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 <net/tls_prot.h>
51#include <net/handshake.h>
52
53#include <linux/bvec.h>
54#include <linux/highmem.h>
55#include <linux/uio.h>
56#include <linux/sched/mm.h>
57
58#include <trace/events/sock.h>
59#include <trace/events/sunrpc.h>
60
61#include "socklib.h"
62#include "sunrpc.h"
63
64static void xs_close(struct rpc_xprt *xprt);
65static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock);
66static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
67 struct socket *sock);
68
69/*
70 * xprtsock tunables
71 */
72static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
73static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
74static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
75
76static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
77static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
78
79#define XS_TCP_LINGER_TO (15U * HZ)
80static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
81
82/*
83 * We can register our own files under /proc/sys/sunrpc by
84 * calling register_sysctl() again. The files in that
85 * directory become the union of all files registered there.
86 *
87 * We simply need to make sure that we don't collide with
88 * someone else's file names!
89 */
90
91static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
92static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
93static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
94static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
95static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
96
97static struct ctl_table_header *sunrpc_table_header;
98
99static struct xprt_class xs_local_transport;
100static struct xprt_class xs_udp_transport;
101static struct xprt_class xs_tcp_transport;
102static struct xprt_class xs_tcp_tls_transport;
103static struct xprt_class xs_bc_tcp_transport;
104
105/*
106 * FIXME: changing the UDP slot table size should also resize the UDP
107 * socket buffers for existing UDP transports
108 */
109static struct ctl_table xs_tunables_table[] = {
110 {
111 .procname = "udp_slot_table_entries",
112 .data = &xprt_udp_slot_table_entries,
113 .maxlen = sizeof(unsigned int),
114 .mode = 0644,
115 .proc_handler = proc_dointvec_minmax,
116 .extra1 = &min_slot_table_size,
117 .extra2 = &max_slot_table_size
118 },
119 {
120 .procname = "tcp_slot_table_entries",
121 .data = &xprt_tcp_slot_table_entries,
122 .maxlen = sizeof(unsigned int),
123 .mode = 0644,
124 .proc_handler = proc_dointvec_minmax,
125 .extra1 = &min_slot_table_size,
126 .extra2 = &max_slot_table_size
127 },
128 {
129 .procname = "tcp_max_slot_table_entries",
130 .data = &xprt_max_tcp_slot_table_entries,
131 .maxlen = sizeof(unsigned int),
132 .mode = 0644,
133 .proc_handler = proc_dointvec_minmax,
134 .extra1 = &min_slot_table_size,
135 .extra2 = &max_tcp_slot_table_limit
136 },
137 {
138 .procname = "min_resvport",
139 .data = &xprt_min_resvport,
140 .maxlen = sizeof(unsigned int),
141 .mode = 0644,
142 .proc_handler = proc_dointvec_minmax,
143 .extra1 = &xprt_min_resvport_limit,
144 .extra2 = &xprt_max_resvport_limit
145 },
146 {
147 .procname = "max_resvport",
148 .data = &xprt_max_resvport,
149 .maxlen = sizeof(unsigned int),
150 .mode = 0644,
151 .proc_handler = proc_dointvec_minmax,
152 .extra1 = &xprt_min_resvport_limit,
153 .extra2 = &xprt_max_resvport_limit
154 },
155 {
156 .procname = "tcp_fin_timeout",
157 .data = &xs_tcp_fin_timeout,
158 .maxlen = sizeof(xs_tcp_fin_timeout),
159 .mode = 0644,
160 .proc_handler = proc_dointvec_jiffies,
161 },
162 { },
163};
164
165/*
166 * Wait duration for a reply from the RPC portmapper.
167 */
168#define XS_BIND_TO (60U * HZ)
169
170/*
171 * Delay if a UDP socket connect error occurs. This is most likely some
172 * kind of resource problem on the local host.
173 */
174#define XS_UDP_REEST_TO (2U * HZ)
175
176/*
177 * The reestablish timeout allows clients to delay for a bit before attempting
178 * to reconnect to a server that just dropped our connection.
179 *
180 * We implement an exponential backoff when trying to reestablish a TCP
181 * transport connection with the server. Some servers like to drop a TCP
182 * connection when they are overworked, so we start with a short timeout and
183 * increase over time if the server is down or not responding.
184 */
185#define XS_TCP_INIT_REEST_TO (3U * HZ)
186
187/*
188 * TCP idle timeout; client drops the transport socket if it is idle
189 * for this long. Note that we also timeout UDP sockets to prevent
190 * holding port numbers when there is no RPC traffic.
191 */
192#define XS_IDLE_DISC_TO (5U * 60 * HZ)
193
194/*
195 * TLS handshake timeout.
196 */
197#define XS_TLS_HANDSHAKE_TO (10U * HZ)
198
199#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
200# undef RPC_DEBUG_DATA
201# define RPCDBG_FACILITY RPCDBG_TRANS
202#endif
203
204#ifdef RPC_DEBUG_DATA
205static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
206{
207 u8 *buf = (u8 *) packet;
208 int j;
209
210 dprintk("RPC: %s\n", msg);
211 for (j = 0; j < count && j < 128; j += 4) {
212 if (!(j & 31)) {
213 if (j)
214 dprintk("\n");
215 dprintk("0x%04x ", j);
216 }
217 dprintk("%02x%02x%02x%02x ",
218 buf[j], buf[j+1], buf[j+2], buf[j+3]);
219 }
220 dprintk("\n");
221}
222#else
223static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
224{
225 /* NOP */
226}
227#endif
228
229static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
230{
231 return (struct rpc_xprt *) sk->sk_user_data;
232}
233
234static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
235{
236 return (struct sockaddr *) &xprt->addr;
237}
238
239static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
240{
241 return (struct sockaddr_un *) &xprt->addr;
242}
243
244static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
245{
246 return (struct sockaddr_in *) &xprt->addr;
247}
248
249static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
250{
251 return (struct sockaddr_in6 *) &xprt->addr;
252}
253
254static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
255{
256 struct sockaddr *sap = xs_addr(xprt);
257 struct sockaddr_in6 *sin6;
258 struct sockaddr_in *sin;
259 struct sockaddr_un *sun;
260 char buf[128];
261
262 switch (sap->sa_family) {
263 case AF_LOCAL:
264 sun = xs_addr_un(xprt);
265 if (sun->sun_path[0]) {
266 strscpy(buf, sun->sun_path, sizeof(buf));
267 } else {
268 buf[0] = '@';
269 strscpy(buf+1, sun->sun_path+1, sizeof(buf)-1);
270 }
271 xprt->address_strings[RPC_DISPLAY_ADDR] =
272 kstrdup(buf, GFP_KERNEL);
273 break;
274 case AF_INET:
275 (void)rpc_ntop(sap, buf, sizeof(buf));
276 xprt->address_strings[RPC_DISPLAY_ADDR] =
277 kstrdup(buf, GFP_KERNEL);
278 sin = xs_addr_in(xprt);
279 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
280 break;
281 case AF_INET6:
282 (void)rpc_ntop(sap, buf, sizeof(buf));
283 xprt->address_strings[RPC_DISPLAY_ADDR] =
284 kstrdup(buf, GFP_KERNEL);
285 sin6 = xs_addr_in6(xprt);
286 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
287 break;
288 default:
289 BUG();
290 }
291
292 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
293}
294
295static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
296{
297 struct sockaddr *sap = xs_addr(xprt);
298 char buf[128];
299
300 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
301 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
302
303 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
304 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
305}
306
307static void xs_format_peer_addresses(struct rpc_xprt *xprt,
308 const char *protocol,
309 const char *netid)
310{
311 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
312 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
313 xs_format_common_peer_addresses(xprt);
314 xs_format_common_peer_ports(xprt);
315}
316
317static void xs_update_peer_port(struct rpc_xprt *xprt)
318{
319 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
320 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
321
322 xs_format_common_peer_ports(xprt);
323}
324
325static void xs_free_peer_addresses(struct rpc_xprt *xprt)
326{
327 unsigned int i;
328
329 for (i = 0; i < RPC_DISPLAY_MAX; i++)
330 switch (i) {
331 case RPC_DISPLAY_PROTO:
332 case RPC_DISPLAY_NETID:
333 continue;
334 default:
335 kfree(xprt->address_strings[i]);
336 }
337}
338
339static size_t
340xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp)
341{
342 size_t i,n;
343
344 if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES))
345 return want;
346 n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT;
347 for (i = 0; i < n; i++) {
348 if (buf->pages[i])
349 continue;
350 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp);
351 if (!buf->pages[i]) {
352 i *= PAGE_SIZE;
353 return i > buf->page_base ? i - buf->page_base : 0;
354 }
355 }
356 return want;
357}
358
359static int
360xs_sock_process_cmsg(struct socket *sock, struct msghdr *msg,
361 struct cmsghdr *cmsg, int ret)
362{
363 u8 content_type = tls_get_record_type(sock->sk, cmsg);
364 u8 level, description;
365
366 switch (content_type) {
367 case 0:
368 break;
369 case TLS_RECORD_TYPE_DATA:
370 /* TLS sets EOR at the end of each application data
371 * record, even though there might be more frames
372 * waiting to be decrypted.
373 */
374 msg->msg_flags &= ~MSG_EOR;
375 break;
376 case TLS_RECORD_TYPE_ALERT:
377 tls_alert_recv(sock->sk, msg, &level, &description);
378 ret = (level == TLS_ALERT_LEVEL_FATAL) ?
379 -EACCES : -EAGAIN;
380 break;
381 default:
382 /* discard this record type */
383 ret = -EAGAIN;
384 }
385 return ret;
386}
387
388static int
389xs_sock_recv_cmsg(struct socket *sock, struct msghdr *msg, int flags)
390{
391 union {
392 struct cmsghdr cmsg;
393 u8 buf[CMSG_SPACE(sizeof(u8))];
394 } u;
395 int ret;
396
397 msg->msg_control = &u;
398 msg->msg_controllen = sizeof(u);
399 ret = sock_recvmsg(sock, msg, flags);
400 if (msg->msg_controllen != sizeof(u))
401 ret = xs_sock_process_cmsg(sock, msg, &u.cmsg, ret);
402 return ret;
403}
404
405static ssize_t
406xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek)
407{
408 ssize_t ret;
409 if (seek != 0)
410 iov_iter_advance(&msg->msg_iter, seek);
411 ret = xs_sock_recv_cmsg(sock, msg, flags);
412 return ret > 0 ? ret + seek : ret;
413}
414
415static ssize_t
416xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags,
417 struct kvec *kvec, size_t count, size_t seek)
418{
419 iov_iter_kvec(&msg->msg_iter, ITER_DEST, kvec, 1, count);
420 return xs_sock_recvmsg(sock, msg, flags, seek);
421}
422
423static ssize_t
424xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags,
425 struct bio_vec *bvec, unsigned long nr, size_t count,
426 size_t seek)
427{
428 iov_iter_bvec(&msg->msg_iter, ITER_DEST, bvec, nr, count);
429 return xs_sock_recvmsg(sock, msg, flags, seek);
430}
431
432static ssize_t
433xs_read_discard(struct socket *sock, struct msghdr *msg, int flags,
434 size_t count)
435{
436 iov_iter_discard(&msg->msg_iter, ITER_DEST, count);
437 return xs_sock_recv_cmsg(sock, msg, flags);
438}
439
440#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
441static void
442xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
443{
444 struct bvec_iter bi = {
445 .bi_size = count,
446 };
447 struct bio_vec bv;
448
449 bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
450 for_each_bvec(bv, bvec, bi, bi)
451 flush_dcache_page(bv.bv_page);
452}
453#else
454static inline void
455xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
456{
457}
458#endif
459
460static ssize_t
461xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
462 struct xdr_buf *buf, size_t count, size_t seek, size_t *read)
463{
464 size_t want, seek_init = seek, offset = 0;
465 ssize_t ret;
466
467 want = min_t(size_t, count, buf->head[0].iov_len);
468 if (seek < want) {
469 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek);
470 if (ret <= 0)
471 goto sock_err;
472 offset += ret;
473 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
474 goto out;
475 if (ret != want)
476 goto out;
477 seek = 0;
478 } else {
479 seek -= want;
480 offset += want;
481 }
482
483 want = xs_alloc_sparse_pages(
484 buf, min_t(size_t, count - offset, buf->page_len),
485 GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN);
486 if (seek < want) {
487 ret = xs_read_bvec(sock, msg, flags, buf->bvec,
488 xdr_buf_pagecount(buf),
489 want + buf->page_base,
490 seek + buf->page_base);
491 if (ret <= 0)
492 goto sock_err;
493 xs_flush_bvec(buf->bvec, ret, seek + buf->page_base);
494 ret -= buf->page_base;
495 offset += ret;
496 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
497 goto out;
498 if (ret != want)
499 goto out;
500 seek = 0;
501 } else {
502 seek -= want;
503 offset += want;
504 }
505
506 want = min_t(size_t, count - offset, buf->tail[0].iov_len);
507 if (seek < want) {
508 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek);
509 if (ret <= 0)
510 goto sock_err;
511 offset += ret;
512 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
513 goto out;
514 if (ret != want)
515 goto out;
516 } else if (offset < seek_init)
517 offset = seek_init;
518 ret = -EMSGSIZE;
519out:
520 *read = offset - seek_init;
521 return ret;
522sock_err:
523 offset += seek;
524 goto out;
525}
526
527static void
528xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf)
529{
530 if (!transport->recv.copied) {
531 if (buf->head[0].iov_len >= transport->recv.offset)
532 memcpy(buf->head[0].iov_base,
533 &transport->recv.xid,
534 transport->recv.offset);
535 transport->recv.copied = transport->recv.offset;
536 }
537}
538
539static bool
540xs_read_stream_request_done(struct sock_xprt *transport)
541{
542 return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT);
543}
544
545static void
546xs_read_stream_check_eor(struct sock_xprt *transport,
547 struct msghdr *msg)
548{
549 if (xs_read_stream_request_done(transport))
550 msg->msg_flags |= MSG_EOR;
551}
552
553static ssize_t
554xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg,
555 int flags, struct rpc_rqst *req)
556{
557 struct xdr_buf *buf = &req->rq_private_buf;
558 size_t want, read;
559 ssize_t ret;
560
561 xs_read_header(transport, buf);
562
563 want = transport->recv.len - transport->recv.offset;
564 if (want != 0) {
565 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf,
566 transport->recv.copied + want,
567 transport->recv.copied,
568 &read);
569 transport->recv.offset += read;
570 transport->recv.copied += read;
571 }
572
573 if (transport->recv.offset == transport->recv.len)
574 xs_read_stream_check_eor(transport, msg);
575
576 if (want == 0)
577 return 0;
578
579 switch (ret) {
580 default:
581 break;
582 case -EFAULT:
583 case -EMSGSIZE:
584 msg->msg_flags |= MSG_TRUNC;
585 return read;
586 case 0:
587 return -ESHUTDOWN;
588 }
589 return ret < 0 ? ret : read;
590}
591
592static size_t
593xs_read_stream_headersize(bool isfrag)
594{
595 if (isfrag)
596 return sizeof(__be32);
597 return 3 * sizeof(__be32);
598}
599
600static ssize_t
601xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg,
602 int flags, size_t want, size_t seek)
603{
604 struct kvec kvec = {
605 .iov_base = &transport->recv.fraghdr,
606 .iov_len = want,
607 };
608 return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek);
609}
610
611#if defined(CONFIG_SUNRPC_BACKCHANNEL)
612static ssize_t
613xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
614{
615 struct rpc_xprt *xprt = &transport->xprt;
616 struct rpc_rqst *req;
617 ssize_t ret;
618
619 /* Is this transport associated with the backchannel? */
620 if (!xprt->bc_serv)
621 return -ESHUTDOWN;
622
623 /* Look up and lock the request corresponding to the given XID */
624 req = xprt_lookup_bc_request(xprt, transport->recv.xid);
625 if (!req) {
626 printk(KERN_WARNING "Callback slot table overflowed\n");
627 return -ESHUTDOWN;
628 }
629 if (transport->recv.copied && !req->rq_private_buf.len)
630 return -ESHUTDOWN;
631
632 ret = xs_read_stream_request(transport, msg, flags, req);
633 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
634 xprt_complete_bc_request(req, transport->recv.copied);
635 else
636 req->rq_private_buf.len = transport->recv.copied;
637
638 return ret;
639}
640#else /* CONFIG_SUNRPC_BACKCHANNEL */
641static ssize_t
642xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
643{
644 return -ESHUTDOWN;
645}
646#endif /* CONFIG_SUNRPC_BACKCHANNEL */
647
648static ssize_t
649xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags)
650{
651 struct rpc_xprt *xprt = &transport->xprt;
652 struct rpc_rqst *req;
653 ssize_t ret = 0;
654
655 /* Look up and lock the request corresponding to the given XID */
656 spin_lock(&xprt->queue_lock);
657 req = xprt_lookup_rqst(xprt, transport->recv.xid);
658 if (!req || (transport->recv.copied && !req->rq_private_buf.len)) {
659 msg->msg_flags |= MSG_TRUNC;
660 goto out;
661 }
662 xprt_pin_rqst(req);
663 spin_unlock(&xprt->queue_lock);
664
665 ret = xs_read_stream_request(transport, msg, flags, req);
666
667 spin_lock(&xprt->queue_lock);
668 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
669 xprt_complete_rqst(req->rq_task, transport->recv.copied);
670 else
671 req->rq_private_buf.len = transport->recv.copied;
672 xprt_unpin_rqst(req);
673out:
674 spin_unlock(&xprt->queue_lock);
675 return ret;
676}
677
678static ssize_t
679xs_read_stream(struct sock_xprt *transport, int flags)
680{
681 struct msghdr msg = { 0 };
682 size_t want, read = 0;
683 ssize_t ret = 0;
684
685 if (transport->recv.len == 0) {
686 want = xs_read_stream_headersize(transport->recv.copied != 0);
687 ret = xs_read_stream_header(transport, &msg, flags, want,
688 transport->recv.offset);
689 if (ret <= 0)
690 goto out_err;
691 transport->recv.offset = ret;
692 if (transport->recv.offset != want)
693 return transport->recv.offset;
694 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) &
695 RPC_FRAGMENT_SIZE_MASK;
696 transport->recv.offset -= sizeof(transport->recv.fraghdr);
697 read = ret;
698 }
699
700 switch (be32_to_cpu(transport->recv.calldir)) {
701 default:
702 msg.msg_flags |= MSG_TRUNC;
703 break;
704 case RPC_CALL:
705 ret = xs_read_stream_call(transport, &msg, flags);
706 break;
707 case RPC_REPLY:
708 ret = xs_read_stream_reply(transport, &msg, flags);
709 }
710 if (msg.msg_flags & MSG_TRUNC) {
711 transport->recv.calldir = cpu_to_be32(-1);
712 transport->recv.copied = -1;
713 }
714 if (ret < 0)
715 goto out_err;
716 read += ret;
717 if (transport->recv.offset < transport->recv.len) {
718 if (!(msg.msg_flags & MSG_TRUNC))
719 return read;
720 msg.msg_flags = 0;
721 ret = xs_read_discard(transport->sock, &msg, flags,
722 transport->recv.len - transport->recv.offset);
723 if (ret <= 0)
724 goto out_err;
725 transport->recv.offset += ret;
726 read += ret;
727 if (transport->recv.offset != transport->recv.len)
728 return read;
729 }
730 if (xs_read_stream_request_done(transport)) {
731 trace_xs_stream_read_request(transport);
732 transport->recv.copied = 0;
733 }
734 transport->recv.offset = 0;
735 transport->recv.len = 0;
736 return read;
737out_err:
738 return ret != 0 ? ret : -ESHUTDOWN;
739}
740
741static __poll_t xs_poll_socket(struct sock_xprt *transport)
742{
743 return transport->sock->ops->poll(transport->file, transport->sock,
744 NULL);
745}
746
747static bool xs_poll_socket_readable(struct sock_xprt *transport)
748{
749 __poll_t events = xs_poll_socket(transport);
750
751 return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP);
752}
753
754static void xs_poll_check_readable(struct sock_xprt *transport)
755{
756
757 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
758 if (test_bit(XPRT_SOCK_IGNORE_RECV, &transport->sock_state))
759 return;
760 if (!xs_poll_socket_readable(transport))
761 return;
762 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
763 queue_work(xprtiod_workqueue, &transport->recv_worker);
764}
765
766static void xs_stream_data_receive(struct sock_xprt *transport)
767{
768 size_t read = 0;
769 ssize_t ret = 0;
770
771 mutex_lock(&transport->recv_mutex);
772 if (transport->sock == NULL)
773 goto out;
774 for (;;) {
775 ret = xs_read_stream(transport, MSG_DONTWAIT);
776 if (ret < 0)
777 break;
778 read += ret;
779 cond_resched();
780 }
781 if (ret == -ESHUTDOWN)
782 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
783 else if (ret == -EACCES)
784 xprt_wake_pending_tasks(&transport->xprt, -EACCES);
785 else
786 xs_poll_check_readable(transport);
787out:
788 mutex_unlock(&transport->recv_mutex);
789 trace_xs_stream_read_data(&transport->xprt, ret, read);
790}
791
792static void xs_stream_data_receive_workfn(struct work_struct *work)
793{
794 struct sock_xprt *transport =
795 container_of(work, struct sock_xprt, recv_worker);
796 unsigned int pflags = memalloc_nofs_save();
797
798 xs_stream_data_receive(transport);
799 memalloc_nofs_restore(pflags);
800}
801
802static void
803xs_stream_reset_connect(struct sock_xprt *transport)
804{
805 transport->recv.offset = 0;
806 transport->recv.len = 0;
807 transport->recv.copied = 0;
808 transport->xmit.offset = 0;
809}
810
811static void
812xs_stream_start_connect(struct sock_xprt *transport)
813{
814 transport->xprt.stat.connect_count++;
815 transport->xprt.stat.connect_start = jiffies;
816}
817
818#define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
819
820/**
821 * xs_nospace - handle transmit was incomplete
822 * @req: pointer to RPC request
823 * @transport: pointer to struct sock_xprt
824 *
825 */
826static int xs_nospace(struct rpc_rqst *req, struct sock_xprt *transport)
827{
828 struct rpc_xprt *xprt = &transport->xprt;
829 struct sock *sk = transport->inet;
830 int ret = -EAGAIN;
831
832 trace_rpc_socket_nospace(req, transport);
833
834 /* Protect against races with write_space */
835 spin_lock(&xprt->transport_lock);
836
837 /* Don't race with disconnect */
838 if (xprt_connected(xprt)) {
839 /* wait for more buffer space */
840 set_bit(XPRT_SOCK_NOSPACE, &transport->sock_state);
841 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
842 sk->sk_write_pending++;
843 xprt_wait_for_buffer_space(xprt);
844 } else
845 ret = -ENOTCONN;
846
847 spin_unlock(&xprt->transport_lock);
848 return ret;
849}
850
851static int xs_sock_nospace(struct rpc_rqst *req)
852{
853 struct sock_xprt *transport =
854 container_of(req->rq_xprt, struct sock_xprt, xprt);
855 struct sock *sk = transport->inet;
856 int ret = -EAGAIN;
857
858 lock_sock(sk);
859 if (!sock_writeable(sk))
860 ret = xs_nospace(req, transport);
861 release_sock(sk);
862 return ret;
863}
864
865static int xs_stream_nospace(struct rpc_rqst *req, bool vm_wait)
866{
867 struct sock_xprt *transport =
868 container_of(req->rq_xprt, struct sock_xprt, xprt);
869 struct sock *sk = transport->inet;
870 int ret = -EAGAIN;
871
872 if (vm_wait)
873 return -ENOBUFS;
874 lock_sock(sk);
875 if (!sk_stream_memory_free(sk))
876 ret = xs_nospace(req, transport);
877 release_sock(sk);
878 return ret;
879}
880
881static int xs_stream_prepare_request(struct rpc_rqst *req, struct xdr_buf *buf)
882{
883 return xdr_alloc_bvec(buf, rpc_task_gfp_mask());
884}
885
886/*
887 * Determine if the previous message in the stream was aborted before it
888 * could complete transmission.
889 */
890static bool
891xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req)
892{
893 return transport->xmit.offset != 0 && req->rq_bytes_sent == 0;
894}
895
896/*
897 * Return the stream record marker field for a record of length < 2^31-1
898 */
899static rpc_fraghdr
900xs_stream_record_marker(struct xdr_buf *xdr)
901{
902 if (!xdr->len)
903 return 0;
904 return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len);
905}
906
907/**
908 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
909 * @req: pointer to RPC request
910 *
911 * Return values:
912 * 0: The request has been sent
913 * EAGAIN: The socket was blocked, please call again later to
914 * complete the request
915 * ENOTCONN: Caller needs to invoke connect logic then call again
916 * other: Some other error occurred, the request was not sent
917 */
918static int xs_local_send_request(struct rpc_rqst *req)
919{
920 struct rpc_xprt *xprt = req->rq_xprt;
921 struct sock_xprt *transport =
922 container_of(xprt, struct sock_xprt, xprt);
923 struct xdr_buf *xdr = &req->rq_snd_buf;
924 rpc_fraghdr rm = xs_stream_record_marker(xdr);
925 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
926 struct msghdr msg = {
927 .msg_flags = XS_SENDMSG_FLAGS,
928 };
929 bool vm_wait;
930 unsigned int sent;
931 int status;
932
933 /* Close the stream if the previous transmission was incomplete */
934 if (xs_send_request_was_aborted(transport, req)) {
935 xprt_force_disconnect(xprt);
936 return -ENOTCONN;
937 }
938
939 xs_pktdump("packet data:",
940 req->rq_svec->iov_base, req->rq_svec->iov_len);
941
942 vm_wait = sk_stream_is_writeable(transport->inet) ? true : false;
943
944 req->rq_xtime = ktime_get();
945 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
946 transport->xmit.offset, rm, &sent);
947 dprintk("RPC: %s(%u) = %d\n",
948 __func__, xdr->len - transport->xmit.offset, status);
949
950 if (likely(sent > 0) || status == 0) {
951 transport->xmit.offset += sent;
952 req->rq_bytes_sent = transport->xmit.offset;
953 if (likely(req->rq_bytes_sent >= msglen)) {
954 req->rq_xmit_bytes_sent += transport->xmit.offset;
955 transport->xmit.offset = 0;
956 return 0;
957 }
958 status = -EAGAIN;
959 vm_wait = false;
960 }
961
962 switch (status) {
963 case -EAGAIN:
964 status = xs_stream_nospace(req, vm_wait);
965 break;
966 default:
967 dprintk("RPC: sendmsg returned unrecognized error %d\n",
968 -status);
969 fallthrough;
970 case -EPIPE:
971 xprt_force_disconnect(xprt);
972 status = -ENOTCONN;
973 }
974
975 return status;
976}
977
978/**
979 * xs_udp_send_request - write an RPC request to a UDP socket
980 * @req: pointer to RPC request
981 *
982 * Return values:
983 * 0: The request has been sent
984 * EAGAIN: The socket was blocked, please call again later to
985 * complete the request
986 * ENOTCONN: Caller needs to invoke connect logic then call again
987 * other: Some other error occurred, the request was not sent
988 */
989static int xs_udp_send_request(struct rpc_rqst *req)
990{
991 struct rpc_xprt *xprt = req->rq_xprt;
992 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
993 struct xdr_buf *xdr = &req->rq_snd_buf;
994 struct msghdr msg = {
995 .msg_name = xs_addr(xprt),
996 .msg_namelen = xprt->addrlen,
997 .msg_flags = XS_SENDMSG_FLAGS,
998 };
999 unsigned int sent;
1000 int status;
1001
1002 xs_pktdump("packet data:",
1003 req->rq_svec->iov_base,
1004 req->rq_svec->iov_len);
1005
1006 if (!xprt_bound(xprt))
1007 return -ENOTCONN;
1008
1009 if (!xprt_request_get_cong(xprt, req))
1010 return -EBADSLT;
1011
1012 status = xdr_alloc_bvec(xdr, rpc_task_gfp_mask());
1013 if (status < 0)
1014 return status;
1015 req->rq_xtime = ktime_get();
1016 status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, 0, &sent);
1017
1018 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
1019 xdr->len, status);
1020
1021 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
1022 if (status == -EPERM)
1023 goto process_status;
1024
1025 if (status == -EAGAIN && sock_writeable(transport->inet))
1026 status = -ENOBUFS;
1027
1028 if (sent > 0 || status == 0) {
1029 req->rq_xmit_bytes_sent += sent;
1030 if (sent >= req->rq_slen)
1031 return 0;
1032 /* Still some bytes left; set up for a retry later. */
1033 status = -EAGAIN;
1034 }
1035
1036process_status:
1037 switch (status) {
1038 case -ENOTSOCK:
1039 status = -ENOTCONN;
1040 /* Should we call xs_close() here? */
1041 break;
1042 case -EAGAIN:
1043 status = xs_sock_nospace(req);
1044 break;
1045 case -ENETUNREACH:
1046 case -ENOBUFS:
1047 case -EPIPE:
1048 case -ECONNREFUSED:
1049 case -EPERM:
1050 /* When the server has died, an ICMP port unreachable message
1051 * prompts ECONNREFUSED. */
1052 break;
1053 default:
1054 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1055 -status);
1056 }
1057
1058 return status;
1059}
1060
1061/**
1062 * xs_tcp_send_request - write an RPC request to a TCP socket
1063 * @req: pointer to RPC request
1064 *
1065 * Return values:
1066 * 0: The request has been sent
1067 * EAGAIN: The socket was blocked, please call again later to
1068 * complete the request
1069 * ENOTCONN: Caller needs to invoke connect logic then call again
1070 * other: Some other error occurred, the request was not sent
1071 *
1072 * XXX: In the case of soft timeouts, should we eventually give up
1073 * if sendmsg is not able to make progress?
1074 */
1075static int xs_tcp_send_request(struct rpc_rqst *req)
1076{
1077 struct rpc_xprt *xprt = req->rq_xprt;
1078 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1079 struct xdr_buf *xdr = &req->rq_snd_buf;
1080 rpc_fraghdr rm = xs_stream_record_marker(xdr);
1081 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
1082 struct msghdr msg = {
1083 .msg_flags = XS_SENDMSG_FLAGS,
1084 };
1085 bool vm_wait;
1086 unsigned int sent;
1087 int status;
1088
1089 /* Close the stream if the previous transmission was incomplete */
1090 if (xs_send_request_was_aborted(transport, req)) {
1091 if (transport->sock != NULL)
1092 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
1093 return -ENOTCONN;
1094 }
1095 if (!transport->inet)
1096 return -ENOTCONN;
1097
1098 xs_pktdump("packet data:",
1099 req->rq_svec->iov_base,
1100 req->rq_svec->iov_len);
1101
1102 if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
1103 xs_tcp_set_socket_timeouts(xprt, transport->sock);
1104
1105 xs_set_srcport(transport, transport->sock);
1106
1107 /* Continue transmitting the packet/record. We must be careful
1108 * to cope with writespace callbacks arriving _after_ we have
1109 * called sendmsg(). */
1110 req->rq_xtime = ktime_get();
1111 tcp_sock_set_cork(transport->inet, true);
1112
1113 vm_wait = sk_stream_is_writeable(transport->inet) ? true : false;
1114
1115 do {
1116 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
1117 transport->xmit.offset, rm, &sent);
1118
1119 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
1120 xdr->len - transport->xmit.offset, status);
1121
1122 /* If we've sent the entire packet, immediately
1123 * reset the count of bytes sent. */
1124 transport->xmit.offset += sent;
1125 req->rq_bytes_sent = transport->xmit.offset;
1126 if (likely(req->rq_bytes_sent >= msglen)) {
1127 req->rq_xmit_bytes_sent += transport->xmit.offset;
1128 transport->xmit.offset = 0;
1129 if (atomic_long_read(&xprt->xmit_queuelen) == 1)
1130 tcp_sock_set_cork(transport->inet, false);
1131 return 0;
1132 }
1133
1134 WARN_ON_ONCE(sent == 0 && status == 0);
1135
1136 if (sent > 0)
1137 vm_wait = false;
1138
1139 } while (status == 0);
1140
1141 switch (status) {
1142 case -ENOTSOCK:
1143 status = -ENOTCONN;
1144 /* Should we call xs_close() here? */
1145 break;
1146 case -EAGAIN:
1147 status = xs_stream_nospace(req, vm_wait);
1148 break;
1149 case -ECONNRESET:
1150 case -ECONNREFUSED:
1151 case -ENOTCONN:
1152 case -EADDRINUSE:
1153 case -ENOBUFS:
1154 case -EPIPE:
1155 break;
1156 default:
1157 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1158 -status);
1159 }
1160
1161 return status;
1162}
1163
1164static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1165{
1166 transport->old_data_ready = sk->sk_data_ready;
1167 transport->old_state_change = sk->sk_state_change;
1168 transport->old_write_space = sk->sk_write_space;
1169 transport->old_error_report = sk->sk_error_report;
1170}
1171
1172static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1173{
1174 sk->sk_data_ready = transport->old_data_ready;
1175 sk->sk_state_change = transport->old_state_change;
1176 sk->sk_write_space = transport->old_write_space;
1177 sk->sk_error_report = transport->old_error_report;
1178}
1179
1180static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
1181{
1182 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1183
1184 transport->xprt_err = 0;
1185 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1186 clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state);
1187 clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state);
1188 clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state);
1189 clear_bit(XPRT_SOCK_NOSPACE, &transport->sock_state);
1190}
1191
1192static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr)
1193{
1194 set_bit(nr, &transport->sock_state);
1195 queue_work(xprtiod_workqueue, &transport->error_worker);
1196}
1197
1198static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1199{
1200 xprt->connect_cookie++;
1201 smp_mb__before_atomic();
1202 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1203 clear_bit(XPRT_CLOSING, &xprt->state);
1204 xs_sock_reset_state_flags(xprt);
1205 smp_mb__after_atomic();
1206}
1207
1208/**
1209 * xs_error_report - callback to handle TCP socket state errors
1210 * @sk: socket
1211 *
1212 * Note: we don't call sock_error() since there may be a rpc_task
1213 * using the socket, and so we don't want to clear sk->sk_err.
1214 */
1215static void xs_error_report(struct sock *sk)
1216{
1217 struct sock_xprt *transport;
1218 struct rpc_xprt *xprt;
1219
1220 if (!(xprt = xprt_from_sock(sk)))
1221 return;
1222
1223 transport = container_of(xprt, struct sock_xprt, xprt);
1224 transport->xprt_err = -sk->sk_err;
1225 if (transport->xprt_err == 0)
1226 return;
1227 dprintk("RPC: xs_error_report client %p, error=%d...\n",
1228 xprt, -transport->xprt_err);
1229 trace_rpc_socket_error(xprt, sk->sk_socket, transport->xprt_err);
1230
1231 /* barrier ensures xprt_err is set before XPRT_SOCK_WAKE_ERROR */
1232 smp_mb__before_atomic();
1233 xs_run_error_worker(transport, XPRT_SOCK_WAKE_ERROR);
1234}
1235
1236static void xs_reset_transport(struct sock_xprt *transport)
1237{
1238 struct socket *sock = transport->sock;
1239 struct sock *sk = transport->inet;
1240 struct rpc_xprt *xprt = &transport->xprt;
1241 struct file *filp = transport->file;
1242
1243 if (sk == NULL)
1244 return;
1245 /*
1246 * Make sure we're calling this in a context from which it is safe
1247 * to call __fput_sync(). In practice that means rpciod and the
1248 * system workqueue.
1249 */
1250 if (!(current->flags & PF_WQ_WORKER)) {
1251 WARN_ON_ONCE(1);
1252 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
1253 return;
1254 }
1255
1256 if (atomic_read(&transport->xprt.swapper))
1257 sk_clear_memalloc(sk);
1258
1259 tls_handshake_cancel(sk);
1260
1261 kernel_sock_shutdown(sock, SHUT_RDWR);
1262
1263 mutex_lock(&transport->recv_mutex);
1264 lock_sock(sk);
1265 transport->inet = NULL;
1266 transport->sock = NULL;
1267 transport->file = NULL;
1268
1269 sk->sk_user_data = NULL;
1270
1271 xs_restore_old_callbacks(transport, sk);
1272 xprt_clear_connected(xprt);
1273 xs_sock_reset_connection_flags(xprt);
1274 /* Reset stream record info */
1275 xs_stream_reset_connect(transport);
1276 release_sock(sk);
1277 mutex_unlock(&transport->recv_mutex);
1278
1279 trace_rpc_socket_close(xprt, sock);
1280 __fput_sync(filp);
1281
1282 xprt_disconnect_done(xprt);
1283}
1284
1285/**
1286 * xs_close - close a socket
1287 * @xprt: transport
1288 *
1289 * This is used when all requests are complete; ie, no DRC state remains
1290 * on the server we want to save.
1291 *
1292 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1293 * xs_reset_transport() zeroing the socket from underneath a writer.
1294 */
1295static void xs_close(struct rpc_xprt *xprt)
1296{
1297 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1298
1299 dprintk("RPC: xs_close xprt %p\n", xprt);
1300
1301 if (transport->sock)
1302 tls_handshake_close(transport->sock);
1303 xs_reset_transport(transport);
1304 xprt->reestablish_timeout = 0;
1305}
1306
1307static void xs_inject_disconnect(struct rpc_xprt *xprt)
1308{
1309 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
1310 xprt);
1311 xprt_disconnect_done(xprt);
1312}
1313
1314static void xs_xprt_free(struct rpc_xprt *xprt)
1315{
1316 xs_free_peer_addresses(xprt);
1317 xprt_free(xprt);
1318}
1319
1320/**
1321 * xs_destroy - prepare to shutdown a transport
1322 * @xprt: doomed transport
1323 *
1324 */
1325static void xs_destroy(struct rpc_xprt *xprt)
1326{
1327 struct sock_xprt *transport = container_of(xprt,
1328 struct sock_xprt, xprt);
1329 dprintk("RPC: xs_destroy xprt %p\n", xprt);
1330
1331 cancel_delayed_work_sync(&transport->connect_worker);
1332 xs_close(xprt);
1333 cancel_work_sync(&transport->recv_worker);
1334 cancel_work_sync(&transport->error_worker);
1335 xs_xprt_free(xprt);
1336 module_put(THIS_MODULE);
1337}
1338
1339/**
1340 * xs_udp_data_read_skb - receive callback for UDP sockets
1341 * @xprt: transport
1342 * @sk: socket
1343 * @skb: skbuff
1344 *
1345 */
1346static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1347 struct sock *sk,
1348 struct sk_buff *skb)
1349{
1350 struct rpc_task *task;
1351 struct rpc_rqst *rovr;
1352 int repsize, copied;
1353 u32 _xid;
1354 __be32 *xp;
1355
1356 repsize = skb->len;
1357 if (repsize < 4) {
1358 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
1359 return;
1360 }
1361
1362 /* Copy the XID from the skb... */
1363 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1364 if (xp == NULL)
1365 return;
1366
1367 /* Look up and lock the request corresponding to the given XID */
1368 spin_lock(&xprt->queue_lock);
1369 rovr = xprt_lookup_rqst(xprt, *xp);
1370 if (!rovr)
1371 goto out_unlock;
1372 xprt_pin_rqst(rovr);
1373 xprt_update_rtt(rovr->rq_task);
1374 spin_unlock(&xprt->queue_lock);
1375 task = rovr->rq_task;
1376
1377 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1378 copied = repsize;
1379
1380 /* Suck it into the iovec, verify checksum if not done by hw. */
1381 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1382 spin_lock(&xprt->queue_lock);
1383 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1384 goto out_unpin;
1385 }
1386
1387
1388 spin_lock(&xprt->transport_lock);
1389 xprt_adjust_cwnd(xprt, task, copied);
1390 spin_unlock(&xprt->transport_lock);
1391 spin_lock(&xprt->queue_lock);
1392 xprt_complete_rqst(task, copied);
1393 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1394out_unpin:
1395 xprt_unpin_rqst(rovr);
1396 out_unlock:
1397 spin_unlock(&xprt->queue_lock);
1398}
1399
1400static void xs_udp_data_receive(struct sock_xprt *transport)
1401{
1402 struct sk_buff *skb;
1403 struct sock *sk;
1404 int err;
1405
1406 mutex_lock(&transport->recv_mutex);
1407 sk = transport->inet;
1408 if (sk == NULL)
1409 goto out;
1410 for (;;) {
1411 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err);
1412 if (skb == NULL)
1413 break;
1414 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1415 consume_skb(skb);
1416 cond_resched();
1417 }
1418 xs_poll_check_readable(transport);
1419out:
1420 mutex_unlock(&transport->recv_mutex);
1421}
1422
1423static void xs_udp_data_receive_workfn(struct work_struct *work)
1424{
1425 struct sock_xprt *transport =
1426 container_of(work, struct sock_xprt, recv_worker);
1427 unsigned int pflags = memalloc_nofs_save();
1428
1429 xs_udp_data_receive(transport);
1430 memalloc_nofs_restore(pflags);
1431}
1432
1433/**
1434 * xs_data_ready - "data ready" callback for sockets
1435 * @sk: socket with data to read
1436 *
1437 */
1438static void xs_data_ready(struct sock *sk)
1439{
1440 struct rpc_xprt *xprt;
1441
1442 trace_sk_data_ready(sk);
1443
1444 xprt = xprt_from_sock(sk);
1445 if (xprt != NULL) {
1446 struct sock_xprt *transport = container_of(xprt,
1447 struct sock_xprt, xprt);
1448
1449 trace_xs_data_ready(xprt);
1450
1451 transport->old_data_ready(sk);
1452
1453 if (test_bit(XPRT_SOCK_IGNORE_RECV, &transport->sock_state))
1454 return;
1455
1456 /* Any data means we had a useful conversation, so
1457 * then we don't need to delay the next reconnect
1458 */
1459 if (xprt->reestablish_timeout)
1460 xprt->reestablish_timeout = 0;
1461 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1462 queue_work(xprtiod_workqueue, &transport->recv_worker);
1463 }
1464}
1465
1466/*
1467 * Helper function to force a TCP close if the server is sending
1468 * junk and/or it has put us in CLOSE_WAIT
1469 */
1470static void xs_tcp_force_close(struct rpc_xprt *xprt)
1471{
1472 xprt_force_disconnect(xprt);
1473}
1474
1475#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1476static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1477{
1478 return PAGE_SIZE;
1479}
1480#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1481
1482/**
1483 * xs_local_state_change - callback to handle AF_LOCAL socket state changes
1484 * @sk: socket whose state has changed
1485 *
1486 */
1487static void xs_local_state_change(struct sock *sk)
1488{
1489 struct rpc_xprt *xprt;
1490 struct sock_xprt *transport;
1491
1492 if (!(xprt = xprt_from_sock(sk)))
1493 return;
1494 transport = container_of(xprt, struct sock_xprt, xprt);
1495 if (sk->sk_shutdown & SHUTDOWN_MASK) {
1496 clear_bit(XPRT_CONNECTED, &xprt->state);
1497 /* Trigger the socket release */
1498 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1499 }
1500}
1501
1502/**
1503 * xs_tcp_state_change - callback to handle TCP socket state changes
1504 * @sk: socket whose state has changed
1505 *
1506 */
1507static void xs_tcp_state_change(struct sock *sk)
1508{
1509 struct rpc_xprt *xprt;
1510 struct sock_xprt *transport;
1511
1512 if (!(xprt = xprt_from_sock(sk)))
1513 return;
1514 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1515 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1516 sk->sk_state, xprt_connected(xprt),
1517 sock_flag(sk, SOCK_DEAD),
1518 sock_flag(sk, SOCK_ZAPPED),
1519 sk->sk_shutdown);
1520
1521 transport = container_of(xprt, struct sock_xprt, xprt);
1522 trace_rpc_socket_state_change(xprt, sk->sk_socket);
1523 switch (sk->sk_state) {
1524 case TCP_ESTABLISHED:
1525 if (!xprt_test_and_set_connected(xprt)) {
1526 xprt->connect_cookie++;
1527 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1528 xprt_clear_connecting(xprt);
1529
1530 xprt->stat.connect_count++;
1531 xprt->stat.connect_time += (long)jiffies -
1532 xprt->stat.connect_start;
1533 xs_run_error_worker(transport, XPRT_SOCK_WAKE_PENDING);
1534 }
1535 break;
1536 case TCP_FIN_WAIT1:
1537 /* The client initiated a shutdown of the socket */
1538 xprt->connect_cookie++;
1539 xprt->reestablish_timeout = 0;
1540 set_bit(XPRT_CLOSING, &xprt->state);
1541 smp_mb__before_atomic();
1542 clear_bit(XPRT_CONNECTED, &xprt->state);
1543 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1544 smp_mb__after_atomic();
1545 break;
1546 case TCP_CLOSE_WAIT:
1547 /* The server initiated a shutdown of the socket */
1548 xprt->connect_cookie++;
1549 clear_bit(XPRT_CONNECTED, &xprt->state);
1550 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1551 fallthrough;
1552 case TCP_CLOSING:
1553 /*
1554 * If the server closed down the connection, make sure that
1555 * we back off before reconnecting
1556 */
1557 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1558 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1559 break;
1560 case TCP_LAST_ACK:
1561 set_bit(XPRT_CLOSING, &xprt->state);
1562 smp_mb__before_atomic();
1563 clear_bit(XPRT_CONNECTED, &xprt->state);
1564 smp_mb__after_atomic();
1565 break;
1566 case TCP_CLOSE:
1567 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1568 &transport->sock_state))
1569 xprt_clear_connecting(xprt);
1570 clear_bit(XPRT_CLOSING, &xprt->state);
1571 /* Trigger the socket release */
1572 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1573 }
1574}
1575
1576static void xs_write_space(struct sock *sk)
1577{
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 if (!test_and_clear_bit(XPRT_SOCK_NOSPACE, &transport->sock_state))
1589 return;
1590 xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE);
1591 sk->sk_write_pending--;
1592}
1593
1594/**
1595 * xs_udp_write_space - callback invoked when socket buffer space
1596 * becomes available
1597 * @sk: socket whose state has changed
1598 *
1599 * Called when more output buffer space is available for this socket.
1600 * We try not to wake our writers until they can make "significant"
1601 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1602 * with a bunch of small requests.
1603 */
1604static void xs_udp_write_space(struct sock *sk)
1605{
1606 /* from net/core/sock.c:sock_def_write_space */
1607 if (sock_writeable(sk))
1608 xs_write_space(sk);
1609}
1610
1611/**
1612 * xs_tcp_write_space - callback invoked when socket buffer space
1613 * becomes available
1614 * @sk: socket whose state has changed
1615 *
1616 * Called when more output buffer space is available for this socket.
1617 * We try not to wake our writers until they can make "significant"
1618 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1619 * with a bunch of small requests.
1620 */
1621static void xs_tcp_write_space(struct sock *sk)
1622{
1623 /* from net/core/stream.c:sk_stream_write_space */
1624 if (sk_stream_is_writeable(sk))
1625 xs_write_space(sk);
1626}
1627
1628static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1629{
1630 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1631 struct sock *sk = transport->inet;
1632
1633 if (transport->rcvsize) {
1634 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1635 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1636 }
1637 if (transport->sndsize) {
1638 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1639 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1640 sk->sk_write_space(sk);
1641 }
1642}
1643
1644/**
1645 * xs_udp_set_buffer_size - set send and receive limits
1646 * @xprt: generic transport
1647 * @sndsize: requested size of send buffer, in bytes
1648 * @rcvsize: requested size of receive buffer, in bytes
1649 *
1650 * Set socket send and receive buffer size limits.
1651 */
1652static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1653{
1654 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1655
1656 transport->sndsize = 0;
1657 if (sndsize)
1658 transport->sndsize = sndsize + 1024;
1659 transport->rcvsize = 0;
1660 if (rcvsize)
1661 transport->rcvsize = rcvsize + 1024;
1662
1663 xs_udp_do_set_buffer_size(xprt);
1664}
1665
1666/**
1667 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1668 * @xprt: controlling transport
1669 * @task: task that timed out
1670 *
1671 * Adjust the congestion window after a retransmit timeout has occurred.
1672 */
1673static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1674{
1675 spin_lock(&xprt->transport_lock);
1676 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1677 spin_unlock(&xprt->transport_lock);
1678}
1679
1680static int xs_get_random_port(void)
1681{
1682 unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1683 unsigned short range;
1684 unsigned short rand;
1685
1686 if (max < min)
1687 return -EADDRINUSE;
1688 range = max - min + 1;
1689 rand = get_random_u32_below(range);
1690 return rand + min;
1691}
1692
1693static unsigned short xs_sock_getport(struct socket *sock)
1694{
1695 struct sockaddr_storage buf;
1696 unsigned short port = 0;
1697
1698 if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1699 goto out;
1700 switch (buf.ss_family) {
1701 case AF_INET6:
1702 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1703 break;
1704 case AF_INET:
1705 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1706 }
1707out:
1708 return port;
1709}
1710
1711/**
1712 * xs_set_port - reset the port number in the remote endpoint address
1713 * @xprt: generic transport
1714 * @port: new port number
1715 *
1716 */
1717static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1718{
1719 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1720
1721 rpc_set_port(xs_addr(xprt), port);
1722 xs_update_peer_port(xprt);
1723}
1724
1725static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1726{
1727 if (transport->srcport == 0 && transport->xprt.reuseport)
1728 transport->srcport = xs_sock_getport(sock);
1729}
1730
1731static int xs_get_srcport(struct sock_xprt *transport)
1732{
1733 int port = transport->srcport;
1734
1735 if (port == 0 && transport->xprt.resvport)
1736 port = xs_get_random_port();
1737 return port;
1738}
1739
1740static unsigned short xs_sock_srcport(struct rpc_xprt *xprt)
1741{
1742 struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt);
1743 unsigned short ret = 0;
1744 mutex_lock(&sock->recv_mutex);
1745 if (sock->sock)
1746 ret = xs_sock_getport(sock->sock);
1747 mutex_unlock(&sock->recv_mutex);
1748 return ret;
1749}
1750
1751static int xs_sock_srcaddr(struct rpc_xprt *xprt, char *buf, size_t buflen)
1752{
1753 struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt);
1754 union {
1755 struct sockaddr sa;
1756 struct sockaddr_storage st;
1757 } saddr;
1758 int ret = -ENOTCONN;
1759
1760 mutex_lock(&sock->recv_mutex);
1761 if (sock->sock) {
1762 ret = kernel_getsockname(sock->sock, &saddr.sa);
1763 if (ret >= 0)
1764 ret = snprintf(buf, buflen, "%pISc", &saddr.sa);
1765 }
1766 mutex_unlock(&sock->recv_mutex);
1767 return ret;
1768}
1769
1770static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1771{
1772 if (transport->srcport != 0)
1773 transport->srcport = 0;
1774 if (!transport->xprt.resvport)
1775 return 0;
1776 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1777 return xprt_max_resvport;
1778 return --port;
1779}
1780static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1781{
1782 struct sockaddr_storage myaddr;
1783 int err, nloop = 0;
1784 int port = xs_get_srcport(transport);
1785 unsigned short last;
1786
1787 /*
1788 * If we are asking for any ephemeral port (i.e. port == 0 &&
1789 * transport->xprt.resvport == 0), don't bind. Let the local
1790 * port selection happen implicitly when the socket is used
1791 * (for example at connect time).
1792 *
1793 * This ensures that we can continue to establish TCP
1794 * connections even when all local ephemeral ports are already
1795 * a part of some TCP connection. This makes no difference
1796 * for UDP sockets, but also doesn't harm them.
1797 *
1798 * If we're asking for any reserved port (i.e. port == 0 &&
1799 * transport->xprt.resvport == 1) xs_get_srcport above will
1800 * ensure that port is non-zero and we will bind as needed.
1801 */
1802 if (port <= 0)
1803 return port;
1804
1805 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1806 do {
1807 rpc_set_port((struct sockaddr *)&myaddr, port);
1808 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1809 transport->xprt.addrlen);
1810 if (err == 0) {
1811 if (transport->xprt.reuseport)
1812 transport->srcport = port;
1813 break;
1814 }
1815 last = port;
1816 port = xs_next_srcport(transport, port);
1817 if (port > last)
1818 nloop++;
1819 } while (err == -EADDRINUSE && nloop != 2);
1820
1821 if (myaddr.ss_family == AF_INET)
1822 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1823 &((struct sockaddr_in *)&myaddr)->sin_addr,
1824 port, err ? "failed" : "ok", err);
1825 else
1826 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1827 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1828 port, err ? "failed" : "ok", err);
1829 return err;
1830}
1831
1832/*
1833 * We don't support autobind on AF_LOCAL sockets
1834 */
1835static void xs_local_rpcbind(struct rpc_task *task)
1836{
1837 xprt_set_bound(task->tk_xprt);
1838}
1839
1840static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1841{
1842}
1843
1844#ifdef CONFIG_DEBUG_LOCK_ALLOC
1845static struct lock_class_key xs_key[3];
1846static struct lock_class_key xs_slock_key[3];
1847
1848static inline void xs_reclassify_socketu(struct socket *sock)
1849{
1850 struct sock *sk = sock->sk;
1851
1852 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1853 &xs_slock_key[0], "sk_lock-AF_LOCAL-RPC", &xs_key[0]);
1854}
1855
1856static inline void xs_reclassify_socket4(struct socket *sock)
1857{
1858 struct sock *sk = sock->sk;
1859
1860 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1861 &xs_slock_key[1], "sk_lock-AF_INET-RPC", &xs_key[1]);
1862}
1863
1864static inline void xs_reclassify_socket6(struct socket *sock)
1865{
1866 struct sock *sk = sock->sk;
1867
1868 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1869 &xs_slock_key[2], "sk_lock-AF_INET6-RPC", &xs_key[2]);
1870}
1871
1872static inline void xs_reclassify_socket(int family, struct socket *sock)
1873{
1874 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1875 return;
1876
1877 switch (family) {
1878 case AF_LOCAL:
1879 xs_reclassify_socketu(sock);
1880 break;
1881 case AF_INET:
1882 xs_reclassify_socket4(sock);
1883 break;
1884 case AF_INET6:
1885 xs_reclassify_socket6(sock);
1886 break;
1887 }
1888}
1889#else
1890static inline void xs_reclassify_socket(int family, struct socket *sock)
1891{
1892}
1893#endif
1894
1895static void xs_dummy_setup_socket(struct work_struct *work)
1896{
1897}
1898
1899static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1900 struct sock_xprt *transport, int family, int type,
1901 int protocol, bool reuseport)
1902{
1903 struct file *filp;
1904 struct socket *sock;
1905 int err;
1906
1907 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1908 if (err < 0) {
1909 dprintk("RPC: can't create %d transport socket (%d).\n",
1910 protocol, -err);
1911 goto out;
1912 }
1913 xs_reclassify_socket(family, sock);
1914
1915 if (reuseport)
1916 sock_set_reuseport(sock->sk);
1917
1918 err = xs_bind(transport, sock);
1919 if (err) {
1920 sock_release(sock);
1921 goto out;
1922 }
1923
1924 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1925 if (IS_ERR(filp))
1926 return ERR_CAST(filp);
1927 transport->file = filp;
1928
1929 return sock;
1930out:
1931 return ERR_PTR(err);
1932}
1933
1934static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1935 struct socket *sock)
1936{
1937 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1938 xprt);
1939
1940 if (!transport->inet) {
1941 struct sock *sk = sock->sk;
1942
1943 lock_sock(sk);
1944
1945 xs_save_old_callbacks(transport, sk);
1946
1947 sk->sk_user_data = xprt;
1948 sk->sk_data_ready = xs_data_ready;
1949 sk->sk_write_space = xs_udp_write_space;
1950 sk->sk_state_change = xs_local_state_change;
1951 sk->sk_error_report = xs_error_report;
1952 sk->sk_use_task_frag = false;
1953
1954 xprt_clear_connected(xprt);
1955
1956 /* Reset to new socket */
1957 transport->sock = sock;
1958 transport->inet = sk;
1959
1960 release_sock(sk);
1961 }
1962
1963 xs_stream_start_connect(transport);
1964
1965 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1966}
1967
1968/**
1969 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1970 * @transport: socket transport to connect
1971 */
1972static int xs_local_setup_socket(struct sock_xprt *transport)
1973{
1974 struct rpc_xprt *xprt = &transport->xprt;
1975 struct file *filp;
1976 struct socket *sock;
1977 int status;
1978
1979 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1980 SOCK_STREAM, 0, &sock, 1);
1981 if (status < 0) {
1982 dprintk("RPC: can't create AF_LOCAL "
1983 "transport socket (%d).\n", -status);
1984 goto out;
1985 }
1986 xs_reclassify_socket(AF_LOCAL, sock);
1987
1988 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1989 if (IS_ERR(filp)) {
1990 status = PTR_ERR(filp);
1991 goto out;
1992 }
1993 transport->file = filp;
1994
1995 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1996 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1997
1998 status = xs_local_finish_connecting(xprt, sock);
1999 trace_rpc_socket_connect(xprt, sock, status);
2000 switch (status) {
2001 case 0:
2002 dprintk("RPC: xprt %p connected to %s\n",
2003 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2004 xprt->stat.connect_count++;
2005 xprt->stat.connect_time += (long)jiffies -
2006 xprt->stat.connect_start;
2007 xprt_set_connected(xprt);
2008 break;
2009 case -ENOBUFS:
2010 break;
2011 case -ENOENT:
2012 dprintk("RPC: xprt %p: socket %s does not exist\n",
2013 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2014 break;
2015 case -ECONNREFUSED:
2016 dprintk("RPC: xprt %p: connection refused for %s\n",
2017 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2018 break;
2019 default:
2020 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2021 __func__, -status,
2022 xprt->address_strings[RPC_DISPLAY_ADDR]);
2023 }
2024
2025out:
2026 xprt_clear_connecting(xprt);
2027 xprt_wake_pending_tasks(xprt, status);
2028 return status;
2029}
2030
2031static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2032{
2033 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2034 int ret;
2035
2036 if (transport->file)
2037 goto force_disconnect;
2038
2039 if (RPC_IS_ASYNC(task)) {
2040 /*
2041 * We want the AF_LOCAL connect to be resolved in the
2042 * filesystem namespace of the process making the rpc
2043 * call. Thus we connect synchronously.
2044 *
2045 * If we want to support asynchronous AF_LOCAL calls,
2046 * we'll need to figure out how to pass a namespace to
2047 * connect.
2048 */
2049 rpc_task_set_rpc_status(task, -ENOTCONN);
2050 goto out_wake;
2051 }
2052 ret = xs_local_setup_socket(transport);
2053 if (ret && !RPC_IS_SOFTCONN(task))
2054 msleep_interruptible(15000);
2055 return;
2056force_disconnect:
2057 xprt_force_disconnect(xprt);
2058out_wake:
2059 xprt_clear_connecting(xprt);
2060 xprt_wake_pending_tasks(xprt, -ENOTCONN);
2061}
2062
2063#if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2064/*
2065 * Note that this should be called with XPRT_LOCKED held, or recv_mutex
2066 * held, or when we otherwise know that we have exclusive access to the
2067 * socket, to guard against races with xs_reset_transport.
2068 */
2069static void xs_set_memalloc(struct rpc_xprt *xprt)
2070{
2071 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2072 xprt);
2073
2074 /*
2075 * If there's no sock, then we have nothing to set. The
2076 * reconnecting process will get it for us.
2077 */
2078 if (!transport->inet)
2079 return;
2080 if (atomic_read(&xprt->swapper))
2081 sk_set_memalloc(transport->inet);
2082}
2083
2084/**
2085 * xs_enable_swap - Tag this transport as being used for swap.
2086 * @xprt: transport to tag
2087 *
2088 * Take a reference to this transport on behalf of the rpc_clnt, and
2089 * optionally mark it for swapping if it wasn't already.
2090 */
2091static int
2092xs_enable_swap(struct rpc_xprt *xprt)
2093{
2094 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2095
2096 mutex_lock(&xs->recv_mutex);
2097 if (atomic_inc_return(&xprt->swapper) == 1 &&
2098 xs->inet)
2099 sk_set_memalloc(xs->inet);
2100 mutex_unlock(&xs->recv_mutex);
2101 return 0;
2102}
2103
2104/**
2105 * xs_disable_swap - Untag this transport as being used for swap.
2106 * @xprt: transport to tag
2107 *
2108 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2109 * swapper refcount goes to 0, untag the socket as a memalloc socket.
2110 */
2111static void
2112xs_disable_swap(struct rpc_xprt *xprt)
2113{
2114 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2115
2116 mutex_lock(&xs->recv_mutex);
2117 if (atomic_dec_and_test(&xprt->swapper) &&
2118 xs->inet)
2119 sk_clear_memalloc(xs->inet);
2120 mutex_unlock(&xs->recv_mutex);
2121}
2122#else
2123static void xs_set_memalloc(struct rpc_xprt *xprt)
2124{
2125}
2126
2127static int
2128xs_enable_swap(struct rpc_xprt *xprt)
2129{
2130 return -EINVAL;
2131}
2132
2133static void
2134xs_disable_swap(struct rpc_xprt *xprt)
2135{
2136}
2137#endif
2138
2139static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2140{
2141 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2142
2143 if (!transport->inet) {
2144 struct sock *sk = sock->sk;
2145
2146 lock_sock(sk);
2147
2148 xs_save_old_callbacks(transport, sk);
2149
2150 sk->sk_user_data = xprt;
2151 sk->sk_data_ready = xs_data_ready;
2152 sk->sk_write_space = xs_udp_write_space;
2153 sk->sk_use_task_frag = false;
2154
2155 xprt_set_connected(xprt);
2156
2157 /* Reset to new socket */
2158 transport->sock = sock;
2159 transport->inet = sk;
2160
2161 xs_set_memalloc(xprt);
2162
2163 release_sock(sk);
2164 }
2165 xs_udp_do_set_buffer_size(xprt);
2166
2167 xprt->stat.connect_start = jiffies;
2168}
2169
2170static void xs_udp_setup_socket(struct work_struct *work)
2171{
2172 struct sock_xprt *transport =
2173 container_of(work, struct sock_xprt, connect_worker.work);
2174 struct rpc_xprt *xprt = &transport->xprt;
2175 struct socket *sock;
2176 int status = -EIO;
2177 unsigned int pflags = current->flags;
2178
2179 if (atomic_read(&xprt->swapper))
2180 current->flags |= PF_MEMALLOC;
2181 sock = xs_create_sock(xprt, transport,
2182 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2183 IPPROTO_UDP, false);
2184 if (IS_ERR(sock))
2185 goto out;
2186
2187 dprintk("RPC: worker connecting xprt %p via %s to "
2188 "%s (port %s)\n", xprt,
2189 xprt->address_strings[RPC_DISPLAY_PROTO],
2190 xprt->address_strings[RPC_DISPLAY_ADDR],
2191 xprt->address_strings[RPC_DISPLAY_PORT]);
2192
2193 xs_udp_finish_connecting(xprt, sock);
2194 trace_rpc_socket_connect(xprt, sock, 0);
2195 status = 0;
2196out:
2197 xprt_clear_connecting(xprt);
2198 xprt_unlock_connect(xprt, transport);
2199 xprt_wake_pending_tasks(xprt, status);
2200 current_restore_flags(pflags, PF_MEMALLOC);
2201}
2202
2203/**
2204 * xs_tcp_shutdown - gracefully shut down a TCP socket
2205 * @xprt: transport
2206 *
2207 * Initiates a graceful shutdown of the TCP socket by calling the
2208 * equivalent of shutdown(SHUT_RDWR);
2209 */
2210static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2211{
2212 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2213 struct socket *sock = transport->sock;
2214 int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2215
2216 if (sock == NULL)
2217 return;
2218 if (!xprt->reuseport) {
2219 xs_close(xprt);
2220 return;
2221 }
2222 switch (skst) {
2223 case TCP_FIN_WAIT1:
2224 case TCP_FIN_WAIT2:
2225 case TCP_LAST_ACK:
2226 break;
2227 case TCP_ESTABLISHED:
2228 case TCP_CLOSE_WAIT:
2229 kernel_sock_shutdown(sock, SHUT_RDWR);
2230 trace_rpc_socket_shutdown(xprt, sock);
2231 break;
2232 default:
2233 xs_reset_transport(transport);
2234 }
2235}
2236
2237static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2238 struct socket *sock)
2239{
2240 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2241 struct net *net = sock_net(sock->sk);
2242 unsigned long connect_timeout;
2243 unsigned long syn_retries;
2244 unsigned int keepidle;
2245 unsigned int keepcnt;
2246 unsigned int timeo;
2247 unsigned long t;
2248
2249 spin_lock(&xprt->transport_lock);
2250 keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2251 keepcnt = xprt->timeout->to_retries + 1;
2252 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2253 (xprt->timeout->to_retries + 1);
2254 clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2255 spin_unlock(&xprt->transport_lock);
2256
2257 /* TCP Keepalive options */
2258 sock_set_keepalive(sock->sk);
2259 tcp_sock_set_keepidle(sock->sk, keepidle);
2260 tcp_sock_set_keepintvl(sock->sk, keepidle);
2261 tcp_sock_set_keepcnt(sock->sk, keepcnt);
2262
2263 /* TCP user timeout (see RFC5482) */
2264 tcp_sock_set_user_timeout(sock->sk, timeo);
2265
2266 /* Connect timeout */
2267 connect_timeout = max_t(unsigned long,
2268 DIV_ROUND_UP(xprt->connect_timeout, HZ), 1);
2269 syn_retries = max_t(unsigned long,
2270 READ_ONCE(net->ipv4.sysctl_tcp_syn_retries), 1);
2271 for (t = 0; t <= syn_retries && (1UL << t) < connect_timeout; t++)
2272 ;
2273 if (t <= syn_retries)
2274 tcp_sock_set_syncnt(sock->sk, t - 1);
2275}
2276
2277static void xs_tcp_do_set_connect_timeout(struct rpc_xprt *xprt,
2278 unsigned long connect_timeout)
2279{
2280 struct sock_xprt *transport =
2281 container_of(xprt, struct sock_xprt, xprt);
2282 struct rpc_timeout to;
2283 unsigned long initval;
2284
2285 memcpy(&to, xprt->timeout, sizeof(to));
2286 /* Arbitrary lower limit */
2287 initval = max_t(unsigned long, connect_timeout, XS_TCP_INIT_REEST_TO);
2288 to.to_initval = initval;
2289 to.to_maxval = initval;
2290 to.to_retries = 0;
2291 memcpy(&transport->tcp_timeout, &to, sizeof(transport->tcp_timeout));
2292 xprt->timeout = &transport->tcp_timeout;
2293 xprt->connect_timeout = connect_timeout;
2294}
2295
2296static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2297 unsigned long connect_timeout,
2298 unsigned long reconnect_timeout)
2299{
2300 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2301
2302 spin_lock(&xprt->transport_lock);
2303 if (reconnect_timeout < xprt->max_reconnect_timeout)
2304 xprt->max_reconnect_timeout = reconnect_timeout;
2305 if (connect_timeout < xprt->connect_timeout)
2306 xs_tcp_do_set_connect_timeout(xprt, connect_timeout);
2307 set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2308 spin_unlock(&xprt->transport_lock);
2309}
2310
2311static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2312{
2313 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2314
2315 if (!transport->inet) {
2316 struct sock *sk = sock->sk;
2317
2318 /* Avoid temporary address, they are bad for long-lived
2319 * connections such as NFS mounts.
2320 * RFC4941, section 3.6 suggests that:
2321 * Individual applications, which have specific
2322 * knowledge about the normal duration of connections,
2323 * MAY override this as appropriate.
2324 */
2325 if (xs_addr(xprt)->sa_family == PF_INET6) {
2326 ip6_sock_set_addr_preferences(sk,
2327 IPV6_PREFER_SRC_PUBLIC);
2328 }
2329
2330 xs_tcp_set_socket_timeouts(xprt, sock);
2331 tcp_sock_set_nodelay(sk);
2332
2333 lock_sock(sk);
2334
2335 xs_save_old_callbacks(transport, sk);
2336
2337 sk->sk_user_data = xprt;
2338 sk->sk_data_ready = xs_data_ready;
2339 sk->sk_state_change = xs_tcp_state_change;
2340 sk->sk_write_space = xs_tcp_write_space;
2341 sk->sk_error_report = xs_error_report;
2342 sk->sk_use_task_frag = false;
2343
2344 /* socket options */
2345 sock_reset_flag(sk, SOCK_LINGER);
2346
2347 xprt_clear_connected(xprt);
2348
2349 /* Reset to new socket */
2350 transport->sock = sock;
2351 transport->inet = sk;
2352
2353 release_sock(sk);
2354 }
2355
2356 if (!xprt_bound(xprt))
2357 return -ENOTCONN;
2358
2359 xs_set_memalloc(xprt);
2360
2361 xs_stream_start_connect(transport);
2362
2363 /* Tell the socket layer to start connecting... */
2364 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2365 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2366}
2367
2368/**
2369 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2370 * @work: queued work item
2371 *
2372 * Invoked by a work queue tasklet.
2373 */
2374static void xs_tcp_setup_socket(struct work_struct *work)
2375{
2376 struct sock_xprt *transport =
2377 container_of(work, struct sock_xprt, connect_worker.work);
2378 struct socket *sock = transport->sock;
2379 struct rpc_xprt *xprt = &transport->xprt;
2380 int status;
2381 unsigned int pflags = current->flags;
2382
2383 if (atomic_read(&xprt->swapper))
2384 current->flags |= PF_MEMALLOC;
2385
2386 if (xprt_connected(xprt))
2387 goto out;
2388 if (test_and_clear_bit(XPRT_SOCK_CONNECT_SENT,
2389 &transport->sock_state) ||
2390 !sock) {
2391 xs_reset_transport(transport);
2392 sock = xs_create_sock(xprt, transport, xs_addr(xprt)->sa_family,
2393 SOCK_STREAM, IPPROTO_TCP, true);
2394 if (IS_ERR(sock)) {
2395 xprt_wake_pending_tasks(xprt, PTR_ERR(sock));
2396 goto out;
2397 }
2398 }
2399
2400 dprintk("RPC: worker connecting xprt %p via %s to "
2401 "%s (port %s)\n", xprt,
2402 xprt->address_strings[RPC_DISPLAY_PROTO],
2403 xprt->address_strings[RPC_DISPLAY_ADDR],
2404 xprt->address_strings[RPC_DISPLAY_PORT]);
2405
2406 status = xs_tcp_finish_connecting(xprt, sock);
2407 trace_rpc_socket_connect(xprt, sock, status);
2408 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2409 xprt, -status, xprt_connected(xprt),
2410 sock->sk->sk_state);
2411 switch (status) {
2412 case 0:
2413 case -EINPROGRESS:
2414 /* SYN_SENT! */
2415 set_bit(XPRT_SOCK_CONNECT_SENT, &transport->sock_state);
2416 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2417 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2418 fallthrough;
2419 case -EALREADY:
2420 goto out_unlock;
2421 case -EADDRNOTAVAIL:
2422 /* Source port number is unavailable. Try a new one! */
2423 transport->srcport = 0;
2424 status = -EAGAIN;
2425 break;
2426 case -EINVAL:
2427 /* Happens, for instance, if the user specified a link
2428 * local IPv6 address without a scope-id.
2429 */
2430 case -ECONNREFUSED:
2431 case -ECONNRESET:
2432 case -ENETDOWN:
2433 case -ENETUNREACH:
2434 case -EHOSTUNREACH:
2435 case -EADDRINUSE:
2436 case -ENOBUFS:
2437 break;
2438 default:
2439 printk("%s: connect returned unhandled error %d\n",
2440 __func__, status);
2441 status = -EAGAIN;
2442 }
2443
2444 /* xs_tcp_force_close() wakes tasks with a fixed error code.
2445 * We need to wake them first to ensure the correct error code.
2446 */
2447 xprt_wake_pending_tasks(xprt, status);
2448 xs_tcp_force_close(xprt);
2449out:
2450 xprt_clear_connecting(xprt);
2451out_unlock:
2452 xprt_unlock_connect(xprt, transport);
2453 current_restore_flags(pflags, PF_MEMALLOC);
2454}
2455
2456/*
2457 * Transfer the connected socket to @upper_transport, then mark that
2458 * xprt CONNECTED.
2459 */
2460static int xs_tcp_tls_finish_connecting(struct rpc_xprt *lower_xprt,
2461 struct sock_xprt *upper_transport)
2462{
2463 struct sock_xprt *lower_transport =
2464 container_of(lower_xprt, struct sock_xprt, xprt);
2465 struct rpc_xprt *upper_xprt = &upper_transport->xprt;
2466
2467 if (!upper_transport->inet) {
2468 struct socket *sock = lower_transport->sock;
2469 struct sock *sk = sock->sk;
2470
2471 /* Avoid temporary address, they are bad for long-lived
2472 * connections such as NFS mounts.
2473 * RFC4941, section 3.6 suggests that:
2474 * Individual applications, which have specific
2475 * knowledge about the normal duration of connections,
2476 * MAY override this as appropriate.
2477 */
2478 if (xs_addr(upper_xprt)->sa_family == PF_INET6)
2479 ip6_sock_set_addr_preferences(sk, IPV6_PREFER_SRC_PUBLIC);
2480
2481 xs_tcp_set_socket_timeouts(upper_xprt, sock);
2482 tcp_sock_set_nodelay(sk);
2483
2484 lock_sock(sk);
2485
2486 /* @sk is already connected, so it now has the RPC callbacks.
2487 * Reach into @lower_transport to save the original ones.
2488 */
2489 upper_transport->old_data_ready = lower_transport->old_data_ready;
2490 upper_transport->old_state_change = lower_transport->old_state_change;
2491 upper_transport->old_write_space = lower_transport->old_write_space;
2492 upper_transport->old_error_report = lower_transport->old_error_report;
2493 sk->sk_user_data = upper_xprt;
2494
2495 /* socket options */
2496 sock_reset_flag(sk, SOCK_LINGER);
2497
2498 xprt_clear_connected(upper_xprt);
2499
2500 upper_transport->sock = sock;
2501 upper_transport->inet = sk;
2502 upper_transport->file = lower_transport->file;
2503
2504 release_sock(sk);
2505
2506 /* Reset lower_transport before shutting down its clnt */
2507 mutex_lock(&lower_transport->recv_mutex);
2508 lower_transport->inet = NULL;
2509 lower_transport->sock = NULL;
2510 lower_transport->file = NULL;
2511
2512 xprt_clear_connected(lower_xprt);
2513 xs_sock_reset_connection_flags(lower_xprt);
2514 xs_stream_reset_connect(lower_transport);
2515 mutex_unlock(&lower_transport->recv_mutex);
2516 }
2517
2518 if (!xprt_bound(upper_xprt))
2519 return -ENOTCONN;
2520
2521 xs_set_memalloc(upper_xprt);
2522
2523 if (!xprt_test_and_set_connected(upper_xprt)) {
2524 upper_xprt->connect_cookie++;
2525 clear_bit(XPRT_SOCK_CONNECTING, &upper_transport->sock_state);
2526 xprt_clear_connecting(upper_xprt);
2527
2528 upper_xprt->stat.connect_count++;
2529 upper_xprt->stat.connect_time += (long)jiffies -
2530 upper_xprt->stat.connect_start;
2531 xs_run_error_worker(upper_transport, XPRT_SOCK_WAKE_PENDING);
2532 }
2533 return 0;
2534}
2535
2536/**
2537 * xs_tls_handshake_done - TLS handshake completion handler
2538 * @data: address of xprt to wake
2539 * @status: status of handshake
2540 * @peerid: serial number of key containing the remote's identity
2541 *
2542 */
2543static void xs_tls_handshake_done(void *data, int status, key_serial_t peerid)
2544{
2545 struct rpc_xprt *lower_xprt = data;
2546 struct sock_xprt *lower_transport =
2547 container_of(lower_xprt, struct sock_xprt, xprt);
2548
2549 lower_transport->xprt_err = status ? -EACCES : 0;
2550 complete(&lower_transport->handshake_done);
2551 xprt_put(lower_xprt);
2552}
2553
2554static int xs_tls_handshake_sync(struct rpc_xprt *lower_xprt, struct xprtsec_parms *xprtsec)
2555{
2556 struct sock_xprt *lower_transport =
2557 container_of(lower_xprt, struct sock_xprt, xprt);
2558 struct tls_handshake_args args = {
2559 .ta_sock = lower_transport->sock,
2560 .ta_done = xs_tls_handshake_done,
2561 .ta_data = xprt_get(lower_xprt),
2562 .ta_peername = lower_xprt->servername,
2563 };
2564 struct sock *sk = lower_transport->inet;
2565 int rc;
2566
2567 init_completion(&lower_transport->handshake_done);
2568 set_bit(XPRT_SOCK_IGNORE_RECV, &lower_transport->sock_state);
2569 lower_transport->xprt_err = -ETIMEDOUT;
2570 switch (xprtsec->policy) {
2571 case RPC_XPRTSEC_TLS_ANON:
2572 rc = tls_client_hello_anon(&args, GFP_KERNEL);
2573 if (rc)
2574 goto out_put_xprt;
2575 break;
2576 case RPC_XPRTSEC_TLS_X509:
2577 args.ta_my_cert = xprtsec->cert_serial;
2578 args.ta_my_privkey = xprtsec->privkey_serial;
2579 rc = tls_client_hello_x509(&args, GFP_KERNEL);
2580 if (rc)
2581 goto out_put_xprt;
2582 break;
2583 default:
2584 rc = -EACCES;
2585 goto out_put_xprt;
2586 }
2587
2588 rc = wait_for_completion_interruptible_timeout(&lower_transport->handshake_done,
2589 XS_TLS_HANDSHAKE_TO);
2590 if (rc <= 0) {
2591 if (!tls_handshake_cancel(sk)) {
2592 if (rc == 0)
2593 rc = -ETIMEDOUT;
2594 goto out_put_xprt;
2595 }
2596 }
2597
2598 rc = lower_transport->xprt_err;
2599
2600out:
2601 xs_stream_reset_connect(lower_transport);
2602 clear_bit(XPRT_SOCK_IGNORE_RECV, &lower_transport->sock_state);
2603 return rc;
2604
2605out_put_xprt:
2606 xprt_put(lower_xprt);
2607 goto out;
2608}
2609
2610/**
2611 * xs_tcp_tls_setup_socket - establish a TLS session on a TCP socket
2612 * @work: queued work item
2613 *
2614 * Invoked by a work queue tasklet.
2615 *
2616 * For RPC-with-TLS, there is a two-stage connection process.
2617 *
2618 * The "upper-layer xprt" is visible to the RPC consumer. Once it has
2619 * been marked connected, the consumer knows that a TCP connection and
2620 * a TLS session have been established.
2621 *
2622 * A "lower-layer xprt", created in this function, handles the mechanics
2623 * of connecting the TCP socket, performing the RPC_AUTH_TLS probe, and
2624 * then driving the TLS handshake. Once all that is complete, the upper
2625 * layer xprt is marked connected.
2626 */
2627static void xs_tcp_tls_setup_socket(struct work_struct *work)
2628{
2629 struct sock_xprt *upper_transport =
2630 container_of(work, struct sock_xprt, connect_worker.work);
2631 struct rpc_clnt *upper_clnt = upper_transport->clnt;
2632 struct rpc_xprt *upper_xprt = &upper_transport->xprt;
2633 struct rpc_create_args args = {
2634 .net = upper_xprt->xprt_net,
2635 .protocol = upper_xprt->prot,
2636 .address = (struct sockaddr *)&upper_xprt->addr,
2637 .addrsize = upper_xprt->addrlen,
2638 .timeout = upper_clnt->cl_timeout,
2639 .servername = upper_xprt->servername,
2640 .program = upper_clnt->cl_program,
2641 .prognumber = upper_clnt->cl_prog,
2642 .version = upper_clnt->cl_vers,
2643 .authflavor = RPC_AUTH_TLS,
2644 .cred = upper_clnt->cl_cred,
2645 .xprtsec = {
2646 .policy = RPC_XPRTSEC_NONE,
2647 },
2648 };
2649 unsigned int pflags = current->flags;
2650 struct rpc_clnt *lower_clnt;
2651 struct rpc_xprt *lower_xprt;
2652 int status;
2653
2654 if (atomic_read(&upper_xprt->swapper))
2655 current->flags |= PF_MEMALLOC;
2656
2657 xs_stream_start_connect(upper_transport);
2658
2659 /* This implicitly sends an RPC_AUTH_TLS probe */
2660 lower_clnt = rpc_create(&args);
2661 if (IS_ERR(lower_clnt)) {
2662 trace_rpc_tls_unavailable(upper_clnt, upper_xprt);
2663 clear_bit(XPRT_SOCK_CONNECTING, &upper_transport->sock_state);
2664 xprt_clear_connecting(upper_xprt);
2665 xprt_wake_pending_tasks(upper_xprt, PTR_ERR(lower_clnt));
2666 xs_run_error_worker(upper_transport, XPRT_SOCK_WAKE_PENDING);
2667 goto out_unlock;
2668 }
2669
2670 /* RPC_AUTH_TLS probe was successful. Try a TLS handshake on
2671 * the lower xprt.
2672 */
2673 rcu_read_lock();
2674 lower_xprt = rcu_dereference(lower_clnt->cl_xprt);
2675 rcu_read_unlock();
2676
2677 if (wait_on_bit_lock(&lower_xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2678 goto out_unlock;
2679
2680 status = xs_tls_handshake_sync(lower_xprt, &upper_xprt->xprtsec);
2681 if (status) {
2682 trace_rpc_tls_not_started(upper_clnt, upper_xprt);
2683 goto out_close;
2684 }
2685
2686 status = xs_tcp_tls_finish_connecting(lower_xprt, upper_transport);
2687 if (status)
2688 goto out_close;
2689 xprt_release_write(lower_xprt, NULL);
2690
2691 trace_rpc_socket_connect(upper_xprt, upper_transport->sock, 0);
2692 if (!xprt_test_and_set_connected(upper_xprt)) {
2693 upper_xprt->connect_cookie++;
2694 clear_bit(XPRT_SOCK_CONNECTING, &upper_transport->sock_state);
2695 xprt_clear_connecting(upper_xprt);
2696
2697 upper_xprt->stat.connect_count++;
2698 upper_xprt->stat.connect_time += (long)jiffies -
2699 upper_xprt->stat.connect_start;
2700 xs_run_error_worker(upper_transport, XPRT_SOCK_WAKE_PENDING);
2701 }
2702 rpc_shutdown_client(lower_clnt);
2703
2704out_unlock:
2705 current_restore_flags(pflags, PF_MEMALLOC);
2706 upper_transport->clnt = NULL;
2707 xprt_unlock_connect(upper_xprt, upper_transport);
2708 return;
2709
2710out_close:
2711 xprt_release_write(lower_xprt, NULL);
2712 rpc_shutdown_client(lower_clnt);
2713
2714 /* xprt_force_disconnect() wakes tasks with a fixed tk_status code.
2715 * Wake them first here to ensure they get our tk_status code.
2716 */
2717 xprt_wake_pending_tasks(upper_xprt, status);
2718 xs_tcp_force_close(upper_xprt);
2719 xprt_clear_connecting(upper_xprt);
2720 goto out_unlock;
2721}
2722
2723/**
2724 * xs_connect - connect a socket to a remote endpoint
2725 * @xprt: pointer to transport structure
2726 * @task: address of RPC task that manages state of connect request
2727 *
2728 * TCP: If the remote end dropped the connection, delay reconnecting.
2729 *
2730 * UDP socket connects are synchronous, but we use a work queue anyway
2731 * to guarantee that even unprivileged user processes can set up a
2732 * socket on a privileged port.
2733 *
2734 * If a UDP socket connect fails, the delay behavior here prevents
2735 * retry floods (hard mounts).
2736 */
2737static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2738{
2739 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2740 unsigned long delay = 0;
2741
2742 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2743
2744 if (transport->sock != NULL) {
2745 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2746 "seconds\n", xprt, xprt->reestablish_timeout / HZ);
2747
2748 delay = xprt_reconnect_delay(xprt);
2749 xprt_reconnect_backoff(xprt, XS_TCP_INIT_REEST_TO);
2750
2751 } else
2752 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2753
2754 transport->clnt = task->tk_client;
2755 queue_delayed_work(xprtiod_workqueue,
2756 &transport->connect_worker,
2757 delay);
2758}
2759
2760static void xs_wake_disconnect(struct sock_xprt *transport)
2761{
2762 if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state))
2763 xs_tcp_force_close(&transport->xprt);
2764}
2765
2766static void xs_wake_write(struct sock_xprt *transport)
2767{
2768 if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state))
2769 xprt_write_space(&transport->xprt);
2770}
2771
2772static void xs_wake_error(struct sock_xprt *transport)
2773{
2774 int sockerr;
2775
2776 if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2777 return;
2778 sockerr = xchg(&transport->xprt_err, 0);
2779 if (sockerr < 0) {
2780 xprt_wake_pending_tasks(&transport->xprt, sockerr);
2781 xs_tcp_force_close(&transport->xprt);
2782 }
2783}
2784
2785static void xs_wake_pending(struct sock_xprt *transport)
2786{
2787 if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING, &transport->sock_state))
2788 xprt_wake_pending_tasks(&transport->xprt, -EAGAIN);
2789}
2790
2791static void xs_error_handle(struct work_struct *work)
2792{
2793 struct sock_xprt *transport = container_of(work,
2794 struct sock_xprt, error_worker);
2795
2796 xs_wake_disconnect(transport);
2797 xs_wake_write(transport);
2798 xs_wake_error(transport);
2799 xs_wake_pending(transport);
2800}
2801
2802/**
2803 * xs_local_print_stats - display AF_LOCAL socket-specific stats
2804 * @xprt: rpc_xprt struct containing statistics
2805 * @seq: output file
2806 *
2807 */
2808static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2809{
2810 long idle_time = 0;
2811
2812 if (xprt_connected(xprt))
2813 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2814
2815 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2816 "%llu %llu %lu %llu %llu\n",
2817 xprt->stat.bind_count,
2818 xprt->stat.connect_count,
2819 xprt->stat.connect_time / HZ,
2820 idle_time,
2821 xprt->stat.sends,
2822 xprt->stat.recvs,
2823 xprt->stat.bad_xids,
2824 xprt->stat.req_u,
2825 xprt->stat.bklog_u,
2826 xprt->stat.max_slots,
2827 xprt->stat.sending_u,
2828 xprt->stat.pending_u);
2829}
2830
2831/**
2832 * xs_udp_print_stats - display UDP socket-specific stats
2833 * @xprt: rpc_xprt struct containing statistics
2834 * @seq: output file
2835 *
2836 */
2837static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2838{
2839 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2840
2841 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2842 "%lu %llu %llu\n",
2843 transport->srcport,
2844 xprt->stat.bind_count,
2845 xprt->stat.sends,
2846 xprt->stat.recvs,
2847 xprt->stat.bad_xids,
2848 xprt->stat.req_u,
2849 xprt->stat.bklog_u,
2850 xprt->stat.max_slots,
2851 xprt->stat.sending_u,
2852 xprt->stat.pending_u);
2853}
2854
2855/**
2856 * xs_tcp_print_stats - display TCP socket-specific stats
2857 * @xprt: rpc_xprt struct containing statistics
2858 * @seq: output file
2859 *
2860 */
2861static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2862{
2863 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2864 long idle_time = 0;
2865
2866 if (xprt_connected(xprt))
2867 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2868
2869 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2870 "%llu %llu %lu %llu %llu\n",
2871 transport->srcport,
2872 xprt->stat.bind_count,
2873 xprt->stat.connect_count,
2874 xprt->stat.connect_time / HZ,
2875 idle_time,
2876 xprt->stat.sends,
2877 xprt->stat.recvs,
2878 xprt->stat.bad_xids,
2879 xprt->stat.req_u,
2880 xprt->stat.bklog_u,
2881 xprt->stat.max_slots,
2882 xprt->stat.sending_u,
2883 xprt->stat.pending_u);
2884}
2885
2886/*
2887 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2888 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2889 * to use the server side send routines.
2890 */
2891static int bc_malloc(struct rpc_task *task)
2892{
2893 struct rpc_rqst *rqst = task->tk_rqstp;
2894 size_t size = rqst->rq_callsize;
2895 struct page *page;
2896 struct rpc_buffer *buf;
2897
2898 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2899 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2900 size);
2901 return -EINVAL;
2902 }
2903
2904 page = alloc_page(GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN);
2905 if (!page)
2906 return -ENOMEM;
2907
2908 buf = page_address(page);
2909 buf->len = PAGE_SIZE;
2910
2911 rqst->rq_buffer = buf->data;
2912 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2913 return 0;
2914}
2915
2916/*
2917 * Free the space allocated in the bc_alloc routine
2918 */
2919static void bc_free(struct rpc_task *task)
2920{
2921 void *buffer = task->tk_rqstp->rq_buffer;
2922 struct rpc_buffer *buf;
2923
2924 buf = container_of(buffer, struct rpc_buffer, data);
2925 free_page((unsigned long)buf);
2926}
2927
2928static int bc_sendto(struct rpc_rqst *req)
2929{
2930 struct xdr_buf *xdr = &req->rq_snd_buf;
2931 struct sock_xprt *transport =
2932 container_of(req->rq_xprt, struct sock_xprt, xprt);
2933 struct msghdr msg = {
2934 .msg_flags = 0,
2935 };
2936 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
2937 (u32)xdr->len);
2938 unsigned int sent = 0;
2939 int err;
2940
2941 req->rq_xtime = ktime_get();
2942 err = xdr_alloc_bvec(xdr, rpc_task_gfp_mask());
2943 if (err < 0)
2944 return err;
2945 err = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, marker, &sent);
2946 xdr_free_bvec(xdr);
2947 if (err < 0 || sent != (xdr->len + sizeof(marker)))
2948 return -EAGAIN;
2949 return sent;
2950}
2951
2952/**
2953 * bc_send_request - Send a backchannel Call on a TCP socket
2954 * @req: rpc_rqst containing Call message to be sent
2955 *
2956 * xpt_mutex ensures @rqstp's whole message is written to the socket
2957 * without interruption.
2958 *
2959 * Return values:
2960 * %0 if the message was sent successfully
2961 * %ENOTCONN if the message was not sent
2962 */
2963static int bc_send_request(struct rpc_rqst *req)
2964{
2965 struct svc_xprt *xprt;
2966 int len;
2967
2968 /*
2969 * Get the server socket associated with this callback xprt
2970 */
2971 xprt = req->rq_xprt->bc_xprt;
2972
2973 /*
2974 * Grab the mutex to serialize data as the connection is shared
2975 * with the fore channel
2976 */
2977 mutex_lock(&xprt->xpt_mutex);
2978 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2979 len = -ENOTCONN;
2980 else
2981 len = bc_sendto(req);
2982 mutex_unlock(&xprt->xpt_mutex);
2983
2984 if (len > 0)
2985 len = 0;
2986
2987 return len;
2988}
2989
2990/*
2991 * The close routine. Since this is client initiated, we do nothing
2992 */
2993
2994static void bc_close(struct rpc_xprt *xprt)
2995{
2996 xprt_disconnect_done(xprt);
2997}
2998
2999/*
3000 * The xprt destroy routine. Again, because this connection is client
3001 * initiated, we do nothing
3002 */
3003
3004static void bc_destroy(struct rpc_xprt *xprt)
3005{
3006 dprintk("RPC: bc_destroy xprt %p\n", xprt);
3007
3008 xs_xprt_free(xprt);
3009 module_put(THIS_MODULE);
3010}
3011
3012static const struct rpc_xprt_ops xs_local_ops = {
3013 .reserve_xprt = xprt_reserve_xprt,
3014 .release_xprt = xprt_release_xprt,
3015 .alloc_slot = xprt_alloc_slot,
3016 .free_slot = xprt_free_slot,
3017 .rpcbind = xs_local_rpcbind,
3018 .set_port = xs_local_set_port,
3019 .connect = xs_local_connect,
3020 .buf_alloc = rpc_malloc,
3021 .buf_free = rpc_free,
3022 .prepare_request = xs_stream_prepare_request,
3023 .send_request = xs_local_send_request,
3024 .wait_for_reply_request = xprt_wait_for_reply_request_def,
3025 .close = xs_close,
3026 .destroy = xs_destroy,
3027 .print_stats = xs_local_print_stats,
3028 .enable_swap = xs_enable_swap,
3029 .disable_swap = xs_disable_swap,
3030};
3031
3032static const struct rpc_xprt_ops xs_udp_ops = {
3033 .set_buffer_size = xs_udp_set_buffer_size,
3034 .reserve_xprt = xprt_reserve_xprt_cong,
3035 .release_xprt = xprt_release_xprt_cong,
3036 .alloc_slot = xprt_alloc_slot,
3037 .free_slot = xprt_free_slot,
3038 .rpcbind = rpcb_getport_async,
3039 .set_port = xs_set_port,
3040 .connect = xs_connect,
3041 .get_srcaddr = xs_sock_srcaddr,
3042 .get_srcport = xs_sock_srcport,
3043 .buf_alloc = rpc_malloc,
3044 .buf_free = rpc_free,
3045 .send_request = xs_udp_send_request,
3046 .wait_for_reply_request = xprt_wait_for_reply_request_rtt,
3047 .timer = xs_udp_timer,
3048 .release_request = xprt_release_rqst_cong,
3049 .close = xs_close,
3050 .destroy = xs_destroy,
3051 .print_stats = xs_udp_print_stats,
3052 .enable_swap = xs_enable_swap,
3053 .disable_swap = xs_disable_swap,
3054 .inject_disconnect = xs_inject_disconnect,
3055};
3056
3057static const struct rpc_xprt_ops xs_tcp_ops = {
3058 .reserve_xprt = xprt_reserve_xprt,
3059 .release_xprt = xprt_release_xprt,
3060 .alloc_slot = xprt_alloc_slot,
3061 .free_slot = xprt_free_slot,
3062 .rpcbind = rpcb_getport_async,
3063 .set_port = xs_set_port,
3064 .connect = xs_connect,
3065 .get_srcaddr = xs_sock_srcaddr,
3066 .get_srcport = xs_sock_srcport,
3067 .buf_alloc = rpc_malloc,
3068 .buf_free = rpc_free,
3069 .prepare_request = xs_stream_prepare_request,
3070 .send_request = xs_tcp_send_request,
3071 .wait_for_reply_request = xprt_wait_for_reply_request_def,
3072 .close = xs_tcp_shutdown,
3073 .destroy = xs_destroy,
3074 .set_connect_timeout = xs_tcp_set_connect_timeout,
3075 .print_stats = xs_tcp_print_stats,
3076 .enable_swap = xs_enable_swap,
3077 .disable_swap = xs_disable_swap,
3078 .inject_disconnect = xs_inject_disconnect,
3079#ifdef CONFIG_SUNRPC_BACKCHANNEL
3080 .bc_setup = xprt_setup_bc,
3081 .bc_maxpayload = xs_tcp_bc_maxpayload,
3082 .bc_num_slots = xprt_bc_max_slots,
3083 .bc_free_rqst = xprt_free_bc_rqst,
3084 .bc_destroy = xprt_destroy_bc,
3085#endif
3086};
3087
3088/*
3089 * The rpc_xprt_ops for the server backchannel
3090 */
3091
3092static const struct rpc_xprt_ops bc_tcp_ops = {
3093 .reserve_xprt = xprt_reserve_xprt,
3094 .release_xprt = xprt_release_xprt,
3095 .alloc_slot = xprt_alloc_slot,
3096 .free_slot = xprt_free_slot,
3097 .buf_alloc = bc_malloc,
3098 .buf_free = bc_free,
3099 .send_request = bc_send_request,
3100 .wait_for_reply_request = xprt_wait_for_reply_request_def,
3101 .close = bc_close,
3102 .destroy = bc_destroy,
3103 .print_stats = xs_tcp_print_stats,
3104 .enable_swap = xs_enable_swap,
3105 .disable_swap = xs_disable_swap,
3106 .inject_disconnect = xs_inject_disconnect,
3107};
3108
3109static int xs_init_anyaddr(const int family, struct sockaddr *sap)
3110{
3111 static const struct sockaddr_in sin = {
3112 .sin_family = AF_INET,
3113 .sin_addr.s_addr = htonl(INADDR_ANY),
3114 };
3115 static const struct sockaddr_in6 sin6 = {
3116 .sin6_family = AF_INET6,
3117 .sin6_addr = IN6ADDR_ANY_INIT,
3118 };
3119
3120 switch (family) {
3121 case AF_LOCAL:
3122 break;
3123 case AF_INET:
3124 memcpy(sap, &sin, sizeof(sin));
3125 break;
3126 case AF_INET6:
3127 memcpy(sap, &sin6, sizeof(sin6));
3128 break;
3129 default:
3130 dprintk("RPC: %s: Bad address family\n", __func__);
3131 return -EAFNOSUPPORT;
3132 }
3133 return 0;
3134}
3135
3136static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
3137 unsigned int slot_table_size,
3138 unsigned int max_slot_table_size)
3139{
3140 struct rpc_xprt *xprt;
3141 struct sock_xprt *new;
3142
3143 if (args->addrlen > sizeof(xprt->addr)) {
3144 dprintk("RPC: xs_setup_xprt: address too large\n");
3145 return ERR_PTR(-EBADF);
3146 }
3147
3148 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
3149 max_slot_table_size);
3150 if (xprt == NULL) {
3151 dprintk("RPC: xs_setup_xprt: couldn't allocate "
3152 "rpc_xprt\n");
3153 return ERR_PTR(-ENOMEM);
3154 }
3155
3156 new = container_of(xprt, struct sock_xprt, xprt);
3157 mutex_init(&new->recv_mutex);
3158 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
3159 xprt->addrlen = args->addrlen;
3160 if (args->srcaddr)
3161 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
3162 else {
3163 int err;
3164 err = xs_init_anyaddr(args->dstaddr->sa_family,
3165 (struct sockaddr *)&new->srcaddr);
3166 if (err != 0) {
3167 xprt_free(xprt);
3168 return ERR_PTR(err);
3169 }
3170 }
3171
3172 return xprt;
3173}
3174
3175static const struct rpc_timeout xs_local_default_timeout = {
3176 .to_initval = 10 * HZ,
3177 .to_maxval = 10 * HZ,
3178 .to_retries = 2,
3179};
3180
3181/**
3182 * xs_setup_local - Set up transport to use an AF_LOCAL socket
3183 * @args: rpc transport creation arguments
3184 *
3185 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
3186 */
3187static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
3188{
3189 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
3190 struct sock_xprt *transport;
3191 struct rpc_xprt *xprt;
3192 struct rpc_xprt *ret;
3193
3194 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3195 xprt_max_tcp_slot_table_entries);
3196 if (IS_ERR(xprt))
3197 return xprt;
3198 transport = container_of(xprt, struct sock_xprt, xprt);
3199
3200 xprt->prot = 0;
3201 xprt->xprt_class = &xs_local_transport;
3202 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3203
3204 xprt->bind_timeout = XS_BIND_TO;
3205 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3206 xprt->idle_timeout = XS_IDLE_DISC_TO;
3207
3208 xprt->ops = &xs_local_ops;
3209 xprt->timeout = &xs_local_default_timeout;
3210
3211 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3212 INIT_WORK(&transport->error_worker, xs_error_handle);
3213 INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket);
3214
3215 switch (sun->sun_family) {
3216 case AF_LOCAL:
3217 if (sun->sun_path[0] != '/' && sun->sun_path[0] != '\0') {
3218 dprintk("RPC: bad AF_LOCAL address: %s\n",
3219 sun->sun_path);
3220 ret = ERR_PTR(-EINVAL);
3221 goto out_err;
3222 }
3223 xprt_set_bound(xprt);
3224 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
3225 break;
3226 default:
3227 ret = ERR_PTR(-EAFNOSUPPORT);
3228 goto out_err;
3229 }
3230
3231 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
3232 xprt->address_strings[RPC_DISPLAY_ADDR]);
3233
3234 if (try_module_get(THIS_MODULE))
3235 return xprt;
3236 ret = ERR_PTR(-EINVAL);
3237out_err:
3238 xs_xprt_free(xprt);
3239 return ret;
3240}
3241
3242static const struct rpc_timeout xs_udp_default_timeout = {
3243 .to_initval = 5 * HZ,
3244 .to_maxval = 30 * HZ,
3245 .to_increment = 5 * HZ,
3246 .to_retries = 5,
3247};
3248
3249/**
3250 * xs_setup_udp - Set up transport to use a UDP socket
3251 * @args: rpc transport creation arguments
3252 *
3253 */
3254static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
3255{
3256 struct sockaddr *addr = args->dstaddr;
3257 struct rpc_xprt *xprt;
3258 struct sock_xprt *transport;
3259 struct rpc_xprt *ret;
3260
3261 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
3262 xprt_udp_slot_table_entries);
3263 if (IS_ERR(xprt))
3264 return xprt;
3265 transport = container_of(xprt, struct sock_xprt, xprt);
3266
3267 xprt->prot = IPPROTO_UDP;
3268 xprt->xprt_class = &xs_udp_transport;
3269 /* XXX: header size can vary due to auth type, IPv6, etc. */
3270 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
3271
3272 xprt->bind_timeout = XS_BIND_TO;
3273 xprt->reestablish_timeout = XS_UDP_REEST_TO;
3274 xprt->idle_timeout = XS_IDLE_DISC_TO;
3275
3276 xprt->ops = &xs_udp_ops;
3277
3278 xprt->timeout = &xs_udp_default_timeout;
3279
3280 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
3281 INIT_WORK(&transport->error_worker, xs_error_handle);
3282 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
3283
3284 switch (addr->sa_family) {
3285 case AF_INET:
3286 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3287 xprt_set_bound(xprt);
3288
3289 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
3290 break;
3291 case AF_INET6:
3292 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3293 xprt_set_bound(xprt);
3294
3295 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
3296 break;
3297 default:
3298 ret = ERR_PTR(-EAFNOSUPPORT);
3299 goto out_err;
3300 }
3301
3302 if (xprt_bound(xprt))
3303 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3304 xprt->address_strings[RPC_DISPLAY_ADDR],
3305 xprt->address_strings[RPC_DISPLAY_PORT],
3306 xprt->address_strings[RPC_DISPLAY_PROTO]);
3307 else
3308 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
3309 xprt->address_strings[RPC_DISPLAY_ADDR],
3310 xprt->address_strings[RPC_DISPLAY_PROTO]);
3311
3312 if (try_module_get(THIS_MODULE))
3313 return xprt;
3314 ret = ERR_PTR(-EINVAL);
3315out_err:
3316 xs_xprt_free(xprt);
3317 return ret;
3318}
3319
3320static const struct rpc_timeout xs_tcp_default_timeout = {
3321 .to_initval = 60 * HZ,
3322 .to_maxval = 60 * HZ,
3323 .to_retries = 2,
3324};
3325
3326/**
3327 * xs_setup_tcp - Set up transport to use a TCP socket
3328 * @args: rpc transport creation arguments
3329 *
3330 */
3331static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
3332{
3333 struct sockaddr *addr = args->dstaddr;
3334 struct rpc_xprt *xprt;
3335 struct sock_xprt *transport;
3336 struct rpc_xprt *ret;
3337 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
3338
3339 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
3340 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
3341
3342 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3343 max_slot_table_size);
3344 if (IS_ERR(xprt))
3345 return xprt;
3346 transport = container_of(xprt, struct sock_xprt, xprt);
3347
3348 xprt->prot = IPPROTO_TCP;
3349 xprt->xprt_class = &xs_tcp_transport;
3350 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3351
3352 xprt->bind_timeout = XS_BIND_TO;
3353 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3354 xprt->idle_timeout = XS_IDLE_DISC_TO;
3355
3356 xprt->ops = &xs_tcp_ops;
3357 xprt->timeout = &xs_tcp_default_timeout;
3358
3359 xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3360 if (args->reconnect_timeout)
3361 xprt->max_reconnect_timeout = args->reconnect_timeout;
3362
3363 xprt->connect_timeout = xprt->timeout->to_initval *
3364 (xprt->timeout->to_retries + 1);
3365 if (args->connect_timeout)
3366 xs_tcp_do_set_connect_timeout(xprt, args->connect_timeout);
3367
3368 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3369 INIT_WORK(&transport->error_worker, xs_error_handle);
3370 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
3371
3372 switch (addr->sa_family) {
3373 case AF_INET:
3374 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3375 xprt_set_bound(xprt);
3376
3377 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3378 break;
3379 case AF_INET6:
3380 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3381 xprt_set_bound(xprt);
3382
3383 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3384 break;
3385 default:
3386 ret = ERR_PTR(-EAFNOSUPPORT);
3387 goto out_err;
3388 }
3389
3390 if (xprt_bound(xprt))
3391 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3392 xprt->address_strings[RPC_DISPLAY_ADDR],
3393 xprt->address_strings[RPC_DISPLAY_PORT],
3394 xprt->address_strings[RPC_DISPLAY_PROTO]);
3395 else
3396 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
3397 xprt->address_strings[RPC_DISPLAY_ADDR],
3398 xprt->address_strings[RPC_DISPLAY_PROTO]);
3399
3400 if (try_module_get(THIS_MODULE))
3401 return xprt;
3402 ret = ERR_PTR(-EINVAL);
3403out_err:
3404 xs_xprt_free(xprt);
3405 return ret;
3406}
3407
3408/**
3409 * xs_setup_tcp_tls - Set up transport to use a TCP with TLS
3410 * @args: rpc transport creation arguments
3411 *
3412 */
3413static struct rpc_xprt *xs_setup_tcp_tls(struct xprt_create *args)
3414{
3415 struct sockaddr *addr = args->dstaddr;
3416 struct rpc_xprt *xprt;
3417 struct sock_xprt *transport;
3418 struct rpc_xprt *ret;
3419 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
3420
3421 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
3422 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
3423
3424 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3425 max_slot_table_size);
3426 if (IS_ERR(xprt))
3427 return xprt;
3428 transport = container_of(xprt, struct sock_xprt, xprt);
3429
3430 xprt->prot = IPPROTO_TCP;
3431 xprt->xprt_class = &xs_tcp_transport;
3432 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3433
3434 xprt->bind_timeout = XS_BIND_TO;
3435 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3436 xprt->idle_timeout = XS_IDLE_DISC_TO;
3437
3438 xprt->ops = &xs_tcp_ops;
3439 xprt->timeout = &xs_tcp_default_timeout;
3440
3441 xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3442 xprt->connect_timeout = xprt->timeout->to_initval *
3443 (xprt->timeout->to_retries + 1);
3444
3445 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3446 INIT_WORK(&transport->error_worker, xs_error_handle);
3447
3448 switch (args->xprtsec.policy) {
3449 case RPC_XPRTSEC_TLS_ANON:
3450 case RPC_XPRTSEC_TLS_X509:
3451 xprt->xprtsec = args->xprtsec;
3452 INIT_DELAYED_WORK(&transport->connect_worker,
3453 xs_tcp_tls_setup_socket);
3454 break;
3455 default:
3456 ret = ERR_PTR(-EACCES);
3457 goto out_err;
3458 }
3459
3460 switch (addr->sa_family) {
3461 case AF_INET:
3462 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3463 xprt_set_bound(xprt);
3464
3465 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3466 break;
3467 case AF_INET6:
3468 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3469 xprt_set_bound(xprt);
3470
3471 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3472 break;
3473 default:
3474 ret = ERR_PTR(-EAFNOSUPPORT);
3475 goto out_err;
3476 }
3477
3478 if (xprt_bound(xprt))
3479 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3480 xprt->address_strings[RPC_DISPLAY_ADDR],
3481 xprt->address_strings[RPC_DISPLAY_PORT],
3482 xprt->address_strings[RPC_DISPLAY_PROTO]);
3483 else
3484 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
3485 xprt->address_strings[RPC_DISPLAY_ADDR],
3486 xprt->address_strings[RPC_DISPLAY_PROTO]);
3487
3488 if (try_module_get(THIS_MODULE))
3489 return xprt;
3490 ret = ERR_PTR(-EINVAL);
3491out_err:
3492 xs_xprt_free(xprt);
3493 return ret;
3494}
3495
3496/**
3497 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3498 * @args: rpc transport creation arguments
3499 *
3500 */
3501static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3502{
3503 struct sockaddr *addr = args->dstaddr;
3504 struct rpc_xprt *xprt;
3505 struct sock_xprt *transport;
3506 struct svc_sock *bc_sock;
3507 struct rpc_xprt *ret;
3508
3509 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3510 xprt_tcp_slot_table_entries);
3511 if (IS_ERR(xprt))
3512 return xprt;
3513 transport = container_of(xprt, struct sock_xprt, xprt);
3514
3515 xprt->prot = IPPROTO_TCP;
3516 xprt->xprt_class = &xs_bc_tcp_transport;
3517 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3518 xprt->timeout = &xs_tcp_default_timeout;
3519
3520 /* backchannel */
3521 xprt_set_bound(xprt);
3522 xprt->bind_timeout = 0;
3523 xprt->reestablish_timeout = 0;
3524 xprt->idle_timeout = 0;
3525
3526 xprt->ops = &bc_tcp_ops;
3527
3528 switch (addr->sa_family) {
3529 case AF_INET:
3530 xs_format_peer_addresses(xprt, "tcp",
3531 RPCBIND_NETID_TCP);
3532 break;
3533 case AF_INET6:
3534 xs_format_peer_addresses(xprt, "tcp",
3535 RPCBIND_NETID_TCP6);
3536 break;
3537 default:
3538 ret = ERR_PTR(-EAFNOSUPPORT);
3539 goto out_err;
3540 }
3541
3542 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3543 xprt->address_strings[RPC_DISPLAY_ADDR],
3544 xprt->address_strings[RPC_DISPLAY_PORT],
3545 xprt->address_strings[RPC_DISPLAY_PROTO]);
3546
3547 /*
3548 * Once we've associated a backchannel xprt with a connection,
3549 * we want to keep it around as long as the connection lasts,
3550 * in case we need to start using it for a backchannel again;
3551 * this reference won't be dropped until bc_xprt is destroyed.
3552 */
3553 xprt_get(xprt);
3554 args->bc_xprt->xpt_bc_xprt = xprt;
3555 xprt->bc_xprt = args->bc_xprt;
3556 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3557 transport->sock = bc_sock->sk_sock;
3558 transport->inet = bc_sock->sk_sk;
3559
3560 /*
3561 * Since we don't want connections for the backchannel, we set
3562 * the xprt status to connected
3563 */
3564 xprt_set_connected(xprt);
3565
3566 if (try_module_get(THIS_MODULE))
3567 return xprt;
3568
3569 args->bc_xprt->xpt_bc_xprt = NULL;
3570 args->bc_xprt->xpt_bc_xps = NULL;
3571 xprt_put(xprt);
3572 ret = ERR_PTR(-EINVAL);
3573out_err:
3574 xs_xprt_free(xprt);
3575 return ret;
3576}
3577
3578static struct xprt_class xs_local_transport = {
3579 .list = LIST_HEAD_INIT(xs_local_transport.list),
3580 .name = "named UNIX socket",
3581 .owner = THIS_MODULE,
3582 .ident = XPRT_TRANSPORT_LOCAL,
3583 .setup = xs_setup_local,
3584 .netid = { "" },
3585};
3586
3587static struct xprt_class xs_udp_transport = {
3588 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3589 .name = "udp",
3590 .owner = THIS_MODULE,
3591 .ident = XPRT_TRANSPORT_UDP,
3592 .setup = xs_setup_udp,
3593 .netid = { "udp", "udp6", "" },
3594};
3595
3596static struct xprt_class xs_tcp_transport = {
3597 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3598 .name = "tcp",
3599 .owner = THIS_MODULE,
3600 .ident = XPRT_TRANSPORT_TCP,
3601 .setup = xs_setup_tcp,
3602 .netid = { "tcp", "tcp6", "" },
3603};
3604
3605static struct xprt_class xs_tcp_tls_transport = {
3606 .list = LIST_HEAD_INIT(xs_tcp_tls_transport.list),
3607 .name = "tcp-with-tls",
3608 .owner = THIS_MODULE,
3609 .ident = XPRT_TRANSPORT_TCP_TLS,
3610 .setup = xs_setup_tcp_tls,
3611 .netid = { "tcp", "tcp6", "" },
3612};
3613
3614static struct xprt_class xs_bc_tcp_transport = {
3615 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3616 .name = "tcp NFSv4.1 backchannel",
3617 .owner = THIS_MODULE,
3618 .ident = XPRT_TRANSPORT_BC_TCP,
3619 .setup = xs_setup_bc_tcp,
3620 .netid = { "" },
3621};
3622
3623/**
3624 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3625 *
3626 */
3627int init_socket_xprt(void)
3628{
3629 if (!sunrpc_table_header)
3630 sunrpc_table_header = register_sysctl("sunrpc", xs_tunables_table);
3631
3632 xprt_register_transport(&xs_local_transport);
3633 xprt_register_transport(&xs_udp_transport);
3634 xprt_register_transport(&xs_tcp_transport);
3635 xprt_register_transport(&xs_tcp_tls_transport);
3636 xprt_register_transport(&xs_bc_tcp_transport);
3637
3638 return 0;
3639}
3640
3641/**
3642 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3643 *
3644 */
3645void cleanup_socket_xprt(void)
3646{
3647 if (sunrpc_table_header) {
3648 unregister_sysctl_table(sunrpc_table_header);
3649 sunrpc_table_header = NULL;
3650 }
3651
3652 xprt_unregister_transport(&xs_local_transport);
3653 xprt_unregister_transport(&xs_udp_transport);
3654 xprt_unregister_transport(&xs_tcp_transport);
3655 xprt_unregister_transport(&xs_tcp_tls_transport);
3656 xprt_unregister_transport(&xs_bc_tcp_transport);
3657}
3658
3659static int param_set_portnr(const char *val, const struct kernel_param *kp)
3660{
3661 return param_set_uint_minmax(val, kp,
3662 RPC_MIN_RESVPORT,
3663 RPC_MAX_RESVPORT);
3664}
3665
3666static const struct kernel_param_ops param_ops_portnr = {
3667 .set = param_set_portnr,
3668 .get = param_get_uint,
3669};
3670
3671#define param_check_portnr(name, p) \
3672 __param_check(name, p, unsigned int);
3673
3674module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3675module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3676
3677static int param_set_slot_table_size(const char *val,
3678 const struct kernel_param *kp)
3679{
3680 return param_set_uint_minmax(val, kp,
3681 RPC_MIN_SLOT_TABLE,
3682 RPC_MAX_SLOT_TABLE);
3683}
3684
3685static const struct kernel_param_ops param_ops_slot_table_size = {
3686 .set = param_set_slot_table_size,
3687 .get = param_get_uint,
3688};
3689
3690#define param_check_slot_table_size(name, p) \
3691 __param_check(name, p, unsigned int);
3692
3693static int param_set_max_slot_table_size(const char *val,
3694 const struct kernel_param *kp)
3695{
3696 return param_set_uint_minmax(val, kp,
3697 RPC_MIN_SLOT_TABLE,
3698 RPC_MAX_SLOT_TABLE_LIMIT);
3699}
3700
3701static const struct kernel_param_ops param_ops_max_slot_table_size = {
3702 .set = param_set_max_slot_table_size,
3703 .get = param_get_uint,
3704};
3705
3706#define param_check_max_slot_table_size(name, p) \
3707 __param_check(name, p, unsigned int);
3708
3709module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3710 slot_table_size, 0644);
3711module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3712 max_slot_table_size, 0644);
3713module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3714 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
51#include <trace/events/sunrpc.h>
52
53#include "sunrpc.h"
54
55#define RPC_TCP_READ_CHUNK_SZ (3*512*1024)
56
57static void xs_close(struct rpc_xprt *xprt);
58static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
59 struct socket *sock);
60
61/*
62 * xprtsock tunables
63 */
64static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
65static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
66static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
67
68static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
69static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
70
71#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
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
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,
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#endif
163
164/*
165 * Wait duration for a reply from the RPC portmapper.
166 */
167#define XS_BIND_TO (60U * HZ)
168
169/*
170 * Delay if a UDP socket connect error occurs. This is most likely some
171 * kind of resource problem on the local host.
172 */
173#define XS_UDP_REEST_TO (2U * HZ)
174
175/*
176 * The reestablish timeout allows clients to delay for a bit before attempting
177 * to reconnect to a server that just dropped our connection.
178 *
179 * We implement an exponential backoff when trying to reestablish a TCP
180 * transport connection with the server. Some servers like to drop a TCP
181 * connection when they are overworked, so we start with a short timeout and
182 * increase over time if the server is down or not responding.
183 */
184#define XS_TCP_INIT_REEST_TO (3U * HZ)
185
186/*
187 * TCP idle timeout; client drops the transport socket if it is idle
188 * for this long. Note that we also timeout UDP sockets to prevent
189 * holding port numbers when there is no RPC traffic.
190 */
191#define XS_IDLE_DISC_TO (5U * 60 * HZ)
192
193#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
194# undef RPC_DEBUG_DATA
195# define RPCDBG_FACILITY RPCDBG_TRANS
196#endif
197
198#ifdef RPC_DEBUG_DATA
199static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
200{
201 u8 *buf = (u8 *) packet;
202 int j;
203
204 dprintk("RPC: %s\n", msg);
205 for (j = 0; j < count && j < 128; j += 4) {
206 if (!(j & 31)) {
207 if (j)
208 dprintk("\n");
209 dprintk("0x%04x ", j);
210 }
211 dprintk("%02x%02x%02x%02x ",
212 buf[j], buf[j+1], buf[j+2], buf[j+3]);
213 }
214 dprintk("\n");
215}
216#else
217static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
218{
219 /* NOP */
220}
221#endif
222
223static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
224{
225 return (struct rpc_xprt *) sk->sk_user_data;
226}
227
228static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
229{
230 return (struct sockaddr *) &xprt->addr;
231}
232
233static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
234{
235 return (struct sockaddr_un *) &xprt->addr;
236}
237
238static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
239{
240 return (struct sockaddr_in *) &xprt->addr;
241}
242
243static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
244{
245 return (struct sockaddr_in6 *) &xprt->addr;
246}
247
248static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
249{
250 struct sockaddr *sap = xs_addr(xprt);
251 struct sockaddr_in6 *sin6;
252 struct sockaddr_in *sin;
253 struct sockaddr_un *sun;
254 char buf[128];
255
256 switch (sap->sa_family) {
257 case AF_LOCAL:
258 sun = xs_addr_un(xprt);
259 strlcpy(buf, sun->sun_path, sizeof(buf));
260 xprt->address_strings[RPC_DISPLAY_ADDR] =
261 kstrdup(buf, GFP_KERNEL);
262 break;
263 case AF_INET:
264 (void)rpc_ntop(sap, buf, sizeof(buf));
265 xprt->address_strings[RPC_DISPLAY_ADDR] =
266 kstrdup(buf, GFP_KERNEL);
267 sin = xs_addr_in(xprt);
268 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
269 break;
270 case AF_INET6:
271 (void)rpc_ntop(sap, buf, sizeof(buf));
272 xprt->address_strings[RPC_DISPLAY_ADDR] =
273 kstrdup(buf, GFP_KERNEL);
274 sin6 = xs_addr_in6(xprt);
275 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
276 break;
277 default:
278 BUG();
279 }
280
281 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
282}
283
284static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
285{
286 struct sockaddr *sap = xs_addr(xprt);
287 char buf[128];
288
289 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
290 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
291
292 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
293 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
294}
295
296static void xs_format_peer_addresses(struct rpc_xprt *xprt,
297 const char *protocol,
298 const char *netid)
299{
300 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
301 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
302 xs_format_common_peer_addresses(xprt);
303 xs_format_common_peer_ports(xprt);
304}
305
306static void xs_update_peer_port(struct rpc_xprt *xprt)
307{
308 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
309 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
310
311 xs_format_common_peer_ports(xprt);
312}
313
314static void xs_free_peer_addresses(struct rpc_xprt *xprt)
315{
316 unsigned int i;
317
318 for (i = 0; i < RPC_DISPLAY_MAX; i++)
319 switch (i) {
320 case RPC_DISPLAY_PROTO:
321 case RPC_DISPLAY_NETID:
322 continue;
323 default:
324 kfree(xprt->address_strings[i]);
325 }
326}
327
328#define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
329
330static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
331{
332 struct msghdr msg = {
333 .msg_name = addr,
334 .msg_namelen = addrlen,
335 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
336 };
337 struct kvec iov = {
338 .iov_base = vec->iov_base + base,
339 .iov_len = vec->iov_len - base,
340 };
341
342 if (iov.iov_len != 0)
343 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
344 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
345}
346
347static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p)
348{
349 ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
350 int offset, size_t size, int flags);
351 struct page **ppage;
352 unsigned int remainder;
353 int err;
354
355 remainder = xdr->page_len - base;
356 base += xdr->page_base;
357 ppage = xdr->pages + (base >> PAGE_SHIFT);
358 base &= ~PAGE_MASK;
359 do_sendpage = sock->ops->sendpage;
360 if (!zerocopy)
361 do_sendpage = sock_no_sendpage;
362 for(;;) {
363 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
364 int flags = XS_SENDMSG_FLAGS;
365
366 remainder -= len;
367 if (more)
368 flags |= MSG_MORE;
369 if (remainder != 0)
370 flags |= MSG_SENDPAGE_NOTLAST | MSG_MORE;
371 err = do_sendpage(sock, *ppage, base, len, flags);
372 if (remainder == 0 || err != len)
373 break;
374 *sent_p += err;
375 ppage++;
376 base = 0;
377 }
378 if (err > 0) {
379 *sent_p += err;
380 err = 0;
381 }
382 return err;
383}
384
385/**
386 * xs_sendpages - write pages directly to a socket
387 * @sock: socket to send on
388 * @addr: UDP only -- address of destination
389 * @addrlen: UDP only -- length of destination address
390 * @xdr: buffer containing this request
391 * @base: starting position in the buffer
392 * @zerocopy: true if it is safe to use sendpage()
393 * @sent_p: return the total number of bytes successfully queued for sending
394 *
395 */
396static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p)
397{
398 unsigned int remainder = xdr->len - base;
399 int err = 0;
400 int sent = 0;
401
402 if (unlikely(!sock))
403 return -ENOTSOCK;
404
405 if (base != 0) {
406 addr = NULL;
407 addrlen = 0;
408 }
409
410 if (base < xdr->head[0].iov_len || addr != NULL) {
411 unsigned int len = xdr->head[0].iov_len - base;
412 remainder -= len;
413 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
414 if (remainder == 0 || err != len)
415 goto out;
416 *sent_p += err;
417 base = 0;
418 } else
419 base -= xdr->head[0].iov_len;
420
421 if (base < xdr->page_len) {
422 unsigned int len = xdr->page_len - base;
423 remainder -= len;
424 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy, &sent);
425 *sent_p += sent;
426 if (remainder == 0 || sent != len)
427 goto out;
428 base = 0;
429 } else
430 base -= xdr->page_len;
431
432 if (base >= xdr->tail[0].iov_len)
433 return 0;
434 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
435out:
436 if (err > 0) {
437 *sent_p += err;
438 err = 0;
439 }
440 return err;
441}
442
443static void xs_nospace_callback(struct rpc_task *task)
444{
445 struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
446
447 transport->inet->sk_write_pending--;
448}
449
450/**
451 * xs_nospace - place task on wait queue if transmit was incomplete
452 * @task: task to put to sleep
453 *
454 */
455static int xs_nospace(struct rpc_task *task)
456{
457 struct rpc_rqst *req = task->tk_rqstp;
458 struct rpc_xprt *xprt = req->rq_xprt;
459 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
460 struct sock *sk = transport->inet;
461 int ret = -EAGAIN;
462
463 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
464 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
465 req->rq_slen);
466
467 /* Protect against races with write_space */
468 spin_lock_bh(&xprt->transport_lock);
469
470 /* Don't race with disconnect */
471 if (xprt_connected(xprt)) {
472 /* wait for more buffer space */
473 sk->sk_write_pending++;
474 xprt_wait_for_buffer_space(task, xs_nospace_callback);
475 } else
476 ret = -ENOTCONN;
477
478 spin_unlock_bh(&xprt->transport_lock);
479
480 /* Race breaker in case memory is freed before above code is called */
481 if (ret == -EAGAIN) {
482 struct socket_wq *wq;
483
484 rcu_read_lock();
485 wq = rcu_dereference(sk->sk_wq);
486 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
487 rcu_read_unlock();
488
489 sk->sk_write_space(sk);
490 }
491 return ret;
492}
493
494/*
495 * Construct a stream transport record marker in @buf.
496 */
497static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
498{
499 u32 reclen = buf->len - sizeof(rpc_fraghdr);
500 rpc_fraghdr *base = buf->head[0].iov_base;
501 *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
502}
503
504/**
505 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
506 * @task: RPC task that manages the state of an RPC request
507 *
508 * Return values:
509 * 0: The request has been sent
510 * EAGAIN: The socket was blocked, please call again later to
511 * complete the request
512 * ENOTCONN: Caller needs to invoke connect logic then call again
513 * other: Some other error occured, the request was not sent
514 */
515static int xs_local_send_request(struct rpc_task *task)
516{
517 struct rpc_rqst *req = task->tk_rqstp;
518 struct rpc_xprt *xprt = req->rq_xprt;
519 struct sock_xprt *transport =
520 container_of(xprt, struct sock_xprt, xprt);
521 struct xdr_buf *xdr = &req->rq_snd_buf;
522 int status;
523 int sent = 0;
524
525 xs_encode_stream_record_marker(&req->rq_snd_buf);
526
527 xs_pktdump("packet data:",
528 req->rq_svec->iov_base, req->rq_svec->iov_len);
529
530 req->rq_xtime = ktime_get();
531 status = xs_sendpages(transport->sock, NULL, 0, xdr, req->rq_bytes_sent,
532 true, &sent);
533 dprintk("RPC: %s(%u) = %d\n",
534 __func__, xdr->len - req->rq_bytes_sent, status);
535
536 if (status == -EAGAIN && sock_writeable(transport->inet))
537 status = -ENOBUFS;
538
539 if (likely(sent > 0) || status == 0) {
540 req->rq_bytes_sent += sent;
541 req->rq_xmit_bytes_sent += sent;
542 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
543 req->rq_bytes_sent = 0;
544 return 0;
545 }
546 status = -EAGAIN;
547 }
548
549 switch (status) {
550 case -ENOBUFS:
551 break;
552 case -EAGAIN:
553 status = xs_nospace(task);
554 break;
555 default:
556 dprintk("RPC: sendmsg returned unrecognized error %d\n",
557 -status);
558 /* fall through */
559 case -EPIPE:
560 xs_close(xprt);
561 status = -ENOTCONN;
562 }
563
564 return status;
565}
566
567/**
568 * xs_udp_send_request - write an RPC request to a UDP socket
569 * @task: address of RPC task that manages the state of an RPC request
570 *
571 * Return values:
572 * 0: The request has been sent
573 * EAGAIN: The socket was blocked, please call again later to
574 * complete the request
575 * ENOTCONN: Caller needs to invoke connect logic then call again
576 * other: Some other error occurred, the request was not sent
577 */
578static int xs_udp_send_request(struct rpc_task *task)
579{
580 struct rpc_rqst *req = task->tk_rqstp;
581 struct rpc_xprt *xprt = req->rq_xprt;
582 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
583 struct xdr_buf *xdr = &req->rq_snd_buf;
584 int sent = 0;
585 int status;
586
587 xs_pktdump("packet data:",
588 req->rq_svec->iov_base,
589 req->rq_svec->iov_len);
590
591 if (!xprt_bound(xprt))
592 return -ENOTCONN;
593 req->rq_xtime = ktime_get();
594 status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
595 xdr, req->rq_bytes_sent, true, &sent);
596
597 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
598 xdr->len - req->rq_bytes_sent, status);
599
600 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
601 if (status == -EPERM)
602 goto process_status;
603
604 if (status == -EAGAIN && sock_writeable(transport->inet))
605 status = -ENOBUFS;
606
607 if (sent > 0 || status == 0) {
608 req->rq_xmit_bytes_sent += sent;
609 if (sent >= req->rq_slen)
610 return 0;
611 /* Still some bytes left; set up for a retry later. */
612 status = -EAGAIN;
613 }
614
615process_status:
616 switch (status) {
617 case -ENOTSOCK:
618 status = -ENOTCONN;
619 /* Should we call xs_close() here? */
620 break;
621 case -EAGAIN:
622 status = xs_nospace(task);
623 break;
624 case -ENETUNREACH:
625 case -ENOBUFS:
626 case -EPIPE:
627 case -ECONNREFUSED:
628 case -EPERM:
629 /* When the server has died, an ICMP port unreachable message
630 * prompts ECONNREFUSED. */
631 break;
632 default:
633 dprintk("RPC: sendmsg returned unrecognized error %d\n",
634 -status);
635 }
636
637 return status;
638}
639
640/**
641 * xs_tcp_send_request - write an RPC request to a TCP socket
642 * @task: address of RPC task that manages the state of an RPC request
643 *
644 * Return values:
645 * 0: The request has been sent
646 * EAGAIN: The socket was blocked, please call again later to
647 * complete the request
648 * ENOTCONN: Caller needs to invoke connect logic then call again
649 * other: Some other error occurred, the request was not sent
650 *
651 * XXX: In the case of soft timeouts, should we eventually give up
652 * if sendmsg is not able to make progress?
653 */
654static int xs_tcp_send_request(struct rpc_task *task)
655{
656 struct rpc_rqst *req = task->tk_rqstp;
657 struct rpc_xprt *xprt = req->rq_xprt;
658 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
659 struct xdr_buf *xdr = &req->rq_snd_buf;
660 bool zerocopy = true;
661 bool vm_wait = false;
662 int status;
663 int sent;
664
665 xs_encode_stream_record_marker(&req->rq_snd_buf);
666
667 xs_pktdump("packet data:",
668 req->rq_svec->iov_base,
669 req->rq_svec->iov_len);
670 /* Don't use zero copy if this is a resend. If the RPC call
671 * completes while the socket holds a reference to the pages,
672 * then we may end up resending corrupted data.
673 */
674 if (task->tk_flags & RPC_TASK_SENT)
675 zerocopy = false;
676
677 if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
678 xs_tcp_set_socket_timeouts(xprt, transport->sock);
679
680 /* Continue transmitting the packet/record. We must be careful
681 * to cope with writespace callbacks arriving _after_ we have
682 * called sendmsg(). */
683 req->rq_xtime = ktime_get();
684 while (1) {
685 sent = 0;
686 status = xs_sendpages(transport->sock, NULL, 0, xdr,
687 req->rq_bytes_sent, zerocopy, &sent);
688
689 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
690 xdr->len - req->rq_bytes_sent, status);
691
692 /* If we've sent the entire packet, immediately
693 * reset the count of bytes sent. */
694 req->rq_bytes_sent += sent;
695 req->rq_xmit_bytes_sent += sent;
696 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
697 req->rq_bytes_sent = 0;
698 return 0;
699 }
700
701 WARN_ON_ONCE(sent == 0 && status == 0);
702
703 if (status == -EAGAIN ) {
704 /*
705 * Return EAGAIN if we're sure we're hitting the
706 * socket send buffer limits.
707 */
708 if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
709 break;
710 /*
711 * Did we hit a memory allocation failure?
712 */
713 if (sent == 0) {
714 status = -ENOBUFS;
715 if (vm_wait)
716 break;
717 /* Retry, knowing now that we're below the
718 * socket send buffer limit
719 */
720 vm_wait = true;
721 }
722 continue;
723 }
724 if (status < 0)
725 break;
726 vm_wait = false;
727 }
728
729 switch (status) {
730 case -ENOTSOCK:
731 status = -ENOTCONN;
732 /* Should we call xs_close() here? */
733 break;
734 case -EAGAIN:
735 status = xs_nospace(task);
736 break;
737 case -ECONNRESET:
738 case -ECONNREFUSED:
739 case -ENOTCONN:
740 case -EADDRINUSE:
741 case -ENOBUFS:
742 case -EPIPE:
743 break;
744 default:
745 dprintk("RPC: sendmsg returned unrecognized error %d\n",
746 -status);
747 }
748
749 return status;
750}
751
752/**
753 * xs_tcp_release_xprt - clean up after a tcp transmission
754 * @xprt: transport
755 * @task: rpc task
756 *
757 * This cleans up if an error causes us to abort the transmission of a request.
758 * In this case, the socket may need to be reset in order to avoid confusing
759 * the server.
760 */
761static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
762{
763 struct rpc_rqst *req;
764
765 if (task != xprt->snd_task)
766 return;
767 if (task == NULL)
768 goto out_release;
769 req = task->tk_rqstp;
770 if (req == NULL)
771 goto out_release;
772 if (req->rq_bytes_sent == 0)
773 goto out_release;
774 if (req->rq_bytes_sent == req->rq_snd_buf.len)
775 goto out_release;
776 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
777out_release:
778 xprt_release_xprt(xprt, task);
779}
780
781static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
782{
783 transport->old_data_ready = sk->sk_data_ready;
784 transport->old_state_change = sk->sk_state_change;
785 transport->old_write_space = sk->sk_write_space;
786 transport->old_error_report = sk->sk_error_report;
787}
788
789static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
790{
791 sk->sk_data_ready = transport->old_data_ready;
792 sk->sk_state_change = transport->old_state_change;
793 sk->sk_write_space = transport->old_write_space;
794 sk->sk_error_report = transport->old_error_report;
795}
796
797static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
798{
799 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
800
801 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
802}
803
804static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
805{
806 smp_mb__before_atomic();
807 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
808 clear_bit(XPRT_CLOSING, &xprt->state);
809 xs_sock_reset_state_flags(xprt);
810 smp_mb__after_atomic();
811}
812
813/**
814 * xs_error_report - callback to handle TCP socket state errors
815 * @sk: socket
816 *
817 * Note: we don't call sock_error() since there may be a rpc_task
818 * using the socket, and so we don't want to clear sk->sk_err.
819 */
820static void xs_error_report(struct sock *sk)
821{
822 struct rpc_xprt *xprt;
823 int err;
824
825 read_lock_bh(&sk->sk_callback_lock);
826 if (!(xprt = xprt_from_sock(sk)))
827 goto out;
828
829 err = -sk->sk_err;
830 if (err == 0)
831 goto out;
832 dprintk("RPC: xs_error_report client %p, error=%d...\n",
833 xprt, -err);
834 trace_rpc_socket_error(xprt, sk->sk_socket, err);
835 xprt_wake_pending_tasks(xprt, err);
836 out:
837 read_unlock_bh(&sk->sk_callback_lock);
838}
839
840static void xs_reset_transport(struct sock_xprt *transport)
841{
842 struct socket *sock = transport->sock;
843 struct sock *sk = transport->inet;
844 struct rpc_xprt *xprt = &transport->xprt;
845
846 if (sk == NULL)
847 return;
848
849 if (atomic_read(&transport->xprt.swapper))
850 sk_clear_memalloc(sk);
851
852 kernel_sock_shutdown(sock, SHUT_RDWR);
853
854 mutex_lock(&transport->recv_mutex);
855 write_lock_bh(&sk->sk_callback_lock);
856 transport->inet = NULL;
857 transport->sock = NULL;
858
859 sk->sk_user_data = NULL;
860
861 xs_restore_old_callbacks(transport, sk);
862 xprt_clear_connected(xprt);
863 write_unlock_bh(&sk->sk_callback_lock);
864 xs_sock_reset_connection_flags(xprt);
865 mutex_unlock(&transport->recv_mutex);
866
867 trace_rpc_socket_close(xprt, sock);
868 sock_release(sock);
869}
870
871/**
872 * xs_close - close a socket
873 * @xprt: transport
874 *
875 * This is used when all requests are complete; ie, no DRC state remains
876 * on the server we want to save.
877 *
878 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
879 * xs_reset_transport() zeroing the socket from underneath a writer.
880 */
881static void xs_close(struct rpc_xprt *xprt)
882{
883 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
884
885 dprintk("RPC: xs_close xprt %p\n", xprt);
886
887 xs_reset_transport(transport);
888 xprt->reestablish_timeout = 0;
889
890 xprt_disconnect_done(xprt);
891}
892
893static void xs_inject_disconnect(struct rpc_xprt *xprt)
894{
895 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
896 xprt);
897 xprt_disconnect_done(xprt);
898}
899
900static void xs_xprt_free(struct rpc_xprt *xprt)
901{
902 xs_free_peer_addresses(xprt);
903 xprt_free(xprt);
904}
905
906/**
907 * xs_destroy - prepare to shutdown a transport
908 * @xprt: doomed transport
909 *
910 */
911static void xs_destroy(struct rpc_xprt *xprt)
912{
913 struct sock_xprt *transport = container_of(xprt,
914 struct sock_xprt, xprt);
915 dprintk("RPC: xs_destroy xprt %p\n", xprt);
916
917 cancel_delayed_work_sync(&transport->connect_worker);
918 xs_close(xprt);
919 cancel_work_sync(&transport->recv_worker);
920 xs_xprt_free(xprt);
921 module_put(THIS_MODULE);
922}
923
924static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
925{
926 struct xdr_skb_reader desc = {
927 .skb = skb,
928 .offset = sizeof(rpc_fraghdr),
929 .count = skb->len - sizeof(rpc_fraghdr),
930 };
931
932 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
933 return -1;
934 if (desc.count)
935 return -1;
936 return 0;
937}
938
939/**
940 * xs_local_data_read_skb
941 * @xprt: transport
942 * @sk: socket
943 * @skb: skbuff
944 *
945 * Currently this assumes we can read the whole reply in a single gulp.
946 */
947static void xs_local_data_read_skb(struct rpc_xprt *xprt,
948 struct sock *sk,
949 struct sk_buff *skb)
950{
951 struct rpc_task *task;
952 struct rpc_rqst *rovr;
953 int repsize, copied;
954 u32 _xid;
955 __be32 *xp;
956
957 repsize = skb->len - sizeof(rpc_fraghdr);
958 if (repsize < 4) {
959 dprintk("RPC: impossible RPC reply size %d\n", repsize);
960 return;
961 }
962
963 /* Copy the XID from the skb... */
964 xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
965 if (xp == NULL)
966 return;
967
968 /* Look up and lock the request corresponding to the given XID */
969 spin_lock(&xprt->recv_lock);
970 rovr = xprt_lookup_rqst(xprt, *xp);
971 if (!rovr)
972 goto out_unlock;
973 xprt_pin_rqst(rovr);
974 spin_unlock(&xprt->recv_lock);
975 task = rovr->rq_task;
976
977 copied = rovr->rq_private_buf.buflen;
978 if (copied > repsize)
979 copied = repsize;
980
981 if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
982 dprintk("RPC: sk_buff copy failed\n");
983 spin_lock(&xprt->recv_lock);
984 goto out_unpin;
985 }
986
987 spin_lock(&xprt->recv_lock);
988 xprt_complete_rqst(task, copied);
989out_unpin:
990 xprt_unpin_rqst(rovr);
991 out_unlock:
992 spin_unlock(&xprt->recv_lock);
993}
994
995static void xs_local_data_receive(struct sock_xprt *transport)
996{
997 struct sk_buff *skb;
998 struct sock *sk;
999 int err;
1000
1001restart:
1002 mutex_lock(&transport->recv_mutex);
1003 sk = transport->inet;
1004 if (sk == NULL)
1005 goto out;
1006 for (;;) {
1007 skb = skb_recv_datagram(sk, 0, 1, &err);
1008 if (skb != NULL) {
1009 xs_local_data_read_skb(&transport->xprt, sk, skb);
1010 skb_free_datagram(sk, skb);
1011 continue;
1012 }
1013 if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1014 break;
1015 if (need_resched()) {
1016 mutex_unlock(&transport->recv_mutex);
1017 cond_resched();
1018 goto restart;
1019 }
1020 }
1021out:
1022 mutex_unlock(&transport->recv_mutex);
1023}
1024
1025static void xs_local_data_receive_workfn(struct work_struct *work)
1026{
1027 struct sock_xprt *transport =
1028 container_of(work, struct sock_xprt, recv_worker);
1029 xs_local_data_receive(transport);
1030}
1031
1032/**
1033 * xs_udp_data_read_skb - receive callback for UDP sockets
1034 * @xprt: transport
1035 * @sk: socket
1036 * @skb: skbuff
1037 *
1038 */
1039static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1040 struct sock *sk,
1041 struct sk_buff *skb)
1042{
1043 struct rpc_task *task;
1044 struct rpc_rqst *rovr;
1045 int repsize, copied;
1046 u32 _xid;
1047 __be32 *xp;
1048
1049 repsize = skb->len;
1050 if (repsize < 4) {
1051 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
1052 return;
1053 }
1054
1055 /* Copy the XID from the skb... */
1056 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1057 if (xp == NULL)
1058 return;
1059
1060 /* Look up and lock the request corresponding to the given XID */
1061 spin_lock(&xprt->recv_lock);
1062 rovr = xprt_lookup_rqst(xprt, *xp);
1063 if (!rovr)
1064 goto out_unlock;
1065 xprt_pin_rqst(rovr);
1066 xprt_update_rtt(rovr->rq_task);
1067 spin_unlock(&xprt->recv_lock);
1068 task = rovr->rq_task;
1069
1070 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1071 copied = repsize;
1072
1073 /* Suck it into the iovec, verify checksum if not done by hw. */
1074 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1075 spin_lock(&xprt->recv_lock);
1076 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1077 goto out_unpin;
1078 }
1079
1080
1081 spin_lock_bh(&xprt->transport_lock);
1082 xprt_adjust_cwnd(xprt, task, copied);
1083 spin_unlock_bh(&xprt->transport_lock);
1084 spin_lock(&xprt->recv_lock);
1085 xprt_complete_rqst(task, copied);
1086 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1087out_unpin:
1088 xprt_unpin_rqst(rovr);
1089 out_unlock:
1090 spin_unlock(&xprt->recv_lock);
1091}
1092
1093static void xs_udp_data_receive(struct sock_xprt *transport)
1094{
1095 struct sk_buff *skb;
1096 struct sock *sk;
1097 int err;
1098
1099restart:
1100 mutex_lock(&transport->recv_mutex);
1101 sk = transport->inet;
1102 if (sk == NULL)
1103 goto out;
1104 for (;;) {
1105 skb = skb_recv_udp(sk, 0, 1, &err);
1106 if (skb != NULL) {
1107 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1108 consume_skb(skb);
1109 continue;
1110 }
1111 if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1112 break;
1113 if (need_resched()) {
1114 mutex_unlock(&transport->recv_mutex);
1115 cond_resched();
1116 goto restart;
1117 }
1118 }
1119out:
1120 mutex_unlock(&transport->recv_mutex);
1121}
1122
1123static void xs_udp_data_receive_workfn(struct work_struct *work)
1124{
1125 struct sock_xprt *transport =
1126 container_of(work, struct sock_xprt, recv_worker);
1127 xs_udp_data_receive(transport);
1128}
1129
1130/**
1131 * xs_data_ready - "data ready" callback for UDP sockets
1132 * @sk: socket with data to read
1133 *
1134 */
1135static void xs_data_ready(struct sock *sk)
1136{
1137 struct rpc_xprt *xprt;
1138
1139 read_lock_bh(&sk->sk_callback_lock);
1140 dprintk("RPC: xs_data_ready...\n");
1141 xprt = xprt_from_sock(sk);
1142 if (xprt != NULL) {
1143 struct sock_xprt *transport = container_of(xprt,
1144 struct sock_xprt, xprt);
1145 transport->old_data_ready(sk);
1146 /* Any data means we had a useful conversation, so
1147 * then we don't need to delay the next reconnect
1148 */
1149 if (xprt->reestablish_timeout)
1150 xprt->reestablish_timeout = 0;
1151 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1152 queue_work(xprtiod_workqueue, &transport->recv_worker);
1153 }
1154 read_unlock_bh(&sk->sk_callback_lock);
1155}
1156
1157/*
1158 * Helper function to force a TCP close if the server is sending
1159 * junk and/or it has put us in CLOSE_WAIT
1160 */
1161static void xs_tcp_force_close(struct rpc_xprt *xprt)
1162{
1163 xprt_force_disconnect(xprt);
1164}
1165
1166static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1167{
1168 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1169 size_t len, used;
1170 char *p;
1171
1172 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1173 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1174 used = xdr_skb_read_bits(desc, p, len);
1175 transport->tcp_offset += used;
1176 if (used != len)
1177 return;
1178
1179 transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1180 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1181 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1182 else
1183 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1184 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1185
1186 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1187 transport->tcp_offset = 0;
1188
1189 /* Sanity check of the record length */
1190 if (unlikely(transport->tcp_reclen < 8)) {
1191 dprintk("RPC: invalid TCP record fragment length\n");
1192 xs_tcp_force_close(xprt);
1193 return;
1194 }
1195 dprintk("RPC: reading TCP record fragment of length %d\n",
1196 transport->tcp_reclen);
1197}
1198
1199static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1200{
1201 if (transport->tcp_offset == transport->tcp_reclen) {
1202 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1203 transport->tcp_offset = 0;
1204 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1205 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1206 transport->tcp_flags |= TCP_RCV_COPY_XID;
1207 transport->tcp_copied = 0;
1208 }
1209 }
1210}
1211
1212static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1213{
1214 size_t len, used;
1215 char *p;
1216
1217 len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1218 dprintk("RPC: reading XID (%zu bytes)\n", len);
1219 p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1220 used = xdr_skb_read_bits(desc, p, len);
1221 transport->tcp_offset += used;
1222 if (used != len)
1223 return;
1224 transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1225 transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1226 transport->tcp_copied = 4;
1227 dprintk("RPC: reading %s XID %08x\n",
1228 (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1229 : "request with",
1230 ntohl(transport->tcp_xid));
1231 xs_tcp_check_fraghdr(transport);
1232}
1233
1234static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1235 struct xdr_skb_reader *desc)
1236{
1237 size_t len, used;
1238 u32 offset;
1239 char *p;
1240
1241 /*
1242 * We want transport->tcp_offset to be 8 at the end of this routine
1243 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1244 * When this function is called for the first time,
1245 * transport->tcp_offset is 4 (after having already read the xid).
1246 */
1247 offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1248 len = sizeof(transport->tcp_calldir) - offset;
1249 dprintk("RPC: reading CALL/REPLY flag (%zu bytes)\n", len);
1250 p = ((char *) &transport->tcp_calldir) + offset;
1251 used = xdr_skb_read_bits(desc, p, len);
1252 transport->tcp_offset += used;
1253 if (used != len)
1254 return;
1255 transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1256 /*
1257 * We don't yet have the XDR buffer, so we will write the calldir
1258 * out after we get the buffer from the 'struct rpc_rqst'
1259 */
1260 switch (ntohl(transport->tcp_calldir)) {
1261 case RPC_REPLY:
1262 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1263 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1264 transport->tcp_flags |= TCP_RPC_REPLY;
1265 break;
1266 case RPC_CALL:
1267 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1268 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1269 transport->tcp_flags &= ~TCP_RPC_REPLY;
1270 break;
1271 default:
1272 dprintk("RPC: invalid request message type\n");
1273 xs_tcp_force_close(&transport->xprt);
1274 }
1275 xs_tcp_check_fraghdr(transport);
1276}
1277
1278static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1279 struct xdr_skb_reader *desc,
1280 struct rpc_rqst *req)
1281{
1282 struct sock_xprt *transport =
1283 container_of(xprt, struct sock_xprt, xprt);
1284 struct xdr_buf *rcvbuf;
1285 size_t len;
1286 ssize_t r;
1287
1288 rcvbuf = &req->rq_private_buf;
1289
1290 if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1291 /*
1292 * Save the RPC direction in the XDR buffer
1293 */
1294 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1295 &transport->tcp_calldir,
1296 sizeof(transport->tcp_calldir));
1297 transport->tcp_copied += sizeof(transport->tcp_calldir);
1298 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1299 }
1300
1301 len = desc->count;
1302 if (len > transport->tcp_reclen - transport->tcp_offset)
1303 desc->count = transport->tcp_reclen - transport->tcp_offset;
1304 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1305 desc, xdr_skb_read_bits);
1306
1307 if (desc->count) {
1308 /* Error when copying to the receive buffer,
1309 * usually because we weren't able to allocate
1310 * additional buffer pages. All we can do now
1311 * is turn off TCP_RCV_COPY_DATA, so the request
1312 * will not receive any additional updates,
1313 * and time out.
1314 * Any remaining data from this record will
1315 * be discarded.
1316 */
1317 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1318 dprintk("RPC: XID %08x truncated request\n",
1319 ntohl(transport->tcp_xid));
1320 dprintk("RPC: xprt = %p, tcp_copied = %lu, "
1321 "tcp_offset = %u, tcp_reclen = %u\n",
1322 xprt, transport->tcp_copied,
1323 transport->tcp_offset, transport->tcp_reclen);
1324 return;
1325 }
1326
1327 transport->tcp_copied += r;
1328 transport->tcp_offset += r;
1329 desc->count = len - r;
1330
1331 dprintk("RPC: XID %08x read %zd bytes\n",
1332 ntohl(transport->tcp_xid), r);
1333 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1334 "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1335 transport->tcp_offset, transport->tcp_reclen);
1336
1337 if (transport->tcp_copied == req->rq_private_buf.buflen)
1338 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1339 else if (transport->tcp_offset == transport->tcp_reclen) {
1340 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1341 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1342 }
1343}
1344
1345/*
1346 * Finds the request corresponding to the RPC xid and invokes the common
1347 * tcp read code to read the data.
1348 */
1349static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1350 struct xdr_skb_reader *desc)
1351{
1352 struct sock_xprt *transport =
1353 container_of(xprt, struct sock_xprt, xprt);
1354 struct rpc_rqst *req;
1355
1356 dprintk("RPC: read reply XID %08x\n", ntohl(transport->tcp_xid));
1357
1358 /* Find and lock the request corresponding to this xid */
1359 spin_lock(&xprt->recv_lock);
1360 req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1361 if (!req) {
1362 dprintk("RPC: XID %08x request not found!\n",
1363 ntohl(transport->tcp_xid));
1364 spin_unlock(&xprt->recv_lock);
1365 return -1;
1366 }
1367 xprt_pin_rqst(req);
1368 spin_unlock(&xprt->recv_lock);
1369
1370 xs_tcp_read_common(xprt, desc, req);
1371
1372 spin_lock(&xprt->recv_lock);
1373 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1374 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1375 xprt_unpin_rqst(req);
1376 spin_unlock(&xprt->recv_lock);
1377 return 0;
1378}
1379
1380#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1381/*
1382 * Obtains an rpc_rqst previously allocated and invokes the common
1383 * tcp read code to read the data. The result is placed in the callback
1384 * queue.
1385 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1386 * connection and return -1.
1387 */
1388static int xs_tcp_read_callback(struct rpc_xprt *xprt,
1389 struct xdr_skb_reader *desc)
1390{
1391 struct sock_xprt *transport =
1392 container_of(xprt, struct sock_xprt, xprt);
1393 struct rpc_rqst *req;
1394
1395 /* Look up the request corresponding to the given XID */
1396 req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
1397 if (req == NULL) {
1398 printk(KERN_WARNING "Callback slot table overflowed\n");
1399 xprt_force_disconnect(xprt);
1400 return -1;
1401 }
1402
1403 dprintk("RPC: read callback XID %08x\n", ntohl(req->rq_xid));
1404 xs_tcp_read_common(xprt, desc, req);
1405
1406 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1407 xprt_complete_bc_request(req, transport->tcp_copied);
1408
1409 return 0;
1410}
1411
1412static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1413 struct xdr_skb_reader *desc)
1414{
1415 struct sock_xprt *transport =
1416 container_of(xprt, struct sock_xprt, xprt);
1417
1418 return (transport->tcp_flags & TCP_RPC_REPLY) ?
1419 xs_tcp_read_reply(xprt, desc) :
1420 xs_tcp_read_callback(xprt, desc);
1421}
1422
1423static int xs_tcp_bc_up(struct svc_serv *serv, struct net *net)
1424{
1425 int ret;
1426
1427 ret = svc_create_xprt(serv, "tcp-bc", net, PF_INET, 0,
1428 SVC_SOCK_ANONYMOUS);
1429 if (ret < 0)
1430 return ret;
1431 return 0;
1432}
1433
1434static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1435{
1436 return PAGE_SIZE;
1437}
1438#else
1439static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1440 struct xdr_skb_reader *desc)
1441{
1442 return xs_tcp_read_reply(xprt, desc);
1443}
1444#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1445
1446/*
1447 * Read data off the transport. This can be either an RPC_CALL or an
1448 * RPC_REPLY. Relay the processing to helper functions.
1449 */
1450static void xs_tcp_read_data(struct rpc_xprt *xprt,
1451 struct xdr_skb_reader *desc)
1452{
1453 struct sock_xprt *transport =
1454 container_of(xprt, struct sock_xprt, xprt);
1455
1456 if (_xs_tcp_read_data(xprt, desc) == 0)
1457 xs_tcp_check_fraghdr(transport);
1458 else {
1459 /*
1460 * The transport_lock protects the request handling.
1461 * There's no need to hold it to update the tcp_flags.
1462 */
1463 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1464 }
1465}
1466
1467static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1468{
1469 size_t len;
1470
1471 len = transport->tcp_reclen - transport->tcp_offset;
1472 if (len > desc->count)
1473 len = desc->count;
1474 desc->count -= len;
1475 desc->offset += len;
1476 transport->tcp_offset += len;
1477 dprintk("RPC: discarded %zu bytes\n", len);
1478 xs_tcp_check_fraghdr(transport);
1479}
1480
1481static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1482{
1483 struct rpc_xprt *xprt = rd_desc->arg.data;
1484 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1485 struct xdr_skb_reader desc = {
1486 .skb = skb,
1487 .offset = offset,
1488 .count = len,
1489 };
1490 size_t ret;
1491
1492 dprintk("RPC: xs_tcp_data_recv started\n");
1493 do {
1494 trace_xs_tcp_data_recv(transport);
1495 /* Read in a new fragment marker if necessary */
1496 /* Can we ever really expect to get completely empty fragments? */
1497 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1498 xs_tcp_read_fraghdr(xprt, &desc);
1499 continue;
1500 }
1501 /* Read in the xid if necessary */
1502 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1503 xs_tcp_read_xid(transport, &desc);
1504 continue;
1505 }
1506 /* Read in the call/reply flag */
1507 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1508 xs_tcp_read_calldir(transport, &desc);
1509 continue;
1510 }
1511 /* Read in the request data */
1512 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1513 xs_tcp_read_data(xprt, &desc);
1514 continue;
1515 }
1516 /* Skip over any trailing bytes on short reads */
1517 xs_tcp_read_discard(transport, &desc);
1518 } while (desc.count);
1519 ret = len - desc.count;
1520 if (ret < rd_desc->count)
1521 rd_desc->count -= ret;
1522 else
1523 rd_desc->count = 0;
1524 trace_xs_tcp_data_recv(transport);
1525 dprintk("RPC: xs_tcp_data_recv done\n");
1526 return ret;
1527}
1528
1529static void xs_tcp_data_receive(struct sock_xprt *transport)
1530{
1531 struct rpc_xprt *xprt = &transport->xprt;
1532 struct sock *sk;
1533 read_descriptor_t rd_desc = {
1534 .arg.data = xprt,
1535 };
1536 unsigned long total = 0;
1537 int read = 0;
1538
1539restart:
1540 mutex_lock(&transport->recv_mutex);
1541 sk = transport->inet;
1542 if (sk == NULL)
1543 goto out;
1544
1545 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1546 for (;;) {
1547 rd_desc.count = RPC_TCP_READ_CHUNK_SZ;
1548 lock_sock(sk);
1549 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1550 if (rd_desc.count != 0 || read < 0) {
1551 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1552 release_sock(sk);
1553 break;
1554 }
1555 release_sock(sk);
1556 total += read;
1557 if (need_resched()) {
1558 mutex_unlock(&transport->recv_mutex);
1559 cond_resched();
1560 goto restart;
1561 }
1562 }
1563 if (test_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1564 queue_work(xprtiod_workqueue, &transport->recv_worker);
1565out:
1566 mutex_unlock(&transport->recv_mutex);
1567 trace_xs_tcp_data_ready(xprt, read, total);
1568}
1569
1570static void xs_tcp_data_receive_workfn(struct work_struct *work)
1571{
1572 struct sock_xprt *transport =
1573 container_of(work, struct sock_xprt, recv_worker);
1574 xs_tcp_data_receive(transport);
1575}
1576
1577/**
1578 * xs_tcp_state_change - callback to handle TCP socket state changes
1579 * @sk: socket whose state has changed
1580 *
1581 */
1582static void xs_tcp_state_change(struct sock *sk)
1583{
1584 struct rpc_xprt *xprt;
1585 struct sock_xprt *transport;
1586
1587 read_lock_bh(&sk->sk_callback_lock);
1588 if (!(xprt = xprt_from_sock(sk)))
1589 goto out;
1590 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1591 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1592 sk->sk_state, xprt_connected(xprt),
1593 sock_flag(sk, SOCK_DEAD),
1594 sock_flag(sk, SOCK_ZAPPED),
1595 sk->sk_shutdown);
1596
1597 transport = container_of(xprt, struct sock_xprt, xprt);
1598 trace_rpc_socket_state_change(xprt, sk->sk_socket);
1599 switch (sk->sk_state) {
1600 case TCP_ESTABLISHED:
1601 spin_lock(&xprt->transport_lock);
1602 if (!xprt_test_and_set_connected(xprt)) {
1603
1604 /* Reset TCP record info */
1605 transport->tcp_offset = 0;
1606 transport->tcp_reclen = 0;
1607 transport->tcp_copied = 0;
1608 transport->tcp_flags =
1609 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1610 xprt->connect_cookie++;
1611 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1612 xprt_clear_connecting(xprt);
1613
1614 xprt_wake_pending_tasks(xprt, -EAGAIN);
1615 }
1616 spin_unlock(&xprt->transport_lock);
1617 break;
1618 case TCP_FIN_WAIT1:
1619 /* The client initiated a shutdown of the socket */
1620 xprt->connect_cookie++;
1621 xprt->reestablish_timeout = 0;
1622 set_bit(XPRT_CLOSING, &xprt->state);
1623 smp_mb__before_atomic();
1624 clear_bit(XPRT_CONNECTED, &xprt->state);
1625 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1626 smp_mb__after_atomic();
1627 break;
1628 case TCP_CLOSE_WAIT:
1629 /* The server initiated a shutdown of the socket */
1630 xprt->connect_cookie++;
1631 clear_bit(XPRT_CONNECTED, &xprt->state);
1632 xs_tcp_force_close(xprt);
1633 /* fall through */
1634 case TCP_CLOSING:
1635 /*
1636 * If the server closed down the connection, make sure that
1637 * we back off before reconnecting
1638 */
1639 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1640 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1641 break;
1642 case TCP_LAST_ACK:
1643 set_bit(XPRT_CLOSING, &xprt->state);
1644 smp_mb__before_atomic();
1645 clear_bit(XPRT_CONNECTED, &xprt->state);
1646 smp_mb__after_atomic();
1647 break;
1648 case TCP_CLOSE:
1649 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1650 &transport->sock_state))
1651 xprt_clear_connecting(xprt);
1652 clear_bit(XPRT_CLOSING, &xprt->state);
1653 if (sk->sk_err)
1654 xprt_wake_pending_tasks(xprt, -sk->sk_err);
1655 /* Trigger the socket release */
1656 xs_tcp_force_close(xprt);
1657 }
1658 out:
1659 read_unlock_bh(&sk->sk_callback_lock);
1660}
1661
1662static void xs_write_space(struct sock *sk)
1663{
1664 struct socket_wq *wq;
1665 struct rpc_xprt *xprt;
1666
1667 if (!sk->sk_socket)
1668 return;
1669 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1670
1671 if (unlikely(!(xprt = xprt_from_sock(sk))))
1672 return;
1673 rcu_read_lock();
1674 wq = rcu_dereference(sk->sk_wq);
1675 if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1676 goto out;
1677
1678 xprt_write_space(xprt);
1679out:
1680 rcu_read_unlock();
1681}
1682
1683/**
1684 * xs_udp_write_space - callback invoked when socket buffer space
1685 * becomes available
1686 * @sk: socket whose state has changed
1687 *
1688 * Called when more output buffer space is available for this socket.
1689 * We try not to wake our writers until they can make "significant"
1690 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1691 * with a bunch of small requests.
1692 */
1693static void xs_udp_write_space(struct sock *sk)
1694{
1695 read_lock_bh(&sk->sk_callback_lock);
1696
1697 /* from net/core/sock.c:sock_def_write_space */
1698 if (sock_writeable(sk))
1699 xs_write_space(sk);
1700
1701 read_unlock_bh(&sk->sk_callback_lock);
1702}
1703
1704/**
1705 * xs_tcp_write_space - callback invoked when socket buffer space
1706 * becomes available
1707 * @sk: socket whose state has changed
1708 *
1709 * Called when more output buffer space is available for this socket.
1710 * We try not to wake our writers until they can make "significant"
1711 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1712 * with a bunch of small requests.
1713 */
1714static void xs_tcp_write_space(struct sock *sk)
1715{
1716 read_lock_bh(&sk->sk_callback_lock);
1717
1718 /* from net/core/stream.c:sk_stream_write_space */
1719 if (sk_stream_is_writeable(sk))
1720 xs_write_space(sk);
1721
1722 read_unlock_bh(&sk->sk_callback_lock);
1723}
1724
1725static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1726{
1727 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1728 struct sock *sk = transport->inet;
1729
1730 if (transport->rcvsize) {
1731 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1732 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1733 }
1734 if (transport->sndsize) {
1735 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1736 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1737 sk->sk_write_space(sk);
1738 }
1739}
1740
1741/**
1742 * xs_udp_set_buffer_size - set send and receive limits
1743 * @xprt: generic transport
1744 * @sndsize: requested size of send buffer, in bytes
1745 * @rcvsize: requested size of receive buffer, in bytes
1746 *
1747 * Set socket send and receive buffer size limits.
1748 */
1749static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1750{
1751 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1752
1753 transport->sndsize = 0;
1754 if (sndsize)
1755 transport->sndsize = sndsize + 1024;
1756 transport->rcvsize = 0;
1757 if (rcvsize)
1758 transport->rcvsize = rcvsize + 1024;
1759
1760 xs_udp_do_set_buffer_size(xprt);
1761}
1762
1763/**
1764 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1765 * @task: task that timed out
1766 *
1767 * Adjust the congestion window after a retransmit timeout has occurred.
1768 */
1769static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1770{
1771 spin_lock_bh(&xprt->transport_lock);
1772 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1773 spin_unlock_bh(&xprt->transport_lock);
1774}
1775
1776static unsigned short xs_get_random_port(void)
1777{
1778 unsigned short range = xprt_max_resvport - xprt_min_resvport + 1;
1779 unsigned short rand = (unsigned short) prandom_u32() % range;
1780 return rand + xprt_min_resvport;
1781}
1782
1783/**
1784 * xs_set_reuseaddr_port - set the socket's port and address reuse options
1785 * @sock: socket
1786 *
1787 * Note that this function has to be called on all sockets that share the
1788 * same port, and it must be called before binding.
1789 */
1790static void xs_sock_set_reuseport(struct socket *sock)
1791{
1792 int opt = 1;
1793
1794 kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1795 (char *)&opt, sizeof(opt));
1796}
1797
1798static unsigned short xs_sock_getport(struct socket *sock)
1799{
1800 struct sockaddr_storage buf;
1801 unsigned short port = 0;
1802
1803 if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1804 goto out;
1805 switch (buf.ss_family) {
1806 case AF_INET6:
1807 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1808 break;
1809 case AF_INET:
1810 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1811 }
1812out:
1813 return port;
1814}
1815
1816/**
1817 * xs_set_port - reset the port number in the remote endpoint address
1818 * @xprt: generic transport
1819 * @port: new port number
1820 *
1821 */
1822static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1823{
1824 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1825
1826 rpc_set_port(xs_addr(xprt), port);
1827 xs_update_peer_port(xprt);
1828}
1829
1830static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1831{
1832 if (transport->srcport == 0)
1833 transport->srcport = xs_sock_getport(sock);
1834}
1835
1836static unsigned short xs_get_srcport(struct sock_xprt *transport)
1837{
1838 unsigned short port = transport->srcport;
1839
1840 if (port == 0 && transport->xprt.resvport)
1841 port = xs_get_random_port();
1842 return port;
1843}
1844
1845static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1846{
1847 if (transport->srcport != 0)
1848 transport->srcport = 0;
1849 if (!transport->xprt.resvport)
1850 return 0;
1851 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1852 return xprt_max_resvport;
1853 return --port;
1854}
1855static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1856{
1857 struct sockaddr_storage myaddr;
1858 int err, nloop = 0;
1859 unsigned short port = xs_get_srcport(transport);
1860 unsigned short last;
1861
1862 /*
1863 * If we are asking for any ephemeral port (i.e. port == 0 &&
1864 * transport->xprt.resvport == 0), don't bind. Let the local
1865 * port selection happen implicitly when the socket is used
1866 * (for example at connect time).
1867 *
1868 * This ensures that we can continue to establish TCP
1869 * connections even when all local ephemeral ports are already
1870 * a part of some TCP connection. This makes no difference
1871 * for UDP sockets, but also doens't harm them.
1872 *
1873 * If we're asking for any reserved port (i.e. port == 0 &&
1874 * transport->xprt.resvport == 1) xs_get_srcport above will
1875 * ensure that port is non-zero and we will bind as needed.
1876 */
1877 if (port == 0)
1878 return 0;
1879
1880 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1881 do {
1882 rpc_set_port((struct sockaddr *)&myaddr, port);
1883 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1884 transport->xprt.addrlen);
1885 if (err == 0) {
1886 transport->srcport = port;
1887 break;
1888 }
1889 last = port;
1890 port = xs_next_srcport(transport, port);
1891 if (port > last)
1892 nloop++;
1893 } while (err == -EADDRINUSE && nloop != 2);
1894
1895 if (myaddr.ss_family == AF_INET)
1896 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1897 &((struct sockaddr_in *)&myaddr)->sin_addr,
1898 port, err ? "failed" : "ok", err);
1899 else
1900 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1901 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1902 port, err ? "failed" : "ok", err);
1903 return err;
1904}
1905
1906/*
1907 * We don't support autobind on AF_LOCAL sockets
1908 */
1909static void xs_local_rpcbind(struct rpc_task *task)
1910{
1911 xprt_set_bound(task->tk_xprt);
1912}
1913
1914static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1915{
1916}
1917
1918#ifdef CONFIG_DEBUG_LOCK_ALLOC
1919static struct lock_class_key xs_key[2];
1920static struct lock_class_key xs_slock_key[2];
1921
1922static inline void xs_reclassify_socketu(struct socket *sock)
1923{
1924 struct sock *sk = sock->sk;
1925
1926 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1927 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1928}
1929
1930static inline void xs_reclassify_socket4(struct socket *sock)
1931{
1932 struct sock *sk = sock->sk;
1933
1934 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1935 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1936}
1937
1938static inline void xs_reclassify_socket6(struct socket *sock)
1939{
1940 struct sock *sk = sock->sk;
1941
1942 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1943 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1944}
1945
1946static inline void xs_reclassify_socket(int family, struct socket *sock)
1947{
1948 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1949 return;
1950
1951 switch (family) {
1952 case AF_LOCAL:
1953 xs_reclassify_socketu(sock);
1954 break;
1955 case AF_INET:
1956 xs_reclassify_socket4(sock);
1957 break;
1958 case AF_INET6:
1959 xs_reclassify_socket6(sock);
1960 break;
1961 }
1962}
1963#else
1964static inline void xs_reclassify_socket(int family, struct socket *sock)
1965{
1966}
1967#endif
1968
1969static void xs_dummy_setup_socket(struct work_struct *work)
1970{
1971}
1972
1973static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1974 struct sock_xprt *transport, int family, int type,
1975 int protocol, bool reuseport)
1976{
1977 struct socket *sock;
1978 int err;
1979
1980 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1981 if (err < 0) {
1982 dprintk("RPC: can't create %d transport socket (%d).\n",
1983 protocol, -err);
1984 goto out;
1985 }
1986 xs_reclassify_socket(family, sock);
1987
1988 if (reuseport)
1989 xs_sock_set_reuseport(sock);
1990
1991 err = xs_bind(transport, sock);
1992 if (err) {
1993 sock_release(sock);
1994 goto out;
1995 }
1996
1997 return sock;
1998out:
1999 return ERR_PTR(err);
2000}
2001
2002static int xs_local_finish_connecting(struct rpc_xprt *xprt,
2003 struct socket *sock)
2004{
2005 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2006 xprt);
2007
2008 if (!transport->inet) {
2009 struct sock *sk = sock->sk;
2010
2011 write_lock_bh(&sk->sk_callback_lock);
2012
2013 xs_save_old_callbacks(transport, sk);
2014
2015 sk->sk_user_data = xprt;
2016 sk->sk_data_ready = xs_data_ready;
2017 sk->sk_write_space = xs_udp_write_space;
2018 sock_set_flag(sk, SOCK_FASYNC);
2019 sk->sk_error_report = xs_error_report;
2020 sk->sk_allocation = GFP_NOIO;
2021
2022 xprt_clear_connected(xprt);
2023
2024 /* Reset to new socket */
2025 transport->sock = sock;
2026 transport->inet = sk;
2027
2028 write_unlock_bh(&sk->sk_callback_lock);
2029 }
2030
2031 /* Tell the socket layer to start connecting... */
2032 xprt->stat.connect_count++;
2033 xprt->stat.connect_start = jiffies;
2034 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
2035}
2036
2037/**
2038 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
2039 * @transport: socket transport to connect
2040 */
2041static int xs_local_setup_socket(struct sock_xprt *transport)
2042{
2043 struct rpc_xprt *xprt = &transport->xprt;
2044 struct socket *sock;
2045 int status = -EIO;
2046
2047 status = __sock_create(xprt->xprt_net, AF_LOCAL,
2048 SOCK_STREAM, 0, &sock, 1);
2049 if (status < 0) {
2050 dprintk("RPC: can't create AF_LOCAL "
2051 "transport socket (%d).\n", -status);
2052 goto out;
2053 }
2054 xs_reclassify_socket(AF_LOCAL, sock);
2055
2056 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
2057 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2058
2059 status = xs_local_finish_connecting(xprt, sock);
2060 trace_rpc_socket_connect(xprt, sock, status);
2061 switch (status) {
2062 case 0:
2063 dprintk("RPC: xprt %p connected to %s\n",
2064 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2065 xprt_set_connected(xprt);
2066 case -ENOBUFS:
2067 break;
2068 case -ENOENT:
2069 dprintk("RPC: xprt %p: socket %s does not exist\n",
2070 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2071 break;
2072 case -ECONNREFUSED:
2073 dprintk("RPC: xprt %p: connection refused for %s\n",
2074 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2075 break;
2076 default:
2077 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2078 __func__, -status,
2079 xprt->address_strings[RPC_DISPLAY_ADDR]);
2080 }
2081
2082out:
2083 xprt_clear_connecting(xprt);
2084 xprt_wake_pending_tasks(xprt, status);
2085 return status;
2086}
2087
2088static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2089{
2090 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2091 int ret;
2092
2093 if (RPC_IS_ASYNC(task)) {
2094 /*
2095 * We want the AF_LOCAL connect to be resolved in the
2096 * filesystem namespace of the process making the rpc
2097 * call. Thus we connect synchronously.
2098 *
2099 * If we want to support asynchronous AF_LOCAL calls,
2100 * we'll need to figure out how to pass a namespace to
2101 * connect.
2102 */
2103 rpc_exit(task, -ENOTCONN);
2104 return;
2105 }
2106 ret = xs_local_setup_socket(transport);
2107 if (ret && !RPC_IS_SOFTCONN(task))
2108 msleep_interruptible(15000);
2109}
2110
2111#if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2112/*
2113 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2114 * know that we have exclusive access to the socket), to guard against
2115 * races with xs_reset_transport.
2116 */
2117static void xs_set_memalloc(struct rpc_xprt *xprt)
2118{
2119 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2120 xprt);
2121
2122 /*
2123 * If there's no sock, then we have nothing to set. The
2124 * reconnecting process will get it for us.
2125 */
2126 if (!transport->inet)
2127 return;
2128 if (atomic_read(&xprt->swapper))
2129 sk_set_memalloc(transport->inet);
2130}
2131
2132/**
2133 * xs_enable_swap - Tag this transport as being used for swap.
2134 * @xprt: transport to tag
2135 *
2136 * Take a reference to this transport on behalf of the rpc_clnt, and
2137 * optionally mark it for swapping if it wasn't already.
2138 */
2139static int
2140xs_enable_swap(struct rpc_xprt *xprt)
2141{
2142 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2143
2144 if (atomic_inc_return(&xprt->swapper) != 1)
2145 return 0;
2146 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2147 return -ERESTARTSYS;
2148 if (xs->inet)
2149 sk_set_memalloc(xs->inet);
2150 xprt_release_xprt(xprt, NULL);
2151 return 0;
2152}
2153
2154/**
2155 * xs_disable_swap - Untag this transport as being used for swap.
2156 * @xprt: transport to tag
2157 *
2158 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2159 * swapper refcount goes to 0, untag the socket as a memalloc socket.
2160 */
2161static void
2162xs_disable_swap(struct rpc_xprt *xprt)
2163{
2164 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2165
2166 if (!atomic_dec_and_test(&xprt->swapper))
2167 return;
2168 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2169 return;
2170 if (xs->inet)
2171 sk_clear_memalloc(xs->inet);
2172 xprt_release_xprt(xprt, NULL);
2173}
2174#else
2175static void xs_set_memalloc(struct rpc_xprt *xprt)
2176{
2177}
2178
2179static int
2180xs_enable_swap(struct rpc_xprt *xprt)
2181{
2182 return -EINVAL;
2183}
2184
2185static void
2186xs_disable_swap(struct rpc_xprt *xprt)
2187{
2188}
2189#endif
2190
2191static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2192{
2193 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2194
2195 if (!transport->inet) {
2196 struct sock *sk = sock->sk;
2197
2198 write_lock_bh(&sk->sk_callback_lock);
2199
2200 xs_save_old_callbacks(transport, sk);
2201
2202 sk->sk_user_data = xprt;
2203 sk->sk_data_ready = xs_data_ready;
2204 sk->sk_write_space = xs_udp_write_space;
2205 sock_set_flag(sk, SOCK_FASYNC);
2206 sk->sk_allocation = GFP_NOIO;
2207
2208 xprt_set_connected(xprt);
2209
2210 /* Reset to new socket */
2211 transport->sock = sock;
2212 transport->inet = sk;
2213
2214 xs_set_memalloc(xprt);
2215
2216 write_unlock_bh(&sk->sk_callback_lock);
2217 }
2218 xs_udp_do_set_buffer_size(xprt);
2219
2220 xprt->stat.connect_start = jiffies;
2221}
2222
2223static void xs_udp_setup_socket(struct work_struct *work)
2224{
2225 struct sock_xprt *transport =
2226 container_of(work, struct sock_xprt, connect_worker.work);
2227 struct rpc_xprt *xprt = &transport->xprt;
2228 struct socket *sock;
2229 int status = -EIO;
2230
2231 sock = xs_create_sock(xprt, transport,
2232 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2233 IPPROTO_UDP, false);
2234 if (IS_ERR(sock))
2235 goto out;
2236
2237 dprintk("RPC: worker connecting xprt %p via %s to "
2238 "%s (port %s)\n", xprt,
2239 xprt->address_strings[RPC_DISPLAY_PROTO],
2240 xprt->address_strings[RPC_DISPLAY_ADDR],
2241 xprt->address_strings[RPC_DISPLAY_PORT]);
2242
2243 xs_udp_finish_connecting(xprt, sock);
2244 trace_rpc_socket_connect(xprt, sock, 0);
2245 status = 0;
2246out:
2247 xprt_unlock_connect(xprt, transport);
2248 xprt_clear_connecting(xprt);
2249 xprt_wake_pending_tasks(xprt, status);
2250}
2251
2252/**
2253 * xs_tcp_shutdown - gracefully shut down a TCP socket
2254 * @xprt: transport
2255 *
2256 * Initiates a graceful shutdown of the TCP socket by calling the
2257 * equivalent of shutdown(SHUT_RDWR);
2258 */
2259static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2260{
2261 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2262 struct socket *sock = transport->sock;
2263 int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2264
2265 if (sock == NULL)
2266 return;
2267 switch (skst) {
2268 default:
2269 kernel_sock_shutdown(sock, SHUT_RDWR);
2270 trace_rpc_socket_shutdown(xprt, sock);
2271 break;
2272 case TCP_CLOSE:
2273 case TCP_TIME_WAIT:
2274 xs_reset_transport(transport);
2275 }
2276}
2277
2278static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2279 struct socket *sock)
2280{
2281 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2282 unsigned int keepidle;
2283 unsigned int keepcnt;
2284 unsigned int opt_on = 1;
2285 unsigned int timeo;
2286
2287 spin_lock_bh(&xprt->transport_lock);
2288 keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2289 keepcnt = xprt->timeout->to_retries + 1;
2290 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2291 (xprt->timeout->to_retries + 1);
2292 clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2293 spin_unlock_bh(&xprt->transport_lock);
2294
2295 /* TCP Keepalive options */
2296 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2297 (char *)&opt_on, sizeof(opt_on));
2298 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2299 (char *)&keepidle, sizeof(keepidle));
2300 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2301 (char *)&keepidle, sizeof(keepidle));
2302 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2303 (char *)&keepcnt, sizeof(keepcnt));
2304
2305 /* TCP user timeout (see RFC5482) */
2306 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2307 (char *)&timeo, sizeof(timeo));
2308}
2309
2310static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2311 unsigned long connect_timeout,
2312 unsigned long reconnect_timeout)
2313{
2314 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2315 struct rpc_timeout to;
2316 unsigned long initval;
2317
2318 spin_lock_bh(&xprt->transport_lock);
2319 if (reconnect_timeout < xprt->max_reconnect_timeout)
2320 xprt->max_reconnect_timeout = reconnect_timeout;
2321 if (connect_timeout < xprt->connect_timeout) {
2322 memcpy(&to, xprt->timeout, sizeof(to));
2323 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2324 /* Arbitrary lower limit */
2325 if (initval < XS_TCP_INIT_REEST_TO << 1)
2326 initval = XS_TCP_INIT_REEST_TO << 1;
2327 to.to_initval = initval;
2328 to.to_maxval = initval;
2329 memcpy(&transport->tcp_timeout, &to,
2330 sizeof(transport->tcp_timeout));
2331 xprt->timeout = &transport->tcp_timeout;
2332 xprt->connect_timeout = connect_timeout;
2333 }
2334 set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2335 spin_unlock_bh(&xprt->transport_lock);
2336}
2337
2338static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2339{
2340 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2341 int ret = -ENOTCONN;
2342
2343 if (!transport->inet) {
2344 struct sock *sk = sock->sk;
2345 unsigned int addr_pref = IPV6_PREFER_SRC_PUBLIC;
2346
2347 /* Avoid temporary address, they are bad for long-lived
2348 * connections such as NFS mounts.
2349 * RFC4941, section 3.6 suggests that:
2350 * Individual applications, which have specific
2351 * knowledge about the normal duration of connections,
2352 * MAY override this as appropriate.
2353 */
2354 kernel_setsockopt(sock, SOL_IPV6, IPV6_ADDR_PREFERENCES,
2355 (char *)&addr_pref, sizeof(addr_pref));
2356
2357 xs_tcp_set_socket_timeouts(xprt, sock);
2358
2359 write_lock_bh(&sk->sk_callback_lock);
2360
2361 xs_save_old_callbacks(transport, sk);
2362
2363 sk->sk_user_data = xprt;
2364 sk->sk_data_ready = xs_data_ready;
2365 sk->sk_state_change = xs_tcp_state_change;
2366 sk->sk_write_space = xs_tcp_write_space;
2367 sock_set_flag(sk, SOCK_FASYNC);
2368 sk->sk_error_report = xs_error_report;
2369 sk->sk_allocation = GFP_NOIO;
2370
2371 /* socket options */
2372 sock_reset_flag(sk, SOCK_LINGER);
2373 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2374
2375 xprt_clear_connected(xprt);
2376
2377 /* Reset to new socket */
2378 transport->sock = sock;
2379 transport->inet = sk;
2380
2381 write_unlock_bh(&sk->sk_callback_lock);
2382 }
2383
2384 if (!xprt_bound(xprt))
2385 goto out;
2386
2387 xs_set_memalloc(xprt);
2388
2389 /* Tell the socket layer to start connecting... */
2390 xprt->stat.connect_count++;
2391 xprt->stat.connect_start = jiffies;
2392 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2393 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2394 switch (ret) {
2395 case 0:
2396 xs_set_srcport(transport, sock);
2397 /* fall through */
2398 case -EINPROGRESS:
2399 /* SYN_SENT! */
2400 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2401 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2402 break;
2403 case -EADDRNOTAVAIL:
2404 /* Source port number is unavailable. Try a new one! */
2405 transport->srcport = 0;
2406 }
2407out:
2408 return ret;
2409}
2410
2411/**
2412 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2413 *
2414 * Invoked by a work queue tasklet.
2415 */
2416static void xs_tcp_setup_socket(struct work_struct *work)
2417{
2418 struct sock_xprt *transport =
2419 container_of(work, struct sock_xprt, connect_worker.work);
2420 struct socket *sock = transport->sock;
2421 struct rpc_xprt *xprt = &transport->xprt;
2422 int status = -EIO;
2423
2424 if (!sock) {
2425 sock = xs_create_sock(xprt, transport,
2426 xs_addr(xprt)->sa_family, SOCK_STREAM,
2427 IPPROTO_TCP, true);
2428 if (IS_ERR(sock)) {
2429 status = PTR_ERR(sock);
2430 goto out;
2431 }
2432 }
2433
2434 dprintk("RPC: worker connecting xprt %p via %s to "
2435 "%s (port %s)\n", xprt,
2436 xprt->address_strings[RPC_DISPLAY_PROTO],
2437 xprt->address_strings[RPC_DISPLAY_ADDR],
2438 xprt->address_strings[RPC_DISPLAY_PORT]);
2439
2440 status = xs_tcp_finish_connecting(xprt, sock);
2441 trace_rpc_socket_connect(xprt, sock, status);
2442 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2443 xprt, -status, xprt_connected(xprt),
2444 sock->sk->sk_state);
2445 switch (status) {
2446 default:
2447 printk("%s: connect returned unhandled error %d\n",
2448 __func__, status);
2449 /* fall through */
2450 case -EADDRNOTAVAIL:
2451 /* We're probably in TIME_WAIT. Get rid of existing socket,
2452 * and retry
2453 */
2454 xs_tcp_force_close(xprt);
2455 break;
2456 case 0:
2457 case -EINPROGRESS:
2458 case -EALREADY:
2459 xprt_unlock_connect(xprt, transport);
2460 return;
2461 case -EINVAL:
2462 /* Happens, for instance, if the user specified a link
2463 * local IPv6 address without a scope-id.
2464 */
2465 case -ECONNREFUSED:
2466 case -ECONNRESET:
2467 case -ENETDOWN:
2468 case -ENETUNREACH:
2469 case -EHOSTUNREACH:
2470 case -EADDRINUSE:
2471 case -ENOBUFS:
2472 /*
2473 * xs_tcp_force_close() wakes tasks with -EIO.
2474 * We need to wake them first to ensure the
2475 * correct error code.
2476 */
2477 xprt_wake_pending_tasks(xprt, status);
2478 xs_tcp_force_close(xprt);
2479 goto out;
2480 }
2481 status = -EAGAIN;
2482out:
2483 xprt_unlock_connect(xprt, transport);
2484 xprt_clear_connecting(xprt);
2485 xprt_wake_pending_tasks(xprt, status);
2486}
2487
2488static unsigned long xs_reconnect_delay(const struct rpc_xprt *xprt)
2489{
2490 unsigned long start, now = jiffies;
2491
2492 start = xprt->stat.connect_start + xprt->reestablish_timeout;
2493 if (time_after(start, now))
2494 return start - now;
2495 return 0;
2496}
2497
2498static void xs_reconnect_backoff(struct rpc_xprt *xprt)
2499{
2500 xprt->reestablish_timeout <<= 1;
2501 if (xprt->reestablish_timeout > xprt->max_reconnect_timeout)
2502 xprt->reestablish_timeout = xprt->max_reconnect_timeout;
2503 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2504 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2505}
2506
2507/**
2508 * xs_connect - connect a socket to a remote endpoint
2509 * @xprt: pointer to transport structure
2510 * @task: address of RPC task that manages state of connect request
2511 *
2512 * TCP: If the remote end dropped the connection, delay reconnecting.
2513 *
2514 * UDP socket connects are synchronous, but we use a work queue anyway
2515 * to guarantee that even unprivileged user processes can set up a
2516 * socket on a privileged port.
2517 *
2518 * If a UDP socket connect fails, the delay behavior here prevents
2519 * retry floods (hard mounts).
2520 */
2521static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2522{
2523 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2524 unsigned long delay = 0;
2525
2526 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2527
2528 if (transport->sock != NULL) {
2529 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2530 "seconds\n",
2531 xprt, xprt->reestablish_timeout / HZ);
2532
2533 /* Start by resetting any existing state */
2534 xs_reset_transport(transport);
2535
2536 delay = xs_reconnect_delay(xprt);
2537 xs_reconnect_backoff(xprt);
2538
2539 } else
2540 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2541
2542 queue_delayed_work(xprtiod_workqueue,
2543 &transport->connect_worker,
2544 delay);
2545}
2546
2547/**
2548 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2549 * @xprt: rpc_xprt struct containing statistics
2550 * @seq: output file
2551 *
2552 */
2553static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2554{
2555 long idle_time = 0;
2556
2557 if (xprt_connected(xprt))
2558 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2559
2560 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2561 "%llu %llu %lu %llu %llu\n",
2562 xprt->stat.bind_count,
2563 xprt->stat.connect_count,
2564 xprt->stat.connect_time,
2565 idle_time,
2566 xprt->stat.sends,
2567 xprt->stat.recvs,
2568 xprt->stat.bad_xids,
2569 xprt->stat.req_u,
2570 xprt->stat.bklog_u,
2571 xprt->stat.max_slots,
2572 xprt->stat.sending_u,
2573 xprt->stat.pending_u);
2574}
2575
2576/**
2577 * xs_udp_print_stats - display UDP socket-specifc stats
2578 * @xprt: rpc_xprt struct containing statistics
2579 * @seq: output file
2580 *
2581 */
2582static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2583{
2584 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2585
2586 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2587 "%lu %llu %llu\n",
2588 transport->srcport,
2589 xprt->stat.bind_count,
2590 xprt->stat.sends,
2591 xprt->stat.recvs,
2592 xprt->stat.bad_xids,
2593 xprt->stat.req_u,
2594 xprt->stat.bklog_u,
2595 xprt->stat.max_slots,
2596 xprt->stat.sending_u,
2597 xprt->stat.pending_u);
2598}
2599
2600/**
2601 * xs_tcp_print_stats - display TCP socket-specifc stats
2602 * @xprt: rpc_xprt struct containing statistics
2603 * @seq: output file
2604 *
2605 */
2606static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2607{
2608 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2609 long idle_time = 0;
2610
2611 if (xprt_connected(xprt))
2612 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2613
2614 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2615 "%llu %llu %lu %llu %llu\n",
2616 transport->srcport,
2617 xprt->stat.bind_count,
2618 xprt->stat.connect_count,
2619 xprt->stat.connect_time,
2620 idle_time,
2621 xprt->stat.sends,
2622 xprt->stat.recvs,
2623 xprt->stat.bad_xids,
2624 xprt->stat.req_u,
2625 xprt->stat.bklog_u,
2626 xprt->stat.max_slots,
2627 xprt->stat.sending_u,
2628 xprt->stat.pending_u);
2629}
2630
2631/*
2632 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2633 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2634 * to use the server side send routines.
2635 */
2636static int bc_malloc(struct rpc_task *task)
2637{
2638 struct rpc_rqst *rqst = task->tk_rqstp;
2639 size_t size = rqst->rq_callsize;
2640 struct page *page;
2641 struct rpc_buffer *buf;
2642
2643 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2644 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2645 size);
2646 return -EINVAL;
2647 }
2648
2649 page = alloc_page(GFP_KERNEL);
2650 if (!page)
2651 return -ENOMEM;
2652
2653 buf = page_address(page);
2654 buf->len = PAGE_SIZE;
2655
2656 rqst->rq_buffer = buf->data;
2657 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2658 return 0;
2659}
2660
2661/*
2662 * Free the space allocated in the bc_alloc routine
2663 */
2664static void bc_free(struct rpc_task *task)
2665{
2666 void *buffer = task->tk_rqstp->rq_buffer;
2667 struct rpc_buffer *buf;
2668
2669 buf = container_of(buffer, struct rpc_buffer, data);
2670 free_page((unsigned long)buf);
2671}
2672
2673/*
2674 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2675 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2676 */
2677static int bc_sendto(struct rpc_rqst *req)
2678{
2679 int len;
2680 struct xdr_buf *xbufp = &req->rq_snd_buf;
2681 struct rpc_xprt *xprt = req->rq_xprt;
2682 struct sock_xprt *transport =
2683 container_of(xprt, struct sock_xprt, xprt);
2684 struct socket *sock = transport->sock;
2685 unsigned long headoff;
2686 unsigned long tailoff;
2687
2688 xs_encode_stream_record_marker(xbufp);
2689
2690 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2691 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2692 len = svc_send_common(sock, xbufp,
2693 virt_to_page(xbufp->head[0].iov_base), headoff,
2694 xbufp->tail[0].iov_base, tailoff);
2695
2696 if (len != xbufp->len) {
2697 printk(KERN_NOTICE "Error sending entire callback!\n");
2698 len = -EAGAIN;
2699 }
2700
2701 return len;
2702}
2703
2704/*
2705 * The send routine. Borrows from svc_send
2706 */
2707static int bc_send_request(struct rpc_task *task)
2708{
2709 struct rpc_rqst *req = task->tk_rqstp;
2710 struct svc_xprt *xprt;
2711 int len;
2712
2713 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2714 /*
2715 * Get the server socket associated with this callback xprt
2716 */
2717 xprt = req->rq_xprt->bc_xprt;
2718
2719 /*
2720 * Grab the mutex to serialize data as the connection is shared
2721 * with the fore channel
2722 */
2723 if (!mutex_trylock(&xprt->xpt_mutex)) {
2724 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2725 if (!mutex_trylock(&xprt->xpt_mutex))
2726 return -EAGAIN;
2727 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2728 }
2729 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2730 len = -ENOTCONN;
2731 else
2732 len = bc_sendto(req);
2733 mutex_unlock(&xprt->xpt_mutex);
2734
2735 if (len > 0)
2736 len = 0;
2737
2738 return len;
2739}
2740
2741/*
2742 * The close routine. Since this is client initiated, we do nothing
2743 */
2744
2745static void bc_close(struct rpc_xprt *xprt)
2746{
2747}
2748
2749/*
2750 * The xprt destroy routine. Again, because this connection is client
2751 * initiated, we do nothing
2752 */
2753
2754static void bc_destroy(struct rpc_xprt *xprt)
2755{
2756 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2757
2758 xs_xprt_free(xprt);
2759 module_put(THIS_MODULE);
2760}
2761
2762static const struct rpc_xprt_ops xs_local_ops = {
2763 .reserve_xprt = xprt_reserve_xprt,
2764 .release_xprt = xs_tcp_release_xprt,
2765 .alloc_slot = xprt_alloc_slot,
2766 .rpcbind = xs_local_rpcbind,
2767 .set_port = xs_local_set_port,
2768 .connect = xs_local_connect,
2769 .buf_alloc = rpc_malloc,
2770 .buf_free = rpc_free,
2771 .send_request = xs_local_send_request,
2772 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2773 .close = xs_close,
2774 .destroy = xs_destroy,
2775 .print_stats = xs_local_print_stats,
2776 .enable_swap = xs_enable_swap,
2777 .disable_swap = xs_disable_swap,
2778};
2779
2780static const struct rpc_xprt_ops xs_udp_ops = {
2781 .set_buffer_size = xs_udp_set_buffer_size,
2782 .reserve_xprt = xprt_reserve_xprt_cong,
2783 .release_xprt = xprt_release_xprt_cong,
2784 .alloc_slot = xprt_alloc_slot,
2785 .rpcbind = rpcb_getport_async,
2786 .set_port = xs_set_port,
2787 .connect = xs_connect,
2788 .buf_alloc = rpc_malloc,
2789 .buf_free = rpc_free,
2790 .send_request = xs_udp_send_request,
2791 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
2792 .timer = xs_udp_timer,
2793 .release_request = xprt_release_rqst_cong,
2794 .close = xs_close,
2795 .destroy = xs_destroy,
2796 .print_stats = xs_udp_print_stats,
2797 .enable_swap = xs_enable_swap,
2798 .disable_swap = xs_disable_swap,
2799 .inject_disconnect = xs_inject_disconnect,
2800};
2801
2802static const struct rpc_xprt_ops xs_tcp_ops = {
2803 .reserve_xprt = xprt_reserve_xprt,
2804 .release_xprt = xs_tcp_release_xprt,
2805 .alloc_slot = xprt_lock_and_alloc_slot,
2806 .rpcbind = rpcb_getport_async,
2807 .set_port = xs_set_port,
2808 .connect = xs_connect,
2809 .buf_alloc = rpc_malloc,
2810 .buf_free = rpc_free,
2811 .send_request = xs_tcp_send_request,
2812 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2813 .close = xs_tcp_shutdown,
2814 .destroy = xs_destroy,
2815 .set_connect_timeout = xs_tcp_set_connect_timeout,
2816 .print_stats = xs_tcp_print_stats,
2817 .enable_swap = xs_enable_swap,
2818 .disable_swap = xs_disable_swap,
2819 .inject_disconnect = xs_inject_disconnect,
2820#ifdef CONFIG_SUNRPC_BACKCHANNEL
2821 .bc_setup = xprt_setup_bc,
2822 .bc_up = xs_tcp_bc_up,
2823 .bc_maxpayload = xs_tcp_bc_maxpayload,
2824 .bc_free_rqst = xprt_free_bc_rqst,
2825 .bc_destroy = xprt_destroy_bc,
2826#endif
2827};
2828
2829/*
2830 * The rpc_xprt_ops for the server backchannel
2831 */
2832
2833static const struct rpc_xprt_ops bc_tcp_ops = {
2834 .reserve_xprt = xprt_reserve_xprt,
2835 .release_xprt = xprt_release_xprt,
2836 .alloc_slot = xprt_alloc_slot,
2837 .buf_alloc = bc_malloc,
2838 .buf_free = bc_free,
2839 .send_request = bc_send_request,
2840 .set_retrans_timeout = xprt_set_retrans_timeout_def,
2841 .close = bc_close,
2842 .destroy = bc_destroy,
2843 .print_stats = xs_tcp_print_stats,
2844 .enable_swap = xs_enable_swap,
2845 .disable_swap = xs_disable_swap,
2846 .inject_disconnect = xs_inject_disconnect,
2847};
2848
2849static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2850{
2851 static const struct sockaddr_in sin = {
2852 .sin_family = AF_INET,
2853 .sin_addr.s_addr = htonl(INADDR_ANY),
2854 };
2855 static const struct sockaddr_in6 sin6 = {
2856 .sin6_family = AF_INET6,
2857 .sin6_addr = IN6ADDR_ANY_INIT,
2858 };
2859
2860 switch (family) {
2861 case AF_LOCAL:
2862 break;
2863 case AF_INET:
2864 memcpy(sap, &sin, sizeof(sin));
2865 break;
2866 case AF_INET6:
2867 memcpy(sap, &sin6, sizeof(sin6));
2868 break;
2869 default:
2870 dprintk("RPC: %s: Bad address family\n", __func__);
2871 return -EAFNOSUPPORT;
2872 }
2873 return 0;
2874}
2875
2876static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2877 unsigned int slot_table_size,
2878 unsigned int max_slot_table_size)
2879{
2880 struct rpc_xprt *xprt;
2881 struct sock_xprt *new;
2882
2883 if (args->addrlen > sizeof(xprt->addr)) {
2884 dprintk("RPC: xs_setup_xprt: address too large\n");
2885 return ERR_PTR(-EBADF);
2886 }
2887
2888 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2889 max_slot_table_size);
2890 if (xprt == NULL) {
2891 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2892 "rpc_xprt\n");
2893 return ERR_PTR(-ENOMEM);
2894 }
2895
2896 new = container_of(xprt, struct sock_xprt, xprt);
2897 mutex_init(&new->recv_mutex);
2898 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2899 xprt->addrlen = args->addrlen;
2900 if (args->srcaddr)
2901 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2902 else {
2903 int err;
2904 err = xs_init_anyaddr(args->dstaddr->sa_family,
2905 (struct sockaddr *)&new->srcaddr);
2906 if (err != 0) {
2907 xprt_free(xprt);
2908 return ERR_PTR(err);
2909 }
2910 }
2911
2912 return xprt;
2913}
2914
2915static const struct rpc_timeout xs_local_default_timeout = {
2916 .to_initval = 10 * HZ,
2917 .to_maxval = 10 * HZ,
2918 .to_retries = 2,
2919};
2920
2921/**
2922 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2923 * @args: rpc transport creation arguments
2924 *
2925 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2926 */
2927static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2928{
2929 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2930 struct sock_xprt *transport;
2931 struct rpc_xprt *xprt;
2932 struct rpc_xprt *ret;
2933
2934 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2935 xprt_max_tcp_slot_table_entries);
2936 if (IS_ERR(xprt))
2937 return xprt;
2938 transport = container_of(xprt, struct sock_xprt, xprt);
2939
2940 xprt->prot = 0;
2941 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2942 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2943
2944 xprt->bind_timeout = XS_BIND_TO;
2945 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2946 xprt->idle_timeout = XS_IDLE_DISC_TO;
2947
2948 xprt->ops = &xs_local_ops;
2949 xprt->timeout = &xs_local_default_timeout;
2950
2951 INIT_WORK(&transport->recv_worker, xs_local_data_receive_workfn);
2952 INIT_DELAYED_WORK(&transport->connect_worker,
2953 xs_dummy_setup_socket);
2954
2955 switch (sun->sun_family) {
2956 case AF_LOCAL:
2957 if (sun->sun_path[0] != '/') {
2958 dprintk("RPC: bad AF_LOCAL address: %s\n",
2959 sun->sun_path);
2960 ret = ERR_PTR(-EINVAL);
2961 goto out_err;
2962 }
2963 xprt_set_bound(xprt);
2964 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2965 ret = ERR_PTR(xs_local_setup_socket(transport));
2966 if (ret)
2967 goto out_err;
2968 break;
2969 default:
2970 ret = ERR_PTR(-EAFNOSUPPORT);
2971 goto out_err;
2972 }
2973
2974 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2975 xprt->address_strings[RPC_DISPLAY_ADDR]);
2976
2977 if (try_module_get(THIS_MODULE))
2978 return xprt;
2979 ret = ERR_PTR(-EINVAL);
2980out_err:
2981 xs_xprt_free(xprt);
2982 return ret;
2983}
2984
2985static const struct rpc_timeout xs_udp_default_timeout = {
2986 .to_initval = 5 * HZ,
2987 .to_maxval = 30 * HZ,
2988 .to_increment = 5 * HZ,
2989 .to_retries = 5,
2990};
2991
2992/**
2993 * xs_setup_udp - Set up transport to use a UDP socket
2994 * @args: rpc transport creation arguments
2995 *
2996 */
2997static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2998{
2999 struct sockaddr *addr = args->dstaddr;
3000 struct rpc_xprt *xprt;
3001 struct sock_xprt *transport;
3002 struct rpc_xprt *ret;
3003
3004 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
3005 xprt_udp_slot_table_entries);
3006 if (IS_ERR(xprt))
3007 return xprt;
3008 transport = container_of(xprt, struct sock_xprt, xprt);
3009
3010 xprt->prot = IPPROTO_UDP;
3011 xprt->tsh_size = 0;
3012 /* XXX: header size can vary due to auth type, IPv6, etc. */
3013 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
3014
3015 xprt->bind_timeout = XS_BIND_TO;
3016 xprt->reestablish_timeout = XS_UDP_REEST_TO;
3017 xprt->idle_timeout = XS_IDLE_DISC_TO;
3018
3019 xprt->ops = &xs_udp_ops;
3020
3021 xprt->timeout = &xs_udp_default_timeout;
3022
3023 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
3024 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
3025
3026 switch (addr->sa_family) {
3027 case AF_INET:
3028 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3029 xprt_set_bound(xprt);
3030
3031 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
3032 break;
3033 case AF_INET6:
3034 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3035 xprt_set_bound(xprt);
3036
3037 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
3038 break;
3039 default:
3040 ret = ERR_PTR(-EAFNOSUPPORT);
3041 goto out_err;
3042 }
3043
3044 if (xprt_bound(xprt))
3045 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3046 xprt->address_strings[RPC_DISPLAY_ADDR],
3047 xprt->address_strings[RPC_DISPLAY_PORT],
3048 xprt->address_strings[RPC_DISPLAY_PROTO]);
3049 else
3050 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
3051 xprt->address_strings[RPC_DISPLAY_ADDR],
3052 xprt->address_strings[RPC_DISPLAY_PROTO]);
3053
3054 if (try_module_get(THIS_MODULE))
3055 return xprt;
3056 ret = ERR_PTR(-EINVAL);
3057out_err:
3058 xs_xprt_free(xprt);
3059 return ret;
3060}
3061
3062static const struct rpc_timeout xs_tcp_default_timeout = {
3063 .to_initval = 60 * HZ,
3064 .to_maxval = 60 * HZ,
3065 .to_retries = 2,
3066};
3067
3068/**
3069 * xs_setup_tcp - Set up transport to use a TCP socket
3070 * @args: rpc transport creation arguments
3071 *
3072 */
3073static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
3074{
3075 struct sockaddr *addr = args->dstaddr;
3076 struct rpc_xprt *xprt;
3077 struct sock_xprt *transport;
3078 struct rpc_xprt *ret;
3079 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
3080
3081 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
3082 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
3083
3084 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3085 max_slot_table_size);
3086 if (IS_ERR(xprt))
3087 return xprt;
3088 transport = container_of(xprt, struct sock_xprt, xprt);
3089
3090 xprt->prot = IPPROTO_TCP;
3091 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3092 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3093
3094 xprt->bind_timeout = XS_BIND_TO;
3095 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3096 xprt->idle_timeout = XS_IDLE_DISC_TO;
3097
3098 xprt->ops = &xs_tcp_ops;
3099 xprt->timeout = &xs_tcp_default_timeout;
3100
3101 xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3102 xprt->connect_timeout = xprt->timeout->to_initval *
3103 (xprt->timeout->to_retries + 1);
3104
3105 INIT_WORK(&transport->recv_worker, xs_tcp_data_receive_workfn);
3106 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
3107
3108 switch (addr->sa_family) {
3109 case AF_INET:
3110 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3111 xprt_set_bound(xprt);
3112
3113 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3114 break;
3115 case AF_INET6:
3116 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3117 xprt_set_bound(xprt);
3118
3119 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3120 break;
3121 default:
3122 ret = ERR_PTR(-EAFNOSUPPORT);
3123 goto out_err;
3124 }
3125
3126 if (xprt_bound(xprt))
3127 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3128 xprt->address_strings[RPC_DISPLAY_ADDR],
3129 xprt->address_strings[RPC_DISPLAY_PORT],
3130 xprt->address_strings[RPC_DISPLAY_PROTO]);
3131 else
3132 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
3133 xprt->address_strings[RPC_DISPLAY_ADDR],
3134 xprt->address_strings[RPC_DISPLAY_PROTO]);
3135
3136 if (try_module_get(THIS_MODULE))
3137 return xprt;
3138 ret = ERR_PTR(-EINVAL);
3139out_err:
3140 xs_xprt_free(xprt);
3141 return ret;
3142}
3143
3144/**
3145 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3146 * @args: rpc transport creation arguments
3147 *
3148 */
3149static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3150{
3151 struct sockaddr *addr = args->dstaddr;
3152 struct rpc_xprt *xprt;
3153 struct sock_xprt *transport;
3154 struct svc_sock *bc_sock;
3155 struct rpc_xprt *ret;
3156
3157 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3158 xprt_tcp_slot_table_entries);
3159 if (IS_ERR(xprt))
3160 return xprt;
3161 transport = container_of(xprt, struct sock_xprt, xprt);
3162
3163 xprt->prot = IPPROTO_TCP;
3164 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3165 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3166 xprt->timeout = &xs_tcp_default_timeout;
3167
3168 /* backchannel */
3169 xprt_set_bound(xprt);
3170 xprt->bind_timeout = 0;
3171 xprt->reestablish_timeout = 0;
3172 xprt->idle_timeout = 0;
3173
3174 xprt->ops = &bc_tcp_ops;
3175
3176 switch (addr->sa_family) {
3177 case AF_INET:
3178 xs_format_peer_addresses(xprt, "tcp",
3179 RPCBIND_NETID_TCP);
3180 break;
3181 case AF_INET6:
3182 xs_format_peer_addresses(xprt, "tcp",
3183 RPCBIND_NETID_TCP6);
3184 break;
3185 default:
3186 ret = ERR_PTR(-EAFNOSUPPORT);
3187 goto out_err;
3188 }
3189
3190 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3191 xprt->address_strings[RPC_DISPLAY_ADDR],
3192 xprt->address_strings[RPC_DISPLAY_PORT],
3193 xprt->address_strings[RPC_DISPLAY_PROTO]);
3194
3195 /*
3196 * Once we've associated a backchannel xprt with a connection,
3197 * we want to keep it around as long as the connection lasts,
3198 * in case we need to start using it for a backchannel again;
3199 * this reference won't be dropped until bc_xprt is destroyed.
3200 */
3201 xprt_get(xprt);
3202 args->bc_xprt->xpt_bc_xprt = xprt;
3203 xprt->bc_xprt = args->bc_xprt;
3204 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3205 transport->sock = bc_sock->sk_sock;
3206 transport->inet = bc_sock->sk_sk;
3207
3208 /*
3209 * Since we don't want connections for the backchannel, we set
3210 * the xprt status to connected
3211 */
3212 xprt_set_connected(xprt);
3213
3214 if (try_module_get(THIS_MODULE))
3215 return xprt;
3216
3217 args->bc_xprt->xpt_bc_xprt = NULL;
3218 args->bc_xprt->xpt_bc_xps = NULL;
3219 xprt_put(xprt);
3220 ret = ERR_PTR(-EINVAL);
3221out_err:
3222 xs_xprt_free(xprt);
3223 return ret;
3224}
3225
3226static struct xprt_class xs_local_transport = {
3227 .list = LIST_HEAD_INIT(xs_local_transport.list),
3228 .name = "named UNIX socket",
3229 .owner = THIS_MODULE,
3230 .ident = XPRT_TRANSPORT_LOCAL,
3231 .setup = xs_setup_local,
3232};
3233
3234static struct xprt_class xs_udp_transport = {
3235 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3236 .name = "udp",
3237 .owner = THIS_MODULE,
3238 .ident = XPRT_TRANSPORT_UDP,
3239 .setup = xs_setup_udp,
3240};
3241
3242static struct xprt_class xs_tcp_transport = {
3243 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3244 .name = "tcp",
3245 .owner = THIS_MODULE,
3246 .ident = XPRT_TRANSPORT_TCP,
3247 .setup = xs_setup_tcp,
3248};
3249
3250static struct xprt_class xs_bc_tcp_transport = {
3251 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3252 .name = "tcp NFSv4.1 backchannel",
3253 .owner = THIS_MODULE,
3254 .ident = XPRT_TRANSPORT_BC_TCP,
3255 .setup = xs_setup_bc_tcp,
3256};
3257
3258/**
3259 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3260 *
3261 */
3262int init_socket_xprt(void)
3263{
3264#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3265 if (!sunrpc_table_header)
3266 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3267#endif
3268
3269 xprt_register_transport(&xs_local_transport);
3270 xprt_register_transport(&xs_udp_transport);
3271 xprt_register_transport(&xs_tcp_transport);
3272 xprt_register_transport(&xs_bc_tcp_transport);
3273
3274 return 0;
3275}
3276
3277/**
3278 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3279 *
3280 */
3281void cleanup_socket_xprt(void)
3282{
3283#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3284 if (sunrpc_table_header) {
3285 unregister_sysctl_table(sunrpc_table_header);
3286 sunrpc_table_header = NULL;
3287 }
3288#endif
3289
3290 xprt_unregister_transport(&xs_local_transport);
3291 xprt_unregister_transport(&xs_udp_transport);
3292 xprt_unregister_transport(&xs_tcp_transport);
3293 xprt_unregister_transport(&xs_bc_tcp_transport);
3294}
3295
3296static int param_set_uint_minmax(const char *val,
3297 const struct kernel_param *kp,
3298 unsigned int min, unsigned int max)
3299{
3300 unsigned int num;
3301 int ret;
3302
3303 if (!val)
3304 return -EINVAL;
3305 ret = kstrtouint(val, 0, &num);
3306 if (ret)
3307 return ret;
3308 if (num < min || num > max)
3309 return -EINVAL;
3310 *((unsigned int *)kp->arg) = num;
3311 return 0;
3312}
3313
3314static int param_set_portnr(const char *val, const struct kernel_param *kp)
3315{
3316 if (kp->arg == &xprt_min_resvport)
3317 return param_set_uint_minmax(val, kp,
3318 RPC_MIN_RESVPORT,
3319 xprt_max_resvport);
3320 return param_set_uint_minmax(val, kp,
3321 xprt_min_resvport,
3322 RPC_MAX_RESVPORT);
3323}
3324
3325static const struct kernel_param_ops param_ops_portnr = {
3326 .set = param_set_portnr,
3327 .get = param_get_uint,
3328};
3329
3330#define param_check_portnr(name, p) \
3331 __param_check(name, p, unsigned int);
3332
3333module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3334module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3335
3336static int param_set_slot_table_size(const char *val,
3337 const struct kernel_param *kp)
3338{
3339 return param_set_uint_minmax(val, kp,
3340 RPC_MIN_SLOT_TABLE,
3341 RPC_MAX_SLOT_TABLE);
3342}
3343
3344static const struct kernel_param_ops param_ops_slot_table_size = {
3345 .set = param_set_slot_table_size,
3346 .get = param_get_uint,
3347};
3348
3349#define param_check_slot_table_size(name, p) \
3350 __param_check(name, p, unsigned int);
3351
3352static int param_set_max_slot_table_size(const char *val,
3353 const struct kernel_param *kp)
3354{
3355 return param_set_uint_minmax(val, kp,
3356 RPC_MIN_SLOT_TABLE,
3357 RPC_MAX_SLOT_TABLE_LIMIT);
3358}
3359
3360static const struct kernel_param_ops param_ops_max_slot_table_size = {
3361 .set = param_set_max_slot_table_size,
3362 .get = param_get_uint,
3363};
3364
3365#define param_check_max_slot_table_size(name, p) \
3366 __param_check(name, p, unsigned int);
3367
3368module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3369 slot_table_size, 0644);
3370module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3371 max_slot_table_size, 0644);
3372module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3373 slot_table_size, 0644);
3374