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 <net/tls_prot.h>
51#include <net/handshake.h>
52
53#include <linux/bvec.h>
54#include <linux/highmem.h>
55#include <linux/uio.h>
56#include <linux/sched/mm.h>
57
58#include <trace/events/sock.h>
59#include <trace/events/sunrpc.h>
60
61#include "socklib.h"
62#include "sunrpc.h"
63
64static void xs_close(struct rpc_xprt *xprt);
65static void xs_reset_srcport(struct sock_xprt *transport);
66static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock);
67static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
68 struct socket *sock);
69
70/*
71 * xprtsock tunables
72 */
73static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
74static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
75static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
76
77static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
78static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
79
80#define XS_TCP_LINGER_TO (15U * HZ)
81static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
82
83/*
84 * We can register our own files under /proc/sys/sunrpc by
85 * calling register_sysctl() again. The files in that
86 * directory become the union of all files registered there.
87 *
88 * We simply need to make sure that we don't collide with
89 * someone else's file names!
90 */
91
92static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
93static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
94static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
95static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
96static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
97
98static struct ctl_table_header *sunrpc_table_header;
99
100static struct xprt_class xs_local_transport;
101static struct xprt_class xs_udp_transport;
102static struct xprt_class xs_tcp_transport;
103static struct xprt_class xs_tcp_tls_transport;
104static struct xprt_class xs_bc_tcp_transport;
105
106/*
107 * FIXME: changing the UDP slot table size should also resize the UDP
108 * socket buffers for existing UDP transports
109 */
110static struct ctl_table xs_tunables_table[] = {
111 {
112 .procname = "udp_slot_table_entries",
113 .data = &xprt_udp_slot_table_entries,
114 .maxlen = sizeof(unsigned int),
115 .mode = 0644,
116 .proc_handler = proc_dointvec_minmax,
117 .extra1 = &min_slot_table_size,
118 .extra2 = &max_slot_table_size
119 },
120 {
121 .procname = "tcp_slot_table_entries",
122 .data = &xprt_tcp_slot_table_entries,
123 .maxlen = sizeof(unsigned int),
124 .mode = 0644,
125 .proc_handler = proc_dointvec_minmax,
126 .extra1 = &min_slot_table_size,
127 .extra2 = &max_slot_table_size
128 },
129 {
130 .procname = "tcp_max_slot_table_entries",
131 .data = &xprt_max_tcp_slot_table_entries,
132 .maxlen = sizeof(unsigned int),
133 .mode = 0644,
134 .proc_handler = proc_dointvec_minmax,
135 .extra1 = &min_slot_table_size,
136 .extra2 = &max_tcp_slot_table_limit
137 },
138 {
139 .procname = "min_resvport",
140 .data = &xprt_min_resvport,
141 .maxlen = sizeof(unsigned int),
142 .mode = 0644,
143 .proc_handler = proc_dointvec_minmax,
144 .extra1 = &xprt_min_resvport_limit,
145 .extra2 = &xprt_max_resvport_limit
146 },
147 {
148 .procname = "max_resvport",
149 .data = &xprt_max_resvport,
150 .maxlen = sizeof(unsigned int),
151 .mode = 0644,
152 .proc_handler = proc_dointvec_minmax,
153 .extra1 = &xprt_min_resvport_limit,
154 .extra2 = &xprt_max_resvport_limit
155 },
156 {
157 .procname = "tcp_fin_timeout",
158 .data = &xs_tcp_fin_timeout,
159 .maxlen = sizeof(xs_tcp_fin_timeout),
160 .mode = 0644,
161 .proc_handler = proc_dointvec_jiffies,
162 },
163 { },
164};
165
166/*
167 * Wait duration for a reply from the RPC portmapper.
168 */
169#define XS_BIND_TO (60U * HZ)
170
171/*
172 * Delay if a UDP socket connect error occurs. This is most likely some
173 * kind of resource problem on the local host.
174 */
175#define XS_UDP_REEST_TO (2U * HZ)
176
177/*
178 * The reestablish timeout allows clients to delay for a bit before attempting
179 * to reconnect to a server that just dropped our connection.
180 *
181 * We implement an exponential backoff when trying to reestablish a TCP
182 * transport connection with the server. Some servers like to drop a TCP
183 * connection when they are overworked, so we start with a short timeout and
184 * increase over time if the server is down or not responding.
185 */
186#define XS_TCP_INIT_REEST_TO (3U * HZ)
187
188/*
189 * TCP idle timeout; client drops the transport socket if it is idle
190 * for this long. Note that we also timeout UDP sockets to prevent
191 * holding port numbers when there is no RPC traffic.
192 */
193#define XS_IDLE_DISC_TO (5U * 60 * HZ)
194
195/*
196 * TLS handshake timeout.
197 */
198#define XS_TLS_HANDSHAKE_TO (10U * HZ)
199
200#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
201# undef RPC_DEBUG_DATA
202# define RPCDBG_FACILITY RPCDBG_TRANS
203#endif
204
205#ifdef RPC_DEBUG_DATA
206static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
207{
208 u8 *buf = (u8 *) packet;
209 int j;
210
211 dprintk("RPC: %s\n", msg);
212 for (j = 0; j < count && j < 128; j += 4) {
213 if (!(j & 31)) {
214 if (j)
215 dprintk("\n");
216 dprintk("0x%04x ", j);
217 }
218 dprintk("%02x%02x%02x%02x ",
219 buf[j], buf[j+1], buf[j+2], buf[j+3]);
220 }
221 dprintk("\n");
222}
223#else
224static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
225{
226 /* NOP */
227}
228#endif
229
230static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
231{
232 return (struct rpc_xprt *) sk->sk_user_data;
233}
234
235static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
236{
237 return (struct sockaddr *) &xprt->addr;
238}
239
240static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
241{
242 return (struct sockaddr_un *) &xprt->addr;
243}
244
245static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
246{
247 return (struct sockaddr_in *) &xprt->addr;
248}
249
250static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
251{
252 return (struct sockaddr_in6 *) &xprt->addr;
253}
254
255static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
256{
257 struct sockaddr *sap = xs_addr(xprt);
258 struct sockaddr_in6 *sin6;
259 struct sockaddr_in *sin;
260 struct sockaddr_un *sun;
261 char buf[128];
262
263 switch (sap->sa_family) {
264 case AF_LOCAL:
265 sun = xs_addr_un(xprt);
266 if (sun->sun_path[0]) {
267 strscpy(buf, sun->sun_path, sizeof(buf));
268 } else {
269 buf[0] = '@';
270 strscpy(buf+1, sun->sun_path+1, sizeof(buf)-1);
271 }
272 xprt->address_strings[RPC_DISPLAY_ADDR] =
273 kstrdup(buf, GFP_KERNEL);
274 break;
275 case AF_INET:
276 (void)rpc_ntop(sap, buf, sizeof(buf));
277 xprt->address_strings[RPC_DISPLAY_ADDR] =
278 kstrdup(buf, GFP_KERNEL);
279 sin = xs_addr_in(xprt);
280 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
281 break;
282 case AF_INET6:
283 (void)rpc_ntop(sap, buf, sizeof(buf));
284 xprt->address_strings[RPC_DISPLAY_ADDR] =
285 kstrdup(buf, GFP_KERNEL);
286 sin6 = xs_addr_in6(xprt);
287 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
288 break;
289 default:
290 BUG();
291 }
292
293 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
294}
295
296static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
297{
298 struct sockaddr *sap = xs_addr(xprt);
299 char buf[128];
300
301 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
302 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
303
304 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
305 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
306}
307
308static void xs_format_peer_addresses(struct rpc_xprt *xprt,
309 const char *protocol,
310 const char *netid)
311{
312 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
313 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
314 xs_format_common_peer_addresses(xprt);
315 xs_format_common_peer_ports(xprt);
316}
317
318static void xs_update_peer_port(struct rpc_xprt *xprt)
319{
320 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
321 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
322
323 xs_format_common_peer_ports(xprt);
324}
325
326static void xs_free_peer_addresses(struct rpc_xprt *xprt)
327{
328 unsigned int i;
329
330 for (i = 0; i < RPC_DISPLAY_MAX; i++)
331 switch (i) {
332 case RPC_DISPLAY_PROTO:
333 case RPC_DISPLAY_NETID:
334 continue;
335 default:
336 kfree(xprt->address_strings[i]);
337 }
338}
339
340static size_t
341xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp)
342{
343 size_t i,n;
344
345 if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES))
346 return want;
347 n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT;
348 for (i = 0; i < n; i++) {
349 if (buf->pages[i])
350 continue;
351 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp);
352 if (!buf->pages[i]) {
353 i *= PAGE_SIZE;
354 return i > buf->page_base ? i - buf->page_base : 0;
355 }
356 }
357 return want;
358}
359
360static int
361xs_sock_process_cmsg(struct socket *sock, struct msghdr *msg,
362 struct cmsghdr *cmsg, int ret)
363{
364 u8 content_type = tls_get_record_type(sock->sk, cmsg);
365 u8 level, description;
366
367 switch (content_type) {
368 case 0:
369 break;
370 case TLS_RECORD_TYPE_DATA:
371 /* TLS sets EOR at the end of each application data
372 * record, even though there might be more frames
373 * waiting to be decrypted.
374 */
375 msg->msg_flags &= ~MSG_EOR;
376 break;
377 case TLS_RECORD_TYPE_ALERT:
378 tls_alert_recv(sock->sk, msg, &level, &description);
379 ret = (level == TLS_ALERT_LEVEL_FATAL) ?
380 -EACCES : -EAGAIN;
381 break;
382 default:
383 /* discard this record type */
384 ret = -EAGAIN;
385 }
386 return ret;
387}
388
389static int
390xs_sock_recv_cmsg(struct socket *sock, struct msghdr *msg, int flags)
391{
392 union {
393 struct cmsghdr cmsg;
394 u8 buf[CMSG_SPACE(sizeof(u8))];
395 } u;
396 int ret;
397
398 msg->msg_control = &u;
399 msg->msg_controllen = sizeof(u);
400 ret = sock_recvmsg(sock, msg, flags);
401 if (msg->msg_controllen != sizeof(u))
402 ret = xs_sock_process_cmsg(sock, msg, &u.cmsg, ret);
403 return ret;
404}
405
406static ssize_t
407xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek)
408{
409 ssize_t ret;
410 if (seek != 0)
411 iov_iter_advance(&msg->msg_iter, seek);
412 ret = xs_sock_recv_cmsg(sock, msg, flags);
413 return ret > 0 ? ret + seek : ret;
414}
415
416static ssize_t
417xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags,
418 struct kvec *kvec, size_t count, size_t seek)
419{
420 iov_iter_kvec(&msg->msg_iter, ITER_DEST, kvec, 1, count);
421 return xs_sock_recvmsg(sock, msg, flags, seek);
422}
423
424static ssize_t
425xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags,
426 struct bio_vec *bvec, unsigned long nr, size_t count,
427 size_t seek)
428{
429 iov_iter_bvec(&msg->msg_iter, ITER_DEST, bvec, nr, count);
430 return xs_sock_recvmsg(sock, msg, flags, seek);
431}
432
433static ssize_t
434xs_read_discard(struct socket *sock, struct msghdr *msg, int flags,
435 size_t count)
436{
437 iov_iter_discard(&msg->msg_iter, ITER_DEST, count);
438 return xs_sock_recv_cmsg(sock, msg, flags);
439}
440
441#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
442static void
443xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
444{
445 struct bvec_iter bi = {
446 .bi_size = count,
447 };
448 struct bio_vec bv;
449
450 bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
451 for_each_bvec(bv, bvec, bi, bi)
452 flush_dcache_page(bv.bv_page);
453}
454#else
455static inline void
456xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
457{
458}
459#endif
460
461static ssize_t
462xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
463 struct xdr_buf *buf, size_t count, size_t seek, size_t *read)
464{
465 size_t want, seek_init = seek, offset = 0;
466 ssize_t ret;
467
468 want = min_t(size_t, count, buf->head[0].iov_len);
469 if (seek < want) {
470 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek);
471 if (ret <= 0)
472 goto sock_err;
473 offset += ret;
474 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
475 goto out;
476 if (ret != want)
477 goto out;
478 seek = 0;
479 } else {
480 seek -= want;
481 offset += want;
482 }
483
484 want = xs_alloc_sparse_pages(
485 buf, min_t(size_t, count - offset, buf->page_len),
486 GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN);
487 if (seek < want) {
488 ret = xs_read_bvec(sock, msg, flags, buf->bvec,
489 xdr_buf_pagecount(buf),
490 want + buf->page_base,
491 seek + buf->page_base);
492 if (ret <= 0)
493 goto sock_err;
494 xs_flush_bvec(buf->bvec, ret, seek + buf->page_base);
495 ret -= buf->page_base;
496 offset += ret;
497 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
498 goto out;
499 if (ret != want)
500 goto out;
501 seek = 0;
502 } else {
503 seek -= want;
504 offset += want;
505 }
506
507 want = min_t(size_t, count - offset, buf->tail[0].iov_len);
508 if (seek < want) {
509 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek);
510 if (ret <= 0)
511 goto sock_err;
512 offset += ret;
513 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
514 goto out;
515 if (ret != want)
516 goto out;
517 } else if (offset < seek_init)
518 offset = seek_init;
519 ret = -EMSGSIZE;
520out:
521 *read = offset - seek_init;
522 return ret;
523sock_err:
524 offset += seek;
525 goto out;
526}
527
528static void
529xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf)
530{
531 if (!transport->recv.copied) {
532 if (buf->head[0].iov_len >= transport->recv.offset)
533 memcpy(buf->head[0].iov_base,
534 &transport->recv.xid,
535 transport->recv.offset);
536 transport->recv.copied = transport->recv.offset;
537 }
538}
539
540static bool
541xs_read_stream_request_done(struct sock_xprt *transport)
542{
543 return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT);
544}
545
546static void
547xs_read_stream_check_eor(struct sock_xprt *transport,
548 struct msghdr *msg)
549{
550 if (xs_read_stream_request_done(transport))
551 msg->msg_flags |= MSG_EOR;
552}
553
554static ssize_t
555xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg,
556 int flags, struct rpc_rqst *req)
557{
558 struct xdr_buf *buf = &req->rq_private_buf;
559 size_t want, read;
560 ssize_t ret;
561
562 xs_read_header(transport, buf);
563
564 want = transport->recv.len - transport->recv.offset;
565 if (want != 0) {
566 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf,
567 transport->recv.copied + want,
568 transport->recv.copied,
569 &read);
570 transport->recv.offset += read;
571 transport->recv.copied += read;
572 }
573
574 if (transport->recv.offset == transport->recv.len)
575 xs_read_stream_check_eor(transport, msg);
576
577 if (want == 0)
578 return 0;
579
580 switch (ret) {
581 default:
582 break;
583 case -EFAULT:
584 case -EMSGSIZE:
585 msg->msg_flags |= MSG_TRUNC;
586 return read;
587 case 0:
588 return -ESHUTDOWN;
589 }
590 return ret < 0 ? ret : read;
591}
592
593static size_t
594xs_read_stream_headersize(bool isfrag)
595{
596 if (isfrag)
597 return sizeof(__be32);
598 return 3 * sizeof(__be32);
599}
600
601static ssize_t
602xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg,
603 int flags, size_t want, size_t seek)
604{
605 struct kvec kvec = {
606 .iov_base = &transport->recv.fraghdr,
607 .iov_len = want,
608 };
609 return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek);
610}
611
612#if defined(CONFIG_SUNRPC_BACKCHANNEL)
613static ssize_t
614xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
615{
616 struct rpc_xprt *xprt = &transport->xprt;
617 struct rpc_rqst *req;
618 ssize_t ret;
619
620 /* Is this transport associated with the backchannel? */
621 if (!xprt->bc_serv)
622 return -ESHUTDOWN;
623
624 /* Look up and lock the request corresponding to the given XID */
625 req = xprt_lookup_bc_request(xprt, transport->recv.xid);
626 if (!req) {
627 printk(KERN_WARNING "Callback slot table overflowed\n");
628 return -ESHUTDOWN;
629 }
630 if (transport->recv.copied && !req->rq_private_buf.len)
631 return -ESHUTDOWN;
632
633 ret = xs_read_stream_request(transport, msg, flags, req);
634 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
635 xprt_complete_bc_request(req, transport->recv.copied);
636 else
637 req->rq_private_buf.len = transport->recv.copied;
638
639 return ret;
640}
641#else /* CONFIG_SUNRPC_BACKCHANNEL */
642static ssize_t
643xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
644{
645 return -ESHUTDOWN;
646}
647#endif /* CONFIG_SUNRPC_BACKCHANNEL */
648
649static ssize_t
650xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags)
651{
652 struct rpc_xprt *xprt = &transport->xprt;
653 struct rpc_rqst *req;
654 ssize_t ret = 0;
655
656 /* Look up and lock the request corresponding to the given XID */
657 spin_lock(&xprt->queue_lock);
658 req = xprt_lookup_rqst(xprt, transport->recv.xid);
659 if (!req || (transport->recv.copied && !req->rq_private_buf.len)) {
660 msg->msg_flags |= MSG_TRUNC;
661 goto out;
662 }
663 xprt_pin_rqst(req);
664 spin_unlock(&xprt->queue_lock);
665
666 ret = xs_read_stream_request(transport, msg, flags, req);
667
668 spin_lock(&xprt->queue_lock);
669 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
670 xprt_complete_rqst(req->rq_task, transport->recv.copied);
671 else
672 req->rq_private_buf.len = transport->recv.copied;
673 xprt_unpin_rqst(req);
674out:
675 spin_unlock(&xprt->queue_lock);
676 return ret;
677}
678
679static ssize_t
680xs_read_stream(struct sock_xprt *transport, int flags)
681{
682 struct msghdr msg = { 0 };
683 size_t want, read = 0;
684 ssize_t ret = 0;
685
686 if (transport->recv.len == 0) {
687 want = xs_read_stream_headersize(transport->recv.copied != 0);
688 ret = xs_read_stream_header(transport, &msg, flags, want,
689 transport->recv.offset);
690 if (ret <= 0)
691 goto out_err;
692 transport->recv.offset = ret;
693 if (transport->recv.offset != want)
694 return transport->recv.offset;
695 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) &
696 RPC_FRAGMENT_SIZE_MASK;
697 transport->recv.offset -= sizeof(transport->recv.fraghdr);
698 read = ret;
699 }
700
701 switch (be32_to_cpu(transport->recv.calldir)) {
702 default:
703 msg.msg_flags |= MSG_TRUNC;
704 break;
705 case RPC_CALL:
706 ret = xs_read_stream_call(transport, &msg, flags);
707 break;
708 case RPC_REPLY:
709 ret = xs_read_stream_reply(transport, &msg, flags);
710 }
711 if (msg.msg_flags & MSG_TRUNC) {
712 transport->recv.calldir = cpu_to_be32(-1);
713 transport->recv.copied = -1;
714 }
715 if (ret < 0)
716 goto out_err;
717 read += ret;
718 if (transport->recv.offset < transport->recv.len) {
719 if (!(msg.msg_flags & MSG_TRUNC))
720 return read;
721 msg.msg_flags = 0;
722 ret = xs_read_discard(transport->sock, &msg, flags,
723 transport->recv.len - transport->recv.offset);
724 if (ret <= 0)
725 goto out_err;
726 transport->recv.offset += ret;
727 read += ret;
728 if (transport->recv.offset != transport->recv.len)
729 return read;
730 }
731 if (xs_read_stream_request_done(transport)) {
732 trace_xs_stream_read_request(transport);
733 transport->recv.copied = 0;
734 }
735 transport->recv.offset = 0;
736 transport->recv.len = 0;
737 return read;
738out_err:
739 return ret != 0 ? ret : -ESHUTDOWN;
740}
741
742static __poll_t xs_poll_socket(struct sock_xprt *transport)
743{
744 return transport->sock->ops->poll(transport->file, transport->sock,
745 NULL);
746}
747
748static bool xs_poll_socket_readable(struct sock_xprt *transport)
749{
750 __poll_t events = xs_poll_socket(transport);
751
752 return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP);
753}
754
755static void xs_poll_check_readable(struct sock_xprt *transport)
756{
757
758 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
759 if (test_bit(XPRT_SOCK_IGNORE_RECV, &transport->sock_state))
760 return;
761 if (!xs_poll_socket_readable(transport))
762 return;
763 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
764 queue_work(xprtiod_workqueue, &transport->recv_worker);
765}
766
767static void xs_stream_data_receive(struct sock_xprt *transport)
768{
769 size_t read = 0;
770 ssize_t ret = 0;
771
772 mutex_lock(&transport->recv_mutex);
773 if (transport->sock == NULL)
774 goto out;
775 for (;;) {
776 ret = xs_read_stream(transport, MSG_DONTWAIT);
777 if (ret < 0)
778 break;
779 read += ret;
780 cond_resched();
781 }
782 if (ret == -ESHUTDOWN)
783 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
784 else if (ret == -EACCES)
785 xprt_wake_pending_tasks(&transport->xprt, -EACCES);
786 else
787 xs_poll_check_readable(transport);
788out:
789 mutex_unlock(&transport->recv_mutex);
790 trace_xs_stream_read_data(&transport->xprt, ret, read);
791}
792
793static void xs_stream_data_receive_workfn(struct work_struct *work)
794{
795 struct sock_xprt *transport =
796 container_of(work, struct sock_xprt, recv_worker);
797 unsigned int pflags = memalloc_nofs_save();
798
799 xs_stream_data_receive(transport);
800 memalloc_nofs_restore(pflags);
801}
802
803static void
804xs_stream_reset_connect(struct sock_xprt *transport)
805{
806 transport->recv.offset = 0;
807 transport->recv.len = 0;
808 transport->recv.copied = 0;
809 transport->xmit.offset = 0;
810}
811
812static void
813xs_stream_start_connect(struct sock_xprt *transport)
814{
815 transport->xprt.stat.connect_count++;
816 transport->xprt.stat.connect_start = jiffies;
817}
818
819#define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
820
821/**
822 * xs_nospace - handle transmit was incomplete
823 * @req: pointer to RPC request
824 * @transport: pointer to struct sock_xprt
825 *
826 */
827static int xs_nospace(struct rpc_rqst *req, struct sock_xprt *transport)
828{
829 struct rpc_xprt *xprt = &transport->xprt;
830 struct sock *sk = transport->inet;
831 int ret = -EAGAIN;
832
833 trace_rpc_socket_nospace(req, transport);
834
835 /* Protect against races with write_space */
836 spin_lock(&xprt->transport_lock);
837
838 /* Don't race with disconnect */
839 if (xprt_connected(xprt)) {
840 /* wait for more buffer space */
841 set_bit(XPRT_SOCK_NOSPACE, &transport->sock_state);
842 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
843 sk->sk_write_pending++;
844 xprt_wait_for_buffer_space(xprt);
845 } else
846 ret = -ENOTCONN;
847
848 spin_unlock(&xprt->transport_lock);
849 return ret;
850}
851
852static int xs_sock_nospace(struct rpc_rqst *req)
853{
854 struct sock_xprt *transport =
855 container_of(req->rq_xprt, struct sock_xprt, xprt);
856 struct sock *sk = transport->inet;
857 int ret = -EAGAIN;
858
859 lock_sock(sk);
860 if (!sock_writeable(sk))
861 ret = xs_nospace(req, transport);
862 release_sock(sk);
863 return ret;
864}
865
866static int xs_stream_nospace(struct rpc_rqst *req, bool vm_wait)
867{
868 struct sock_xprt *transport =
869 container_of(req->rq_xprt, struct sock_xprt, xprt);
870 struct sock *sk = transport->inet;
871 int ret = -EAGAIN;
872
873 if (vm_wait)
874 return -ENOBUFS;
875 lock_sock(sk);
876 if (!sk_stream_memory_free(sk))
877 ret = xs_nospace(req, transport);
878 release_sock(sk);
879 return ret;
880}
881
882static int xs_stream_prepare_request(struct rpc_rqst *req, struct xdr_buf *buf)
883{
884 return xdr_alloc_bvec(buf, rpc_task_gfp_mask());
885}
886
887static void xs_stream_abort_send_request(struct rpc_rqst *req)
888{
889 struct rpc_xprt *xprt = req->rq_xprt;
890 struct sock_xprt *transport =
891 container_of(xprt, struct sock_xprt, xprt);
892
893 if (transport->xmit.offset != 0 &&
894 !test_bit(XPRT_CLOSE_WAIT, &xprt->state))
895 xprt_force_disconnect(xprt);
896}
897
898/*
899 * Determine if the previous message in the stream was aborted before it
900 * could complete transmission.
901 */
902static bool
903xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req)
904{
905 return transport->xmit.offset != 0 && req->rq_bytes_sent == 0;
906}
907
908/*
909 * Return the stream record marker field for a record of length < 2^31-1
910 */
911static rpc_fraghdr
912xs_stream_record_marker(struct xdr_buf *xdr)
913{
914 if (!xdr->len)
915 return 0;
916 return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len);
917}
918
919/**
920 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
921 * @req: pointer to RPC request
922 *
923 * Return values:
924 * 0: The request has been sent
925 * EAGAIN: The socket was blocked, please call again later to
926 * complete the request
927 * ENOTCONN: Caller needs to invoke connect logic then call again
928 * other: Some other error occurred, the request was not sent
929 */
930static int xs_local_send_request(struct rpc_rqst *req)
931{
932 struct rpc_xprt *xprt = req->rq_xprt;
933 struct sock_xprt *transport =
934 container_of(xprt, struct sock_xprt, xprt);
935 struct xdr_buf *xdr = &req->rq_snd_buf;
936 rpc_fraghdr rm = xs_stream_record_marker(xdr);
937 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
938 struct msghdr msg = {
939 .msg_flags = XS_SENDMSG_FLAGS,
940 };
941 bool vm_wait;
942 unsigned int sent;
943 int status;
944
945 /* Close the stream if the previous transmission was incomplete */
946 if (xs_send_request_was_aborted(transport, req)) {
947 xprt_force_disconnect(xprt);
948 return -ENOTCONN;
949 }
950
951 xs_pktdump("packet data:",
952 req->rq_svec->iov_base, req->rq_svec->iov_len);
953
954 vm_wait = sk_stream_is_writeable(transport->inet) ? true : false;
955
956 req->rq_xtime = ktime_get();
957 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
958 transport->xmit.offset, rm, &sent);
959 dprintk("RPC: %s(%u) = %d\n",
960 __func__, xdr->len - transport->xmit.offset, status);
961
962 if (likely(sent > 0) || status == 0) {
963 transport->xmit.offset += sent;
964 req->rq_bytes_sent = transport->xmit.offset;
965 if (likely(req->rq_bytes_sent >= msglen)) {
966 req->rq_xmit_bytes_sent += transport->xmit.offset;
967 transport->xmit.offset = 0;
968 return 0;
969 }
970 status = -EAGAIN;
971 vm_wait = false;
972 }
973
974 switch (status) {
975 case -EAGAIN:
976 status = xs_stream_nospace(req, vm_wait);
977 break;
978 default:
979 dprintk("RPC: sendmsg returned unrecognized error %d\n",
980 -status);
981 fallthrough;
982 case -EPIPE:
983 xprt_force_disconnect(xprt);
984 status = -ENOTCONN;
985 }
986
987 return status;
988}
989
990/**
991 * xs_udp_send_request - write an RPC request to a UDP socket
992 * @req: pointer to RPC request
993 *
994 * Return values:
995 * 0: The request has been sent
996 * EAGAIN: The socket was blocked, please call again later to
997 * complete the request
998 * ENOTCONN: Caller needs to invoke connect logic then call again
999 * other: Some other error occurred, the request was not sent
1000 */
1001static int xs_udp_send_request(struct rpc_rqst *req)
1002{
1003 struct rpc_xprt *xprt = req->rq_xprt;
1004 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1005 struct xdr_buf *xdr = &req->rq_snd_buf;
1006 struct msghdr msg = {
1007 .msg_name = xs_addr(xprt),
1008 .msg_namelen = xprt->addrlen,
1009 .msg_flags = XS_SENDMSG_FLAGS,
1010 };
1011 unsigned int sent;
1012 int status;
1013
1014 xs_pktdump("packet data:",
1015 req->rq_svec->iov_base,
1016 req->rq_svec->iov_len);
1017
1018 if (!xprt_bound(xprt))
1019 return -ENOTCONN;
1020
1021 if (!xprt_request_get_cong(xprt, req))
1022 return -EBADSLT;
1023
1024 status = xdr_alloc_bvec(xdr, rpc_task_gfp_mask());
1025 if (status < 0)
1026 return status;
1027 req->rq_xtime = ktime_get();
1028 status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, 0, &sent);
1029
1030 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
1031 xdr->len, status);
1032
1033 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
1034 if (status == -EPERM)
1035 goto process_status;
1036
1037 if (status == -EAGAIN && sock_writeable(transport->inet))
1038 status = -ENOBUFS;
1039
1040 if (sent > 0 || status == 0) {
1041 req->rq_xmit_bytes_sent += sent;
1042 if (sent >= req->rq_slen)
1043 return 0;
1044 /* Still some bytes left; set up for a retry later. */
1045 status = -EAGAIN;
1046 }
1047
1048process_status:
1049 switch (status) {
1050 case -ENOTSOCK:
1051 status = -ENOTCONN;
1052 /* Should we call xs_close() here? */
1053 break;
1054 case -EAGAIN:
1055 status = xs_sock_nospace(req);
1056 break;
1057 case -ENETUNREACH:
1058 case -ENOBUFS:
1059 case -EPIPE:
1060 case -ECONNREFUSED:
1061 case -EPERM:
1062 /* When the server has died, an ICMP port unreachable message
1063 * prompts ECONNREFUSED. */
1064 break;
1065 default:
1066 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1067 -status);
1068 }
1069
1070 return status;
1071}
1072
1073/**
1074 * xs_tcp_send_request - write an RPC request to a TCP socket
1075 * @req: pointer to RPC request
1076 *
1077 * Return values:
1078 * 0: The request has been sent
1079 * EAGAIN: The socket was blocked, please call again later to
1080 * complete the request
1081 * ENOTCONN: Caller needs to invoke connect logic then call again
1082 * other: Some other error occurred, the request was not sent
1083 *
1084 * XXX: In the case of soft timeouts, should we eventually give up
1085 * if sendmsg is not able to make progress?
1086 */
1087static int xs_tcp_send_request(struct rpc_rqst *req)
1088{
1089 struct rpc_xprt *xprt = req->rq_xprt;
1090 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1091 struct xdr_buf *xdr = &req->rq_snd_buf;
1092 rpc_fraghdr rm = xs_stream_record_marker(xdr);
1093 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
1094 struct msghdr msg = {
1095 .msg_flags = XS_SENDMSG_FLAGS,
1096 };
1097 bool vm_wait;
1098 unsigned int sent;
1099 int status;
1100
1101 /* Close the stream if the previous transmission was incomplete */
1102 if (xs_send_request_was_aborted(transport, req)) {
1103 if (transport->sock != NULL)
1104 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
1105 return -ENOTCONN;
1106 }
1107 if (!transport->inet)
1108 return -ENOTCONN;
1109
1110 xs_pktdump("packet data:",
1111 req->rq_svec->iov_base,
1112 req->rq_svec->iov_len);
1113
1114 if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
1115 xs_tcp_set_socket_timeouts(xprt, transport->sock);
1116
1117 xs_set_srcport(transport, transport->sock);
1118
1119 /* Continue transmitting the packet/record. We must be careful
1120 * to cope with writespace callbacks arriving _after_ we have
1121 * called sendmsg(). */
1122 req->rq_xtime = ktime_get();
1123 tcp_sock_set_cork(transport->inet, true);
1124
1125 vm_wait = sk_stream_is_writeable(transport->inet) ? true : false;
1126
1127 do {
1128 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
1129 transport->xmit.offset, rm, &sent);
1130
1131 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
1132 xdr->len - transport->xmit.offset, status);
1133
1134 /* If we've sent the entire packet, immediately
1135 * reset the count of bytes sent. */
1136 transport->xmit.offset += sent;
1137 req->rq_bytes_sent = transport->xmit.offset;
1138 if (likely(req->rq_bytes_sent >= msglen)) {
1139 req->rq_xmit_bytes_sent += transport->xmit.offset;
1140 transport->xmit.offset = 0;
1141 if (atomic_long_read(&xprt->xmit_queuelen) == 1)
1142 tcp_sock_set_cork(transport->inet, false);
1143 return 0;
1144 }
1145
1146 WARN_ON_ONCE(sent == 0 && status == 0);
1147
1148 if (sent > 0)
1149 vm_wait = false;
1150
1151 } while (status == 0);
1152
1153 switch (status) {
1154 case -ENOTSOCK:
1155 status = -ENOTCONN;
1156 /* Should we call xs_close() here? */
1157 break;
1158 case -EAGAIN:
1159 status = xs_stream_nospace(req, vm_wait);
1160 break;
1161 case -ECONNRESET:
1162 case -ECONNREFUSED:
1163 case -ENOTCONN:
1164 case -EADDRINUSE:
1165 case -ENOBUFS:
1166 case -EPIPE:
1167 break;
1168 default:
1169 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1170 -status);
1171 }
1172
1173 return status;
1174}
1175
1176static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1177{
1178 transport->old_data_ready = sk->sk_data_ready;
1179 transport->old_state_change = sk->sk_state_change;
1180 transport->old_write_space = sk->sk_write_space;
1181 transport->old_error_report = sk->sk_error_report;
1182}
1183
1184static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1185{
1186 sk->sk_data_ready = transport->old_data_ready;
1187 sk->sk_state_change = transport->old_state_change;
1188 sk->sk_write_space = transport->old_write_space;
1189 sk->sk_error_report = transport->old_error_report;
1190}
1191
1192static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
1193{
1194 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1195
1196 transport->xprt_err = 0;
1197 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1198 clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state);
1199 clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state);
1200 clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state);
1201 clear_bit(XPRT_SOCK_NOSPACE, &transport->sock_state);
1202}
1203
1204static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr)
1205{
1206 set_bit(nr, &transport->sock_state);
1207 queue_work(xprtiod_workqueue, &transport->error_worker);
1208}
1209
1210static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1211{
1212 xprt->connect_cookie++;
1213 smp_mb__before_atomic();
1214 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1215 clear_bit(XPRT_CLOSING, &xprt->state);
1216 xs_sock_reset_state_flags(xprt);
1217 smp_mb__after_atomic();
1218}
1219
1220/**
1221 * xs_error_report - callback to handle TCP socket state errors
1222 * @sk: socket
1223 *
1224 * Note: we don't call sock_error() since there may be a rpc_task
1225 * using the socket, and so we don't want to clear sk->sk_err.
1226 */
1227static void xs_error_report(struct sock *sk)
1228{
1229 struct sock_xprt *transport;
1230 struct rpc_xprt *xprt;
1231
1232 if (!(xprt = xprt_from_sock(sk)))
1233 return;
1234
1235 transport = container_of(xprt, struct sock_xprt, xprt);
1236 transport->xprt_err = -sk->sk_err;
1237 if (transport->xprt_err == 0)
1238 return;
1239 dprintk("RPC: xs_error_report client %p, error=%d...\n",
1240 xprt, -transport->xprt_err);
1241 trace_rpc_socket_error(xprt, sk->sk_socket, transport->xprt_err);
1242
1243 /* barrier ensures xprt_err is set before XPRT_SOCK_WAKE_ERROR */
1244 smp_mb__before_atomic();
1245 xs_run_error_worker(transport, XPRT_SOCK_WAKE_ERROR);
1246}
1247
1248static void xs_reset_transport(struct sock_xprt *transport)
1249{
1250 struct socket *sock = transport->sock;
1251 struct sock *sk = transport->inet;
1252 struct rpc_xprt *xprt = &transport->xprt;
1253 struct file *filp = transport->file;
1254
1255 if (sk == NULL)
1256 return;
1257 /*
1258 * Make sure we're calling this in a context from which it is safe
1259 * to call __fput_sync(). In practice that means rpciod and the
1260 * system workqueue.
1261 */
1262 if (!(current->flags & PF_WQ_WORKER)) {
1263 WARN_ON_ONCE(1);
1264 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
1265 return;
1266 }
1267
1268 if (atomic_read(&transport->xprt.swapper))
1269 sk_clear_memalloc(sk);
1270
1271 tls_handshake_cancel(sk);
1272
1273 kernel_sock_shutdown(sock, SHUT_RDWR);
1274
1275 mutex_lock(&transport->recv_mutex);
1276 lock_sock(sk);
1277 transport->inet = NULL;
1278 transport->sock = NULL;
1279 transport->file = NULL;
1280
1281 sk->sk_user_data = NULL;
1282
1283 xs_restore_old_callbacks(transport, sk);
1284 xprt_clear_connected(xprt);
1285 xs_sock_reset_connection_flags(xprt);
1286 /* Reset stream record info */
1287 xs_stream_reset_connect(transport);
1288 release_sock(sk);
1289 mutex_unlock(&transport->recv_mutex);
1290
1291 trace_rpc_socket_close(xprt, sock);
1292 __fput_sync(filp);
1293
1294 xprt_disconnect_done(xprt);
1295}
1296
1297/**
1298 * xs_close - close a socket
1299 * @xprt: transport
1300 *
1301 * This is used when all requests are complete; ie, no DRC state remains
1302 * on the server we want to save.
1303 *
1304 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1305 * xs_reset_transport() zeroing the socket from underneath a writer.
1306 */
1307static void xs_close(struct rpc_xprt *xprt)
1308{
1309 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1310
1311 dprintk("RPC: xs_close xprt %p\n", xprt);
1312
1313 if (transport->sock)
1314 tls_handshake_close(transport->sock);
1315 xs_reset_transport(transport);
1316 xprt->reestablish_timeout = 0;
1317}
1318
1319static void xs_inject_disconnect(struct rpc_xprt *xprt)
1320{
1321 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
1322 xprt);
1323 xprt_disconnect_done(xprt);
1324}
1325
1326static void xs_xprt_free(struct rpc_xprt *xprt)
1327{
1328 xs_free_peer_addresses(xprt);
1329 xprt_free(xprt);
1330}
1331
1332/**
1333 * xs_destroy - prepare to shutdown a transport
1334 * @xprt: doomed transport
1335 *
1336 */
1337static void xs_destroy(struct rpc_xprt *xprt)
1338{
1339 struct sock_xprt *transport = container_of(xprt,
1340 struct sock_xprt, xprt);
1341 dprintk("RPC: xs_destroy xprt %p\n", xprt);
1342
1343 cancel_delayed_work_sync(&transport->connect_worker);
1344 xs_close(xprt);
1345 cancel_work_sync(&transport->recv_worker);
1346 cancel_work_sync(&transport->error_worker);
1347 xs_xprt_free(xprt);
1348 module_put(THIS_MODULE);
1349}
1350
1351/**
1352 * xs_udp_data_read_skb - receive callback for UDP sockets
1353 * @xprt: transport
1354 * @sk: socket
1355 * @skb: skbuff
1356 *
1357 */
1358static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1359 struct sock *sk,
1360 struct sk_buff *skb)
1361{
1362 struct rpc_task *task;
1363 struct rpc_rqst *rovr;
1364 int repsize, copied;
1365 u32 _xid;
1366 __be32 *xp;
1367
1368 repsize = skb->len;
1369 if (repsize < 4) {
1370 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
1371 return;
1372 }
1373
1374 /* Copy the XID from the skb... */
1375 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1376 if (xp == NULL)
1377 return;
1378
1379 /* Look up and lock the request corresponding to the given XID */
1380 spin_lock(&xprt->queue_lock);
1381 rovr = xprt_lookup_rqst(xprt, *xp);
1382 if (!rovr)
1383 goto out_unlock;
1384 xprt_pin_rqst(rovr);
1385 xprt_update_rtt(rovr->rq_task);
1386 spin_unlock(&xprt->queue_lock);
1387 task = rovr->rq_task;
1388
1389 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1390 copied = repsize;
1391
1392 /* Suck it into the iovec, verify checksum if not done by hw. */
1393 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1394 spin_lock(&xprt->queue_lock);
1395 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1396 goto out_unpin;
1397 }
1398
1399
1400 spin_lock(&xprt->transport_lock);
1401 xprt_adjust_cwnd(xprt, task, copied);
1402 spin_unlock(&xprt->transport_lock);
1403 spin_lock(&xprt->queue_lock);
1404 xprt_complete_rqst(task, copied);
1405 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1406out_unpin:
1407 xprt_unpin_rqst(rovr);
1408 out_unlock:
1409 spin_unlock(&xprt->queue_lock);
1410}
1411
1412static void xs_udp_data_receive(struct sock_xprt *transport)
1413{
1414 struct sk_buff *skb;
1415 struct sock *sk;
1416 int err;
1417
1418 mutex_lock(&transport->recv_mutex);
1419 sk = transport->inet;
1420 if (sk == NULL)
1421 goto out;
1422 for (;;) {
1423 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err);
1424 if (skb == NULL)
1425 break;
1426 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1427 consume_skb(skb);
1428 cond_resched();
1429 }
1430 xs_poll_check_readable(transport);
1431out:
1432 mutex_unlock(&transport->recv_mutex);
1433}
1434
1435static void xs_udp_data_receive_workfn(struct work_struct *work)
1436{
1437 struct sock_xprt *transport =
1438 container_of(work, struct sock_xprt, recv_worker);
1439 unsigned int pflags = memalloc_nofs_save();
1440
1441 xs_udp_data_receive(transport);
1442 memalloc_nofs_restore(pflags);
1443}
1444
1445/**
1446 * xs_data_ready - "data ready" callback for sockets
1447 * @sk: socket with data to read
1448 *
1449 */
1450static void xs_data_ready(struct sock *sk)
1451{
1452 struct rpc_xprt *xprt;
1453
1454 trace_sk_data_ready(sk);
1455
1456 xprt = xprt_from_sock(sk);
1457 if (xprt != NULL) {
1458 struct sock_xprt *transport = container_of(xprt,
1459 struct sock_xprt, xprt);
1460
1461 trace_xs_data_ready(xprt);
1462
1463 transport->old_data_ready(sk);
1464
1465 if (test_bit(XPRT_SOCK_IGNORE_RECV, &transport->sock_state))
1466 return;
1467
1468 /* Any data means we had a useful conversation, so
1469 * then we don't need to delay the next reconnect
1470 */
1471 if (xprt->reestablish_timeout)
1472 xprt->reestablish_timeout = 0;
1473 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1474 queue_work(xprtiod_workqueue, &transport->recv_worker);
1475 }
1476}
1477
1478/*
1479 * Helper function to force a TCP close if the server is sending
1480 * junk and/or it has put us in CLOSE_WAIT
1481 */
1482static void xs_tcp_force_close(struct rpc_xprt *xprt)
1483{
1484 xprt_force_disconnect(xprt);
1485}
1486
1487#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1488static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1489{
1490 return PAGE_SIZE;
1491}
1492#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1493
1494/**
1495 * xs_local_state_change - callback to handle AF_LOCAL socket state changes
1496 * @sk: socket whose state has changed
1497 *
1498 */
1499static void xs_local_state_change(struct sock *sk)
1500{
1501 struct rpc_xprt *xprt;
1502 struct sock_xprt *transport;
1503
1504 if (!(xprt = xprt_from_sock(sk)))
1505 return;
1506 transport = container_of(xprt, struct sock_xprt, xprt);
1507 if (sk->sk_shutdown & SHUTDOWN_MASK) {
1508 clear_bit(XPRT_CONNECTED, &xprt->state);
1509 /* Trigger the socket release */
1510 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1511 }
1512}
1513
1514/**
1515 * xs_tcp_state_change - callback to handle TCP socket state changes
1516 * @sk: socket whose state has changed
1517 *
1518 */
1519static void xs_tcp_state_change(struct sock *sk)
1520{
1521 struct rpc_xprt *xprt;
1522 struct sock_xprt *transport;
1523
1524 if (!(xprt = xprt_from_sock(sk)))
1525 return;
1526 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1527 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1528 sk->sk_state, xprt_connected(xprt),
1529 sock_flag(sk, SOCK_DEAD),
1530 sock_flag(sk, SOCK_ZAPPED),
1531 sk->sk_shutdown);
1532
1533 transport = container_of(xprt, struct sock_xprt, xprt);
1534 trace_rpc_socket_state_change(xprt, sk->sk_socket);
1535 switch (sk->sk_state) {
1536 case TCP_ESTABLISHED:
1537 if (!xprt_test_and_set_connected(xprt)) {
1538 xprt->connect_cookie++;
1539 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1540 xprt_clear_connecting(xprt);
1541
1542 xprt->stat.connect_count++;
1543 xprt->stat.connect_time += (long)jiffies -
1544 xprt->stat.connect_start;
1545 xs_run_error_worker(transport, XPRT_SOCK_WAKE_PENDING);
1546 }
1547 break;
1548 case TCP_FIN_WAIT1:
1549 /* The client initiated a shutdown of the socket */
1550 xprt->connect_cookie++;
1551 xprt->reestablish_timeout = 0;
1552 set_bit(XPRT_CLOSING, &xprt->state);
1553 smp_mb__before_atomic();
1554 clear_bit(XPRT_CONNECTED, &xprt->state);
1555 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1556 smp_mb__after_atomic();
1557 break;
1558 case TCP_CLOSE_WAIT:
1559 /* The server initiated a shutdown of the socket */
1560 xprt->connect_cookie++;
1561 clear_bit(XPRT_CONNECTED, &xprt->state);
1562 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1563 fallthrough;
1564 case TCP_CLOSING:
1565 /*
1566 * If the server closed down the connection, make sure that
1567 * we back off before reconnecting
1568 */
1569 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1570 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1571 break;
1572 case TCP_LAST_ACK:
1573 set_bit(XPRT_CLOSING, &xprt->state);
1574 smp_mb__before_atomic();
1575 clear_bit(XPRT_CONNECTED, &xprt->state);
1576 smp_mb__after_atomic();
1577 break;
1578 case TCP_CLOSE:
1579 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1580 &transport->sock_state)) {
1581 xs_reset_srcport(transport);
1582 xprt_clear_connecting(xprt);
1583 }
1584 clear_bit(XPRT_CLOSING, &xprt->state);
1585 /* Trigger the socket release */
1586 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1587 }
1588}
1589
1590static void xs_write_space(struct sock *sk)
1591{
1592 struct sock_xprt *transport;
1593 struct rpc_xprt *xprt;
1594
1595 if (!sk->sk_socket)
1596 return;
1597 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1598
1599 if (unlikely(!(xprt = xprt_from_sock(sk))))
1600 return;
1601 transport = container_of(xprt, struct sock_xprt, xprt);
1602 if (!test_and_clear_bit(XPRT_SOCK_NOSPACE, &transport->sock_state))
1603 return;
1604 xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE);
1605 sk->sk_write_pending--;
1606}
1607
1608/**
1609 * xs_udp_write_space - callback invoked when socket buffer space
1610 * becomes available
1611 * @sk: socket whose state has changed
1612 *
1613 * Called when more output buffer space is available for this socket.
1614 * We try not to wake our writers until they can make "significant"
1615 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1616 * with a bunch of small requests.
1617 */
1618static void xs_udp_write_space(struct sock *sk)
1619{
1620 /* from net/core/sock.c:sock_def_write_space */
1621 if (sock_writeable(sk))
1622 xs_write_space(sk);
1623}
1624
1625/**
1626 * xs_tcp_write_space - callback invoked when socket buffer space
1627 * becomes available
1628 * @sk: socket whose state has changed
1629 *
1630 * Called when more output buffer space is available for this socket.
1631 * We try not to wake our writers until they can make "significant"
1632 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1633 * with a bunch of small requests.
1634 */
1635static void xs_tcp_write_space(struct sock *sk)
1636{
1637 /* from net/core/stream.c:sk_stream_write_space */
1638 if (sk_stream_is_writeable(sk))
1639 xs_write_space(sk);
1640}
1641
1642static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1643{
1644 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1645 struct sock *sk = transport->inet;
1646
1647 if (transport->rcvsize) {
1648 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1649 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1650 }
1651 if (transport->sndsize) {
1652 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1653 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1654 sk->sk_write_space(sk);
1655 }
1656}
1657
1658/**
1659 * xs_udp_set_buffer_size - set send and receive limits
1660 * @xprt: generic transport
1661 * @sndsize: requested size of send buffer, in bytes
1662 * @rcvsize: requested size of receive buffer, in bytes
1663 *
1664 * Set socket send and receive buffer size limits.
1665 */
1666static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1667{
1668 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1669
1670 transport->sndsize = 0;
1671 if (sndsize)
1672 transport->sndsize = sndsize + 1024;
1673 transport->rcvsize = 0;
1674 if (rcvsize)
1675 transport->rcvsize = rcvsize + 1024;
1676
1677 xs_udp_do_set_buffer_size(xprt);
1678}
1679
1680/**
1681 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1682 * @xprt: controlling transport
1683 * @task: task that timed out
1684 *
1685 * Adjust the congestion window after a retransmit timeout has occurred.
1686 */
1687static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1688{
1689 spin_lock(&xprt->transport_lock);
1690 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1691 spin_unlock(&xprt->transport_lock);
1692}
1693
1694static int xs_get_random_port(void)
1695{
1696 unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1697 unsigned short range;
1698 unsigned short rand;
1699
1700 if (max < min)
1701 return -EADDRINUSE;
1702 range = max - min + 1;
1703 rand = get_random_u32_below(range);
1704 return rand + min;
1705}
1706
1707static unsigned short xs_sock_getport(struct socket *sock)
1708{
1709 struct sockaddr_storage buf;
1710 unsigned short port = 0;
1711
1712 if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1713 goto out;
1714 switch (buf.ss_family) {
1715 case AF_INET6:
1716 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1717 break;
1718 case AF_INET:
1719 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1720 }
1721out:
1722 return port;
1723}
1724
1725/**
1726 * xs_set_port - reset the port number in the remote endpoint address
1727 * @xprt: generic transport
1728 * @port: new port number
1729 *
1730 */
1731static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1732{
1733 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1734
1735 rpc_set_port(xs_addr(xprt), port);
1736 xs_update_peer_port(xprt);
1737}
1738
1739static void xs_reset_srcport(struct sock_xprt *transport)
1740{
1741 transport->srcport = 0;
1742}
1743
1744static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1745{
1746 if (transport->srcport == 0 && transport->xprt.reuseport)
1747 transport->srcport = xs_sock_getport(sock);
1748}
1749
1750static int xs_get_srcport(struct sock_xprt *transport)
1751{
1752 int port = transport->srcport;
1753
1754 if (port == 0 && transport->xprt.resvport)
1755 port = xs_get_random_port();
1756 return port;
1757}
1758
1759static unsigned short xs_sock_srcport(struct rpc_xprt *xprt)
1760{
1761 struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt);
1762 unsigned short ret = 0;
1763 mutex_lock(&sock->recv_mutex);
1764 if (sock->sock)
1765 ret = xs_sock_getport(sock->sock);
1766 mutex_unlock(&sock->recv_mutex);
1767 return ret;
1768}
1769
1770static int xs_sock_srcaddr(struct rpc_xprt *xprt, char *buf, size_t buflen)
1771{
1772 struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt);
1773 union {
1774 struct sockaddr sa;
1775 struct sockaddr_storage st;
1776 } saddr;
1777 int ret = -ENOTCONN;
1778
1779 mutex_lock(&sock->recv_mutex);
1780 if (sock->sock) {
1781 ret = kernel_getsockname(sock->sock, &saddr.sa);
1782 if (ret >= 0)
1783 ret = snprintf(buf, buflen, "%pISc", &saddr.sa);
1784 }
1785 mutex_unlock(&sock->recv_mutex);
1786 return ret;
1787}
1788
1789static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1790{
1791 if (transport->srcport != 0)
1792 transport->srcport = 0;
1793 if (!transport->xprt.resvport)
1794 return 0;
1795 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1796 return xprt_max_resvport;
1797 return --port;
1798}
1799static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1800{
1801 struct sockaddr_storage myaddr;
1802 int err, nloop = 0;
1803 int port = xs_get_srcport(transport);
1804 unsigned short last;
1805
1806 /*
1807 * If we are asking for any ephemeral port (i.e. port == 0 &&
1808 * transport->xprt.resvport == 0), don't bind. Let the local
1809 * port selection happen implicitly when the socket is used
1810 * (for example at connect time).
1811 *
1812 * This ensures that we can continue to establish TCP
1813 * connections even when all local ephemeral ports are already
1814 * a part of some TCP connection. This makes no difference
1815 * for UDP sockets, but also doesn't harm them.
1816 *
1817 * If we're asking for any reserved port (i.e. port == 0 &&
1818 * transport->xprt.resvport == 1) xs_get_srcport above will
1819 * ensure that port is non-zero and we will bind as needed.
1820 */
1821 if (port <= 0)
1822 return port;
1823
1824 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1825 do {
1826 rpc_set_port((struct sockaddr *)&myaddr, port);
1827 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1828 transport->xprt.addrlen);
1829 if (err == 0) {
1830 if (transport->xprt.reuseport)
1831 transport->srcport = port;
1832 break;
1833 }
1834 last = port;
1835 port = xs_next_srcport(transport, port);
1836 if (port > last)
1837 nloop++;
1838 } while (err == -EADDRINUSE && nloop != 2);
1839
1840 if (myaddr.ss_family == AF_INET)
1841 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1842 &((struct sockaddr_in *)&myaddr)->sin_addr,
1843 port, err ? "failed" : "ok", err);
1844 else
1845 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1846 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1847 port, err ? "failed" : "ok", err);
1848 return err;
1849}
1850
1851/*
1852 * We don't support autobind on AF_LOCAL sockets
1853 */
1854static void xs_local_rpcbind(struct rpc_task *task)
1855{
1856 xprt_set_bound(task->tk_xprt);
1857}
1858
1859static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1860{
1861}
1862
1863#ifdef CONFIG_DEBUG_LOCK_ALLOC
1864static struct lock_class_key xs_key[3];
1865static struct lock_class_key xs_slock_key[3];
1866
1867static inline void xs_reclassify_socketu(struct socket *sock)
1868{
1869 struct sock *sk = sock->sk;
1870
1871 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1872 &xs_slock_key[0], "sk_lock-AF_LOCAL-RPC", &xs_key[0]);
1873}
1874
1875static inline void xs_reclassify_socket4(struct socket *sock)
1876{
1877 struct sock *sk = sock->sk;
1878
1879 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1880 &xs_slock_key[1], "sk_lock-AF_INET-RPC", &xs_key[1]);
1881}
1882
1883static inline void xs_reclassify_socket6(struct socket *sock)
1884{
1885 struct sock *sk = sock->sk;
1886
1887 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1888 &xs_slock_key[2], "sk_lock-AF_INET6-RPC", &xs_key[2]);
1889}
1890
1891static inline void xs_reclassify_socket(int family, struct socket *sock)
1892{
1893 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1894 return;
1895
1896 switch (family) {
1897 case AF_LOCAL:
1898 xs_reclassify_socketu(sock);
1899 break;
1900 case AF_INET:
1901 xs_reclassify_socket4(sock);
1902 break;
1903 case AF_INET6:
1904 xs_reclassify_socket6(sock);
1905 break;
1906 }
1907}
1908#else
1909static inline void xs_reclassify_socket(int family, struct socket *sock)
1910{
1911}
1912#endif
1913
1914static void xs_dummy_setup_socket(struct work_struct *work)
1915{
1916}
1917
1918static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1919 struct sock_xprt *transport, int family, int type,
1920 int protocol, bool reuseport)
1921{
1922 struct file *filp;
1923 struct socket *sock;
1924 int err;
1925
1926 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1927 if (err < 0) {
1928 dprintk("RPC: can't create %d transport socket (%d).\n",
1929 protocol, -err);
1930 goto out;
1931 }
1932 xs_reclassify_socket(family, sock);
1933
1934 if (reuseport)
1935 sock_set_reuseport(sock->sk);
1936
1937 err = xs_bind(transport, sock);
1938 if (err) {
1939 sock_release(sock);
1940 goto out;
1941 }
1942
1943 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1944 if (IS_ERR(filp))
1945 return ERR_CAST(filp);
1946 transport->file = filp;
1947
1948 return sock;
1949out:
1950 return ERR_PTR(err);
1951}
1952
1953static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1954 struct socket *sock)
1955{
1956 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1957 xprt);
1958
1959 if (!transport->inet) {
1960 struct sock *sk = sock->sk;
1961
1962 lock_sock(sk);
1963
1964 xs_save_old_callbacks(transport, sk);
1965
1966 sk->sk_user_data = xprt;
1967 sk->sk_data_ready = xs_data_ready;
1968 sk->sk_write_space = xs_udp_write_space;
1969 sk->sk_state_change = xs_local_state_change;
1970 sk->sk_error_report = xs_error_report;
1971 sk->sk_use_task_frag = false;
1972
1973 xprt_clear_connected(xprt);
1974
1975 /* Reset to new socket */
1976 transport->sock = sock;
1977 transport->inet = sk;
1978
1979 release_sock(sk);
1980 }
1981
1982 xs_stream_start_connect(transport);
1983
1984 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1985}
1986
1987/**
1988 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1989 * @transport: socket transport to connect
1990 */
1991static int xs_local_setup_socket(struct sock_xprt *transport)
1992{
1993 struct rpc_xprt *xprt = &transport->xprt;
1994 struct file *filp;
1995 struct socket *sock;
1996 int status;
1997
1998 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1999 SOCK_STREAM, 0, &sock, 1);
2000 if (status < 0) {
2001 dprintk("RPC: can't create AF_LOCAL "
2002 "transport socket (%d).\n", -status);
2003 goto out;
2004 }
2005 xs_reclassify_socket(AF_LOCAL, sock);
2006
2007 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
2008 if (IS_ERR(filp)) {
2009 status = PTR_ERR(filp);
2010 goto out;
2011 }
2012 transport->file = filp;
2013
2014 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
2015 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2016
2017 status = xs_local_finish_connecting(xprt, sock);
2018 trace_rpc_socket_connect(xprt, sock, status);
2019 switch (status) {
2020 case 0:
2021 dprintk("RPC: xprt %p connected to %s\n",
2022 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2023 xprt->stat.connect_count++;
2024 xprt->stat.connect_time += (long)jiffies -
2025 xprt->stat.connect_start;
2026 xprt_set_connected(xprt);
2027 break;
2028 case -ENOBUFS:
2029 break;
2030 case -ENOENT:
2031 dprintk("RPC: xprt %p: socket %s does not exist\n",
2032 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2033 break;
2034 case -ECONNREFUSED:
2035 dprintk("RPC: xprt %p: connection refused for %s\n",
2036 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2037 break;
2038 default:
2039 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2040 __func__, -status,
2041 xprt->address_strings[RPC_DISPLAY_ADDR]);
2042 }
2043
2044out:
2045 xprt_clear_connecting(xprt);
2046 xprt_wake_pending_tasks(xprt, status);
2047 return status;
2048}
2049
2050static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2051{
2052 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2053 int ret;
2054
2055 if (transport->file)
2056 goto force_disconnect;
2057
2058 if (RPC_IS_ASYNC(task)) {
2059 /*
2060 * We want the AF_LOCAL connect to be resolved in the
2061 * filesystem namespace of the process making the rpc
2062 * call. Thus we connect synchronously.
2063 *
2064 * If we want to support asynchronous AF_LOCAL calls,
2065 * we'll need to figure out how to pass a namespace to
2066 * connect.
2067 */
2068 rpc_task_set_rpc_status(task, -ENOTCONN);
2069 goto out_wake;
2070 }
2071 ret = xs_local_setup_socket(transport);
2072 if (ret && !RPC_IS_SOFTCONN(task))
2073 msleep_interruptible(15000);
2074 return;
2075force_disconnect:
2076 xprt_force_disconnect(xprt);
2077out_wake:
2078 xprt_clear_connecting(xprt);
2079 xprt_wake_pending_tasks(xprt, -ENOTCONN);
2080}
2081
2082#if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2083/*
2084 * Note that this should be called with XPRT_LOCKED held, or recv_mutex
2085 * held, or when we otherwise know that we have exclusive access to the
2086 * socket, to guard against races with xs_reset_transport.
2087 */
2088static void xs_set_memalloc(struct rpc_xprt *xprt)
2089{
2090 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2091 xprt);
2092
2093 /*
2094 * If there's no sock, then we have nothing to set. The
2095 * reconnecting process will get it for us.
2096 */
2097 if (!transport->inet)
2098 return;
2099 if (atomic_read(&xprt->swapper))
2100 sk_set_memalloc(transport->inet);
2101}
2102
2103/**
2104 * xs_enable_swap - Tag this transport as being used for swap.
2105 * @xprt: transport to tag
2106 *
2107 * Take a reference to this transport on behalf of the rpc_clnt, and
2108 * optionally mark it for swapping if it wasn't already.
2109 */
2110static int
2111xs_enable_swap(struct rpc_xprt *xprt)
2112{
2113 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2114
2115 mutex_lock(&xs->recv_mutex);
2116 if (atomic_inc_return(&xprt->swapper) == 1 &&
2117 xs->inet)
2118 sk_set_memalloc(xs->inet);
2119 mutex_unlock(&xs->recv_mutex);
2120 return 0;
2121}
2122
2123/**
2124 * xs_disable_swap - Untag this transport as being used for swap.
2125 * @xprt: transport to tag
2126 *
2127 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2128 * swapper refcount goes to 0, untag the socket as a memalloc socket.
2129 */
2130static void
2131xs_disable_swap(struct rpc_xprt *xprt)
2132{
2133 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2134
2135 mutex_lock(&xs->recv_mutex);
2136 if (atomic_dec_and_test(&xprt->swapper) &&
2137 xs->inet)
2138 sk_clear_memalloc(xs->inet);
2139 mutex_unlock(&xs->recv_mutex);
2140}
2141#else
2142static void xs_set_memalloc(struct rpc_xprt *xprt)
2143{
2144}
2145
2146static int
2147xs_enable_swap(struct rpc_xprt *xprt)
2148{
2149 return -EINVAL;
2150}
2151
2152static void
2153xs_disable_swap(struct rpc_xprt *xprt)
2154{
2155}
2156#endif
2157
2158static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2159{
2160 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2161
2162 if (!transport->inet) {
2163 struct sock *sk = sock->sk;
2164
2165 lock_sock(sk);
2166
2167 xs_save_old_callbacks(transport, sk);
2168
2169 sk->sk_user_data = xprt;
2170 sk->sk_data_ready = xs_data_ready;
2171 sk->sk_write_space = xs_udp_write_space;
2172 sk->sk_use_task_frag = false;
2173
2174 xprt_set_connected(xprt);
2175
2176 /* Reset to new socket */
2177 transport->sock = sock;
2178 transport->inet = sk;
2179
2180 xs_set_memalloc(xprt);
2181
2182 release_sock(sk);
2183 }
2184 xs_udp_do_set_buffer_size(xprt);
2185
2186 xprt->stat.connect_start = jiffies;
2187}
2188
2189static void xs_udp_setup_socket(struct work_struct *work)
2190{
2191 struct sock_xprt *transport =
2192 container_of(work, struct sock_xprt, connect_worker.work);
2193 struct rpc_xprt *xprt = &transport->xprt;
2194 struct socket *sock;
2195 int status = -EIO;
2196 unsigned int pflags = current->flags;
2197
2198 if (atomic_read(&xprt->swapper))
2199 current->flags |= PF_MEMALLOC;
2200 sock = xs_create_sock(xprt, transport,
2201 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2202 IPPROTO_UDP, false);
2203 if (IS_ERR(sock))
2204 goto out;
2205
2206 dprintk("RPC: worker connecting xprt %p via %s to "
2207 "%s (port %s)\n", xprt,
2208 xprt->address_strings[RPC_DISPLAY_PROTO],
2209 xprt->address_strings[RPC_DISPLAY_ADDR],
2210 xprt->address_strings[RPC_DISPLAY_PORT]);
2211
2212 xs_udp_finish_connecting(xprt, sock);
2213 trace_rpc_socket_connect(xprt, sock, 0);
2214 status = 0;
2215out:
2216 xprt_clear_connecting(xprt);
2217 xprt_unlock_connect(xprt, transport);
2218 xprt_wake_pending_tasks(xprt, status);
2219 current_restore_flags(pflags, PF_MEMALLOC);
2220}
2221
2222/**
2223 * xs_tcp_shutdown - gracefully shut down a TCP socket
2224 * @xprt: transport
2225 *
2226 * Initiates a graceful shutdown of the TCP socket by calling the
2227 * equivalent of shutdown(SHUT_RDWR);
2228 */
2229static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2230{
2231 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2232 struct socket *sock = transport->sock;
2233 int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2234
2235 if (sock == NULL)
2236 return;
2237 if (!xprt->reuseport) {
2238 xs_close(xprt);
2239 return;
2240 }
2241 switch (skst) {
2242 case TCP_FIN_WAIT1:
2243 case TCP_FIN_WAIT2:
2244 case TCP_LAST_ACK:
2245 break;
2246 case TCP_ESTABLISHED:
2247 case TCP_CLOSE_WAIT:
2248 kernel_sock_shutdown(sock, SHUT_RDWR);
2249 trace_rpc_socket_shutdown(xprt, sock);
2250 break;
2251 default:
2252 xs_reset_transport(transport);
2253 }
2254}
2255
2256static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2257 struct socket *sock)
2258{
2259 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2260 struct net *net = sock_net(sock->sk);
2261 unsigned long connect_timeout;
2262 unsigned long syn_retries;
2263 unsigned int keepidle;
2264 unsigned int keepcnt;
2265 unsigned int timeo;
2266 unsigned long t;
2267
2268 spin_lock(&xprt->transport_lock);
2269 keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2270 keepcnt = xprt->timeout->to_retries + 1;
2271 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2272 (xprt->timeout->to_retries + 1);
2273 clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2274 spin_unlock(&xprt->transport_lock);
2275
2276 /* TCP Keepalive options */
2277 sock_set_keepalive(sock->sk);
2278 tcp_sock_set_keepidle(sock->sk, keepidle);
2279 tcp_sock_set_keepintvl(sock->sk, keepidle);
2280 tcp_sock_set_keepcnt(sock->sk, keepcnt);
2281
2282 /* TCP user timeout (see RFC5482) */
2283 tcp_sock_set_user_timeout(sock->sk, timeo);
2284
2285 /* Connect timeout */
2286 connect_timeout = max_t(unsigned long,
2287 DIV_ROUND_UP(xprt->connect_timeout, HZ), 1);
2288 syn_retries = max_t(unsigned long,
2289 READ_ONCE(net->ipv4.sysctl_tcp_syn_retries), 1);
2290 for (t = 0; t <= syn_retries && (1UL << t) < connect_timeout; t++)
2291 ;
2292 if (t <= syn_retries)
2293 tcp_sock_set_syncnt(sock->sk, t - 1);
2294}
2295
2296static void xs_tcp_do_set_connect_timeout(struct rpc_xprt *xprt,
2297 unsigned long connect_timeout)
2298{
2299 struct sock_xprt *transport =
2300 container_of(xprt, struct sock_xprt, xprt);
2301 struct rpc_timeout to;
2302 unsigned long initval;
2303
2304 memcpy(&to, xprt->timeout, sizeof(to));
2305 /* Arbitrary lower limit */
2306 initval = max_t(unsigned long, connect_timeout, XS_TCP_INIT_REEST_TO);
2307 to.to_initval = initval;
2308 to.to_maxval = initval;
2309 to.to_retries = 0;
2310 memcpy(&transport->tcp_timeout, &to, sizeof(transport->tcp_timeout));
2311 xprt->timeout = &transport->tcp_timeout;
2312 xprt->connect_timeout = connect_timeout;
2313}
2314
2315static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2316 unsigned long connect_timeout,
2317 unsigned long reconnect_timeout)
2318{
2319 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2320
2321 spin_lock(&xprt->transport_lock);
2322 if (reconnect_timeout < xprt->max_reconnect_timeout)
2323 xprt->max_reconnect_timeout = reconnect_timeout;
2324 if (connect_timeout < xprt->connect_timeout)
2325 xs_tcp_do_set_connect_timeout(xprt, connect_timeout);
2326 set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2327 spin_unlock(&xprt->transport_lock);
2328}
2329
2330static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2331{
2332 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2333
2334 if (!transport->inet) {
2335 struct sock *sk = sock->sk;
2336
2337 /* Avoid temporary address, they are bad for long-lived
2338 * connections such as NFS mounts.
2339 * RFC4941, section 3.6 suggests that:
2340 * Individual applications, which have specific
2341 * knowledge about the normal duration of connections,
2342 * MAY override this as appropriate.
2343 */
2344 if (xs_addr(xprt)->sa_family == PF_INET6) {
2345 ip6_sock_set_addr_preferences(sk,
2346 IPV6_PREFER_SRC_PUBLIC);
2347 }
2348
2349 xs_tcp_set_socket_timeouts(xprt, sock);
2350 tcp_sock_set_nodelay(sk);
2351
2352 lock_sock(sk);
2353
2354 xs_save_old_callbacks(transport, sk);
2355
2356 sk->sk_user_data = xprt;
2357 sk->sk_data_ready = xs_data_ready;
2358 sk->sk_state_change = xs_tcp_state_change;
2359 sk->sk_write_space = xs_tcp_write_space;
2360 sk->sk_error_report = xs_error_report;
2361 sk->sk_use_task_frag = false;
2362
2363 /* socket options */
2364 sock_reset_flag(sk, SOCK_LINGER);
2365
2366 xprt_clear_connected(xprt);
2367
2368 /* Reset to new socket */
2369 transport->sock = sock;
2370 transport->inet = sk;
2371
2372 release_sock(sk);
2373 }
2374
2375 if (!xprt_bound(xprt))
2376 return -ENOTCONN;
2377
2378 xs_set_memalloc(xprt);
2379
2380 xs_stream_start_connect(transport);
2381
2382 /* Tell the socket layer to start connecting... */
2383 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2384 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2385}
2386
2387/**
2388 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2389 * @work: queued work item
2390 *
2391 * Invoked by a work queue tasklet.
2392 */
2393static void xs_tcp_setup_socket(struct work_struct *work)
2394{
2395 struct sock_xprt *transport =
2396 container_of(work, struct sock_xprt, connect_worker.work);
2397 struct socket *sock = transport->sock;
2398 struct rpc_xprt *xprt = &transport->xprt;
2399 int status;
2400 unsigned int pflags = current->flags;
2401
2402 if (atomic_read(&xprt->swapper))
2403 current->flags |= PF_MEMALLOC;
2404
2405 if (xprt_connected(xprt))
2406 goto out;
2407 if (test_and_clear_bit(XPRT_SOCK_CONNECT_SENT,
2408 &transport->sock_state) ||
2409 !sock) {
2410 xs_reset_transport(transport);
2411 sock = xs_create_sock(xprt, transport, xs_addr(xprt)->sa_family,
2412 SOCK_STREAM, IPPROTO_TCP, true);
2413 if (IS_ERR(sock)) {
2414 xprt_wake_pending_tasks(xprt, PTR_ERR(sock));
2415 goto out;
2416 }
2417 }
2418
2419 dprintk("RPC: worker connecting xprt %p via %s to "
2420 "%s (port %s)\n", xprt,
2421 xprt->address_strings[RPC_DISPLAY_PROTO],
2422 xprt->address_strings[RPC_DISPLAY_ADDR],
2423 xprt->address_strings[RPC_DISPLAY_PORT]);
2424
2425 status = xs_tcp_finish_connecting(xprt, sock);
2426 trace_rpc_socket_connect(xprt, sock, status);
2427 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2428 xprt, -status, xprt_connected(xprt),
2429 sock->sk->sk_state);
2430 switch (status) {
2431 case 0:
2432 case -EINPROGRESS:
2433 /* SYN_SENT! */
2434 set_bit(XPRT_SOCK_CONNECT_SENT, &transport->sock_state);
2435 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2436 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2437 fallthrough;
2438 case -EALREADY:
2439 goto out_unlock;
2440 case -EADDRNOTAVAIL:
2441 /* Source port number is unavailable. Try a new one! */
2442 transport->srcport = 0;
2443 status = -EAGAIN;
2444 break;
2445 case -EINVAL:
2446 /* Happens, for instance, if the user specified a link
2447 * local IPv6 address without a scope-id.
2448 */
2449 case -ECONNREFUSED:
2450 case -ECONNRESET:
2451 case -ENETDOWN:
2452 case -ENETUNREACH:
2453 case -EHOSTUNREACH:
2454 case -EADDRINUSE:
2455 case -ENOBUFS:
2456 break;
2457 default:
2458 printk("%s: connect returned unhandled error %d\n",
2459 __func__, status);
2460 status = -EAGAIN;
2461 }
2462
2463 /* xs_tcp_force_close() wakes tasks with a fixed error code.
2464 * We need to wake them first to ensure the correct error code.
2465 */
2466 xprt_wake_pending_tasks(xprt, status);
2467 xs_tcp_force_close(xprt);
2468out:
2469 xprt_clear_connecting(xprt);
2470out_unlock:
2471 xprt_unlock_connect(xprt, transport);
2472 current_restore_flags(pflags, PF_MEMALLOC);
2473}
2474
2475/*
2476 * Transfer the connected socket to @upper_transport, then mark that
2477 * xprt CONNECTED.
2478 */
2479static int xs_tcp_tls_finish_connecting(struct rpc_xprt *lower_xprt,
2480 struct sock_xprt *upper_transport)
2481{
2482 struct sock_xprt *lower_transport =
2483 container_of(lower_xprt, struct sock_xprt, xprt);
2484 struct rpc_xprt *upper_xprt = &upper_transport->xprt;
2485
2486 if (!upper_transport->inet) {
2487 struct socket *sock = lower_transport->sock;
2488 struct sock *sk = sock->sk;
2489
2490 /* Avoid temporary address, they are bad for long-lived
2491 * connections such as NFS mounts.
2492 * RFC4941, section 3.6 suggests that:
2493 * Individual applications, which have specific
2494 * knowledge about the normal duration of connections,
2495 * MAY override this as appropriate.
2496 */
2497 if (xs_addr(upper_xprt)->sa_family == PF_INET6)
2498 ip6_sock_set_addr_preferences(sk, IPV6_PREFER_SRC_PUBLIC);
2499
2500 xs_tcp_set_socket_timeouts(upper_xprt, sock);
2501 tcp_sock_set_nodelay(sk);
2502
2503 lock_sock(sk);
2504
2505 /* @sk is already connected, so it now has the RPC callbacks.
2506 * Reach into @lower_transport to save the original ones.
2507 */
2508 upper_transport->old_data_ready = lower_transport->old_data_ready;
2509 upper_transport->old_state_change = lower_transport->old_state_change;
2510 upper_transport->old_write_space = lower_transport->old_write_space;
2511 upper_transport->old_error_report = lower_transport->old_error_report;
2512 sk->sk_user_data = upper_xprt;
2513
2514 /* socket options */
2515 sock_reset_flag(sk, SOCK_LINGER);
2516
2517 xprt_clear_connected(upper_xprt);
2518
2519 upper_transport->sock = sock;
2520 upper_transport->inet = sk;
2521 upper_transport->file = lower_transport->file;
2522
2523 release_sock(sk);
2524
2525 /* Reset lower_transport before shutting down its clnt */
2526 mutex_lock(&lower_transport->recv_mutex);
2527 lower_transport->inet = NULL;
2528 lower_transport->sock = NULL;
2529 lower_transport->file = NULL;
2530
2531 xprt_clear_connected(lower_xprt);
2532 xs_sock_reset_connection_flags(lower_xprt);
2533 xs_stream_reset_connect(lower_transport);
2534 mutex_unlock(&lower_transport->recv_mutex);
2535 }
2536
2537 if (!xprt_bound(upper_xprt))
2538 return -ENOTCONN;
2539
2540 xs_set_memalloc(upper_xprt);
2541
2542 if (!xprt_test_and_set_connected(upper_xprt)) {
2543 upper_xprt->connect_cookie++;
2544 clear_bit(XPRT_SOCK_CONNECTING, &upper_transport->sock_state);
2545 xprt_clear_connecting(upper_xprt);
2546
2547 upper_xprt->stat.connect_count++;
2548 upper_xprt->stat.connect_time += (long)jiffies -
2549 upper_xprt->stat.connect_start;
2550 xs_run_error_worker(upper_transport, XPRT_SOCK_WAKE_PENDING);
2551 }
2552 return 0;
2553}
2554
2555/**
2556 * xs_tls_handshake_done - TLS handshake completion handler
2557 * @data: address of xprt to wake
2558 * @status: status of handshake
2559 * @peerid: serial number of key containing the remote's identity
2560 *
2561 */
2562static void xs_tls_handshake_done(void *data, int status, key_serial_t peerid)
2563{
2564 struct rpc_xprt *lower_xprt = data;
2565 struct sock_xprt *lower_transport =
2566 container_of(lower_xprt, struct sock_xprt, xprt);
2567
2568 lower_transport->xprt_err = status ? -EACCES : 0;
2569 complete(&lower_transport->handshake_done);
2570 xprt_put(lower_xprt);
2571}
2572
2573static int xs_tls_handshake_sync(struct rpc_xprt *lower_xprt, struct xprtsec_parms *xprtsec)
2574{
2575 struct sock_xprt *lower_transport =
2576 container_of(lower_xprt, struct sock_xprt, xprt);
2577 struct tls_handshake_args args = {
2578 .ta_sock = lower_transport->sock,
2579 .ta_done = xs_tls_handshake_done,
2580 .ta_data = xprt_get(lower_xprt),
2581 .ta_peername = lower_xprt->servername,
2582 };
2583 struct sock *sk = lower_transport->inet;
2584 int rc;
2585
2586 init_completion(&lower_transport->handshake_done);
2587 set_bit(XPRT_SOCK_IGNORE_RECV, &lower_transport->sock_state);
2588 lower_transport->xprt_err = -ETIMEDOUT;
2589 switch (xprtsec->policy) {
2590 case RPC_XPRTSEC_TLS_ANON:
2591 rc = tls_client_hello_anon(&args, GFP_KERNEL);
2592 if (rc)
2593 goto out_put_xprt;
2594 break;
2595 case RPC_XPRTSEC_TLS_X509:
2596 args.ta_my_cert = xprtsec->cert_serial;
2597 args.ta_my_privkey = xprtsec->privkey_serial;
2598 rc = tls_client_hello_x509(&args, GFP_KERNEL);
2599 if (rc)
2600 goto out_put_xprt;
2601 break;
2602 default:
2603 rc = -EACCES;
2604 goto out_put_xprt;
2605 }
2606
2607 rc = wait_for_completion_interruptible_timeout(&lower_transport->handshake_done,
2608 XS_TLS_HANDSHAKE_TO);
2609 if (rc <= 0) {
2610 if (!tls_handshake_cancel(sk)) {
2611 if (rc == 0)
2612 rc = -ETIMEDOUT;
2613 goto out_put_xprt;
2614 }
2615 }
2616
2617 rc = lower_transport->xprt_err;
2618
2619out:
2620 xs_stream_reset_connect(lower_transport);
2621 clear_bit(XPRT_SOCK_IGNORE_RECV, &lower_transport->sock_state);
2622 return rc;
2623
2624out_put_xprt:
2625 xprt_put(lower_xprt);
2626 goto out;
2627}
2628
2629/**
2630 * xs_tcp_tls_setup_socket - establish a TLS session on a TCP socket
2631 * @work: queued work item
2632 *
2633 * Invoked by a work queue tasklet.
2634 *
2635 * For RPC-with-TLS, there is a two-stage connection process.
2636 *
2637 * The "upper-layer xprt" is visible to the RPC consumer. Once it has
2638 * been marked connected, the consumer knows that a TCP connection and
2639 * a TLS session have been established.
2640 *
2641 * A "lower-layer xprt", created in this function, handles the mechanics
2642 * of connecting the TCP socket, performing the RPC_AUTH_TLS probe, and
2643 * then driving the TLS handshake. Once all that is complete, the upper
2644 * layer xprt is marked connected.
2645 */
2646static void xs_tcp_tls_setup_socket(struct work_struct *work)
2647{
2648 struct sock_xprt *upper_transport =
2649 container_of(work, struct sock_xprt, connect_worker.work);
2650 struct rpc_clnt *upper_clnt = upper_transport->clnt;
2651 struct rpc_xprt *upper_xprt = &upper_transport->xprt;
2652 struct rpc_create_args args = {
2653 .net = upper_xprt->xprt_net,
2654 .protocol = upper_xprt->prot,
2655 .address = (struct sockaddr *)&upper_xprt->addr,
2656 .addrsize = upper_xprt->addrlen,
2657 .timeout = upper_clnt->cl_timeout,
2658 .servername = upper_xprt->servername,
2659 .program = upper_clnt->cl_program,
2660 .prognumber = upper_clnt->cl_prog,
2661 .version = upper_clnt->cl_vers,
2662 .authflavor = RPC_AUTH_TLS,
2663 .cred = upper_clnt->cl_cred,
2664 .xprtsec = {
2665 .policy = RPC_XPRTSEC_NONE,
2666 },
2667 .stats = upper_clnt->cl_stats,
2668 };
2669 unsigned int pflags = current->flags;
2670 struct rpc_clnt *lower_clnt;
2671 struct rpc_xprt *lower_xprt;
2672 int status;
2673
2674 if (atomic_read(&upper_xprt->swapper))
2675 current->flags |= PF_MEMALLOC;
2676
2677 xs_stream_start_connect(upper_transport);
2678
2679 /* This implicitly sends an RPC_AUTH_TLS probe */
2680 lower_clnt = rpc_create(&args);
2681 if (IS_ERR(lower_clnt)) {
2682 trace_rpc_tls_unavailable(upper_clnt, upper_xprt);
2683 clear_bit(XPRT_SOCK_CONNECTING, &upper_transport->sock_state);
2684 xprt_clear_connecting(upper_xprt);
2685 xprt_wake_pending_tasks(upper_xprt, PTR_ERR(lower_clnt));
2686 xs_run_error_worker(upper_transport, XPRT_SOCK_WAKE_PENDING);
2687 goto out_unlock;
2688 }
2689
2690 /* RPC_AUTH_TLS probe was successful. Try a TLS handshake on
2691 * the lower xprt.
2692 */
2693 rcu_read_lock();
2694 lower_xprt = rcu_dereference(lower_clnt->cl_xprt);
2695 rcu_read_unlock();
2696
2697 if (wait_on_bit_lock(&lower_xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2698 goto out_unlock;
2699
2700 status = xs_tls_handshake_sync(lower_xprt, &upper_xprt->xprtsec);
2701 if (status) {
2702 trace_rpc_tls_not_started(upper_clnt, upper_xprt);
2703 goto out_close;
2704 }
2705
2706 status = xs_tcp_tls_finish_connecting(lower_xprt, upper_transport);
2707 if (status)
2708 goto out_close;
2709 xprt_release_write(lower_xprt, NULL);
2710
2711 trace_rpc_socket_connect(upper_xprt, upper_transport->sock, 0);
2712 if (!xprt_test_and_set_connected(upper_xprt)) {
2713 upper_xprt->connect_cookie++;
2714 clear_bit(XPRT_SOCK_CONNECTING, &upper_transport->sock_state);
2715 xprt_clear_connecting(upper_xprt);
2716
2717 upper_xprt->stat.connect_count++;
2718 upper_xprt->stat.connect_time += (long)jiffies -
2719 upper_xprt->stat.connect_start;
2720 xs_run_error_worker(upper_transport, XPRT_SOCK_WAKE_PENDING);
2721 }
2722 rpc_shutdown_client(lower_clnt);
2723
2724out_unlock:
2725 current_restore_flags(pflags, PF_MEMALLOC);
2726 upper_transport->clnt = NULL;
2727 xprt_unlock_connect(upper_xprt, upper_transport);
2728 return;
2729
2730out_close:
2731 xprt_release_write(lower_xprt, NULL);
2732 rpc_shutdown_client(lower_clnt);
2733
2734 /* xprt_force_disconnect() wakes tasks with a fixed tk_status code.
2735 * Wake them first here to ensure they get our tk_status code.
2736 */
2737 xprt_wake_pending_tasks(upper_xprt, status);
2738 xs_tcp_force_close(upper_xprt);
2739 xprt_clear_connecting(upper_xprt);
2740 goto out_unlock;
2741}
2742
2743/**
2744 * xs_connect - connect a socket to a remote endpoint
2745 * @xprt: pointer to transport structure
2746 * @task: address of RPC task that manages state of connect request
2747 *
2748 * TCP: If the remote end dropped the connection, delay reconnecting.
2749 *
2750 * UDP socket connects are synchronous, but we use a work queue anyway
2751 * to guarantee that even unprivileged user processes can set up a
2752 * socket on a privileged port.
2753 *
2754 * If a UDP socket connect fails, the delay behavior here prevents
2755 * retry floods (hard mounts).
2756 */
2757static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2758{
2759 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2760 unsigned long delay = 0;
2761
2762 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2763
2764 if (transport->sock != NULL) {
2765 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2766 "seconds\n", xprt, xprt->reestablish_timeout / HZ);
2767
2768 delay = xprt_reconnect_delay(xprt);
2769 xprt_reconnect_backoff(xprt, XS_TCP_INIT_REEST_TO);
2770
2771 } else
2772 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2773
2774 transport->clnt = task->tk_client;
2775 queue_delayed_work(xprtiod_workqueue,
2776 &transport->connect_worker,
2777 delay);
2778}
2779
2780static void xs_wake_disconnect(struct sock_xprt *transport)
2781{
2782 if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state))
2783 xs_tcp_force_close(&transport->xprt);
2784}
2785
2786static void xs_wake_write(struct sock_xprt *transport)
2787{
2788 if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state))
2789 xprt_write_space(&transport->xprt);
2790}
2791
2792static void xs_wake_error(struct sock_xprt *transport)
2793{
2794 int sockerr;
2795
2796 if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2797 return;
2798 sockerr = xchg(&transport->xprt_err, 0);
2799 if (sockerr < 0) {
2800 xprt_wake_pending_tasks(&transport->xprt, sockerr);
2801 xs_tcp_force_close(&transport->xprt);
2802 }
2803}
2804
2805static void xs_wake_pending(struct sock_xprt *transport)
2806{
2807 if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING, &transport->sock_state))
2808 xprt_wake_pending_tasks(&transport->xprt, -EAGAIN);
2809}
2810
2811static void xs_error_handle(struct work_struct *work)
2812{
2813 struct sock_xprt *transport = container_of(work,
2814 struct sock_xprt, error_worker);
2815
2816 xs_wake_disconnect(transport);
2817 xs_wake_write(transport);
2818 xs_wake_error(transport);
2819 xs_wake_pending(transport);
2820}
2821
2822/**
2823 * xs_local_print_stats - display AF_LOCAL socket-specific stats
2824 * @xprt: rpc_xprt struct containing statistics
2825 * @seq: output file
2826 *
2827 */
2828static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2829{
2830 long idle_time = 0;
2831
2832 if (xprt_connected(xprt))
2833 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2834
2835 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2836 "%llu %llu %lu %llu %llu\n",
2837 xprt->stat.bind_count,
2838 xprt->stat.connect_count,
2839 xprt->stat.connect_time / HZ,
2840 idle_time,
2841 xprt->stat.sends,
2842 xprt->stat.recvs,
2843 xprt->stat.bad_xids,
2844 xprt->stat.req_u,
2845 xprt->stat.bklog_u,
2846 xprt->stat.max_slots,
2847 xprt->stat.sending_u,
2848 xprt->stat.pending_u);
2849}
2850
2851/**
2852 * xs_udp_print_stats - display UDP socket-specific stats
2853 * @xprt: rpc_xprt struct containing statistics
2854 * @seq: output file
2855 *
2856 */
2857static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2858{
2859 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2860
2861 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2862 "%lu %llu %llu\n",
2863 transport->srcport,
2864 xprt->stat.bind_count,
2865 xprt->stat.sends,
2866 xprt->stat.recvs,
2867 xprt->stat.bad_xids,
2868 xprt->stat.req_u,
2869 xprt->stat.bklog_u,
2870 xprt->stat.max_slots,
2871 xprt->stat.sending_u,
2872 xprt->stat.pending_u);
2873}
2874
2875/**
2876 * xs_tcp_print_stats - display TCP socket-specific stats
2877 * @xprt: rpc_xprt struct containing statistics
2878 * @seq: output file
2879 *
2880 */
2881static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2882{
2883 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2884 long idle_time = 0;
2885
2886 if (xprt_connected(xprt))
2887 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2888
2889 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2890 "%llu %llu %lu %llu %llu\n",
2891 transport->srcport,
2892 xprt->stat.bind_count,
2893 xprt->stat.connect_count,
2894 xprt->stat.connect_time / HZ,
2895 idle_time,
2896 xprt->stat.sends,
2897 xprt->stat.recvs,
2898 xprt->stat.bad_xids,
2899 xprt->stat.req_u,
2900 xprt->stat.bklog_u,
2901 xprt->stat.max_slots,
2902 xprt->stat.sending_u,
2903 xprt->stat.pending_u);
2904}
2905
2906/*
2907 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2908 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2909 * to use the server side send routines.
2910 */
2911static int bc_malloc(struct rpc_task *task)
2912{
2913 struct rpc_rqst *rqst = task->tk_rqstp;
2914 size_t size = rqst->rq_callsize;
2915 struct page *page;
2916 struct rpc_buffer *buf;
2917
2918 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2919 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2920 size);
2921 return -EINVAL;
2922 }
2923
2924 page = alloc_page(GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN);
2925 if (!page)
2926 return -ENOMEM;
2927
2928 buf = page_address(page);
2929 buf->len = PAGE_SIZE;
2930
2931 rqst->rq_buffer = buf->data;
2932 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2933 return 0;
2934}
2935
2936/*
2937 * Free the space allocated in the bc_alloc routine
2938 */
2939static void bc_free(struct rpc_task *task)
2940{
2941 void *buffer = task->tk_rqstp->rq_buffer;
2942 struct rpc_buffer *buf;
2943
2944 buf = container_of(buffer, struct rpc_buffer, data);
2945 free_page((unsigned long)buf);
2946}
2947
2948static int bc_sendto(struct rpc_rqst *req)
2949{
2950 struct xdr_buf *xdr = &req->rq_snd_buf;
2951 struct sock_xprt *transport =
2952 container_of(req->rq_xprt, struct sock_xprt, xprt);
2953 struct msghdr msg = {
2954 .msg_flags = 0,
2955 };
2956 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
2957 (u32)xdr->len);
2958 unsigned int sent = 0;
2959 int err;
2960
2961 req->rq_xtime = ktime_get();
2962 err = xdr_alloc_bvec(xdr, rpc_task_gfp_mask());
2963 if (err < 0)
2964 return err;
2965 err = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, marker, &sent);
2966 xdr_free_bvec(xdr);
2967 if (err < 0 || sent != (xdr->len + sizeof(marker)))
2968 return -EAGAIN;
2969 return sent;
2970}
2971
2972/**
2973 * bc_send_request - Send a backchannel Call on a TCP socket
2974 * @req: rpc_rqst containing Call message to be sent
2975 *
2976 * xpt_mutex ensures @rqstp's whole message is written to the socket
2977 * without interruption.
2978 *
2979 * Return values:
2980 * %0 if the message was sent successfully
2981 * %ENOTCONN if the message was not sent
2982 */
2983static int bc_send_request(struct rpc_rqst *req)
2984{
2985 struct svc_xprt *xprt;
2986 int len;
2987
2988 /*
2989 * Get the server socket associated with this callback xprt
2990 */
2991 xprt = req->rq_xprt->bc_xprt;
2992
2993 /*
2994 * Grab the mutex to serialize data as the connection is shared
2995 * with the fore channel
2996 */
2997 mutex_lock(&xprt->xpt_mutex);
2998 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2999 len = -ENOTCONN;
3000 else
3001 len = bc_sendto(req);
3002 mutex_unlock(&xprt->xpt_mutex);
3003
3004 if (len > 0)
3005 len = 0;
3006
3007 return len;
3008}
3009
3010static void bc_close(struct rpc_xprt *xprt)
3011{
3012 xprt_disconnect_done(xprt);
3013}
3014
3015static void bc_destroy(struct rpc_xprt *xprt)
3016{
3017 dprintk("RPC: bc_destroy xprt %p\n", xprt);
3018
3019 xs_xprt_free(xprt);
3020 module_put(THIS_MODULE);
3021}
3022
3023static const struct rpc_xprt_ops xs_local_ops = {
3024 .reserve_xprt = xprt_reserve_xprt,
3025 .release_xprt = xprt_release_xprt,
3026 .alloc_slot = xprt_alloc_slot,
3027 .free_slot = xprt_free_slot,
3028 .rpcbind = xs_local_rpcbind,
3029 .set_port = xs_local_set_port,
3030 .connect = xs_local_connect,
3031 .buf_alloc = rpc_malloc,
3032 .buf_free = rpc_free,
3033 .prepare_request = xs_stream_prepare_request,
3034 .send_request = xs_local_send_request,
3035 .abort_send_request = xs_stream_abort_send_request,
3036 .wait_for_reply_request = xprt_wait_for_reply_request_def,
3037 .close = xs_close,
3038 .destroy = xs_destroy,
3039 .print_stats = xs_local_print_stats,
3040 .enable_swap = xs_enable_swap,
3041 .disable_swap = xs_disable_swap,
3042};
3043
3044static const struct rpc_xprt_ops xs_udp_ops = {
3045 .set_buffer_size = xs_udp_set_buffer_size,
3046 .reserve_xprt = xprt_reserve_xprt_cong,
3047 .release_xprt = xprt_release_xprt_cong,
3048 .alloc_slot = xprt_alloc_slot,
3049 .free_slot = xprt_free_slot,
3050 .rpcbind = rpcb_getport_async,
3051 .set_port = xs_set_port,
3052 .connect = xs_connect,
3053 .get_srcaddr = xs_sock_srcaddr,
3054 .get_srcport = xs_sock_srcport,
3055 .buf_alloc = rpc_malloc,
3056 .buf_free = rpc_free,
3057 .send_request = xs_udp_send_request,
3058 .wait_for_reply_request = xprt_wait_for_reply_request_rtt,
3059 .timer = xs_udp_timer,
3060 .release_request = xprt_release_rqst_cong,
3061 .close = xs_close,
3062 .destroy = xs_destroy,
3063 .print_stats = xs_udp_print_stats,
3064 .enable_swap = xs_enable_swap,
3065 .disable_swap = xs_disable_swap,
3066 .inject_disconnect = xs_inject_disconnect,
3067};
3068
3069static const struct rpc_xprt_ops xs_tcp_ops = {
3070 .reserve_xprt = xprt_reserve_xprt,
3071 .release_xprt = xprt_release_xprt,
3072 .alloc_slot = xprt_alloc_slot,
3073 .free_slot = xprt_free_slot,
3074 .rpcbind = rpcb_getport_async,
3075 .set_port = xs_set_port,
3076 .connect = xs_connect,
3077 .get_srcaddr = xs_sock_srcaddr,
3078 .get_srcport = xs_sock_srcport,
3079 .buf_alloc = rpc_malloc,
3080 .buf_free = rpc_free,
3081 .prepare_request = xs_stream_prepare_request,
3082 .send_request = xs_tcp_send_request,
3083 .abort_send_request = xs_stream_abort_send_request,
3084 .wait_for_reply_request = xprt_wait_for_reply_request_def,
3085 .close = xs_tcp_shutdown,
3086 .destroy = xs_destroy,
3087 .set_connect_timeout = xs_tcp_set_connect_timeout,
3088 .print_stats = xs_tcp_print_stats,
3089 .enable_swap = xs_enable_swap,
3090 .disable_swap = xs_disable_swap,
3091 .inject_disconnect = xs_inject_disconnect,
3092#ifdef CONFIG_SUNRPC_BACKCHANNEL
3093 .bc_setup = xprt_setup_bc,
3094 .bc_maxpayload = xs_tcp_bc_maxpayload,
3095 .bc_num_slots = xprt_bc_max_slots,
3096 .bc_free_rqst = xprt_free_bc_rqst,
3097 .bc_destroy = xprt_destroy_bc,
3098#endif
3099};
3100
3101/*
3102 * The rpc_xprt_ops for the server backchannel
3103 */
3104
3105static const struct rpc_xprt_ops bc_tcp_ops = {
3106 .reserve_xprt = xprt_reserve_xprt,
3107 .release_xprt = xprt_release_xprt,
3108 .alloc_slot = xprt_alloc_slot,
3109 .free_slot = xprt_free_slot,
3110 .buf_alloc = bc_malloc,
3111 .buf_free = bc_free,
3112 .send_request = bc_send_request,
3113 .wait_for_reply_request = xprt_wait_for_reply_request_def,
3114 .close = bc_close,
3115 .destroy = bc_destroy,
3116 .print_stats = xs_tcp_print_stats,
3117 .enable_swap = xs_enable_swap,
3118 .disable_swap = xs_disable_swap,
3119 .inject_disconnect = xs_inject_disconnect,
3120};
3121
3122static int xs_init_anyaddr(const int family, struct sockaddr *sap)
3123{
3124 static const struct sockaddr_in sin = {
3125 .sin_family = AF_INET,
3126 .sin_addr.s_addr = htonl(INADDR_ANY),
3127 };
3128 static const struct sockaddr_in6 sin6 = {
3129 .sin6_family = AF_INET6,
3130 .sin6_addr = IN6ADDR_ANY_INIT,
3131 };
3132
3133 switch (family) {
3134 case AF_LOCAL:
3135 break;
3136 case AF_INET:
3137 memcpy(sap, &sin, sizeof(sin));
3138 break;
3139 case AF_INET6:
3140 memcpy(sap, &sin6, sizeof(sin6));
3141 break;
3142 default:
3143 dprintk("RPC: %s: Bad address family\n", __func__);
3144 return -EAFNOSUPPORT;
3145 }
3146 return 0;
3147}
3148
3149static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
3150 unsigned int slot_table_size,
3151 unsigned int max_slot_table_size)
3152{
3153 struct rpc_xprt *xprt;
3154 struct sock_xprt *new;
3155
3156 if (args->addrlen > sizeof(xprt->addr)) {
3157 dprintk("RPC: xs_setup_xprt: address too large\n");
3158 return ERR_PTR(-EBADF);
3159 }
3160
3161 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
3162 max_slot_table_size);
3163 if (xprt == NULL) {
3164 dprintk("RPC: xs_setup_xprt: couldn't allocate "
3165 "rpc_xprt\n");
3166 return ERR_PTR(-ENOMEM);
3167 }
3168
3169 new = container_of(xprt, struct sock_xprt, xprt);
3170 mutex_init(&new->recv_mutex);
3171 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
3172 xprt->addrlen = args->addrlen;
3173 if (args->srcaddr)
3174 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
3175 else {
3176 int err;
3177 err = xs_init_anyaddr(args->dstaddr->sa_family,
3178 (struct sockaddr *)&new->srcaddr);
3179 if (err != 0) {
3180 xprt_free(xprt);
3181 return ERR_PTR(err);
3182 }
3183 }
3184
3185 return xprt;
3186}
3187
3188static const struct rpc_timeout xs_local_default_timeout = {
3189 .to_initval = 10 * HZ,
3190 .to_maxval = 10 * HZ,
3191 .to_retries = 2,
3192};
3193
3194/**
3195 * xs_setup_local - Set up transport to use an AF_LOCAL socket
3196 * @args: rpc transport creation arguments
3197 *
3198 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
3199 */
3200static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
3201{
3202 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
3203 struct sock_xprt *transport;
3204 struct rpc_xprt *xprt;
3205 struct rpc_xprt *ret;
3206
3207 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3208 xprt_max_tcp_slot_table_entries);
3209 if (IS_ERR(xprt))
3210 return xprt;
3211 transport = container_of(xprt, struct sock_xprt, xprt);
3212
3213 xprt->prot = 0;
3214 xprt->xprt_class = &xs_local_transport;
3215 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3216
3217 xprt->bind_timeout = XS_BIND_TO;
3218 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3219 xprt->idle_timeout = XS_IDLE_DISC_TO;
3220
3221 xprt->ops = &xs_local_ops;
3222 xprt->timeout = &xs_local_default_timeout;
3223
3224 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3225 INIT_WORK(&transport->error_worker, xs_error_handle);
3226 INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket);
3227
3228 switch (sun->sun_family) {
3229 case AF_LOCAL:
3230 if (sun->sun_path[0] != '/' && sun->sun_path[0] != '\0') {
3231 dprintk("RPC: bad AF_LOCAL address: %s\n",
3232 sun->sun_path);
3233 ret = ERR_PTR(-EINVAL);
3234 goto out_err;
3235 }
3236 xprt_set_bound(xprt);
3237 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
3238 break;
3239 default:
3240 ret = ERR_PTR(-EAFNOSUPPORT);
3241 goto out_err;
3242 }
3243
3244 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
3245 xprt->address_strings[RPC_DISPLAY_ADDR]);
3246
3247 if (try_module_get(THIS_MODULE))
3248 return xprt;
3249 ret = ERR_PTR(-EINVAL);
3250out_err:
3251 xs_xprt_free(xprt);
3252 return ret;
3253}
3254
3255static const struct rpc_timeout xs_udp_default_timeout = {
3256 .to_initval = 5 * HZ,
3257 .to_maxval = 30 * HZ,
3258 .to_increment = 5 * HZ,
3259 .to_retries = 5,
3260};
3261
3262/**
3263 * xs_setup_udp - Set up transport to use a UDP socket
3264 * @args: rpc transport creation arguments
3265 *
3266 */
3267static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
3268{
3269 struct sockaddr *addr = args->dstaddr;
3270 struct rpc_xprt *xprt;
3271 struct sock_xprt *transport;
3272 struct rpc_xprt *ret;
3273
3274 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
3275 xprt_udp_slot_table_entries);
3276 if (IS_ERR(xprt))
3277 return xprt;
3278 transport = container_of(xprt, struct sock_xprt, xprt);
3279
3280 xprt->prot = IPPROTO_UDP;
3281 xprt->xprt_class = &xs_udp_transport;
3282 /* XXX: header size can vary due to auth type, IPv6, etc. */
3283 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
3284
3285 xprt->bind_timeout = XS_BIND_TO;
3286 xprt->reestablish_timeout = XS_UDP_REEST_TO;
3287 xprt->idle_timeout = XS_IDLE_DISC_TO;
3288
3289 xprt->ops = &xs_udp_ops;
3290
3291 xprt->timeout = &xs_udp_default_timeout;
3292
3293 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
3294 INIT_WORK(&transport->error_worker, xs_error_handle);
3295 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
3296
3297 switch (addr->sa_family) {
3298 case AF_INET:
3299 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3300 xprt_set_bound(xprt);
3301
3302 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
3303 break;
3304 case AF_INET6:
3305 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3306 xprt_set_bound(xprt);
3307
3308 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
3309 break;
3310 default:
3311 ret = ERR_PTR(-EAFNOSUPPORT);
3312 goto out_err;
3313 }
3314
3315 if (xprt_bound(xprt))
3316 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3317 xprt->address_strings[RPC_DISPLAY_ADDR],
3318 xprt->address_strings[RPC_DISPLAY_PORT],
3319 xprt->address_strings[RPC_DISPLAY_PROTO]);
3320 else
3321 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
3322 xprt->address_strings[RPC_DISPLAY_ADDR],
3323 xprt->address_strings[RPC_DISPLAY_PROTO]);
3324
3325 if (try_module_get(THIS_MODULE))
3326 return xprt;
3327 ret = ERR_PTR(-EINVAL);
3328out_err:
3329 xs_xprt_free(xprt);
3330 return ret;
3331}
3332
3333static const struct rpc_timeout xs_tcp_default_timeout = {
3334 .to_initval = 60 * HZ,
3335 .to_maxval = 60 * HZ,
3336 .to_retries = 2,
3337};
3338
3339/**
3340 * xs_setup_tcp - Set up transport to use a TCP socket
3341 * @args: rpc transport creation arguments
3342 *
3343 */
3344static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
3345{
3346 struct sockaddr *addr = args->dstaddr;
3347 struct rpc_xprt *xprt;
3348 struct sock_xprt *transport;
3349 struct rpc_xprt *ret;
3350 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
3351
3352 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
3353 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
3354
3355 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3356 max_slot_table_size);
3357 if (IS_ERR(xprt))
3358 return xprt;
3359 transport = container_of(xprt, struct sock_xprt, xprt);
3360
3361 xprt->prot = IPPROTO_TCP;
3362 xprt->xprt_class = &xs_tcp_transport;
3363 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3364
3365 xprt->bind_timeout = XS_BIND_TO;
3366 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3367 xprt->idle_timeout = XS_IDLE_DISC_TO;
3368
3369 xprt->ops = &xs_tcp_ops;
3370 xprt->timeout = &xs_tcp_default_timeout;
3371
3372 xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3373 if (args->reconnect_timeout)
3374 xprt->max_reconnect_timeout = args->reconnect_timeout;
3375
3376 xprt->connect_timeout = xprt->timeout->to_initval *
3377 (xprt->timeout->to_retries + 1);
3378 if (args->connect_timeout)
3379 xs_tcp_do_set_connect_timeout(xprt, args->connect_timeout);
3380
3381 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3382 INIT_WORK(&transport->error_worker, xs_error_handle);
3383 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
3384
3385 switch (addr->sa_family) {
3386 case AF_INET:
3387 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3388 xprt_set_bound(xprt);
3389
3390 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3391 break;
3392 case AF_INET6:
3393 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3394 xprt_set_bound(xprt);
3395
3396 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3397 break;
3398 default:
3399 ret = ERR_PTR(-EAFNOSUPPORT);
3400 goto out_err;
3401 }
3402
3403 if (xprt_bound(xprt))
3404 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3405 xprt->address_strings[RPC_DISPLAY_ADDR],
3406 xprt->address_strings[RPC_DISPLAY_PORT],
3407 xprt->address_strings[RPC_DISPLAY_PROTO]);
3408 else
3409 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
3410 xprt->address_strings[RPC_DISPLAY_ADDR],
3411 xprt->address_strings[RPC_DISPLAY_PROTO]);
3412
3413 if (try_module_get(THIS_MODULE))
3414 return xprt;
3415 ret = ERR_PTR(-EINVAL);
3416out_err:
3417 xs_xprt_free(xprt);
3418 return ret;
3419}
3420
3421/**
3422 * xs_setup_tcp_tls - Set up transport to use a TCP with TLS
3423 * @args: rpc transport creation arguments
3424 *
3425 */
3426static struct rpc_xprt *xs_setup_tcp_tls(struct xprt_create *args)
3427{
3428 struct sockaddr *addr = args->dstaddr;
3429 struct rpc_xprt *xprt;
3430 struct sock_xprt *transport;
3431 struct rpc_xprt *ret;
3432 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
3433
3434 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
3435 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
3436
3437 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3438 max_slot_table_size);
3439 if (IS_ERR(xprt))
3440 return xprt;
3441 transport = container_of(xprt, struct sock_xprt, xprt);
3442
3443 xprt->prot = IPPROTO_TCP;
3444 xprt->xprt_class = &xs_tcp_transport;
3445 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3446
3447 xprt->bind_timeout = XS_BIND_TO;
3448 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3449 xprt->idle_timeout = XS_IDLE_DISC_TO;
3450
3451 xprt->ops = &xs_tcp_ops;
3452 xprt->timeout = &xs_tcp_default_timeout;
3453
3454 xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3455 xprt->connect_timeout = xprt->timeout->to_initval *
3456 (xprt->timeout->to_retries + 1);
3457
3458 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3459 INIT_WORK(&transport->error_worker, xs_error_handle);
3460
3461 switch (args->xprtsec.policy) {
3462 case RPC_XPRTSEC_TLS_ANON:
3463 case RPC_XPRTSEC_TLS_X509:
3464 xprt->xprtsec = args->xprtsec;
3465 INIT_DELAYED_WORK(&transport->connect_worker,
3466 xs_tcp_tls_setup_socket);
3467 break;
3468 default:
3469 ret = ERR_PTR(-EACCES);
3470 goto out_err;
3471 }
3472
3473 switch (addr->sa_family) {
3474 case AF_INET:
3475 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3476 xprt_set_bound(xprt);
3477
3478 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3479 break;
3480 case AF_INET6:
3481 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3482 xprt_set_bound(xprt);
3483
3484 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3485 break;
3486 default:
3487 ret = ERR_PTR(-EAFNOSUPPORT);
3488 goto out_err;
3489 }
3490
3491 if (xprt_bound(xprt))
3492 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3493 xprt->address_strings[RPC_DISPLAY_ADDR],
3494 xprt->address_strings[RPC_DISPLAY_PORT],
3495 xprt->address_strings[RPC_DISPLAY_PROTO]);
3496 else
3497 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
3498 xprt->address_strings[RPC_DISPLAY_ADDR],
3499 xprt->address_strings[RPC_DISPLAY_PROTO]);
3500
3501 if (try_module_get(THIS_MODULE))
3502 return xprt;
3503 ret = ERR_PTR(-EINVAL);
3504out_err:
3505 xs_xprt_free(xprt);
3506 return ret;
3507}
3508
3509/**
3510 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3511 * @args: rpc transport creation arguments
3512 *
3513 */
3514static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3515{
3516 struct sockaddr *addr = args->dstaddr;
3517 struct rpc_xprt *xprt;
3518 struct sock_xprt *transport;
3519 struct svc_sock *bc_sock;
3520 struct rpc_xprt *ret;
3521
3522 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3523 xprt_tcp_slot_table_entries);
3524 if (IS_ERR(xprt))
3525 return xprt;
3526 transport = container_of(xprt, struct sock_xprt, xprt);
3527
3528 xprt->prot = IPPROTO_TCP;
3529 xprt->xprt_class = &xs_bc_tcp_transport;
3530 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3531 xprt->timeout = &xs_tcp_default_timeout;
3532
3533 /* backchannel */
3534 xprt_set_bound(xprt);
3535 xprt->bind_timeout = 0;
3536 xprt->reestablish_timeout = 0;
3537 xprt->idle_timeout = 0;
3538
3539 xprt->ops = &bc_tcp_ops;
3540
3541 switch (addr->sa_family) {
3542 case AF_INET:
3543 xs_format_peer_addresses(xprt, "tcp",
3544 RPCBIND_NETID_TCP);
3545 break;
3546 case AF_INET6:
3547 xs_format_peer_addresses(xprt, "tcp",
3548 RPCBIND_NETID_TCP6);
3549 break;
3550 default:
3551 ret = ERR_PTR(-EAFNOSUPPORT);
3552 goto out_err;
3553 }
3554
3555 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3556 xprt->address_strings[RPC_DISPLAY_ADDR],
3557 xprt->address_strings[RPC_DISPLAY_PORT],
3558 xprt->address_strings[RPC_DISPLAY_PROTO]);
3559
3560 /*
3561 * Once we've associated a backchannel xprt with a connection,
3562 * we want to keep it around as long as the connection lasts,
3563 * in case we need to start using it for a backchannel again;
3564 * this reference won't be dropped until bc_xprt is destroyed.
3565 */
3566 xprt_get(xprt);
3567 args->bc_xprt->xpt_bc_xprt = xprt;
3568 xprt->bc_xprt = args->bc_xprt;
3569 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3570 transport->sock = bc_sock->sk_sock;
3571 transport->inet = bc_sock->sk_sk;
3572
3573 /*
3574 * Since we don't want connections for the backchannel, we set
3575 * the xprt status to connected
3576 */
3577 xprt_set_connected(xprt);
3578
3579 if (try_module_get(THIS_MODULE))
3580 return xprt;
3581
3582 args->bc_xprt->xpt_bc_xprt = NULL;
3583 args->bc_xprt->xpt_bc_xps = NULL;
3584 xprt_put(xprt);
3585 ret = ERR_PTR(-EINVAL);
3586out_err:
3587 xs_xprt_free(xprt);
3588 return ret;
3589}
3590
3591static struct xprt_class xs_local_transport = {
3592 .list = LIST_HEAD_INIT(xs_local_transport.list),
3593 .name = "named UNIX socket",
3594 .owner = THIS_MODULE,
3595 .ident = XPRT_TRANSPORT_LOCAL,
3596 .setup = xs_setup_local,
3597 .netid = { "" },
3598};
3599
3600static struct xprt_class xs_udp_transport = {
3601 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3602 .name = "udp",
3603 .owner = THIS_MODULE,
3604 .ident = XPRT_TRANSPORT_UDP,
3605 .setup = xs_setup_udp,
3606 .netid = { "udp", "udp6", "" },
3607};
3608
3609static struct xprt_class xs_tcp_transport = {
3610 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3611 .name = "tcp",
3612 .owner = THIS_MODULE,
3613 .ident = XPRT_TRANSPORT_TCP,
3614 .setup = xs_setup_tcp,
3615 .netid = { "tcp", "tcp6", "" },
3616};
3617
3618static struct xprt_class xs_tcp_tls_transport = {
3619 .list = LIST_HEAD_INIT(xs_tcp_tls_transport.list),
3620 .name = "tcp-with-tls",
3621 .owner = THIS_MODULE,
3622 .ident = XPRT_TRANSPORT_TCP_TLS,
3623 .setup = xs_setup_tcp_tls,
3624 .netid = { "tcp", "tcp6", "" },
3625};
3626
3627static struct xprt_class xs_bc_tcp_transport = {
3628 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3629 .name = "tcp NFSv4.1 backchannel",
3630 .owner = THIS_MODULE,
3631 .ident = XPRT_TRANSPORT_BC_TCP,
3632 .setup = xs_setup_bc_tcp,
3633 .netid = { "" },
3634};
3635
3636/**
3637 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3638 *
3639 */
3640int init_socket_xprt(void)
3641{
3642 if (!sunrpc_table_header)
3643 sunrpc_table_header = register_sysctl("sunrpc", xs_tunables_table);
3644
3645 xprt_register_transport(&xs_local_transport);
3646 xprt_register_transport(&xs_udp_transport);
3647 xprt_register_transport(&xs_tcp_transport);
3648 xprt_register_transport(&xs_tcp_tls_transport);
3649 xprt_register_transport(&xs_bc_tcp_transport);
3650
3651 return 0;
3652}
3653
3654/**
3655 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3656 *
3657 */
3658void cleanup_socket_xprt(void)
3659{
3660 if (sunrpc_table_header) {
3661 unregister_sysctl_table(sunrpc_table_header);
3662 sunrpc_table_header = NULL;
3663 }
3664
3665 xprt_unregister_transport(&xs_local_transport);
3666 xprt_unregister_transport(&xs_udp_transport);
3667 xprt_unregister_transport(&xs_tcp_transport);
3668 xprt_unregister_transport(&xs_tcp_tls_transport);
3669 xprt_unregister_transport(&xs_bc_tcp_transport);
3670}
3671
3672static int param_set_portnr(const char *val, const struct kernel_param *kp)
3673{
3674 return param_set_uint_minmax(val, kp,
3675 RPC_MIN_RESVPORT,
3676 RPC_MAX_RESVPORT);
3677}
3678
3679static const struct kernel_param_ops param_ops_portnr = {
3680 .set = param_set_portnr,
3681 .get = param_get_uint,
3682};
3683
3684#define param_check_portnr(name, p) \
3685 __param_check(name, p, unsigned int);
3686
3687module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3688module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3689
3690static int param_set_slot_table_size(const char *val,
3691 const struct kernel_param *kp)
3692{
3693 return param_set_uint_minmax(val, kp,
3694 RPC_MIN_SLOT_TABLE,
3695 RPC_MAX_SLOT_TABLE);
3696}
3697
3698static const struct kernel_param_ops param_ops_slot_table_size = {
3699 .set = param_set_slot_table_size,
3700 .get = param_get_uint,
3701};
3702
3703#define param_check_slot_table_size(name, p) \
3704 __param_check(name, p, unsigned int);
3705
3706static int param_set_max_slot_table_size(const char *val,
3707 const struct kernel_param *kp)
3708{
3709 return param_set_uint_minmax(val, kp,
3710 RPC_MIN_SLOT_TABLE,
3711 RPC_MAX_SLOT_TABLE_LIMIT);
3712}
3713
3714static const struct kernel_param_ops param_ops_max_slot_table_size = {
3715 .set = param_set_max_slot_table_size,
3716 .get = param_get_uint,
3717};
3718
3719#define param_check_max_slot_table_size(name, p) \
3720 __param_check(name, p, unsigned int);
3721
3722module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3723 slot_table_size, 0644);
3724module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3725 max_slot_table_size, 0644);
3726module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3727 slot_table_size, 0644);