr(xprt)->sa_family, SOCK_STREAM,
2344 IPPROTO_TCP, true);
2345 if (IS_ERR(sock)) {
2346 status = PTR_ERR(sock);
2347 goto out;
2348 }
2349 }
2350
2351 dprintk("RPC: worker connecting xprt %p via %s to "
2352 "%s (port %s)\n", xprt,
2353 xprt->address_strings[RPC_DISPLAY_PROTO],
2354 xprt->address_strings[RPC_DISPLAY_ADDR],
2355 xprt->address_strings[RPC_DISPLAY_PORT]);
2356
2357 status = xs_tcp_finish_connecting(xprt, sock);
2358 trace_rpc_socket_connect(xprt, sock, status);
2359 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2360 xprt, -status, xprt_connected(xprt),
2361 sock->sk->sk_state);
2362 switch (status) {
2363 default:
2364 printk("%s: connect returned unhandled error %d\n",
2365 __func__, status);
2366 case -EADDRNOTAVAIL:
2367 /* We're probably in TIME_WAIT. Get rid of existing socket,
2368 * and retry
2369 */
2370 xs_tcp_force_close(xprt);
2371 break;
2372 case 0:
2373 case -EINPROGRESS:
2374 case -EALREADY:
2375 xprt_unlock_connect(xprt, transport);
2376 return;
2377 case -EINVAL:
2378 /* Happens, for instance, if the user specified a link
2379 * local IPv6 address without a scope-id.
2380 */
2381 case -ECONNREFUSED:
2382 case -ECONNRESET:
2383 case -ENETUNREACH:
2384 case -EADDRINUSE:
2385 case -ENOBUFS:
2386 /* retry with existing socket, after a delay */
2387 xs_tcp_force_close(xprt);
2388 goto out;
2389 }
2390 status = -EAGAIN;
2391out:
2392 xprt_unlock_connect(xprt, transport);
2393 xprt_clear_connecting(xprt);
2394 xprt_wake_pending_tasks(xprt, status);
2395}
2396
2397static unsigned long xs_reconnect_delay(const struct rpc_xprt *xprt)
2398{
2399 unsigned long start, now = jiffies;
2400
2401 start = xprt->stat.connect_start + xprt->reestablish_timeout;
2402 if (time_after(start, now))
2403 return start - now;
2404 return 0;
2405}
2406
2407static void xs_reconnect_backoff(struct rpc_xprt *xprt)
2408{
2409 xprt->reestablish_timeout <<= 1;
2410 if (xprt->reestablish_timeout > xprt->max_reconnect_timeout)
2411 xprt->reestablish_timeout = xprt->max_reconnect_timeout;
2412 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2413 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2414}
2415
2416/**
2417 * xs_connect - connect a socket to a remote endpoint
2418 * @xprt: pointer to transport structure
2419 * @task: address of RPC task that manages state of connect request
2420 *
2421 * TCP: If the remote end dropped the connection, delay reconnecting.
2422 *
2423 * UDP socket connects are synchronous, but we use a work queue anyway
2424 * to guarantee that even unprivileged user processes can set up a
2425 * socket on a privileged port.
2426 *
2427 * If a UDP socket connect fails, the delay behavior here prevents
2428 * retry floods (hard mounts).
2429 */
2430static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2431{
2432 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2433 unsigned long delay = 0;
2434
2435 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2436
2437 if (transport->sock != NULL) {
2438 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2439 "seconds\n",
2440 xprt, xprt->reestablish_timeout / HZ);
2441
2442 /* Start by resetting any existing state */
2443 xs_reset_transport(transport);
2444
2445 delay = xs_reconnect_delay(xprt);
2446 xs_reconnect_backoff(xprt);
2447
2448 } else
2449 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2450
2451 queue_delayed_work(xprtiod_workqueue,
2452 &transport->connect_worker,
2453 delay);
2454}
2455
2456/**
2457 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2458 * @xprt: rpc_xprt struct containing statistics
2459 * @seq: output file
2460 *
2461 */
2462static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2463{
2464 long idle_time = 0;
2465
2466 if (xprt_connected(xprt))
2467 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2468
2469 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2470 "%llu %llu %lu %llu %llu\n",
2471 xprt->stat.bind_count,
2472 xprt->stat.connect_count,
2473 xprt->stat.connect_time,
2474 idle_time,
2475 xprt->stat.sends,
2476 xprt->stat.recvs,
2477 xprt->stat.bad_xids,
2478 xprt->stat.req_u,
2479 xprt->stat.bklog_u,
2480 xprt->stat.max_slots,
2481 xprt->stat.sending_u,
2482 xprt->stat.pending_u);
2483}
2484
2485/**
2486 * xs_udp_print_stats - display UDP socket-specifc stats
2487 * @xprt: rpc_xprt struct containing statistics
2488 * @seq: output file
2489 *
2490 */
2491static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2492{
2493 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2494
2495 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2496 "%lu %llu %llu\n",
2497 transport->srcport,
2498 xprt->stat.bind_count,
2499 xprt->stat.sends,
2500 xprt->stat.recvs,
2501 xprt->stat.bad_xids,
2502 xprt->stat.req_u,
2503 xprt->stat.bklog_u,
2504 xprt->stat.max_slots,
2505 xprt->stat.sending_u,
2506 xprt->stat.pending_u);
2507}
2508
2509/**
2510 * xs_tcp_print_stats - display TCP socket-specifc stats
2511 * @xprt: rpc_xprt struct containing statistics
2512 * @seq: output file
2513 *
2514 */
2515static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2516{
2517 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2518 long idle_time = 0;
2519
2520 if (xprt_connected(xprt))
2521 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2522
2523 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2524 "%llu %llu %lu %llu %llu\n",
2525 transport->srcport,
2526 xprt->stat.bind_count,
2527 xprt->stat.connect_count,
2528 xprt->stat.connect_time,
2529 idle_time,
2530 xprt->stat.sends,
2531 xprt->stat.recvs,
2532 xprt->stat.bad_xids,
2533 xprt->stat.req_u,
2534 xprt->stat.bklog_u,
2535 xprt->stat.max_slots,
2536 xprt->stat.sending_u,
2537 xprt->stat.pending_u);
2538}
2539
2540/*
2541 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2542 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2543 * to use the server side send routines.
2544 */
2545static int bc_malloc(struct rpc_task *task)
2546{
2547 struct rpc_rqst *rqst = task->tk_rqstp;
2548 size_t size = rqst->rq_callsize;
2549 struct page *page;
2550 struct rpc_buffer *buf;
2551
2552 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2553 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2554 size);
2555 return -EINVAL;
2556 }
2557
2558 page = alloc_page(GFP_KERNEL);
2559 if (!page)
2560 return -ENOMEM;
2561
2562 buf = page_address(page);
2563 buf->len = PAGE_SIZE;
2564
2565 rqst->rq_buffer = buf->data;
2566 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2567 return 0;
2568}
2569
2570/*
2571 * Free the space allocated in the bc_alloc routine
2572 */
2573static void bc_free(struct rpc_task *task)
2574{
2575 void *buffer = task->tk_rqstp->rq_buffer;
2576 struct rpc_buffer *buf;
2577
2578 buf = container_of(buffer, struct rpc_buffer, data);
2579 free_page((unsigned long)buf);
2580}
2581
2582/*
2583 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2584 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2585 */
2586static int bc_sendto(struct rpc_rqst *req)
2587{
2588 int len;
2589 struct xdr_buf *xbufp = &req->rq_snd_buf;
2590 struct rpc_xprt *xprt = req->rq_xprt;
2591 struct sock_xprt *transport =
2592 container_of(xprt, struct sock_xprt, xprt);
2593 struct socket *sock = transport->sock;
2594 unsigned long headoff;
2595 unsigned long tailoff;
2596
2597 xs_encode_stream_record_marker(xbufp);
2598
2599 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2600 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2601 len = svc_send_common(sock, xbufp,
2602 virt_to_page(xbufp->head[0].iov_base), headoff,
2603 xbufp->tail[0].iov_base, tailoff);
2604
2605 if (len != xbufp->len) {
2606 printk(KERN_NOTICE "Error sending entire callback!\n");
2607 len = -EAGAIN;
2608 }
2609
2610 return len;
2611}
2612
2613/*
2614 * The send routine. Borrows from svc_send
2615 */
2616static int bc_send_request(struct rpc_task *task)
2617{
2618 struct rpc_rqst *req = task->tk_rqstp;
2619 struct svc_xprt *xprt;
2620 int len;
2621
2622 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2623 /*
2624 * Get the server socket associated with this callback xprt
2625 */
2626 xprt = req->rq_xprt->bc_xprt;
2627
2628 /*
2629 * Grab the mutex to serialize data as the connection is shared
2630 * with the fore channel
2631 */
2632 if (!mutex_trylock(&xprt->xpt_mutex)) {
2633 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2634 if (!mutex_trylock(&xprt->xpt_mutex))
2635 return -EAGAIN;
2636 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2637 }
2638 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2639 len = -ENOTCONN;
2640 else
2641 len = bc_sendto(req);
2642 mutex_unlock(&xprt->xpt_mutex);
2643
2644 if (len > 0)
2645 len = 0;
2646
2647 return len;
2648}
2649
2650/*
2651 * The close routine. Since this is client initiated, we do nothing
2652 */
2653
2654static void bc_close(struct rpc_xprt *xprt)
2655{
2656}
2657
2658/*
2659 * The xprt destroy routine. Again, because this connection is client
2660 * initiated, we do nothing
2661 */
2662
2663static void bc_destroy(struct rpc_xprt *xprt)
2664{
2665 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2666
2667 xs_xprt_free(xprt);
2668 module_put(THIS_MODULE);
2669}
2670
2671static struct rpc_xprt_ops xs_local_ops = {
2672 .reserve_xprt = xprt_reserve_xprt,
2673 .release_xprt = xs_tcp_release_xprt,
2674 .alloc_slot = xprt_alloc_slot,
2675 .rpcbind = xs_local_rpcbind,
2676 .set_port = xs_local_set_port,
2677 .connect = xs_local_connect,
2678 .buf_alloc = rpc_malloc,
2679 .buf_free = rpc_free,
2680 .send_request = xs_local_send_request,
2681 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2682 .close = xs_close,
2683 .destroy = xs_destroy,
2684 .print_stats = xs_local_print_stats,
2685 .enable_swap = xs_enable_swap,
2686 .disable_swap = xs_disable_swap,
2687};
2688
2689static struct rpc_xprt_ops xs_udp_ops = {
2690 .set_buffer_size = xs_udp_set_buffer_size,
2691 .reserve_xprt = xprt_reserve_xprt_cong,
2692 .release_xprt = xprt_release_xprt_cong,
2693 .alloc_slot = xprt_alloc_slot,
2694 .rpcbind = rpcb_getport_async,
2695 .set_port = xs_set_port,
2696 .connect = xs_connect,
2697 .buf_alloc = rpc_malloc,
2698 .buf_free = rpc_free,
2699 .send_request = xs_udp_send_request,
2700 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
2701 .timer = xs_udp_timer,
2702 .release_request = xprt_release_rqst_cong,
2703 .close = xs_close,
2704 .destroy = xs_destroy,
2705 .print_stats = xs_udp_print_stats,
2706 .enable_swap = xs_enable_swap,
2707 .disable_swap = xs_disable_swap,
2708 .inject_disconnect = xs_inject_disconnect,
2709};
2710
2711static struct rpc_xprt_ops xs_tcp_ops = {
2712 .reserve_xprt = xprt_reserve_xprt,
2713 .release_xprt = xs_tcp_release_xprt,
2714 .alloc_slot = xprt_lock_and_alloc_slot,
2715 .rpcbind = rpcb_getport_async,
2716 .set_port = xs_set_port,
2717 .connect = xs_connect,
2718 .buf_alloc = rpc_malloc,
2719 .buf_free = rpc_free,
2720 .send_request = xs_tcp_send_request,
2721 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2722 .close = xs_tcp_shutdown,
2723 .destroy = xs_destroy,
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_up = xs_tcp_bc_up,
2731 .bc_maxpayload = xs_tcp_bc_maxpayload,
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 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 .buf_alloc = bc_malloc,
2746 .buf_free = bc_free,
2747 .send_request = bc_send_request,
2748 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2749 .close = bc_close,
2750 .destroy = bc_destroy,
2751 .print_stats = xs_tcp_print_stats,
2752 .enable_swap = xs_enable_swap,
2753 .disable_swap = xs_disable_swap,
2754 .inject_disconnect = xs_inject_disconnect,
2755};
2756
2757static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2758{
2759 static const struct sockaddr_in sin = {
2760 .sin_family = AF_INET,
2761 .sin_addr.s_addr = htonl(INADDR_ANY),
2762 };
2763 static const struct sockaddr_in6 sin6 = {
2764 .sin6_family = AF_INET6,
2765 .sin6_addr = IN6ADDR_ANY_INIT,
2766 };
2767
2768 switch (family) {
2769 case AF_LOCAL:
2770 break;
2771 case AF_INET:
2772 memcpy(sap, &sin, sizeof(sin));
2773 break;
2774 case AF_INET6:
2775 memcpy(sap, &sin6, sizeof(sin6));
2776 break;
2777 default:
2778 dprintk("RPC: %s: Bad address family\n", __func__);
2779 return -EAFNOSUPPORT;
2780 }
2781 return 0;
2782}
2783
2784static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2785 unsigned int slot_table_size,
2786 unsigned int max_slot_table_size)
2787{
2788 struct rpc_xprt *xprt;
2789 struct sock_xprt *new;
2790
2791 if (args->addrlen > sizeof(xprt->addr)) {
2792 dprintk("RPC: xs_setup_xprt: address too large\n");
2793 return ERR_PTR(-EBADF);
2794 }
2795
2796 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2797 max_slot_table_size);
2798 if (xprt == NULL) {
2799 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2800 "rpc_xprt\n");
2801 return ERR_PTR(-ENOMEM);
2802 }
2803
2804 new = container_of(xprt, struct sock_xprt, xprt);
2805 mutex_init(&new->recv_mutex);
2806 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2807 xprt->addrlen = args->addrlen;
2808 if (args->srcaddr)
2809 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2810 else {
2811 int err;
2812 err = xs_init_anyaddr(args->dstaddr->sa_family,
2813 (struct sockaddr *)&new->srcaddr);
2814 if (err != 0) {
2815 xprt_free(xprt);
2816 return ERR_PTR(err);
2817 }
2818 }
2819
2820 return xprt;
2821}
2822
2823static const struct rpc_timeout xs_local_default_timeout = {
2824 .to_initval = 10 * HZ,
2825 .to_maxval = 10 * HZ,
2826 .to_retries = 2,
2827};
2828
2829/**
2830 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2831 * @args: rpc transport creation arguments
2832 *
2833 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2834 */
2835static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2836{
2837 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2838 struct sock_xprt *transport;
2839 struct rpc_xprt *xprt;
2840 struct rpc_xprt *ret;
2841
2842 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2843 xprt_max_tcp_slot_table_entries);
2844 if (IS_ERR(xprt))
2845 return xprt;
2846 transport = container_of(xprt, struct sock_xprt, xprt);
2847
2848 xprt->prot = 0;
2849 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2850 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2851
2852 xprt->bind_timeout = XS_BIND_TO;
2853 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2854 xprt->idle_timeout = XS_IDLE_DISC_TO;
2855
2856 xprt->ops = &xs_local_ops;
2857 xprt->timeout = &xs_local_default_timeout;
2858
2859 INIT_WORK(&transport->recv_worker, xs_local_data_receive_workfn);
2860 INIT_DELAYED_WORK(&transport->connect_worker,
2861 xs_dummy_setup_socket);
2862
2863 switch (sun->sun_family) {
2864 case AF_LOCAL:
2865 if (sun->sun_path[0] != '/') {
2866 dprintk("RPC: bad AF_LOCAL address: %s\n",
2867 sun->sun_path);
2868 ret = ERR_PTR(-EINVAL);
2869 goto out_err;
2870 }
2871 xprt_set_bound(xprt);
2872 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2873 ret = ERR_PTR(xs_local_setup_socket(transport));
2874 if (ret)
2875 goto out_err;
2876 break;
2877 default:
2878 ret = ERR_PTR(-EAFNOSUPPORT);
2879 goto out_err;
2880 }
2881
2882 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2883 xprt->address_strings[RPC_DISPLAY_ADDR]);
2884
2885 if (try_module_get(THIS_MODULE))
2886 return xprt;
2887 ret = ERR_PTR(-EINVAL);
2888out_err:
2889 xs_xprt_free(xprt);
2890 return ret;
2891}
2892
2893static const struct rpc_timeout xs_udp_default_timeout = {
2894 .to_initval = 5 * HZ,
2895 .to_maxval = 30 * HZ,
2896 .to_increment = 5 * HZ,
2897 .to_retries = 5,
2898};
2899
2900/**
2901 * xs_setup_udp - Set up transport to use a UDP socket
2902 * @args: rpc transport creation arguments
2903 *
2904 */
2905static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2906{
2907 struct sockaddr *addr = args->dstaddr;
2908 struct rpc_xprt *xprt;
2909 struct sock_xprt *transport;
2910 struct rpc_xprt *ret;
2911
2912 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2913 xprt_udp_slot_table_entries);
2914 if (IS_ERR(xprt))
2915 return xprt;
2916 transport = container_of(xprt, struct sock_xprt, xprt);
2917
2918 xprt->prot = IPPROTO_UDP;
2919 xprt->tsh_size = 0;
2920 /* XXX: header size can vary due to auth type, IPv6, etc. */
2921 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2922
2923 xprt->bind_timeout = XS_BIND_TO;
2924 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2925 xprt->idle_timeout = XS_IDLE_DISC_TO;
2926
2927 xprt->ops = &xs_udp_ops;
2928
2929 xprt->timeout = &xs_udp_default_timeout;
2930
2931 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
2932 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2933
2934 switch (addr->sa_family) {
2935 case AF_INET:
2936 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2937 xprt_set_bound(xprt);
2938
2939 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2940 break;
2941 case AF_INET6:
2942 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2943 xprt_set_bound(xprt);
2944
2945 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2946 break;
2947 default:
2948 ret = ERR_PTR(-EAFNOSUPPORT);
2949 goto out_err;
2950 }
2951
2952 if (xprt_bound(xprt))
2953 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2954 xprt->address_strings[RPC_DISPLAY_ADDR],
2955 xprt->address_strings[RPC_DISPLAY_PORT],
2956 xprt->address_strings[RPC_DISPLAY_PROTO]);
2957 else
2958 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2959 xprt->address_strings[RPC_DISPLAY_ADDR],
2960 xprt->address_strings[RPC_DISPLAY_PROTO]);
2961
2962 if (try_module_get(THIS_MODULE))
2963 return xprt;
2964 ret = ERR_PTR(-EINVAL);
2965out_err:
2966 xs_xprt_free(xprt);
2967 return ret;
2968}
2969
2970static const struct rpc_timeout xs_tcp_default_timeout = {
2971 .to_initval = 60 * HZ,
2972 .to_maxval = 60 * HZ,
2973 .to_retries = 2,
2974};
2975
2976/**
2977 * xs_setup_tcp - Set up transport to use a TCP socket
2978 * @args: rpc transport creation arguments
2979 *
2980 */
2981static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2982{
2983 struct sockaddr *addr = args->dstaddr;
2984 struct rpc_xprt *xprt;
2985 struct sock_xprt *transport;
2986 struct rpc_xprt *ret;
2987 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2988
2989 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2990 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2991
2992 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2993 max_slot_table_size);
2994 if (IS_ERR(xprt))
2995 return xprt;
2996 transport = container_of(xprt, struct sock_xprt, xprt);
2997
2998 xprt->prot = IPPROTO_TCP;
2999 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3000 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3001
3002 xprt->bind_timeout = XS_BIND_TO;
3003 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3004 xprt->idle_timeout = XS_IDLE_DISC_TO;
3005
3006 xprt->ops = &xs_tcp_ops;
3007 xprt->timeout = &xs_tcp_default_timeout;
3008
3009 xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3010
3011 INIT_WORK(&transport->recv_worker, xs_tcp_data_receive_workfn);
3012 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
3013
3014 switch (addr->sa_family) {
3015 case AF_INET:
3016 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3017 xprt_set_bound(xprt);
3018
3019 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3020 break;
3021 case AF_INET6:
3022 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3023 xprt_set_bound(xprt);
3024
3025 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3026 break;
3027 default:
3028 ret = ERR_PTR(-EAFNOSUPPORT);
3029 goto out_err;
3030 }
3031
3032 if (xprt_bound(xprt))
3033 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3034 xprt->address_strings[RPC_DISPLAY_ADDR],
3035 xprt->address_strings[RPC_DISPLAY_PORT],
3036 xprt->address_strings[RPC_DISPLAY_PROTO]);
3037 else
3038 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
3039 xprt->address_strings[RPC_DISPLAY_ADDR],
3040 xprt->address_strings[RPC_DISPLAY_PROTO]);
3041
3042 if (try_module_get(THIS_MODULE))
3043 return xprt;
3044 ret = ERR_PTR(-EINVAL);
3045out_err:
3046 xs_xprt_free(xprt);
3047 return ret;
3048}
3049
3050/**
3051 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3052 * @args: rpc transport creation arguments
3053 *
3054 */
3055static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3056{
3057 struct sockaddr *addr = args->dstaddr;
3058 struct rpc_xprt *xprt;
3059 struct sock_xprt *transport;
3060 struct svc_sock *bc_sock;
3061 struct rpc_xprt *ret;
3062
3063 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3064 xprt_tcp_slot_table_entries);
3065 if (IS_ERR(xprt))
3066 return xprt;
3067 transport = container_of(xprt, struct sock_xprt, xprt);
3068
3069 xprt->prot = IPPROTO_TCP;
3070 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3071 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3072 xprt->timeout = &xs_tcp_default_timeout;
3073
3074 /* backchannel */
3075 xprt_set_bound(xprt);
3076 xprt->bind_timeout = 0;
3077 xprt->reestablish_timeout = 0;
3078 xprt->idle_timeout = 0;
3079
3080 xprt->ops = &bc_tcp_ops;
3081
3082 switch (addr->sa_family) {
3083 case AF_INET:
3084 xs_format_peer_addresses(xprt, "tcp",
3085 RPCBIND_NETID_TCP);
3086 break;
3087 case AF_INET6:
3088 xs_format_peer_addresses(xprt, "tcp",
3089 RPCBIND_NETID_TCP6);
3090 break;
3091 default:
3092 ret = ERR_PTR(-EAFNOSUPPORT);
3093 goto out_err;
3094 }
3095
3096 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3097 xprt->address_strings[RPC_DISPLAY_ADDR],
3098 xprt->address_strings[RPC_DISPLAY_PORT],
3099 xprt->address_strings[RPC_DISPLAY_PROTO]);
3100
3101 /*
3102 * Once we've associated a backchannel xprt with a connection,
3103 * we want to keep it around as long as the connection lasts,
3104 * in case we need to start using it for a backchannel again;
3105 * this reference won't be dropped until bc_xprt is destroyed.
3106 */
3107 xprt_get(xprt);
3108 args->bc_xprt->xpt_bc_xprt = xprt;
3109 xprt->bc_xprt = args->bc_xprt;
3110 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3111 transport->sock = bc_sock->sk_sock;
3112 transport->inet = bc_sock->sk_sk;
3113
3114 /*
3115 * Since we don't want connections for the backchannel, we set
3116 * the xprt status to connected
3117 */
3118 xprt_set_connected(xprt);
3119
3120 if (try_module_get(THIS_MODULE))
3121 return xprt;
3122
3123 args->bc_xprt->xpt_bc_xprt = NULL;
3124 args->bc_xprt->xpt_bc_xps = NULL;
3125 xprt_put(xprt);
3126 ret = ERR_PTR(-EINVAL);
3127out_err:
3128 xs_xprt_free(xprt);
3129 return ret;
3130}
3131
3132static struct xprt_class xs_local_transport = {
3133 .list = LIST_HEAD_INIT(xs_local_transport.list),
3134 .name = "named UNIX socket",
3135 .owner = THIS_MODULE,
3136 .ident = XPRT_TRANSPORT_LOCAL,
3137 .setup = xs_setup_local,
3138};
3139
3140static struct xprt_class xs_udp_transport = {
3141 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3142 .name = "udp",
3143 .owner = THIS_MODULE,
3144 .ident = XPRT_TRANSPORT_UDP,
3145 .setup = xs_setup_udp,
3146};
3147
3148static struct xprt_class xs_tcp_transport = {
3149 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3150 .name = "tcp",
3151 .owner = THIS_MODULE,
3152 .ident = XPRT_TRANSPORT_TCP,
3153 .setup = xs_setup_tcp,
3154};
3155
3156static struct xprt_class xs_bc_tcp_transport = {
3157 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3158 .name = "tcp NFSv4.1 backchannel",
3159 .owner = THIS_MODULE,
3160 .ident = XPRT_TRANSPORT_BC_TCP,
3161 .setup = xs_setup_bc_tcp,
3162};
3163
3164/**
3165 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3166 *
3167 */
3168int init_socket_xprt(void)
3169{
3170#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3171 if (!sunrpc_table_header)
3172 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3173#endif
3174
3175 xprt_register_transport(&xs_local_transport);
3176 xprt_register_transport(&xs_udp_transport);
3177 xprt_register_transport(&xs_tcp_transport);
3178 xprt_register_transport(&xs_bc_tcp_transport);
3179
3180 return 0;
3181}
3182
3183/**
3184 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3185 *
3186 */
3187void cleanup_socket_xprt(void)
3188{
3189#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3190 if (sunrpc_table_header) {
3191 unregister_sysctl_table(sunrpc_table_header);
3192 sunrpc_table_header = NULL;
3193 }
3194#endif
3195
3196 xprt_unregister_transport(&xs_local_transport);
3197 xprt_unregister_transport(&xs_udp_transport);
3198 xprt_unregister_transport(&xs_tcp_transport);
3199 xprt_unregister_transport(&xs_bc_tcp_transport);
3200}
3201
3202static int param_set_uint_minmax(const char *val,
3203 const struct kernel_param *kp,
3204 unsigned int min, unsigned int max)
3205{
3206 unsigned int num;
3207 int ret;
3208
3209 if (!val)
3210 return -EINVAL;
3211 ret = kstrtouint(val, 0, &num);
3212 if (ret == -EINVAL || num < min || num > max)
3213 return -EINVAL;
3214 *((unsigned int *)kp->arg) = num;
3215 return 0;
3216}
3217
3218static int param_set_portnr(const char *val, const struct kernel_param *kp)
3219{
3220 if (kp->arg == &xprt_min_resvport)
3221 return param_set_uint_minmax(val, kp,
3222 RPC_MIN_RESVPORT,
3223 xprt_max_resvport);
3224 return param_set_uint_minmax(val, kp,
3225 xprt_min_resvport,
3226 RPC_MAX_RESVPORT);
3227}
3228
3229static const struct kernel_param_ops param_ops_portnr = {
3230 .set = param_set_portnr,
3231 .get = param_get_uint,
3232};
3233
3234#define param_check_portnr(name, p) \
3235 __param_check(name, p, unsigned int);
3236
3237module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3238module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3239
3240static int param_set_slot_table_size(const char *val,
3241 const struct kernel_param *kp)
3242{
3243 return param_set_uint_minmax(val, kp,
3244 RPC_MIN_SLOT_TABLE,
3245 RPC_MAX_SLOT_TABLE);
3246}
3247
3248static const struct kernel_param_ops param_ops_slot_table_size = {
3249 .set = param_set_slot_table_size,
3250 .get = param_get_uint,
3251};
3252
3253#define param_check_slot_table_size(name, p) \
3254 __param_check(name, p, unsigned int);
3255
3256static int param_set_max_slot_table_size(const char *val,
3257 const struct kernel_param *kp)
3258{
3259 return param_set_uint_minmax(val, kp,
3260 RPC_MIN_SLOT_TABLE,
3261 RPC_MAX_SLOT_TABLE_LIMIT);
3262}
3263
3264static const struct kernel_param_ops param_ops_max_slot_table_size = {
3265 .set = param_set_max_slot_table_size,
3266 .get = param_get_uint,
3267};
3268
3269#define param_check_max_slot_table_size(name, p) \
3270 __param_check(name, p, unsigned int);
3271
3272module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3273 slot_table_size, 0644);
3274module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3275 max_slot_table_size, 0644);
3276module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3277 slot_table_size, 0644);
3278