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