Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/net/sunrpc/xprtsock.c
4 *
5 * Client-side transport implementation for sockets.
6 *
7 * TCP callback races fixes (C) 1998 Red Hat
8 * TCP send fixes (C) 1998 Red Hat
9 * TCP NFS related read + write fixes
10 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
11 *
12 * Rewrite of larges part of the code in order to stabilize TCP stuff.
13 * Fix behaviour when socket buffer is full.
14 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
15 *
16 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
17 *
18 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
19 * <gilles.quillard@bull.net>
20 */
21
22#include <linux/types.h>
23#include <linux/string.h>
24#include <linux/slab.h>
25#include <linux/module.h>
26#include <linux/capability.h>
27#include <linux/pagemap.h>
28#include <linux/errno.h>
29#include <linux/socket.h>
30#include <linux/in.h>
31#include <linux/net.h>
32#include <linux/mm.h>
33#include <linux/un.h>
34#include <linux/udp.h>
35#include <linux/tcp.h>
36#include <linux/sunrpc/clnt.h>
37#include <linux/sunrpc/addr.h>
38#include <linux/sunrpc/sched.h>
39#include <linux/sunrpc/svcsock.h>
40#include <linux/sunrpc/xprtsock.h>
41#include <linux/file.h>
42#ifdef CONFIG_SUNRPC_BACKCHANNEL
43#include <linux/sunrpc/bc_xprt.h>
44#endif
45
46#include <net/sock.h>
47#include <net/checksum.h>
48#include <net/udp.h>
49#include <net/tcp.h>
50#include <linux/bvec.h>
51#include <linux/highmem.h>
52#include <linux/uio.h>
53#include <linux/sched/mm.h>
54
55#include <trace/events/sunrpc.h>
56
57#include "socklib.h"
58#include "sunrpc.h"
59
60static void xs_close(struct rpc_xprt *xprt);
61static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
62 struct socket *sock);
63
64/*
65 * xprtsock tunables
66 */
67static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
68static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
69static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
70
71static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
72static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
73
74#define XS_TCP_LINGER_TO (15U * HZ)
75static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
76
77/*
78 * We can register our own files under /proc/sys/sunrpc by
79 * calling register_sysctl_table() again. The files in that
80 * directory become the union of all files registered there.
81 *
82 * We simply need to make sure that we don't collide with
83 * someone else's file names!
84 */
85
86static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
87static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
88static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
89static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
90static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
91
92static struct ctl_table_header *sunrpc_table_header;
93
94/*
95 * FIXME: changing the UDP slot table size should also resize the UDP
96 * socket buffers for existing UDP transports
97 */
98static struct ctl_table xs_tunables_table[] = {
99 {
100 .procname = "udp_slot_table_entries",
101 .data = &xprt_udp_slot_table_entries,
102 .maxlen = sizeof(unsigned int),
103 .mode = 0644,
104 .proc_handler = proc_dointvec_minmax,
105 .extra1 = &min_slot_table_size,
106 .extra2 = &max_slot_table_size
107 },
108 {
109 .procname = "tcp_slot_table_entries",
110 .data = &xprt_tcp_slot_table_entries,
111 .maxlen = sizeof(unsigned int),
112 .mode = 0644,
113 .proc_handler = proc_dointvec_minmax,
114 .extra1 = &min_slot_table_size,
115 .extra2 = &max_slot_table_size
116 },
117 {
118 .procname = "tcp_max_slot_table_entries",
119 .data = &xprt_max_tcp_slot_table_entries,
120 .maxlen = sizeof(unsigned int),
121 .mode = 0644,
122 .proc_handler = proc_dointvec_minmax,
123 .extra1 = &min_slot_table_size,
124 .extra2 = &max_tcp_slot_table_limit
125 },
126 {
127 .procname = "min_resvport",
128 .data = &xprt_min_resvport,
129 .maxlen = sizeof(unsigned int),
130 .mode = 0644,
131 .proc_handler = proc_dointvec_minmax,
132 .extra1 = &xprt_min_resvport_limit,
133 .extra2 = &xprt_max_resvport_limit
134 },
135 {
136 .procname = "max_resvport",
137 .data = &xprt_max_resvport,
138 .maxlen = sizeof(unsigned int),
139 .mode = 0644,
140 .proc_handler = proc_dointvec_minmax,
141 .extra1 = &xprt_min_resvport_limit,
142 .extra2 = &xprt_max_resvport_limit
143 },
144 {
145 .procname = "tcp_fin_timeout",
146 .data = &xs_tcp_fin_timeout,
147 .maxlen = sizeof(xs_tcp_fin_timeout),
148 .mode = 0644,
149 .proc_handler = proc_dointvec_jiffies,
150 },
151 { },
152};
153
154static struct ctl_table sunrpc_table[] = {
155 {
156 .procname = "sunrpc",
157 .mode = 0555,
158 .child = xs_tunables_table
159 },
160 { },
161};
162
163/*
164 * Wait duration for a reply from the RPC portmapper.
165 */
166#define XS_BIND_TO (60U * HZ)
167
168/*
169 * Delay if a UDP socket connect error occurs. This is most likely some
170 * kind of resource problem on the local host.
171 */
172#define XS_UDP_REEST_TO (2U * HZ)
173
174/*
175 * The reestablish timeout allows clients to delay for a bit before attempting
176 * to reconnect to a server that just dropped our connection.
177 *
178 * We implement an exponential backoff when trying to reestablish a TCP
179 * transport connection with the server. Some servers like to drop a TCP
180 * connection when they are overworked, so we start with a short timeout and
181 * increase over time if the server is down or not responding.
182 */
183#define XS_TCP_INIT_REEST_TO (3U * HZ)
184
185/*
186 * TCP idle timeout; client drops the transport socket if it is idle
187 * for this long. Note that we also timeout UDP sockets to prevent
188 * holding port numbers when there is no RPC traffic.
189 */
190#define XS_IDLE_DISC_TO (5U * 60 * HZ)
191
192#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
193# undef RPC_DEBUG_DATA
194# define RPCDBG_FACILITY RPCDBG_TRANS
195#endif
196
197#ifdef RPC_DEBUG_DATA
198static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
199{
200 u8 *buf = (u8 *) packet;
201 int j;
202
203 dprintk("RPC: %s\n", msg);
204 for (j = 0; j < count && j < 128; j += 4) {
205 if (!(j & 31)) {
206 if (j)
207 dprintk("\n");
208 dprintk("0x%04x ", j);
209 }
210 dprintk("%02x%02x%02x%02x ",
211 buf[j], buf[j+1], buf[j+2], buf[j+3]);
212 }
213 dprintk("\n");
214}
215#else
216static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
217{
218 /* NOP */
219}
220#endif
221
222static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
223{
224 return (struct rpc_xprt *) sk->sk_user_data;
225}
226
227static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
228{
229 return (struct sockaddr *) &xprt->addr;
230}
231
232static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
233{
234 return (struct sockaddr_un *) &xprt->addr;
235}
236
237static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
238{
239 return (struct sockaddr_in *) &xprt->addr;
240}
241
242static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
243{
244 return (struct sockaddr_in6 *) &xprt->addr;
245}
246
247static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
248{
249 struct sockaddr *sap = xs_addr(xprt);
250 struct sockaddr_in6 *sin6;
251 struct sockaddr_in *sin;
252 struct sockaddr_un *sun;
253 char buf[128];
254
255 switch (sap->sa_family) {
256 case AF_LOCAL:
257 sun = xs_addr_un(xprt);
258 strlcpy(buf, sun->sun_path, sizeof(buf));
259 xprt->address_strings[RPC_DISPLAY_ADDR] =
260 kstrdup(buf, GFP_KERNEL);
261 break;
262 case AF_INET:
263 (void)rpc_ntop(sap, buf, sizeof(buf));
264 xprt->address_strings[RPC_DISPLAY_ADDR] =
265 kstrdup(buf, GFP_KERNEL);
266 sin = xs_addr_in(xprt);
267 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
268 break;
269 case AF_INET6:
270 (void)rpc_ntop(sap, buf, sizeof(buf));
271 xprt->address_strings[RPC_DISPLAY_ADDR] =
272 kstrdup(buf, GFP_KERNEL);
273 sin6 = xs_addr_in6(xprt);
274 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
275 break;
276 default:
277 BUG();
278 }
279
280 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
281}
282
283static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
284{
285 struct sockaddr *sap = xs_addr(xprt);
286 char buf[128];
287
288 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
289 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
290
291 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
292 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
293}
294
295static void xs_format_peer_addresses(struct rpc_xprt *xprt,
296 const char *protocol,
297 const char *netid)
298{
299 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
300 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
301 xs_format_common_peer_addresses(xprt);
302 xs_format_common_peer_ports(xprt);
303}
304
305static void xs_update_peer_port(struct rpc_xprt *xprt)
306{
307 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
308 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
309
310 xs_format_common_peer_ports(xprt);
311}
312
313static void xs_free_peer_addresses(struct rpc_xprt *xprt)
314{
315 unsigned int i;
316
317 for (i = 0; i < RPC_DISPLAY_MAX; i++)
318 switch (i) {
319 case RPC_DISPLAY_PROTO:
320 case RPC_DISPLAY_NETID:
321 continue;
322 default:
323 kfree(xprt->address_strings[i]);
324 }
325}
326
327static size_t
328xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp)
329{
330 size_t i,n;
331
332 if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES))
333 return want;
334 n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT;
335 for (i = 0; i < n; i++) {
336 if (buf->pages[i])
337 continue;
338 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp);
339 if (!buf->pages[i]) {
340 i *= PAGE_SIZE;
341 return i > buf->page_base ? i - buf->page_base : 0;
342 }
343 }
344 return want;
345}
346
347static ssize_t
348xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek)
349{
350 ssize_t ret;
351 if (seek != 0)
352 iov_iter_advance(&msg->msg_iter, seek);
353 ret = sock_recvmsg(sock, msg, flags);
354 return ret > 0 ? ret + seek : ret;
355}
356
357static ssize_t
358xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags,
359 struct kvec *kvec, size_t count, size_t seek)
360{
361 iov_iter_kvec(&msg->msg_iter, READ, kvec, 1, count);
362 return xs_sock_recvmsg(sock, msg, flags, seek);
363}
364
365static ssize_t
366xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags,
367 struct bio_vec *bvec, unsigned long nr, size_t count,
368 size_t seek)
369{
370 iov_iter_bvec(&msg->msg_iter, READ, bvec, nr, count);
371 return xs_sock_recvmsg(sock, msg, flags, seek);
372}
373
374static ssize_t
375xs_read_discard(struct socket *sock, struct msghdr *msg, int flags,
376 size_t count)
377{
378 iov_iter_discard(&msg->msg_iter, READ, count);
379 return sock_recvmsg(sock, msg, flags);
380}
381
382#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
383static void
384xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
385{
386 struct bvec_iter bi = {
387 .bi_size = count,
388 };
389 struct bio_vec bv;
390
391 bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
392 for_each_bvec(bv, bvec, bi, bi)
393 flush_dcache_page(bv.bv_page);
394}
395#else
396static inline void
397xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
398{
399}
400#endif
401
402static ssize_t
403xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
404 struct xdr_buf *buf, size_t count, size_t seek, size_t *read)
405{
406 size_t want, seek_init = seek, offset = 0;
407 ssize_t ret;
408
409 want = min_t(size_t, count, buf->head[0].iov_len);
410 if (seek < want) {
411 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek);
412 if (ret <= 0)
413 goto sock_err;
414 offset += ret;
415 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
416 goto out;
417 if (ret != want)
418 goto out;
419 seek = 0;
420 } else {
421 seek -= want;
422 offset += want;
423 }
424
425 want = xs_alloc_sparse_pages(buf,
426 min_t(size_t, count - offset, buf->page_len),
427 GFP_KERNEL);
428 if (seek < want) {
429 ret = xs_read_bvec(sock, msg, flags, buf->bvec,
430 xdr_buf_pagecount(buf),
431 want + buf->page_base,
432 seek + buf->page_base);
433 if (ret <= 0)
434 goto sock_err;
435 xs_flush_bvec(buf->bvec, ret, seek + buf->page_base);
436 offset += ret - buf->page_base;
437 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
438 goto out;
439 if (ret != want)
440 goto out;
441 seek = 0;
442 } else {
443 seek -= want;
444 offset += want;
445 }
446
447 want = min_t(size_t, count - offset, buf->tail[0].iov_len);
448 if (seek < want) {
449 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek);
450 if (ret <= 0)
451 goto sock_err;
452 offset += ret;
453 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
454 goto out;
455 if (ret != want)
456 goto out;
457 } else if (offset < seek_init)
458 offset = seek_init;
459 ret = -EMSGSIZE;
460out:
461 *read = offset - seek_init;
462 return ret;
463sock_err:
464 offset += seek;
465 goto out;
466}
467
468static void
469xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf)
470{
471 if (!transport->recv.copied) {
472 if (buf->head[0].iov_len >= transport->recv.offset)
473 memcpy(buf->head[0].iov_base,
474 &transport->recv.xid,
475 transport->recv.offset);
476 transport->recv.copied = transport->recv.offset;
477 }
478}
479
480static bool
481xs_read_stream_request_done(struct sock_xprt *transport)
482{
483 return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT);
484}
485
486static void
487xs_read_stream_check_eor(struct sock_xprt *transport,
488 struct msghdr *msg)
489{
490 if (xs_read_stream_request_done(transport))
491 msg->msg_flags |= MSG_EOR;
492}
493
494static ssize_t
495xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg,
496 int flags, struct rpc_rqst *req)
497{
498 struct xdr_buf *buf = &req->rq_private_buf;
499 size_t want, read;
500 ssize_t ret;
501
502 xs_read_header(transport, buf);
503
504 want = transport->recv.len - transport->recv.offset;
505 if (want != 0) {
506 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf,
507 transport->recv.copied + want,
508 transport->recv.copied,
509 &read);
510 transport->recv.offset += read;
511 transport->recv.copied += read;
512 }
513
514 if (transport->recv.offset == transport->recv.len)
515 xs_read_stream_check_eor(transport, msg);
516
517 if (want == 0)
518 return 0;
519
520 switch (ret) {
521 default:
522 break;
523 case -EFAULT:
524 case -EMSGSIZE:
525 msg->msg_flags |= MSG_TRUNC;
526 return read;
527 case 0:
528 return -ESHUTDOWN;
529 }
530 return ret < 0 ? ret : read;
531}
532
533static size_t
534xs_read_stream_headersize(bool isfrag)
535{
536 if (isfrag)
537 return sizeof(__be32);
538 return 3 * sizeof(__be32);
539}
540
541static ssize_t
542xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg,
543 int flags, size_t want, size_t seek)
544{
545 struct kvec kvec = {
546 .iov_base = &transport->recv.fraghdr,
547 .iov_len = want,
548 };
549 return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek);
550}
551
552#if defined(CONFIG_SUNRPC_BACKCHANNEL)
553static ssize_t
554xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
555{
556 struct rpc_xprt *xprt = &transport->xprt;
557 struct rpc_rqst *req;
558 ssize_t ret;
559
560 /* Look up and lock the request corresponding to the given XID */
561 req = xprt_lookup_bc_request(xprt, transport->recv.xid);
562 if (!req) {
563 printk(KERN_WARNING "Callback slot table overflowed\n");
564 return -ESHUTDOWN;
565 }
566 if (transport->recv.copied && !req->rq_private_buf.len)
567 return -ESHUTDOWN;
568
569 ret = xs_read_stream_request(transport, msg, flags, req);
570 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
571 xprt_complete_bc_request(req, transport->recv.copied);
572 else
573 req->rq_private_buf.len = transport->recv.copied;
574
575 return ret;
576}
577#else /* CONFIG_SUNRPC_BACKCHANNEL */
578static ssize_t
579xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
580{
581 return -ESHUTDOWN;
582}
583#endif /* CONFIG_SUNRPC_BACKCHANNEL */
584
585static ssize_t
586xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags)
587{
588 struct rpc_xprt *xprt = &transport->xprt;
589 struct rpc_rqst *req;
590 ssize_t ret = 0;
591
592 /* Look up and lock the request corresponding to the given XID */
593 spin_lock(&xprt->queue_lock);
594 req = xprt_lookup_rqst(xprt, transport->recv.xid);
595 if (!req || (transport->recv.copied && !req->rq_private_buf.len)) {
596 msg->msg_flags |= MSG_TRUNC;
597 goto out;
598 }
599 xprt_pin_rqst(req);
600 spin_unlock(&xprt->queue_lock);
601
602 ret = xs_read_stream_request(transport, msg, flags, req);
603
604 spin_lock(&xprt->queue_lock);
605 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
606 xprt_complete_rqst(req->rq_task, transport->recv.copied);
607 else
608 req->rq_private_buf.len = transport->recv.copied;
609 xprt_unpin_rqst(req);
610out:
611 spin_unlock(&xprt->queue_lock);
612 return ret;
613}
614
615static ssize_t
616xs_read_stream(struct sock_xprt *transport, int flags)
617{
618 struct msghdr msg = { 0 };
619 size_t want, read = 0;
620 ssize_t ret = 0;
621
622 if (transport->recv.len == 0) {
623 want = xs_read_stream_headersize(transport->recv.copied != 0);
624 ret = xs_read_stream_header(transport, &msg, flags, want,
625 transport->recv.offset);
626 if (ret <= 0)
627 goto out_err;
628 transport->recv.offset = ret;
629 if (transport->recv.offset != want)
630 return transport->recv.offset;
631 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) &
632 RPC_FRAGMENT_SIZE_MASK;
633 transport->recv.offset -= sizeof(transport->recv.fraghdr);
634 read = ret;
635 }
636
637 switch (be32_to_cpu(transport->recv.calldir)) {
638 default:
639 msg.msg_flags |= MSG_TRUNC;
640 break;
641 case RPC_CALL:
642 ret = xs_read_stream_call(transport, &msg, flags);
643 break;
644 case RPC_REPLY:
645 ret = xs_read_stream_reply(transport, &msg, flags);
646 }
647 if (msg.msg_flags & MSG_TRUNC) {
648 transport->recv.calldir = cpu_to_be32(-1);
649 transport->recv.copied = -1;
650 }
651 if (ret < 0)
652 goto out_err;
653 read += ret;
654 if (transport->recv.offset < transport->recv.len) {
655 if (!(msg.msg_flags & MSG_TRUNC))
656 return read;
657 msg.msg_flags = 0;
658 ret = xs_read_discard(transport->sock, &msg, flags,
659 transport->recv.len - transport->recv.offset);
660 if (ret <= 0)
661 goto out_err;
662 transport->recv.offset += ret;
663 read += ret;
664 if (transport->recv.offset != transport->recv.len)
665 return read;
666 }
667 if (xs_read_stream_request_done(transport)) {
668 trace_xs_stream_read_request(transport);
669 transport->recv.copied = 0;
670 }
671 transport->recv.offset = 0;
672 transport->recv.len = 0;
673 return read;
674out_err:
675 return ret != 0 ? ret : -ESHUTDOWN;
676}
677
678static __poll_t xs_poll_socket(struct sock_xprt *transport)
679{
680 return transport->sock->ops->poll(transport->file, transport->sock,
681 NULL);
682}
683
684static bool xs_poll_socket_readable(struct sock_xprt *transport)
685{
686 __poll_t events = xs_poll_socket(transport);
687
688 return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP);
689}
690
691static void xs_poll_check_readable(struct sock_xprt *transport)
692{
693
694 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
695 if (!xs_poll_socket_readable(transport))
696 return;
697 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
698 queue_work(xprtiod_workqueue, &transport->recv_worker);
699}
700
701static void xs_stream_data_receive(struct sock_xprt *transport)
702{
703 size_t read = 0;
704 ssize_t ret = 0;
705
706 mutex_lock(&transport->recv_mutex);
707 if (transport->sock == NULL)
708 goto out;
709 for (;;) {
710 ret = xs_read_stream(transport, MSG_DONTWAIT);
711 if (ret < 0)
712 break;
713 read += ret;
714 cond_resched();
715 }
716 if (ret == -ESHUTDOWN)
717 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
718 else
719 xs_poll_check_readable(transport);
720out:
721 mutex_unlock(&transport->recv_mutex);
722 trace_xs_stream_read_data(&transport->xprt, ret, read);
723}
724
725static void xs_stream_data_receive_workfn(struct work_struct *work)
726{
727 struct sock_xprt *transport =
728 container_of(work, struct sock_xprt, recv_worker);
729 unsigned int pflags = memalloc_nofs_save();
730
731 xs_stream_data_receive(transport);
732 memalloc_nofs_restore(pflags);
733}
734
735static void
736xs_stream_reset_connect(struct sock_xprt *transport)
737{
738 transport->recv.offset = 0;
739 transport->recv.len = 0;
740 transport->recv.copied = 0;
741 transport->xmit.offset = 0;
742}
743
744static void
745xs_stream_start_connect(struct sock_xprt *transport)
746{
747 transport->xprt.stat.connect_count++;
748 transport->xprt.stat.connect_start = jiffies;
749}
750
751#define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
752
753/**
754 * xs_nospace - handle transmit was incomplete
755 * @req: pointer to RPC request
756 *
757 */
758static int xs_nospace(struct rpc_rqst *req)
759{
760 struct rpc_xprt *xprt = req->rq_xprt;
761 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
762 struct sock *sk = transport->inet;
763 int ret = -EAGAIN;
764
765 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
766 req->rq_task->tk_pid,
767 req->rq_slen - transport->xmit.offset,
768 req->rq_slen);
769
770 /* Protect against races with write_space */
771 spin_lock(&xprt->transport_lock);
772
773 /* Don't race with disconnect */
774 if (xprt_connected(xprt)) {
775 /* wait for more buffer space */
776 sk->sk_write_pending++;
777 xprt_wait_for_buffer_space(xprt);
778 } else
779 ret = -ENOTCONN;
780
781 spin_unlock(&xprt->transport_lock);
782
783 /* Race breaker in case memory is freed before above code is called */
784 if (ret == -EAGAIN) {
785 struct socket_wq *wq;
786
787 rcu_read_lock();
788 wq = rcu_dereference(sk->sk_wq);
789 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
790 rcu_read_unlock();
791
792 sk->sk_write_space(sk);
793 }
794 return ret;
795}
796
797static void
798xs_stream_prepare_request(struct rpc_rqst *req)
799{
800 xdr_free_bvec(&req->rq_rcv_buf);
801 req->rq_task->tk_status = xdr_alloc_bvec(&req->rq_rcv_buf, GFP_KERNEL);
802}
803
804/*
805 * Determine if the previous message in the stream was aborted before it
806 * could complete transmission.
807 */
808static bool
809xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req)
810{
811 return transport->xmit.offset != 0 && req->rq_bytes_sent == 0;
812}
813
814/*
815 * Return the stream record marker field for a record of length < 2^31-1
816 */
817static rpc_fraghdr
818xs_stream_record_marker(struct xdr_buf *xdr)
819{
820 if (!xdr->len)
821 return 0;
822 return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len);
823}
824
825/**
826 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
827 * @req: pointer to RPC request
828 *
829 * Return values:
830 * 0: The request has been sent
831 * EAGAIN: The socket was blocked, please call again later to
832 * complete the request
833 * ENOTCONN: Caller needs to invoke connect logic then call again
834 * other: Some other error occured, the request was not sent
835 */
836static int xs_local_send_request(struct rpc_rqst *req)
837{
838 struct rpc_xprt *xprt = req->rq_xprt;
839 struct sock_xprt *transport =
840 container_of(xprt, struct sock_xprt, xprt);
841 struct xdr_buf *xdr = &req->rq_snd_buf;
842 rpc_fraghdr rm = xs_stream_record_marker(xdr);
843 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
844 struct msghdr msg = {
845 .msg_flags = XS_SENDMSG_FLAGS,
846 };
847 unsigned int sent;
848 int status;
849
850 /* Close the stream if the previous transmission was incomplete */
851 if (xs_send_request_was_aborted(transport, req)) {
852 xs_close(xprt);
853 return -ENOTCONN;
854 }
855
856 xs_pktdump("packet data:",
857 req->rq_svec->iov_base, req->rq_svec->iov_len);
858
859 req->rq_xtime = ktime_get();
860 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
861 transport->xmit.offset, rm, &sent);
862 dprintk("RPC: %s(%u) = %d\n",
863 __func__, xdr->len - transport->xmit.offset, status);
864
865 if (status == -EAGAIN && sock_writeable(transport->inet))
866 status = -ENOBUFS;
867
868 if (likely(sent > 0) || status == 0) {
869 transport->xmit.offset += sent;
870 req->rq_bytes_sent = transport->xmit.offset;
871 if (likely(req->rq_bytes_sent >= msglen)) {
872 req->rq_xmit_bytes_sent += transport->xmit.offset;
873 transport->xmit.offset = 0;
874 return 0;
875 }
876 status = -EAGAIN;
877 }
878
879 switch (status) {
880 case -ENOBUFS:
881 break;
882 case -EAGAIN:
883 status = xs_nospace(req);
884 break;
885 default:
886 dprintk("RPC: sendmsg returned unrecognized error %d\n",
887 -status);
888 fallthrough;
889 case -EPIPE:
890 xs_close(xprt);
891 status = -ENOTCONN;
892 }
893
894 return status;
895}
896
897/**
898 * xs_udp_send_request - write an RPC request to a UDP socket
899 * @req: pointer to RPC request
900 *
901 * Return values:
902 * 0: The request has been sent
903 * EAGAIN: The socket was blocked, please call again later to
904 * complete the request
905 * ENOTCONN: Caller needs to invoke connect logic then call again
906 * other: Some other error occurred, the request was not sent
907 */
908static int xs_udp_send_request(struct rpc_rqst *req)
909{
910 struct rpc_xprt *xprt = req->rq_xprt;
911 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
912 struct xdr_buf *xdr = &req->rq_snd_buf;
913 struct msghdr msg = {
914 .msg_name = xs_addr(xprt),
915 .msg_namelen = xprt->addrlen,
916 .msg_flags = XS_SENDMSG_FLAGS,
917 };
918 unsigned int sent;
919 int status;
920
921 xs_pktdump("packet data:",
922 req->rq_svec->iov_base,
923 req->rq_svec->iov_len);
924
925 if (!xprt_bound(xprt))
926 return -ENOTCONN;
927
928 if (!xprt_request_get_cong(xprt, req))
929 return -EBADSLT;
930
931 req->rq_xtime = ktime_get();
932 status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, 0, &sent);
933
934 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
935 xdr->len, status);
936
937 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
938 if (status == -EPERM)
939 goto process_status;
940
941 if (status == -EAGAIN && sock_writeable(transport->inet))
942 status = -ENOBUFS;
943
944 if (sent > 0 || status == 0) {
945 req->rq_xmit_bytes_sent += sent;
946 if (sent >= req->rq_slen)
947 return 0;
948 /* Still some bytes left; set up for a retry later. */
949 status = -EAGAIN;
950 }
951
952process_status:
953 switch (status) {
954 case -ENOTSOCK:
955 status = -ENOTCONN;
956 /* Should we call xs_close() here? */
957 break;
958 case -EAGAIN:
959 status = xs_nospace(req);
960 break;
961 case -ENETUNREACH:
962 case -ENOBUFS:
963 case -EPIPE:
964 case -ECONNREFUSED:
965 case -EPERM:
966 /* When the server has died, an ICMP port unreachable message
967 * prompts ECONNREFUSED. */
968 break;
969 default:
970 dprintk("RPC: sendmsg returned unrecognized error %d\n",
971 -status);
972 }
973
974 return status;
975}
976
977/**
978 * xs_tcp_send_request - write an RPC request to a TCP socket
979 * @req: pointer to RPC request
980 *
981 * Return values:
982 * 0: The request has been sent
983 * EAGAIN: The socket was blocked, please call again later to
984 * complete the request
985 * ENOTCONN: Caller needs to invoke connect logic then call again
986 * other: Some other error occurred, the request was not sent
987 *
988 * XXX: In the case of soft timeouts, should we eventually give up
989 * if sendmsg is not able to make progress?
990 */
991static int xs_tcp_send_request(struct rpc_rqst *req)
992{
993 struct rpc_xprt *xprt = req->rq_xprt;
994 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
995 struct xdr_buf *xdr = &req->rq_snd_buf;
996 rpc_fraghdr rm = xs_stream_record_marker(xdr);
997 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
998 struct msghdr msg = {
999 .msg_flags = XS_SENDMSG_FLAGS,
1000 };
1001 bool vm_wait = false;
1002 unsigned int sent;
1003 int status;
1004
1005 /* Close the stream if the previous transmission was incomplete */
1006 if (xs_send_request_was_aborted(transport, req)) {
1007 if (transport->sock != NULL)
1008 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
1009 return -ENOTCONN;
1010 }
1011
1012 xs_pktdump("packet data:",
1013 req->rq_svec->iov_base,
1014 req->rq_svec->iov_len);
1015
1016 if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
1017 xs_tcp_set_socket_timeouts(xprt, transport->sock);
1018
1019 /* Continue transmitting the packet/record. We must be careful
1020 * to cope with writespace callbacks arriving _after_ we have
1021 * called sendmsg(). */
1022 req->rq_xtime = ktime_get();
1023 while (1) {
1024 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
1025 transport->xmit.offset, rm, &sent);
1026
1027 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
1028 xdr->len - transport->xmit.offset, status);
1029
1030 /* If we've sent the entire packet, immediately
1031 * reset the count of bytes sent. */
1032 transport->xmit.offset += sent;
1033 req->rq_bytes_sent = transport->xmit.offset;
1034 if (likely(req->rq_bytes_sent >= msglen)) {
1035 req->rq_xmit_bytes_sent += transport->xmit.offset;
1036 transport->xmit.offset = 0;
1037 return 0;
1038 }
1039
1040 WARN_ON_ONCE(sent == 0 && status == 0);
1041
1042 if (status == -EAGAIN ) {
1043 /*
1044 * Return EAGAIN if we're sure we're hitting the
1045 * socket send buffer limits.
1046 */
1047 if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
1048 break;
1049 /*
1050 * Did we hit a memory allocation failure?
1051 */
1052 if (sent == 0) {
1053 status = -ENOBUFS;
1054 if (vm_wait)
1055 break;
1056 /* Retry, knowing now that we're below the
1057 * socket send buffer limit
1058 */
1059 vm_wait = true;
1060 }
1061 continue;
1062 }
1063 if (status < 0)
1064 break;
1065 vm_wait = false;
1066 }
1067
1068 switch (status) {
1069 case -ENOTSOCK:
1070 status = -ENOTCONN;
1071 /* Should we call xs_close() here? */
1072 break;
1073 case -EAGAIN:
1074 status = xs_nospace(req);
1075 break;
1076 case -ECONNRESET:
1077 case -ECONNREFUSED:
1078 case -ENOTCONN:
1079 case -EADDRINUSE:
1080 case -ENOBUFS:
1081 case -EPIPE:
1082 break;
1083 default:
1084 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1085 -status);
1086 }
1087
1088 return status;
1089}
1090
1091static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1092{
1093 transport->old_data_ready = sk->sk_data_ready;
1094 transport->old_state_change = sk->sk_state_change;
1095 transport->old_write_space = sk->sk_write_space;
1096 transport->old_error_report = sk->sk_error_report;
1097}
1098
1099static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1100{
1101 sk->sk_data_ready = transport->old_data_ready;
1102 sk->sk_state_change = transport->old_state_change;
1103 sk->sk_write_space = transport->old_write_space;
1104 sk->sk_error_report = transport->old_error_report;
1105}
1106
1107static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
1108{
1109 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1110
1111 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1112 clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state);
1113 clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state);
1114 clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state);
1115}
1116
1117static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr)
1118{
1119 set_bit(nr, &transport->sock_state);
1120 queue_work(xprtiod_workqueue, &transport->error_worker);
1121}
1122
1123static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1124{
1125 smp_mb__before_atomic();
1126 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1127 clear_bit(XPRT_CLOSING, &xprt->state);
1128 xs_sock_reset_state_flags(xprt);
1129 smp_mb__after_atomic();
1130}
1131
1132/**
1133 * xs_error_report - callback to handle TCP socket state errors
1134 * @sk: socket
1135 *
1136 * Note: we don't call sock_error() since there may be a rpc_task
1137 * using the socket, and so we don't want to clear sk->sk_err.
1138 */
1139static void xs_error_report(struct sock *sk)
1140{
1141 struct sock_xprt *transport;
1142 struct rpc_xprt *xprt;
1143
1144 read_lock_bh(&sk->sk_callback_lock);
1145 if (!(xprt = xprt_from_sock(sk)))
1146 goto out;
1147
1148 transport = container_of(xprt, struct sock_xprt, xprt);
1149 transport->xprt_err = -sk->sk_err;
1150 if (transport->xprt_err == 0)
1151 goto out;
1152 dprintk("RPC: xs_error_report client %p, error=%d...\n",
1153 xprt, -transport->xprt_err);
1154 trace_rpc_socket_error(xprt, sk->sk_socket, transport->xprt_err);
1155
1156 /* barrier ensures xprt_err is set before XPRT_SOCK_WAKE_ERROR */
1157 smp_mb__before_atomic();
1158 xs_run_error_worker(transport, XPRT_SOCK_WAKE_ERROR);
1159 out:
1160 read_unlock_bh(&sk->sk_callback_lock);
1161}
1162
1163static void xs_reset_transport(struct sock_xprt *transport)
1164{
1165 struct socket *sock = transport->sock;
1166 struct sock *sk = transport->inet;
1167 struct rpc_xprt *xprt = &transport->xprt;
1168 struct file *filp = transport->file;
1169
1170 if (sk == NULL)
1171 return;
1172
1173 if (atomic_read(&transport->xprt.swapper))
1174 sk_clear_memalloc(sk);
1175
1176 kernel_sock_shutdown(sock, SHUT_RDWR);
1177
1178 mutex_lock(&transport->recv_mutex);
1179 write_lock_bh(&sk->sk_callback_lock);
1180 transport->inet = NULL;
1181 transport->sock = NULL;
1182 transport->file = NULL;
1183
1184 sk->sk_user_data = NULL;
1185
1186 xs_restore_old_callbacks(transport, sk);
1187 xprt_clear_connected(xprt);
1188 write_unlock_bh(&sk->sk_callback_lock);
1189 xs_sock_reset_connection_flags(xprt);
1190 /* Reset stream record info */
1191 xs_stream_reset_connect(transport);
1192 mutex_unlock(&transport->recv_mutex);
1193
1194 trace_rpc_socket_close(xprt, sock);
1195 fput(filp);
1196
1197 xprt_disconnect_done(xprt);
1198}
1199
1200/**
1201 * xs_close - close a socket
1202 * @xprt: transport
1203 *
1204 * This is used when all requests are complete; ie, no DRC state remains
1205 * on the server we want to save.
1206 *
1207 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1208 * xs_reset_transport() zeroing the socket from underneath a writer.
1209 */
1210static void xs_close(struct rpc_xprt *xprt)
1211{
1212 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1213
1214 dprintk("RPC: xs_close xprt %p\n", xprt);
1215
1216 xs_reset_transport(transport);
1217 xprt->reestablish_timeout = 0;
1218}
1219
1220static void xs_inject_disconnect(struct rpc_xprt *xprt)
1221{
1222 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
1223 xprt);
1224 xprt_disconnect_done(xprt);
1225}
1226
1227static void xs_xprt_free(struct rpc_xprt *xprt)
1228{
1229 xs_free_peer_addresses(xprt);
1230 xprt_free(xprt);
1231}
1232
1233/**
1234 * xs_destroy - prepare to shutdown a transport
1235 * @xprt: doomed transport
1236 *
1237 */
1238static void xs_destroy(struct rpc_xprt *xprt)
1239{
1240 struct sock_xprt *transport = container_of(xprt,
1241 struct sock_xprt, xprt);
1242 dprintk("RPC: xs_destroy xprt %p\n", xprt);
1243
1244 cancel_delayed_work_sync(&transport->connect_worker);
1245 xs_close(xprt);
1246 cancel_work_sync(&transport->recv_worker);
1247 cancel_work_sync(&transport->error_worker);
1248 xs_xprt_free(xprt);
1249 module_put(THIS_MODULE);
1250}
1251
1252/**
1253 * xs_udp_data_read_skb - receive callback for UDP sockets
1254 * @xprt: transport
1255 * @sk: socket
1256 * @skb: skbuff
1257 *
1258 */
1259static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1260 struct sock *sk,
1261 struct sk_buff *skb)
1262{
1263 struct rpc_task *task;
1264 struct rpc_rqst *rovr;
1265 int repsize, copied;
1266 u32 _xid;
1267 __be32 *xp;
1268
1269 repsize = skb->len;
1270 if (repsize < 4) {
1271 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
1272 return;
1273 }
1274
1275 /* Copy the XID from the skb... */
1276 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1277 if (xp == NULL)
1278 return;
1279
1280 /* Look up and lock the request corresponding to the given XID */
1281 spin_lock(&xprt->queue_lock);
1282 rovr = xprt_lookup_rqst(xprt, *xp);
1283 if (!rovr)
1284 goto out_unlock;
1285 xprt_pin_rqst(rovr);
1286 xprt_update_rtt(rovr->rq_task);
1287 spin_unlock(&xprt->queue_lock);
1288 task = rovr->rq_task;
1289
1290 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1291 copied = repsize;
1292
1293 /* Suck it into the iovec, verify checksum if not done by hw. */
1294 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1295 spin_lock(&xprt->queue_lock);
1296 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1297 goto out_unpin;
1298 }
1299
1300
1301 spin_lock(&xprt->transport_lock);
1302 xprt_adjust_cwnd(xprt, task, copied);
1303 spin_unlock(&xprt->transport_lock);
1304 spin_lock(&xprt->queue_lock);
1305 xprt_complete_rqst(task, copied);
1306 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1307out_unpin:
1308 xprt_unpin_rqst(rovr);
1309 out_unlock:
1310 spin_unlock(&xprt->queue_lock);
1311}
1312
1313static void xs_udp_data_receive(struct sock_xprt *transport)
1314{
1315 struct sk_buff *skb;
1316 struct sock *sk;
1317 int err;
1318
1319 mutex_lock(&transport->recv_mutex);
1320 sk = transport->inet;
1321 if (sk == NULL)
1322 goto out;
1323 for (;;) {
1324 skb = skb_recv_udp(sk, 0, 1, &err);
1325 if (skb == NULL)
1326 break;
1327 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1328 consume_skb(skb);
1329 cond_resched();
1330 }
1331 xs_poll_check_readable(transport);
1332out:
1333 mutex_unlock(&transport->recv_mutex);
1334}
1335
1336static void xs_udp_data_receive_workfn(struct work_struct *work)
1337{
1338 struct sock_xprt *transport =
1339 container_of(work, struct sock_xprt, recv_worker);
1340 unsigned int pflags = memalloc_nofs_save();
1341
1342 xs_udp_data_receive(transport);
1343 memalloc_nofs_restore(pflags);
1344}
1345
1346/**
1347 * xs_data_ready - "data ready" callback for UDP sockets
1348 * @sk: socket with data to read
1349 *
1350 */
1351static void xs_data_ready(struct sock *sk)
1352{
1353 struct rpc_xprt *xprt;
1354
1355 read_lock_bh(&sk->sk_callback_lock);
1356 dprintk("RPC: xs_data_ready...\n");
1357 xprt = xprt_from_sock(sk);
1358 if (xprt != NULL) {
1359 struct sock_xprt *transport = container_of(xprt,
1360 struct sock_xprt, xprt);
1361 transport->old_data_ready(sk);
1362 /* Any data means we had a useful conversation, so
1363 * then we don't need to delay the next reconnect
1364 */
1365 if (xprt->reestablish_timeout)
1366 xprt->reestablish_timeout = 0;
1367 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1368 queue_work(xprtiod_workqueue, &transport->recv_worker);
1369 }
1370 read_unlock_bh(&sk->sk_callback_lock);
1371}
1372
1373/*
1374 * Helper function to force a TCP close if the server is sending
1375 * junk and/or it has put us in CLOSE_WAIT
1376 */
1377static void xs_tcp_force_close(struct rpc_xprt *xprt)
1378{
1379 xprt_force_disconnect(xprt);
1380}
1381
1382#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1383static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1384{
1385 return PAGE_SIZE;
1386}
1387#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1388
1389/**
1390 * xs_tcp_state_change - callback to handle TCP socket state changes
1391 * @sk: socket whose state has changed
1392 *
1393 */
1394static void xs_tcp_state_change(struct sock *sk)
1395{
1396 struct rpc_xprt *xprt;
1397 struct sock_xprt *transport;
1398
1399 read_lock_bh(&sk->sk_callback_lock);
1400 if (!(xprt = xprt_from_sock(sk)))
1401 goto out;
1402 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1403 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1404 sk->sk_state, xprt_connected(xprt),
1405 sock_flag(sk, SOCK_DEAD),
1406 sock_flag(sk, SOCK_ZAPPED),
1407 sk->sk_shutdown);
1408
1409 transport = container_of(xprt, struct sock_xprt, xprt);
1410 trace_rpc_socket_state_change(xprt, sk->sk_socket);
1411 switch (sk->sk_state) {
1412 case TCP_ESTABLISHED:
1413 if (!xprt_test_and_set_connected(xprt)) {
1414 xprt->connect_cookie++;
1415 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1416 xprt_clear_connecting(xprt);
1417
1418 xprt->stat.connect_count++;
1419 xprt->stat.connect_time += (long)jiffies -
1420 xprt->stat.connect_start;
1421 xs_run_error_worker(transport, XPRT_SOCK_WAKE_PENDING);
1422 }
1423 break;
1424 case TCP_FIN_WAIT1:
1425 /* The client initiated a shutdown of the socket */
1426 xprt->connect_cookie++;
1427 xprt->reestablish_timeout = 0;
1428 set_bit(XPRT_CLOSING, &xprt->state);
1429 smp_mb__before_atomic();
1430 clear_bit(XPRT_CONNECTED, &xprt->state);
1431 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1432 smp_mb__after_atomic();
1433 break;
1434 case TCP_CLOSE_WAIT:
1435 /* The server initiated a shutdown of the socket */
1436 xprt->connect_cookie++;
1437 clear_bit(XPRT_CONNECTED, &xprt->state);
1438 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1439 fallthrough;
1440 case TCP_CLOSING:
1441 /*
1442 * If the server closed down the connection, make sure that
1443 * we back off before reconnecting
1444 */
1445 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1446 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1447 break;
1448 case TCP_LAST_ACK:
1449 set_bit(XPRT_CLOSING, &xprt->state);
1450 smp_mb__before_atomic();
1451 clear_bit(XPRT_CONNECTED, &xprt->state);
1452 smp_mb__after_atomic();
1453 break;
1454 case TCP_CLOSE:
1455 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1456 &transport->sock_state))
1457 xprt_clear_connecting(xprt);
1458 clear_bit(XPRT_CLOSING, &xprt->state);
1459 /* Trigger the socket release */
1460 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1461 }
1462 out:
1463 read_unlock_bh(&sk->sk_callback_lock);
1464}
1465
1466static void xs_write_space(struct sock *sk)
1467{
1468 struct socket_wq *wq;
1469 struct sock_xprt *transport;
1470 struct rpc_xprt *xprt;
1471
1472 if (!sk->sk_socket)
1473 return;
1474 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1475
1476 if (unlikely(!(xprt = xprt_from_sock(sk))))
1477 return;
1478 transport = container_of(xprt, struct sock_xprt, xprt);
1479 rcu_read_lock();
1480 wq = rcu_dereference(sk->sk_wq);
1481 if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1482 goto out;
1483
1484 xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE);
1485 sk->sk_write_pending--;
1486out:
1487 rcu_read_unlock();
1488}
1489
1490/**
1491 * xs_udp_write_space - callback invoked when socket buffer space
1492 * becomes available
1493 * @sk: socket whose state has changed
1494 *
1495 * Called when more output buffer space is available for this socket.
1496 * We try not to wake our writers until they can make "significant"
1497 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1498 * with a bunch of small requests.
1499 */
1500static void xs_udp_write_space(struct sock *sk)
1501{
1502 read_lock_bh(&sk->sk_callback_lock);
1503
1504 /* from net/core/sock.c:sock_def_write_space */
1505 if (sock_writeable(sk))
1506 xs_write_space(sk);
1507
1508 read_unlock_bh(&sk->sk_callback_lock);
1509}
1510
1511/**
1512 * xs_tcp_write_space - callback invoked when socket buffer space
1513 * becomes available
1514 * @sk: socket whose state has changed
1515 *
1516 * Called when more output buffer space is available for this socket.
1517 * We try not to wake our writers until they can make "significant"
1518 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1519 * with a bunch of small requests.
1520 */
1521static void xs_tcp_write_space(struct sock *sk)
1522{
1523 read_lock_bh(&sk->sk_callback_lock);
1524
1525 /* from net/core/stream.c:sk_stream_write_space */
1526 if (sk_stream_is_writeable(sk))
1527 xs_write_space(sk);
1528
1529 read_unlock_bh(&sk->sk_callback_lock);
1530}
1531
1532static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1533{
1534 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1535 struct sock *sk = transport->inet;
1536
1537 if (transport->rcvsize) {
1538 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1539 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1540 }
1541 if (transport->sndsize) {
1542 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1543 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1544 sk->sk_write_space(sk);
1545 }
1546}
1547
1548/**
1549 * xs_udp_set_buffer_size - set send and receive limits
1550 * @xprt: generic transport
1551 * @sndsize: requested size of send buffer, in bytes
1552 * @rcvsize: requested size of receive buffer, in bytes
1553 *
1554 * Set socket send and receive buffer size limits.
1555 */
1556static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1557{
1558 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1559
1560 transport->sndsize = 0;
1561 if (sndsize)
1562 transport->sndsize = sndsize + 1024;
1563 transport->rcvsize = 0;
1564 if (rcvsize)
1565 transport->rcvsize = rcvsize + 1024;
1566
1567 xs_udp_do_set_buffer_size(xprt);
1568}
1569
1570/**
1571 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1572 * @xprt: controlling transport
1573 * @task: task that timed out
1574 *
1575 * Adjust the congestion window after a retransmit timeout has occurred.
1576 */
1577static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1578{
1579 spin_lock(&xprt->transport_lock);
1580 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1581 spin_unlock(&xprt->transport_lock);
1582}
1583
1584static int xs_get_random_port(void)
1585{
1586 unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1587 unsigned short range;
1588 unsigned short rand;
1589
1590 if (max < min)
1591 return -EADDRINUSE;
1592 range = max - min + 1;
1593 rand = (unsigned short) prandom_u32() % range;
1594 return rand + min;
1595}
1596
1597static unsigned short xs_sock_getport(struct socket *sock)
1598{
1599 struct sockaddr_storage buf;
1600 unsigned short port = 0;
1601
1602 if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1603 goto out;
1604 switch (buf.ss_family) {
1605 case AF_INET6:
1606 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1607 break;
1608 case AF_INET:
1609 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1610 }
1611out:
1612 return port;
1613}
1614
1615/**
1616 * xs_set_port - reset the port number in the remote endpoint address
1617 * @xprt: generic transport
1618 * @port: new port number
1619 *
1620 */
1621static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1622{
1623 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1624
1625 rpc_set_port(xs_addr(xprt), port);
1626 xs_update_peer_port(xprt);
1627}
1628
1629static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1630{
1631 if (transport->srcport == 0 && transport->xprt.reuseport)
1632 transport->srcport = xs_sock_getport(sock);
1633}
1634
1635static int xs_get_srcport(struct sock_xprt *transport)
1636{
1637 int port = transport->srcport;
1638
1639 if (port == 0 && transport->xprt.resvport)
1640 port = xs_get_random_port();
1641 return port;
1642}
1643
1644static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1645{
1646 if (transport->srcport != 0)
1647 transport->srcport = 0;
1648 if (!transport->xprt.resvport)
1649 return 0;
1650 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1651 return xprt_max_resvport;
1652 return --port;
1653}
1654static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1655{
1656 struct sockaddr_storage myaddr;
1657 int err, nloop = 0;
1658 int port = xs_get_srcport(transport);
1659 unsigned short last;
1660
1661 /*
1662 * If we are asking for any ephemeral port (i.e. port == 0 &&
1663 * transport->xprt.resvport == 0), don't bind. Let the local
1664 * port selection happen implicitly when the socket is used
1665 * (for example at connect time).
1666 *
1667 * This ensures that we can continue to establish TCP
1668 * connections even when all local ephemeral ports are already
1669 * a part of some TCP connection. This makes no difference
1670 * for UDP sockets, but also doens't harm them.
1671 *
1672 * If we're asking for any reserved port (i.e. port == 0 &&
1673 * transport->xprt.resvport == 1) xs_get_srcport above will
1674 * ensure that port is non-zero and we will bind as needed.
1675 */
1676 if (port <= 0)
1677 return port;
1678
1679 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1680 do {
1681 rpc_set_port((struct sockaddr *)&myaddr, port);
1682 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1683 transport->xprt.addrlen);
1684 if (err == 0) {
1685 transport->srcport = port;
1686 break;
1687 }
1688 last = port;
1689 port = xs_next_srcport(transport, port);
1690 if (port > last)
1691 nloop++;
1692 } while (err == -EADDRINUSE && nloop != 2);
1693
1694 if (myaddr.ss_family == AF_INET)
1695 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1696 &((struct sockaddr_in *)&myaddr)->sin_addr,
1697 port, err ? "failed" : "ok", err);
1698 else
1699 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1700 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1701 port, err ? "failed" : "ok", err);
1702 return err;
1703}
1704
1705/*
1706 * We don't support autobind on AF_LOCAL sockets
1707 */
1708static void xs_local_rpcbind(struct rpc_task *task)
1709{
1710 xprt_set_bound(task->tk_xprt);
1711}
1712
1713static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1714{
1715}
1716
1717#ifdef CONFIG_DEBUG_LOCK_ALLOC
1718static struct lock_class_key xs_key[2];
1719static struct lock_class_key xs_slock_key[2];
1720
1721static inline void xs_reclassify_socketu(struct socket *sock)
1722{
1723 struct sock *sk = sock->sk;
1724
1725 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1726 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1727}
1728
1729static inline void xs_reclassify_socket4(struct socket *sock)
1730{
1731 struct sock *sk = sock->sk;
1732
1733 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1734 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1735}
1736
1737static inline void xs_reclassify_socket6(struct socket *sock)
1738{
1739 struct sock *sk = sock->sk;
1740
1741 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1742 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1743}
1744
1745static inline void xs_reclassify_socket(int family, struct socket *sock)
1746{
1747 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1748 return;
1749
1750 switch (family) {
1751 case AF_LOCAL:
1752 xs_reclassify_socketu(sock);
1753 break;
1754 case AF_INET:
1755 xs_reclassify_socket4(sock);
1756 break;
1757 case AF_INET6:
1758 xs_reclassify_socket6(sock);
1759 break;
1760 }
1761}
1762#else
1763static inline void xs_reclassify_socket(int family, struct socket *sock)
1764{
1765}
1766#endif
1767
1768static void xs_dummy_setup_socket(struct work_struct *work)
1769{
1770}
1771
1772static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1773 struct sock_xprt *transport, int family, int type,
1774 int protocol, bool reuseport)
1775{
1776 struct file *filp;
1777 struct socket *sock;
1778 int err;
1779
1780 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1781 if (err < 0) {
1782 dprintk("RPC: can't create %d transport socket (%d).\n",
1783 protocol, -err);
1784 goto out;
1785 }
1786 xs_reclassify_socket(family, sock);
1787
1788 if (reuseport)
1789 sock_set_reuseport(sock->sk);
1790
1791 err = xs_bind(transport, sock);
1792 if (err) {
1793 sock_release(sock);
1794 goto out;
1795 }
1796
1797 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1798 if (IS_ERR(filp))
1799 return ERR_CAST(filp);
1800 transport->file = filp;
1801
1802 return sock;
1803out:
1804 return ERR_PTR(err);
1805}
1806
1807static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1808 struct socket *sock)
1809{
1810 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1811 xprt);
1812
1813 if (!transport->inet) {
1814 struct sock *sk = sock->sk;
1815
1816 write_lock_bh(&sk->sk_callback_lock);
1817
1818 xs_save_old_callbacks(transport, sk);
1819
1820 sk->sk_user_data = xprt;
1821 sk->sk_data_ready = xs_data_ready;
1822 sk->sk_write_space = xs_udp_write_space;
1823 sock_set_flag(sk, SOCK_FASYNC);
1824 sk->sk_error_report = xs_error_report;
1825
1826 xprt_clear_connected(xprt);
1827
1828 /* Reset to new socket */
1829 transport->sock = sock;
1830 transport->inet = sk;
1831
1832 write_unlock_bh(&sk->sk_callback_lock);
1833 }
1834
1835 xs_stream_start_connect(transport);
1836
1837 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1838}
1839
1840/**
1841 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1842 * @transport: socket transport to connect
1843 */
1844static int xs_local_setup_socket(struct sock_xprt *transport)
1845{
1846 struct rpc_xprt *xprt = &transport->xprt;
1847 struct file *filp;
1848 struct socket *sock;
1849 int status;
1850
1851 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1852 SOCK_STREAM, 0, &sock, 1);
1853 if (status < 0) {
1854 dprintk("RPC: can't create AF_LOCAL "
1855 "transport socket (%d).\n", -status);
1856 goto out;
1857 }
1858 xs_reclassify_socket(AF_LOCAL, sock);
1859
1860 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1861 if (IS_ERR(filp)) {
1862 status = PTR_ERR(filp);
1863 goto out;
1864 }
1865 transport->file = filp;
1866
1867 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1868 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1869
1870 status = xs_local_finish_connecting(xprt, sock);
1871 trace_rpc_socket_connect(xprt, sock, status);
1872 switch (status) {
1873 case 0:
1874 dprintk("RPC: xprt %p connected to %s\n",
1875 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1876 xprt->stat.connect_count++;
1877 xprt->stat.connect_time += (long)jiffies -
1878 xprt->stat.connect_start;
1879 xprt_set_connected(xprt);
1880 case -ENOBUFS:
1881 break;
1882 case -ENOENT:
1883 dprintk("RPC: xprt %p: socket %s does not exist\n",
1884 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1885 break;
1886 case -ECONNREFUSED:
1887 dprintk("RPC: xprt %p: connection refused for %s\n",
1888 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1889 break;
1890 default:
1891 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1892 __func__, -status,
1893 xprt->address_strings[RPC_DISPLAY_ADDR]);
1894 }
1895
1896out:
1897 xprt_clear_connecting(xprt);
1898 xprt_wake_pending_tasks(xprt, status);
1899 return status;
1900}
1901
1902static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1903{
1904 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1905 int ret;
1906
1907 if (RPC_IS_ASYNC(task)) {
1908 /*
1909 * We want the AF_LOCAL connect to be resolved in the
1910 * filesystem namespace of the process making the rpc
1911 * call. Thus we connect synchronously.
1912 *
1913 * If we want to support asynchronous AF_LOCAL calls,
1914 * we'll need to figure out how to pass a namespace to
1915 * connect.
1916 */
1917 task->tk_rpc_status = -ENOTCONN;
1918 rpc_exit(task, -ENOTCONN);
1919 return;
1920 }
1921 ret = xs_local_setup_socket(transport);
1922 if (ret && !RPC_IS_SOFTCONN(task))
1923 msleep_interruptible(15000);
1924}
1925
1926#if IS_ENABLED(CONFIG_SUNRPC_SWAP)
1927/*
1928 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
1929 * know that we have exclusive access to the socket), to guard against
1930 * races with xs_reset_transport.
1931 */
1932static void xs_set_memalloc(struct rpc_xprt *xprt)
1933{
1934 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1935 xprt);
1936
1937 /*
1938 * If there's no sock, then we have nothing to set. The
1939 * reconnecting process will get it for us.
1940 */
1941 if (!transport->inet)
1942 return;
1943 if (atomic_read(&xprt->swapper))
1944 sk_set_memalloc(transport->inet);
1945}
1946
1947/**
1948 * xs_enable_swap - Tag this transport as being used for swap.
1949 * @xprt: transport to tag
1950 *
1951 * Take a reference to this transport on behalf of the rpc_clnt, and
1952 * optionally mark it for swapping if it wasn't already.
1953 */
1954static int
1955xs_enable_swap(struct rpc_xprt *xprt)
1956{
1957 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
1958
1959 if (atomic_inc_return(&xprt->swapper) != 1)
1960 return 0;
1961 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
1962 return -ERESTARTSYS;
1963 if (xs->inet)
1964 sk_set_memalloc(xs->inet);
1965 xprt_release_xprt(xprt, NULL);
1966 return 0;
1967}
1968
1969/**
1970 * xs_disable_swap - Untag this transport as being used for swap.
1971 * @xprt: transport to tag
1972 *
1973 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
1974 * swapper refcount goes to 0, untag the socket as a memalloc socket.
1975 */
1976static void
1977xs_disable_swap(struct rpc_xprt *xprt)
1978{
1979 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
1980
1981 if (!atomic_dec_and_test(&xprt->swapper))
1982 return;
1983 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
1984 return;
1985 if (xs->inet)
1986 sk_clear_memalloc(xs->inet);
1987 xprt_release_xprt(xprt, NULL);
1988}
1989#else
1990static void xs_set_memalloc(struct rpc_xprt *xprt)
1991{
1992}
1993
1994static int
1995xs_enable_swap(struct rpc_xprt *xprt)
1996{
1997 return -EINVAL;
1998}
1999
2000static void
2001xs_disable_swap(struct rpc_xprt *xprt)
2002{
2003}
2004#endif
2005
2006static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2007{
2008 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2009
2010 if (!transport->inet) {
2011 struct sock *sk = sock->sk;
2012
2013 write_lock_bh(&sk->sk_callback_lock);
2014
2015 xs_save_old_callbacks(transport, sk);
2016
2017 sk->sk_user_data = xprt;
2018 sk->sk_data_ready = xs_data_ready;
2019 sk->sk_write_space = xs_udp_write_space;
2020 sock_set_flag(sk, SOCK_FASYNC);
2021
2022 xprt_set_connected(xprt);
2023
2024 /* Reset to new socket */
2025 transport->sock = sock;
2026 transport->inet = sk;
2027
2028 xs_set_memalloc(xprt);
2029
2030 write_unlock_bh(&sk->sk_callback_lock);
2031 }
2032 xs_udp_do_set_buffer_size(xprt);
2033
2034 xprt->stat.connect_start = jiffies;
2035}
2036
2037static void xs_udp_setup_socket(struct work_struct *work)
2038{
2039 struct sock_xprt *transport =
2040 container_of(work, struct sock_xprt, connect_worker.work);
2041 struct rpc_xprt *xprt = &transport->xprt;
2042 struct socket *sock;
2043 int status = -EIO;
2044
2045 sock = xs_create_sock(xprt, transport,
2046 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2047 IPPROTO_UDP, false);
2048 if (IS_ERR(sock))
2049 goto out;
2050
2051 dprintk("RPC: worker connecting xprt %p via %s to "
2052 "%s (port %s)\n", xprt,
2053 xprt->address_strings[RPC_DISPLAY_PROTO],
2054 xprt->address_strings[RPC_DISPLAY_ADDR],
2055 xprt->address_strings[RPC_DISPLAY_PORT]);
2056
2057 xs_udp_finish_connecting(xprt, sock);
2058 trace_rpc_socket_connect(xprt, sock, 0);
2059 status = 0;
2060out:
2061 xprt_clear_connecting(xprt);
2062 xprt_unlock_connect(xprt, transport);
2063 xprt_wake_pending_tasks(xprt, status);
2064}
2065
2066/**
2067 * xs_tcp_shutdown - gracefully shut down a TCP socket
2068 * @xprt: transport
2069 *
2070 * Initiates a graceful shutdown of the TCP socket by calling the
2071 * equivalent of shutdown(SHUT_RDWR);
2072 */
2073static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2074{
2075 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2076 struct socket *sock = transport->sock;
2077 int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2078
2079 if (sock == NULL)
2080 return;
2081 switch (skst) {
2082 default:
2083 kernel_sock_shutdown(sock, SHUT_RDWR);
2084 trace_rpc_socket_shutdown(xprt, sock);
2085 break;
2086 case TCP_CLOSE:
2087 case TCP_TIME_WAIT:
2088 xs_reset_transport(transport);
2089 }
2090}
2091
2092static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2093 struct socket *sock)
2094{
2095 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2096 unsigned int keepidle;
2097 unsigned int keepcnt;
2098 unsigned int timeo;
2099
2100 spin_lock(&xprt->transport_lock);
2101 keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2102 keepcnt = xprt->timeout->to_retries + 1;
2103 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2104 (xprt->timeout->to_retries + 1);
2105 clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2106 spin_unlock(&xprt->transport_lock);
2107
2108 /* TCP Keepalive options */
2109 sock_set_keepalive(sock->sk);
2110 tcp_sock_set_keepidle(sock->sk, keepidle);
2111 tcp_sock_set_keepintvl(sock->sk, keepidle);
2112 tcp_sock_set_keepcnt(sock->sk, keepcnt);
2113
2114 /* TCP user timeout (see RFC5482) */
2115 tcp_sock_set_user_timeout(sock->sk, timeo);
2116}
2117
2118static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2119 unsigned long connect_timeout,
2120 unsigned long reconnect_timeout)
2121{
2122 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2123 struct rpc_timeout to;
2124 unsigned long initval;
2125
2126 spin_lock(&xprt->transport_lock);
2127 if (reconnect_timeout < xprt->max_reconnect_timeout)
2128 xprt->max_reconnect_timeout = reconnect_timeout;
2129 if (connect_timeout < xprt->connect_timeout) {
2130 memcpy(&to, xprt->timeout, sizeof(to));
2131 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2132 /* Arbitrary lower limit */
2133 if (initval < XS_TCP_INIT_REEST_TO << 1)
2134 initval = XS_TCP_INIT_REEST_TO << 1;
2135 to.to_initval = initval;
2136 to.to_maxval = initval;
2137 memcpy(&transport->tcp_timeout, &to,
2138 sizeof(transport->tcp_timeout));
2139 xprt->timeout = &transport->tcp_timeout;
2140 xprt->connect_timeout = connect_timeout;
2141 }
2142 set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2143 spin_unlock(&xprt->transport_lock);
2144}
2145
2146static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2147{
2148 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2149 int ret = -ENOTCONN;
2150
2151 if (!transport->inet) {
2152 struct sock *sk = sock->sk;
2153
2154 /* Avoid temporary address, they are bad for long-lived
2155 * connections such as NFS mounts.
2156 * RFC4941, section 3.6 suggests that:
2157 * Individual applications, which have specific
2158 * knowledge about the normal duration of connections,
2159 * MAY override this as appropriate.
2160 */
2161 if (xs_addr(xprt)->sa_family == PF_INET6) {
2162 ip6_sock_set_addr_preferences(sk,
2163 IPV6_PREFER_SRC_PUBLIC);
2164 }
2165
2166 xs_tcp_set_socket_timeouts(xprt, sock);
2167
2168 write_lock_bh(&sk->sk_callback_lock);
2169
2170 xs_save_old_callbacks(transport, sk);
2171
2172 sk->sk_user_data = xprt;
2173 sk->sk_data_ready = xs_data_ready;
2174 sk->sk_state_change = xs_tcp_state_change;
2175 sk->sk_write_space = xs_tcp_write_space;
2176 sock_set_flag(sk, SOCK_FASYNC);
2177 sk->sk_error_report = xs_error_report;
2178
2179 /* socket options */
2180 sock_reset_flag(sk, SOCK_LINGER);
2181 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2182
2183 xprt_clear_connected(xprt);
2184
2185 /* Reset to new socket */
2186 transport->sock = sock;
2187 transport->inet = sk;
2188
2189 write_unlock_bh(&sk->sk_callback_lock);
2190 }
2191
2192 if (!xprt_bound(xprt))
2193 goto out;
2194
2195 xs_set_memalloc(xprt);
2196
2197 xs_stream_start_connect(transport);
2198
2199 /* Tell the socket layer to start connecting... */
2200 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2201 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2202 switch (ret) {
2203 case 0:
2204 xs_set_srcport(transport, sock);
2205 fallthrough;
2206 case -EINPROGRESS:
2207 /* SYN_SENT! */
2208 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2209 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2210 break;
2211 case -EADDRNOTAVAIL:
2212 /* Source port number is unavailable. Try a new one! */
2213 transport->srcport = 0;
2214 }
2215out:
2216 return ret;
2217}
2218
2219/**
2220 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2221 * @work: queued work item
2222 *
2223 * Invoked by a work queue tasklet.
2224 */
2225static void xs_tcp_setup_socket(struct work_struct *work)
2226{
2227 struct sock_xprt *transport =
2228 container_of(work, struct sock_xprt, connect_worker.work);
2229 struct socket *sock = transport->sock;
2230 struct rpc_xprt *xprt = &transport->xprt;
2231 int status = -EIO;
2232
2233 if (!sock) {
2234 sock = xs_create_sock(xprt, transport,
2235 xs_addr(xprt)->sa_family, SOCK_STREAM,
2236 IPPROTO_TCP, true);
2237 if (IS_ERR(sock)) {
2238 status = PTR_ERR(sock);
2239 goto out;
2240 }
2241 }
2242
2243 dprintk("RPC: worker connecting xprt %p via %s to "
2244 "%s (port %s)\n", xprt,
2245 xprt->address_strings[RPC_DISPLAY_PROTO],
2246 xprt->address_strings[RPC_DISPLAY_ADDR],
2247 xprt->address_strings[RPC_DISPLAY_PORT]);
2248
2249 status = xs_tcp_finish_connecting(xprt, sock);
2250 trace_rpc_socket_connect(xprt, sock, status);
2251 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2252 xprt, -status, xprt_connected(xprt),
2253 sock->sk->sk_state);
2254 switch (status) {
2255 default:
2256 printk("%s: connect returned unhandled error %d\n",
2257 __func__, status);
2258 fallthrough;
2259 case -EADDRNOTAVAIL:
2260 /* We're probably in TIME_WAIT. Get rid of existing socket,
2261 * and retry
2262 */
2263 xs_tcp_force_close(xprt);
2264 break;
2265 case 0:
2266 case -EINPROGRESS:
2267 case -EALREADY:
2268 xprt_unlock_connect(xprt, transport);
2269 return;
2270 case -EINVAL:
2271 /* Happens, for instance, if the user specified a link
2272 * local IPv6 address without a scope-id.
2273 */
2274 case -ECONNREFUSED:
2275 case -ECONNRESET:
2276 case -ENETDOWN:
2277 case -ENETUNREACH:
2278 case -EHOSTUNREACH:
2279 case -EADDRINUSE:
2280 case -ENOBUFS:
2281 /*
2282 * xs_tcp_force_close() wakes tasks with -EIO.
2283 * We need to wake them first to ensure the
2284 * correct error code.
2285 */
2286 xprt_wake_pending_tasks(xprt, status);
2287 xs_tcp_force_close(xprt);
2288 goto out;
2289 }
2290 status = -EAGAIN;
2291out:
2292 xprt_clear_connecting(xprt);
2293 xprt_unlock_connect(xprt, transport);
2294 xprt_wake_pending_tasks(xprt, status);
2295}
2296
2297/**
2298 * xs_connect - connect a socket to a remote endpoint
2299 * @xprt: pointer to transport structure
2300 * @task: address of RPC task that manages state of connect request
2301 *
2302 * TCP: If the remote end dropped the connection, delay reconnecting.
2303 *
2304 * UDP socket connects are synchronous, but we use a work queue anyway
2305 * to guarantee that even unprivileged user processes can set up a
2306 * socket on a privileged port.
2307 *
2308 * If a UDP socket connect fails, the delay behavior here prevents
2309 * retry floods (hard mounts).
2310 */
2311static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2312{
2313 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2314 unsigned long delay = 0;
2315
2316 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2317
2318 if (transport->sock != NULL) {
2319 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2320 "seconds\n",
2321 xprt, xprt->reestablish_timeout / HZ);
2322
2323 /* Start by resetting any existing state */
2324 xs_reset_transport(transport);
2325
2326 delay = xprt_reconnect_delay(xprt);
2327 xprt_reconnect_backoff(xprt, XS_TCP_INIT_REEST_TO);
2328
2329 } else
2330 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2331
2332 queue_delayed_work(xprtiod_workqueue,
2333 &transport->connect_worker,
2334 delay);
2335}
2336
2337static void xs_wake_disconnect(struct sock_xprt *transport)
2338{
2339 if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state))
2340 xs_tcp_force_close(&transport->xprt);
2341}
2342
2343static void xs_wake_write(struct sock_xprt *transport)
2344{
2345 if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state))
2346 xprt_write_space(&transport->xprt);
2347}
2348
2349static void xs_wake_error(struct sock_xprt *transport)
2350{
2351 int sockerr;
2352
2353 if (!test_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2354 return;
2355 mutex_lock(&transport->recv_mutex);
2356 if (transport->sock == NULL)
2357 goto out;
2358 if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2359 goto out;
2360 sockerr = xchg(&transport->xprt_err, 0);
2361 if (sockerr < 0)
2362 xprt_wake_pending_tasks(&transport->xprt, sockerr);
2363out:
2364 mutex_unlock(&transport->recv_mutex);
2365}
2366
2367static void xs_wake_pending(struct sock_xprt *transport)
2368{
2369 if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING, &transport->sock_state))
2370 xprt_wake_pending_tasks(&transport->xprt, -EAGAIN);
2371}
2372
2373static void xs_error_handle(struct work_struct *work)
2374{
2375 struct sock_xprt *transport = container_of(work,
2376 struct sock_xprt, error_worker);
2377
2378 xs_wake_disconnect(transport);
2379 xs_wake_write(transport);
2380 xs_wake_error(transport);
2381 xs_wake_pending(transport);
2382}
2383
2384/**
2385 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2386 * @xprt: rpc_xprt struct containing statistics
2387 * @seq: output file
2388 *
2389 */
2390static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2391{
2392 long idle_time = 0;
2393
2394 if (xprt_connected(xprt))
2395 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2396
2397 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2398 "%llu %llu %lu %llu %llu\n",
2399 xprt->stat.bind_count,
2400 xprt->stat.connect_count,
2401 xprt->stat.connect_time / HZ,
2402 idle_time,
2403 xprt->stat.sends,
2404 xprt->stat.recvs,
2405 xprt->stat.bad_xids,
2406 xprt->stat.req_u,
2407 xprt->stat.bklog_u,
2408 xprt->stat.max_slots,
2409 xprt->stat.sending_u,
2410 xprt->stat.pending_u);
2411}
2412
2413/**
2414 * xs_udp_print_stats - display UDP socket-specifc stats
2415 * @xprt: rpc_xprt struct containing statistics
2416 * @seq: output file
2417 *
2418 */
2419static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2420{
2421 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2422
2423 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2424 "%lu %llu %llu\n",
2425 transport->srcport,
2426 xprt->stat.bind_count,
2427 xprt->stat.sends,
2428 xprt->stat.recvs,
2429 xprt->stat.bad_xids,
2430 xprt->stat.req_u,
2431 xprt->stat.bklog_u,
2432 xprt->stat.max_slots,
2433 xprt->stat.sending_u,
2434 xprt->stat.pending_u);
2435}
2436
2437/**
2438 * xs_tcp_print_stats - display TCP socket-specifc stats
2439 * @xprt: rpc_xprt struct containing statistics
2440 * @seq: output file
2441 *
2442 */
2443static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2444{
2445 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2446 long idle_time = 0;
2447
2448 if (xprt_connected(xprt))
2449 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2450
2451 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2452 "%llu %llu %lu %llu %llu\n",
2453 transport->srcport,
2454 xprt->stat.bind_count,
2455 xprt->stat.connect_count,
2456 xprt->stat.connect_time / HZ,
2457 idle_time,
2458 xprt->stat.sends,
2459 xprt->stat.recvs,
2460 xprt->stat.bad_xids,
2461 xprt->stat.req_u,
2462 xprt->stat.bklog_u,
2463 xprt->stat.max_slots,
2464 xprt->stat.sending_u,
2465 xprt->stat.pending_u);
2466}
2467
2468/*
2469 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2470 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2471 * to use the server side send routines.
2472 */
2473static int bc_malloc(struct rpc_task *task)
2474{
2475 struct rpc_rqst *rqst = task->tk_rqstp;
2476 size_t size = rqst->rq_callsize;
2477 struct page *page;
2478 struct rpc_buffer *buf;
2479
2480 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2481 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2482 size);
2483 return -EINVAL;
2484 }
2485
2486 page = alloc_page(GFP_KERNEL);
2487 if (!page)
2488 return -ENOMEM;
2489
2490 buf = page_address(page);
2491 buf->len = PAGE_SIZE;
2492
2493 rqst->rq_buffer = buf->data;
2494 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2495 return 0;
2496}
2497
2498/*
2499 * Free the space allocated in the bc_alloc routine
2500 */
2501static void bc_free(struct rpc_task *task)
2502{
2503 void *buffer = task->tk_rqstp->rq_buffer;
2504 struct rpc_buffer *buf;
2505
2506 buf = container_of(buffer, struct rpc_buffer, data);
2507 free_page((unsigned long)buf);
2508}
2509
2510static int bc_sendto(struct rpc_rqst *req)
2511{
2512 struct xdr_buf *xdr = &req->rq_snd_buf;
2513 struct sock_xprt *transport =
2514 container_of(req->rq_xprt, struct sock_xprt, xprt);
2515 struct msghdr msg = {
2516 .msg_flags = 0,
2517 };
2518 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
2519 (u32)xdr->len);
2520 unsigned int sent = 0;
2521 int err;
2522
2523 req->rq_xtime = ktime_get();
2524 err = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, marker, &sent);
2525 xdr_free_bvec(xdr);
2526 if (err < 0 || sent != (xdr->len + sizeof(marker)))
2527 return -EAGAIN;
2528 return sent;
2529}
2530
2531/**
2532 * bc_send_request - Send a backchannel Call on a TCP socket
2533 * @req: rpc_rqst containing Call message to be sent
2534 *
2535 * xpt_mutex ensures @rqstp's whole message is written to the socket
2536 * without interruption.
2537 *
2538 * Return values:
2539 * %0 if the message was sent successfully
2540 * %ENOTCONN if the message was not sent
2541 */
2542static int bc_send_request(struct rpc_rqst *req)
2543{
2544 struct svc_xprt *xprt;
2545 int len;
2546
2547 /*
2548 * Get the server socket associated with this callback xprt
2549 */
2550 xprt = req->rq_xprt->bc_xprt;
2551
2552 /*
2553 * Grab the mutex to serialize data as the connection is shared
2554 * with the fore channel
2555 */
2556 mutex_lock(&xprt->xpt_mutex);
2557 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2558 len = -ENOTCONN;
2559 else
2560 len = bc_sendto(req);
2561 mutex_unlock(&xprt->xpt_mutex);
2562
2563 if (len > 0)
2564 len = 0;
2565
2566 return len;
2567}
2568
2569/*
2570 * The close routine. Since this is client initiated, we do nothing
2571 */
2572
2573static void bc_close(struct rpc_xprt *xprt)
2574{
2575 xprt_disconnect_done(xprt);
2576}
2577
2578/*
2579 * The xprt destroy routine. Again, because this connection is client
2580 * initiated, we do nothing
2581 */
2582
2583static void bc_destroy(struct rpc_xprt *xprt)
2584{
2585 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2586
2587 xs_xprt_free(xprt);
2588 module_put(THIS_MODULE);
2589}
2590
2591static const struct rpc_xprt_ops xs_local_ops = {
2592 .reserve_xprt = xprt_reserve_xprt,
2593 .release_xprt = xprt_release_xprt,
2594 .alloc_slot = xprt_alloc_slot,
2595 .free_slot = xprt_free_slot,
2596 .rpcbind = xs_local_rpcbind,
2597 .set_port = xs_local_set_port,
2598 .connect = xs_local_connect,
2599 .buf_alloc = rpc_malloc,
2600 .buf_free = rpc_free,
2601 .prepare_request = xs_stream_prepare_request,
2602 .send_request = xs_local_send_request,
2603 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2604 .close = xs_close,
2605 .destroy = xs_destroy,
2606 .print_stats = xs_local_print_stats,
2607 .enable_swap = xs_enable_swap,
2608 .disable_swap = xs_disable_swap,
2609};
2610
2611static const struct rpc_xprt_ops xs_udp_ops = {
2612 .set_buffer_size = xs_udp_set_buffer_size,
2613 .reserve_xprt = xprt_reserve_xprt_cong,
2614 .release_xprt = xprt_release_xprt_cong,
2615 .alloc_slot = xprt_alloc_slot,
2616 .free_slot = xprt_free_slot,
2617 .rpcbind = rpcb_getport_async,
2618 .set_port = xs_set_port,
2619 .connect = xs_connect,
2620 .buf_alloc = rpc_malloc,
2621 .buf_free = rpc_free,
2622 .send_request = xs_udp_send_request,
2623 .wait_for_reply_request = xprt_wait_for_reply_request_rtt,
2624 .timer = xs_udp_timer,
2625 .release_request = xprt_release_rqst_cong,
2626 .close = xs_close,
2627 .destroy = xs_destroy,
2628 .print_stats = xs_udp_print_stats,
2629 .enable_swap = xs_enable_swap,
2630 .disable_swap = xs_disable_swap,
2631 .inject_disconnect = xs_inject_disconnect,
2632};
2633
2634static const struct rpc_xprt_ops xs_tcp_ops = {
2635 .reserve_xprt = xprt_reserve_xprt,
2636 .release_xprt = xprt_release_xprt,
2637 .alloc_slot = xprt_alloc_slot,
2638 .free_slot = xprt_free_slot,
2639 .rpcbind = rpcb_getport_async,
2640 .set_port = xs_set_port,
2641 .connect = xs_connect,
2642 .buf_alloc = rpc_malloc,
2643 .buf_free = rpc_free,
2644 .prepare_request = xs_stream_prepare_request,
2645 .send_request = xs_tcp_send_request,
2646 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2647 .close = xs_tcp_shutdown,
2648 .destroy = xs_destroy,
2649 .set_connect_timeout = xs_tcp_set_connect_timeout,
2650 .print_stats = xs_tcp_print_stats,
2651 .enable_swap = xs_enable_swap,
2652 .disable_swap = xs_disable_swap,
2653 .inject_disconnect = xs_inject_disconnect,
2654#ifdef CONFIG_SUNRPC_BACKCHANNEL
2655 .bc_setup = xprt_setup_bc,
2656 .bc_maxpayload = xs_tcp_bc_maxpayload,
2657 .bc_num_slots = xprt_bc_max_slots,
2658 .bc_free_rqst = xprt_free_bc_rqst,
2659 .bc_destroy = xprt_destroy_bc,
2660#endif
2661};
2662
2663/*
2664 * The rpc_xprt_ops for the server backchannel
2665 */
2666
2667static const struct rpc_xprt_ops bc_tcp_ops = {
2668 .reserve_xprt = xprt_reserve_xprt,
2669 .release_xprt = xprt_release_xprt,
2670 .alloc_slot = xprt_alloc_slot,
2671 .free_slot = xprt_free_slot,
2672 .buf_alloc = bc_malloc,
2673 .buf_free = bc_free,
2674 .send_request = bc_send_request,
2675 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2676 .close = bc_close,
2677 .destroy = bc_destroy,
2678 .print_stats = xs_tcp_print_stats,
2679 .enable_swap = xs_enable_swap,
2680 .disable_swap = xs_disable_swap,
2681 .inject_disconnect = xs_inject_disconnect,
2682};
2683
2684static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2685{
2686 static const struct sockaddr_in sin = {
2687 .sin_family = AF_INET,
2688 .sin_addr.s_addr = htonl(INADDR_ANY),
2689 };
2690 static const struct sockaddr_in6 sin6 = {
2691 .sin6_family = AF_INET6,
2692 .sin6_addr = IN6ADDR_ANY_INIT,
2693 };
2694
2695 switch (family) {
2696 case AF_LOCAL:
2697 break;
2698 case AF_INET:
2699 memcpy(sap, &sin, sizeof(sin));
2700 break;
2701 case AF_INET6:
2702 memcpy(sap, &sin6, sizeof(sin6));
2703 break;
2704 default:
2705 dprintk("RPC: %s: Bad address family\n", __func__);
2706 return -EAFNOSUPPORT;
2707 }
2708 return 0;
2709}
2710
2711static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2712 unsigned int slot_table_size,
2713 unsigned int max_slot_table_size)
2714{
2715 struct rpc_xprt *xprt;
2716 struct sock_xprt *new;
2717
2718 if (args->addrlen > sizeof(xprt->addr)) {
2719 dprintk("RPC: xs_setup_xprt: address too large\n");
2720 return ERR_PTR(-EBADF);
2721 }
2722
2723 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2724 max_slot_table_size);
2725 if (xprt == NULL) {
2726 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2727 "rpc_xprt\n");
2728 return ERR_PTR(-ENOMEM);
2729 }
2730
2731 new = container_of(xprt, struct sock_xprt, xprt);
2732 mutex_init(&new->recv_mutex);
2733 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2734 xprt->addrlen = args->addrlen;
2735 if (args->srcaddr)
2736 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2737 else {
2738 int err;
2739 err = xs_init_anyaddr(args->dstaddr->sa_family,
2740 (struct sockaddr *)&new->srcaddr);
2741 if (err != 0) {
2742 xprt_free(xprt);
2743 return ERR_PTR(err);
2744 }
2745 }
2746
2747 return xprt;
2748}
2749
2750static const struct rpc_timeout xs_local_default_timeout = {
2751 .to_initval = 10 * HZ,
2752 .to_maxval = 10 * HZ,
2753 .to_retries = 2,
2754};
2755
2756/**
2757 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2758 * @args: rpc transport creation arguments
2759 *
2760 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2761 */
2762static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2763{
2764 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2765 struct sock_xprt *transport;
2766 struct rpc_xprt *xprt;
2767 struct rpc_xprt *ret;
2768
2769 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2770 xprt_max_tcp_slot_table_entries);
2771 if (IS_ERR(xprt))
2772 return xprt;
2773 transport = container_of(xprt, struct sock_xprt, xprt);
2774
2775 xprt->prot = 0;
2776 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2777
2778 xprt->bind_timeout = XS_BIND_TO;
2779 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2780 xprt->idle_timeout = XS_IDLE_DISC_TO;
2781
2782 xprt->ops = &xs_local_ops;
2783 xprt->timeout = &xs_local_default_timeout;
2784
2785 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2786 INIT_WORK(&transport->error_worker, xs_error_handle);
2787 INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket);
2788
2789 switch (sun->sun_family) {
2790 case AF_LOCAL:
2791 if (sun->sun_path[0] != '/') {
2792 dprintk("RPC: bad AF_LOCAL address: %s\n",
2793 sun->sun_path);
2794 ret = ERR_PTR(-EINVAL);
2795 goto out_err;
2796 }
2797 xprt_set_bound(xprt);
2798 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2799 ret = ERR_PTR(xs_local_setup_socket(transport));
2800 if (ret)
2801 goto out_err;
2802 break;
2803 default:
2804 ret = ERR_PTR(-EAFNOSUPPORT);
2805 goto out_err;
2806 }
2807
2808 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2809 xprt->address_strings[RPC_DISPLAY_ADDR]);
2810
2811 if (try_module_get(THIS_MODULE))
2812 return xprt;
2813 ret = ERR_PTR(-EINVAL);
2814out_err:
2815 xs_xprt_free(xprt);
2816 return ret;
2817}
2818
2819static const struct rpc_timeout xs_udp_default_timeout = {
2820 .to_initval = 5 * HZ,
2821 .to_maxval = 30 * HZ,
2822 .to_increment = 5 * HZ,
2823 .to_retries = 5,
2824};
2825
2826/**
2827 * xs_setup_udp - Set up transport to use a UDP socket
2828 * @args: rpc transport creation arguments
2829 *
2830 */
2831static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2832{
2833 struct sockaddr *addr = args->dstaddr;
2834 struct rpc_xprt *xprt;
2835 struct sock_xprt *transport;
2836 struct rpc_xprt *ret;
2837
2838 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2839 xprt_udp_slot_table_entries);
2840 if (IS_ERR(xprt))
2841 return xprt;
2842 transport = container_of(xprt, struct sock_xprt, xprt);
2843
2844 xprt->prot = IPPROTO_UDP;
2845 /* XXX: header size can vary due to auth type, IPv6, etc. */
2846 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2847
2848 xprt->bind_timeout = XS_BIND_TO;
2849 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2850 xprt->idle_timeout = XS_IDLE_DISC_TO;
2851
2852 xprt->ops = &xs_udp_ops;
2853
2854 xprt->timeout = &xs_udp_default_timeout;
2855
2856 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
2857 INIT_WORK(&transport->error_worker, xs_error_handle);
2858 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2859
2860 switch (addr->sa_family) {
2861 case AF_INET:
2862 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2863 xprt_set_bound(xprt);
2864
2865 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2866 break;
2867 case AF_INET6:
2868 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2869 xprt_set_bound(xprt);
2870
2871 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2872 break;
2873 default:
2874 ret = ERR_PTR(-EAFNOSUPPORT);
2875 goto out_err;
2876 }
2877
2878 if (xprt_bound(xprt))
2879 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2880 xprt->address_strings[RPC_DISPLAY_ADDR],
2881 xprt->address_strings[RPC_DISPLAY_PORT],
2882 xprt->address_strings[RPC_DISPLAY_PROTO]);
2883 else
2884 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2885 xprt->address_strings[RPC_DISPLAY_ADDR],
2886 xprt->address_strings[RPC_DISPLAY_PROTO]);
2887
2888 if (try_module_get(THIS_MODULE))
2889 return xprt;
2890 ret = ERR_PTR(-EINVAL);
2891out_err:
2892 xs_xprt_free(xprt);
2893 return ret;
2894}
2895
2896static const struct rpc_timeout xs_tcp_default_timeout = {
2897 .to_initval = 60 * HZ,
2898 .to_maxval = 60 * HZ,
2899 .to_retries = 2,
2900};
2901
2902/**
2903 * xs_setup_tcp - Set up transport to use a TCP socket
2904 * @args: rpc transport creation arguments
2905 *
2906 */
2907static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2908{
2909 struct sockaddr *addr = args->dstaddr;
2910 struct rpc_xprt *xprt;
2911 struct sock_xprt *transport;
2912 struct rpc_xprt *ret;
2913 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2914
2915 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2916 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2917
2918 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2919 max_slot_table_size);
2920 if (IS_ERR(xprt))
2921 return xprt;
2922 transport = container_of(xprt, struct sock_xprt, xprt);
2923
2924 xprt->prot = IPPROTO_TCP;
2925 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2926
2927 xprt->bind_timeout = XS_BIND_TO;
2928 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2929 xprt->idle_timeout = XS_IDLE_DISC_TO;
2930
2931 xprt->ops = &xs_tcp_ops;
2932 xprt->timeout = &xs_tcp_default_timeout;
2933
2934 xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
2935 xprt->connect_timeout = xprt->timeout->to_initval *
2936 (xprt->timeout->to_retries + 1);
2937
2938 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2939 INIT_WORK(&transport->error_worker, xs_error_handle);
2940 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
2941
2942 switch (addr->sa_family) {
2943 case AF_INET:
2944 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2945 xprt_set_bound(xprt);
2946
2947 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2948 break;
2949 case AF_INET6:
2950 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2951 xprt_set_bound(xprt);
2952
2953 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2954 break;
2955 default:
2956 ret = ERR_PTR(-EAFNOSUPPORT);
2957 goto out_err;
2958 }
2959
2960 if (xprt_bound(xprt))
2961 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2962 xprt->address_strings[RPC_DISPLAY_ADDR],
2963 xprt->address_strings[RPC_DISPLAY_PORT],
2964 xprt->address_strings[RPC_DISPLAY_PROTO]);
2965 else
2966 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2967 xprt->address_strings[RPC_DISPLAY_ADDR],
2968 xprt->address_strings[RPC_DISPLAY_PROTO]);
2969
2970 if (try_module_get(THIS_MODULE))
2971 return xprt;
2972 ret = ERR_PTR(-EINVAL);
2973out_err:
2974 xs_xprt_free(xprt);
2975 return ret;
2976}
2977
2978/**
2979 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2980 * @args: rpc transport creation arguments
2981 *
2982 */
2983static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2984{
2985 struct sockaddr *addr = args->dstaddr;
2986 struct rpc_xprt *xprt;
2987 struct sock_xprt *transport;
2988 struct svc_sock *bc_sock;
2989 struct rpc_xprt *ret;
2990
2991 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2992 xprt_tcp_slot_table_entries);
2993 if (IS_ERR(xprt))
2994 return xprt;
2995 transport = container_of(xprt, struct sock_xprt, xprt);
2996
2997 xprt->prot = IPPROTO_TCP;
2998 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2999 xprt->timeout = &xs_tcp_default_timeout;
3000
3001 /* backchannel */
3002 xprt_set_bound(xprt);
3003 xprt->bind_timeout = 0;
3004 xprt->reestablish_timeout = 0;
3005 xprt->idle_timeout = 0;
3006
3007 xprt->ops = &bc_tcp_ops;
3008
3009 switch (addr->sa_family) {
3010 case AF_INET:
3011 xs_format_peer_addresses(xprt, "tcp",
3012 RPCBIND_NETID_TCP);
3013 break;
3014 case AF_INET6:
3015 xs_format_peer_addresses(xprt, "tcp",
3016 RPCBIND_NETID_TCP6);
3017 break;
3018 default:
3019 ret = ERR_PTR(-EAFNOSUPPORT);
3020 goto out_err;
3021 }
3022
3023 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3024 xprt->address_strings[RPC_DISPLAY_ADDR],
3025 xprt->address_strings[RPC_DISPLAY_PORT],
3026 xprt->address_strings[RPC_DISPLAY_PROTO]);
3027
3028 /*
3029 * Once we've associated a backchannel xprt with a connection,
3030 * we want to keep it around as long as the connection lasts,
3031 * in case we need to start using it for a backchannel again;
3032 * this reference won't be dropped until bc_xprt is destroyed.
3033 */
3034 xprt_get(xprt);
3035 args->bc_xprt->xpt_bc_xprt = xprt;
3036 xprt->bc_xprt = args->bc_xprt;
3037 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3038 transport->sock = bc_sock->sk_sock;
3039 transport->inet = bc_sock->sk_sk;
3040
3041 /*
3042 * Since we don't want connections for the backchannel, we set
3043 * the xprt status to connected
3044 */
3045 xprt_set_connected(xprt);
3046
3047 if (try_module_get(THIS_MODULE))
3048 return xprt;
3049
3050 args->bc_xprt->xpt_bc_xprt = NULL;
3051 args->bc_xprt->xpt_bc_xps = NULL;
3052 xprt_put(xprt);
3053 ret = ERR_PTR(-EINVAL);
3054out_err:
3055 xs_xprt_free(xprt);
3056 return ret;
3057}
3058
3059static struct xprt_class xs_local_transport = {
3060 .list = LIST_HEAD_INIT(xs_local_transport.list),
3061 .name = "named UNIX socket",
3062 .owner = THIS_MODULE,
3063 .ident = XPRT_TRANSPORT_LOCAL,
3064 .setup = xs_setup_local,
3065};
3066
3067static struct xprt_class xs_udp_transport = {
3068 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3069 .name = "udp",
3070 .owner = THIS_MODULE,
3071 .ident = XPRT_TRANSPORT_UDP,
3072 .setup = xs_setup_udp,
3073};
3074
3075static struct xprt_class xs_tcp_transport = {
3076 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3077 .name = "tcp",
3078 .owner = THIS_MODULE,
3079 .ident = XPRT_TRANSPORT_TCP,
3080 .setup = xs_setup_tcp,
3081};
3082
3083static struct xprt_class xs_bc_tcp_transport = {
3084 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3085 .name = "tcp NFSv4.1 backchannel",
3086 .owner = THIS_MODULE,
3087 .ident = XPRT_TRANSPORT_BC_TCP,
3088 .setup = xs_setup_bc_tcp,
3089};
3090
3091/**
3092 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3093 *
3094 */
3095int init_socket_xprt(void)
3096{
3097 if (!sunrpc_table_header)
3098 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3099
3100 xprt_register_transport(&xs_local_transport);
3101 xprt_register_transport(&xs_udp_transport);
3102 xprt_register_transport(&xs_tcp_transport);
3103 xprt_register_transport(&xs_bc_tcp_transport);
3104
3105 return 0;
3106}
3107
3108/**
3109 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3110 *
3111 */
3112void cleanup_socket_xprt(void)
3113{
3114 if (sunrpc_table_header) {
3115 unregister_sysctl_table(sunrpc_table_header);
3116 sunrpc_table_header = NULL;
3117 }
3118
3119 xprt_unregister_transport(&xs_local_transport);
3120 xprt_unregister_transport(&xs_udp_transport);
3121 xprt_unregister_transport(&xs_tcp_transport);
3122 xprt_unregister_transport(&xs_bc_tcp_transport);
3123}
3124
3125static int param_set_uint_minmax(const char *val,
3126 const struct kernel_param *kp,
3127 unsigned int min, unsigned int max)
3128{
3129 unsigned int num;
3130 int ret;
3131
3132 if (!val)
3133 return -EINVAL;
3134 ret = kstrtouint(val, 0, &num);
3135 if (ret)
3136 return ret;
3137 if (num < min || num > max)
3138 return -EINVAL;
3139 *((unsigned int *)kp->arg) = num;
3140 return 0;
3141}
3142
3143static int param_set_portnr(const char *val, const struct kernel_param *kp)
3144{
3145 return param_set_uint_minmax(val, kp,
3146 RPC_MIN_RESVPORT,
3147 RPC_MAX_RESVPORT);
3148}
3149
3150static const struct kernel_param_ops param_ops_portnr = {
3151 .set = param_set_portnr,
3152 .get = param_get_uint,
3153};
3154
3155#define param_check_portnr(name, p) \
3156 __param_check(name, p, unsigned int);
3157
3158module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3159module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3160
3161static int param_set_slot_table_size(const char *val,
3162 const struct kernel_param *kp)
3163{
3164 return param_set_uint_minmax(val, kp,
3165 RPC_MIN_SLOT_TABLE,
3166 RPC_MAX_SLOT_TABLE);
3167}
3168
3169static const struct kernel_param_ops param_ops_slot_table_size = {
3170 .set = param_set_slot_table_size,
3171 .get = param_get_uint,
3172};
3173
3174#define param_check_slot_table_size(name, p) \
3175 __param_check(name, p, unsigned int);
3176
3177static int param_set_max_slot_table_size(const char *val,
3178 const struct kernel_param *kp)
3179{
3180 return param_set_uint_minmax(val, kp,
3181 RPC_MIN_SLOT_TABLE,
3182 RPC_MAX_SLOT_TABLE_LIMIT);
3183}
3184
3185static const struct kernel_param_ops param_ops_max_slot_table_size = {
3186 .set = param_set_max_slot_table_size,
3187 .get = param_get_uint,
3188};
3189
3190#define param_check_max_slot_table_size(name, p) \
3191 __param_check(name, p, unsigned int);
3192
3193module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3194 slot_table_size, 0644);
3195module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3196 max_slot_table_size, 0644);
3197module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3198 slot_table_size, 0644);
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/net/sunrpc/xprtsock.c
4 *
5 * Client-side transport implementation for sockets.
6 *
7 * TCP callback races fixes (C) 1998 Red Hat
8 * TCP send fixes (C) 1998 Red Hat
9 * TCP NFS related read + write fixes
10 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
11 *
12 * Rewrite of larges part of the code in order to stabilize TCP stuff.
13 * Fix behaviour when socket buffer is full.
14 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
15 *
16 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
17 *
18 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
19 * <gilles.quillard@bull.net>
20 */
21
22#include <linux/types.h>
23#include <linux/string.h>
24#include <linux/slab.h>
25#include <linux/module.h>
26#include <linux/capability.h>
27#include <linux/pagemap.h>
28#include <linux/errno.h>
29#include <linux/socket.h>
30#include <linux/in.h>
31#include <linux/net.h>
32#include <linux/mm.h>
33#include <linux/un.h>
34#include <linux/udp.h>
35#include <linux/tcp.h>
36#include <linux/sunrpc/clnt.h>
37#include <linux/sunrpc/addr.h>
38#include <linux/sunrpc/sched.h>
39#include <linux/sunrpc/svcsock.h>
40#include <linux/sunrpc/xprtsock.h>
41#include <linux/file.h>
42#ifdef CONFIG_SUNRPC_BACKCHANNEL
43#include <linux/sunrpc/bc_xprt.h>
44#endif
45
46#include <net/sock.h>
47#include <net/checksum.h>
48#include <net/udp.h>
49#include <net/tcp.h>
50#include <linux/bvec.h>
51#include <linux/highmem.h>
52#include <linux/uio.h>
53#include <linux/sched/mm.h>
54
55#include <trace/events/sunrpc.h>
56
57#include "sunrpc.h"
58
59static void xs_close(struct rpc_xprt *xprt);
60static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
61 struct socket *sock);
62
63/*
64 * xprtsock tunables
65 */
66static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
67static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
68static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
69
70static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
71static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
72
73#define XS_TCP_LINGER_TO (15U * HZ)
74static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
75
76/*
77 * We can register our own files under /proc/sys/sunrpc by
78 * calling register_sysctl_table() again. The files in that
79 * directory become the union of all files registered there.
80 *
81 * We simply need to make sure that we don't collide with
82 * someone else's file names!
83 */
84
85static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
86static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
87static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
88static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
89static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
90
91static struct ctl_table_header *sunrpc_table_header;
92
93/*
94 * FIXME: changing the UDP slot table size should also resize the UDP
95 * socket buffers for existing UDP transports
96 */
97static struct ctl_table xs_tunables_table[] = {
98 {
99 .procname = "udp_slot_table_entries",
100 .data = &xprt_udp_slot_table_entries,
101 .maxlen = sizeof(unsigned int),
102 .mode = 0644,
103 .proc_handler = proc_dointvec_minmax,
104 .extra1 = &min_slot_table_size,
105 .extra2 = &max_slot_table_size
106 },
107 {
108 .procname = "tcp_slot_table_entries",
109 .data = &xprt_tcp_slot_table_entries,
110 .maxlen = sizeof(unsigned int),
111 .mode = 0644,
112 .proc_handler = proc_dointvec_minmax,
113 .extra1 = &min_slot_table_size,
114 .extra2 = &max_slot_table_size
115 },
116 {
117 .procname = "tcp_max_slot_table_entries",
118 .data = &xprt_max_tcp_slot_table_entries,
119 .maxlen = sizeof(unsigned int),
120 .mode = 0644,
121 .proc_handler = proc_dointvec_minmax,
122 .extra1 = &min_slot_table_size,
123 .extra2 = &max_tcp_slot_table_limit
124 },
125 {
126 .procname = "min_resvport",
127 .data = &xprt_min_resvport,
128 .maxlen = sizeof(unsigned int),
129 .mode = 0644,
130 .proc_handler = proc_dointvec_minmax,
131 .extra1 = &xprt_min_resvport_limit,
132 .extra2 = &xprt_max_resvport_limit
133 },
134 {
135 .procname = "max_resvport",
136 .data = &xprt_max_resvport,
137 .maxlen = sizeof(unsigned int),
138 .mode = 0644,
139 .proc_handler = proc_dointvec_minmax,
140 .extra1 = &xprt_min_resvport_limit,
141 .extra2 = &xprt_max_resvport_limit
142 },
143 {
144 .procname = "tcp_fin_timeout",
145 .data = &xs_tcp_fin_timeout,
146 .maxlen = sizeof(xs_tcp_fin_timeout),
147 .mode = 0644,
148 .proc_handler = proc_dointvec_jiffies,
149 },
150 { },
151};
152
153static struct ctl_table sunrpc_table[] = {
154 {
155 .procname = "sunrpc",
156 .mode = 0555,
157 .child = xs_tunables_table
158 },
159 { },
160};
161
162/*
163 * Wait duration for a reply from the RPC portmapper.
164 */
165#define XS_BIND_TO (60U * HZ)
166
167/*
168 * Delay if a UDP socket connect error occurs. This is most likely some
169 * kind of resource problem on the local host.
170 */
171#define XS_UDP_REEST_TO (2U * HZ)
172
173/*
174 * The reestablish timeout allows clients to delay for a bit before attempting
175 * to reconnect to a server that just dropped our connection.
176 *
177 * We implement an exponential backoff when trying to reestablish a TCP
178 * transport connection with the server. Some servers like to drop a TCP
179 * connection when they are overworked, so we start with a short timeout and
180 * increase over time if the server is down or not responding.
181 */
182#define XS_TCP_INIT_REEST_TO (3U * HZ)
183
184/*
185 * TCP idle timeout; client drops the transport socket if it is idle
186 * for this long. Note that we also timeout UDP sockets to prevent
187 * holding port numbers when there is no RPC traffic.
188 */
189#define XS_IDLE_DISC_TO (5U * 60 * HZ)
190
191#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
192# undef RPC_DEBUG_DATA
193# define RPCDBG_FACILITY RPCDBG_TRANS
194#endif
195
196#ifdef RPC_DEBUG_DATA
197static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
198{
199 u8 *buf = (u8 *) packet;
200 int j;
201
202 dprintk("RPC: %s\n", msg);
203 for (j = 0; j < count && j < 128; j += 4) {
204 if (!(j & 31)) {
205 if (j)
206 dprintk("\n");
207 dprintk("0x%04x ", j);
208 }
209 dprintk("%02x%02x%02x%02x ",
210 buf[j], buf[j+1], buf[j+2], buf[j+3]);
211 }
212 dprintk("\n");
213}
214#else
215static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
216{
217 /* NOP */
218}
219#endif
220
221static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
222{
223 return (struct rpc_xprt *) sk->sk_user_data;
224}
225
226static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
227{
228 return (struct sockaddr *) &xprt->addr;
229}
230
231static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
232{
233 return (struct sockaddr_un *) &xprt->addr;
234}
235
236static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
237{
238 return (struct sockaddr_in *) &xprt->addr;
239}
240
241static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
242{
243 return (struct sockaddr_in6 *) &xprt->addr;
244}
245
246static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
247{
248 struct sockaddr *sap = xs_addr(xprt);
249 struct sockaddr_in6 *sin6;
250 struct sockaddr_in *sin;
251 struct sockaddr_un *sun;
252 char buf[128];
253
254 switch (sap->sa_family) {
255 case AF_LOCAL:
256 sun = xs_addr_un(xprt);
257 strlcpy(buf, sun->sun_path, sizeof(buf));
258 xprt->address_strings[RPC_DISPLAY_ADDR] =
259 kstrdup(buf, GFP_KERNEL);
260 break;
261 case AF_INET:
262 (void)rpc_ntop(sap, buf, sizeof(buf));
263 xprt->address_strings[RPC_DISPLAY_ADDR] =
264 kstrdup(buf, GFP_KERNEL);
265 sin = xs_addr_in(xprt);
266 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
267 break;
268 case AF_INET6:
269 (void)rpc_ntop(sap, buf, sizeof(buf));
270 xprt->address_strings[RPC_DISPLAY_ADDR] =
271 kstrdup(buf, GFP_KERNEL);
272 sin6 = xs_addr_in6(xprt);
273 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
274 break;
275 default:
276 BUG();
277 }
278
279 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
280}
281
282static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
283{
284 struct sockaddr *sap = xs_addr(xprt);
285 char buf[128];
286
287 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
288 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
289
290 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
291 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
292}
293
294static void xs_format_peer_addresses(struct rpc_xprt *xprt,
295 const char *protocol,
296 const char *netid)
297{
298 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
299 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
300 xs_format_common_peer_addresses(xprt);
301 xs_format_common_peer_ports(xprt);
302}
303
304static void xs_update_peer_port(struct rpc_xprt *xprt)
305{
306 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
307 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
308
309 xs_format_common_peer_ports(xprt);
310}
311
312static void xs_free_peer_addresses(struct rpc_xprt *xprt)
313{
314 unsigned int i;
315
316 for (i = 0; i < RPC_DISPLAY_MAX; i++)
317 switch (i) {
318 case RPC_DISPLAY_PROTO:
319 case RPC_DISPLAY_NETID:
320 continue;
321 default:
322 kfree(xprt->address_strings[i]);
323 }
324}
325
326static size_t
327xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp)
328{
329 size_t i,n;
330
331 if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES))
332 return want;
333 n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT;
334 for (i = 0; i < n; i++) {
335 if (buf->pages[i])
336 continue;
337 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp);
338 if (!buf->pages[i]) {
339 i *= PAGE_SIZE;
340 return i > buf->page_base ? i - buf->page_base : 0;
341 }
342 }
343 return want;
344}
345
346static ssize_t
347xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek)
348{
349 ssize_t ret;
350 if (seek != 0)
351 iov_iter_advance(&msg->msg_iter, seek);
352 ret = sock_recvmsg(sock, msg, flags);
353 return ret > 0 ? ret + seek : ret;
354}
355
356static ssize_t
357xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags,
358 struct kvec *kvec, size_t count, size_t seek)
359{
360 iov_iter_kvec(&msg->msg_iter, READ, kvec, 1, count);
361 return xs_sock_recvmsg(sock, msg, flags, seek);
362}
363
364static ssize_t
365xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags,
366 struct bio_vec *bvec, unsigned long nr, size_t count,
367 size_t seek)
368{
369 iov_iter_bvec(&msg->msg_iter, READ, bvec, nr, count);
370 return xs_sock_recvmsg(sock, msg, flags, seek);
371}
372
373static ssize_t
374xs_read_discard(struct socket *sock, struct msghdr *msg, int flags,
375 size_t count)
376{
377 iov_iter_discard(&msg->msg_iter, READ, count);
378 return sock_recvmsg(sock, msg, flags);
379}
380
381#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
382static void
383xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
384{
385 struct bvec_iter bi = {
386 .bi_size = count,
387 };
388 struct bio_vec bv;
389
390 bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
391 for_each_bvec(bv, bvec, bi, bi)
392 flush_dcache_page(bv.bv_page);
393}
394#else
395static inline void
396xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
397{
398}
399#endif
400
401static ssize_t
402xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
403 struct xdr_buf *buf, size_t count, size_t seek, size_t *read)
404{
405 size_t want, seek_init = seek, offset = 0;
406 ssize_t ret;
407
408 want = min_t(size_t, count, buf->head[0].iov_len);
409 if (seek < want) {
410 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek);
411 if (ret <= 0)
412 goto sock_err;
413 offset += ret;
414 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
415 goto out;
416 if (ret != want)
417 goto out;
418 seek = 0;
419 } else {
420 seek -= want;
421 offset += want;
422 }
423
424 want = xs_alloc_sparse_pages(buf,
425 min_t(size_t, count - offset, buf->page_len),
426 GFP_KERNEL);
427 if (seek < want) {
428 ret = xs_read_bvec(sock, msg, flags, buf->bvec,
429 xdr_buf_pagecount(buf),
430 want + buf->page_base,
431 seek + buf->page_base);
432 if (ret <= 0)
433 goto sock_err;
434 xs_flush_bvec(buf->bvec, ret, seek + buf->page_base);
435 offset += ret - buf->page_base;
436 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
437 goto out;
438 if (ret != want)
439 goto out;
440 seek = 0;
441 } else {
442 seek -= want;
443 offset += want;
444 }
445
446 want = min_t(size_t, count - offset, buf->tail[0].iov_len);
447 if (seek < want) {
448 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek);
449 if (ret <= 0)
450 goto sock_err;
451 offset += ret;
452 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
453 goto out;
454 if (ret != want)
455 goto out;
456 } else if (offset < seek_init)
457 offset = seek_init;
458 ret = -EMSGSIZE;
459out:
460 *read = offset - seek_init;
461 return ret;
462sock_err:
463 offset += seek;
464 goto out;
465}
466
467static void
468xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf)
469{
470 if (!transport->recv.copied) {
471 if (buf->head[0].iov_len >= transport->recv.offset)
472 memcpy(buf->head[0].iov_base,
473 &transport->recv.xid,
474 transport->recv.offset);
475 transport->recv.copied = transport->recv.offset;
476 }
477}
478
479static bool
480xs_read_stream_request_done(struct sock_xprt *transport)
481{
482 return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT);
483}
484
485static void
486xs_read_stream_check_eor(struct sock_xprt *transport,
487 struct msghdr *msg)
488{
489 if (xs_read_stream_request_done(transport))
490 msg->msg_flags |= MSG_EOR;
491}
492
493static ssize_t
494xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg,
495 int flags, struct rpc_rqst *req)
496{
497 struct xdr_buf *buf = &req->rq_private_buf;
498 size_t want, uninitialized_var(read);
499 ssize_t uninitialized_var(ret);
500
501 xs_read_header(transport, buf);
502
503 want = transport->recv.len - transport->recv.offset;
504 if (want != 0) {
505 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf,
506 transport->recv.copied + want,
507 transport->recv.copied,
508 &read);
509 transport->recv.offset += read;
510 transport->recv.copied += read;
511 }
512
513 if (transport->recv.offset == transport->recv.len)
514 xs_read_stream_check_eor(transport, msg);
515
516 if (want == 0)
517 return 0;
518
519 switch (ret) {
520 default:
521 break;
522 case -EFAULT:
523 case -EMSGSIZE:
524 msg->msg_flags |= MSG_TRUNC;
525 return read;
526 case 0:
527 return -ESHUTDOWN;
528 }
529 return ret < 0 ? ret : read;
530}
531
532static size_t
533xs_read_stream_headersize(bool isfrag)
534{
535 if (isfrag)
536 return sizeof(__be32);
537 return 3 * sizeof(__be32);
538}
539
540static ssize_t
541xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg,
542 int flags, size_t want, size_t seek)
543{
544 struct kvec kvec = {
545 .iov_base = &transport->recv.fraghdr,
546 .iov_len = want,
547 };
548 return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek);
549}
550
551#if defined(CONFIG_SUNRPC_BACKCHANNEL)
552static ssize_t
553xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
554{
555 struct rpc_xprt *xprt = &transport->xprt;
556 struct rpc_rqst *req;
557 ssize_t ret;
558
559 /* Look up and lock the request corresponding to the given XID */
560 req = xprt_lookup_bc_request(xprt, transport->recv.xid);
561 if (!req) {
562 printk(KERN_WARNING "Callback slot table overflowed\n");
563 return -ESHUTDOWN;
564 }
565 if (transport->recv.copied && !req->rq_private_buf.len)
566 return -ESHUTDOWN;
567
568 ret = xs_read_stream_request(transport, msg, flags, req);
569 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
570 xprt_complete_bc_request(req, transport->recv.copied);
571 else
572 req->rq_private_buf.len = transport->recv.copied;
573
574 return ret;
575}
576#else /* CONFIG_SUNRPC_BACKCHANNEL */
577static ssize_t
578xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
579{
580 return -ESHUTDOWN;
581}
582#endif /* CONFIG_SUNRPC_BACKCHANNEL */
583
584static ssize_t
585xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags)
586{
587 struct rpc_xprt *xprt = &transport->xprt;
588 struct rpc_rqst *req;
589 ssize_t ret = 0;
590
591 /* Look up and lock the request corresponding to the given XID */
592 spin_lock(&xprt->queue_lock);
593 req = xprt_lookup_rqst(xprt, transport->recv.xid);
594 if (!req || (transport->recv.copied && !req->rq_private_buf.len)) {
595 msg->msg_flags |= MSG_TRUNC;
596 goto out;
597 }
598 xprt_pin_rqst(req);
599 spin_unlock(&xprt->queue_lock);
600
601 ret = xs_read_stream_request(transport, msg, flags, req);
602
603 spin_lock(&xprt->queue_lock);
604 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
605 xprt_complete_rqst(req->rq_task, transport->recv.copied);
606 else
607 req->rq_private_buf.len = transport->recv.copied;
608 xprt_unpin_rqst(req);
609out:
610 spin_unlock(&xprt->queue_lock);
611 return ret;
612}
613
614static ssize_t
615xs_read_stream(struct sock_xprt *transport, int flags)
616{
617 struct msghdr msg = { 0 };
618 size_t want, read = 0;
619 ssize_t ret = 0;
620
621 if (transport->recv.len == 0) {
622 want = xs_read_stream_headersize(transport->recv.copied != 0);
623 ret = xs_read_stream_header(transport, &msg, flags, want,
624 transport->recv.offset);
625 if (ret <= 0)
626 goto out_err;
627 transport->recv.offset = ret;
628 if (transport->recv.offset != want)
629 return transport->recv.offset;
630 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) &
631 RPC_FRAGMENT_SIZE_MASK;
632 transport->recv.offset -= sizeof(transport->recv.fraghdr);
633 read = ret;
634 }
635
636 switch (be32_to_cpu(transport->recv.calldir)) {
637 default:
638 msg.msg_flags |= MSG_TRUNC;
639 break;
640 case RPC_CALL:
641 ret = xs_read_stream_call(transport, &msg, flags);
642 break;
643 case RPC_REPLY:
644 ret = xs_read_stream_reply(transport, &msg, flags);
645 }
646 if (msg.msg_flags & MSG_TRUNC) {
647 transport->recv.calldir = cpu_to_be32(-1);
648 transport->recv.copied = -1;
649 }
650 if (ret < 0)
651 goto out_err;
652 read += ret;
653 if (transport->recv.offset < transport->recv.len) {
654 if (!(msg.msg_flags & MSG_TRUNC))
655 return read;
656 msg.msg_flags = 0;
657 ret = xs_read_discard(transport->sock, &msg, flags,
658 transport->recv.len - transport->recv.offset);
659 if (ret <= 0)
660 goto out_err;
661 transport->recv.offset += ret;
662 read += ret;
663 if (transport->recv.offset != transport->recv.len)
664 return read;
665 }
666 if (xs_read_stream_request_done(transport)) {
667 trace_xs_stream_read_request(transport);
668 transport->recv.copied = 0;
669 }
670 transport->recv.offset = 0;
671 transport->recv.len = 0;
672 return read;
673out_err:
674 return ret != 0 ? ret : -ESHUTDOWN;
675}
676
677static __poll_t xs_poll_socket(struct sock_xprt *transport)
678{
679 return transport->sock->ops->poll(transport->file, transport->sock,
680 NULL);
681}
682
683static bool xs_poll_socket_readable(struct sock_xprt *transport)
684{
685 __poll_t events = xs_poll_socket(transport);
686
687 return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP);
688}
689
690static void xs_poll_check_readable(struct sock_xprt *transport)
691{
692
693 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
694 if (!xs_poll_socket_readable(transport))
695 return;
696 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
697 queue_work(xprtiod_workqueue, &transport->recv_worker);
698}
699
700static void xs_stream_data_receive(struct sock_xprt *transport)
701{
702 size_t read = 0;
703 ssize_t ret = 0;
704
705 mutex_lock(&transport->recv_mutex);
706 if (transport->sock == NULL)
707 goto out;
708 for (;;) {
709 ret = xs_read_stream(transport, MSG_DONTWAIT);
710 if (ret < 0)
711 break;
712 read += ret;
713 cond_resched();
714 }
715 if (ret == -ESHUTDOWN)
716 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
717 else
718 xs_poll_check_readable(transport);
719out:
720 mutex_unlock(&transport->recv_mutex);
721 trace_xs_stream_read_data(&transport->xprt, ret, read);
722}
723
724static void xs_stream_data_receive_workfn(struct work_struct *work)
725{
726 struct sock_xprt *transport =
727 container_of(work, struct sock_xprt, recv_worker);
728 unsigned int pflags = memalloc_nofs_save();
729
730 xs_stream_data_receive(transport);
731 memalloc_nofs_restore(pflags);
732}
733
734static void
735xs_stream_reset_connect(struct sock_xprt *transport)
736{
737 transport->recv.offset = 0;
738 transport->recv.len = 0;
739 transport->recv.copied = 0;
740 transport->xmit.offset = 0;
741}
742
743static void
744xs_stream_start_connect(struct sock_xprt *transport)
745{
746 transport->xprt.stat.connect_count++;
747 transport->xprt.stat.connect_start = jiffies;
748}
749
750#define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
751
752static int xs_sendmsg(struct socket *sock, struct msghdr *msg, size_t seek)
753{
754 if (seek)
755 iov_iter_advance(&msg->msg_iter, seek);
756 return sock_sendmsg(sock, msg);
757}
758
759static int xs_send_kvec(struct socket *sock, struct msghdr *msg, struct kvec *vec, size_t seek)
760{
761 iov_iter_kvec(&msg->msg_iter, WRITE, vec, 1, vec->iov_len);
762 return xs_sendmsg(sock, msg, seek);
763}
764
765static int xs_send_pagedata(struct socket *sock, struct msghdr *msg, struct xdr_buf *xdr, size_t base)
766{
767 int err;
768
769 err = xdr_alloc_bvec(xdr, GFP_KERNEL);
770 if (err < 0)
771 return err;
772
773 iov_iter_bvec(&msg->msg_iter, WRITE, xdr->bvec,
774 xdr_buf_pagecount(xdr),
775 xdr->page_len + xdr->page_base);
776 return xs_sendmsg(sock, msg, base + xdr->page_base);
777}
778
779#define xs_record_marker_len() sizeof(rpc_fraghdr)
780
781/* Common case:
782 * - stream transport
783 * - sending from byte 0 of the message
784 * - the message is wholly contained in @xdr's head iovec
785 */
786static int xs_send_rm_and_kvec(struct socket *sock, struct msghdr *msg,
787 rpc_fraghdr marker, struct kvec *vec, size_t base)
788{
789 struct kvec iov[2] = {
790 [0] = {
791 .iov_base = &marker,
792 .iov_len = sizeof(marker)
793 },
794 [1] = *vec,
795 };
796 size_t len = iov[0].iov_len + iov[1].iov_len;
797
798 iov_iter_kvec(&msg->msg_iter, WRITE, iov, 2, len);
799 return xs_sendmsg(sock, msg, base);
800}
801
802/**
803 * xs_sendpages - write pages directly to a socket
804 * @sock: socket to send on
805 * @addr: UDP only -- address of destination
806 * @addrlen: UDP only -- length of destination address
807 * @xdr: buffer containing this request
808 * @base: starting position in the buffer
809 * @rm: stream record marker field
810 * @sent_p: return the total number of bytes successfully queued for sending
811 *
812 */
813static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, rpc_fraghdr rm, int *sent_p)
814{
815 struct msghdr msg = {
816 .msg_name = addr,
817 .msg_namelen = addrlen,
818 .msg_flags = XS_SENDMSG_FLAGS | MSG_MORE,
819 };
820 unsigned int rmsize = rm ? sizeof(rm) : 0;
821 unsigned int remainder = rmsize + xdr->len - base;
822 unsigned int want;
823 int err = 0;
824
825 if (unlikely(!sock))
826 return -ENOTSOCK;
827
828 want = xdr->head[0].iov_len + rmsize;
829 if (base < want) {
830 unsigned int len = want - base;
831 remainder -= len;
832 if (remainder == 0)
833 msg.msg_flags &= ~MSG_MORE;
834 if (rmsize)
835 err = xs_send_rm_and_kvec(sock, &msg, rm,
836 &xdr->head[0], base);
837 else
838 err = xs_send_kvec(sock, &msg, &xdr->head[0], base);
839 if (remainder == 0 || err != len)
840 goto out;
841 *sent_p += err;
842 base = 0;
843 } else
844 base -= want;
845
846 if (base < xdr->page_len) {
847 unsigned int len = xdr->page_len - base;
848 remainder -= len;
849 if (remainder == 0)
850 msg.msg_flags &= ~MSG_MORE;
851 err = xs_send_pagedata(sock, &msg, xdr, base);
852 if (remainder == 0 || err != len)
853 goto out;
854 *sent_p += err;
855 base = 0;
856 } else
857 base -= xdr->page_len;
858
859 if (base >= xdr->tail[0].iov_len)
860 return 0;
861 msg.msg_flags &= ~MSG_MORE;
862 err = xs_send_kvec(sock, &msg, &xdr->tail[0], base);
863out:
864 if (err > 0) {
865 *sent_p += err;
866 err = 0;
867 }
868 return err;
869}
870
871/**
872 * xs_nospace - handle transmit was incomplete
873 * @req: pointer to RPC request
874 *
875 */
876static int xs_nospace(struct rpc_rqst *req)
877{
878 struct rpc_xprt *xprt = req->rq_xprt;
879 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
880 struct sock *sk = transport->inet;
881 int ret = -EAGAIN;
882
883 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
884 req->rq_task->tk_pid,
885 req->rq_slen - transport->xmit.offset,
886 req->rq_slen);
887
888 /* Protect against races with write_space */
889 spin_lock(&xprt->transport_lock);
890
891 /* Don't race with disconnect */
892 if (xprt_connected(xprt)) {
893 /* wait for more buffer space */
894 sk->sk_write_pending++;
895 xprt_wait_for_buffer_space(xprt);
896 } else
897 ret = -ENOTCONN;
898
899 spin_unlock(&xprt->transport_lock);
900
901 /* Race breaker in case memory is freed before above code is called */
902 if (ret == -EAGAIN) {
903 struct socket_wq *wq;
904
905 rcu_read_lock();
906 wq = rcu_dereference(sk->sk_wq);
907 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
908 rcu_read_unlock();
909
910 sk->sk_write_space(sk);
911 }
912 return ret;
913}
914
915static void
916xs_stream_prepare_request(struct rpc_rqst *req)
917{
918 xdr_free_bvec(&req->rq_rcv_buf);
919 req->rq_task->tk_status = xdr_alloc_bvec(&req->rq_rcv_buf, GFP_KERNEL);
920}
921
922/*
923 * Determine if the previous message in the stream was aborted before it
924 * could complete transmission.
925 */
926static bool
927xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req)
928{
929 return transport->xmit.offset != 0 && req->rq_bytes_sent == 0;
930}
931
932/*
933 * Return the stream record marker field for a record of length < 2^31-1
934 */
935static rpc_fraghdr
936xs_stream_record_marker(struct xdr_buf *xdr)
937{
938 if (!xdr->len)
939 return 0;
940 return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len);
941}
942
943/**
944 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
945 * @req: pointer to RPC request
946 *
947 * Return values:
948 * 0: The request has been sent
949 * EAGAIN: The socket was blocked, please call again later to
950 * complete the request
951 * ENOTCONN: Caller needs to invoke connect logic then call again
952 * other: Some other error occured, the request was not sent
953 */
954static int xs_local_send_request(struct rpc_rqst *req)
955{
956 struct rpc_xprt *xprt = req->rq_xprt;
957 struct sock_xprt *transport =
958 container_of(xprt, struct sock_xprt, xprt);
959 struct xdr_buf *xdr = &req->rq_snd_buf;
960 rpc_fraghdr rm = xs_stream_record_marker(xdr);
961 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
962 int status;
963 int sent = 0;
964
965 /* Close the stream if the previous transmission was incomplete */
966 if (xs_send_request_was_aborted(transport, req)) {
967 xs_close(xprt);
968 return -ENOTCONN;
969 }
970
971 xs_pktdump("packet data:",
972 req->rq_svec->iov_base, req->rq_svec->iov_len);
973
974 req->rq_xtime = ktime_get();
975 status = xs_sendpages(transport->sock, NULL, 0, xdr,
976 transport->xmit.offset, rm, &sent);
977 dprintk("RPC: %s(%u) = %d\n",
978 __func__, xdr->len - transport->xmit.offset, status);
979
980 if (status == -EAGAIN && sock_writeable(transport->inet))
981 status = -ENOBUFS;
982
983 if (likely(sent > 0) || status == 0) {
984 transport->xmit.offset += sent;
985 req->rq_bytes_sent = transport->xmit.offset;
986 if (likely(req->rq_bytes_sent >= msglen)) {
987 req->rq_xmit_bytes_sent += transport->xmit.offset;
988 transport->xmit.offset = 0;
989 return 0;
990 }
991 status = -EAGAIN;
992 }
993
994 switch (status) {
995 case -ENOBUFS:
996 break;
997 case -EAGAIN:
998 status = xs_nospace(req);
999 break;
1000 default:
1001 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1002 -status);
1003 /* fall through */
1004 case -EPIPE:
1005 xs_close(xprt);
1006 status = -ENOTCONN;
1007 }
1008
1009 return status;
1010}
1011
1012/**
1013 * xs_udp_send_request - write an RPC request to a UDP socket
1014 * @req: pointer to RPC request
1015 *
1016 * Return values:
1017 * 0: The request has been sent
1018 * EAGAIN: The socket was blocked, please call again later to
1019 * complete the request
1020 * ENOTCONN: Caller needs to invoke connect logic then call again
1021 * other: Some other error occurred, the request was not sent
1022 */
1023static int xs_udp_send_request(struct rpc_rqst *req)
1024{
1025 struct rpc_xprt *xprt = req->rq_xprt;
1026 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1027 struct xdr_buf *xdr = &req->rq_snd_buf;
1028 int sent = 0;
1029 int status;
1030
1031 xs_pktdump("packet data:",
1032 req->rq_svec->iov_base,
1033 req->rq_svec->iov_len);
1034
1035 if (!xprt_bound(xprt))
1036 return -ENOTCONN;
1037
1038 if (!xprt_request_get_cong(xprt, req))
1039 return -EBADSLT;
1040
1041 req->rq_xtime = ktime_get();
1042 status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
1043 xdr, 0, 0, &sent);
1044
1045 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
1046 xdr->len, status);
1047
1048 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
1049 if (status == -EPERM)
1050 goto process_status;
1051
1052 if (status == -EAGAIN && sock_writeable(transport->inet))
1053 status = -ENOBUFS;
1054
1055 if (sent > 0 || status == 0) {
1056 req->rq_xmit_bytes_sent += sent;
1057 if (sent >= req->rq_slen)
1058 return 0;
1059 /* Still some bytes left; set up for a retry later. */
1060 status = -EAGAIN;
1061 }
1062
1063process_status:
1064 switch (status) {
1065 case -ENOTSOCK:
1066 status = -ENOTCONN;
1067 /* Should we call xs_close() here? */
1068 break;
1069 case -EAGAIN:
1070 status = xs_nospace(req);
1071 break;
1072 case -ENETUNREACH:
1073 case -ENOBUFS:
1074 case -EPIPE:
1075 case -ECONNREFUSED:
1076 case -EPERM:
1077 /* When the server has died, an ICMP port unreachable message
1078 * prompts ECONNREFUSED. */
1079 break;
1080 default:
1081 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1082 -status);
1083 }
1084
1085 return status;
1086}
1087
1088/**
1089 * xs_tcp_send_request - write an RPC request to a TCP socket
1090 * @req: pointer to RPC request
1091 *
1092 * Return values:
1093 * 0: The request has been sent
1094 * EAGAIN: The socket was blocked, please call again later to
1095 * complete the request
1096 * ENOTCONN: Caller needs to invoke connect logic then call again
1097 * other: Some other error occurred, the request was not sent
1098 *
1099 * XXX: In the case of soft timeouts, should we eventually give up
1100 * if sendmsg is not able to make progress?
1101 */
1102static int xs_tcp_send_request(struct rpc_rqst *req)
1103{
1104 struct rpc_xprt *xprt = req->rq_xprt;
1105 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1106 struct xdr_buf *xdr = &req->rq_snd_buf;
1107 rpc_fraghdr rm = xs_stream_record_marker(xdr);
1108 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
1109 bool vm_wait = false;
1110 int status;
1111 int sent;
1112
1113 /* Close the stream if the previous transmission was incomplete */
1114 if (xs_send_request_was_aborted(transport, req)) {
1115 if (transport->sock != NULL)
1116 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
1117 return -ENOTCONN;
1118 }
1119
1120 xs_pktdump("packet data:",
1121 req->rq_svec->iov_base,
1122 req->rq_svec->iov_len);
1123
1124 if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
1125 xs_tcp_set_socket_timeouts(xprt, transport->sock);
1126
1127 /* Continue transmitting the packet/record. We must be careful
1128 * to cope with writespace callbacks arriving _after_ we have
1129 * called sendmsg(). */
1130 req->rq_xtime = ktime_get();
1131 while (1) {
1132 sent = 0;
1133 status = xs_sendpages(transport->sock, NULL, 0, xdr,
1134 transport->xmit.offset, rm, &sent);
1135
1136 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
1137 xdr->len - transport->xmit.offset, status);
1138
1139 /* If we've sent the entire packet, immediately
1140 * reset the count of bytes sent. */
1141 transport->xmit.offset += sent;
1142 req->rq_bytes_sent = transport->xmit.offset;
1143 if (likely(req->rq_bytes_sent >= msglen)) {
1144 req->rq_xmit_bytes_sent += transport->xmit.offset;
1145 transport->xmit.offset = 0;
1146 return 0;
1147 }
1148
1149 WARN_ON_ONCE(sent == 0 && status == 0);
1150
1151 if (status == -EAGAIN ) {
1152 /*
1153 * Return EAGAIN if we're sure we're hitting the
1154 * socket send buffer limits.
1155 */
1156 if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
1157 break;
1158 /*
1159 * Did we hit a memory allocation failure?
1160 */
1161 if (sent == 0) {
1162 status = -ENOBUFS;
1163 if (vm_wait)
1164 break;
1165 /* Retry, knowing now that we're below the
1166 * socket send buffer limit
1167 */
1168 vm_wait = true;
1169 }
1170 continue;
1171 }
1172 if (status < 0)
1173 break;
1174 vm_wait = false;
1175 }
1176
1177 switch (status) {
1178 case -ENOTSOCK:
1179 status = -ENOTCONN;
1180 /* Should we call xs_close() here? */
1181 break;
1182 case -EAGAIN:
1183 status = xs_nospace(req);
1184 break;
1185 case -ECONNRESET:
1186 case -ECONNREFUSED:
1187 case -ENOTCONN:
1188 case -EADDRINUSE:
1189 case -ENOBUFS:
1190 case -EPIPE:
1191 break;
1192 default:
1193 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1194 -status);
1195 }
1196
1197 return status;
1198}
1199
1200static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1201{
1202 transport->old_data_ready = sk->sk_data_ready;
1203 transport->old_state_change = sk->sk_state_change;
1204 transport->old_write_space = sk->sk_write_space;
1205 transport->old_error_report = sk->sk_error_report;
1206}
1207
1208static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1209{
1210 sk->sk_data_ready = transport->old_data_ready;
1211 sk->sk_state_change = transport->old_state_change;
1212 sk->sk_write_space = transport->old_write_space;
1213 sk->sk_error_report = transport->old_error_report;
1214}
1215
1216static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
1217{
1218 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1219
1220 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1221 clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state);
1222 clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state);
1223 clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state);
1224}
1225
1226static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr)
1227{
1228 set_bit(nr, &transport->sock_state);
1229 queue_work(xprtiod_workqueue, &transport->error_worker);
1230}
1231
1232static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1233{
1234 smp_mb__before_atomic();
1235 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1236 clear_bit(XPRT_CLOSING, &xprt->state);
1237 xs_sock_reset_state_flags(xprt);
1238 smp_mb__after_atomic();
1239}
1240
1241/**
1242 * xs_error_report - callback to handle TCP socket state errors
1243 * @sk: socket
1244 *
1245 * Note: we don't call sock_error() since there may be a rpc_task
1246 * using the socket, and so we don't want to clear sk->sk_err.
1247 */
1248static void xs_error_report(struct sock *sk)
1249{
1250 struct sock_xprt *transport;
1251 struct rpc_xprt *xprt;
1252
1253 read_lock_bh(&sk->sk_callback_lock);
1254 if (!(xprt = xprt_from_sock(sk)))
1255 goto out;
1256
1257 transport = container_of(xprt, struct sock_xprt, xprt);
1258 transport->xprt_err = -sk->sk_err;
1259 if (transport->xprt_err == 0)
1260 goto out;
1261 dprintk("RPC: xs_error_report client %p, error=%d...\n",
1262 xprt, -transport->xprt_err);
1263 trace_rpc_socket_error(xprt, sk->sk_socket, transport->xprt_err);
1264
1265 /* barrier ensures xprt_err is set before XPRT_SOCK_WAKE_ERROR */
1266 smp_mb__before_atomic();
1267 xs_run_error_worker(transport, XPRT_SOCK_WAKE_ERROR);
1268 out:
1269 read_unlock_bh(&sk->sk_callback_lock);
1270}
1271
1272static void xs_reset_transport(struct sock_xprt *transport)
1273{
1274 struct socket *sock = transport->sock;
1275 struct sock *sk = transport->inet;
1276 struct rpc_xprt *xprt = &transport->xprt;
1277 struct file *filp = transport->file;
1278
1279 if (sk == NULL)
1280 return;
1281
1282 if (atomic_read(&transport->xprt.swapper))
1283 sk_clear_memalloc(sk);
1284
1285 kernel_sock_shutdown(sock, SHUT_RDWR);
1286
1287 mutex_lock(&transport->recv_mutex);
1288 write_lock_bh(&sk->sk_callback_lock);
1289 transport->inet = NULL;
1290 transport->sock = NULL;
1291 transport->file = NULL;
1292
1293 sk->sk_user_data = NULL;
1294
1295 xs_restore_old_callbacks(transport, sk);
1296 xprt_clear_connected(xprt);
1297 write_unlock_bh(&sk->sk_callback_lock);
1298 xs_sock_reset_connection_flags(xprt);
1299 /* Reset stream record info */
1300 xs_stream_reset_connect(transport);
1301 mutex_unlock(&transport->recv_mutex);
1302
1303 trace_rpc_socket_close(xprt, sock);
1304 fput(filp);
1305
1306 xprt_disconnect_done(xprt);
1307}
1308
1309/**
1310 * xs_close - close a socket
1311 * @xprt: transport
1312 *
1313 * This is used when all requests are complete; ie, no DRC state remains
1314 * on the server we want to save.
1315 *
1316 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1317 * xs_reset_transport() zeroing the socket from underneath a writer.
1318 */
1319static void xs_close(struct rpc_xprt *xprt)
1320{
1321 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1322
1323 dprintk("RPC: xs_close xprt %p\n", xprt);
1324
1325 xs_reset_transport(transport);
1326 xprt->reestablish_timeout = 0;
1327}
1328
1329static void xs_inject_disconnect(struct rpc_xprt *xprt)
1330{
1331 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
1332 xprt);
1333 xprt_disconnect_done(xprt);
1334}
1335
1336static void xs_xprt_free(struct rpc_xprt *xprt)
1337{
1338 xs_free_peer_addresses(xprt);
1339 xprt_free(xprt);
1340}
1341
1342/**
1343 * xs_destroy - prepare to shutdown a transport
1344 * @xprt: doomed transport
1345 *
1346 */
1347static void xs_destroy(struct rpc_xprt *xprt)
1348{
1349 struct sock_xprt *transport = container_of(xprt,
1350 struct sock_xprt, xprt);
1351 dprintk("RPC: xs_destroy xprt %p\n", xprt);
1352
1353 cancel_delayed_work_sync(&transport->connect_worker);
1354 xs_close(xprt);
1355 cancel_work_sync(&transport->recv_worker);
1356 cancel_work_sync(&transport->error_worker);
1357 xs_xprt_free(xprt);
1358 module_put(THIS_MODULE);
1359}
1360
1361/**
1362 * xs_udp_data_read_skb - receive callback for UDP sockets
1363 * @xprt: transport
1364 * @sk: socket
1365 * @skb: skbuff
1366 *
1367 */
1368static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1369 struct sock *sk,
1370 struct sk_buff *skb)
1371{
1372 struct rpc_task *task;
1373 struct rpc_rqst *rovr;
1374 int repsize, copied;
1375 u32 _xid;
1376 __be32 *xp;
1377
1378 repsize = skb->len;
1379 if (repsize < 4) {
1380 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
1381 return;
1382 }
1383
1384 /* Copy the XID from the skb... */
1385 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1386 if (xp == NULL)
1387 return;
1388
1389 /* Look up and lock the request corresponding to the given XID */
1390 spin_lock(&xprt->queue_lock);
1391 rovr = xprt_lookup_rqst(xprt, *xp);
1392 if (!rovr)
1393 goto out_unlock;
1394 xprt_pin_rqst(rovr);
1395 xprt_update_rtt(rovr->rq_task);
1396 spin_unlock(&xprt->queue_lock);
1397 task = rovr->rq_task;
1398
1399 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1400 copied = repsize;
1401
1402 /* Suck it into the iovec, verify checksum if not done by hw. */
1403 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1404 spin_lock(&xprt->queue_lock);
1405 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1406 goto out_unpin;
1407 }
1408
1409
1410 spin_lock(&xprt->transport_lock);
1411 xprt_adjust_cwnd(xprt, task, copied);
1412 spin_unlock(&xprt->transport_lock);
1413 spin_lock(&xprt->queue_lock);
1414 xprt_complete_rqst(task, copied);
1415 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1416out_unpin:
1417 xprt_unpin_rqst(rovr);
1418 out_unlock:
1419 spin_unlock(&xprt->queue_lock);
1420}
1421
1422static void xs_udp_data_receive(struct sock_xprt *transport)
1423{
1424 struct sk_buff *skb;
1425 struct sock *sk;
1426 int err;
1427
1428 mutex_lock(&transport->recv_mutex);
1429 sk = transport->inet;
1430 if (sk == NULL)
1431 goto out;
1432 for (;;) {
1433 skb = skb_recv_udp(sk, 0, 1, &err);
1434 if (skb == NULL)
1435 break;
1436 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1437 consume_skb(skb);
1438 cond_resched();
1439 }
1440 xs_poll_check_readable(transport);
1441out:
1442 mutex_unlock(&transport->recv_mutex);
1443}
1444
1445static void xs_udp_data_receive_workfn(struct work_struct *work)
1446{
1447 struct sock_xprt *transport =
1448 container_of(work, struct sock_xprt, recv_worker);
1449 unsigned int pflags = memalloc_nofs_save();
1450
1451 xs_udp_data_receive(transport);
1452 memalloc_nofs_restore(pflags);
1453}
1454
1455/**
1456 * xs_data_ready - "data ready" callback for UDP sockets
1457 * @sk: socket with data to read
1458 *
1459 */
1460static void xs_data_ready(struct sock *sk)
1461{
1462 struct rpc_xprt *xprt;
1463
1464 read_lock_bh(&sk->sk_callback_lock);
1465 dprintk("RPC: xs_data_ready...\n");
1466 xprt = xprt_from_sock(sk);
1467 if (xprt != NULL) {
1468 struct sock_xprt *transport = container_of(xprt,
1469 struct sock_xprt, xprt);
1470 transport->old_data_ready(sk);
1471 /* Any data means we had a useful conversation, so
1472 * then we don't need to delay the next reconnect
1473 */
1474 if (xprt->reestablish_timeout)
1475 xprt->reestablish_timeout = 0;
1476 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1477 queue_work(xprtiod_workqueue, &transport->recv_worker);
1478 }
1479 read_unlock_bh(&sk->sk_callback_lock);
1480}
1481
1482/*
1483 * Helper function to force a TCP close if the server is sending
1484 * junk and/or it has put us in CLOSE_WAIT
1485 */
1486static void xs_tcp_force_close(struct rpc_xprt *xprt)
1487{
1488 xprt_force_disconnect(xprt);
1489}
1490
1491#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1492static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1493{
1494 return PAGE_SIZE;
1495}
1496#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1497
1498/**
1499 * xs_tcp_state_change - callback to handle TCP socket state changes
1500 * @sk: socket whose state has changed
1501 *
1502 */
1503static void xs_tcp_state_change(struct sock *sk)
1504{
1505 struct rpc_xprt *xprt;
1506 struct sock_xprt *transport;
1507
1508 read_lock_bh(&sk->sk_callback_lock);
1509 if (!(xprt = xprt_from_sock(sk)))
1510 goto out;
1511 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1512 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1513 sk->sk_state, xprt_connected(xprt),
1514 sock_flag(sk, SOCK_DEAD),
1515 sock_flag(sk, SOCK_ZAPPED),
1516 sk->sk_shutdown);
1517
1518 transport = container_of(xprt, struct sock_xprt, xprt);
1519 trace_rpc_socket_state_change(xprt, sk->sk_socket);
1520 switch (sk->sk_state) {
1521 case TCP_ESTABLISHED:
1522 if (!xprt_test_and_set_connected(xprt)) {
1523 xprt->connect_cookie++;
1524 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1525 xprt_clear_connecting(xprt);
1526
1527 xprt->stat.connect_count++;
1528 xprt->stat.connect_time += (long)jiffies -
1529 xprt->stat.connect_start;
1530 xs_run_error_worker(transport, XPRT_SOCK_WAKE_PENDING);
1531 }
1532 break;
1533 case TCP_FIN_WAIT1:
1534 /* The client initiated a shutdown of the socket */
1535 xprt->connect_cookie++;
1536 xprt->reestablish_timeout = 0;
1537 set_bit(XPRT_CLOSING, &xprt->state);
1538 smp_mb__before_atomic();
1539 clear_bit(XPRT_CONNECTED, &xprt->state);
1540 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1541 smp_mb__after_atomic();
1542 break;
1543 case TCP_CLOSE_WAIT:
1544 /* The server initiated a shutdown of the socket */
1545 xprt->connect_cookie++;
1546 clear_bit(XPRT_CONNECTED, &xprt->state);
1547 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1548 /* fall through */
1549 case TCP_CLOSING:
1550 /*
1551 * If the server closed down the connection, make sure that
1552 * we back off before reconnecting
1553 */
1554 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1555 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1556 break;
1557 case TCP_LAST_ACK:
1558 set_bit(XPRT_CLOSING, &xprt->state);
1559 smp_mb__before_atomic();
1560 clear_bit(XPRT_CONNECTED, &xprt->state);
1561 smp_mb__after_atomic();
1562 break;
1563 case TCP_CLOSE:
1564 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1565 &transport->sock_state))
1566 xprt_clear_connecting(xprt);
1567 clear_bit(XPRT_CLOSING, &xprt->state);
1568 /* Trigger the socket release */
1569 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1570 }
1571 out:
1572 read_unlock_bh(&sk->sk_callback_lock);
1573}
1574
1575static void xs_write_space(struct sock *sk)
1576{
1577 struct socket_wq *wq;
1578 struct sock_xprt *transport;
1579 struct rpc_xprt *xprt;
1580
1581 if (!sk->sk_socket)
1582 return;
1583 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1584
1585 if (unlikely(!(xprt = xprt_from_sock(sk))))
1586 return;
1587 transport = container_of(xprt, struct sock_xprt, xprt);
1588 rcu_read_lock();
1589 wq = rcu_dereference(sk->sk_wq);
1590 if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1591 goto out;
1592
1593 xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE);
1594 sk->sk_write_pending--;
1595out:
1596 rcu_read_unlock();
1597}
1598
1599/**
1600 * xs_udp_write_space - callback invoked when socket buffer space
1601 * becomes available
1602 * @sk: socket whose state has changed
1603 *
1604 * Called when more output buffer space is available for this socket.
1605 * We try not to wake our writers until they can make "significant"
1606 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1607 * with a bunch of small requests.
1608 */
1609static void xs_udp_write_space(struct sock *sk)
1610{
1611 read_lock_bh(&sk->sk_callback_lock);
1612
1613 /* from net/core/sock.c:sock_def_write_space */
1614 if (sock_writeable(sk))
1615 xs_write_space(sk);
1616
1617 read_unlock_bh(&sk->sk_callback_lock);
1618}
1619
1620/**
1621 * xs_tcp_write_space - callback invoked when socket buffer space
1622 * becomes available
1623 * @sk: socket whose state has changed
1624 *
1625 * Called when more output buffer space is available for this socket.
1626 * We try not to wake our writers until they can make "significant"
1627 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1628 * with a bunch of small requests.
1629 */
1630static void xs_tcp_write_space(struct sock *sk)
1631{
1632 read_lock_bh(&sk->sk_callback_lock);
1633
1634 /* from net/core/stream.c:sk_stream_write_space */
1635 if (sk_stream_is_writeable(sk))
1636 xs_write_space(sk);
1637
1638 read_unlock_bh(&sk->sk_callback_lock);
1639}
1640
1641static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1642{
1643 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1644 struct sock *sk = transport->inet;
1645
1646 if (transport->rcvsize) {
1647 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1648 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1649 }
1650 if (transport->sndsize) {
1651 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1652 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1653 sk->sk_write_space(sk);
1654 }
1655}
1656
1657/**
1658 * xs_udp_set_buffer_size - set send and receive limits
1659 * @xprt: generic transport
1660 * @sndsize: requested size of send buffer, in bytes
1661 * @rcvsize: requested size of receive buffer, in bytes
1662 *
1663 * Set socket send and receive buffer size limits.
1664 */
1665static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1666{
1667 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1668
1669 transport->sndsize = 0;
1670 if (sndsize)
1671 transport->sndsize = sndsize + 1024;
1672 transport->rcvsize = 0;
1673 if (rcvsize)
1674 transport->rcvsize = rcvsize + 1024;
1675
1676 xs_udp_do_set_buffer_size(xprt);
1677}
1678
1679/**
1680 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1681 * @xprt: controlling transport
1682 * @task: task that timed out
1683 *
1684 * Adjust the congestion window after a retransmit timeout has occurred.
1685 */
1686static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1687{
1688 spin_lock(&xprt->transport_lock);
1689 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1690 spin_unlock(&xprt->transport_lock);
1691}
1692
1693static int xs_get_random_port(void)
1694{
1695 unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1696 unsigned short range;
1697 unsigned short rand;
1698
1699 if (max < min)
1700 return -EADDRINUSE;
1701 range = max - min + 1;
1702 rand = (unsigned short) prandom_u32() % range;
1703 return rand + min;
1704}
1705
1706/**
1707 * xs_set_reuseaddr_port - set the socket's port and address reuse options
1708 * @sock: socket
1709 *
1710 * Note that this function has to be called on all sockets that share the
1711 * same port, and it must be called before binding.
1712 */
1713static void xs_sock_set_reuseport(struct socket *sock)
1714{
1715 int opt = 1;
1716
1717 kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1718 (char *)&opt, sizeof(opt));
1719}
1720
1721static unsigned short xs_sock_getport(struct socket *sock)
1722{
1723 struct sockaddr_storage buf;
1724 unsigned short port = 0;
1725
1726 if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1727 goto out;
1728 switch (buf.ss_family) {
1729 case AF_INET6:
1730 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1731 break;
1732 case AF_INET:
1733 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1734 }
1735out:
1736 return port;
1737}
1738
1739/**
1740 * xs_set_port - reset the port number in the remote endpoint address
1741 * @xprt: generic transport
1742 * @port: new port number
1743 *
1744 */
1745static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1746{
1747 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1748
1749 rpc_set_port(xs_addr(xprt), port);
1750 xs_update_peer_port(xprt);
1751}
1752
1753static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1754{
1755 if (transport->srcport == 0)
1756 transport->srcport = xs_sock_getport(sock);
1757}
1758
1759static int xs_get_srcport(struct sock_xprt *transport)
1760{
1761 int port = transport->srcport;
1762
1763 if (port == 0 && transport->xprt.resvport)
1764 port = xs_get_random_port();
1765 return port;
1766}
1767
1768static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1769{
1770 if (transport->srcport != 0)
1771 transport->srcport = 0;
1772 if (!transport->xprt.resvport)
1773 return 0;
1774 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1775 return xprt_max_resvport;
1776 return --port;
1777}
1778static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1779{
1780 struct sockaddr_storage myaddr;
1781 int err, nloop = 0;
1782 int port = xs_get_srcport(transport);
1783 unsigned short last;
1784
1785 /*
1786 * If we are asking for any ephemeral port (i.e. port == 0 &&
1787 * transport->xprt.resvport == 0), don't bind. Let the local
1788 * port selection happen implicitly when the socket is used
1789 * (for example at connect time).
1790 *
1791 * This ensures that we can continue to establish TCP
1792 * connections even when all local ephemeral ports are already
1793 * a part of some TCP connection. This makes no difference
1794 * for UDP sockets, but also doens't harm them.
1795 *
1796 * If we're asking for any reserved port (i.e. port == 0 &&
1797 * transport->xprt.resvport == 1) xs_get_srcport above will
1798 * ensure that port is non-zero and we will bind as needed.
1799 */
1800 if (port <= 0)
1801 return port;
1802
1803 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1804 do {
1805 rpc_set_port((struct sockaddr *)&myaddr, port);
1806 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1807 transport->xprt.addrlen);
1808 if (err == 0) {
1809 transport->srcport = port;
1810 break;
1811 }
1812 last = port;
1813 port = xs_next_srcport(transport, port);
1814 if (port > last)
1815 nloop++;
1816 } while (err == -EADDRINUSE && nloop != 2);
1817
1818 if (myaddr.ss_family == AF_INET)
1819 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1820 &((struct sockaddr_in *)&myaddr)->sin_addr,
1821 port, err ? "failed" : "ok", err);
1822 else
1823 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1824 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1825 port, err ? "failed" : "ok", err);
1826 return err;
1827}
1828
1829/*
1830 * We don't support autobind on AF_LOCAL sockets
1831 */
1832static void xs_local_rpcbind(struct rpc_task *task)
1833{
1834 xprt_set_bound(task->tk_xprt);
1835}
1836
1837static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1838{
1839}
1840
1841#ifdef CONFIG_DEBUG_LOCK_ALLOC
1842static struct lock_class_key xs_key[2];
1843static struct lock_class_key xs_slock_key[2];
1844
1845static inline void xs_reclassify_socketu(struct socket *sock)
1846{
1847 struct sock *sk = sock->sk;
1848
1849 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1850 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1851}
1852
1853static inline void xs_reclassify_socket4(struct socket *sock)
1854{
1855 struct sock *sk = sock->sk;
1856
1857 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1858 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1859}
1860
1861static inline void xs_reclassify_socket6(struct socket *sock)
1862{
1863 struct sock *sk = sock->sk;
1864
1865 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1866 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1867}
1868
1869static inline void xs_reclassify_socket(int family, struct socket *sock)
1870{
1871 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1872 return;
1873
1874 switch (family) {
1875 case AF_LOCAL:
1876 xs_reclassify_socketu(sock);
1877 break;
1878 case AF_INET:
1879 xs_reclassify_socket4(sock);
1880 break;
1881 case AF_INET6:
1882 xs_reclassify_socket6(sock);
1883 break;
1884 }
1885}
1886#else
1887static inline void xs_reclassify_socket(int family, struct socket *sock)
1888{
1889}
1890#endif
1891
1892static void xs_dummy_setup_socket(struct work_struct *work)
1893{
1894}
1895
1896static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1897 struct sock_xprt *transport, int family, int type,
1898 int protocol, bool reuseport)
1899{
1900 struct file *filp;
1901 struct socket *sock;
1902 int err;
1903
1904 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1905 if (err < 0) {
1906 dprintk("RPC: can't create %d transport socket (%d).\n",
1907 protocol, -err);
1908 goto out;
1909 }
1910 xs_reclassify_socket(family, sock);
1911
1912 if (reuseport)
1913 xs_sock_set_reuseport(sock);
1914
1915 err = xs_bind(transport, sock);
1916 if (err) {
1917 sock_release(sock);
1918 goto out;
1919 }
1920
1921 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1922 if (IS_ERR(filp))
1923 return ERR_CAST(filp);
1924 transport->file = filp;
1925
1926 return sock;
1927out:
1928 return ERR_PTR(err);
1929}
1930
1931static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1932 struct socket *sock)
1933{
1934 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1935 xprt);
1936
1937 if (!transport->inet) {
1938 struct sock *sk = sock->sk;
1939
1940 write_lock_bh(&sk->sk_callback_lock);
1941
1942 xs_save_old_callbacks(transport, sk);
1943
1944 sk->sk_user_data = xprt;
1945 sk->sk_data_ready = xs_data_ready;
1946 sk->sk_write_space = xs_udp_write_space;
1947 sock_set_flag(sk, SOCK_FASYNC);
1948 sk->sk_error_report = xs_error_report;
1949
1950 xprt_clear_connected(xprt);
1951
1952 /* Reset to new socket */
1953 transport->sock = sock;
1954 transport->inet = sk;
1955
1956 write_unlock_bh(&sk->sk_callback_lock);
1957 }
1958
1959 xs_stream_start_connect(transport);
1960
1961 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1962}
1963
1964/**
1965 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1966 * @transport: socket transport to connect
1967 */
1968static int xs_local_setup_socket(struct sock_xprt *transport)
1969{
1970 struct rpc_xprt *xprt = &transport->xprt;
1971 struct file *filp;
1972 struct socket *sock;
1973 int status = -EIO;
1974
1975 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1976 SOCK_STREAM, 0, &sock, 1);
1977 if (status < 0) {
1978 dprintk("RPC: can't create AF_LOCAL "
1979 "transport socket (%d).\n", -status);
1980 goto out;
1981 }
1982 xs_reclassify_socket(AF_LOCAL, sock);
1983
1984 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1985 if (IS_ERR(filp)) {
1986 status = PTR_ERR(filp);
1987 goto out;
1988 }
1989 transport->file = filp;
1990
1991 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1992 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1993
1994 status = xs_local_finish_connecting(xprt, sock);
1995 trace_rpc_socket_connect(xprt, sock, status);
1996 switch (status) {
1997 case 0:
1998 dprintk("RPC: xprt %p connected to %s\n",
1999 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2000 xprt->stat.connect_count++;
2001 xprt->stat.connect_time += (long)jiffies -
2002 xprt->stat.connect_start;
2003 xprt_set_connected(xprt);
2004 case -ENOBUFS:
2005 break;
2006 case -ENOENT:
2007 dprintk("RPC: xprt %p: socket %s does not exist\n",
2008 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2009 break;
2010 case -ECONNREFUSED:
2011 dprintk("RPC: xprt %p: connection refused for %s\n",
2012 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2013 break;
2014 default:
2015 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2016 __func__, -status,
2017 xprt->address_strings[RPC_DISPLAY_ADDR]);
2018 }
2019
2020out:
2021 xprt_clear_connecting(xprt);
2022 xprt_wake_pending_tasks(xprt, status);
2023 return status;
2024}
2025
2026static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2027{
2028 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2029 int ret;
2030
2031 if (RPC_IS_ASYNC(task)) {
2032 /*
2033 * We want the AF_LOCAL connect to be resolved in the
2034 * filesystem namespace of the process making the rpc
2035 * call. Thus we connect synchronously.
2036 *
2037 * If we want to support asynchronous AF_LOCAL calls,
2038 * we'll need to figure out how to pass a namespace to
2039 * connect.
2040 */
2041 task->tk_rpc_status = -ENOTCONN;
2042 rpc_exit(task, -ENOTCONN);
2043 return;
2044 }
2045 ret = xs_local_setup_socket(transport);
2046 if (ret && !RPC_IS_SOFTCONN(task))
2047 msleep_interruptible(15000);
2048}
2049
2050#if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2051/*
2052 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2053 * know that we have exclusive access to the socket), to guard against
2054 * races with xs_reset_transport.
2055 */
2056static void xs_set_memalloc(struct rpc_xprt *xprt)
2057{
2058 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2059 xprt);
2060
2061 /*
2062 * If there's no sock, then we have nothing to set. The
2063 * reconnecting process will get it for us.
2064 */
2065 if (!transport->inet)
2066 return;
2067 if (atomic_read(&xprt->swapper))
2068 sk_set_memalloc(transport->inet);
2069}
2070
2071/**
2072 * xs_enable_swap - Tag this transport as being used for swap.
2073 * @xprt: transport to tag
2074 *
2075 * Take a reference to this transport on behalf of the rpc_clnt, and
2076 * optionally mark it for swapping if it wasn't already.
2077 */
2078static int
2079xs_enable_swap(struct rpc_xprt *xprt)
2080{
2081 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2082
2083 if (atomic_inc_return(&xprt->swapper) != 1)
2084 return 0;
2085 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2086 return -ERESTARTSYS;
2087 if (xs->inet)
2088 sk_set_memalloc(xs->inet);
2089 xprt_release_xprt(xprt, NULL);
2090 return 0;
2091}
2092
2093/**
2094 * xs_disable_swap - Untag this transport as being used for swap.
2095 * @xprt: transport to tag
2096 *
2097 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2098 * swapper refcount goes to 0, untag the socket as a memalloc socket.
2099 */
2100static void
2101xs_disable_swap(struct rpc_xprt *xprt)
2102{
2103 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2104
2105 if (!atomic_dec_and_test(&xprt->swapper))
2106 return;
2107 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2108 return;
2109 if (xs->inet)
2110 sk_clear_memalloc(xs->inet);
2111 xprt_release_xprt(xprt, NULL);
2112}
2113#else
2114static void xs_set_memalloc(struct rpc_xprt *xprt)
2115{
2116}
2117
2118static int
2119xs_enable_swap(struct rpc_xprt *xprt)
2120{
2121 return -EINVAL;
2122}
2123
2124static void
2125xs_disable_swap(struct rpc_xprt *xprt)
2126{
2127}
2128#endif
2129
2130static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2131{
2132 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2133
2134 if (!transport->inet) {
2135 struct sock *sk = sock->sk;
2136
2137 write_lock_bh(&sk->sk_callback_lock);
2138
2139 xs_save_old_callbacks(transport, sk);
2140
2141 sk->sk_user_data = xprt;
2142 sk->sk_data_ready = xs_data_ready;
2143 sk->sk_write_space = xs_udp_write_space;
2144 sock_set_flag(sk, SOCK_FASYNC);
2145
2146 xprt_set_connected(xprt);
2147
2148 /* Reset to new socket */
2149 transport->sock = sock;
2150 transport->inet = sk;
2151
2152 xs_set_memalloc(xprt);
2153
2154 write_unlock_bh(&sk->sk_callback_lock);
2155 }
2156 xs_udp_do_set_buffer_size(xprt);
2157
2158 xprt->stat.connect_start = jiffies;
2159}
2160
2161static void xs_udp_setup_socket(struct work_struct *work)
2162{
2163 struct sock_xprt *transport =
2164 container_of(work, struct sock_xprt, connect_worker.work);
2165 struct rpc_xprt *xprt = &transport->xprt;
2166 struct socket *sock;
2167 int status = -EIO;
2168
2169 sock = xs_create_sock(xprt, transport,
2170 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2171 IPPROTO_UDP, false);
2172 if (IS_ERR(sock))
2173 goto out;
2174
2175 dprintk("RPC: worker connecting xprt %p via %s to "
2176 "%s (port %s)\n", xprt,
2177 xprt->address_strings[RPC_DISPLAY_PROTO],
2178 xprt->address_strings[RPC_DISPLAY_ADDR],
2179 xprt->address_strings[RPC_DISPLAY_PORT]);
2180
2181 xs_udp_finish_connecting(xprt, sock);
2182 trace_rpc_socket_connect(xprt, sock, 0);
2183 status = 0;
2184out:
2185 xprt_clear_connecting(xprt);
2186 xprt_unlock_connect(xprt, transport);
2187 xprt_wake_pending_tasks(xprt, status);
2188}
2189
2190/**
2191 * xs_tcp_shutdown - gracefully shut down a TCP socket
2192 * @xprt: transport
2193 *
2194 * Initiates a graceful shutdown of the TCP socket by calling the
2195 * equivalent of shutdown(SHUT_RDWR);
2196 */
2197static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2198{
2199 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2200 struct socket *sock = transport->sock;
2201 int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2202
2203 if (sock == NULL)
2204 return;
2205 switch (skst) {
2206 default:
2207 kernel_sock_shutdown(sock, SHUT_RDWR);
2208 trace_rpc_socket_shutdown(xprt, sock);
2209 break;
2210 case TCP_CLOSE:
2211 case TCP_TIME_WAIT:
2212 xs_reset_transport(transport);
2213 }
2214}
2215
2216static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2217 struct socket *sock)
2218{
2219 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2220 unsigned int keepidle;
2221 unsigned int keepcnt;
2222 unsigned int opt_on = 1;
2223 unsigned int timeo;
2224
2225 spin_lock(&xprt->transport_lock);
2226 keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2227 keepcnt = xprt->timeout->to_retries + 1;
2228 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2229 (xprt->timeout->to_retries + 1);
2230 clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2231 spin_unlock(&xprt->transport_lock);
2232
2233 /* TCP Keepalive options */
2234 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2235 (char *)&opt_on, sizeof(opt_on));
2236 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2237 (char *)&keepidle, sizeof(keepidle));
2238 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2239 (char *)&keepidle, sizeof(keepidle));
2240 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2241 (char *)&keepcnt, sizeof(keepcnt));
2242
2243 /* TCP user timeout (see RFC5482) */
2244 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2245 (char *)&timeo, sizeof(timeo));
2246}
2247
2248static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2249 unsigned long connect_timeout,
2250 unsigned long reconnect_timeout)
2251{
2252 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2253 struct rpc_timeout to;
2254 unsigned long initval;
2255
2256 spin_lock(&xprt->transport_lock);
2257 if (reconnect_timeout < xprt->max_reconnect_timeout)
2258 xprt->max_reconnect_timeout = reconnect_timeout;
2259 if (connect_timeout < xprt->connect_timeout) {
2260 memcpy(&to, xprt->timeout, sizeof(to));
2261 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2262 /* Arbitrary lower limit */
2263 if (initval < XS_TCP_INIT_REEST_TO << 1)
2264 initval = XS_TCP_INIT_REEST_TO << 1;
2265 to.to_initval = initval;
2266 to.to_maxval = initval;
2267 memcpy(&transport->tcp_timeout, &to,
2268 sizeof(transport->tcp_timeout));
2269 xprt->timeout = &transport->tcp_timeout;
2270 xprt->connect_timeout = connect_timeout;
2271 }
2272 set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2273 spin_unlock(&xprt->transport_lock);
2274}
2275
2276static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2277{
2278 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2279 int ret = -ENOTCONN;
2280
2281 if (!transport->inet) {
2282 struct sock *sk = sock->sk;
2283 unsigned int addr_pref = IPV6_PREFER_SRC_PUBLIC;
2284
2285 /* Avoid temporary address, they are bad for long-lived
2286 * connections such as NFS mounts.
2287 * RFC4941, section 3.6 suggests that:
2288 * Individual applications, which have specific
2289 * knowledge about the normal duration of connections,
2290 * MAY override this as appropriate.
2291 */
2292 kernel_setsockopt(sock, SOL_IPV6, IPV6_ADDR_PREFERENCES,
2293 (char *)&addr_pref, sizeof(addr_pref));
2294
2295 xs_tcp_set_socket_timeouts(xprt, sock);
2296
2297 write_lock_bh(&sk->sk_callback_lock);
2298
2299 xs_save_old_callbacks(transport, sk);
2300
2301 sk->sk_user_data = xprt;
2302 sk->sk_data_ready = xs_data_ready;
2303 sk->sk_state_change = xs_tcp_state_change;
2304 sk->sk_write_space = xs_tcp_write_space;
2305 sock_set_flag(sk, SOCK_FASYNC);
2306 sk->sk_error_report = xs_error_report;
2307
2308 /* socket options */
2309 sock_reset_flag(sk, SOCK_LINGER);
2310 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2311
2312 xprt_clear_connected(xprt);
2313
2314 /* Reset to new socket */
2315 transport->sock = sock;
2316 transport->inet = sk;
2317
2318 write_unlock_bh(&sk->sk_callback_lock);
2319 }
2320
2321 if (!xprt_bound(xprt))
2322 goto out;
2323
2324 xs_set_memalloc(xprt);
2325
2326 xs_stream_start_connect(transport);
2327
2328 /* Tell the socket layer to start connecting... */
2329 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2330 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2331 switch (ret) {
2332 case 0:
2333 xs_set_srcport(transport, sock);
2334 /* fall through */
2335 case -EINPROGRESS:
2336 /* SYN_SENT! */
2337 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2338 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2339 break;
2340 case -EADDRNOTAVAIL:
2341 /* Source port number is unavailable. Try a new one! */
2342 transport->srcport = 0;
2343 }
2344out:
2345 return ret;
2346}
2347
2348/**
2349 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2350 * @work: queued work item
2351 *
2352 * Invoked by a work queue tasklet.
2353 */
2354static void xs_tcp_setup_socket(struct work_struct *work)
2355{
2356 struct sock_xprt *transport =
2357 container_of(work, struct sock_xprt, connect_worker.work);
2358 struct socket *sock = transport->sock;
2359 struct rpc_xprt *xprt = &transport->xprt;
2360 int status = -EIO;
2361
2362 if (!sock) {
2363 sock = xs_create_sock(xprt, transport,
2364 xs_addr(xprt)->sa_family, SOCK_STREAM,
2365 IPPROTO_TCP, true);
2366 if (IS_ERR(sock)) {
2367 status = PTR_ERR(sock);
2368 goto out;
2369 }
2370 }
2371
2372 dprintk("RPC: worker connecting xprt %p via %s to "
2373 "%s (port %s)\n", xprt,
2374 xprt->address_strings[RPC_DISPLAY_PROTO],
2375 xprt->address_strings[RPC_DISPLAY_ADDR],
2376 xprt->address_strings[RPC_DISPLAY_PORT]);
2377
2378 status = xs_tcp_finish_connecting(xprt, sock);
2379 trace_rpc_socket_connect(xprt, sock, status);
2380 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2381 xprt, -status, xprt_connected(xprt),
2382 sock->sk->sk_state);
2383 switch (status) {
2384 default:
2385 printk("%s: connect returned unhandled error %d\n",
2386 __func__, status);
2387 /* fall through */
2388 case -EADDRNOTAVAIL:
2389 /* We're probably in TIME_WAIT. Get rid of existing socket,
2390 * and retry
2391 */
2392 xs_tcp_force_close(xprt);
2393 break;
2394 case 0:
2395 case -EINPROGRESS:
2396 case -EALREADY:
2397 xprt_unlock_connect(xprt, transport);
2398 return;
2399 case -EINVAL:
2400 /* Happens, for instance, if the user specified a link
2401 * local IPv6 address without a scope-id.
2402 */
2403 case -ECONNREFUSED:
2404 case -ECONNRESET:
2405 case -ENETDOWN:
2406 case -ENETUNREACH:
2407 case -EHOSTUNREACH:
2408 case -EADDRINUSE:
2409 case -ENOBUFS:
2410 /*
2411 * xs_tcp_force_close() wakes tasks with -EIO.
2412 * We need to wake them first to ensure the
2413 * correct error code.
2414 */
2415 xprt_wake_pending_tasks(xprt, status);
2416 xs_tcp_force_close(xprt);
2417 goto out;
2418 }
2419 status = -EAGAIN;
2420out:
2421 xprt_clear_connecting(xprt);
2422 xprt_unlock_connect(xprt, transport);
2423 xprt_wake_pending_tasks(xprt, status);
2424}
2425
2426/**
2427 * xs_connect - connect a socket to a remote endpoint
2428 * @xprt: pointer to transport structure
2429 * @task: address of RPC task that manages state of connect request
2430 *
2431 * TCP: If the remote end dropped the connection, delay reconnecting.
2432 *
2433 * UDP socket connects are synchronous, but we use a work queue anyway
2434 * to guarantee that even unprivileged user processes can set up a
2435 * socket on a privileged port.
2436 *
2437 * If a UDP socket connect fails, the delay behavior here prevents
2438 * retry floods (hard mounts).
2439 */
2440static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2441{
2442 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2443 unsigned long delay = 0;
2444
2445 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2446
2447 if (transport->sock != NULL) {
2448 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2449 "seconds\n",
2450 xprt, xprt->reestablish_timeout / HZ);
2451
2452 /* Start by resetting any existing state */
2453 xs_reset_transport(transport);
2454
2455 delay = xprt_reconnect_delay(xprt);
2456 xprt_reconnect_backoff(xprt, XS_TCP_INIT_REEST_TO);
2457
2458 } else
2459 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2460
2461 queue_delayed_work(xprtiod_workqueue,
2462 &transport->connect_worker,
2463 delay);
2464}
2465
2466static void xs_wake_disconnect(struct sock_xprt *transport)
2467{
2468 if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state))
2469 xs_tcp_force_close(&transport->xprt);
2470}
2471
2472static void xs_wake_write(struct sock_xprt *transport)
2473{
2474 if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state))
2475 xprt_write_space(&transport->xprt);
2476}
2477
2478static void xs_wake_error(struct sock_xprt *transport)
2479{
2480 int sockerr;
2481
2482 if (!test_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2483 return;
2484 mutex_lock(&transport->recv_mutex);
2485 if (transport->sock == NULL)
2486 goto out;
2487 if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2488 goto out;
2489 sockerr = xchg(&transport->xprt_err, 0);
2490 if (sockerr < 0)
2491 xprt_wake_pending_tasks(&transport->xprt, sockerr);
2492out:
2493 mutex_unlock(&transport->recv_mutex);
2494}
2495
2496static void xs_wake_pending(struct sock_xprt *transport)
2497{
2498 if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING, &transport->sock_state))
2499 xprt_wake_pending_tasks(&transport->xprt, -EAGAIN);
2500}
2501
2502static void xs_error_handle(struct work_struct *work)
2503{
2504 struct sock_xprt *transport = container_of(work,
2505 struct sock_xprt, error_worker);
2506
2507 xs_wake_disconnect(transport);
2508 xs_wake_write(transport);
2509 xs_wake_error(transport);
2510 xs_wake_pending(transport);
2511}
2512
2513/**
2514 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2515 * @xprt: rpc_xprt struct containing statistics
2516 * @seq: output file
2517 *
2518 */
2519static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2520{
2521 long idle_time = 0;
2522
2523 if (xprt_connected(xprt))
2524 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2525
2526 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2527 "%llu %llu %lu %llu %llu\n",
2528 xprt->stat.bind_count,
2529 xprt->stat.connect_count,
2530 xprt->stat.connect_time / HZ,
2531 idle_time,
2532 xprt->stat.sends,
2533 xprt->stat.recvs,
2534 xprt->stat.bad_xids,
2535 xprt->stat.req_u,
2536 xprt->stat.bklog_u,
2537 xprt->stat.max_slots,
2538 xprt->stat.sending_u,
2539 xprt->stat.pending_u);
2540}
2541
2542/**
2543 * xs_udp_print_stats - display UDP socket-specifc stats
2544 * @xprt: rpc_xprt struct containing statistics
2545 * @seq: output file
2546 *
2547 */
2548static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2549{
2550 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2551
2552 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2553 "%lu %llu %llu\n",
2554 transport->srcport,
2555 xprt->stat.bind_count,
2556 xprt->stat.sends,
2557 xprt->stat.recvs,
2558 xprt->stat.bad_xids,
2559 xprt->stat.req_u,
2560 xprt->stat.bklog_u,
2561 xprt->stat.max_slots,
2562 xprt->stat.sending_u,
2563 xprt->stat.pending_u);
2564}
2565
2566/**
2567 * xs_tcp_print_stats - display TCP socket-specifc stats
2568 * @xprt: rpc_xprt struct containing statistics
2569 * @seq: output file
2570 *
2571 */
2572static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2573{
2574 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2575 long idle_time = 0;
2576
2577 if (xprt_connected(xprt))
2578 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2579
2580 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2581 "%llu %llu %lu %llu %llu\n",
2582 transport->srcport,
2583 xprt->stat.bind_count,
2584 xprt->stat.connect_count,
2585 xprt->stat.connect_time / HZ,
2586 idle_time,
2587 xprt->stat.sends,
2588 xprt->stat.recvs,
2589 xprt->stat.bad_xids,
2590 xprt->stat.req_u,
2591 xprt->stat.bklog_u,
2592 xprt->stat.max_slots,
2593 xprt->stat.sending_u,
2594 xprt->stat.pending_u);
2595}
2596
2597/*
2598 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2599 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2600 * to use the server side send routines.
2601 */
2602static int bc_malloc(struct rpc_task *task)
2603{
2604 struct rpc_rqst *rqst = task->tk_rqstp;
2605 size_t size = rqst->rq_callsize;
2606 struct page *page;
2607 struct rpc_buffer *buf;
2608
2609 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2610 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2611 size);
2612 return -EINVAL;
2613 }
2614
2615 page = alloc_page(GFP_KERNEL);
2616 if (!page)
2617 return -ENOMEM;
2618
2619 buf = page_address(page);
2620 buf->len = PAGE_SIZE;
2621
2622 rqst->rq_buffer = buf->data;
2623 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2624 return 0;
2625}
2626
2627/*
2628 * Free the space allocated in the bc_alloc routine
2629 */
2630static void bc_free(struct rpc_task *task)
2631{
2632 void *buffer = task->tk_rqstp->rq_buffer;
2633 struct rpc_buffer *buf;
2634
2635 buf = container_of(buffer, struct rpc_buffer, data);
2636 free_page((unsigned long)buf);
2637}
2638
2639/*
2640 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2641 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2642 */
2643static int bc_sendto(struct rpc_rqst *req)
2644{
2645 int len;
2646 struct xdr_buf *xbufp = &req->rq_snd_buf;
2647 struct sock_xprt *transport =
2648 container_of(req->rq_xprt, struct sock_xprt, xprt);
2649 unsigned long headoff;
2650 unsigned long tailoff;
2651 struct page *tailpage;
2652 struct msghdr msg = {
2653 .msg_flags = MSG_MORE
2654 };
2655 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
2656 (u32)xbufp->len);
2657 struct kvec iov = {
2658 .iov_base = &marker,
2659 .iov_len = sizeof(marker),
2660 };
2661
2662 len = kernel_sendmsg(transport->sock, &msg, &iov, 1, iov.iov_len);
2663 if (len != iov.iov_len)
2664 return -EAGAIN;
2665
2666 tailpage = NULL;
2667 if (xbufp->tail[0].iov_len)
2668 tailpage = virt_to_page(xbufp->tail[0].iov_base);
2669 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2670 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2671 len = svc_send_common(transport->sock, xbufp,
2672 virt_to_page(xbufp->head[0].iov_base), headoff,
2673 tailpage, tailoff);
2674 if (len != xbufp->len)
2675 return -EAGAIN;
2676 return len;
2677}
2678
2679/*
2680 * The send routine. Borrows from svc_send
2681 */
2682static int bc_send_request(struct rpc_rqst *req)
2683{
2684 struct svc_xprt *xprt;
2685 int len;
2686
2687 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2688 /*
2689 * Get the server socket associated with this callback xprt
2690 */
2691 xprt = req->rq_xprt->bc_xprt;
2692
2693 /*
2694 * Grab the mutex to serialize data as the connection is shared
2695 * with the fore channel
2696 */
2697 mutex_lock(&xprt->xpt_mutex);
2698 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2699 len = -ENOTCONN;
2700 else
2701 len = bc_sendto(req);
2702 mutex_unlock(&xprt->xpt_mutex);
2703
2704 if (len > 0)
2705 len = 0;
2706
2707 return len;
2708}
2709
2710/*
2711 * The close routine. Since this is client initiated, we do nothing
2712 */
2713
2714static void bc_close(struct rpc_xprt *xprt)
2715{
2716}
2717
2718/*
2719 * The xprt destroy routine. Again, because this connection is client
2720 * initiated, we do nothing
2721 */
2722
2723static void bc_destroy(struct rpc_xprt *xprt)
2724{
2725 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2726
2727 xs_xprt_free(xprt);
2728 module_put(THIS_MODULE);
2729}
2730
2731static const struct rpc_xprt_ops xs_local_ops = {
2732 .reserve_xprt = xprt_reserve_xprt,
2733 .release_xprt = xprt_release_xprt,
2734 .alloc_slot = xprt_alloc_slot,
2735 .free_slot = xprt_free_slot,
2736 .rpcbind = xs_local_rpcbind,
2737 .set_port = xs_local_set_port,
2738 .connect = xs_local_connect,
2739 .buf_alloc = rpc_malloc,
2740 .buf_free = rpc_free,
2741 .prepare_request = xs_stream_prepare_request,
2742 .send_request = xs_local_send_request,
2743 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2744 .close = xs_close,
2745 .destroy = xs_destroy,
2746 .print_stats = xs_local_print_stats,
2747 .enable_swap = xs_enable_swap,
2748 .disable_swap = xs_disable_swap,
2749};
2750
2751static const struct rpc_xprt_ops xs_udp_ops = {
2752 .set_buffer_size = xs_udp_set_buffer_size,
2753 .reserve_xprt = xprt_reserve_xprt_cong,
2754 .release_xprt = xprt_release_xprt_cong,
2755 .alloc_slot = xprt_alloc_slot,
2756 .free_slot = xprt_free_slot,
2757 .rpcbind = rpcb_getport_async,
2758 .set_port = xs_set_port,
2759 .connect = xs_connect,
2760 .buf_alloc = rpc_malloc,
2761 .buf_free = rpc_free,
2762 .send_request = xs_udp_send_request,
2763 .wait_for_reply_request = xprt_wait_for_reply_request_rtt,
2764 .timer = xs_udp_timer,
2765 .release_request = xprt_release_rqst_cong,
2766 .close = xs_close,
2767 .destroy = xs_destroy,
2768 .print_stats = xs_udp_print_stats,
2769 .enable_swap = xs_enable_swap,
2770 .disable_swap = xs_disable_swap,
2771 .inject_disconnect = xs_inject_disconnect,
2772};
2773
2774static const struct rpc_xprt_ops xs_tcp_ops = {
2775 .reserve_xprt = xprt_reserve_xprt,
2776 .release_xprt = xprt_release_xprt,
2777 .alloc_slot = xprt_alloc_slot,
2778 .free_slot = xprt_free_slot,
2779 .rpcbind = rpcb_getport_async,
2780 .set_port = xs_set_port,
2781 .connect = xs_connect,
2782 .buf_alloc = rpc_malloc,
2783 .buf_free = rpc_free,
2784 .prepare_request = xs_stream_prepare_request,
2785 .send_request = xs_tcp_send_request,
2786 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2787 .close = xs_tcp_shutdown,
2788 .destroy = xs_destroy,
2789 .set_connect_timeout = xs_tcp_set_connect_timeout,
2790 .print_stats = xs_tcp_print_stats,
2791 .enable_swap = xs_enable_swap,
2792 .disable_swap = xs_disable_swap,
2793 .inject_disconnect = xs_inject_disconnect,
2794#ifdef CONFIG_SUNRPC_BACKCHANNEL
2795 .bc_setup = xprt_setup_bc,
2796 .bc_maxpayload = xs_tcp_bc_maxpayload,
2797 .bc_num_slots = xprt_bc_max_slots,
2798 .bc_free_rqst = xprt_free_bc_rqst,
2799 .bc_destroy = xprt_destroy_bc,
2800#endif
2801};
2802
2803/*
2804 * The rpc_xprt_ops for the server backchannel
2805 */
2806
2807static const struct rpc_xprt_ops bc_tcp_ops = {
2808 .reserve_xprt = xprt_reserve_xprt,
2809 .release_xprt = xprt_release_xprt,
2810 .alloc_slot = xprt_alloc_slot,
2811 .free_slot = xprt_free_slot,
2812 .buf_alloc = bc_malloc,
2813 .buf_free = bc_free,
2814 .send_request = bc_send_request,
2815 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2816 .close = bc_close,
2817 .destroy = bc_destroy,
2818 .print_stats = xs_tcp_print_stats,
2819 .enable_swap = xs_enable_swap,
2820 .disable_swap = xs_disable_swap,
2821 .inject_disconnect = xs_inject_disconnect,
2822};
2823
2824static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2825{
2826 static const struct sockaddr_in sin = {
2827 .sin_family = AF_INET,
2828 .sin_addr.s_addr = htonl(INADDR_ANY),
2829 };
2830 static const struct sockaddr_in6 sin6 = {
2831 .sin6_family = AF_INET6,
2832 .sin6_addr = IN6ADDR_ANY_INIT,
2833 };
2834
2835 switch (family) {
2836 case AF_LOCAL:
2837 break;
2838 case AF_INET:
2839 memcpy(sap, &sin, sizeof(sin));
2840 break;
2841 case AF_INET6:
2842 memcpy(sap, &sin6, sizeof(sin6));
2843 break;
2844 default:
2845 dprintk("RPC: %s: Bad address family\n", __func__);
2846 return -EAFNOSUPPORT;
2847 }
2848 return 0;
2849}
2850
2851static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2852 unsigned int slot_table_size,
2853 unsigned int max_slot_table_size)
2854{
2855 struct rpc_xprt *xprt;
2856 struct sock_xprt *new;
2857
2858 if (args->addrlen > sizeof(xprt->addr)) {
2859 dprintk("RPC: xs_setup_xprt: address too large\n");
2860 return ERR_PTR(-EBADF);
2861 }
2862
2863 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2864 max_slot_table_size);
2865 if (xprt == NULL) {
2866 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2867 "rpc_xprt\n");
2868 return ERR_PTR(-ENOMEM);
2869 }
2870
2871 new = container_of(xprt, struct sock_xprt, xprt);
2872 mutex_init(&new->recv_mutex);
2873 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2874 xprt->addrlen = args->addrlen;
2875 if (args->srcaddr)
2876 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2877 else {
2878 int err;
2879 err = xs_init_anyaddr(args->dstaddr->sa_family,
2880 (struct sockaddr *)&new->srcaddr);
2881 if (err != 0) {
2882 xprt_free(xprt);
2883 return ERR_PTR(err);
2884 }
2885 }
2886
2887 return xprt;
2888}
2889
2890static const struct rpc_timeout xs_local_default_timeout = {
2891 .to_initval = 10 * HZ,
2892 .to_maxval = 10 * HZ,
2893 .to_retries = 2,
2894};
2895
2896/**
2897 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2898 * @args: rpc transport creation arguments
2899 *
2900 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2901 */
2902static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2903{
2904 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2905 struct sock_xprt *transport;
2906 struct rpc_xprt *xprt;
2907 struct rpc_xprt *ret;
2908
2909 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2910 xprt_max_tcp_slot_table_entries);
2911 if (IS_ERR(xprt))
2912 return xprt;
2913 transport = container_of(xprt, struct sock_xprt, xprt);
2914
2915 xprt->prot = 0;
2916 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2917
2918 xprt->bind_timeout = XS_BIND_TO;
2919 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2920 xprt->idle_timeout = XS_IDLE_DISC_TO;
2921
2922 xprt->ops = &xs_local_ops;
2923 xprt->timeout = &xs_local_default_timeout;
2924
2925 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2926 INIT_WORK(&transport->error_worker, xs_error_handle);
2927 INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket);
2928
2929 switch (sun->sun_family) {
2930 case AF_LOCAL:
2931 if (sun->sun_path[0] != '/') {
2932 dprintk("RPC: bad AF_LOCAL address: %s\n",
2933 sun->sun_path);
2934 ret = ERR_PTR(-EINVAL);
2935 goto out_err;
2936 }
2937 xprt_set_bound(xprt);
2938 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2939 ret = ERR_PTR(xs_local_setup_socket(transport));
2940 if (ret)
2941 goto out_err;
2942 break;
2943 default:
2944 ret = ERR_PTR(-EAFNOSUPPORT);
2945 goto out_err;
2946 }
2947
2948 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2949 xprt->address_strings[RPC_DISPLAY_ADDR]);
2950
2951 if (try_module_get(THIS_MODULE))
2952 return xprt;
2953 ret = ERR_PTR(-EINVAL);
2954out_err:
2955 xs_xprt_free(xprt);
2956 return ret;
2957}
2958
2959static const struct rpc_timeout xs_udp_default_timeout = {
2960 .to_initval = 5 * HZ,
2961 .to_maxval = 30 * HZ,
2962 .to_increment = 5 * HZ,
2963 .to_retries = 5,
2964};
2965
2966/**
2967 * xs_setup_udp - Set up transport to use a UDP socket
2968 * @args: rpc transport creation arguments
2969 *
2970 */
2971static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2972{
2973 struct sockaddr *addr = args->dstaddr;
2974 struct rpc_xprt *xprt;
2975 struct sock_xprt *transport;
2976 struct rpc_xprt *ret;
2977
2978 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2979 xprt_udp_slot_table_entries);
2980 if (IS_ERR(xprt))
2981 return xprt;
2982 transport = container_of(xprt, struct sock_xprt, xprt);
2983
2984 xprt->prot = IPPROTO_UDP;
2985 /* XXX: header size can vary due to auth type, IPv6, etc. */
2986 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2987
2988 xprt->bind_timeout = XS_BIND_TO;
2989 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2990 xprt->idle_timeout = XS_IDLE_DISC_TO;
2991
2992 xprt->ops = &xs_udp_ops;
2993
2994 xprt->timeout = &xs_udp_default_timeout;
2995
2996 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
2997 INIT_WORK(&transport->error_worker, xs_error_handle);
2998 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2999
3000 switch (addr->sa_family) {
3001 case AF_INET:
3002 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3003 xprt_set_bound(xprt);
3004
3005 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
3006 break;
3007 case AF_INET6:
3008 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3009 xprt_set_bound(xprt);
3010
3011 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
3012 break;
3013 default:
3014 ret = ERR_PTR(-EAFNOSUPPORT);
3015 goto out_err;
3016 }
3017
3018 if (xprt_bound(xprt))
3019 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3020 xprt->address_strings[RPC_DISPLAY_ADDR],
3021 xprt->address_strings[RPC_DISPLAY_PORT],
3022 xprt->address_strings[RPC_DISPLAY_PROTO]);
3023 else
3024 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
3025 xprt->address_strings[RPC_DISPLAY_ADDR],
3026 xprt->address_strings[RPC_DISPLAY_PROTO]);
3027
3028 if (try_module_get(THIS_MODULE))
3029 return xprt;
3030 ret = ERR_PTR(-EINVAL);
3031out_err:
3032 xs_xprt_free(xprt);
3033 return ret;
3034}
3035
3036static const struct rpc_timeout xs_tcp_default_timeout = {
3037 .to_initval = 60 * HZ,
3038 .to_maxval = 60 * HZ,
3039 .to_retries = 2,
3040};
3041
3042/**
3043 * xs_setup_tcp - Set up transport to use a TCP socket
3044 * @args: rpc transport creation arguments
3045 *
3046 */
3047static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
3048{
3049 struct sockaddr *addr = args->dstaddr;
3050 struct rpc_xprt *xprt;
3051 struct sock_xprt *transport;
3052 struct rpc_xprt *ret;
3053 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
3054
3055 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
3056 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
3057
3058 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3059 max_slot_table_size);
3060 if (IS_ERR(xprt))
3061 return xprt;
3062 transport = container_of(xprt, struct sock_xprt, xprt);
3063
3064 xprt->prot = IPPROTO_TCP;
3065 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3066
3067 xprt->bind_timeout = XS_BIND_TO;
3068 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3069 xprt->idle_timeout = XS_IDLE_DISC_TO;
3070
3071 xprt->ops = &xs_tcp_ops;
3072 xprt->timeout = &xs_tcp_default_timeout;
3073
3074 xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3075 xprt->connect_timeout = xprt->timeout->to_initval *
3076 (xprt->timeout->to_retries + 1);
3077
3078 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3079 INIT_WORK(&transport->error_worker, xs_error_handle);
3080 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
3081
3082 switch (addr->sa_family) {
3083 case AF_INET:
3084 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3085 xprt_set_bound(xprt);
3086
3087 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3088 break;
3089 case AF_INET6:
3090 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3091 xprt_set_bound(xprt);
3092
3093 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3094 break;
3095 default:
3096 ret = ERR_PTR(-EAFNOSUPPORT);
3097 goto out_err;
3098 }
3099
3100 if (xprt_bound(xprt))
3101 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3102 xprt->address_strings[RPC_DISPLAY_ADDR],
3103 xprt->address_strings[RPC_DISPLAY_PORT],
3104 xprt->address_strings[RPC_DISPLAY_PROTO]);
3105 else
3106 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
3107 xprt->address_strings[RPC_DISPLAY_ADDR],
3108 xprt->address_strings[RPC_DISPLAY_PROTO]);
3109
3110 if (try_module_get(THIS_MODULE))
3111 return xprt;
3112 ret = ERR_PTR(-EINVAL);
3113out_err:
3114 xs_xprt_free(xprt);
3115 return ret;
3116}
3117
3118/**
3119 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3120 * @args: rpc transport creation arguments
3121 *
3122 */
3123static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3124{
3125 struct sockaddr *addr = args->dstaddr;
3126 struct rpc_xprt *xprt;
3127 struct sock_xprt *transport;
3128 struct svc_sock *bc_sock;
3129 struct rpc_xprt *ret;
3130
3131 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3132 xprt_tcp_slot_table_entries);
3133 if (IS_ERR(xprt))
3134 return xprt;
3135 transport = container_of(xprt, struct sock_xprt, xprt);
3136
3137 xprt->prot = IPPROTO_TCP;
3138 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3139 xprt->timeout = &xs_tcp_default_timeout;
3140
3141 /* backchannel */
3142 xprt_set_bound(xprt);
3143 xprt->bind_timeout = 0;
3144 xprt->reestablish_timeout = 0;
3145 xprt->idle_timeout = 0;
3146
3147 xprt->ops = &bc_tcp_ops;
3148
3149 switch (addr->sa_family) {
3150 case AF_INET:
3151 xs_format_peer_addresses(xprt, "tcp",
3152 RPCBIND_NETID_TCP);
3153 break;
3154 case AF_INET6:
3155 xs_format_peer_addresses(xprt, "tcp",
3156 RPCBIND_NETID_TCP6);
3157 break;
3158 default:
3159 ret = ERR_PTR(-EAFNOSUPPORT);
3160 goto out_err;
3161 }
3162
3163 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3164 xprt->address_strings[RPC_DISPLAY_ADDR],
3165 xprt->address_strings[RPC_DISPLAY_PORT],
3166 xprt->address_strings[RPC_DISPLAY_PROTO]);
3167
3168 /*
3169 * Once we've associated a backchannel xprt with a connection,
3170 * we want to keep it around as long as the connection lasts,
3171 * in case we need to start using it for a backchannel again;
3172 * this reference won't be dropped until bc_xprt is destroyed.
3173 */
3174 xprt_get(xprt);
3175 args->bc_xprt->xpt_bc_xprt = xprt;
3176 xprt->bc_xprt = args->bc_xprt;
3177 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3178 transport->sock = bc_sock->sk_sock;
3179 transport->inet = bc_sock->sk_sk;
3180
3181 /*
3182 * Since we don't want connections for the backchannel, we set
3183 * the xprt status to connected
3184 */
3185 xprt_set_connected(xprt);
3186
3187 if (try_module_get(THIS_MODULE))
3188 return xprt;
3189
3190 args->bc_xprt->xpt_bc_xprt = NULL;
3191 args->bc_xprt->xpt_bc_xps = NULL;
3192 xprt_put(xprt);
3193 ret = ERR_PTR(-EINVAL);
3194out_err:
3195 xs_xprt_free(xprt);
3196 return ret;
3197}
3198
3199static struct xprt_class xs_local_transport = {
3200 .list = LIST_HEAD_INIT(xs_local_transport.list),
3201 .name = "named UNIX socket",
3202 .owner = THIS_MODULE,
3203 .ident = XPRT_TRANSPORT_LOCAL,
3204 .setup = xs_setup_local,
3205};
3206
3207static struct xprt_class xs_udp_transport = {
3208 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3209 .name = "udp",
3210 .owner = THIS_MODULE,
3211 .ident = XPRT_TRANSPORT_UDP,
3212 .setup = xs_setup_udp,
3213};
3214
3215static struct xprt_class xs_tcp_transport = {
3216 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3217 .name = "tcp",
3218 .owner = THIS_MODULE,
3219 .ident = XPRT_TRANSPORT_TCP,
3220 .setup = xs_setup_tcp,
3221};
3222
3223static struct xprt_class xs_bc_tcp_transport = {
3224 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3225 .name = "tcp NFSv4.1 backchannel",
3226 .owner = THIS_MODULE,
3227 .ident = XPRT_TRANSPORT_BC_TCP,
3228 .setup = xs_setup_bc_tcp,
3229};
3230
3231/**
3232 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3233 *
3234 */
3235int init_socket_xprt(void)
3236{
3237 if (!sunrpc_table_header)
3238 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3239
3240 xprt_register_transport(&xs_local_transport);
3241 xprt_register_transport(&xs_udp_transport);
3242 xprt_register_transport(&xs_tcp_transport);
3243 xprt_register_transport(&xs_bc_tcp_transport);
3244
3245 return 0;
3246}
3247
3248/**
3249 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3250 *
3251 */
3252void cleanup_socket_xprt(void)
3253{
3254 if (sunrpc_table_header) {
3255 unregister_sysctl_table(sunrpc_table_header);
3256 sunrpc_table_header = NULL;
3257 }
3258
3259 xprt_unregister_transport(&xs_local_transport);
3260 xprt_unregister_transport(&xs_udp_transport);
3261 xprt_unregister_transport(&xs_tcp_transport);
3262 xprt_unregister_transport(&xs_bc_tcp_transport);
3263}
3264
3265static int param_set_uint_minmax(const char *val,
3266 const struct kernel_param *kp,
3267 unsigned int min, unsigned int max)
3268{
3269 unsigned int num;
3270 int ret;
3271
3272 if (!val)
3273 return -EINVAL;
3274 ret = kstrtouint(val, 0, &num);
3275 if (ret)
3276 return ret;
3277 if (num < min || num > max)
3278 return -EINVAL;
3279 *((unsigned int *)kp->arg) = num;
3280 return 0;
3281}
3282
3283static int param_set_portnr(const char *val, const struct kernel_param *kp)
3284{
3285 return param_set_uint_minmax(val, kp,
3286 RPC_MIN_RESVPORT,
3287 RPC_MAX_RESVPORT);
3288}
3289
3290static const struct kernel_param_ops param_ops_portnr = {
3291 .set = param_set_portnr,
3292 .get = param_get_uint,
3293};
3294
3295#define param_check_portnr(name, p) \
3296 __param_check(name, p, unsigned int);
3297
3298module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3299module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3300
3301static int param_set_slot_table_size(const char *val,
3302 const struct kernel_param *kp)
3303{
3304 return param_set_uint_minmax(val, kp,
3305 RPC_MIN_SLOT_TABLE,
3306 RPC_MAX_SLOT_TABLE);
3307}
3308
3309static const struct kernel_param_ops param_ops_slot_table_size = {
3310 .set = param_set_slot_table_size,
3311 .get = param_get_uint,
3312};
3313
3314#define param_check_slot_table_size(name, p) \
3315 __param_check(name, p, unsigned int);
3316
3317static int param_set_max_slot_table_size(const char *val,
3318 const struct kernel_param *kp)
3319{
3320 return param_set_uint_minmax(val, kp,
3321 RPC_MIN_SLOT_TABLE,
3322 RPC_MAX_SLOT_TABLE_LIMIT);
3323}
3324
3325static const struct kernel_param_ops param_ops_max_slot_table_size = {
3326 .set = param_set_max_slot_table_size,
3327 .get = param_get_uint,
3328};
3329
3330#define param_check_max_slot_table_size(name, p) \
3331 __param_check(name, p, unsigned int);
3332
3333module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3334 slot_table_size, 0644);
3335module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3336 max_slot_table_size, 0644);
3337module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3338 slot_table_size, 0644);