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