Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/net/sunrpc/xprtsock.c
4 *
5 * Client-side transport implementation for sockets.
6 *
7 * TCP callback races fixes (C) 1998 Red Hat
8 * TCP send fixes (C) 1998 Red Hat
9 * TCP NFS related read + write fixes
10 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
11 *
12 * Rewrite of larges part of the code in order to stabilize TCP stuff.
13 * Fix behaviour when socket buffer is full.
14 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
15 *
16 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
17 *
18 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
19 * <gilles.quillard@bull.net>
20 */
21
22#include <linux/types.h>
23#include <linux/string.h>
24#include <linux/slab.h>
25#include <linux/module.h>
26#include <linux/capability.h>
27#include <linux/pagemap.h>
28#include <linux/errno.h>
29#include <linux/socket.h>
30#include <linux/in.h>
31#include <linux/net.h>
32#include <linux/mm.h>
33#include <linux/un.h>
34#include <linux/udp.h>
35#include <linux/tcp.h>
36#include <linux/sunrpc/clnt.h>
37#include <linux/sunrpc/addr.h>
38#include <linux/sunrpc/sched.h>
39#include <linux/sunrpc/svcsock.h>
40#include <linux/sunrpc/xprtsock.h>
41#include <linux/file.h>
42#ifdef CONFIG_SUNRPC_BACKCHANNEL
43#include <linux/sunrpc/bc_xprt.h>
44#endif
45
46#include <net/sock.h>
47#include <net/checksum.h>
48#include <net/udp.h>
49#include <net/tcp.h>
50#include <linux/bvec.h>
51#include <linux/highmem.h>
52#include <linux/uio.h>
53#include <linux/sched/mm.h>
54
55#include <trace/events/sunrpc.h>
56
57#include "socklib.h"
58#include "sunrpc.h"
59
60static void xs_close(struct rpc_xprt *xprt);
61static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock);
62static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
63 struct socket *sock);
64
65/*
66 * xprtsock tunables
67 */
68static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
69static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
70static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
71
72static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
73static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
74
75#define XS_TCP_LINGER_TO (15U * HZ)
76static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
77
78/*
79 * We can register our own files under /proc/sys/sunrpc by
80 * calling register_sysctl_table() again. The files in that
81 * directory become the union of all files registered there.
82 *
83 * We simply need to make sure that we don't collide with
84 * someone else's file names!
85 */
86
87static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
88static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
89static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
90static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
91static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
92
93static struct ctl_table_header *sunrpc_table_header;
94
95static struct xprt_class xs_local_transport;
96static struct xprt_class xs_udp_transport;
97static struct xprt_class xs_tcp_transport;
98static struct xprt_class xs_bc_tcp_transport;
99
100/*
101 * FIXME: changing the UDP slot table size should also resize the UDP
102 * socket buffers for existing UDP transports
103 */
104static struct ctl_table xs_tunables_table[] = {
105 {
106 .procname = "udp_slot_table_entries",
107 .data = &xprt_udp_slot_table_entries,
108 .maxlen = sizeof(unsigned int),
109 .mode = 0644,
110 .proc_handler = proc_dointvec_minmax,
111 .extra1 = &min_slot_table_size,
112 .extra2 = &max_slot_table_size
113 },
114 {
115 .procname = "tcp_slot_table_entries",
116 .data = &xprt_tcp_slot_table_entries,
117 .maxlen = sizeof(unsigned int),
118 .mode = 0644,
119 .proc_handler = proc_dointvec_minmax,
120 .extra1 = &min_slot_table_size,
121 .extra2 = &max_slot_table_size
122 },
123 {
124 .procname = "tcp_max_slot_table_entries",
125 .data = &xprt_max_tcp_slot_table_entries,
126 .maxlen = sizeof(unsigned int),
127 .mode = 0644,
128 .proc_handler = proc_dointvec_minmax,
129 .extra1 = &min_slot_table_size,
130 .extra2 = &max_tcp_slot_table_limit
131 },
132 {
133 .procname = "min_resvport",
134 .data = &xprt_min_resvport,
135 .maxlen = sizeof(unsigned int),
136 .mode = 0644,
137 .proc_handler = proc_dointvec_minmax,
138 .extra1 = &xprt_min_resvport_limit,
139 .extra2 = &xprt_max_resvport_limit
140 },
141 {
142 .procname = "max_resvport",
143 .data = &xprt_max_resvport,
144 .maxlen = sizeof(unsigned int),
145 .mode = 0644,
146 .proc_handler = proc_dointvec_minmax,
147 .extra1 = &xprt_min_resvport_limit,
148 .extra2 = &xprt_max_resvport_limit
149 },
150 {
151 .procname = "tcp_fin_timeout",
152 .data = &xs_tcp_fin_timeout,
153 .maxlen = sizeof(xs_tcp_fin_timeout),
154 .mode = 0644,
155 .proc_handler = proc_dointvec_jiffies,
156 },
157 { },
158};
159
160static struct ctl_table sunrpc_table[] = {
161 {
162 .procname = "sunrpc",
163 .mode = 0555,
164 .child = xs_tunables_table
165 },
166 { },
167};
168
169/*
170 * Wait duration for a reply from the RPC portmapper.
171 */
172#define XS_BIND_TO (60U * HZ)
173
174/*
175 * Delay if a UDP socket connect error occurs. This is most likely some
176 * kind of resource problem on the local host.
177 */
178#define XS_UDP_REEST_TO (2U * HZ)
179
180/*
181 * The reestablish timeout allows clients to delay for a bit before attempting
182 * to reconnect to a server that just dropped our connection.
183 *
184 * We implement an exponential backoff when trying to reestablish a TCP
185 * transport connection with the server. Some servers like to drop a TCP
186 * connection when they are overworked, so we start with a short timeout and
187 * increase over time if the server is down or not responding.
188 */
189#define XS_TCP_INIT_REEST_TO (3U * HZ)
190
191/*
192 * TCP idle timeout; client drops the transport socket if it is idle
193 * for this long. Note that we also timeout UDP sockets to prevent
194 * holding port numbers when there is no RPC traffic.
195 */
196#define XS_IDLE_DISC_TO (5U * 60 * HZ)
197
198#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
199# undef RPC_DEBUG_DATA
200# define RPCDBG_FACILITY RPCDBG_TRANS
201#endif
202
203#ifdef RPC_DEBUG_DATA
204static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
205{
206 u8 *buf = (u8 *) packet;
207 int j;
208
209 dprintk("RPC: %s\n", msg);
210 for (j = 0; j < count && j < 128; j += 4) {
211 if (!(j & 31)) {
212 if (j)
213 dprintk("\n");
214 dprintk("0x%04x ", j);
215 }
216 dprintk("%02x%02x%02x%02x ",
217 buf[j], buf[j+1], buf[j+2], buf[j+3]);
218 }
219 dprintk("\n");
220}
221#else
222static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
223{
224 /* NOP */
225}
226#endif
227
228static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
229{
230 return (struct rpc_xprt *) sk->sk_user_data;
231}
232
233static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
234{
235 return (struct sockaddr *) &xprt->addr;
236}
237
238static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
239{
240 return (struct sockaddr_un *) &xprt->addr;
241}
242
243static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
244{
245 return (struct sockaddr_in *) &xprt->addr;
246}
247
248static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
249{
250 return (struct sockaddr_in6 *) &xprt->addr;
251}
252
253static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
254{
255 struct sockaddr *sap = xs_addr(xprt);
256 struct sockaddr_in6 *sin6;
257 struct sockaddr_in *sin;
258 struct sockaddr_un *sun;
259 char buf[128];
260
261 switch (sap->sa_family) {
262 case AF_LOCAL:
263 sun = xs_addr_un(xprt);
264 strscpy(buf, sun->sun_path, sizeof(buf));
265 xprt->address_strings[RPC_DISPLAY_ADDR] =
266 kstrdup(buf, GFP_KERNEL);
267 break;
268 case AF_INET:
269 (void)rpc_ntop(sap, buf, sizeof(buf));
270 xprt->address_strings[RPC_DISPLAY_ADDR] =
271 kstrdup(buf, GFP_KERNEL);
272 sin = xs_addr_in(xprt);
273 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
274 break;
275 case AF_INET6:
276 (void)rpc_ntop(sap, buf, sizeof(buf));
277 xprt->address_strings[RPC_DISPLAY_ADDR] =
278 kstrdup(buf, GFP_KERNEL);
279 sin6 = xs_addr_in6(xprt);
280 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
281 break;
282 default:
283 BUG();
284 }
285
286 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
287}
288
289static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
290{
291 struct sockaddr *sap = xs_addr(xprt);
292 char buf[128];
293
294 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
295 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
296
297 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
298 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
299}
300
301static void xs_format_peer_addresses(struct rpc_xprt *xprt,
302 const char *protocol,
303 const char *netid)
304{
305 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
306 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
307 xs_format_common_peer_addresses(xprt);
308 xs_format_common_peer_ports(xprt);
309}
310
311static void xs_update_peer_port(struct rpc_xprt *xprt)
312{
313 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
314 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
315
316 xs_format_common_peer_ports(xprt);
317}
318
319static void xs_free_peer_addresses(struct rpc_xprt *xprt)
320{
321 unsigned int i;
322
323 for (i = 0; i < RPC_DISPLAY_MAX; i++)
324 switch (i) {
325 case RPC_DISPLAY_PROTO:
326 case RPC_DISPLAY_NETID:
327 continue;
328 default:
329 kfree(xprt->address_strings[i]);
330 }
331}
332
333static size_t
334xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp)
335{
336 size_t i,n;
337
338 if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES))
339 return want;
340 n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT;
341 for (i = 0; i < n; i++) {
342 if (buf->pages[i])
343 continue;
344 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp);
345 if (!buf->pages[i]) {
346 i *= PAGE_SIZE;
347 return i > buf->page_base ? i - buf->page_base : 0;
348 }
349 }
350 return want;
351}
352
353static ssize_t
354xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek)
355{
356 ssize_t ret;
357 if (seek != 0)
358 iov_iter_advance(&msg->msg_iter, seek);
359 ret = sock_recvmsg(sock, msg, flags);
360 return ret > 0 ? ret + seek : ret;
361}
362
363static ssize_t
364xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags,
365 struct kvec *kvec, size_t count, size_t seek)
366{
367 iov_iter_kvec(&msg->msg_iter, ITER_DEST, kvec, 1, count);
368 return xs_sock_recvmsg(sock, msg, flags, seek);
369}
370
371static ssize_t
372xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags,
373 struct bio_vec *bvec, unsigned long nr, size_t count,
374 size_t seek)
375{
376 iov_iter_bvec(&msg->msg_iter, ITER_DEST, bvec, nr, count);
377 return xs_sock_recvmsg(sock, msg, flags, seek);
378}
379
380static ssize_t
381xs_read_discard(struct socket *sock, struct msghdr *msg, int flags,
382 size_t count)
383{
384 iov_iter_discard(&msg->msg_iter, ITER_DEST, count);
385 return sock_recvmsg(sock, msg, flags);
386}
387
388#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
389static void
390xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
391{
392 struct bvec_iter bi = {
393 .bi_size = count,
394 };
395 struct bio_vec bv;
396
397 bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
398 for_each_bvec(bv, bvec, bi, bi)
399 flush_dcache_page(bv.bv_page);
400}
401#else
402static inline void
403xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
404{
405}
406#endif
407
408static ssize_t
409xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
410 struct xdr_buf *buf, size_t count, size_t seek, size_t *read)
411{
412 size_t want, seek_init = seek, offset = 0;
413 ssize_t ret;
414
415 want = min_t(size_t, count, buf->head[0].iov_len);
416 if (seek < want) {
417 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek);
418 if (ret <= 0)
419 goto sock_err;
420 offset += ret;
421 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
422 goto out;
423 if (ret != want)
424 goto out;
425 seek = 0;
426 } else {
427 seek -= want;
428 offset += want;
429 }
430
431 want = xs_alloc_sparse_pages(
432 buf, min_t(size_t, count - offset, buf->page_len),
433 GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN);
434 if (seek < want) {
435 ret = xs_read_bvec(sock, msg, flags, buf->bvec,
436 xdr_buf_pagecount(buf),
437 want + buf->page_base,
438 seek + buf->page_base);
439 if (ret <= 0)
440 goto sock_err;
441 xs_flush_bvec(buf->bvec, ret, seek + buf->page_base);
442 ret -= buf->page_base;
443 offset += ret;
444 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
445 goto out;
446 if (ret != want)
447 goto out;
448 seek = 0;
449 } else {
450 seek -= want;
451 offset += want;
452 }
453
454 want = min_t(size_t, count - offset, buf->tail[0].iov_len);
455 if (seek < want) {
456 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek);
457 if (ret <= 0)
458 goto sock_err;
459 offset += ret;
460 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
461 goto out;
462 if (ret != want)
463 goto out;
464 } else if (offset < seek_init)
465 offset = seek_init;
466 ret = -EMSGSIZE;
467out:
468 *read = offset - seek_init;
469 return ret;
470sock_err:
471 offset += seek;
472 goto out;
473}
474
475static void
476xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf)
477{
478 if (!transport->recv.copied) {
479 if (buf->head[0].iov_len >= transport->recv.offset)
480 memcpy(buf->head[0].iov_base,
481 &transport->recv.xid,
482 transport->recv.offset);
483 transport->recv.copied = transport->recv.offset;
484 }
485}
486
487static bool
488xs_read_stream_request_done(struct sock_xprt *transport)
489{
490 return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT);
491}
492
493static void
494xs_read_stream_check_eor(struct sock_xprt *transport,
495 struct msghdr *msg)
496{
497 if (xs_read_stream_request_done(transport))
498 msg->msg_flags |= MSG_EOR;
499}
500
501static ssize_t
502xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg,
503 int flags, struct rpc_rqst *req)
504{
505 struct xdr_buf *buf = &req->rq_private_buf;
506 size_t want, read;
507 ssize_t ret;
508
509 xs_read_header(transport, buf);
510
511 want = transport->recv.len - transport->recv.offset;
512 if (want != 0) {
513 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf,
514 transport->recv.copied + want,
515 transport->recv.copied,
516 &read);
517 transport->recv.offset += read;
518 transport->recv.copied += read;
519 }
520
521 if (transport->recv.offset == transport->recv.len)
522 xs_read_stream_check_eor(transport, msg);
523
524 if (want == 0)
525 return 0;
526
527 switch (ret) {
528 default:
529 break;
530 case -EFAULT:
531 case -EMSGSIZE:
532 msg->msg_flags |= MSG_TRUNC;
533 return read;
534 case 0:
535 return -ESHUTDOWN;
536 }
537 return ret < 0 ? ret : read;
538}
539
540static size_t
541xs_read_stream_headersize(bool isfrag)
542{
543 if (isfrag)
544 return sizeof(__be32);
545 return 3 * sizeof(__be32);
546}
547
548static ssize_t
549xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg,
550 int flags, size_t want, size_t seek)
551{
552 struct kvec kvec = {
553 .iov_base = &transport->recv.fraghdr,
554 .iov_len = want,
555 };
556 return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek);
557}
558
559#if defined(CONFIG_SUNRPC_BACKCHANNEL)
560static ssize_t
561xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
562{
563 struct rpc_xprt *xprt = &transport->xprt;
564 struct rpc_rqst *req;
565 ssize_t ret;
566
567 /* Is this transport associated with the backchannel? */
568 if (!xprt->bc_serv)
569 return -ESHUTDOWN;
570
571 /* Look up and lock the request corresponding to the given XID */
572 req = xprt_lookup_bc_request(xprt, transport->recv.xid);
573 if (!req) {
574 printk(KERN_WARNING "Callback slot table overflowed\n");
575 return -ESHUTDOWN;
576 }
577 if (transport->recv.copied && !req->rq_private_buf.len)
578 return -ESHUTDOWN;
579
580 ret = xs_read_stream_request(transport, msg, flags, req);
581 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
582 xprt_complete_bc_request(req, transport->recv.copied);
583 else
584 req->rq_private_buf.len = transport->recv.copied;
585
586 return ret;
587}
588#else /* CONFIG_SUNRPC_BACKCHANNEL */
589static ssize_t
590xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
591{
592 return -ESHUTDOWN;
593}
594#endif /* CONFIG_SUNRPC_BACKCHANNEL */
595
596static ssize_t
597xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags)
598{
599 struct rpc_xprt *xprt = &transport->xprt;
600 struct rpc_rqst *req;
601 ssize_t ret = 0;
602
603 /* Look up and lock the request corresponding to the given XID */
604 spin_lock(&xprt->queue_lock);
605 req = xprt_lookup_rqst(xprt, transport->recv.xid);
606 if (!req || (transport->recv.copied && !req->rq_private_buf.len)) {
607 msg->msg_flags |= MSG_TRUNC;
608 goto out;
609 }
610 xprt_pin_rqst(req);
611 spin_unlock(&xprt->queue_lock);
612
613 ret = xs_read_stream_request(transport, msg, flags, req);
614
615 spin_lock(&xprt->queue_lock);
616 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
617 xprt_complete_rqst(req->rq_task, transport->recv.copied);
618 else
619 req->rq_private_buf.len = transport->recv.copied;
620 xprt_unpin_rqst(req);
621out:
622 spin_unlock(&xprt->queue_lock);
623 return ret;
624}
625
626static ssize_t
627xs_read_stream(struct sock_xprt *transport, int flags)
628{
629 struct msghdr msg = { 0 };
630 size_t want, read = 0;
631 ssize_t ret = 0;
632
633 if (transport->recv.len == 0) {
634 want = xs_read_stream_headersize(transport->recv.copied != 0);
635 ret = xs_read_stream_header(transport, &msg, flags, want,
636 transport->recv.offset);
637 if (ret <= 0)
638 goto out_err;
639 transport->recv.offset = ret;
640 if (transport->recv.offset != want)
641 return transport->recv.offset;
642 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) &
643 RPC_FRAGMENT_SIZE_MASK;
644 transport->recv.offset -= sizeof(transport->recv.fraghdr);
645 read = ret;
646 }
647
648 switch (be32_to_cpu(transport->recv.calldir)) {
649 default:
650 msg.msg_flags |= MSG_TRUNC;
651 break;
652 case RPC_CALL:
653 ret = xs_read_stream_call(transport, &msg, flags);
654 break;
655 case RPC_REPLY:
656 ret = xs_read_stream_reply(transport, &msg, flags);
657 }
658 if (msg.msg_flags & MSG_TRUNC) {
659 transport->recv.calldir = cpu_to_be32(-1);
660 transport->recv.copied = -1;
661 }
662 if (ret < 0)
663 goto out_err;
664 read += ret;
665 if (transport->recv.offset < transport->recv.len) {
666 if (!(msg.msg_flags & MSG_TRUNC))
667 return read;
668 msg.msg_flags = 0;
669 ret = xs_read_discard(transport->sock, &msg, flags,
670 transport->recv.len - transport->recv.offset);
671 if (ret <= 0)
672 goto out_err;
673 transport->recv.offset += ret;
674 read += ret;
675 if (transport->recv.offset != transport->recv.len)
676 return read;
677 }
678 if (xs_read_stream_request_done(transport)) {
679 trace_xs_stream_read_request(transport);
680 transport->recv.copied = 0;
681 }
682 transport->recv.offset = 0;
683 transport->recv.len = 0;
684 return read;
685out_err:
686 return ret != 0 ? ret : -ESHUTDOWN;
687}
688
689static __poll_t xs_poll_socket(struct sock_xprt *transport)
690{
691 return transport->sock->ops->poll(transport->file, transport->sock,
692 NULL);
693}
694
695static bool xs_poll_socket_readable(struct sock_xprt *transport)
696{
697 __poll_t events = xs_poll_socket(transport);
698
699 return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP);
700}
701
702static void xs_poll_check_readable(struct sock_xprt *transport)
703{
704
705 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
706 if (!xs_poll_socket_readable(transport))
707 return;
708 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
709 queue_work(xprtiod_workqueue, &transport->recv_worker);
710}
711
712static void xs_stream_data_receive(struct sock_xprt *transport)
713{
714 size_t read = 0;
715 ssize_t ret = 0;
716
717 mutex_lock(&transport->recv_mutex);
718 if (transport->sock == NULL)
719 goto out;
720 for (;;) {
721 ret = xs_read_stream(transport, MSG_DONTWAIT);
722 if (ret < 0)
723 break;
724 read += ret;
725 cond_resched();
726 }
727 if (ret == -ESHUTDOWN)
728 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
729 else
730 xs_poll_check_readable(transport);
731out:
732 mutex_unlock(&transport->recv_mutex);
733 trace_xs_stream_read_data(&transport->xprt, ret, read);
734}
735
736static void xs_stream_data_receive_workfn(struct work_struct *work)
737{
738 struct sock_xprt *transport =
739 container_of(work, struct sock_xprt, recv_worker);
740 unsigned int pflags = memalloc_nofs_save();
741
742 xs_stream_data_receive(transport);
743 memalloc_nofs_restore(pflags);
744}
745
746static void
747xs_stream_reset_connect(struct sock_xprt *transport)
748{
749 transport->recv.offset = 0;
750 transport->recv.len = 0;
751 transport->recv.copied = 0;
752 transport->xmit.offset = 0;
753}
754
755static void
756xs_stream_start_connect(struct sock_xprt *transport)
757{
758 transport->xprt.stat.connect_count++;
759 transport->xprt.stat.connect_start = jiffies;
760}
761
762#define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
763
764/**
765 * xs_nospace - handle transmit was incomplete
766 * @req: pointer to RPC request
767 * @transport: pointer to struct sock_xprt
768 *
769 */
770static int xs_nospace(struct rpc_rqst *req, struct sock_xprt *transport)
771{
772 struct rpc_xprt *xprt = &transport->xprt;
773 struct sock *sk = transport->inet;
774 int ret = -EAGAIN;
775
776 trace_rpc_socket_nospace(req, transport);
777
778 /* Protect against races with write_space */
779 spin_lock(&xprt->transport_lock);
780
781 /* Don't race with disconnect */
782 if (xprt_connected(xprt)) {
783 /* wait for more buffer space */
784 set_bit(XPRT_SOCK_NOSPACE, &transport->sock_state);
785 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
786 sk->sk_write_pending++;
787 xprt_wait_for_buffer_space(xprt);
788 } else
789 ret = -ENOTCONN;
790
791 spin_unlock(&xprt->transport_lock);
792 return ret;
793}
794
795static int xs_sock_nospace(struct rpc_rqst *req)
796{
797 struct sock_xprt *transport =
798 container_of(req->rq_xprt, struct sock_xprt, xprt);
799 struct sock *sk = transport->inet;
800 int ret = -EAGAIN;
801
802 lock_sock(sk);
803 if (!sock_writeable(sk))
804 ret = xs_nospace(req, transport);
805 release_sock(sk);
806 return ret;
807}
808
809static int xs_stream_nospace(struct rpc_rqst *req, bool vm_wait)
810{
811 struct sock_xprt *transport =
812 container_of(req->rq_xprt, struct sock_xprt, xprt);
813 struct sock *sk = transport->inet;
814 int ret = -EAGAIN;
815
816 if (vm_wait)
817 return -ENOBUFS;
818 lock_sock(sk);
819 if (!sk_stream_memory_free(sk))
820 ret = xs_nospace(req, transport);
821 release_sock(sk);
822 return ret;
823}
824
825static int xs_stream_prepare_request(struct rpc_rqst *req, struct xdr_buf *buf)
826{
827 return xdr_alloc_bvec(buf, rpc_task_gfp_mask());
828}
829
830/*
831 * Determine if the previous message in the stream was aborted before it
832 * could complete transmission.
833 */
834static bool
835xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req)
836{
837 return transport->xmit.offset != 0 && req->rq_bytes_sent == 0;
838}
839
840/*
841 * Return the stream record marker field for a record of length < 2^31-1
842 */
843static rpc_fraghdr
844xs_stream_record_marker(struct xdr_buf *xdr)
845{
846 if (!xdr->len)
847 return 0;
848 return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len);
849}
850
851/**
852 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
853 * @req: pointer to RPC request
854 *
855 * Return values:
856 * 0: The request has been sent
857 * EAGAIN: The socket was blocked, please call again later to
858 * complete the request
859 * ENOTCONN: Caller needs to invoke connect logic then call again
860 * other: Some other error occurred, the request was not sent
861 */
862static int xs_local_send_request(struct rpc_rqst *req)
863{
864 struct rpc_xprt *xprt = req->rq_xprt;
865 struct sock_xprt *transport =
866 container_of(xprt, struct sock_xprt, xprt);
867 struct xdr_buf *xdr = &req->rq_snd_buf;
868 rpc_fraghdr rm = xs_stream_record_marker(xdr);
869 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
870 struct msghdr msg = {
871 .msg_flags = XS_SENDMSG_FLAGS,
872 };
873 bool vm_wait;
874 unsigned int sent;
875 int status;
876
877 /* Close the stream if the previous transmission was incomplete */
878 if (xs_send_request_was_aborted(transport, req)) {
879 xprt_force_disconnect(xprt);
880 return -ENOTCONN;
881 }
882
883 xs_pktdump("packet data:",
884 req->rq_svec->iov_base, req->rq_svec->iov_len);
885
886 vm_wait = sk_stream_is_writeable(transport->inet) ? true : false;
887
888 req->rq_xtime = ktime_get();
889 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
890 transport->xmit.offset, rm, &sent);
891 dprintk("RPC: %s(%u) = %d\n",
892 __func__, xdr->len - transport->xmit.offset, status);
893
894 if (likely(sent > 0) || status == 0) {
895 transport->xmit.offset += sent;
896 req->rq_bytes_sent = transport->xmit.offset;
897 if (likely(req->rq_bytes_sent >= msglen)) {
898 req->rq_xmit_bytes_sent += transport->xmit.offset;
899 transport->xmit.offset = 0;
900 return 0;
901 }
902 status = -EAGAIN;
903 vm_wait = false;
904 }
905
906 switch (status) {
907 case -EAGAIN:
908 status = xs_stream_nospace(req, vm_wait);
909 break;
910 default:
911 dprintk("RPC: sendmsg returned unrecognized error %d\n",
912 -status);
913 fallthrough;
914 case -EPIPE:
915 xprt_force_disconnect(xprt);
916 status = -ENOTCONN;
917 }
918
919 return status;
920}
921
922/**
923 * xs_udp_send_request - write an RPC request to a UDP socket
924 * @req: pointer to RPC request
925 *
926 * Return values:
927 * 0: The request has been sent
928 * EAGAIN: The socket was blocked, please call again later to
929 * complete the request
930 * ENOTCONN: Caller needs to invoke connect logic then call again
931 * other: Some other error occurred, the request was not sent
932 */
933static int xs_udp_send_request(struct rpc_rqst *req)
934{
935 struct rpc_xprt *xprt = req->rq_xprt;
936 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
937 struct xdr_buf *xdr = &req->rq_snd_buf;
938 struct msghdr msg = {
939 .msg_name = xs_addr(xprt),
940 .msg_namelen = xprt->addrlen,
941 .msg_flags = XS_SENDMSG_FLAGS,
942 };
943 unsigned int sent;
944 int status;
945
946 xs_pktdump("packet data:",
947 req->rq_svec->iov_base,
948 req->rq_svec->iov_len);
949
950 if (!xprt_bound(xprt))
951 return -ENOTCONN;
952
953 if (!xprt_request_get_cong(xprt, req))
954 return -EBADSLT;
955
956 status = xdr_alloc_bvec(xdr, rpc_task_gfp_mask());
957 if (status < 0)
958 return status;
959 req->rq_xtime = ktime_get();
960 status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, 0, &sent);
961
962 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
963 xdr->len, status);
964
965 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
966 if (status == -EPERM)
967 goto process_status;
968
969 if (status == -EAGAIN && sock_writeable(transport->inet))
970 status = -ENOBUFS;
971
972 if (sent > 0 || status == 0) {
973 req->rq_xmit_bytes_sent += sent;
974 if (sent >= req->rq_slen)
975 return 0;
976 /* Still some bytes left; set up for a retry later. */
977 status = -EAGAIN;
978 }
979
980process_status:
981 switch (status) {
982 case -ENOTSOCK:
983 status = -ENOTCONN;
984 /* Should we call xs_close() here? */
985 break;
986 case -EAGAIN:
987 status = xs_sock_nospace(req);
988 break;
989 case -ENETUNREACH:
990 case -ENOBUFS:
991 case -EPIPE:
992 case -ECONNREFUSED:
993 case -EPERM:
994 /* When the server has died, an ICMP port unreachable message
995 * prompts ECONNREFUSED. */
996 break;
997 default:
998 dprintk("RPC: sendmsg returned unrecognized error %d\n",
999 -status);
1000 }
1001
1002 return status;
1003}
1004
1005/**
1006 * xs_tcp_send_request - write an RPC request to a TCP socket
1007 * @req: pointer to RPC request
1008 *
1009 * Return values:
1010 * 0: The request has been sent
1011 * EAGAIN: The socket was blocked, please call again later to
1012 * complete the request
1013 * ENOTCONN: Caller needs to invoke connect logic then call again
1014 * other: Some other error occurred, the request was not sent
1015 *
1016 * XXX: In the case of soft timeouts, should we eventually give up
1017 * if sendmsg is not able to make progress?
1018 */
1019static int xs_tcp_send_request(struct rpc_rqst *req)
1020{
1021 struct rpc_xprt *xprt = req->rq_xprt;
1022 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1023 struct xdr_buf *xdr = &req->rq_snd_buf;
1024 rpc_fraghdr rm = xs_stream_record_marker(xdr);
1025 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
1026 struct msghdr msg = {
1027 .msg_flags = XS_SENDMSG_FLAGS,
1028 };
1029 bool vm_wait;
1030 unsigned int sent;
1031 int status;
1032
1033 /* Close the stream if the previous transmission was incomplete */
1034 if (xs_send_request_was_aborted(transport, req)) {
1035 if (transport->sock != NULL)
1036 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
1037 return -ENOTCONN;
1038 }
1039 if (!transport->inet)
1040 return -ENOTCONN;
1041
1042 xs_pktdump("packet data:",
1043 req->rq_svec->iov_base,
1044 req->rq_svec->iov_len);
1045
1046 if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
1047 xs_tcp_set_socket_timeouts(xprt, transport->sock);
1048
1049 xs_set_srcport(transport, transport->sock);
1050
1051 /* Continue transmitting the packet/record. We must be careful
1052 * to cope with writespace callbacks arriving _after_ we have
1053 * called sendmsg(). */
1054 req->rq_xtime = ktime_get();
1055 tcp_sock_set_cork(transport->inet, true);
1056
1057 vm_wait = sk_stream_is_writeable(transport->inet) ? true : false;
1058
1059 do {
1060 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
1061 transport->xmit.offset, rm, &sent);
1062
1063 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
1064 xdr->len - transport->xmit.offset, status);
1065
1066 /* If we've sent the entire packet, immediately
1067 * reset the count of bytes sent. */
1068 transport->xmit.offset += sent;
1069 req->rq_bytes_sent = transport->xmit.offset;
1070 if (likely(req->rq_bytes_sent >= msglen)) {
1071 req->rq_xmit_bytes_sent += transport->xmit.offset;
1072 transport->xmit.offset = 0;
1073 if (atomic_long_read(&xprt->xmit_queuelen) == 1)
1074 tcp_sock_set_cork(transport->inet, false);
1075 return 0;
1076 }
1077
1078 WARN_ON_ONCE(sent == 0 && status == 0);
1079
1080 if (sent > 0)
1081 vm_wait = false;
1082
1083 } while (status == 0);
1084
1085 switch (status) {
1086 case -ENOTSOCK:
1087 status = -ENOTCONN;
1088 /* Should we call xs_close() here? */
1089 break;
1090 case -EAGAIN:
1091 status = xs_stream_nospace(req, vm_wait);
1092 break;
1093 case -ECONNRESET:
1094 case -ECONNREFUSED:
1095 case -ENOTCONN:
1096 case -EADDRINUSE:
1097 case -ENOBUFS:
1098 case -EPIPE:
1099 break;
1100 default:
1101 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1102 -status);
1103 }
1104
1105 return status;
1106}
1107
1108static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1109{
1110 transport->old_data_ready = sk->sk_data_ready;
1111 transport->old_state_change = sk->sk_state_change;
1112 transport->old_write_space = sk->sk_write_space;
1113 transport->old_error_report = sk->sk_error_report;
1114}
1115
1116static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1117{
1118 sk->sk_data_ready = transport->old_data_ready;
1119 sk->sk_state_change = transport->old_state_change;
1120 sk->sk_write_space = transport->old_write_space;
1121 sk->sk_error_report = transport->old_error_report;
1122}
1123
1124static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
1125{
1126 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1127
1128 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1129 clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state);
1130 clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state);
1131 clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state);
1132 clear_bit(XPRT_SOCK_NOSPACE, &transport->sock_state);
1133}
1134
1135static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr)
1136{
1137 set_bit(nr, &transport->sock_state);
1138 queue_work(xprtiod_workqueue, &transport->error_worker);
1139}
1140
1141static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1142{
1143 xprt->connect_cookie++;
1144 smp_mb__before_atomic();
1145 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1146 clear_bit(XPRT_CLOSING, &xprt->state);
1147 xs_sock_reset_state_flags(xprt);
1148 smp_mb__after_atomic();
1149}
1150
1151/**
1152 * xs_error_report - callback to handle TCP socket state errors
1153 * @sk: socket
1154 *
1155 * Note: we don't call sock_error() since there may be a rpc_task
1156 * using the socket, and so we don't want to clear sk->sk_err.
1157 */
1158static void xs_error_report(struct sock *sk)
1159{
1160 struct sock_xprt *transport;
1161 struct rpc_xprt *xprt;
1162
1163 if (!(xprt = xprt_from_sock(sk)))
1164 return;
1165
1166 transport = container_of(xprt, struct sock_xprt, xprt);
1167 transport->xprt_err = -sk->sk_err;
1168 if (transport->xprt_err == 0)
1169 return;
1170 dprintk("RPC: xs_error_report client %p, error=%d...\n",
1171 xprt, -transport->xprt_err);
1172 trace_rpc_socket_error(xprt, sk->sk_socket, transport->xprt_err);
1173
1174 /* barrier ensures xprt_err is set before XPRT_SOCK_WAKE_ERROR */
1175 smp_mb__before_atomic();
1176 xs_run_error_worker(transport, XPRT_SOCK_WAKE_ERROR);
1177}
1178
1179static void xs_reset_transport(struct sock_xprt *transport)
1180{
1181 struct socket *sock = transport->sock;
1182 struct sock *sk = transport->inet;
1183 struct rpc_xprt *xprt = &transport->xprt;
1184 struct file *filp = transport->file;
1185
1186 if (sk == NULL)
1187 return;
1188 /*
1189 * Make sure we're calling this in a context from which it is safe
1190 * to call __fput_sync(). In practice that means rpciod and the
1191 * system workqueue.
1192 */
1193 if (!(current->flags & PF_WQ_WORKER)) {
1194 WARN_ON_ONCE(1);
1195 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
1196 return;
1197 }
1198
1199 if (atomic_read(&transport->xprt.swapper))
1200 sk_clear_memalloc(sk);
1201
1202 kernel_sock_shutdown(sock, SHUT_RDWR);
1203
1204 mutex_lock(&transport->recv_mutex);
1205 lock_sock(sk);
1206 transport->inet = NULL;
1207 transport->sock = NULL;
1208 transport->file = NULL;
1209
1210 sk->sk_user_data = NULL;
1211
1212 xs_restore_old_callbacks(transport, sk);
1213 xprt_clear_connected(xprt);
1214 xs_sock_reset_connection_flags(xprt);
1215 /* Reset stream record info */
1216 xs_stream_reset_connect(transport);
1217 release_sock(sk);
1218 mutex_unlock(&transport->recv_mutex);
1219
1220 trace_rpc_socket_close(xprt, sock);
1221 __fput_sync(filp);
1222
1223 xprt_disconnect_done(xprt);
1224}
1225
1226/**
1227 * xs_close - close a socket
1228 * @xprt: transport
1229 *
1230 * This is used when all requests are complete; ie, no DRC state remains
1231 * on the server we want to save.
1232 *
1233 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1234 * xs_reset_transport() zeroing the socket from underneath a writer.
1235 */
1236static void xs_close(struct rpc_xprt *xprt)
1237{
1238 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1239
1240 dprintk("RPC: xs_close xprt %p\n", xprt);
1241
1242 xs_reset_transport(transport);
1243 xprt->reestablish_timeout = 0;
1244}
1245
1246static void xs_inject_disconnect(struct rpc_xprt *xprt)
1247{
1248 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
1249 xprt);
1250 xprt_disconnect_done(xprt);
1251}
1252
1253static void xs_xprt_free(struct rpc_xprt *xprt)
1254{
1255 xs_free_peer_addresses(xprt);
1256 xprt_free(xprt);
1257}
1258
1259/**
1260 * xs_destroy - prepare to shutdown a transport
1261 * @xprt: doomed transport
1262 *
1263 */
1264static void xs_destroy(struct rpc_xprt *xprt)
1265{
1266 struct sock_xprt *transport = container_of(xprt,
1267 struct sock_xprt, xprt);
1268 dprintk("RPC: xs_destroy xprt %p\n", xprt);
1269
1270 cancel_delayed_work_sync(&transport->connect_worker);
1271 xs_close(xprt);
1272 cancel_work_sync(&transport->recv_worker);
1273 cancel_work_sync(&transport->error_worker);
1274 xs_xprt_free(xprt);
1275 module_put(THIS_MODULE);
1276}
1277
1278/**
1279 * xs_udp_data_read_skb - receive callback for UDP sockets
1280 * @xprt: transport
1281 * @sk: socket
1282 * @skb: skbuff
1283 *
1284 */
1285static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1286 struct sock *sk,
1287 struct sk_buff *skb)
1288{
1289 struct rpc_task *task;
1290 struct rpc_rqst *rovr;
1291 int repsize, copied;
1292 u32 _xid;
1293 __be32 *xp;
1294
1295 repsize = skb->len;
1296 if (repsize < 4) {
1297 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
1298 return;
1299 }
1300
1301 /* Copy the XID from the skb... */
1302 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1303 if (xp == NULL)
1304 return;
1305
1306 /* Look up and lock the request corresponding to the given XID */
1307 spin_lock(&xprt->queue_lock);
1308 rovr = xprt_lookup_rqst(xprt, *xp);
1309 if (!rovr)
1310 goto out_unlock;
1311 xprt_pin_rqst(rovr);
1312 xprt_update_rtt(rovr->rq_task);
1313 spin_unlock(&xprt->queue_lock);
1314 task = rovr->rq_task;
1315
1316 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1317 copied = repsize;
1318
1319 /* Suck it into the iovec, verify checksum if not done by hw. */
1320 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1321 spin_lock(&xprt->queue_lock);
1322 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1323 goto out_unpin;
1324 }
1325
1326
1327 spin_lock(&xprt->transport_lock);
1328 xprt_adjust_cwnd(xprt, task, copied);
1329 spin_unlock(&xprt->transport_lock);
1330 spin_lock(&xprt->queue_lock);
1331 xprt_complete_rqst(task, copied);
1332 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1333out_unpin:
1334 xprt_unpin_rqst(rovr);
1335 out_unlock:
1336 spin_unlock(&xprt->queue_lock);
1337}
1338
1339static void xs_udp_data_receive(struct sock_xprt *transport)
1340{
1341 struct sk_buff *skb;
1342 struct sock *sk;
1343 int err;
1344
1345 mutex_lock(&transport->recv_mutex);
1346 sk = transport->inet;
1347 if (sk == NULL)
1348 goto out;
1349 for (;;) {
1350 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err);
1351 if (skb == NULL)
1352 break;
1353 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1354 consume_skb(skb);
1355 cond_resched();
1356 }
1357 xs_poll_check_readable(transport);
1358out:
1359 mutex_unlock(&transport->recv_mutex);
1360}
1361
1362static void xs_udp_data_receive_workfn(struct work_struct *work)
1363{
1364 struct sock_xprt *transport =
1365 container_of(work, struct sock_xprt, recv_worker);
1366 unsigned int pflags = memalloc_nofs_save();
1367
1368 xs_udp_data_receive(transport);
1369 memalloc_nofs_restore(pflags);
1370}
1371
1372/**
1373 * xs_data_ready - "data ready" callback for sockets
1374 * @sk: socket with data to read
1375 *
1376 */
1377static void xs_data_ready(struct sock *sk)
1378{
1379 struct rpc_xprt *xprt;
1380
1381 xprt = xprt_from_sock(sk);
1382 if (xprt != NULL) {
1383 struct sock_xprt *transport = container_of(xprt,
1384 struct sock_xprt, xprt);
1385
1386 trace_xs_data_ready(xprt);
1387
1388 transport->old_data_ready(sk);
1389 /* Any data means we had a useful conversation, so
1390 * then we don't need to delay the next reconnect
1391 */
1392 if (xprt->reestablish_timeout)
1393 xprt->reestablish_timeout = 0;
1394 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1395 queue_work(xprtiod_workqueue, &transport->recv_worker);
1396 }
1397}
1398
1399/*
1400 * Helper function to force a TCP close if the server is sending
1401 * junk and/or it has put us in CLOSE_WAIT
1402 */
1403static void xs_tcp_force_close(struct rpc_xprt *xprt)
1404{
1405 xprt_force_disconnect(xprt);
1406}
1407
1408#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1409static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1410{
1411 return PAGE_SIZE;
1412}
1413#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1414
1415/**
1416 * xs_local_state_change - callback to handle AF_LOCAL socket state changes
1417 * @sk: socket whose state has changed
1418 *
1419 */
1420static void xs_local_state_change(struct sock *sk)
1421{
1422 struct rpc_xprt *xprt;
1423 struct sock_xprt *transport;
1424
1425 if (!(xprt = xprt_from_sock(sk)))
1426 return;
1427 transport = container_of(xprt, struct sock_xprt, xprt);
1428 if (sk->sk_shutdown & SHUTDOWN_MASK) {
1429 clear_bit(XPRT_CONNECTED, &xprt->state);
1430 /* Trigger the socket release */
1431 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1432 }
1433}
1434
1435/**
1436 * xs_tcp_state_change - callback to handle TCP socket state changes
1437 * @sk: socket whose state has changed
1438 *
1439 */
1440static void xs_tcp_state_change(struct sock *sk)
1441{
1442 struct rpc_xprt *xprt;
1443 struct sock_xprt *transport;
1444
1445 if (!(xprt = xprt_from_sock(sk)))
1446 return;
1447 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1448 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1449 sk->sk_state, xprt_connected(xprt),
1450 sock_flag(sk, SOCK_DEAD),
1451 sock_flag(sk, SOCK_ZAPPED),
1452 sk->sk_shutdown);
1453
1454 transport = container_of(xprt, struct sock_xprt, xprt);
1455 trace_rpc_socket_state_change(xprt, sk->sk_socket);
1456 switch (sk->sk_state) {
1457 case TCP_ESTABLISHED:
1458 if (!xprt_test_and_set_connected(xprt)) {
1459 xprt->connect_cookie++;
1460 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1461 xprt_clear_connecting(xprt);
1462
1463 xprt->stat.connect_count++;
1464 xprt->stat.connect_time += (long)jiffies -
1465 xprt->stat.connect_start;
1466 xs_run_error_worker(transport, XPRT_SOCK_WAKE_PENDING);
1467 }
1468 break;
1469 case TCP_FIN_WAIT1:
1470 /* The client initiated a shutdown of the socket */
1471 xprt->connect_cookie++;
1472 xprt->reestablish_timeout = 0;
1473 set_bit(XPRT_CLOSING, &xprt->state);
1474 smp_mb__before_atomic();
1475 clear_bit(XPRT_CONNECTED, &xprt->state);
1476 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1477 smp_mb__after_atomic();
1478 break;
1479 case TCP_CLOSE_WAIT:
1480 /* The server initiated a shutdown of the socket */
1481 xprt->connect_cookie++;
1482 clear_bit(XPRT_CONNECTED, &xprt->state);
1483 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1484 fallthrough;
1485 case TCP_CLOSING:
1486 /*
1487 * If the server closed down the connection, make sure that
1488 * we back off before reconnecting
1489 */
1490 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1491 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1492 break;
1493 case TCP_LAST_ACK:
1494 set_bit(XPRT_CLOSING, &xprt->state);
1495 smp_mb__before_atomic();
1496 clear_bit(XPRT_CONNECTED, &xprt->state);
1497 smp_mb__after_atomic();
1498 break;
1499 case TCP_CLOSE:
1500 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1501 &transport->sock_state))
1502 xprt_clear_connecting(xprt);
1503 clear_bit(XPRT_CLOSING, &xprt->state);
1504 /* Trigger the socket release */
1505 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1506 }
1507}
1508
1509static void xs_write_space(struct sock *sk)
1510{
1511 struct sock_xprt *transport;
1512 struct rpc_xprt *xprt;
1513
1514 if (!sk->sk_socket)
1515 return;
1516 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1517
1518 if (unlikely(!(xprt = xprt_from_sock(sk))))
1519 return;
1520 transport = container_of(xprt, struct sock_xprt, xprt);
1521 if (!test_and_clear_bit(XPRT_SOCK_NOSPACE, &transport->sock_state))
1522 return;
1523 xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE);
1524 sk->sk_write_pending--;
1525}
1526
1527/**
1528 * xs_udp_write_space - callback invoked when socket buffer space
1529 * becomes available
1530 * @sk: socket whose state has changed
1531 *
1532 * Called when more output buffer space is available for this socket.
1533 * We try not to wake our writers until they can make "significant"
1534 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1535 * with a bunch of small requests.
1536 */
1537static void xs_udp_write_space(struct sock *sk)
1538{
1539 /* from net/core/sock.c:sock_def_write_space */
1540 if (sock_writeable(sk))
1541 xs_write_space(sk);
1542}
1543
1544/**
1545 * xs_tcp_write_space - callback invoked when socket buffer space
1546 * becomes available
1547 * @sk: socket whose state has changed
1548 *
1549 * Called when more output buffer space is available for this socket.
1550 * We try not to wake our writers until they can make "significant"
1551 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1552 * with a bunch of small requests.
1553 */
1554static void xs_tcp_write_space(struct sock *sk)
1555{
1556 /* from net/core/stream.c:sk_stream_write_space */
1557 if (sk_stream_is_writeable(sk))
1558 xs_write_space(sk);
1559}
1560
1561static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1562{
1563 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1564 struct sock *sk = transport->inet;
1565
1566 if (transport->rcvsize) {
1567 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1568 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1569 }
1570 if (transport->sndsize) {
1571 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1572 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1573 sk->sk_write_space(sk);
1574 }
1575}
1576
1577/**
1578 * xs_udp_set_buffer_size - set send and receive limits
1579 * @xprt: generic transport
1580 * @sndsize: requested size of send buffer, in bytes
1581 * @rcvsize: requested size of receive buffer, in bytes
1582 *
1583 * Set socket send and receive buffer size limits.
1584 */
1585static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1586{
1587 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1588
1589 transport->sndsize = 0;
1590 if (sndsize)
1591 transport->sndsize = sndsize + 1024;
1592 transport->rcvsize = 0;
1593 if (rcvsize)
1594 transport->rcvsize = rcvsize + 1024;
1595
1596 xs_udp_do_set_buffer_size(xprt);
1597}
1598
1599/**
1600 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1601 * @xprt: controlling transport
1602 * @task: task that timed out
1603 *
1604 * Adjust the congestion window after a retransmit timeout has occurred.
1605 */
1606static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1607{
1608 spin_lock(&xprt->transport_lock);
1609 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1610 spin_unlock(&xprt->transport_lock);
1611}
1612
1613static int xs_get_random_port(void)
1614{
1615 unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1616 unsigned short range;
1617 unsigned short rand;
1618
1619 if (max < min)
1620 return -EADDRINUSE;
1621 range = max - min + 1;
1622 rand = get_random_u32_below(range);
1623 return rand + min;
1624}
1625
1626static unsigned short xs_sock_getport(struct socket *sock)
1627{
1628 struct sockaddr_storage buf;
1629 unsigned short port = 0;
1630
1631 if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1632 goto out;
1633 switch (buf.ss_family) {
1634 case AF_INET6:
1635 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1636 break;
1637 case AF_INET:
1638 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1639 }
1640out:
1641 return port;
1642}
1643
1644/**
1645 * xs_set_port - reset the port number in the remote endpoint address
1646 * @xprt: generic transport
1647 * @port: new port number
1648 *
1649 */
1650static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1651{
1652 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1653
1654 rpc_set_port(xs_addr(xprt), port);
1655 xs_update_peer_port(xprt);
1656}
1657
1658static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1659{
1660 if (transport->srcport == 0 && transport->xprt.reuseport)
1661 transport->srcport = xs_sock_getport(sock);
1662}
1663
1664static int xs_get_srcport(struct sock_xprt *transport)
1665{
1666 int port = transport->srcport;
1667
1668 if (port == 0 && transport->xprt.resvport)
1669 port = xs_get_random_port();
1670 return port;
1671}
1672
1673static unsigned short xs_sock_srcport(struct rpc_xprt *xprt)
1674{
1675 struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt);
1676 unsigned short ret = 0;
1677 mutex_lock(&sock->recv_mutex);
1678 if (sock->sock)
1679 ret = xs_sock_getport(sock->sock);
1680 mutex_unlock(&sock->recv_mutex);
1681 return ret;
1682}
1683
1684static int xs_sock_srcaddr(struct rpc_xprt *xprt, char *buf, size_t buflen)
1685{
1686 struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt);
1687 union {
1688 struct sockaddr sa;
1689 struct sockaddr_storage st;
1690 } saddr;
1691 int ret = -ENOTCONN;
1692
1693 mutex_lock(&sock->recv_mutex);
1694 if (sock->sock) {
1695 ret = kernel_getsockname(sock->sock, &saddr.sa);
1696 if (ret >= 0)
1697 ret = snprintf(buf, buflen, "%pISc", &saddr.sa);
1698 }
1699 mutex_unlock(&sock->recv_mutex);
1700 return ret;
1701}
1702
1703static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1704{
1705 if (transport->srcport != 0)
1706 transport->srcport = 0;
1707 if (!transport->xprt.resvport)
1708 return 0;
1709 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1710 return xprt_max_resvport;
1711 return --port;
1712}
1713static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1714{
1715 struct sockaddr_storage myaddr;
1716 int err, nloop = 0;
1717 int port = xs_get_srcport(transport);
1718 unsigned short last;
1719
1720 /*
1721 * If we are asking for any ephemeral port (i.e. port == 0 &&
1722 * transport->xprt.resvport == 0), don't bind. Let the local
1723 * port selection happen implicitly when the socket is used
1724 * (for example at connect time).
1725 *
1726 * This ensures that we can continue to establish TCP
1727 * connections even when all local ephemeral ports are already
1728 * a part of some TCP connection. This makes no difference
1729 * for UDP sockets, but also doesn't harm them.
1730 *
1731 * If we're asking for any reserved port (i.e. port == 0 &&
1732 * transport->xprt.resvport == 1) xs_get_srcport above will
1733 * ensure that port is non-zero and we will bind as needed.
1734 */
1735 if (port <= 0)
1736 return port;
1737
1738 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1739 do {
1740 rpc_set_port((struct sockaddr *)&myaddr, port);
1741 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1742 transport->xprt.addrlen);
1743 if (err == 0) {
1744 if (transport->xprt.reuseport)
1745 transport->srcport = port;
1746 break;
1747 }
1748 last = port;
1749 port = xs_next_srcport(transport, port);
1750 if (port > last)
1751 nloop++;
1752 } while (err == -EADDRINUSE && nloop != 2);
1753
1754 if (myaddr.ss_family == AF_INET)
1755 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1756 &((struct sockaddr_in *)&myaddr)->sin_addr,
1757 port, err ? "failed" : "ok", err);
1758 else
1759 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1760 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1761 port, err ? "failed" : "ok", err);
1762 return err;
1763}
1764
1765/*
1766 * We don't support autobind on AF_LOCAL sockets
1767 */
1768static void xs_local_rpcbind(struct rpc_task *task)
1769{
1770 xprt_set_bound(task->tk_xprt);
1771}
1772
1773static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1774{
1775}
1776
1777#ifdef CONFIG_DEBUG_LOCK_ALLOC
1778static struct lock_class_key xs_key[3];
1779static struct lock_class_key xs_slock_key[3];
1780
1781static inline void xs_reclassify_socketu(struct socket *sock)
1782{
1783 struct sock *sk = sock->sk;
1784
1785 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1786 &xs_slock_key[0], "sk_lock-AF_LOCAL-RPC", &xs_key[0]);
1787}
1788
1789static inline void xs_reclassify_socket4(struct socket *sock)
1790{
1791 struct sock *sk = sock->sk;
1792
1793 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1794 &xs_slock_key[1], "sk_lock-AF_INET-RPC", &xs_key[1]);
1795}
1796
1797static inline void xs_reclassify_socket6(struct socket *sock)
1798{
1799 struct sock *sk = sock->sk;
1800
1801 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1802 &xs_slock_key[2], "sk_lock-AF_INET6-RPC", &xs_key[2]);
1803}
1804
1805static inline void xs_reclassify_socket(int family, struct socket *sock)
1806{
1807 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1808 return;
1809
1810 switch (family) {
1811 case AF_LOCAL:
1812 xs_reclassify_socketu(sock);
1813 break;
1814 case AF_INET:
1815 xs_reclassify_socket4(sock);
1816 break;
1817 case AF_INET6:
1818 xs_reclassify_socket6(sock);
1819 break;
1820 }
1821}
1822#else
1823static inline void xs_reclassify_socket(int family, struct socket *sock)
1824{
1825}
1826#endif
1827
1828static void xs_dummy_setup_socket(struct work_struct *work)
1829{
1830}
1831
1832static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1833 struct sock_xprt *transport, int family, int type,
1834 int protocol, bool reuseport)
1835{
1836 struct file *filp;
1837 struct socket *sock;
1838 int err;
1839
1840 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1841 if (err < 0) {
1842 dprintk("RPC: can't create %d transport socket (%d).\n",
1843 protocol, -err);
1844 goto out;
1845 }
1846 xs_reclassify_socket(family, sock);
1847
1848 if (reuseport)
1849 sock_set_reuseport(sock->sk);
1850
1851 err = xs_bind(transport, sock);
1852 if (err) {
1853 sock_release(sock);
1854 goto out;
1855 }
1856
1857 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1858 if (IS_ERR(filp))
1859 return ERR_CAST(filp);
1860 transport->file = filp;
1861
1862 return sock;
1863out:
1864 return ERR_PTR(err);
1865}
1866
1867static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1868 struct socket *sock)
1869{
1870 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1871 xprt);
1872
1873 if (!transport->inet) {
1874 struct sock *sk = sock->sk;
1875
1876 lock_sock(sk);
1877
1878 xs_save_old_callbacks(transport, sk);
1879
1880 sk->sk_user_data = xprt;
1881 sk->sk_data_ready = xs_data_ready;
1882 sk->sk_write_space = xs_udp_write_space;
1883 sk->sk_state_change = xs_local_state_change;
1884 sk->sk_error_report = xs_error_report;
1885 sk->sk_use_task_frag = false;
1886
1887 xprt_clear_connected(xprt);
1888
1889 /* Reset to new socket */
1890 transport->sock = sock;
1891 transport->inet = sk;
1892
1893 release_sock(sk);
1894 }
1895
1896 xs_stream_start_connect(transport);
1897
1898 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1899}
1900
1901/**
1902 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1903 * @transport: socket transport to connect
1904 */
1905static int xs_local_setup_socket(struct sock_xprt *transport)
1906{
1907 struct rpc_xprt *xprt = &transport->xprt;
1908 struct file *filp;
1909 struct socket *sock;
1910 int status;
1911
1912 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1913 SOCK_STREAM, 0, &sock, 1);
1914 if (status < 0) {
1915 dprintk("RPC: can't create AF_LOCAL "
1916 "transport socket (%d).\n", -status);
1917 goto out;
1918 }
1919 xs_reclassify_socket(AF_LOCAL, sock);
1920
1921 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1922 if (IS_ERR(filp)) {
1923 status = PTR_ERR(filp);
1924 goto out;
1925 }
1926 transport->file = filp;
1927
1928 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1929 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1930
1931 status = xs_local_finish_connecting(xprt, sock);
1932 trace_rpc_socket_connect(xprt, sock, status);
1933 switch (status) {
1934 case 0:
1935 dprintk("RPC: xprt %p connected to %s\n",
1936 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1937 xprt->stat.connect_count++;
1938 xprt->stat.connect_time += (long)jiffies -
1939 xprt->stat.connect_start;
1940 xprt_set_connected(xprt);
1941 break;
1942 case -ENOBUFS:
1943 break;
1944 case -ENOENT:
1945 dprintk("RPC: xprt %p: socket %s does not exist\n",
1946 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1947 break;
1948 case -ECONNREFUSED:
1949 dprintk("RPC: xprt %p: connection refused for %s\n",
1950 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1951 break;
1952 default:
1953 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1954 __func__, -status,
1955 xprt->address_strings[RPC_DISPLAY_ADDR]);
1956 }
1957
1958out:
1959 xprt_clear_connecting(xprt);
1960 xprt_wake_pending_tasks(xprt, status);
1961 return status;
1962}
1963
1964static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1965{
1966 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1967 int ret;
1968
1969 if (transport->file)
1970 goto force_disconnect;
1971
1972 if (RPC_IS_ASYNC(task)) {
1973 /*
1974 * We want the AF_LOCAL connect to be resolved in the
1975 * filesystem namespace of the process making the rpc
1976 * call. Thus we connect synchronously.
1977 *
1978 * If we want to support asynchronous AF_LOCAL calls,
1979 * we'll need to figure out how to pass a namespace to
1980 * connect.
1981 */
1982 rpc_task_set_rpc_status(task, -ENOTCONN);
1983 goto out_wake;
1984 }
1985 ret = xs_local_setup_socket(transport);
1986 if (ret && !RPC_IS_SOFTCONN(task))
1987 msleep_interruptible(15000);
1988 return;
1989force_disconnect:
1990 xprt_force_disconnect(xprt);
1991out_wake:
1992 xprt_clear_connecting(xprt);
1993 xprt_wake_pending_tasks(xprt, -ENOTCONN);
1994}
1995
1996#if IS_ENABLED(CONFIG_SUNRPC_SWAP)
1997/*
1998 * Note that this should be called with XPRT_LOCKED held, or recv_mutex
1999 * held, or when we otherwise know that we have exclusive access to the
2000 * socket, to guard against races with xs_reset_transport.
2001 */
2002static void xs_set_memalloc(struct rpc_xprt *xprt)
2003{
2004 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2005 xprt);
2006
2007 /*
2008 * If there's no sock, then we have nothing to set. The
2009 * reconnecting process will get it for us.
2010 */
2011 if (!transport->inet)
2012 return;
2013 if (atomic_read(&xprt->swapper))
2014 sk_set_memalloc(transport->inet);
2015}
2016
2017/**
2018 * xs_enable_swap - Tag this transport as being used for swap.
2019 * @xprt: transport to tag
2020 *
2021 * Take a reference to this transport on behalf of the rpc_clnt, and
2022 * optionally mark it for swapping if it wasn't already.
2023 */
2024static int
2025xs_enable_swap(struct rpc_xprt *xprt)
2026{
2027 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2028
2029 mutex_lock(&xs->recv_mutex);
2030 if (atomic_inc_return(&xprt->swapper) == 1 &&
2031 xs->inet)
2032 sk_set_memalloc(xs->inet);
2033 mutex_unlock(&xs->recv_mutex);
2034 return 0;
2035}
2036
2037/**
2038 * xs_disable_swap - Untag this transport as being used for swap.
2039 * @xprt: transport to tag
2040 *
2041 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2042 * swapper refcount goes to 0, untag the socket as a memalloc socket.
2043 */
2044static void
2045xs_disable_swap(struct rpc_xprt *xprt)
2046{
2047 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2048
2049 mutex_lock(&xs->recv_mutex);
2050 if (atomic_dec_and_test(&xprt->swapper) &&
2051 xs->inet)
2052 sk_clear_memalloc(xs->inet);
2053 mutex_unlock(&xs->recv_mutex);
2054}
2055#else
2056static void xs_set_memalloc(struct rpc_xprt *xprt)
2057{
2058}
2059
2060static int
2061xs_enable_swap(struct rpc_xprt *xprt)
2062{
2063 return -EINVAL;
2064}
2065
2066static void
2067xs_disable_swap(struct rpc_xprt *xprt)
2068{
2069}
2070#endif
2071
2072static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2073{
2074 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2075
2076 if (!transport->inet) {
2077 struct sock *sk = sock->sk;
2078
2079 lock_sock(sk);
2080
2081 xs_save_old_callbacks(transport, sk);
2082
2083 sk->sk_user_data = xprt;
2084 sk->sk_data_ready = xs_data_ready;
2085 sk->sk_write_space = xs_udp_write_space;
2086 sk->sk_use_task_frag = false;
2087
2088 xprt_set_connected(xprt);
2089
2090 /* Reset to new socket */
2091 transport->sock = sock;
2092 transport->inet = sk;
2093
2094 xs_set_memalloc(xprt);
2095
2096 release_sock(sk);
2097 }
2098 xs_udp_do_set_buffer_size(xprt);
2099
2100 xprt->stat.connect_start = jiffies;
2101}
2102
2103static void xs_udp_setup_socket(struct work_struct *work)
2104{
2105 struct sock_xprt *transport =
2106 container_of(work, struct sock_xprt, connect_worker.work);
2107 struct rpc_xprt *xprt = &transport->xprt;
2108 struct socket *sock;
2109 int status = -EIO;
2110 unsigned int pflags = current->flags;
2111
2112 if (atomic_read(&xprt->swapper))
2113 current->flags |= PF_MEMALLOC;
2114 sock = xs_create_sock(xprt, transport,
2115 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2116 IPPROTO_UDP, false);
2117 if (IS_ERR(sock))
2118 goto out;
2119
2120 dprintk("RPC: worker connecting xprt %p via %s to "
2121 "%s (port %s)\n", xprt,
2122 xprt->address_strings[RPC_DISPLAY_PROTO],
2123 xprt->address_strings[RPC_DISPLAY_ADDR],
2124 xprt->address_strings[RPC_DISPLAY_PORT]);
2125
2126 xs_udp_finish_connecting(xprt, sock);
2127 trace_rpc_socket_connect(xprt, sock, 0);
2128 status = 0;
2129out:
2130 xprt_clear_connecting(xprt);
2131 xprt_unlock_connect(xprt, transport);
2132 xprt_wake_pending_tasks(xprt, status);
2133 current_restore_flags(pflags, PF_MEMALLOC);
2134}
2135
2136/**
2137 * xs_tcp_shutdown - gracefully shut down a TCP socket
2138 * @xprt: transport
2139 *
2140 * Initiates a graceful shutdown of the TCP socket by calling the
2141 * equivalent of shutdown(SHUT_RDWR);
2142 */
2143static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2144{
2145 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2146 struct socket *sock = transport->sock;
2147 int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2148
2149 if (sock == NULL)
2150 return;
2151 if (!xprt->reuseport) {
2152 xs_close(xprt);
2153 return;
2154 }
2155 switch (skst) {
2156 case TCP_FIN_WAIT1:
2157 case TCP_FIN_WAIT2:
2158 break;
2159 case TCP_ESTABLISHED:
2160 case TCP_CLOSE_WAIT:
2161 kernel_sock_shutdown(sock, SHUT_RDWR);
2162 trace_rpc_socket_shutdown(xprt, sock);
2163 break;
2164 default:
2165 xs_reset_transport(transport);
2166 }
2167}
2168
2169static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2170 struct socket *sock)
2171{
2172 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2173 unsigned int keepidle;
2174 unsigned int keepcnt;
2175 unsigned int timeo;
2176
2177 spin_lock(&xprt->transport_lock);
2178 keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2179 keepcnt = xprt->timeout->to_retries + 1;
2180 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2181 (xprt->timeout->to_retries + 1);
2182 clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2183 spin_unlock(&xprt->transport_lock);
2184
2185 /* TCP Keepalive options */
2186 sock_set_keepalive(sock->sk);
2187 tcp_sock_set_keepidle(sock->sk, keepidle);
2188 tcp_sock_set_keepintvl(sock->sk, keepidle);
2189 tcp_sock_set_keepcnt(sock->sk, keepcnt);
2190
2191 /* TCP user timeout (see RFC5482) */
2192 tcp_sock_set_user_timeout(sock->sk, timeo);
2193}
2194
2195static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2196 unsigned long connect_timeout,
2197 unsigned long reconnect_timeout)
2198{
2199 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2200 struct rpc_timeout to;
2201 unsigned long initval;
2202
2203 spin_lock(&xprt->transport_lock);
2204 if (reconnect_timeout < xprt->max_reconnect_timeout)
2205 xprt->max_reconnect_timeout = reconnect_timeout;
2206 if (connect_timeout < xprt->connect_timeout) {
2207 memcpy(&to, xprt->timeout, sizeof(to));
2208 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2209 /* Arbitrary lower limit */
2210 if (initval < XS_TCP_INIT_REEST_TO << 1)
2211 initval = XS_TCP_INIT_REEST_TO << 1;
2212 to.to_initval = initval;
2213 to.to_maxval = initval;
2214 memcpy(&transport->tcp_timeout, &to,
2215 sizeof(transport->tcp_timeout));
2216 xprt->timeout = &transport->tcp_timeout;
2217 xprt->connect_timeout = connect_timeout;
2218 }
2219 set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2220 spin_unlock(&xprt->transport_lock);
2221}
2222
2223static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2224{
2225 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2226
2227 if (!transport->inet) {
2228 struct sock *sk = sock->sk;
2229
2230 /* Avoid temporary address, they are bad for long-lived
2231 * connections such as NFS mounts.
2232 * RFC4941, section 3.6 suggests that:
2233 * Individual applications, which have specific
2234 * knowledge about the normal duration of connections,
2235 * MAY override this as appropriate.
2236 */
2237 if (xs_addr(xprt)->sa_family == PF_INET6) {
2238 ip6_sock_set_addr_preferences(sk,
2239 IPV6_PREFER_SRC_PUBLIC);
2240 }
2241
2242 xs_tcp_set_socket_timeouts(xprt, sock);
2243 tcp_sock_set_nodelay(sk);
2244
2245 lock_sock(sk);
2246
2247 xs_save_old_callbacks(transport, sk);
2248
2249 sk->sk_user_data = xprt;
2250 sk->sk_data_ready = xs_data_ready;
2251 sk->sk_state_change = xs_tcp_state_change;
2252 sk->sk_write_space = xs_tcp_write_space;
2253 sk->sk_error_report = xs_error_report;
2254 sk->sk_use_task_frag = false;
2255
2256 /* socket options */
2257 sock_reset_flag(sk, SOCK_LINGER);
2258
2259 xprt_clear_connected(xprt);
2260
2261 /* Reset to new socket */
2262 transport->sock = sock;
2263 transport->inet = sk;
2264
2265 release_sock(sk);
2266 }
2267
2268 if (!xprt_bound(xprt))
2269 return -ENOTCONN;
2270
2271 xs_set_memalloc(xprt);
2272
2273 xs_stream_start_connect(transport);
2274
2275 /* Tell the socket layer to start connecting... */
2276 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2277 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2278}
2279
2280/**
2281 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2282 * @work: queued work item
2283 *
2284 * Invoked by a work queue tasklet.
2285 */
2286static void xs_tcp_setup_socket(struct work_struct *work)
2287{
2288 struct sock_xprt *transport =
2289 container_of(work, struct sock_xprt, connect_worker.work);
2290 struct socket *sock = transport->sock;
2291 struct rpc_xprt *xprt = &transport->xprt;
2292 int status;
2293 unsigned int pflags = current->flags;
2294
2295 if (atomic_read(&xprt->swapper))
2296 current->flags |= PF_MEMALLOC;
2297
2298 if (xprt_connected(xprt))
2299 goto out;
2300 if (test_and_clear_bit(XPRT_SOCK_CONNECT_SENT,
2301 &transport->sock_state) ||
2302 !sock) {
2303 xs_reset_transport(transport);
2304 sock = xs_create_sock(xprt, transport, xs_addr(xprt)->sa_family,
2305 SOCK_STREAM, IPPROTO_TCP, true);
2306 if (IS_ERR(sock)) {
2307 xprt_wake_pending_tasks(xprt, PTR_ERR(sock));
2308 goto out;
2309 }
2310 }
2311
2312 dprintk("RPC: worker connecting xprt %p via %s to "
2313 "%s (port %s)\n", xprt,
2314 xprt->address_strings[RPC_DISPLAY_PROTO],
2315 xprt->address_strings[RPC_DISPLAY_ADDR],
2316 xprt->address_strings[RPC_DISPLAY_PORT]);
2317
2318 status = xs_tcp_finish_connecting(xprt, sock);
2319 trace_rpc_socket_connect(xprt, sock, status);
2320 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2321 xprt, -status, xprt_connected(xprt),
2322 sock->sk->sk_state);
2323 switch (status) {
2324 case 0:
2325 case -EINPROGRESS:
2326 /* SYN_SENT! */
2327 set_bit(XPRT_SOCK_CONNECT_SENT, &transport->sock_state);
2328 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2329 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2330 fallthrough;
2331 case -EALREADY:
2332 goto out_unlock;
2333 case -EADDRNOTAVAIL:
2334 /* Source port number is unavailable. Try a new one! */
2335 transport->srcport = 0;
2336 status = -EAGAIN;
2337 break;
2338 case -EINVAL:
2339 /* Happens, for instance, if the user specified a link
2340 * local IPv6 address without a scope-id.
2341 */
2342 case -ECONNREFUSED:
2343 case -ECONNRESET:
2344 case -ENETDOWN:
2345 case -ENETUNREACH:
2346 case -EHOSTUNREACH:
2347 case -EADDRINUSE:
2348 case -ENOBUFS:
2349 break;
2350 default:
2351 printk("%s: connect returned unhandled error %d\n",
2352 __func__, status);
2353 status = -EAGAIN;
2354 }
2355
2356 /* xs_tcp_force_close() wakes tasks with a fixed error code.
2357 * We need to wake them first to ensure the correct error code.
2358 */
2359 xprt_wake_pending_tasks(xprt, status);
2360 xs_tcp_force_close(xprt);
2361out:
2362 xprt_clear_connecting(xprt);
2363out_unlock:
2364 xprt_unlock_connect(xprt, transport);
2365 current_restore_flags(pflags, PF_MEMALLOC);
2366}
2367
2368/**
2369 * xs_connect - connect a socket to a remote endpoint
2370 * @xprt: pointer to transport structure
2371 * @task: address of RPC task that manages state of connect request
2372 *
2373 * TCP: If the remote end dropped the connection, delay reconnecting.
2374 *
2375 * UDP socket connects are synchronous, but we use a work queue anyway
2376 * to guarantee that even unprivileged user processes can set up a
2377 * socket on a privileged port.
2378 *
2379 * If a UDP socket connect fails, the delay behavior here prevents
2380 * retry floods (hard mounts).
2381 */
2382static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2383{
2384 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2385 unsigned long delay = 0;
2386
2387 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2388
2389 if (transport->sock != NULL) {
2390 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2391 "seconds\n", xprt, xprt->reestablish_timeout / HZ);
2392
2393 delay = xprt_reconnect_delay(xprt);
2394 xprt_reconnect_backoff(xprt, XS_TCP_INIT_REEST_TO);
2395
2396 } else
2397 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2398
2399 queue_delayed_work(xprtiod_workqueue,
2400 &transport->connect_worker,
2401 delay);
2402}
2403
2404static void xs_wake_disconnect(struct sock_xprt *transport)
2405{
2406 if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state))
2407 xs_tcp_force_close(&transport->xprt);
2408}
2409
2410static void xs_wake_write(struct sock_xprt *transport)
2411{
2412 if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state))
2413 xprt_write_space(&transport->xprt);
2414}
2415
2416static void xs_wake_error(struct sock_xprt *transport)
2417{
2418 int sockerr;
2419
2420 if (!test_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2421 return;
2422 mutex_lock(&transport->recv_mutex);
2423 if (transport->sock == NULL)
2424 goto out;
2425 if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2426 goto out;
2427 sockerr = xchg(&transport->xprt_err, 0);
2428 if (sockerr < 0)
2429 xprt_wake_pending_tasks(&transport->xprt, sockerr);
2430out:
2431 mutex_unlock(&transport->recv_mutex);
2432}
2433
2434static void xs_wake_pending(struct sock_xprt *transport)
2435{
2436 if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING, &transport->sock_state))
2437 xprt_wake_pending_tasks(&transport->xprt, -EAGAIN);
2438}
2439
2440static void xs_error_handle(struct work_struct *work)
2441{
2442 struct sock_xprt *transport = container_of(work,
2443 struct sock_xprt, error_worker);
2444
2445 xs_wake_disconnect(transport);
2446 xs_wake_write(transport);
2447 xs_wake_error(transport);
2448 xs_wake_pending(transport);
2449}
2450
2451/**
2452 * xs_local_print_stats - display AF_LOCAL socket-specific stats
2453 * @xprt: rpc_xprt struct containing statistics
2454 * @seq: output file
2455 *
2456 */
2457static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2458{
2459 long idle_time = 0;
2460
2461 if (xprt_connected(xprt))
2462 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2463
2464 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2465 "%llu %llu %lu %llu %llu\n",
2466 xprt->stat.bind_count,
2467 xprt->stat.connect_count,
2468 xprt->stat.connect_time / HZ,
2469 idle_time,
2470 xprt->stat.sends,
2471 xprt->stat.recvs,
2472 xprt->stat.bad_xids,
2473 xprt->stat.req_u,
2474 xprt->stat.bklog_u,
2475 xprt->stat.max_slots,
2476 xprt->stat.sending_u,
2477 xprt->stat.pending_u);
2478}
2479
2480/**
2481 * xs_udp_print_stats - display UDP socket-specific stats
2482 * @xprt: rpc_xprt struct containing statistics
2483 * @seq: output file
2484 *
2485 */
2486static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2487{
2488 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2489
2490 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2491 "%lu %llu %llu\n",
2492 transport->srcport,
2493 xprt->stat.bind_count,
2494 xprt->stat.sends,
2495 xprt->stat.recvs,
2496 xprt->stat.bad_xids,
2497 xprt->stat.req_u,
2498 xprt->stat.bklog_u,
2499 xprt->stat.max_slots,
2500 xprt->stat.sending_u,
2501 xprt->stat.pending_u);
2502}
2503
2504/**
2505 * xs_tcp_print_stats - display TCP socket-specific stats
2506 * @xprt: rpc_xprt struct containing statistics
2507 * @seq: output file
2508 *
2509 */
2510static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2511{
2512 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2513 long idle_time = 0;
2514
2515 if (xprt_connected(xprt))
2516 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2517
2518 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2519 "%llu %llu %lu %llu %llu\n",
2520 transport->srcport,
2521 xprt->stat.bind_count,
2522 xprt->stat.connect_count,
2523 xprt->stat.connect_time / HZ,
2524 idle_time,
2525 xprt->stat.sends,
2526 xprt->stat.recvs,
2527 xprt->stat.bad_xids,
2528 xprt->stat.req_u,
2529 xprt->stat.bklog_u,
2530 xprt->stat.max_slots,
2531 xprt->stat.sending_u,
2532 xprt->stat.pending_u);
2533}
2534
2535/*
2536 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2537 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2538 * to use the server side send routines.
2539 */
2540static int bc_malloc(struct rpc_task *task)
2541{
2542 struct rpc_rqst *rqst = task->tk_rqstp;
2543 size_t size = rqst->rq_callsize;
2544 struct page *page;
2545 struct rpc_buffer *buf;
2546
2547 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2548 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2549 size);
2550 return -EINVAL;
2551 }
2552
2553 page = alloc_page(GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN);
2554 if (!page)
2555 return -ENOMEM;
2556
2557 buf = page_address(page);
2558 buf->len = PAGE_SIZE;
2559
2560 rqst->rq_buffer = buf->data;
2561 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2562 return 0;
2563}
2564
2565/*
2566 * Free the space allocated in the bc_alloc routine
2567 */
2568static void bc_free(struct rpc_task *task)
2569{
2570 void *buffer = task->tk_rqstp->rq_buffer;
2571 struct rpc_buffer *buf;
2572
2573 buf = container_of(buffer, struct rpc_buffer, data);
2574 free_page((unsigned long)buf);
2575}
2576
2577static int bc_sendto(struct rpc_rqst *req)
2578{
2579 struct xdr_buf *xdr = &req->rq_snd_buf;
2580 struct sock_xprt *transport =
2581 container_of(req->rq_xprt, struct sock_xprt, xprt);
2582 struct msghdr msg = {
2583 .msg_flags = 0,
2584 };
2585 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
2586 (u32)xdr->len);
2587 unsigned int sent = 0;
2588 int err;
2589
2590 req->rq_xtime = ktime_get();
2591 err = xdr_alloc_bvec(xdr, rpc_task_gfp_mask());
2592 if (err < 0)
2593 return err;
2594 err = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, marker, &sent);
2595 xdr_free_bvec(xdr);
2596 if (err < 0 || sent != (xdr->len + sizeof(marker)))
2597 return -EAGAIN;
2598 return sent;
2599}
2600
2601/**
2602 * bc_send_request - Send a backchannel Call on a TCP socket
2603 * @req: rpc_rqst containing Call message to be sent
2604 *
2605 * xpt_mutex ensures @rqstp's whole message is written to the socket
2606 * without interruption.
2607 *
2608 * Return values:
2609 * %0 if the message was sent successfully
2610 * %ENOTCONN if the message was not sent
2611 */
2612static int bc_send_request(struct rpc_rqst *req)
2613{
2614 struct svc_xprt *xprt;
2615 int len;
2616
2617 /*
2618 * Get the server socket associated with this callback xprt
2619 */
2620 xprt = req->rq_xprt->bc_xprt;
2621
2622 /*
2623 * Grab the mutex to serialize data as the connection is shared
2624 * with the fore channel
2625 */
2626 mutex_lock(&xprt->xpt_mutex);
2627 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2628 len = -ENOTCONN;
2629 else
2630 len = bc_sendto(req);
2631 mutex_unlock(&xprt->xpt_mutex);
2632
2633 if (len > 0)
2634 len = 0;
2635
2636 return len;
2637}
2638
2639/*
2640 * The close routine. Since this is client initiated, we do nothing
2641 */
2642
2643static void bc_close(struct rpc_xprt *xprt)
2644{
2645 xprt_disconnect_done(xprt);
2646}
2647
2648/*
2649 * The xprt destroy routine. Again, because this connection is client
2650 * initiated, we do nothing
2651 */
2652
2653static void bc_destroy(struct rpc_xprt *xprt)
2654{
2655 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2656
2657 xs_xprt_free(xprt);
2658 module_put(THIS_MODULE);
2659}
2660
2661static const struct rpc_xprt_ops xs_local_ops = {
2662 .reserve_xprt = xprt_reserve_xprt,
2663 .release_xprt = xprt_release_xprt,
2664 .alloc_slot = xprt_alloc_slot,
2665 .free_slot = xprt_free_slot,
2666 .rpcbind = xs_local_rpcbind,
2667 .set_port = xs_local_set_port,
2668 .connect = xs_local_connect,
2669 .buf_alloc = rpc_malloc,
2670 .buf_free = rpc_free,
2671 .prepare_request = xs_stream_prepare_request,
2672 .send_request = xs_local_send_request,
2673 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2674 .close = xs_close,
2675 .destroy = xs_destroy,
2676 .print_stats = xs_local_print_stats,
2677 .enable_swap = xs_enable_swap,
2678 .disable_swap = xs_disable_swap,
2679};
2680
2681static const struct rpc_xprt_ops xs_udp_ops = {
2682 .set_buffer_size = xs_udp_set_buffer_size,
2683 .reserve_xprt = xprt_reserve_xprt_cong,
2684 .release_xprt = xprt_release_xprt_cong,
2685 .alloc_slot = xprt_alloc_slot,
2686 .free_slot = xprt_free_slot,
2687 .rpcbind = rpcb_getport_async,
2688 .set_port = xs_set_port,
2689 .connect = xs_connect,
2690 .get_srcaddr = xs_sock_srcaddr,
2691 .get_srcport = xs_sock_srcport,
2692 .buf_alloc = rpc_malloc,
2693 .buf_free = rpc_free,
2694 .send_request = xs_udp_send_request,
2695 .wait_for_reply_request = xprt_wait_for_reply_request_rtt,
2696 .timer = xs_udp_timer,
2697 .release_request = xprt_release_rqst_cong,
2698 .close = xs_close,
2699 .destroy = xs_destroy,
2700 .print_stats = xs_udp_print_stats,
2701 .enable_swap = xs_enable_swap,
2702 .disable_swap = xs_disable_swap,
2703 .inject_disconnect = xs_inject_disconnect,
2704};
2705
2706static const struct rpc_xprt_ops xs_tcp_ops = {
2707 .reserve_xprt = xprt_reserve_xprt,
2708 .release_xprt = xprt_release_xprt,
2709 .alloc_slot = xprt_alloc_slot,
2710 .free_slot = xprt_free_slot,
2711 .rpcbind = rpcb_getport_async,
2712 .set_port = xs_set_port,
2713 .connect = xs_connect,
2714 .get_srcaddr = xs_sock_srcaddr,
2715 .get_srcport = xs_sock_srcport,
2716 .buf_alloc = rpc_malloc,
2717 .buf_free = rpc_free,
2718 .prepare_request = xs_stream_prepare_request,
2719 .send_request = xs_tcp_send_request,
2720 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2721 .close = xs_tcp_shutdown,
2722 .destroy = xs_destroy,
2723 .set_connect_timeout = xs_tcp_set_connect_timeout,
2724 .print_stats = xs_tcp_print_stats,
2725 .enable_swap = xs_enable_swap,
2726 .disable_swap = xs_disable_swap,
2727 .inject_disconnect = xs_inject_disconnect,
2728#ifdef CONFIG_SUNRPC_BACKCHANNEL
2729 .bc_setup = xprt_setup_bc,
2730 .bc_maxpayload = xs_tcp_bc_maxpayload,
2731 .bc_num_slots = xprt_bc_max_slots,
2732 .bc_free_rqst = xprt_free_bc_rqst,
2733 .bc_destroy = xprt_destroy_bc,
2734#endif
2735};
2736
2737/*
2738 * The rpc_xprt_ops for the server backchannel
2739 */
2740
2741static const struct rpc_xprt_ops bc_tcp_ops = {
2742 .reserve_xprt = xprt_reserve_xprt,
2743 .release_xprt = xprt_release_xprt,
2744 .alloc_slot = xprt_alloc_slot,
2745 .free_slot = xprt_free_slot,
2746 .buf_alloc = bc_malloc,
2747 .buf_free = bc_free,
2748 .send_request = bc_send_request,
2749 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2750 .close = bc_close,
2751 .destroy = bc_destroy,
2752 .print_stats = xs_tcp_print_stats,
2753 .enable_swap = xs_enable_swap,
2754 .disable_swap = xs_disable_swap,
2755 .inject_disconnect = xs_inject_disconnect,
2756};
2757
2758static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2759{
2760 static const struct sockaddr_in sin = {
2761 .sin_family = AF_INET,
2762 .sin_addr.s_addr = htonl(INADDR_ANY),
2763 };
2764 static const struct sockaddr_in6 sin6 = {
2765 .sin6_family = AF_INET6,
2766 .sin6_addr = IN6ADDR_ANY_INIT,
2767 };
2768
2769 switch (family) {
2770 case AF_LOCAL:
2771 break;
2772 case AF_INET:
2773 memcpy(sap, &sin, sizeof(sin));
2774 break;
2775 case AF_INET6:
2776 memcpy(sap, &sin6, sizeof(sin6));
2777 break;
2778 default:
2779 dprintk("RPC: %s: Bad address family\n", __func__);
2780 return -EAFNOSUPPORT;
2781 }
2782 return 0;
2783}
2784
2785static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2786 unsigned int slot_table_size,
2787 unsigned int max_slot_table_size)
2788{
2789 struct rpc_xprt *xprt;
2790 struct sock_xprt *new;
2791
2792 if (args->addrlen > sizeof(xprt->addr)) {
2793 dprintk("RPC: xs_setup_xprt: address too large\n");
2794 return ERR_PTR(-EBADF);
2795 }
2796
2797 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2798 max_slot_table_size);
2799 if (xprt == NULL) {
2800 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2801 "rpc_xprt\n");
2802 return ERR_PTR(-ENOMEM);
2803 }
2804
2805 new = container_of(xprt, struct sock_xprt, xprt);
2806 mutex_init(&new->recv_mutex);
2807 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2808 xprt->addrlen = args->addrlen;
2809 if (args->srcaddr)
2810 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2811 else {
2812 int err;
2813 err = xs_init_anyaddr(args->dstaddr->sa_family,
2814 (struct sockaddr *)&new->srcaddr);
2815 if (err != 0) {
2816 xprt_free(xprt);
2817 return ERR_PTR(err);
2818 }
2819 }
2820
2821 return xprt;
2822}
2823
2824static const struct rpc_timeout xs_local_default_timeout = {
2825 .to_initval = 10 * HZ,
2826 .to_maxval = 10 * HZ,
2827 .to_retries = 2,
2828};
2829
2830/**
2831 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2832 * @args: rpc transport creation arguments
2833 *
2834 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2835 */
2836static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2837{
2838 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2839 struct sock_xprt *transport;
2840 struct rpc_xprt *xprt;
2841 struct rpc_xprt *ret;
2842
2843 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2844 xprt_max_tcp_slot_table_entries);
2845 if (IS_ERR(xprt))
2846 return xprt;
2847 transport = container_of(xprt, struct sock_xprt, xprt);
2848
2849 xprt->prot = 0;
2850 xprt->xprt_class = &xs_local_transport;
2851 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2852
2853 xprt->bind_timeout = XS_BIND_TO;
2854 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2855 xprt->idle_timeout = XS_IDLE_DISC_TO;
2856
2857 xprt->ops = &xs_local_ops;
2858 xprt->timeout = &xs_local_default_timeout;
2859
2860 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2861 INIT_WORK(&transport->error_worker, xs_error_handle);
2862 INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket);
2863
2864 switch (sun->sun_family) {
2865 case AF_LOCAL:
2866 if (sun->sun_path[0] != '/') {
2867 dprintk("RPC: bad AF_LOCAL address: %s\n",
2868 sun->sun_path);
2869 ret = ERR_PTR(-EINVAL);
2870 goto out_err;
2871 }
2872 xprt_set_bound(xprt);
2873 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2874 break;
2875 default:
2876 ret = ERR_PTR(-EAFNOSUPPORT);
2877 goto out_err;
2878 }
2879
2880 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2881 xprt->address_strings[RPC_DISPLAY_ADDR]);
2882
2883 if (try_module_get(THIS_MODULE))
2884 return xprt;
2885 ret = ERR_PTR(-EINVAL);
2886out_err:
2887 xs_xprt_free(xprt);
2888 return ret;
2889}
2890
2891static const struct rpc_timeout xs_udp_default_timeout = {
2892 .to_initval = 5 * HZ,
2893 .to_maxval = 30 * HZ,
2894 .to_increment = 5 * HZ,
2895 .to_retries = 5,
2896};
2897
2898/**
2899 * xs_setup_udp - Set up transport to use a UDP socket
2900 * @args: rpc transport creation arguments
2901 *
2902 */
2903static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2904{
2905 struct sockaddr *addr = args->dstaddr;
2906 struct rpc_xprt *xprt;
2907 struct sock_xprt *transport;
2908 struct rpc_xprt *ret;
2909
2910 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2911 xprt_udp_slot_table_entries);
2912 if (IS_ERR(xprt))
2913 return xprt;
2914 transport = container_of(xprt, struct sock_xprt, xprt);
2915
2916 xprt->prot = IPPROTO_UDP;
2917 xprt->xprt_class = &xs_udp_transport;
2918 /* XXX: header size can vary due to auth type, IPv6, etc. */
2919 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2920
2921 xprt->bind_timeout = XS_BIND_TO;
2922 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2923 xprt->idle_timeout = XS_IDLE_DISC_TO;
2924
2925 xprt->ops = &xs_udp_ops;
2926
2927 xprt->timeout = &xs_udp_default_timeout;
2928
2929 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
2930 INIT_WORK(&transport->error_worker, xs_error_handle);
2931 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2932
2933 switch (addr->sa_family) {
2934 case AF_INET:
2935 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2936 xprt_set_bound(xprt);
2937
2938 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2939 break;
2940 case AF_INET6:
2941 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2942 xprt_set_bound(xprt);
2943
2944 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2945 break;
2946 default:
2947 ret = ERR_PTR(-EAFNOSUPPORT);
2948 goto out_err;
2949 }
2950
2951 if (xprt_bound(xprt))
2952 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2953 xprt->address_strings[RPC_DISPLAY_ADDR],
2954 xprt->address_strings[RPC_DISPLAY_PORT],
2955 xprt->address_strings[RPC_DISPLAY_PROTO]);
2956 else
2957 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2958 xprt->address_strings[RPC_DISPLAY_ADDR],
2959 xprt->address_strings[RPC_DISPLAY_PROTO]);
2960
2961 if (try_module_get(THIS_MODULE))
2962 return xprt;
2963 ret = ERR_PTR(-EINVAL);
2964out_err:
2965 xs_xprt_free(xprt);
2966 return ret;
2967}
2968
2969static const struct rpc_timeout xs_tcp_default_timeout = {
2970 .to_initval = 60 * HZ,
2971 .to_maxval = 60 * HZ,
2972 .to_retries = 2,
2973};
2974
2975/**
2976 * xs_setup_tcp - Set up transport to use a TCP socket
2977 * @args: rpc transport creation arguments
2978 *
2979 */
2980static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2981{
2982 struct sockaddr *addr = args->dstaddr;
2983 struct rpc_xprt *xprt;
2984 struct sock_xprt *transport;
2985 struct rpc_xprt *ret;
2986 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2987
2988 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2989 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2990
2991 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2992 max_slot_table_size);
2993 if (IS_ERR(xprt))
2994 return xprt;
2995 transport = container_of(xprt, struct sock_xprt, xprt);
2996
2997 xprt->prot = IPPROTO_TCP;
2998 xprt->xprt_class = &xs_tcp_transport;
2999 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3000
3001 xprt->bind_timeout = XS_BIND_TO;
3002 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3003 xprt->idle_timeout = XS_IDLE_DISC_TO;
3004
3005 xprt->ops = &xs_tcp_ops;
3006 xprt->timeout = &xs_tcp_default_timeout;
3007
3008 xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3009 xprt->connect_timeout = xprt->timeout->to_initval *
3010 (xprt->timeout->to_retries + 1);
3011
3012 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3013 INIT_WORK(&transport->error_worker, xs_error_handle);
3014 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
3015
3016 switch (addr->sa_family) {
3017 case AF_INET:
3018 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3019 xprt_set_bound(xprt);
3020
3021 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3022 break;
3023 case AF_INET6:
3024 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3025 xprt_set_bound(xprt);
3026
3027 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3028 break;
3029 default:
3030 ret = ERR_PTR(-EAFNOSUPPORT);
3031 goto out_err;
3032 }
3033
3034 if (xprt_bound(xprt))
3035 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3036 xprt->address_strings[RPC_DISPLAY_ADDR],
3037 xprt->address_strings[RPC_DISPLAY_PORT],
3038 xprt->address_strings[RPC_DISPLAY_PROTO]);
3039 else
3040 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
3041 xprt->address_strings[RPC_DISPLAY_ADDR],
3042 xprt->address_strings[RPC_DISPLAY_PROTO]);
3043
3044 if (try_module_get(THIS_MODULE))
3045 return xprt;
3046 ret = ERR_PTR(-EINVAL);
3047out_err:
3048 xs_xprt_free(xprt);
3049 return ret;
3050}
3051
3052/**
3053 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3054 * @args: rpc transport creation arguments
3055 *
3056 */
3057static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3058{
3059 struct sockaddr *addr = args->dstaddr;
3060 struct rpc_xprt *xprt;
3061 struct sock_xprt *transport;
3062 struct svc_sock *bc_sock;
3063 struct rpc_xprt *ret;
3064
3065 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3066 xprt_tcp_slot_table_entries);
3067 if (IS_ERR(xprt))
3068 return xprt;
3069 transport = container_of(xprt, struct sock_xprt, xprt);
3070
3071 xprt->prot = IPPROTO_TCP;
3072 xprt->xprt_class = &xs_bc_tcp_transport;
3073 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3074 xprt->timeout = &xs_tcp_default_timeout;
3075
3076 /* backchannel */
3077 xprt_set_bound(xprt);
3078 xprt->bind_timeout = 0;
3079 xprt->reestablish_timeout = 0;
3080 xprt->idle_timeout = 0;
3081
3082 xprt->ops = &bc_tcp_ops;
3083
3084 switch (addr->sa_family) {
3085 case AF_INET:
3086 xs_format_peer_addresses(xprt, "tcp",
3087 RPCBIND_NETID_TCP);
3088 break;
3089 case AF_INET6:
3090 xs_format_peer_addresses(xprt, "tcp",
3091 RPCBIND_NETID_TCP6);
3092 break;
3093 default:
3094 ret = ERR_PTR(-EAFNOSUPPORT);
3095 goto out_err;
3096 }
3097
3098 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3099 xprt->address_strings[RPC_DISPLAY_ADDR],
3100 xprt->address_strings[RPC_DISPLAY_PORT],
3101 xprt->address_strings[RPC_DISPLAY_PROTO]);
3102
3103 /*
3104 * Once we've associated a backchannel xprt with a connection,
3105 * we want to keep it around as long as the connection lasts,
3106 * in case we need to start using it for a backchannel again;
3107 * this reference won't be dropped until bc_xprt is destroyed.
3108 */
3109 xprt_get(xprt);
3110 args->bc_xprt->xpt_bc_xprt = xprt;
3111 xprt->bc_xprt = args->bc_xprt;
3112 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3113 transport->sock = bc_sock->sk_sock;
3114 transport->inet = bc_sock->sk_sk;
3115
3116 /*
3117 * Since we don't want connections for the backchannel, we set
3118 * the xprt status to connected
3119 */
3120 xprt_set_connected(xprt);
3121
3122 if (try_module_get(THIS_MODULE))
3123 return xprt;
3124
3125 args->bc_xprt->xpt_bc_xprt = NULL;
3126 args->bc_xprt->xpt_bc_xps = NULL;
3127 xprt_put(xprt);
3128 ret = ERR_PTR(-EINVAL);
3129out_err:
3130 xs_xprt_free(xprt);
3131 return ret;
3132}
3133
3134static struct xprt_class xs_local_transport = {
3135 .list = LIST_HEAD_INIT(xs_local_transport.list),
3136 .name = "named UNIX socket",
3137 .owner = THIS_MODULE,
3138 .ident = XPRT_TRANSPORT_LOCAL,
3139 .setup = xs_setup_local,
3140 .netid = { "" },
3141};
3142
3143static struct xprt_class xs_udp_transport = {
3144 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3145 .name = "udp",
3146 .owner = THIS_MODULE,
3147 .ident = XPRT_TRANSPORT_UDP,
3148 .setup = xs_setup_udp,
3149 .netid = { "udp", "udp6", "" },
3150};
3151
3152static struct xprt_class xs_tcp_transport = {
3153 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3154 .name = "tcp",
3155 .owner = THIS_MODULE,
3156 .ident = XPRT_TRANSPORT_TCP,
3157 .setup = xs_setup_tcp,
3158 .netid = { "tcp", "tcp6", "" },
3159};
3160
3161static struct xprt_class xs_bc_tcp_transport = {
3162 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3163 .name = "tcp NFSv4.1 backchannel",
3164 .owner = THIS_MODULE,
3165 .ident = XPRT_TRANSPORT_BC_TCP,
3166 .setup = xs_setup_bc_tcp,
3167 .netid = { "" },
3168};
3169
3170/**
3171 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3172 *
3173 */
3174int init_socket_xprt(void)
3175{
3176 if (!sunrpc_table_header)
3177 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3178
3179 xprt_register_transport(&xs_local_transport);
3180 xprt_register_transport(&xs_udp_transport);
3181 xprt_register_transport(&xs_tcp_transport);
3182 xprt_register_transport(&xs_bc_tcp_transport);
3183
3184 return 0;
3185}
3186
3187/**
3188 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3189 *
3190 */
3191void cleanup_socket_xprt(void)
3192{
3193 if (sunrpc_table_header) {
3194 unregister_sysctl_table(sunrpc_table_header);
3195 sunrpc_table_header = NULL;
3196 }
3197
3198 xprt_unregister_transport(&xs_local_transport);
3199 xprt_unregister_transport(&xs_udp_transport);
3200 xprt_unregister_transport(&xs_tcp_transport);
3201 xprt_unregister_transport(&xs_bc_tcp_transport);
3202}
3203
3204static int param_set_portnr(const char *val, const struct kernel_param *kp)
3205{
3206 return param_set_uint_minmax(val, kp,
3207 RPC_MIN_RESVPORT,
3208 RPC_MAX_RESVPORT);
3209}
3210
3211static const struct kernel_param_ops param_ops_portnr = {
3212 .set = param_set_portnr,
3213 .get = param_get_uint,
3214};
3215
3216#define param_check_portnr(name, p) \
3217 __param_check(name, p, unsigned int);
3218
3219module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3220module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3221
3222static int param_set_slot_table_size(const char *val,
3223 const struct kernel_param *kp)
3224{
3225 return param_set_uint_minmax(val, kp,
3226 RPC_MIN_SLOT_TABLE,
3227 RPC_MAX_SLOT_TABLE);
3228}
3229
3230static const struct kernel_param_ops param_ops_slot_table_size = {
3231 .set = param_set_slot_table_size,
3232 .get = param_get_uint,
3233};
3234
3235#define param_check_slot_table_size(name, p) \
3236 __param_check(name, p, unsigned int);
3237
3238static int param_set_max_slot_table_size(const char *val,
3239 const struct kernel_param *kp)
3240{
3241 return param_set_uint_minmax(val, kp,
3242 RPC_MIN_SLOT_TABLE,
3243 RPC_MAX_SLOT_TABLE_LIMIT);
3244}
3245
3246static const struct kernel_param_ops param_ops_max_slot_table_size = {
3247 .set = param_set_max_slot_table_size,
3248 .get = param_get_uint,
3249};
3250
3251#define param_check_max_slot_table_size(name, p) \
3252 __param_check(name, p, unsigned int);
3253
3254module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3255 slot_table_size, 0644);
3256module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3257 max_slot_table_size, 0644);
3258module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3259 slot_table_size, 0644);
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/net/sunrpc/xprtsock.c
4 *
5 * Client-side transport implementation for sockets.
6 *
7 * TCP callback races fixes (C) 1998 Red Hat
8 * TCP send fixes (C) 1998 Red Hat
9 * TCP NFS related read + write fixes
10 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
11 *
12 * Rewrite of larges part of the code in order to stabilize TCP stuff.
13 * Fix behaviour when socket buffer is full.
14 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
15 *
16 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
17 *
18 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
19 * <gilles.quillard@bull.net>
20 */
21
22#include <linux/types.h>
23#include <linux/string.h>
24#include <linux/slab.h>
25#include <linux/module.h>
26#include <linux/capability.h>
27#include <linux/pagemap.h>
28#include <linux/errno.h>
29#include <linux/socket.h>
30#include <linux/in.h>
31#include <linux/net.h>
32#include <linux/mm.h>
33#include <linux/un.h>
34#include <linux/udp.h>
35#include <linux/tcp.h>
36#include <linux/sunrpc/clnt.h>
37#include <linux/sunrpc/addr.h>
38#include <linux/sunrpc/sched.h>
39#include <linux/sunrpc/svcsock.h>
40#include <linux/sunrpc/xprtsock.h>
41#include <linux/file.h>
42#ifdef CONFIG_SUNRPC_BACKCHANNEL
43#include <linux/sunrpc/bc_xprt.h>
44#endif
45
46#include <net/sock.h>
47#include <net/checksum.h>
48#include <net/udp.h>
49#include <net/tcp.h>
50#include <linux/bvec.h>
51#include <linux/highmem.h>
52#include <linux/uio.h>
53#include <linux/sched/mm.h>
54
55#include <trace/events/sunrpc.h>
56
57#include "socklib.h"
58#include "sunrpc.h"
59
60static void xs_close(struct rpc_xprt *xprt);
61static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
62 struct socket *sock);
63
64/*
65 * xprtsock tunables
66 */
67static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
68static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
69static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
70
71static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
72static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
73
74#define XS_TCP_LINGER_TO (15U * HZ)
75static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
76
77/*
78 * We can register our own files under /proc/sys/sunrpc by
79 * calling register_sysctl_table() again. The files in that
80 * directory become the union of all files registered there.
81 *
82 * We simply need to make sure that we don't collide with
83 * someone else's file names!
84 */
85
86static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
87static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
88static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
89static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
90static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
91
92static struct ctl_table_header *sunrpc_table_header;
93
94/*
95 * FIXME: changing the UDP slot table size should also resize the UDP
96 * socket buffers for existing UDP transports
97 */
98static struct ctl_table xs_tunables_table[] = {
99 {
100 .procname = "udp_slot_table_entries",
101 .data = &xprt_udp_slot_table_entries,
102 .maxlen = sizeof(unsigned int),
103 .mode = 0644,
104 .proc_handler = proc_dointvec_minmax,
105 .extra1 = &min_slot_table_size,
106 .extra2 = &max_slot_table_size
107 },
108 {
109 .procname = "tcp_slot_table_entries",
110 .data = &xprt_tcp_slot_table_entries,
111 .maxlen = sizeof(unsigned int),
112 .mode = 0644,
113 .proc_handler = proc_dointvec_minmax,
114 .extra1 = &min_slot_table_size,
115 .extra2 = &max_slot_table_size
116 },
117 {
118 .procname = "tcp_max_slot_table_entries",
119 .data = &xprt_max_tcp_slot_table_entries,
120 .maxlen = sizeof(unsigned int),
121 .mode = 0644,
122 .proc_handler = proc_dointvec_minmax,
123 .extra1 = &min_slot_table_size,
124 .extra2 = &max_tcp_slot_table_limit
125 },
126 {
127 .procname = "min_resvport",
128 .data = &xprt_min_resvport,
129 .maxlen = sizeof(unsigned int),
130 .mode = 0644,
131 .proc_handler = proc_dointvec_minmax,
132 .extra1 = &xprt_min_resvport_limit,
133 .extra2 = &xprt_max_resvport_limit
134 },
135 {
136 .procname = "max_resvport",
137 .data = &xprt_max_resvport,
138 .maxlen = sizeof(unsigned int),
139 .mode = 0644,
140 .proc_handler = proc_dointvec_minmax,
141 .extra1 = &xprt_min_resvport_limit,
142 .extra2 = &xprt_max_resvport_limit
143 },
144 {
145 .procname = "tcp_fin_timeout",
146 .data = &xs_tcp_fin_timeout,
147 .maxlen = sizeof(xs_tcp_fin_timeout),
148 .mode = 0644,
149 .proc_handler = proc_dointvec_jiffies,
150 },
151 { },
152};
153
154static struct ctl_table sunrpc_table[] = {
155 {
156 .procname = "sunrpc",
157 .mode = 0555,
158 .child = xs_tunables_table
159 },
160 { },
161};
162
163/*
164 * Wait duration for a reply from the RPC portmapper.
165 */
166#define XS_BIND_TO (60U * HZ)
167
168/*
169 * Delay if a UDP socket connect error occurs. This is most likely some
170 * kind of resource problem on the local host.
171 */
172#define XS_UDP_REEST_TO (2U * HZ)
173
174/*
175 * The reestablish timeout allows clients to delay for a bit before attempting
176 * to reconnect to a server that just dropped our connection.
177 *
178 * We implement an exponential backoff when trying to reestablish a TCP
179 * transport connection with the server. Some servers like to drop a TCP
180 * connection when they are overworked, so we start with a short timeout and
181 * increase over time if the server is down or not responding.
182 */
183#define XS_TCP_INIT_REEST_TO (3U * HZ)
184
185/*
186 * TCP idle timeout; client drops the transport socket if it is idle
187 * for this long. Note that we also timeout UDP sockets to prevent
188 * holding port numbers when there is no RPC traffic.
189 */
190#define XS_IDLE_DISC_TO (5U * 60 * HZ)
191
192#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
193# undef RPC_DEBUG_DATA
194# define RPCDBG_FACILITY RPCDBG_TRANS
195#endif
196
197#ifdef RPC_DEBUG_DATA
198static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
199{
200 u8 *buf = (u8 *) packet;
201 int j;
202
203 dprintk("RPC: %s\n", msg);
204 for (j = 0; j < count && j < 128; j += 4) {
205 if (!(j & 31)) {
206 if (j)
207 dprintk("\n");
208 dprintk("0x%04x ", j);
209 }
210 dprintk("%02x%02x%02x%02x ",
211 buf[j], buf[j+1], buf[j+2], buf[j+3]);
212 }
213 dprintk("\n");
214}
215#else
216static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
217{
218 /* NOP */
219}
220#endif
221
222static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
223{
224 return (struct rpc_xprt *) sk->sk_user_data;
225}
226
227static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
228{
229 return (struct sockaddr *) &xprt->addr;
230}
231
232static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
233{
234 return (struct sockaddr_un *) &xprt->addr;
235}
236
237static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
238{
239 return (struct sockaddr_in *) &xprt->addr;
240}
241
242static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
243{
244 return (struct sockaddr_in6 *) &xprt->addr;
245}
246
247static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
248{
249 struct sockaddr *sap = xs_addr(xprt);
250 struct sockaddr_in6 *sin6;
251 struct sockaddr_in *sin;
252 struct sockaddr_un *sun;
253 char buf[128];
254
255 switch (sap->sa_family) {
256 case AF_LOCAL:
257 sun = xs_addr_un(xprt);
258 strlcpy(buf, sun->sun_path, sizeof(buf));
259 xprt->address_strings[RPC_DISPLAY_ADDR] =
260 kstrdup(buf, GFP_KERNEL);
261 break;
262 case AF_INET:
263 (void)rpc_ntop(sap, buf, sizeof(buf));
264 xprt->address_strings[RPC_DISPLAY_ADDR] =
265 kstrdup(buf, GFP_KERNEL);
266 sin = xs_addr_in(xprt);
267 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
268 break;
269 case AF_INET6:
270 (void)rpc_ntop(sap, buf, sizeof(buf));
271 xprt->address_strings[RPC_DISPLAY_ADDR] =
272 kstrdup(buf, GFP_KERNEL);
273 sin6 = xs_addr_in6(xprt);
274 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
275 break;
276 default:
277 BUG();
278 }
279
280 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
281}
282
283static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
284{
285 struct sockaddr *sap = xs_addr(xprt);
286 char buf[128];
287
288 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
289 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
290
291 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
292 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
293}
294
295static void xs_format_peer_addresses(struct rpc_xprt *xprt,
296 const char *protocol,
297 const char *netid)
298{
299 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
300 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
301 xs_format_common_peer_addresses(xprt);
302 xs_format_common_peer_ports(xprt);
303}
304
305static void xs_update_peer_port(struct rpc_xprt *xprt)
306{
307 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
308 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
309
310 xs_format_common_peer_ports(xprt);
311}
312
313static void xs_free_peer_addresses(struct rpc_xprt *xprt)
314{
315 unsigned int i;
316
317 for (i = 0; i < RPC_DISPLAY_MAX; i++)
318 switch (i) {
319 case RPC_DISPLAY_PROTO:
320 case RPC_DISPLAY_NETID:
321 continue;
322 default:
323 kfree(xprt->address_strings[i]);
324 }
325}
326
327static size_t
328xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp)
329{
330 size_t i,n;
331
332 if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES))
333 return want;
334 n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT;
335 for (i = 0; i < n; i++) {
336 if (buf->pages[i])
337 continue;
338 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp);
339 if (!buf->pages[i]) {
340 i *= PAGE_SIZE;
341 return i > buf->page_base ? i - buf->page_base : 0;
342 }
343 }
344 return want;
345}
346
347static ssize_t
348xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek)
349{
350 ssize_t ret;
351 if (seek != 0)
352 iov_iter_advance(&msg->msg_iter, seek);
353 ret = sock_recvmsg(sock, msg, flags);
354 return ret > 0 ? ret + seek : ret;
355}
356
357static ssize_t
358xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags,
359 struct kvec *kvec, size_t count, size_t seek)
360{
361 iov_iter_kvec(&msg->msg_iter, READ, kvec, 1, count);
362 return xs_sock_recvmsg(sock, msg, flags, seek);
363}
364
365static ssize_t
366xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags,
367 struct bio_vec *bvec, unsigned long nr, size_t count,
368 size_t seek)
369{
370 iov_iter_bvec(&msg->msg_iter, READ, bvec, nr, count);
371 return xs_sock_recvmsg(sock, msg, flags, seek);
372}
373
374static ssize_t
375xs_read_discard(struct socket *sock, struct msghdr *msg, int flags,
376 size_t count)
377{
378 iov_iter_discard(&msg->msg_iter, READ, count);
379 return sock_recvmsg(sock, msg, flags);
380}
381
382#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
383static void
384xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
385{
386 struct bvec_iter bi = {
387 .bi_size = count,
388 };
389 struct bio_vec bv;
390
391 bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
392 for_each_bvec(bv, bvec, bi, bi)
393 flush_dcache_page(bv.bv_page);
394}
395#else
396static inline void
397xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
398{
399}
400#endif
401
402static ssize_t
403xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
404 struct xdr_buf *buf, size_t count, size_t seek, size_t *read)
405{
406 size_t want, seek_init = seek, offset = 0;
407 ssize_t ret;
408
409 want = min_t(size_t, count, buf->head[0].iov_len);
410 if (seek < want) {
411 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek);
412 if (ret <= 0)
413 goto sock_err;
414 offset += ret;
415 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
416 goto out;
417 if (ret != want)
418 goto out;
419 seek = 0;
420 } else {
421 seek -= want;
422 offset += want;
423 }
424
425 want = xs_alloc_sparse_pages(buf,
426 min_t(size_t, count - offset, buf->page_len),
427 GFP_KERNEL);
428 if (seek < want) {
429 ret = xs_read_bvec(sock, msg, flags, buf->bvec,
430 xdr_buf_pagecount(buf),
431 want + buf->page_base,
432 seek + buf->page_base);
433 if (ret <= 0)
434 goto sock_err;
435 xs_flush_bvec(buf->bvec, ret, seek + buf->page_base);
436 offset += ret - buf->page_base;
437 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
438 goto out;
439 if (ret != want)
440 goto out;
441 seek = 0;
442 } else {
443 seek -= want;
444 offset += want;
445 }
446
447 want = min_t(size_t, count - offset, buf->tail[0].iov_len);
448 if (seek < want) {
449 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek);
450 if (ret <= 0)
451 goto sock_err;
452 offset += ret;
453 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
454 goto out;
455 if (ret != want)
456 goto out;
457 } else if (offset < seek_init)
458 offset = seek_init;
459 ret = -EMSGSIZE;
460out:
461 *read = offset - seek_init;
462 return ret;
463sock_err:
464 offset += seek;
465 goto out;
466}
467
468static void
469xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf)
470{
471 if (!transport->recv.copied) {
472 if (buf->head[0].iov_len >= transport->recv.offset)
473 memcpy(buf->head[0].iov_base,
474 &transport->recv.xid,
475 transport->recv.offset);
476 transport->recv.copied = transport->recv.offset;
477 }
478}
479
480static bool
481xs_read_stream_request_done(struct sock_xprt *transport)
482{
483 return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT);
484}
485
486static void
487xs_read_stream_check_eor(struct sock_xprt *transport,
488 struct msghdr *msg)
489{
490 if (xs_read_stream_request_done(transport))
491 msg->msg_flags |= MSG_EOR;
492}
493
494static ssize_t
495xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg,
496 int flags, struct rpc_rqst *req)
497{
498 struct xdr_buf *buf = &req->rq_private_buf;
499 size_t want, read;
500 ssize_t ret;
501
502 xs_read_header(transport, buf);
503
504 want = transport->recv.len - transport->recv.offset;
505 if (want != 0) {
506 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf,
507 transport->recv.copied + want,
508 transport->recv.copied,
509 &read);
510 transport->recv.offset += read;
511 transport->recv.copied += read;
512 }
513
514 if (transport->recv.offset == transport->recv.len)
515 xs_read_stream_check_eor(transport, msg);
516
517 if (want == 0)
518 return 0;
519
520 switch (ret) {
521 default:
522 break;
523 case -EFAULT:
524 case -EMSGSIZE:
525 msg->msg_flags |= MSG_TRUNC;
526 return read;
527 case 0:
528 return -ESHUTDOWN;
529 }
530 return ret < 0 ? ret : read;
531}
532
533static size_t
534xs_read_stream_headersize(bool isfrag)
535{
536 if (isfrag)
537 return sizeof(__be32);
538 return 3 * sizeof(__be32);
539}
540
541static ssize_t
542xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg,
543 int flags, size_t want, size_t seek)
544{
545 struct kvec kvec = {
546 .iov_base = &transport->recv.fraghdr,
547 .iov_len = want,
548 };
549 return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek);
550}
551
552#if defined(CONFIG_SUNRPC_BACKCHANNEL)
553static ssize_t
554xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
555{
556 struct rpc_xprt *xprt = &transport->xprt;
557 struct rpc_rqst *req;
558 ssize_t ret;
559
560 /* Look up and lock the request corresponding to the given XID */
561 req = xprt_lookup_bc_request(xprt, transport->recv.xid);
562 if (!req) {
563 printk(KERN_WARNING "Callback slot table overflowed\n");
564 return -ESHUTDOWN;
565 }
566 if (transport->recv.copied && !req->rq_private_buf.len)
567 return -ESHUTDOWN;
568
569 ret = xs_read_stream_request(transport, msg, flags, req);
570 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
571 xprt_complete_bc_request(req, transport->recv.copied);
572 else
573 req->rq_private_buf.len = transport->recv.copied;
574
575 return ret;
576}
577#else /* CONFIG_SUNRPC_BACKCHANNEL */
578static ssize_t
579xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
580{
581 return -ESHUTDOWN;
582}
583#endif /* CONFIG_SUNRPC_BACKCHANNEL */
584
585static ssize_t
586xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags)
587{
588 struct rpc_xprt *xprt = &transport->xprt;
589 struct rpc_rqst *req;
590 ssize_t ret = 0;
591
592 /* Look up and lock the request corresponding to the given XID */
593 spin_lock(&xprt->queue_lock);
594 req = xprt_lookup_rqst(xprt, transport->recv.xid);
595 if (!req || (transport->recv.copied && !req->rq_private_buf.len)) {
596 msg->msg_flags |= MSG_TRUNC;
597 goto out;
598 }
599 xprt_pin_rqst(req);
600 spin_unlock(&xprt->queue_lock);
601
602 ret = xs_read_stream_request(transport, msg, flags, req);
603
604 spin_lock(&xprt->queue_lock);
605 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
606 xprt_complete_rqst(req->rq_task, transport->recv.copied);
607 else
608 req->rq_private_buf.len = transport->recv.copied;
609 xprt_unpin_rqst(req);
610out:
611 spin_unlock(&xprt->queue_lock);
612 return ret;
613}
614
615static ssize_t
616xs_read_stream(struct sock_xprt *transport, int flags)
617{
618 struct msghdr msg = { 0 };
619 size_t want, read = 0;
620 ssize_t ret = 0;
621
622 if (transport->recv.len == 0) {
623 want = xs_read_stream_headersize(transport->recv.copied != 0);
624 ret = xs_read_stream_header(transport, &msg, flags, want,
625 transport->recv.offset);
626 if (ret <= 0)
627 goto out_err;
628 transport->recv.offset = ret;
629 if (transport->recv.offset != want)
630 return transport->recv.offset;
631 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) &
632 RPC_FRAGMENT_SIZE_MASK;
633 transport->recv.offset -= sizeof(transport->recv.fraghdr);
634 read = ret;
635 }
636
637 switch (be32_to_cpu(transport->recv.calldir)) {
638 default:
639 msg.msg_flags |= MSG_TRUNC;
640 break;
641 case RPC_CALL:
642 ret = xs_read_stream_call(transport, &msg, flags);
643 break;
644 case RPC_REPLY:
645 ret = xs_read_stream_reply(transport, &msg, flags);
646 }
647 if (msg.msg_flags & MSG_TRUNC) {
648 transport->recv.calldir = cpu_to_be32(-1);
649 transport->recv.copied = -1;
650 }
651 if (ret < 0)
652 goto out_err;
653 read += ret;
654 if (transport->recv.offset < transport->recv.len) {
655 if (!(msg.msg_flags & MSG_TRUNC))
656 return read;
657 msg.msg_flags = 0;
658 ret = xs_read_discard(transport->sock, &msg, flags,
659 transport->recv.len - transport->recv.offset);
660 if (ret <= 0)
661 goto out_err;
662 transport->recv.offset += ret;
663 read += ret;
664 if (transport->recv.offset != transport->recv.len)
665 return read;
666 }
667 if (xs_read_stream_request_done(transport)) {
668 trace_xs_stream_read_request(transport);
669 transport->recv.copied = 0;
670 }
671 transport->recv.offset = 0;
672 transport->recv.len = 0;
673 return read;
674out_err:
675 return ret != 0 ? ret : -ESHUTDOWN;
676}
677
678static __poll_t xs_poll_socket(struct sock_xprt *transport)
679{
680 return transport->sock->ops->poll(transport->file, transport->sock,
681 NULL);
682}
683
684static bool xs_poll_socket_readable(struct sock_xprt *transport)
685{
686 __poll_t events = xs_poll_socket(transport);
687
688 return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP);
689}
690
691static void xs_poll_check_readable(struct sock_xprt *transport)
692{
693
694 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
695 if (!xs_poll_socket_readable(transport))
696 return;
697 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
698 queue_work(xprtiod_workqueue, &transport->recv_worker);
699}
700
701static void xs_stream_data_receive(struct sock_xprt *transport)
702{
703 size_t read = 0;
704 ssize_t ret = 0;
705
706 mutex_lock(&transport->recv_mutex);
707 if (transport->sock == NULL)
708 goto out;
709 for (;;) {
710 ret = xs_read_stream(transport, MSG_DONTWAIT);
711 if (ret < 0)
712 break;
713 read += ret;
714 cond_resched();
715 }
716 if (ret == -ESHUTDOWN)
717 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
718 else
719 xs_poll_check_readable(transport);
720out:
721 mutex_unlock(&transport->recv_mutex);
722 trace_xs_stream_read_data(&transport->xprt, ret, read);
723}
724
725static void xs_stream_data_receive_workfn(struct work_struct *work)
726{
727 struct sock_xprt *transport =
728 container_of(work, struct sock_xprt, recv_worker);
729 unsigned int pflags = memalloc_nofs_save();
730
731 xs_stream_data_receive(transport);
732 memalloc_nofs_restore(pflags);
733}
734
735static void
736xs_stream_reset_connect(struct sock_xprt *transport)
737{
738 transport->recv.offset = 0;
739 transport->recv.len = 0;
740 transport->recv.copied = 0;
741 transport->xmit.offset = 0;
742}
743
744static void
745xs_stream_start_connect(struct sock_xprt *transport)
746{
747 transport->xprt.stat.connect_count++;
748 transport->xprt.stat.connect_start = jiffies;
749}
750
751#define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
752
753/**
754 * xs_nospace - handle transmit was incomplete
755 * @req: pointer to RPC request
756 *
757 */
758static int xs_nospace(struct rpc_rqst *req)
759{
760 struct rpc_xprt *xprt = req->rq_xprt;
761 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
762 struct sock *sk = transport->inet;
763 int ret = -EAGAIN;
764
765 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
766 req->rq_task->tk_pid,
767 req->rq_slen - transport->xmit.offset,
768 req->rq_slen);
769
770 /* Protect against races with write_space */
771 spin_lock(&xprt->transport_lock);
772
773 /* Don't race with disconnect */
774 if (xprt_connected(xprt)) {
775 /* wait for more buffer space */
776 sk->sk_write_pending++;
777 xprt_wait_for_buffer_space(xprt);
778 } else
779 ret = -ENOTCONN;
780
781 spin_unlock(&xprt->transport_lock);
782
783 /* Race breaker in case memory is freed before above code is called */
784 if (ret == -EAGAIN) {
785 struct socket_wq *wq;
786
787 rcu_read_lock();
788 wq = rcu_dereference(sk->sk_wq);
789 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
790 rcu_read_unlock();
791
792 sk->sk_write_space(sk);
793 }
794 return ret;
795}
796
797static void
798xs_stream_prepare_request(struct rpc_rqst *req)
799{
800 xdr_free_bvec(&req->rq_rcv_buf);
801 req->rq_task->tk_status = xdr_alloc_bvec(&req->rq_rcv_buf, GFP_KERNEL);
802}
803
804/*
805 * Determine if the previous message in the stream was aborted before it
806 * could complete transmission.
807 */
808static bool
809xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req)
810{
811 return transport->xmit.offset != 0 && req->rq_bytes_sent == 0;
812}
813
814/*
815 * Return the stream record marker field for a record of length < 2^31-1
816 */
817static rpc_fraghdr
818xs_stream_record_marker(struct xdr_buf *xdr)
819{
820 if (!xdr->len)
821 return 0;
822 return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len);
823}
824
825/**
826 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
827 * @req: pointer to RPC request
828 *
829 * Return values:
830 * 0: The request has been sent
831 * EAGAIN: The socket was blocked, please call again later to
832 * complete the request
833 * ENOTCONN: Caller needs to invoke connect logic then call again
834 * other: Some other error occured, the request was not sent
835 */
836static int xs_local_send_request(struct rpc_rqst *req)
837{
838 struct rpc_xprt *xprt = req->rq_xprt;
839 struct sock_xprt *transport =
840 container_of(xprt, struct sock_xprt, xprt);
841 struct xdr_buf *xdr = &req->rq_snd_buf;
842 rpc_fraghdr rm = xs_stream_record_marker(xdr);
843 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
844 struct msghdr msg = {
845 .msg_flags = XS_SENDMSG_FLAGS,
846 };
847 unsigned int sent;
848 int status;
849
850 /* Close the stream if the previous transmission was incomplete */
851 if (xs_send_request_was_aborted(transport, req)) {
852 xs_close(xprt);
853 return -ENOTCONN;
854 }
855
856 xs_pktdump("packet data:",
857 req->rq_svec->iov_base, req->rq_svec->iov_len);
858
859 req->rq_xtime = ktime_get();
860 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
861 transport->xmit.offset, rm, &sent);
862 dprintk("RPC: %s(%u) = %d\n",
863 __func__, xdr->len - transport->xmit.offset, status);
864
865 if (status == -EAGAIN && sock_writeable(transport->inet))
866 status = -ENOBUFS;
867
868 if (likely(sent > 0) || status == 0) {
869 transport->xmit.offset += sent;
870 req->rq_bytes_sent = transport->xmit.offset;
871 if (likely(req->rq_bytes_sent >= msglen)) {
872 req->rq_xmit_bytes_sent += transport->xmit.offset;
873 transport->xmit.offset = 0;
874 return 0;
875 }
876 status = -EAGAIN;
877 }
878
879 switch (status) {
880 case -ENOBUFS:
881 break;
882 case -EAGAIN:
883 status = xs_nospace(req);
884 break;
885 default:
886 dprintk("RPC: sendmsg returned unrecognized error %d\n",
887 -status);
888 fallthrough;
889 case -EPIPE:
890 xs_close(xprt);
891 status = -ENOTCONN;
892 }
893
894 return status;
895}
896
897/**
898 * xs_udp_send_request - write an RPC request to a UDP socket
899 * @req: pointer to RPC request
900 *
901 * Return values:
902 * 0: The request has been sent
903 * EAGAIN: The socket was blocked, please call again later to
904 * complete the request
905 * ENOTCONN: Caller needs to invoke connect logic then call again
906 * other: Some other error occurred, the request was not sent
907 */
908static int xs_udp_send_request(struct rpc_rqst *req)
909{
910 struct rpc_xprt *xprt = req->rq_xprt;
911 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
912 struct xdr_buf *xdr = &req->rq_snd_buf;
913 struct msghdr msg = {
914 .msg_name = xs_addr(xprt),
915 .msg_namelen = xprt->addrlen,
916 .msg_flags = XS_SENDMSG_FLAGS,
917 };
918 unsigned int sent;
919 int status;
920
921 xs_pktdump("packet data:",
922 req->rq_svec->iov_base,
923 req->rq_svec->iov_len);
924
925 if (!xprt_bound(xprt))
926 return -ENOTCONN;
927
928 if (!xprt_request_get_cong(xprt, req))
929 return -EBADSLT;
930
931 req->rq_xtime = ktime_get();
932 status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, 0, &sent);
933
934 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
935 xdr->len, status);
936
937 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
938 if (status == -EPERM)
939 goto process_status;
940
941 if (status == -EAGAIN && sock_writeable(transport->inet))
942 status = -ENOBUFS;
943
944 if (sent > 0 || status == 0) {
945 req->rq_xmit_bytes_sent += sent;
946 if (sent >= req->rq_slen)
947 return 0;
948 /* Still some bytes left; set up for a retry later. */
949 status = -EAGAIN;
950 }
951
952process_status:
953 switch (status) {
954 case -ENOTSOCK:
955 status = -ENOTCONN;
956 /* Should we call xs_close() here? */
957 break;
958 case -EAGAIN:
959 status = xs_nospace(req);
960 break;
961 case -ENETUNREACH:
962 case -ENOBUFS:
963 case -EPIPE:
964 case -ECONNREFUSED:
965 case -EPERM:
966 /* When the server has died, an ICMP port unreachable message
967 * prompts ECONNREFUSED. */
968 break;
969 default:
970 dprintk("RPC: sendmsg returned unrecognized error %d\n",
971 -status);
972 }
973
974 return status;
975}
976
977/**
978 * xs_tcp_send_request - write an RPC request to a TCP socket
979 * @req: pointer to RPC request
980 *
981 * Return values:
982 * 0: The request has been sent
983 * EAGAIN: The socket was blocked, please call again later to
984 * complete the request
985 * ENOTCONN: Caller needs to invoke connect logic then call again
986 * other: Some other error occurred, the request was not sent
987 *
988 * XXX: In the case of soft timeouts, should we eventually give up
989 * if sendmsg is not able to make progress?
990 */
991static int xs_tcp_send_request(struct rpc_rqst *req)
992{
993 struct rpc_xprt *xprt = req->rq_xprt;
994 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
995 struct xdr_buf *xdr = &req->rq_snd_buf;
996 rpc_fraghdr rm = xs_stream_record_marker(xdr);
997 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
998 struct msghdr msg = {
999 .msg_flags = XS_SENDMSG_FLAGS,
1000 };
1001 bool vm_wait = false;
1002 unsigned int sent;
1003 int status;
1004
1005 /* Close the stream if the previous transmission was incomplete */
1006 if (xs_send_request_was_aborted(transport, req)) {
1007 if (transport->sock != NULL)
1008 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
1009 return -ENOTCONN;
1010 }
1011
1012 xs_pktdump("packet data:",
1013 req->rq_svec->iov_base,
1014 req->rq_svec->iov_len);
1015
1016 if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
1017 xs_tcp_set_socket_timeouts(xprt, transport->sock);
1018
1019 /* Continue transmitting the packet/record. We must be careful
1020 * to cope with writespace callbacks arriving _after_ we have
1021 * called sendmsg(). */
1022 req->rq_xtime = ktime_get();
1023 while (1) {
1024 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
1025 transport->xmit.offset, rm, &sent);
1026
1027 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
1028 xdr->len - transport->xmit.offset, status);
1029
1030 /* If we've sent the entire packet, immediately
1031 * reset the count of bytes sent. */
1032 transport->xmit.offset += sent;
1033 req->rq_bytes_sent = transport->xmit.offset;
1034 if (likely(req->rq_bytes_sent >= msglen)) {
1035 req->rq_xmit_bytes_sent += transport->xmit.offset;
1036 transport->xmit.offset = 0;
1037 return 0;
1038 }
1039
1040 WARN_ON_ONCE(sent == 0 && status == 0);
1041
1042 if (status == -EAGAIN ) {
1043 /*
1044 * Return EAGAIN if we're sure we're hitting the
1045 * socket send buffer limits.
1046 */
1047 if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
1048 break;
1049 /*
1050 * Did we hit a memory allocation failure?
1051 */
1052 if (sent == 0) {
1053 status = -ENOBUFS;
1054 if (vm_wait)
1055 break;
1056 /* Retry, knowing now that we're below the
1057 * socket send buffer limit
1058 */
1059 vm_wait = true;
1060 }
1061 continue;
1062 }
1063 if (status < 0)
1064 break;
1065 vm_wait = false;
1066 }
1067
1068 switch (status) {
1069 case -ENOTSOCK:
1070 status = -ENOTCONN;
1071 /* Should we call xs_close() here? */
1072 break;
1073 case -EAGAIN:
1074 status = xs_nospace(req);
1075 break;
1076 case -ECONNRESET:
1077 case -ECONNREFUSED:
1078 case -ENOTCONN:
1079 case -EADDRINUSE:
1080 case -ENOBUFS:
1081 case -EPIPE:
1082 break;
1083 default:
1084 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1085 -status);
1086 }
1087
1088 return status;
1089}
1090
1091static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1092{
1093 transport->old_data_ready = sk->sk_data_ready;
1094 transport->old_state_change = sk->sk_state_change;
1095 transport->old_write_space = sk->sk_write_space;
1096 transport->old_error_report = sk->sk_error_report;
1097}
1098
1099static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1100{
1101 sk->sk_data_ready = transport->old_data_ready;
1102 sk->sk_state_change = transport->old_state_change;
1103 sk->sk_write_space = transport->old_write_space;
1104 sk->sk_error_report = transport->old_error_report;
1105}
1106
1107static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
1108{
1109 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1110
1111 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1112 clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state);
1113 clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state);
1114 clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state);
1115}
1116
1117static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr)
1118{
1119 set_bit(nr, &transport->sock_state);
1120 queue_work(xprtiod_workqueue, &transport->error_worker);
1121}
1122
1123static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1124{
1125 smp_mb__before_atomic();
1126 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1127 clear_bit(XPRT_CLOSING, &xprt->state);
1128 xs_sock_reset_state_flags(xprt);
1129 smp_mb__after_atomic();
1130}
1131
1132/**
1133 * xs_error_report - callback to handle TCP socket state errors
1134 * @sk: socket
1135 *
1136 * Note: we don't call sock_error() since there may be a rpc_task
1137 * using the socket, and so we don't want to clear sk->sk_err.
1138 */
1139static void xs_error_report(struct sock *sk)
1140{
1141 struct sock_xprt *transport;
1142 struct rpc_xprt *xprt;
1143
1144 read_lock_bh(&sk->sk_callback_lock);
1145 if (!(xprt = xprt_from_sock(sk)))
1146 goto out;
1147
1148 transport = container_of(xprt, struct sock_xprt, xprt);
1149 transport->xprt_err = -sk->sk_err;
1150 if (transport->xprt_err == 0)
1151 goto out;
1152 dprintk("RPC: xs_error_report client %p, error=%d...\n",
1153 xprt, -transport->xprt_err);
1154 trace_rpc_socket_error(xprt, sk->sk_socket, transport->xprt_err);
1155
1156 /* barrier ensures xprt_err is set before XPRT_SOCK_WAKE_ERROR */
1157 smp_mb__before_atomic();
1158 xs_run_error_worker(transport, XPRT_SOCK_WAKE_ERROR);
1159 out:
1160 read_unlock_bh(&sk->sk_callback_lock);
1161}
1162
1163static void xs_reset_transport(struct sock_xprt *transport)
1164{
1165 struct socket *sock = transport->sock;
1166 struct sock *sk = transport->inet;
1167 struct rpc_xprt *xprt = &transport->xprt;
1168 struct file *filp = transport->file;
1169
1170 if (sk == NULL)
1171 return;
1172
1173 if (atomic_read(&transport->xprt.swapper))
1174 sk_clear_memalloc(sk);
1175
1176 kernel_sock_shutdown(sock, SHUT_RDWR);
1177
1178 mutex_lock(&transport->recv_mutex);
1179 write_lock_bh(&sk->sk_callback_lock);
1180 transport->inet = NULL;
1181 transport->sock = NULL;
1182 transport->file = NULL;
1183
1184 sk->sk_user_data = NULL;
1185
1186 xs_restore_old_callbacks(transport, sk);
1187 xprt_clear_connected(xprt);
1188 write_unlock_bh(&sk->sk_callback_lock);
1189 xs_sock_reset_connection_flags(xprt);
1190 /* Reset stream record info */
1191 xs_stream_reset_connect(transport);
1192 mutex_unlock(&transport->recv_mutex);
1193
1194 trace_rpc_socket_close(xprt, sock);
1195 fput(filp);
1196
1197 xprt_disconnect_done(xprt);
1198}
1199
1200/**
1201 * xs_close - close a socket
1202 * @xprt: transport
1203 *
1204 * This is used when all requests are complete; ie, no DRC state remains
1205 * on the server we want to save.
1206 *
1207 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1208 * xs_reset_transport() zeroing the socket from underneath a writer.
1209 */
1210static void xs_close(struct rpc_xprt *xprt)
1211{
1212 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1213
1214 dprintk("RPC: xs_close xprt %p\n", xprt);
1215
1216 xs_reset_transport(transport);
1217 xprt->reestablish_timeout = 0;
1218}
1219
1220static void xs_inject_disconnect(struct rpc_xprt *xprt)
1221{
1222 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
1223 xprt);
1224 xprt_disconnect_done(xprt);
1225}
1226
1227static void xs_xprt_free(struct rpc_xprt *xprt)
1228{
1229 xs_free_peer_addresses(xprt);
1230 xprt_free(xprt);
1231}
1232
1233/**
1234 * xs_destroy - prepare to shutdown a transport
1235 * @xprt: doomed transport
1236 *
1237 */
1238static void xs_destroy(struct rpc_xprt *xprt)
1239{
1240 struct sock_xprt *transport = container_of(xprt,
1241 struct sock_xprt, xprt);
1242 dprintk("RPC: xs_destroy xprt %p\n", xprt);
1243
1244 cancel_delayed_work_sync(&transport->connect_worker);
1245 xs_close(xprt);
1246 cancel_work_sync(&transport->recv_worker);
1247 cancel_work_sync(&transport->error_worker);
1248 xs_xprt_free(xprt);
1249 module_put(THIS_MODULE);
1250}
1251
1252/**
1253 * xs_udp_data_read_skb - receive callback for UDP sockets
1254 * @xprt: transport
1255 * @sk: socket
1256 * @skb: skbuff
1257 *
1258 */
1259static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1260 struct sock *sk,
1261 struct sk_buff *skb)
1262{
1263 struct rpc_task *task;
1264 struct rpc_rqst *rovr;
1265 int repsize, copied;
1266 u32 _xid;
1267 __be32 *xp;
1268
1269 repsize = skb->len;
1270 if (repsize < 4) {
1271 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
1272 return;
1273 }
1274
1275 /* Copy the XID from the skb... */
1276 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1277 if (xp == NULL)
1278 return;
1279
1280 /* Look up and lock the request corresponding to the given XID */
1281 spin_lock(&xprt->queue_lock);
1282 rovr = xprt_lookup_rqst(xprt, *xp);
1283 if (!rovr)
1284 goto out_unlock;
1285 xprt_pin_rqst(rovr);
1286 xprt_update_rtt(rovr->rq_task);
1287 spin_unlock(&xprt->queue_lock);
1288 task = rovr->rq_task;
1289
1290 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1291 copied = repsize;
1292
1293 /* Suck it into the iovec, verify checksum if not done by hw. */
1294 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1295 spin_lock(&xprt->queue_lock);
1296 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1297 goto out_unpin;
1298 }
1299
1300
1301 spin_lock(&xprt->transport_lock);
1302 xprt_adjust_cwnd(xprt, task, copied);
1303 spin_unlock(&xprt->transport_lock);
1304 spin_lock(&xprt->queue_lock);
1305 xprt_complete_rqst(task, copied);
1306 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1307out_unpin:
1308 xprt_unpin_rqst(rovr);
1309 out_unlock:
1310 spin_unlock(&xprt->queue_lock);
1311}
1312
1313static void xs_udp_data_receive(struct sock_xprt *transport)
1314{
1315 struct sk_buff *skb;
1316 struct sock *sk;
1317 int err;
1318
1319 mutex_lock(&transport->recv_mutex);
1320 sk = transport->inet;
1321 if (sk == NULL)
1322 goto out;
1323 for (;;) {
1324 skb = skb_recv_udp(sk, 0, 1, &err);
1325 if (skb == NULL)
1326 break;
1327 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1328 consume_skb(skb);
1329 cond_resched();
1330 }
1331 xs_poll_check_readable(transport);
1332out:
1333 mutex_unlock(&transport->recv_mutex);
1334}
1335
1336static void xs_udp_data_receive_workfn(struct work_struct *work)
1337{
1338 struct sock_xprt *transport =
1339 container_of(work, struct sock_xprt, recv_worker);
1340 unsigned int pflags = memalloc_nofs_save();
1341
1342 xs_udp_data_receive(transport);
1343 memalloc_nofs_restore(pflags);
1344}
1345
1346/**
1347 * xs_data_ready - "data ready" callback for UDP sockets
1348 * @sk: socket with data to read
1349 *
1350 */
1351static void xs_data_ready(struct sock *sk)
1352{
1353 struct rpc_xprt *xprt;
1354
1355 read_lock_bh(&sk->sk_callback_lock);
1356 dprintk("RPC: xs_data_ready...\n");
1357 xprt = xprt_from_sock(sk);
1358 if (xprt != NULL) {
1359 struct sock_xprt *transport = container_of(xprt,
1360 struct sock_xprt, xprt);
1361 transport->old_data_ready(sk);
1362 /* Any data means we had a useful conversation, so
1363 * then we don't need to delay the next reconnect
1364 */
1365 if (xprt->reestablish_timeout)
1366 xprt->reestablish_timeout = 0;
1367 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1368 queue_work(xprtiod_workqueue, &transport->recv_worker);
1369 }
1370 read_unlock_bh(&sk->sk_callback_lock);
1371}
1372
1373/*
1374 * Helper function to force a TCP close if the server is sending
1375 * junk and/or it has put us in CLOSE_WAIT
1376 */
1377static void xs_tcp_force_close(struct rpc_xprt *xprt)
1378{
1379 xprt_force_disconnect(xprt);
1380}
1381
1382#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1383static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1384{
1385 return PAGE_SIZE;
1386}
1387#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1388
1389/**
1390 * xs_tcp_state_change - callback to handle TCP socket state changes
1391 * @sk: socket whose state has changed
1392 *
1393 */
1394static void xs_tcp_state_change(struct sock *sk)
1395{
1396 struct rpc_xprt *xprt;
1397 struct sock_xprt *transport;
1398
1399 read_lock_bh(&sk->sk_callback_lock);
1400 if (!(xprt = xprt_from_sock(sk)))
1401 goto out;
1402 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1403 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1404 sk->sk_state, xprt_connected(xprt),
1405 sock_flag(sk, SOCK_DEAD),
1406 sock_flag(sk, SOCK_ZAPPED),
1407 sk->sk_shutdown);
1408
1409 transport = container_of(xprt, struct sock_xprt, xprt);
1410 trace_rpc_socket_state_change(xprt, sk->sk_socket);
1411 switch (sk->sk_state) {
1412 case TCP_ESTABLISHED:
1413 if (!xprt_test_and_set_connected(xprt)) {
1414 xprt->connect_cookie++;
1415 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1416 xprt_clear_connecting(xprt);
1417
1418 xprt->stat.connect_count++;
1419 xprt->stat.connect_time += (long)jiffies -
1420 xprt->stat.connect_start;
1421 xs_run_error_worker(transport, XPRT_SOCK_WAKE_PENDING);
1422 }
1423 break;
1424 case TCP_FIN_WAIT1:
1425 /* The client initiated a shutdown of the socket */
1426 xprt->connect_cookie++;
1427 xprt->reestablish_timeout = 0;
1428 set_bit(XPRT_CLOSING, &xprt->state);
1429 smp_mb__before_atomic();
1430 clear_bit(XPRT_CONNECTED, &xprt->state);
1431 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1432 smp_mb__after_atomic();
1433 break;
1434 case TCP_CLOSE_WAIT:
1435 /* The server initiated a shutdown of the socket */
1436 xprt->connect_cookie++;
1437 clear_bit(XPRT_CONNECTED, &xprt->state);
1438 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1439 fallthrough;
1440 case TCP_CLOSING:
1441 /*
1442 * If the server closed down the connection, make sure that
1443 * we back off before reconnecting
1444 */
1445 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1446 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1447 break;
1448 case TCP_LAST_ACK:
1449 set_bit(XPRT_CLOSING, &xprt->state);
1450 smp_mb__before_atomic();
1451 clear_bit(XPRT_CONNECTED, &xprt->state);
1452 smp_mb__after_atomic();
1453 break;
1454 case TCP_CLOSE:
1455 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1456 &transport->sock_state))
1457 xprt_clear_connecting(xprt);
1458 clear_bit(XPRT_CLOSING, &xprt->state);
1459 /* Trigger the socket release */
1460 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1461 }
1462 out:
1463 read_unlock_bh(&sk->sk_callback_lock);
1464}
1465
1466static void xs_write_space(struct sock *sk)
1467{
1468 struct socket_wq *wq;
1469 struct sock_xprt *transport;
1470 struct rpc_xprt *xprt;
1471
1472 if (!sk->sk_socket)
1473 return;
1474 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1475
1476 if (unlikely(!(xprt = xprt_from_sock(sk))))
1477 return;
1478 transport = container_of(xprt, struct sock_xprt, xprt);
1479 rcu_read_lock();
1480 wq = rcu_dereference(sk->sk_wq);
1481 if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1482 goto out;
1483
1484 xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE);
1485 sk->sk_write_pending--;
1486out:
1487 rcu_read_unlock();
1488}
1489
1490/**
1491 * xs_udp_write_space - callback invoked when socket buffer space
1492 * becomes available
1493 * @sk: socket whose state has changed
1494 *
1495 * Called when more output buffer space is available for this socket.
1496 * We try not to wake our writers until they can make "significant"
1497 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1498 * with a bunch of small requests.
1499 */
1500static void xs_udp_write_space(struct sock *sk)
1501{
1502 read_lock_bh(&sk->sk_callback_lock);
1503
1504 /* from net/core/sock.c:sock_def_write_space */
1505 if (sock_writeable(sk))
1506 xs_write_space(sk);
1507
1508 read_unlock_bh(&sk->sk_callback_lock);
1509}
1510
1511/**
1512 * xs_tcp_write_space - callback invoked when socket buffer space
1513 * becomes available
1514 * @sk: socket whose state has changed
1515 *
1516 * Called when more output buffer space is available for this socket.
1517 * We try not to wake our writers until they can make "significant"
1518 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1519 * with a bunch of small requests.
1520 */
1521static void xs_tcp_write_space(struct sock *sk)
1522{
1523 read_lock_bh(&sk->sk_callback_lock);
1524
1525 /* from net/core/stream.c:sk_stream_write_space */
1526 if (sk_stream_is_writeable(sk))
1527 xs_write_space(sk);
1528
1529 read_unlock_bh(&sk->sk_callback_lock);
1530}
1531
1532static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1533{
1534 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1535 struct sock *sk = transport->inet;
1536
1537 if (transport->rcvsize) {
1538 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1539 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1540 }
1541 if (transport->sndsize) {
1542 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1543 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1544 sk->sk_write_space(sk);
1545 }
1546}
1547
1548/**
1549 * xs_udp_set_buffer_size - set send and receive limits
1550 * @xprt: generic transport
1551 * @sndsize: requested size of send buffer, in bytes
1552 * @rcvsize: requested size of receive buffer, in bytes
1553 *
1554 * Set socket send and receive buffer size limits.
1555 */
1556static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1557{
1558 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1559
1560 transport->sndsize = 0;
1561 if (sndsize)
1562 transport->sndsize = sndsize + 1024;
1563 transport->rcvsize = 0;
1564 if (rcvsize)
1565 transport->rcvsize = rcvsize + 1024;
1566
1567 xs_udp_do_set_buffer_size(xprt);
1568}
1569
1570/**
1571 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1572 * @xprt: controlling transport
1573 * @task: task that timed out
1574 *
1575 * Adjust the congestion window after a retransmit timeout has occurred.
1576 */
1577static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1578{
1579 spin_lock(&xprt->transport_lock);
1580 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1581 spin_unlock(&xprt->transport_lock);
1582}
1583
1584static int xs_get_random_port(void)
1585{
1586 unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1587 unsigned short range;
1588 unsigned short rand;
1589
1590 if (max < min)
1591 return -EADDRINUSE;
1592 range = max - min + 1;
1593 rand = (unsigned short) prandom_u32() % range;
1594 return rand + min;
1595}
1596
1597static unsigned short xs_sock_getport(struct socket *sock)
1598{
1599 struct sockaddr_storage buf;
1600 unsigned short port = 0;
1601
1602 if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1603 goto out;
1604 switch (buf.ss_family) {
1605 case AF_INET6:
1606 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1607 break;
1608 case AF_INET:
1609 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1610 }
1611out:
1612 return port;
1613}
1614
1615/**
1616 * xs_set_port - reset the port number in the remote endpoint address
1617 * @xprt: generic transport
1618 * @port: new port number
1619 *
1620 */
1621static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1622{
1623 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1624
1625 rpc_set_port(xs_addr(xprt), port);
1626 xs_update_peer_port(xprt);
1627}
1628
1629static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1630{
1631 if (transport->srcport == 0 && transport->xprt.reuseport)
1632 transport->srcport = xs_sock_getport(sock);
1633}
1634
1635static int xs_get_srcport(struct sock_xprt *transport)
1636{
1637 int port = transport->srcport;
1638
1639 if (port == 0 && transport->xprt.resvport)
1640 port = xs_get_random_port();
1641 return port;
1642}
1643
1644static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1645{
1646 if (transport->srcport != 0)
1647 transport->srcport = 0;
1648 if (!transport->xprt.resvport)
1649 return 0;
1650 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1651 return xprt_max_resvport;
1652 return --port;
1653}
1654static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1655{
1656 struct sockaddr_storage myaddr;
1657 int err, nloop = 0;
1658 int port = xs_get_srcport(transport);
1659 unsigned short last;
1660
1661 /*
1662 * If we are asking for any ephemeral port (i.e. port == 0 &&
1663 * transport->xprt.resvport == 0), don't bind. Let the local
1664 * port selection happen implicitly when the socket is used
1665 * (for example at connect time).
1666 *
1667 * This ensures that we can continue to establish TCP
1668 * connections even when all local ephemeral ports are already
1669 * a part of some TCP connection. This makes no difference
1670 * for UDP sockets, but also doens't harm them.
1671 *
1672 * If we're asking for any reserved port (i.e. port == 0 &&
1673 * transport->xprt.resvport == 1) xs_get_srcport above will
1674 * ensure that port is non-zero and we will bind as needed.
1675 */
1676 if (port <= 0)
1677 return port;
1678
1679 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1680 do {
1681 rpc_set_port((struct sockaddr *)&myaddr, port);
1682 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1683 transport->xprt.addrlen);
1684 if (err == 0) {
1685 transport->srcport = port;
1686 break;
1687 }
1688 last = port;
1689 port = xs_next_srcport(transport, port);
1690 if (port > last)
1691 nloop++;
1692 } while (err == -EADDRINUSE && nloop != 2);
1693
1694 if (myaddr.ss_family == AF_INET)
1695 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1696 &((struct sockaddr_in *)&myaddr)->sin_addr,
1697 port, err ? "failed" : "ok", err);
1698 else
1699 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1700 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1701 port, err ? "failed" : "ok", err);
1702 return err;
1703}
1704
1705/*
1706 * We don't support autobind on AF_LOCAL sockets
1707 */
1708static void xs_local_rpcbind(struct rpc_task *task)
1709{
1710 xprt_set_bound(task->tk_xprt);
1711}
1712
1713static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1714{
1715}
1716
1717#ifdef CONFIG_DEBUG_LOCK_ALLOC
1718static struct lock_class_key xs_key[2];
1719static struct lock_class_key xs_slock_key[2];
1720
1721static inline void xs_reclassify_socketu(struct socket *sock)
1722{
1723 struct sock *sk = sock->sk;
1724
1725 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1726 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1727}
1728
1729static inline void xs_reclassify_socket4(struct socket *sock)
1730{
1731 struct sock *sk = sock->sk;
1732
1733 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1734 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1735}
1736
1737static inline void xs_reclassify_socket6(struct socket *sock)
1738{
1739 struct sock *sk = sock->sk;
1740
1741 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1742 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1743}
1744
1745static inline void xs_reclassify_socket(int family, struct socket *sock)
1746{
1747 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1748 return;
1749
1750 switch (family) {
1751 case AF_LOCAL:
1752 xs_reclassify_socketu(sock);
1753 break;
1754 case AF_INET:
1755 xs_reclassify_socket4(sock);
1756 break;
1757 case AF_INET6:
1758 xs_reclassify_socket6(sock);
1759 break;
1760 }
1761}
1762#else
1763static inline void xs_reclassify_socket(int family, struct socket *sock)
1764{
1765}
1766#endif
1767
1768static void xs_dummy_setup_socket(struct work_struct *work)
1769{
1770}
1771
1772static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1773 struct sock_xprt *transport, int family, int type,
1774 int protocol, bool reuseport)
1775{
1776 struct file *filp;
1777 struct socket *sock;
1778 int err;
1779
1780 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1781 if (err < 0) {
1782 dprintk("RPC: can't create %d transport socket (%d).\n",
1783 protocol, -err);
1784 goto out;
1785 }
1786 xs_reclassify_socket(family, sock);
1787
1788 if (reuseport)
1789 sock_set_reuseport(sock->sk);
1790
1791 err = xs_bind(transport, sock);
1792 if (err) {
1793 sock_release(sock);
1794 goto out;
1795 }
1796
1797 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1798 if (IS_ERR(filp))
1799 return ERR_CAST(filp);
1800 transport->file = filp;
1801
1802 return sock;
1803out:
1804 return ERR_PTR(err);
1805}
1806
1807static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1808 struct socket *sock)
1809{
1810 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1811 xprt);
1812
1813 if (!transport->inet) {
1814 struct sock *sk = sock->sk;
1815
1816 write_lock_bh(&sk->sk_callback_lock);
1817
1818 xs_save_old_callbacks(transport, sk);
1819
1820 sk->sk_user_data = xprt;
1821 sk->sk_data_ready = xs_data_ready;
1822 sk->sk_write_space = xs_udp_write_space;
1823 sock_set_flag(sk, SOCK_FASYNC);
1824 sk->sk_error_report = xs_error_report;
1825
1826 xprt_clear_connected(xprt);
1827
1828 /* Reset to new socket */
1829 transport->sock = sock;
1830 transport->inet = sk;
1831
1832 write_unlock_bh(&sk->sk_callback_lock);
1833 }
1834
1835 xs_stream_start_connect(transport);
1836
1837 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1838}
1839
1840/**
1841 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1842 * @transport: socket transport to connect
1843 */
1844static int xs_local_setup_socket(struct sock_xprt *transport)
1845{
1846 struct rpc_xprt *xprt = &transport->xprt;
1847 struct file *filp;
1848 struct socket *sock;
1849 int status;
1850
1851 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1852 SOCK_STREAM, 0, &sock, 1);
1853 if (status < 0) {
1854 dprintk("RPC: can't create AF_LOCAL "
1855 "transport socket (%d).\n", -status);
1856 goto out;
1857 }
1858 xs_reclassify_socket(AF_LOCAL, sock);
1859
1860 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1861 if (IS_ERR(filp)) {
1862 status = PTR_ERR(filp);
1863 goto out;
1864 }
1865 transport->file = filp;
1866
1867 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1868 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1869
1870 status = xs_local_finish_connecting(xprt, sock);
1871 trace_rpc_socket_connect(xprt, sock, status);
1872 switch (status) {
1873 case 0:
1874 dprintk("RPC: xprt %p connected to %s\n",
1875 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1876 xprt->stat.connect_count++;
1877 xprt->stat.connect_time += (long)jiffies -
1878 xprt->stat.connect_start;
1879 xprt_set_connected(xprt);
1880 case -ENOBUFS:
1881 break;
1882 case -ENOENT:
1883 dprintk("RPC: xprt %p: socket %s does not exist\n",
1884 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1885 break;
1886 case -ECONNREFUSED:
1887 dprintk("RPC: xprt %p: connection refused for %s\n",
1888 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1889 break;
1890 default:
1891 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1892 __func__, -status,
1893 xprt->address_strings[RPC_DISPLAY_ADDR]);
1894 }
1895
1896out:
1897 xprt_clear_connecting(xprt);
1898 xprt_wake_pending_tasks(xprt, status);
1899 return status;
1900}
1901
1902static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1903{
1904 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1905 int ret;
1906
1907 if (RPC_IS_ASYNC(task)) {
1908 /*
1909 * We want the AF_LOCAL connect to be resolved in the
1910 * filesystem namespace of the process making the rpc
1911 * call. Thus we connect synchronously.
1912 *
1913 * If we want to support asynchronous AF_LOCAL calls,
1914 * we'll need to figure out how to pass a namespace to
1915 * connect.
1916 */
1917 task->tk_rpc_status = -ENOTCONN;
1918 rpc_exit(task, -ENOTCONN);
1919 return;
1920 }
1921 ret = xs_local_setup_socket(transport);
1922 if (ret && !RPC_IS_SOFTCONN(task))
1923 msleep_interruptible(15000);
1924}
1925
1926#if IS_ENABLED(CONFIG_SUNRPC_SWAP)
1927/*
1928 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
1929 * know that we have exclusive access to the socket), to guard against
1930 * races with xs_reset_transport.
1931 */
1932static void xs_set_memalloc(struct rpc_xprt *xprt)
1933{
1934 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1935 xprt);
1936
1937 /*
1938 * If there's no sock, then we have nothing to set. The
1939 * reconnecting process will get it for us.
1940 */
1941 if (!transport->inet)
1942 return;
1943 if (atomic_read(&xprt->swapper))
1944 sk_set_memalloc(transport->inet);
1945}
1946
1947/**
1948 * xs_enable_swap - Tag this transport as being used for swap.
1949 * @xprt: transport to tag
1950 *
1951 * Take a reference to this transport on behalf of the rpc_clnt, and
1952 * optionally mark it for swapping if it wasn't already.
1953 */
1954static int
1955xs_enable_swap(struct rpc_xprt *xprt)
1956{
1957 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
1958
1959 if (atomic_inc_return(&xprt->swapper) != 1)
1960 return 0;
1961 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
1962 return -ERESTARTSYS;
1963 if (xs->inet)
1964 sk_set_memalloc(xs->inet);
1965 xprt_release_xprt(xprt, NULL);
1966 return 0;
1967}
1968
1969/**
1970 * xs_disable_swap - Untag this transport as being used for swap.
1971 * @xprt: transport to tag
1972 *
1973 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
1974 * swapper refcount goes to 0, untag the socket as a memalloc socket.
1975 */
1976static void
1977xs_disable_swap(struct rpc_xprt *xprt)
1978{
1979 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
1980
1981 if (!atomic_dec_and_test(&xprt->swapper))
1982 return;
1983 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
1984 return;
1985 if (xs->inet)
1986 sk_clear_memalloc(xs->inet);
1987 xprt_release_xprt(xprt, NULL);
1988}
1989#else
1990static void xs_set_memalloc(struct rpc_xprt *xprt)
1991{
1992}
1993
1994static int
1995xs_enable_swap(struct rpc_xprt *xprt)
1996{
1997 return -EINVAL;
1998}
1999
2000static void
2001xs_disable_swap(struct rpc_xprt *xprt)
2002{
2003}
2004#endif
2005
2006static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2007{
2008 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2009
2010 if (!transport->inet) {
2011 struct sock *sk = sock->sk;
2012
2013 write_lock_bh(&sk->sk_callback_lock);
2014
2015 xs_save_old_callbacks(transport, sk);
2016
2017 sk->sk_user_data = xprt;
2018 sk->sk_data_ready = xs_data_ready;
2019 sk->sk_write_space = xs_udp_write_space;
2020 sock_set_flag(sk, SOCK_FASYNC);
2021
2022 xprt_set_connected(xprt);
2023
2024 /* Reset to new socket */
2025 transport->sock = sock;
2026 transport->inet = sk;
2027
2028 xs_set_memalloc(xprt);
2029
2030 write_unlock_bh(&sk->sk_callback_lock);
2031 }
2032 xs_udp_do_set_buffer_size(xprt);
2033
2034 xprt->stat.connect_start = jiffies;
2035}
2036
2037static void xs_udp_setup_socket(struct work_struct *work)
2038{
2039 struct sock_xprt *transport =
2040 container_of(work, struct sock_xprt, connect_worker.work);
2041 struct rpc_xprt *xprt = &transport->xprt;
2042 struct socket *sock;
2043 int status = -EIO;
2044
2045 sock = xs_create_sock(xprt, transport,
2046 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2047 IPPROTO_UDP, false);
2048 if (IS_ERR(sock))
2049 goto out;
2050
2051 dprintk("RPC: worker connecting xprt %p via %s to "
2052 "%s (port %s)\n", xprt,
2053 xprt->address_strings[RPC_DISPLAY_PROTO],
2054 xprt->address_strings[RPC_DISPLAY_ADDR],
2055 xprt->address_strings[RPC_DISPLAY_PORT]);
2056
2057 xs_udp_finish_connecting(xprt, sock);
2058 trace_rpc_socket_connect(xprt, sock, 0);
2059 status = 0;
2060out:
2061 xprt_clear_connecting(xprt);
2062 xprt_unlock_connect(xprt, transport);
2063 xprt_wake_pending_tasks(xprt, status);
2064}
2065
2066/**
2067 * xs_tcp_shutdown - gracefully shut down a TCP socket
2068 * @xprt: transport
2069 *
2070 * Initiates a graceful shutdown of the TCP socket by calling the
2071 * equivalent of shutdown(SHUT_RDWR);
2072 */
2073static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2074{
2075 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2076 struct socket *sock = transport->sock;
2077 int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2078
2079 if (sock == NULL)
2080 return;
2081 switch (skst) {
2082 default:
2083 kernel_sock_shutdown(sock, SHUT_RDWR);
2084 trace_rpc_socket_shutdown(xprt, sock);
2085 break;
2086 case TCP_CLOSE:
2087 case TCP_TIME_WAIT:
2088 xs_reset_transport(transport);
2089 }
2090}
2091
2092static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2093 struct socket *sock)
2094{
2095 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2096 unsigned int keepidle;
2097 unsigned int keepcnt;
2098 unsigned int timeo;
2099
2100 spin_lock(&xprt->transport_lock);
2101 keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2102 keepcnt = xprt->timeout->to_retries + 1;
2103 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2104 (xprt->timeout->to_retries + 1);
2105 clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2106 spin_unlock(&xprt->transport_lock);
2107
2108 /* TCP Keepalive options */
2109 sock_set_keepalive(sock->sk);
2110 tcp_sock_set_keepidle(sock->sk, keepidle);
2111 tcp_sock_set_keepintvl(sock->sk, keepidle);
2112 tcp_sock_set_keepcnt(sock->sk, keepcnt);
2113
2114 /* TCP user timeout (see RFC5482) */
2115 tcp_sock_set_user_timeout(sock->sk, timeo);
2116}
2117
2118static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2119 unsigned long connect_timeout,
2120 unsigned long reconnect_timeout)
2121{
2122 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2123 struct rpc_timeout to;
2124 unsigned long initval;
2125
2126 spin_lock(&xprt->transport_lock);
2127 if (reconnect_timeout < xprt->max_reconnect_timeout)
2128 xprt->max_reconnect_timeout = reconnect_timeout;
2129 if (connect_timeout < xprt->connect_timeout) {
2130 memcpy(&to, xprt->timeout, sizeof(to));
2131 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2132 /* Arbitrary lower limit */
2133 if (initval < XS_TCP_INIT_REEST_TO << 1)
2134 initval = XS_TCP_INIT_REEST_TO << 1;
2135 to.to_initval = initval;
2136 to.to_maxval = initval;
2137 memcpy(&transport->tcp_timeout, &to,
2138 sizeof(transport->tcp_timeout));
2139 xprt->timeout = &transport->tcp_timeout;
2140 xprt->connect_timeout = connect_timeout;
2141 }
2142 set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2143 spin_unlock(&xprt->transport_lock);
2144}
2145
2146static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2147{
2148 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2149 int ret = -ENOTCONN;
2150
2151 if (!transport->inet) {
2152 struct sock *sk = sock->sk;
2153
2154 /* Avoid temporary address, they are bad for long-lived
2155 * connections such as NFS mounts.
2156 * RFC4941, section 3.6 suggests that:
2157 * Individual applications, which have specific
2158 * knowledge about the normal duration of connections,
2159 * MAY override this as appropriate.
2160 */
2161 if (xs_addr(xprt)->sa_family == PF_INET6) {
2162 ip6_sock_set_addr_preferences(sk,
2163 IPV6_PREFER_SRC_PUBLIC);
2164 }
2165
2166 xs_tcp_set_socket_timeouts(xprt, sock);
2167
2168 write_lock_bh(&sk->sk_callback_lock);
2169
2170 xs_save_old_callbacks(transport, sk);
2171
2172 sk->sk_user_data = xprt;
2173 sk->sk_data_ready = xs_data_ready;
2174 sk->sk_state_change = xs_tcp_state_change;
2175 sk->sk_write_space = xs_tcp_write_space;
2176 sock_set_flag(sk, SOCK_FASYNC);
2177 sk->sk_error_report = xs_error_report;
2178
2179 /* socket options */
2180 sock_reset_flag(sk, SOCK_LINGER);
2181 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2182
2183 xprt_clear_connected(xprt);
2184
2185 /* Reset to new socket */
2186 transport->sock = sock;
2187 transport->inet = sk;
2188
2189 write_unlock_bh(&sk->sk_callback_lock);
2190 }
2191
2192 if (!xprt_bound(xprt))
2193 goto out;
2194
2195 xs_set_memalloc(xprt);
2196
2197 xs_stream_start_connect(transport);
2198
2199 /* Tell the socket layer to start connecting... */
2200 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2201 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2202 switch (ret) {
2203 case 0:
2204 xs_set_srcport(transport, sock);
2205 fallthrough;
2206 case -EINPROGRESS:
2207 /* SYN_SENT! */
2208 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2209 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2210 break;
2211 case -EADDRNOTAVAIL:
2212 /* Source port number is unavailable. Try a new one! */
2213 transport->srcport = 0;
2214 }
2215out:
2216 return ret;
2217}
2218
2219/**
2220 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2221 * @work: queued work item
2222 *
2223 * Invoked by a work queue tasklet.
2224 */
2225static void xs_tcp_setup_socket(struct work_struct *work)
2226{
2227 struct sock_xprt *transport =
2228 container_of(work, struct sock_xprt, connect_worker.work);
2229 struct socket *sock = transport->sock;
2230 struct rpc_xprt *xprt = &transport->xprt;
2231 int status = -EIO;
2232
2233 if (!sock) {
2234 sock = xs_create_sock(xprt, transport,
2235 xs_addr(xprt)->sa_family, SOCK_STREAM,
2236 IPPROTO_TCP, true);
2237 if (IS_ERR(sock)) {
2238 status = PTR_ERR(sock);
2239 goto out;
2240 }
2241 }
2242
2243 dprintk("RPC: worker connecting xprt %p via %s to "
2244 "%s (port %s)\n", xprt,
2245 xprt->address_strings[RPC_DISPLAY_PROTO],
2246 xprt->address_strings[RPC_DISPLAY_ADDR],
2247 xprt->address_strings[RPC_DISPLAY_PORT]);
2248
2249 status = xs_tcp_finish_connecting(xprt, sock);
2250 trace_rpc_socket_connect(xprt, sock, status);
2251 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2252 xprt, -status, xprt_connected(xprt),
2253 sock->sk->sk_state);
2254 switch (status) {
2255 default:
2256 printk("%s: connect returned unhandled error %d\n",
2257 __func__, status);
2258 fallthrough;
2259 case -EADDRNOTAVAIL:
2260 /* We're probably in TIME_WAIT. Get rid of existing socket,
2261 * and retry
2262 */
2263 xs_tcp_force_close(xprt);
2264 break;
2265 case 0:
2266 case -EINPROGRESS:
2267 case -EALREADY:
2268 xprt_unlock_connect(xprt, transport);
2269 return;
2270 case -EINVAL:
2271 /* Happens, for instance, if the user specified a link
2272 * local IPv6 address without a scope-id.
2273 */
2274 case -ECONNREFUSED:
2275 case -ECONNRESET:
2276 case -ENETDOWN:
2277 case -ENETUNREACH:
2278 case -EHOSTUNREACH:
2279 case -EADDRINUSE:
2280 case -ENOBUFS:
2281 /*
2282 * xs_tcp_force_close() wakes tasks with -EIO.
2283 * We need to wake them first to ensure the
2284 * correct error code.
2285 */
2286 xprt_wake_pending_tasks(xprt, status);
2287 xs_tcp_force_close(xprt);
2288 goto out;
2289 }
2290 status = -EAGAIN;
2291out:
2292 xprt_clear_connecting(xprt);
2293 xprt_unlock_connect(xprt, transport);
2294 xprt_wake_pending_tasks(xprt, status);
2295}
2296
2297/**
2298 * xs_connect - connect a socket to a remote endpoint
2299 * @xprt: pointer to transport structure
2300 * @task: address of RPC task that manages state of connect request
2301 *
2302 * TCP: If the remote end dropped the connection, delay reconnecting.
2303 *
2304 * UDP socket connects are synchronous, but we use a work queue anyway
2305 * to guarantee that even unprivileged user processes can set up a
2306 * socket on a privileged port.
2307 *
2308 * If a UDP socket connect fails, the delay behavior here prevents
2309 * retry floods (hard mounts).
2310 */
2311static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2312{
2313 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2314 unsigned long delay = 0;
2315
2316 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2317
2318 if (transport->sock != NULL) {
2319 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2320 "seconds\n",
2321 xprt, xprt->reestablish_timeout / HZ);
2322
2323 /* Start by resetting any existing state */
2324 xs_reset_transport(transport);
2325
2326 delay = xprt_reconnect_delay(xprt);
2327 xprt_reconnect_backoff(xprt, XS_TCP_INIT_REEST_TO);
2328
2329 } else
2330 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2331
2332 queue_delayed_work(xprtiod_workqueue,
2333 &transport->connect_worker,
2334 delay);
2335}
2336
2337static void xs_wake_disconnect(struct sock_xprt *transport)
2338{
2339 if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state))
2340 xs_tcp_force_close(&transport->xprt);
2341}
2342
2343static void xs_wake_write(struct sock_xprt *transport)
2344{
2345 if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state))
2346 xprt_write_space(&transport->xprt);
2347}
2348
2349static void xs_wake_error(struct sock_xprt *transport)
2350{
2351 int sockerr;
2352
2353 if (!test_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2354 return;
2355 mutex_lock(&transport->recv_mutex);
2356 if (transport->sock == NULL)
2357 goto out;
2358 if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2359 goto out;
2360 sockerr = xchg(&transport->xprt_err, 0);
2361 if (sockerr < 0)
2362 xprt_wake_pending_tasks(&transport->xprt, sockerr);
2363out:
2364 mutex_unlock(&transport->recv_mutex);
2365}
2366
2367static void xs_wake_pending(struct sock_xprt *transport)
2368{
2369 if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING, &transport->sock_state))
2370 xprt_wake_pending_tasks(&transport->xprt, -EAGAIN);
2371}
2372
2373static void xs_error_handle(struct work_struct *work)
2374{
2375 struct sock_xprt *transport = container_of(work,
2376 struct sock_xprt, error_worker);
2377
2378 xs_wake_disconnect(transport);
2379 xs_wake_write(transport);
2380 xs_wake_error(transport);
2381 xs_wake_pending(transport);
2382}
2383
2384/**
2385 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2386 * @xprt: rpc_xprt struct containing statistics
2387 * @seq: output file
2388 *
2389 */
2390static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2391{
2392 long idle_time = 0;
2393
2394 if (xprt_connected(xprt))
2395 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2396
2397 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2398 "%llu %llu %lu %llu %llu\n",
2399 xprt->stat.bind_count,
2400 xprt->stat.connect_count,
2401 xprt->stat.connect_time / HZ,
2402 idle_time,
2403 xprt->stat.sends,
2404 xprt->stat.recvs,
2405 xprt->stat.bad_xids,
2406 xprt->stat.req_u,
2407 xprt->stat.bklog_u,
2408 xprt->stat.max_slots,
2409 xprt->stat.sending_u,
2410 xprt->stat.pending_u);
2411}
2412
2413/**
2414 * xs_udp_print_stats - display UDP socket-specifc stats
2415 * @xprt: rpc_xprt struct containing statistics
2416 * @seq: output file
2417 *
2418 */
2419static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2420{
2421 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2422
2423 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2424 "%lu %llu %llu\n",
2425 transport->srcport,
2426 xprt->stat.bind_count,
2427 xprt->stat.sends,
2428 xprt->stat.recvs,
2429 xprt->stat.bad_xids,
2430 xprt->stat.req_u,
2431 xprt->stat.bklog_u,
2432 xprt->stat.max_slots,
2433 xprt->stat.sending_u,
2434 xprt->stat.pending_u);
2435}
2436
2437/**
2438 * xs_tcp_print_stats - display TCP socket-specifc stats
2439 * @xprt: rpc_xprt struct containing statistics
2440 * @seq: output file
2441 *
2442 */
2443static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2444{
2445 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2446 long idle_time = 0;
2447
2448 if (xprt_connected(xprt))
2449 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2450
2451 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2452 "%llu %llu %lu %llu %llu\n",
2453 transport->srcport,
2454 xprt->stat.bind_count,
2455 xprt->stat.connect_count,
2456 xprt->stat.connect_time / HZ,
2457 idle_time,
2458 xprt->stat.sends,
2459 xprt->stat.recvs,
2460 xprt->stat.bad_xids,
2461 xprt->stat.req_u,
2462 xprt->stat.bklog_u,
2463 xprt->stat.max_slots,
2464 xprt->stat.sending_u,
2465 xprt->stat.pending_u);
2466}
2467
2468/*
2469 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2470 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2471 * to use the server side send routines.
2472 */
2473static int bc_malloc(struct rpc_task *task)
2474{
2475 struct rpc_rqst *rqst = task->tk_rqstp;
2476 size_t size = rqst->rq_callsize;
2477 struct page *page;
2478 struct rpc_buffer *buf;
2479
2480 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2481 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2482 size);
2483 return -EINVAL;
2484 }
2485
2486 page = alloc_page(GFP_KERNEL);
2487 if (!page)
2488 return -ENOMEM;
2489
2490 buf = page_address(page);
2491 buf->len = PAGE_SIZE;
2492
2493 rqst->rq_buffer = buf->data;
2494 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2495 return 0;
2496}
2497
2498/*
2499 * Free the space allocated in the bc_alloc routine
2500 */
2501static void bc_free(struct rpc_task *task)
2502{
2503 void *buffer = task->tk_rqstp->rq_buffer;
2504 struct rpc_buffer *buf;
2505
2506 buf = container_of(buffer, struct rpc_buffer, data);
2507 free_page((unsigned long)buf);
2508}
2509
2510static int bc_sendto(struct rpc_rqst *req)
2511{
2512 struct xdr_buf *xdr = &req->rq_snd_buf;
2513 struct sock_xprt *transport =
2514 container_of(req->rq_xprt, struct sock_xprt, xprt);
2515 struct msghdr msg = {
2516 .msg_flags = 0,
2517 };
2518 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
2519 (u32)xdr->len);
2520 unsigned int sent = 0;
2521 int err;
2522
2523 req->rq_xtime = ktime_get();
2524 err = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, marker, &sent);
2525 xdr_free_bvec(xdr);
2526 if (err < 0 || sent != (xdr->len + sizeof(marker)))
2527 return -EAGAIN;
2528 return sent;
2529}
2530
2531/**
2532 * bc_send_request - Send a backchannel Call on a TCP socket
2533 * @req: rpc_rqst containing Call message to be sent
2534 *
2535 * xpt_mutex ensures @rqstp's whole message is written to the socket
2536 * without interruption.
2537 *
2538 * Return values:
2539 * %0 if the message was sent successfully
2540 * %ENOTCONN if the message was not sent
2541 */
2542static int bc_send_request(struct rpc_rqst *req)
2543{
2544 struct svc_xprt *xprt;
2545 int len;
2546
2547 /*
2548 * Get the server socket associated with this callback xprt
2549 */
2550 xprt = req->rq_xprt->bc_xprt;
2551
2552 /*
2553 * Grab the mutex to serialize data as the connection is shared
2554 * with the fore channel
2555 */
2556 mutex_lock(&xprt->xpt_mutex);
2557 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2558 len = -ENOTCONN;
2559 else
2560 len = bc_sendto(req);
2561 mutex_unlock(&xprt->xpt_mutex);
2562
2563 if (len > 0)
2564 len = 0;
2565
2566 return len;
2567}
2568
2569/*
2570 * The close routine. Since this is client initiated, we do nothing
2571 */
2572
2573static void bc_close(struct rpc_xprt *xprt)
2574{
2575 xprt_disconnect_done(xprt);
2576}
2577
2578/*
2579 * The xprt destroy routine. Again, because this connection is client
2580 * initiated, we do nothing
2581 */
2582
2583static void bc_destroy(struct rpc_xprt *xprt)
2584{
2585 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2586
2587 xs_xprt_free(xprt);
2588 module_put(THIS_MODULE);
2589}
2590
2591static const struct rpc_xprt_ops xs_local_ops = {
2592 .reserve_xprt = xprt_reserve_xprt,
2593 .release_xprt = xprt_release_xprt,
2594 .alloc_slot = xprt_alloc_slot,
2595 .free_slot = xprt_free_slot,
2596 .rpcbind = xs_local_rpcbind,
2597 .set_port = xs_local_set_port,
2598 .connect = xs_local_connect,
2599 .buf_alloc = rpc_malloc,
2600 .buf_free = rpc_free,
2601 .prepare_request = xs_stream_prepare_request,
2602 .send_request = xs_local_send_request,
2603 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2604 .close = xs_close,
2605 .destroy = xs_destroy,
2606 .print_stats = xs_local_print_stats,
2607 .enable_swap = xs_enable_swap,
2608 .disable_swap = xs_disable_swap,
2609};
2610
2611static const struct rpc_xprt_ops xs_udp_ops = {
2612 .set_buffer_size = xs_udp_set_buffer_size,
2613 .reserve_xprt = xprt_reserve_xprt_cong,
2614 .release_xprt = xprt_release_xprt_cong,
2615 .alloc_slot = xprt_alloc_slot,
2616 .free_slot = xprt_free_slot,
2617 .rpcbind = rpcb_getport_async,
2618 .set_port = xs_set_port,
2619 .connect = xs_connect,
2620 .buf_alloc = rpc_malloc,
2621 .buf_free = rpc_free,
2622 .send_request = xs_udp_send_request,
2623 .wait_for_reply_request = xprt_wait_for_reply_request_rtt,
2624 .timer = xs_udp_timer,
2625 .release_request = xprt_release_rqst_cong,
2626 .close = xs_close,
2627 .destroy = xs_destroy,
2628 .print_stats = xs_udp_print_stats,
2629 .enable_swap = xs_enable_swap,
2630 .disable_swap = xs_disable_swap,
2631 .inject_disconnect = xs_inject_disconnect,
2632};
2633
2634static const struct rpc_xprt_ops xs_tcp_ops = {
2635 .reserve_xprt = xprt_reserve_xprt,
2636 .release_xprt = xprt_release_xprt,
2637 .alloc_slot = xprt_alloc_slot,
2638 .free_slot = xprt_free_slot,
2639 .rpcbind = rpcb_getport_async,
2640 .set_port = xs_set_port,
2641 .connect = xs_connect,
2642 .buf_alloc = rpc_malloc,
2643 .buf_free = rpc_free,
2644 .prepare_request = xs_stream_prepare_request,
2645 .send_request = xs_tcp_send_request,
2646 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2647 .close = xs_tcp_shutdown,
2648 .destroy = xs_destroy,
2649 .set_connect_timeout = xs_tcp_set_connect_timeout,
2650 .print_stats = xs_tcp_print_stats,
2651 .enable_swap = xs_enable_swap,
2652 .disable_swap = xs_disable_swap,
2653 .inject_disconnect = xs_inject_disconnect,
2654#ifdef CONFIG_SUNRPC_BACKCHANNEL
2655 .bc_setup = xprt_setup_bc,
2656 .bc_maxpayload = xs_tcp_bc_maxpayload,
2657 .bc_num_slots = xprt_bc_max_slots,
2658 .bc_free_rqst = xprt_free_bc_rqst,
2659 .bc_destroy = xprt_destroy_bc,
2660#endif
2661};
2662
2663/*
2664 * The rpc_xprt_ops for the server backchannel
2665 */
2666
2667static const struct rpc_xprt_ops bc_tcp_ops = {
2668 .reserve_xprt = xprt_reserve_xprt,
2669 .release_xprt = xprt_release_xprt,
2670 .alloc_slot = xprt_alloc_slot,
2671 .free_slot = xprt_free_slot,
2672 .buf_alloc = bc_malloc,
2673 .buf_free = bc_free,
2674 .send_request = bc_send_request,
2675 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2676 .close = bc_close,
2677 .destroy = bc_destroy,
2678 .print_stats = xs_tcp_print_stats,
2679 .enable_swap = xs_enable_swap,
2680 .disable_swap = xs_disable_swap,
2681 .inject_disconnect = xs_inject_disconnect,
2682};
2683
2684static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2685{
2686 static const struct sockaddr_in sin = {
2687 .sin_family = AF_INET,
2688 .sin_addr.s_addr = htonl(INADDR_ANY),
2689 };
2690 static const struct sockaddr_in6 sin6 = {
2691 .sin6_family = AF_INET6,
2692 .sin6_addr = IN6ADDR_ANY_INIT,
2693 };
2694
2695 switch (family) {
2696 case AF_LOCAL:
2697 break;
2698 case AF_INET:
2699 memcpy(sap, &sin, sizeof(sin));
2700 break;
2701 case AF_INET6:
2702 memcpy(sap, &sin6, sizeof(sin6));
2703 break;
2704 default:
2705 dprintk("RPC: %s: Bad address family\n", __func__);
2706 return -EAFNOSUPPORT;
2707 }
2708 return 0;
2709}
2710
2711static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2712 unsigned int slot_table_size,
2713 unsigned int max_slot_table_size)
2714{
2715 struct rpc_xprt *xprt;
2716 struct sock_xprt *new;
2717
2718 if (args->addrlen > sizeof(xprt->addr)) {
2719 dprintk("RPC: xs_setup_xprt: address too large\n");
2720 return ERR_PTR(-EBADF);
2721 }
2722
2723 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2724 max_slot_table_size);
2725 if (xprt == NULL) {
2726 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2727 "rpc_xprt\n");
2728 return ERR_PTR(-ENOMEM);
2729 }
2730
2731 new = container_of(xprt, struct sock_xprt, xprt);
2732 mutex_init(&new->recv_mutex);
2733 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2734 xprt->addrlen = args->addrlen;
2735 if (args->srcaddr)
2736 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2737 else {
2738 int err;
2739 err = xs_init_anyaddr(args->dstaddr->sa_family,
2740 (struct sockaddr *)&new->srcaddr);
2741 if (err != 0) {
2742 xprt_free(xprt);
2743 return ERR_PTR(err);
2744 }
2745 }
2746
2747 return xprt;
2748}
2749
2750static const struct rpc_timeout xs_local_default_timeout = {
2751 .to_initval = 10 * HZ,
2752 .to_maxval = 10 * HZ,
2753 .to_retries = 2,
2754};
2755
2756/**
2757 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2758 * @args: rpc transport creation arguments
2759 *
2760 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2761 */
2762static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2763{
2764 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2765 struct sock_xprt *transport;
2766 struct rpc_xprt *xprt;
2767 struct rpc_xprt *ret;
2768
2769 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2770 xprt_max_tcp_slot_table_entries);
2771 if (IS_ERR(xprt))
2772 return xprt;
2773 transport = container_of(xprt, struct sock_xprt, xprt);
2774
2775 xprt->prot = 0;
2776 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2777
2778 xprt->bind_timeout = XS_BIND_TO;
2779 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2780 xprt->idle_timeout = XS_IDLE_DISC_TO;
2781
2782 xprt->ops = &xs_local_ops;
2783 xprt->timeout = &xs_local_default_timeout;
2784
2785 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2786 INIT_WORK(&transport->error_worker, xs_error_handle);
2787 INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket);
2788
2789 switch (sun->sun_family) {
2790 case AF_LOCAL:
2791 if (sun->sun_path[0] != '/') {
2792 dprintk("RPC: bad AF_LOCAL address: %s\n",
2793 sun->sun_path);
2794 ret = ERR_PTR(-EINVAL);
2795 goto out_err;
2796 }
2797 xprt_set_bound(xprt);
2798 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2799 ret = ERR_PTR(xs_local_setup_socket(transport));
2800 if (ret)
2801 goto out_err;
2802 break;
2803 default:
2804 ret = ERR_PTR(-EAFNOSUPPORT);
2805 goto out_err;
2806 }
2807
2808 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2809 xprt->address_strings[RPC_DISPLAY_ADDR]);
2810
2811 if (try_module_get(THIS_MODULE))
2812 return xprt;
2813 ret = ERR_PTR(-EINVAL);
2814out_err:
2815 xs_xprt_free(xprt);
2816 return ret;
2817}
2818
2819static const struct rpc_timeout xs_udp_default_timeout = {
2820 .to_initval = 5 * HZ,
2821 .to_maxval = 30 * HZ,
2822 .to_increment = 5 * HZ,
2823 .to_retries = 5,
2824};
2825
2826/**
2827 * xs_setup_udp - Set up transport to use a UDP socket
2828 * @args: rpc transport creation arguments
2829 *
2830 */
2831static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2832{
2833 struct sockaddr *addr = args->dstaddr;
2834 struct rpc_xprt *xprt;
2835 struct sock_xprt *transport;
2836 struct rpc_xprt *ret;
2837
2838 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2839 xprt_udp_slot_table_entries);
2840 if (IS_ERR(xprt))
2841 return xprt;
2842 transport = container_of(xprt, struct sock_xprt, xprt);
2843
2844 xprt->prot = IPPROTO_UDP;
2845 /* XXX: header size can vary due to auth type, IPv6, etc. */
2846 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2847
2848 xprt->bind_timeout = XS_BIND_TO;
2849 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2850 xprt->idle_timeout = XS_IDLE_DISC_TO;
2851
2852 xprt->ops = &xs_udp_ops;
2853
2854 xprt->timeout = &xs_udp_default_timeout;
2855
2856 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
2857 INIT_WORK(&transport->error_worker, xs_error_handle);
2858 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2859
2860 switch (addr->sa_family) {
2861 case AF_INET:
2862 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2863 xprt_set_bound(xprt);
2864
2865 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2866 break;
2867 case AF_INET6:
2868 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2869 xprt_set_bound(xprt);
2870
2871 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2872 break;
2873 default:
2874 ret = ERR_PTR(-EAFNOSUPPORT);
2875 goto out_err;
2876 }
2877
2878 if (xprt_bound(xprt))
2879 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2880 xprt->address_strings[RPC_DISPLAY_ADDR],
2881 xprt->address_strings[RPC_DISPLAY_PORT],
2882 xprt->address_strings[RPC_DISPLAY_PROTO]);
2883 else
2884 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2885 xprt->address_strings[RPC_DISPLAY_ADDR],
2886 xprt->address_strings[RPC_DISPLAY_PROTO]);
2887
2888 if (try_module_get(THIS_MODULE))
2889 return xprt;
2890 ret = ERR_PTR(-EINVAL);
2891out_err:
2892 xs_xprt_free(xprt);
2893 return ret;
2894}
2895
2896static const struct rpc_timeout xs_tcp_default_timeout = {
2897 .to_initval = 60 * HZ,
2898 .to_maxval = 60 * HZ,
2899 .to_retries = 2,
2900};
2901
2902/**
2903 * xs_setup_tcp - Set up transport to use a TCP socket
2904 * @args: rpc transport creation arguments
2905 *
2906 */
2907static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2908{
2909 struct sockaddr *addr = args->dstaddr;
2910 struct rpc_xprt *xprt;
2911 struct sock_xprt *transport;
2912 struct rpc_xprt *ret;
2913 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2914
2915 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2916 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2917
2918 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2919 max_slot_table_size);
2920 if (IS_ERR(xprt))
2921 return xprt;
2922 transport = container_of(xprt, struct sock_xprt, xprt);
2923
2924 xprt->prot = IPPROTO_TCP;
2925 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2926
2927 xprt->bind_timeout = XS_BIND_TO;
2928 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2929 xprt->idle_timeout = XS_IDLE_DISC_TO;
2930
2931 xprt->ops = &xs_tcp_ops;
2932 xprt->timeout = &xs_tcp_default_timeout;
2933
2934 xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
2935 xprt->connect_timeout = xprt->timeout->to_initval *
2936 (xprt->timeout->to_retries + 1);
2937
2938 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2939 INIT_WORK(&transport->error_worker, xs_error_handle);
2940 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
2941
2942 switch (addr->sa_family) {
2943 case AF_INET:
2944 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2945 xprt_set_bound(xprt);
2946
2947 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2948 break;
2949 case AF_INET6:
2950 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2951 xprt_set_bound(xprt);
2952
2953 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2954 break;
2955 default:
2956 ret = ERR_PTR(-EAFNOSUPPORT);
2957 goto out_err;
2958 }
2959
2960 if (xprt_bound(xprt))
2961 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
2962 xprt->address_strings[RPC_DISPLAY_ADDR],
2963 xprt->address_strings[RPC_DISPLAY_PORT],
2964 xprt->address_strings[RPC_DISPLAY_PROTO]);
2965 else
2966 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
2967 xprt->address_strings[RPC_DISPLAY_ADDR],
2968 xprt->address_strings[RPC_DISPLAY_PROTO]);
2969
2970 if (try_module_get(THIS_MODULE))
2971 return xprt;
2972 ret = ERR_PTR(-EINVAL);
2973out_err:
2974 xs_xprt_free(xprt);
2975 return ret;
2976}
2977
2978/**
2979 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2980 * @args: rpc transport creation arguments
2981 *
2982 */
2983static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2984{
2985 struct sockaddr *addr = args->dstaddr;
2986 struct rpc_xprt *xprt;
2987 struct sock_xprt *transport;
2988 struct svc_sock *bc_sock;
2989 struct rpc_xprt *ret;
2990
2991 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2992 xprt_tcp_slot_table_entries);
2993 if (IS_ERR(xprt))
2994 return xprt;
2995 transport = container_of(xprt, struct sock_xprt, xprt);
2996
2997 xprt->prot = IPPROTO_TCP;
2998 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2999 xprt->timeout = &xs_tcp_default_timeout;
3000
3001 /* backchannel */
3002 xprt_set_bound(xprt);
3003 xprt->bind_timeout = 0;
3004 xprt->reestablish_timeout = 0;
3005 xprt->idle_timeout = 0;
3006
3007 xprt->ops = &bc_tcp_ops;
3008
3009 switch (addr->sa_family) {
3010 case AF_INET:
3011 xs_format_peer_addresses(xprt, "tcp",
3012 RPCBIND_NETID_TCP);
3013 break;
3014 case AF_INET6:
3015 xs_format_peer_addresses(xprt, "tcp",
3016 RPCBIND_NETID_TCP6);
3017 break;
3018 default:
3019 ret = ERR_PTR(-EAFNOSUPPORT);
3020 goto out_err;
3021 }
3022
3023 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3024 xprt->address_strings[RPC_DISPLAY_ADDR],
3025 xprt->address_strings[RPC_DISPLAY_PORT],
3026 xprt->address_strings[RPC_DISPLAY_PROTO]);
3027
3028 /*
3029 * Once we've associated a backchannel xprt with a connection,
3030 * we want to keep it around as long as the connection lasts,
3031 * in case we need to start using it for a backchannel again;
3032 * this reference won't be dropped until bc_xprt is destroyed.
3033 */
3034 xprt_get(xprt);
3035 args->bc_xprt->xpt_bc_xprt = xprt;
3036 xprt->bc_xprt = args->bc_xprt;
3037 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3038 transport->sock = bc_sock->sk_sock;
3039 transport->inet = bc_sock->sk_sk;
3040
3041 /*
3042 * Since we don't want connections for the backchannel, we set
3043 * the xprt status to connected
3044 */
3045 xprt_set_connected(xprt);
3046
3047 if (try_module_get(THIS_MODULE))
3048 return xprt;
3049
3050 args->bc_xprt->xpt_bc_xprt = NULL;
3051 args->bc_xprt->xpt_bc_xps = NULL;
3052 xprt_put(xprt);
3053 ret = ERR_PTR(-EINVAL);
3054out_err:
3055 xs_xprt_free(xprt);
3056 return ret;
3057}
3058
3059static struct xprt_class xs_local_transport = {
3060 .list = LIST_HEAD_INIT(xs_local_transport.list),
3061 .name = "named UNIX socket",
3062 .owner = THIS_MODULE,
3063 .ident = XPRT_TRANSPORT_LOCAL,
3064 .setup = xs_setup_local,
3065};
3066
3067static struct xprt_class xs_udp_transport = {
3068 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3069 .name = "udp",
3070 .owner = THIS_MODULE,
3071 .ident = XPRT_TRANSPORT_UDP,
3072 .setup = xs_setup_udp,
3073};
3074
3075static struct xprt_class xs_tcp_transport = {
3076 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3077 .name = "tcp",
3078 .owner = THIS_MODULE,
3079 .ident = XPRT_TRANSPORT_TCP,
3080 .setup = xs_setup_tcp,
3081};
3082
3083static struct xprt_class xs_bc_tcp_transport = {
3084 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3085 .name = "tcp NFSv4.1 backchannel",
3086 .owner = THIS_MODULE,
3087 .ident = XPRT_TRANSPORT_BC_TCP,
3088 .setup = xs_setup_bc_tcp,
3089};
3090
3091/**
3092 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3093 *
3094 */
3095int init_socket_xprt(void)
3096{
3097 if (!sunrpc_table_header)
3098 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3099
3100 xprt_register_transport(&xs_local_transport);
3101 xprt_register_transport(&xs_udp_transport);
3102 xprt_register_transport(&xs_tcp_transport);
3103 xprt_register_transport(&xs_bc_tcp_transport);
3104
3105 return 0;
3106}
3107
3108/**
3109 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3110 *
3111 */
3112void cleanup_socket_xprt(void)
3113{
3114 if (sunrpc_table_header) {
3115 unregister_sysctl_table(sunrpc_table_header);
3116 sunrpc_table_header = NULL;
3117 }
3118
3119 xprt_unregister_transport(&xs_local_transport);
3120 xprt_unregister_transport(&xs_udp_transport);
3121 xprt_unregister_transport(&xs_tcp_transport);
3122 xprt_unregister_transport(&xs_bc_tcp_transport);
3123}
3124
3125static int param_set_uint_minmax(const char *val,
3126 const struct kernel_param *kp,
3127 unsigned int min, unsigned int max)
3128{
3129 unsigned int num;
3130 int ret;
3131
3132 if (!val)
3133 return -EINVAL;
3134 ret = kstrtouint(val, 0, &num);
3135 if (ret)
3136 return ret;
3137 if (num < min || num > max)
3138 return -EINVAL;
3139 *((unsigned int *)kp->arg) = num;
3140 return 0;
3141}
3142
3143static int param_set_portnr(const char *val, const struct kernel_param *kp)
3144{
3145 return param_set_uint_minmax(val, kp,
3146 RPC_MIN_RESVPORT,
3147 RPC_MAX_RESVPORT);
3148}
3149
3150static const struct kernel_param_ops param_ops_portnr = {
3151 .set = param_set_portnr,
3152 .get = param_get_uint,
3153};
3154
3155#define param_check_portnr(name, p) \
3156 __param_check(name, p, unsigned int);
3157
3158module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3159module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3160
3161static int param_set_slot_table_size(const char *val,
3162 const struct kernel_param *kp)
3163{
3164 return param_set_uint_minmax(val, kp,
3165 RPC_MIN_SLOT_TABLE,
3166 RPC_MAX_SLOT_TABLE);
3167}
3168
3169static const struct kernel_param_ops param_ops_slot_table_size = {
3170 .set = param_set_slot_table_size,
3171 .get = param_get_uint,
3172};
3173
3174#define param_check_slot_table_size(name, p) \
3175 __param_check(name, p, unsigned int);
3176
3177static int param_set_max_slot_table_size(const char *val,
3178 const struct kernel_param *kp)
3179{
3180 return param_set_uint_minmax(val, kp,
3181 RPC_MIN_SLOT_TABLE,
3182 RPC_MAX_SLOT_TABLE_LIMIT);
3183}
3184
3185static const struct kernel_param_ops param_ops_max_slot_table_size = {
3186 .set = param_set_max_slot_table_size,
3187 .get = param_get_uint,
3188};
3189
3190#define param_check_max_slot_table_size(name, p) \
3191 __param_check(name, p, unsigned int);
3192
3193module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3194 slot_table_size, 0644);
3195module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3196 max_slot_table_size, 0644);
3197module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3198 slot_table_size, 0644);