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 "sunrpc.h"
58
59static void xs_close(struct rpc_xprt *xprt);
60static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
61 struct socket *sock);
62
63/*
64 * xprtsock tunables
65 */
66static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
67static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
68static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
69
70static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
71static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
72
73#define XS_TCP_LINGER_TO (15U * HZ)
74static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
75
76/*
77 * We can register our own files under /proc/sys/sunrpc by
78 * calling register_sysctl_table() again. The files in that
79 * directory become the union of all files registered there.
80 *
81 * We simply need to make sure that we don't collide with
82 * someone else's file names!
83 */
84
85static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
86static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
87static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
88static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
89static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
90
91static struct ctl_table_header *sunrpc_table_header;
92
93/*
94 * FIXME: changing the UDP slot table size should also resize the UDP
95 * socket buffers for existing UDP transports
96 */
97static struct ctl_table xs_tunables_table[] = {
98 {
99 .procname = "udp_slot_table_entries",
100 .data = &xprt_udp_slot_table_entries,
101 .maxlen = sizeof(unsigned int),
102 .mode = 0644,
103 .proc_handler = proc_dointvec_minmax,
104 .extra1 = &min_slot_table_size,
105 .extra2 = &max_slot_table_size
106 },
107 {
108 .procname = "tcp_slot_table_entries",
109 .data = &xprt_tcp_slot_table_entries,
110 .maxlen = sizeof(unsigned int),
111 .mode = 0644,
112 .proc_handler = proc_dointvec_minmax,
113 .extra1 = &min_slot_table_size,
114 .extra2 = &max_slot_table_size
115 },
116 {
117 .procname = "tcp_max_slot_table_entries",
118 .data = &xprt_max_tcp_slot_table_entries,
119 .maxlen = sizeof(unsigned int),
120 .mode = 0644,
121 .proc_handler = proc_dointvec_minmax,
122 .extra1 = &min_slot_table_size,
123 .extra2 = &max_tcp_slot_table_limit
124 },
125 {
126 .procname = "min_resvport",
127 .data = &xprt_min_resvport,
128 .maxlen = sizeof(unsigned int),
129 .mode = 0644,
130 .proc_handler = proc_dointvec_minmax,
131 .extra1 = &xprt_min_resvport_limit,
132 .extra2 = &xprt_max_resvport_limit
133 },
134 {
135 .procname = "max_resvport",
136 .data = &xprt_max_resvport,
137 .maxlen = sizeof(unsigned int),
138 .mode = 0644,
139 .proc_handler = proc_dointvec_minmax,
140 .extra1 = &xprt_min_resvport_limit,
141 .extra2 = &xprt_max_resvport_limit
142 },
143 {
144 .procname = "tcp_fin_timeout",
145 .data = &xs_tcp_fin_timeout,
146 .maxlen = sizeof(xs_tcp_fin_timeout),
147 .mode = 0644,
148 .proc_handler = proc_dointvec_jiffies,
149 },
150 { },
151};
152
153static struct ctl_table sunrpc_table[] = {
154 {
155 .procname = "sunrpc",
156 .mode = 0555,
157 .child = xs_tunables_table
158 },
159 { },
160};
161
162/*
163 * Wait duration for a reply from the RPC portmapper.
164 */
165#define XS_BIND_TO (60U * HZ)
166
167/*
168 * Delay if a UDP socket connect error occurs. This is most likely some
169 * kind of resource problem on the local host.
170 */
171#define XS_UDP_REEST_TO (2U * HZ)
172
173/*
174 * The reestablish timeout allows clients to delay for a bit before attempting
175 * to reconnect to a server that just dropped our connection.
176 *
177 * We implement an exponential backoff when trying to reestablish a TCP
178 * transport connection with the server. Some servers like to drop a TCP
179 * connection when they are overworked, so we start with a short timeout and
180 * increase over time if the server is down or not responding.
181 */
182#define XS_TCP_INIT_REEST_TO (3U * HZ)
183
184/*
185 * TCP idle timeout; client drops the transport socket if it is idle
186 * for this long. Note that we also timeout UDP sockets to prevent
187 * holding port numbers when there is no RPC traffic.
188 */
189#define XS_IDLE_DISC_TO (5U * 60 * HZ)
190
191#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
192# undef RPC_DEBUG_DATA
193# define RPCDBG_FACILITY RPCDBG_TRANS
194#endif
195
196#ifdef RPC_DEBUG_DATA
197static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
198{
199 u8 *buf = (u8 *) packet;
200 int j;
201
202 dprintk("RPC: %s\n", msg);
203 for (j = 0; j < count && j < 128; j += 4) {
204 if (!(j & 31)) {
205 if (j)
206 dprintk("\n");
207 dprintk("0x%04x ", j);
208 }
209 dprintk("%02x%02x%02x%02x ",
210 buf[j], buf[j+1], buf[j+2], buf[j+3]);
211 }
212 dprintk("\n");
213}
214#else
215static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
216{
217 /* NOP */
218}
219#endif
220
221static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
222{
223 return (struct rpc_xprt *) sk->sk_user_data;
224}
225
226static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
227{
228 return (struct sockaddr *) &xprt->addr;
229}
230
231static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
232{
233 return (struct sockaddr_un *) &xprt->addr;
234}
235
236static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
237{
238 return (struct sockaddr_in *) &xprt->addr;
239}
240
241static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
242{
243 return (struct sockaddr_in6 *) &xprt->addr;
244}
245
246static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
247{
248 struct sockaddr *sap = xs_addr(xprt);
249 struct sockaddr_in6 *sin6;
250 struct sockaddr_in *sin;
251 struct sockaddr_un *sun;
252 char buf[128];
253
254 switch (sap->sa_family) {
255 case AF_LOCAL:
256 sun = xs_addr_un(xprt);
257 strlcpy(buf, sun->sun_path, sizeof(buf));
258 xprt->address_strings[RPC_DISPLAY_ADDR] =
259 kstrdup(buf, GFP_KERNEL);
260 break;
261 case AF_INET:
262 (void)rpc_ntop(sap, buf, sizeof(buf));
263 xprt->address_strings[RPC_DISPLAY_ADDR] =
264 kstrdup(buf, GFP_KERNEL);
265 sin = xs_addr_in(xprt);
266 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
267 break;
268 case AF_INET6:
269 (void)rpc_ntop(sap, buf, sizeof(buf));
270 xprt->address_strings[RPC_DISPLAY_ADDR] =
271 kstrdup(buf, GFP_KERNEL);
272 sin6 = xs_addr_in6(xprt);
273 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
274 break;
275 default:
276 BUG();
277 }
278
279 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
280}
281
282static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
283{
284 struct sockaddr *sap = xs_addr(xprt);
285 char buf[128];
286
287 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
288 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
289
290 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
291 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
292}
293
294static void xs_format_peer_addresses(struct rpc_xprt *xprt,
295 const char *protocol,
296 const char *netid)
297{
298 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
299 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
300 xs_format_common_peer_addresses(xprt);
301 xs_format_common_peer_ports(xprt);
302}
303
304static void xs_update_peer_port(struct rpc_xprt *xprt)
305{
306 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
307 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
308
309 xs_format_common_peer_ports(xprt);
310}
311
312static void xs_free_peer_addresses(struct rpc_xprt *xprt)
313{
314 unsigned int i;
315
316 for (i = 0; i < RPC_DISPLAY_MAX; i++)
317 switch (i) {
318 case RPC_DISPLAY_PROTO:
319 case RPC_DISPLAY_NETID:
320 continue;
321 default:
322 kfree(xprt->address_strings[i]);
323 }
324}
325
326static size_t
327xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp)
328{
329 size_t i,n;
330
331 if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES))
332 return want;
333 n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT;
334 for (i = 0; i < n; i++) {
335 if (buf->pages[i])
336 continue;
337 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp);
338 if (!buf->pages[i]) {
339 i *= PAGE_SIZE;
340 return i > buf->page_base ? i - buf->page_base : 0;
341 }
342 }
343 return want;
344}
345
346static ssize_t
347xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek)
348{
349 ssize_t ret;
350 if (seek != 0)
351 iov_iter_advance(&msg->msg_iter, seek);
352 ret = sock_recvmsg(sock, msg, flags);
353 return ret > 0 ? ret + seek : ret;
354}
355
356static ssize_t
357xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags,
358 struct kvec *kvec, size_t count, size_t seek)
359{
360 iov_iter_kvec(&msg->msg_iter, READ, kvec, 1, count);
361 return xs_sock_recvmsg(sock, msg, flags, seek);
362}
363
364static ssize_t
365xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags,
366 struct bio_vec *bvec, unsigned long nr, size_t count,
367 size_t seek)
368{
369 iov_iter_bvec(&msg->msg_iter, READ, bvec, nr, count);
370 return xs_sock_recvmsg(sock, msg, flags, seek);
371}
372
373static ssize_t
374xs_read_discard(struct socket *sock, struct msghdr *msg, int flags,
375 size_t count)
376{
377 iov_iter_discard(&msg->msg_iter, READ, count);
378 return sock_recvmsg(sock, msg, flags);
379}
380
381#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
382static void
383xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
384{
385 struct bvec_iter bi = {
386 .bi_size = count,
387 };
388 struct bio_vec bv;
389
390 bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
391 for_each_bvec(bv, bvec, bi, bi)
392 flush_dcache_page(bv.bv_page);
393}
394#else
395static inline void
396xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
397{
398}
399#endif
400
401static ssize_t
402xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
403 struct xdr_buf *buf, size_t count, size_t seek, size_t *read)
404{
405 size_t want, seek_init = seek, offset = 0;
406 ssize_t ret;
407
408 want = min_t(size_t, count, buf->head[0].iov_len);
409 if (seek < want) {
410 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek);
411 if (ret <= 0)
412 goto sock_err;
413 offset += ret;
414 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
415 goto out;
416 if (ret != want)
417 goto out;
418 seek = 0;
419 } else {
420 seek -= want;
421 offset += want;
422 }
423
424 want = xs_alloc_sparse_pages(buf,
425 min_t(size_t, count - offset, buf->page_len),
426 GFP_KERNEL);
427 if (seek < want) {
428 ret = xs_read_bvec(sock, msg, flags, buf->bvec,
429 xdr_buf_pagecount(buf),
430 want + buf->page_base,
431 seek + buf->page_base);
432 if (ret <= 0)
433 goto sock_err;
434 xs_flush_bvec(buf->bvec, ret, seek + buf->page_base);
435 offset += ret - buf->page_base;
436 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
437 goto out;
438 if (ret != want)
439 goto out;
440 seek = 0;
441 } else {
442 seek -= want;
443 offset += want;
444 }
445
446 want = min_t(size_t, count - offset, buf->tail[0].iov_len);
447 if (seek < want) {
448 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek);
449 if (ret <= 0)
450 goto sock_err;
451 offset += ret;
452 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
453 goto out;
454 if (ret != want)
455 goto out;
456 } else if (offset < seek_init)
457 offset = seek_init;
458 ret = -EMSGSIZE;
459out:
460 *read = offset - seek_init;
461 return ret;
462sock_err:
463 offset += seek;
464 goto out;
465}
466
467static void
468xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf)
469{
470 if (!transport->recv.copied) {
471 if (buf->head[0].iov_len >= transport->recv.offset)
472 memcpy(buf->head[0].iov_base,
473 &transport->recv.xid,
474 transport->recv.offset);
475 transport->recv.copied = transport->recv.offset;
476 }
477}
478
479static bool
480xs_read_stream_request_done(struct sock_xprt *transport)
481{
482 return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT);
483}
484
485static void
486xs_read_stream_check_eor(struct sock_xprt *transport,
487 struct msghdr *msg)
488{
489 if (xs_read_stream_request_done(transport))
490 msg->msg_flags |= MSG_EOR;
491}
492
493static ssize_t
494xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg,
495 int flags, struct rpc_rqst *req)
496{
497 struct xdr_buf *buf = &req->rq_private_buf;
498 size_t want, uninitialized_var(read);
499 ssize_t uninitialized_var(ret);
500
501 xs_read_header(transport, buf);
502
503 want = transport->recv.len - transport->recv.offset;
504 if (want != 0) {
505 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf,
506 transport->recv.copied + want,
507 transport->recv.copied,
508 &read);
509 transport->recv.offset += read;
510 transport->recv.copied += read;
511 }
512
513 if (transport->recv.offset == transport->recv.len)
514 xs_read_stream_check_eor(transport, msg);
515
516 if (want == 0)
517 return 0;
518
519 switch (ret) {
520 default:
521 break;
522 case -EFAULT:
523 case -EMSGSIZE:
524 msg->msg_flags |= MSG_TRUNC;
525 return read;
526 case 0:
527 return -ESHUTDOWN;
528 }
529 return ret < 0 ? ret : read;
530}
531
532static size_t
533xs_read_stream_headersize(bool isfrag)
534{
535 if (isfrag)
536 return sizeof(__be32);
537 return 3 * sizeof(__be32);
538}
539
540static ssize_t
541xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg,
542 int flags, size_t want, size_t seek)
543{
544 struct kvec kvec = {
545 .iov_base = &transport->recv.fraghdr,
546 .iov_len = want,
547 };
548 return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek);
549}
550
551#if defined(CONFIG_SUNRPC_BACKCHANNEL)
552static ssize_t
553xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
554{
555 struct rpc_xprt *xprt = &transport->xprt;
556 struct rpc_rqst *req;
557 ssize_t ret;
558
559 /* Look up and lock the request corresponding to the given XID */
560 req = xprt_lookup_bc_request(xprt, transport->recv.xid);
561 if (!req) {
562 printk(KERN_WARNING "Callback slot table overflowed\n");
563 return -ESHUTDOWN;
564 }
565 if (transport->recv.copied && !req->rq_private_buf.len)
566 return -ESHUTDOWN;
567
568 ret = xs_read_stream_request(transport, msg, flags, req);
569 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
570 xprt_complete_bc_request(req, transport->recv.copied);
571 else
572 req->rq_private_buf.len = transport->recv.copied;
573
574 return ret;
575}
576#else /* CONFIG_SUNRPC_BACKCHANNEL */
577static ssize_t
578xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
579{
580 return -ESHUTDOWN;
581}
582#endif /* CONFIG_SUNRPC_BACKCHANNEL */
583
584static ssize_t
585xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags)
586{
587 struct rpc_xprt *xprt = &transport->xprt;
588 struct rpc_rqst *req;
589 ssize_t ret = 0;
590
591 /* Look up and lock the request corresponding to the given XID */
592 spin_lock(&xprt->queue_lock);
593 req = xprt_lookup_rqst(xprt, transport->recv.xid);
594 if (!req || (transport->recv.copied && !req->rq_private_buf.len)) {
595 msg->msg_flags |= MSG_TRUNC;
596 goto out;
597 }
598 xprt_pin_rqst(req);
599 spin_unlock(&xprt->queue_lock);
600
601 ret = xs_read_stream_request(transport, msg, flags, req);
602
603 spin_lock(&xprt->queue_lock);
604 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
605 xprt_complete_rqst(req->rq_task, transport->recv.copied);
606 else
607 req->rq_private_buf.len = transport->recv.copied;
608 xprt_unpin_rqst(req);
609out:
610 spin_unlock(&xprt->queue_lock);
611 return ret;
612}
613
614static ssize_t
615xs_read_stream(struct sock_xprt *transport, int flags)
616{
617 struct msghdr msg = { 0 };
618 size_t want, read = 0;
619 ssize_t ret = 0;
620
621 if (transport->recv.len == 0) {
622 want = xs_read_stream_headersize(transport->recv.copied != 0);
623 ret = xs_read_stream_header(transport, &msg, flags, want,
624 transport->recv.offset);
625 if (ret <= 0)
626 goto out_err;
627 transport->recv.offset = ret;
628 if (transport->recv.offset != want)
629 return transport->recv.offset;
630 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) &
631 RPC_FRAGMENT_SIZE_MASK;
632 transport->recv.offset -= sizeof(transport->recv.fraghdr);
633 read = ret;
634 }
635
636 switch (be32_to_cpu(transport->recv.calldir)) {
637 default:
638 msg.msg_flags |= MSG_TRUNC;
639 break;
640 case RPC_CALL:
641 ret = xs_read_stream_call(transport, &msg, flags);
642 break;
643 case RPC_REPLY:
644 ret = xs_read_stream_reply(transport, &msg, flags);
645 }
646 if (msg.msg_flags & MSG_TRUNC) {
647 transport->recv.calldir = cpu_to_be32(-1);
648 transport->recv.copied = -1;
649 }
650 if (ret < 0)
651 goto out_err;
652 read += ret;
653 if (transport->recv.offset < transport->recv.len) {
654 if (!(msg.msg_flags & MSG_TRUNC))
655 return read;
656 msg.msg_flags = 0;
657 ret = xs_read_discard(transport->sock, &msg, flags,
658 transport->recv.len - transport->recv.offset);
659 if (ret <= 0)
660 goto out_err;
661 transport->recv.offset += ret;
662 read += ret;
663 if (transport->recv.offset != transport->recv.len)
664 return read;
665 }
666 if (xs_read_stream_request_done(transport)) {
667 trace_xs_stream_read_request(transport);
668 transport->recv.copied = 0;
669 }
670 transport->recv.offset = 0;
671 transport->recv.len = 0;
672 return read;
673out_err:
674 return ret != 0 ? ret : -ESHUTDOWN;
675}
676
677static __poll_t xs_poll_socket(struct sock_xprt *transport)
678{
679 return transport->sock->ops->poll(transport->file, transport->sock,
680 NULL);
681}
682
683static bool xs_poll_socket_readable(struct sock_xprt *transport)
684{
685 __poll_t events = xs_poll_socket(transport);
686
687 return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP);
688}
689
690static void xs_poll_check_readable(struct sock_xprt *transport)
691{
692
693 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
694 if (!xs_poll_socket_readable(transport))
695 return;
696 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
697 queue_work(xprtiod_workqueue, &transport->recv_worker);
698}
699
700static void xs_stream_data_receive(struct sock_xprt *transport)
701{
702 size_t read = 0;
703 ssize_t ret = 0;
704
705 mutex_lock(&transport->recv_mutex);
706 if (transport->sock == NULL)
707 goto out;
708 for (;;) {
709 ret = xs_read_stream(transport, MSG_DONTWAIT);
710 if (ret < 0)
711 break;
712 read += ret;
713 cond_resched();
714 }
715 if (ret == -ESHUTDOWN)
716 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
717 else
718 xs_poll_check_readable(transport);
719out:
720 mutex_unlock(&transport->recv_mutex);
721 trace_xs_stream_read_data(&transport->xprt, ret, read);
722}
723
724static void xs_stream_data_receive_workfn(struct work_struct *work)
725{
726 struct sock_xprt *transport =
727 container_of(work, struct sock_xprt, recv_worker);
728 unsigned int pflags = memalloc_nofs_save();
729
730 xs_stream_data_receive(transport);
731 memalloc_nofs_restore(pflags);
732}
733
734static void
735xs_stream_reset_connect(struct sock_xprt *transport)
736{
737 transport->recv.offset = 0;
738 transport->recv.len = 0;
739 transport->recv.copied = 0;
740 transport->xmit.offset = 0;
741}
742
743static void
744xs_stream_start_connect(struct sock_xprt *transport)
745{
746 transport->xprt.stat.connect_count++;
747 transport->xprt.stat.connect_start = jiffies;
748}
749
750#define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
751
752static int xs_sendmsg(struct socket *sock, struct msghdr *msg, size_t seek)
753{
754 if (seek)
755 iov_iter_advance(&msg->msg_iter, seek);
756 return sock_sendmsg(sock, msg);
757}
758
759static int xs_send_kvec(struct socket *sock, struct msghdr *msg, struct kvec *vec, size_t seek)
760{
761 iov_iter_kvec(&msg->msg_iter, WRITE, vec, 1, vec->iov_len);
762 return xs_sendmsg(sock, msg, seek);
763}
764
765static int xs_send_pagedata(struct socket *sock, struct msghdr *msg, struct xdr_buf *xdr, size_t base)
766{
767 int err;
768
769 err = xdr_alloc_bvec(xdr, GFP_KERNEL);
770 if (err < 0)
771 return err;
772
773 iov_iter_bvec(&msg->msg_iter, WRITE, xdr->bvec,
774 xdr_buf_pagecount(xdr),
775 xdr->page_len + xdr->page_base);
776 return xs_sendmsg(sock, msg, base + xdr->page_base);
777}
778
779#define xs_record_marker_len() sizeof(rpc_fraghdr)
780
781/* Common case:
782 * - stream transport
783 * - sending from byte 0 of the message
784 * - the message is wholly contained in @xdr's head iovec
785 */
786static int xs_send_rm_and_kvec(struct socket *sock, struct msghdr *msg,
787 rpc_fraghdr marker, struct kvec *vec, size_t base)
788{
789 struct kvec iov[2] = {
790 [0] = {
791 .iov_base = &marker,
792 .iov_len = sizeof(marker)
793 },
794 [1] = *vec,
795 };
796 size_t len = iov[0].iov_len + iov[1].iov_len;
797
798 iov_iter_kvec(&msg->msg_iter, WRITE, iov, 2, len);
799 return xs_sendmsg(sock, msg, base);
800}
801
802/**
803 * xs_sendpages - write pages directly to a socket
804 * @sock: socket to send on
805 * @addr: UDP only -- address of destination
806 * @addrlen: UDP only -- length of destination address
807 * @xdr: buffer containing this request
808 * @base: starting position in the buffer
809 * @rm: stream record marker field
810 * @sent_p: return the total number of bytes successfully queued for sending
811 *
812 */
813static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, rpc_fraghdr rm, int *sent_p)
814{
815 struct msghdr msg = {
816 .msg_name = addr,
817 .msg_namelen = addrlen,
818 .msg_flags = XS_SENDMSG_FLAGS | MSG_MORE,
819 };
820 unsigned int rmsize = rm ? sizeof(rm) : 0;
821 unsigned int remainder = rmsize + xdr->len - base;
822 unsigned int want;
823 int err = 0;
824
825 if (unlikely(!sock))
826 return -ENOTSOCK;
827
828 want = xdr->head[0].iov_len + rmsize;
829 if (base < want) {
830 unsigned int len = want - base;
831 remainder -= len;
832 if (remainder == 0)
833 msg.msg_flags &= ~MSG_MORE;
834 if (rmsize)
835 err = xs_send_rm_and_kvec(sock, &msg, rm,
836 &xdr->head[0], base);
837 else
838 err = xs_send_kvec(sock, &msg, &xdr->head[0], base);
839 if (remainder == 0 || err != len)
840 goto out;
841 *sent_p += err;
842 base = 0;
843 } else
844 base -= want;
845
846 if (base < xdr->page_len) {
847 unsigned int len = xdr->page_len - base;
848 remainder -= len;
849 if (remainder == 0)
850 msg.msg_flags &= ~MSG_MORE;
851 err = xs_send_pagedata(sock, &msg, xdr, base);
852 if (remainder == 0 || err != len)
853 goto out;
854 *sent_p += err;
855 base = 0;
856 } else
857 base -= xdr->page_len;
858
859 if (base >= xdr->tail[0].iov_len)
860 return 0;
861 msg.msg_flags &= ~MSG_MORE;
862 err = xs_send_kvec(sock, &msg, &xdr->tail[0], base);
863out:
864 if (err > 0) {
865 *sent_p += err;
866 err = 0;
867 }
868 return err;
869}
870
871/**
872 * xs_nospace - handle transmit was incomplete
873 * @req: pointer to RPC request
874 *
875 */
876static int xs_nospace(struct rpc_rqst *req)
877{
878 struct rpc_xprt *xprt = req->rq_xprt;
879 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
880 struct sock *sk = transport->inet;
881 int ret = -EAGAIN;
882
883 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
884 req->rq_task->tk_pid,
885 req->rq_slen - transport->xmit.offset,
886 req->rq_slen);
887
888 /* Protect against races with write_space */
889 spin_lock(&xprt->transport_lock);
890
891 /* Don't race with disconnect */
892 if (xprt_connected(xprt)) {
893 /* wait for more buffer space */
894 sk->sk_write_pending++;
895 xprt_wait_for_buffer_space(xprt);
896 } else
897 ret = -ENOTCONN;
898
899 spin_unlock(&xprt->transport_lock);
900
901 /* Race breaker in case memory is freed before above code is called */
902 if (ret == -EAGAIN) {
903 struct socket_wq *wq;
904
905 rcu_read_lock();
906 wq = rcu_dereference(sk->sk_wq);
907 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
908 rcu_read_unlock();
909
910 sk->sk_write_space(sk);
911 }
912 return ret;
913}
914
915static void
916xs_stream_prepare_request(struct rpc_rqst *req)
917{
918 xdr_free_bvec(&req->rq_rcv_buf);
919 req->rq_task->tk_status = xdr_alloc_bvec(&req->rq_rcv_buf, GFP_KERNEL);
920}
921
922/*
923 * Determine if the previous message in the stream was aborted before it
924 * could complete transmission.
925 */
926static bool
927xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req)
928{
929 return transport->xmit.offset != 0 && req->rq_bytes_sent == 0;
930}
931
932/*
933 * Return the stream record marker field for a record of length < 2^31-1
934 */
935static rpc_fraghdr
936xs_stream_record_marker(struct xdr_buf *xdr)
937{
938 if (!xdr->len)
939 return 0;
940 return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len);
941}
942
943/**
944 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
945 * @req: pointer to RPC request
946 *
947 * Return values:
948 * 0: The request has been sent
949 * EAGAIN: The socket was blocked, please call again later to
950 * complete the request
951 * ENOTCONN: Caller needs to invoke connect logic then call again
952 * other: Some other error occured, the request was not sent
953 */
954static int xs_local_send_request(struct rpc_rqst *req)
955{
956 struct rpc_xprt *xprt = req->rq_xprt;
957 struct sock_xprt *transport =
958 container_of(xprt, struct sock_xprt, xprt);
959 struct xdr_buf *xdr = &req->rq_snd_buf;
960 rpc_fraghdr rm = xs_stream_record_marker(xdr);
961 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
962 int status;
963 int sent = 0;
964
965 /* Close the stream if the previous transmission was incomplete */
966 if (xs_send_request_was_aborted(transport, req)) {
967 xs_close(xprt);
968 return -ENOTCONN;
969 }
970
971 xs_pktdump("packet data:",
972 req->rq_svec->iov_base, req->rq_svec->iov_len);
973
974 req->rq_xtime = ktime_get();
975 status = xs_sendpages(transport->sock, NULL, 0, xdr,
976 transport->xmit.offset, rm, &sent);
977 dprintk("RPC: %s(%u) = %d\n",
978 __func__, xdr->len - transport->xmit.offset, status);
979
980 if (status == -EAGAIN && sock_writeable(transport->inet))
981 status = -ENOBUFS;
982
983 if (likely(sent > 0) || status == 0) {
984 transport->xmit.offset += sent;
985 req->rq_bytes_sent = transport->xmit.offset;
986 if (likely(req->rq_bytes_sent >= msglen)) {
987 req->rq_xmit_bytes_sent += transport->xmit.offset;
988 transport->xmit.offset = 0;
989 return 0;
990 }
991 status = -EAGAIN;
992 }
993
994 switch (status) {
995 case -ENOBUFS:
996 break;
997 case -EAGAIN:
998 status = xs_nospace(req);
999 break;
1000 default:
1001 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1002 -status);
1003 /* fall through */
1004 case -EPIPE:
1005 xs_close(xprt);
1006 status = -ENOTCONN;
1007 }
1008
1009 return status;
1010}
1011
1012/**
1013 * xs_udp_send_request - write an RPC request to a UDP socket
1014 * @req: pointer to RPC request
1015 *
1016 * Return values:
1017 * 0: The request has been sent
1018 * EAGAIN: The socket was blocked, please call again later to
1019 * complete the request
1020 * ENOTCONN: Caller needs to invoke connect logic then call again
1021 * other: Some other error occurred, the request was not sent
1022 */
1023static int xs_udp_send_request(struct rpc_rqst *req)
1024{
1025 struct rpc_xprt *xprt = req->rq_xprt;
1026 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1027 struct xdr_buf *xdr = &req->rq_snd_buf;
1028 int sent = 0;
1029 int status;
1030
1031 xs_pktdump("packet data:",
1032 req->rq_svec->iov_base,
1033 req->rq_svec->iov_len);
1034
1035 if (!xprt_bound(xprt))
1036 return -ENOTCONN;
1037
1038 if (!xprt_request_get_cong(xprt, req))
1039 return -EBADSLT;
1040
1041 req->rq_xtime = ktime_get();
1042 status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
1043 xdr, 0, 0, &sent);
1044
1045 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
1046 xdr->len, status);
1047
1048 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
1049 if (status == -EPERM)
1050 goto process_status;
1051
1052 if (status == -EAGAIN && sock_writeable(transport->inet))
1053 status = -ENOBUFS;
1054
1055 if (sent > 0 || status == 0) {
1056 req->rq_xmit_bytes_sent += sent;
1057 if (sent >= req->rq_slen)
1058 return 0;
1059 /* Still some bytes left; set up for a retry later. */
1060 status = -EAGAIN;
1061 }
1062
1063process_status:
1064 switch (status) {
1065 case -ENOTSOCK:
1066 status = -ENOTCONN;
1067 /* Should we call xs_close() here? */
1068 break;
1069 case -EAGAIN:
1070 status = xs_nospace(req);
1071 break;
1072 case -ENETUNREACH:
1073 case -ENOBUFS:
1074 case -EPIPE:
1075 case -ECONNREFUSED:
1076 case -EPERM:
1077 /* When the server has died, an ICMP port unreachable message
1078 * prompts ECONNREFUSED. */
1079 break;
1080 default:
1081 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1082 -status);
1083 }
1084
1085 return status;
1086}
1087
1088/**
1089 * xs_tcp_send_request - write an RPC request to a TCP socket
1090 * @req: pointer to RPC request
1091 *
1092 * Return values:
1093 * 0: The request has been sent
1094 * EAGAIN: The socket was blocked, please call again later to
1095 * complete the request
1096 * ENOTCONN: Caller needs to invoke connect logic then call again
1097 * other: Some other error occurred, the request was not sent
1098 *
1099 * XXX: In the case of soft timeouts, should we eventually give up
1100 * if sendmsg is not able to make progress?
1101 */
1102static int xs_tcp_send_request(struct rpc_rqst *req)
1103{
1104 struct rpc_xprt *xprt = req->rq_xprt;
1105 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1106 struct xdr_buf *xdr = &req->rq_snd_buf;
1107 rpc_fraghdr rm = xs_stream_record_marker(xdr);
1108 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
1109 bool vm_wait = false;
1110 int status;
1111 int sent;
1112
1113 /* Close the stream if the previous transmission was incomplete */
1114 if (xs_send_request_was_aborted(transport, req)) {
1115 if (transport->sock != NULL)
1116 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
1117 return -ENOTCONN;
1118 }
1119
1120 xs_pktdump("packet data:",
1121 req->rq_svec->iov_base,
1122 req->rq_svec->iov_len);
1123
1124 if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
1125 xs_tcp_set_socket_timeouts(xprt, transport->sock);
1126
1127 /* Continue transmitting the packet/record. We must be careful
1128 * to cope with writespace callbacks arriving _after_ we have
1129 * called sendmsg(). */
1130 req->rq_xtime = ktime_get();
1131 while (1) {
1132 sent = 0;
1133 status = xs_sendpages(transport->sock, NULL, 0, xdr,
1134 transport->xmit.offset, rm, &sent);
1135
1136 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
1137 xdr->len - transport->xmit.offset, status);
1138
1139 /* If we've sent the entire packet, immediately
1140 * reset the count of bytes sent. */
1141 transport->xmit.offset += sent;
1142 req->rq_bytes_sent = transport->xmit.offset;
1143 if (likely(req->rq_bytes_sent >= msglen)) {
1144 req->rq_xmit_bytes_sent += transport->xmit.offset;
1145 transport->xmit.offset = 0;
1146 return 0;
1147 }
1148
1149 WARN_ON_ONCE(sent == 0 && status == 0);
1150
1151 if (status == -EAGAIN ) {
1152 /*
1153 * Return EAGAIN if we're sure we're hitting the
1154 * socket send buffer limits.
1155 */
1156 if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
1157 break;
1158 /*
1159 * Did we hit a memory allocation failure?
1160 */
1161 if (sent == 0) {
1162 status = -ENOBUFS;
1163 if (vm_wait)
1164 break;
1165 /* Retry, knowing now that we're below the
1166 * socket send buffer limit
1167 */
1168 vm_wait = true;
1169 }
1170 continue;
1171 }
1172 if (status < 0)
1173 break;
1174 vm_wait = false;
1175 }
1176
1177 switch (status) {
1178 case -ENOTSOCK:
1179 status = -ENOTCONN;
1180 /* Should we call xs_close() here? */
1181 break;
1182 case -EAGAIN:
1183 status = xs_nospace(req);
1184 break;
1185 case -ECONNRESET:
1186 case -ECONNREFUSED:
1187 case -ENOTCONN:
1188 case -EADDRINUSE:
1189 case -ENOBUFS:
1190 case -EPIPE:
1191 break;
1192 default:
1193 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1194 -status);
1195 }
1196
1197 return status;
1198}
1199
1200static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1201{
1202 transport->old_data_ready = sk->sk_data_ready;
1203 transport->old_state_change = sk->sk_state_change;
1204 transport->old_write_space = sk->sk_write_space;
1205 transport->old_error_report = sk->sk_error_report;
1206}
1207
1208static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1209{
1210 sk->sk_data_ready = transport->old_data_ready;
1211 sk->sk_state_change = transport->old_state_change;
1212 sk->sk_write_space = transport->old_write_space;
1213 sk->sk_error_report = transport->old_error_report;
1214}
1215
1216static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
1217{
1218 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1219
1220 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1221 clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state);
1222 clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state);
1223 clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state);
1224}
1225
1226static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr)
1227{
1228 set_bit(nr, &transport->sock_state);
1229 queue_work(xprtiod_workqueue, &transport->error_worker);
1230}
1231
1232static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1233{
1234 smp_mb__before_atomic();
1235 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1236 clear_bit(XPRT_CLOSING, &xprt->state);
1237 xs_sock_reset_state_flags(xprt);
1238 smp_mb__after_atomic();
1239}
1240
1241/**
1242 * xs_error_report - callback to handle TCP socket state errors
1243 * @sk: socket
1244 *
1245 * Note: we don't call sock_error() since there may be a rpc_task
1246 * using the socket, and so we don't want to clear sk->sk_err.
1247 */
1248static void xs_error_report(struct sock *sk)
1249{
1250 struct sock_xprt *transport;
1251 struct rpc_xprt *xprt;
1252
1253 read_lock_bh(&sk->sk_callback_lock);
1254 if (!(xprt = xprt_from_sock(sk)))
1255 goto out;
1256
1257 transport = container_of(xprt, struct sock_xprt, xprt);
1258 transport->xprt_err = -sk->sk_err;
1259 if (transport->xprt_err == 0)
1260 goto out;
1261 dprintk("RPC: xs_error_report client %p, error=%d...\n",
1262 xprt, -transport->xprt_err);
1263 trace_rpc_socket_error(xprt, sk->sk_socket, transport->xprt_err);
1264
1265 /* barrier ensures xprt_err is set before XPRT_SOCK_WAKE_ERROR */
1266 smp_mb__before_atomic();
1267 xs_run_error_worker(transport, XPRT_SOCK_WAKE_ERROR);
1268 out:
1269 read_unlock_bh(&sk->sk_callback_lock);
1270}
1271
1272static void xs_reset_transport(struct sock_xprt *transport)
1273{
1274 struct socket *sock = transport->sock;
1275 struct sock *sk = transport->inet;
1276 struct rpc_xprt *xprt = &transport->xprt;
1277 struct file *filp = transport->file;
1278
1279 if (sk == NULL)
1280 return;
1281
1282 if (atomic_read(&transport->xprt.swapper))
1283 sk_clear_memalloc(sk);
1284
1285 kernel_sock_shutdown(sock, SHUT_RDWR);
1286
1287 mutex_lock(&transport->recv_mutex);
1288 write_lock_bh(&sk->sk_callback_lock);
1289 transport->inet = NULL;
1290 transport->sock = NULL;
1291 transport->file = NULL;
1292
1293 sk->sk_user_data = NULL;
1294
1295 xs_restore_old_callbacks(transport, sk);
1296 xprt_clear_connected(xprt);
1297 write_unlock_bh(&sk->sk_callback_lock);
1298 xs_sock_reset_connection_flags(xprt);
1299 /* Reset stream record info */
1300 xs_stream_reset_connect(transport);
1301 mutex_unlock(&transport->recv_mutex);
1302
1303 trace_rpc_socket_close(xprt, sock);
1304 fput(filp);
1305
1306 xprt_disconnect_done(xprt);
1307}
1308
1309/**
1310 * xs_close - close a socket
1311 * @xprt: transport
1312 *
1313 * This is used when all requests are complete; ie, no DRC state remains
1314 * on the server we want to save.
1315 *
1316 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1317 * xs_reset_transport() zeroing the socket from underneath a writer.
1318 */
1319static void xs_close(struct rpc_xprt *xprt)
1320{
1321 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1322
1323 dprintk("RPC: xs_close xprt %p\n", xprt);
1324
1325 xs_reset_transport(transport);
1326 xprt->reestablish_timeout = 0;
1327}
1328
1329static void xs_inject_disconnect(struct rpc_xprt *xprt)
1330{
1331 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
1332 xprt);
1333 xprt_disconnect_done(xprt);
1334}
1335
1336static void xs_xprt_free(struct rpc_xprt *xprt)
1337{
1338 xs_free_peer_addresses(xprt);
1339 xprt_free(xprt);
1340}
1341
1342/**
1343 * xs_destroy - prepare to shutdown a transport
1344 * @xprt: doomed transport
1345 *
1346 */
1347static void xs_destroy(struct rpc_xprt *xprt)
1348{
1349 struct sock_xprt *transport = container_of(xprt,
1350 struct sock_xprt, xprt);
1351 dprintk("RPC: xs_destroy xprt %p\n", xprt);
1352
1353 cancel_delayed_work_sync(&transport->connect_worker);
1354 xs_close(xprt);
1355 cancel_work_sync(&transport->recv_worker);
1356 cancel_work_sync(&transport->error_worker);
1357 xs_xprt_free(xprt);
1358 module_put(THIS_MODULE);
1359}
1360
1361/**
1362 * xs_udp_data_read_skb - receive callback for UDP sockets
1363 * @xprt: transport
1364 * @sk: socket
1365 * @skb: skbuff
1366 *
1367 */
1368static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1369 struct sock *sk,
1370 struct sk_buff *skb)
1371{
1372 struct rpc_task *task;
1373 struct rpc_rqst *rovr;
1374 int repsize, copied;
1375 u32 _xid;
1376 __be32 *xp;
1377
1378 repsize = skb->len;
1379 if (repsize < 4) {
1380 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
1381 return;
1382 }
1383
1384 /* Copy the XID from the skb... */
1385 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1386 if (xp == NULL)
1387 return;
1388
1389 /* Look up and lock the request corresponding to the given XID */
1390 spin_lock(&xprt->queue_lock);
1391 rovr = xprt_lookup_rqst(xprt, *xp);
1392 if (!rovr)
1393 goto out_unlock;
1394 xprt_pin_rqst(rovr);
1395 xprt_update_rtt(rovr->rq_task);
1396 spin_unlock(&xprt->queue_lock);
1397 task = rovr->rq_task;
1398
1399 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1400 copied = repsize;
1401
1402 /* Suck it into the iovec, verify checksum if not done by hw. */
1403 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1404 spin_lock(&xprt->queue_lock);
1405 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1406 goto out_unpin;
1407 }
1408
1409
1410 spin_lock(&xprt->transport_lock);
1411 xprt_adjust_cwnd(xprt, task, copied);
1412 spin_unlock(&xprt->transport_lock);
1413 spin_lock(&xprt->queue_lock);
1414 xprt_complete_rqst(task, copied);
1415 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1416out_unpin:
1417 xprt_unpin_rqst(rovr);
1418 out_unlock:
1419 spin_unlock(&xprt->queue_lock);
1420}
1421
1422static void xs_udp_data_receive(struct sock_xprt *transport)
1423{
1424 struct sk_buff *skb;
1425 struct sock *sk;
1426 int err;
1427
1428 mutex_lock(&transport->recv_mutex);
1429 sk = transport->inet;
1430 if (sk == NULL)
1431 goto out;
1432 for (;;) {
1433 skb = skb_recv_udp(sk, 0, 1, &err);
1434 if (skb == NULL)
1435 break;
1436 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1437 consume_skb(skb);
1438 cond_resched();
1439 }
1440 xs_poll_check_readable(transport);
1441out:
1442 mutex_unlock(&transport->recv_mutex);
1443}
1444
1445static void xs_udp_data_receive_workfn(struct work_struct *work)
1446{
1447 struct sock_xprt *transport =
1448 container_of(work, struct sock_xprt, recv_worker);
1449 unsigned int pflags = memalloc_nofs_save();
1450
1451 xs_udp_data_receive(transport);
1452 memalloc_nofs_restore(pflags);
1453}
1454
1455/**
1456 * xs_data_ready - "data ready" callback for UDP sockets
1457 * @sk: socket with data to read
1458 *
1459 */
1460static void xs_data_ready(struct sock *sk)
1461{
1462 struct rpc_xprt *xprt;
1463
1464 read_lock_bh(&sk->sk_callback_lock);
1465 dprintk("RPC: xs_data_ready...\n");
1466 xprt = xprt_from_sock(sk);
1467 if (xprt != NULL) {
1468 struct sock_xprt *transport = container_of(xprt,
1469 struct sock_xprt, xprt);
1470 transport->old_data_ready(sk);
1471 /* Any data means we had a useful conversation, so
1472 * then we don't need to delay the next reconnect
1473 */
1474 if (xprt->reestablish_timeout)
1475 xprt->reestablish_timeout = 0;
1476 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1477 queue_work(xprtiod_workqueue, &transport->recv_worker);
1478 }
1479 read_unlock_bh(&sk->sk_callback_lock);
1480}
1481
1482/*
1483 * Helper function to force a TCP close if the server is sending
1484 * junk and/or it has put us in CLOSE_WAIT
1485 */
1486static void xs_tcp_force_close(struct rpc_xprt *xprt)
1487{
1488 xprt_force_disconnect(xprt);
1489}
1490
1491#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1492static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1493{
1494 return PAGE_SIZE;
1495}
1496#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1497
1498/**
1499 * xs_tcp_state_change - callback to handle TCP socket state changes
1500 * @sk: socket whose state has changed
1501 *
1502 */
1503static void xs_tcp_state_change(struct sock *sk)
1504{
1505 struct rpc_xprt *xprt;
1506 struct sock_xprt *transport;
1507
1508 read_lock_bh(&sk->sk_callback_lock);
1509 if (!(xprt = xprt_from_sock(sk)))
1510 goto out;
1511 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1512 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1513 sk->sk_state, xprt_connected(xprt),
1514 sock_flag(sk, SOCK_DEAD),
1515 sock_flag(sk, SOCK_ZAPPED),
1516 sk->sk_shutdown);
1517
1518 transport = container_of(xprt, struct sock_xprt, xprt);
1519 trace_rpc_socket_state_change(xprt, sk->sk_socket);
1520 switch (sk->sk_state) {
1521 case TCP_ESTABLISHED:
1522 if (!xprt_test_and_set_connected(xprt)) {
1523 xprt->connect_cookie++;
1524 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1525 xprt_clear_connecting(xprt);
1526
1527 xprt->stat.connect_count++;
1528 xprt->stat.connect_time += (long)jiffies -
1529 xprt->stat.connect_start;
1530 xs_run_error_worker(transport, XPRT_SOCK_WAKE_PENDING);
1531 }
1532 break;
1533 case TCP_FIN_WAIT1:
1534 /* The client initiated a shutdown of the socket */
1535 xprt->connect_cookie++;
1536 xprt->reestablish_timeout = 0;
1537 set_bit(XPRT_CLOSING, &xprt->state);
1538 smp_mb__before_atomic();
1539 clear_bit(XPRT_CONNECTED, &xprt->state);
1540 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1541 smp_mb__after_atomic();
1542 break;
1543 case TCP_CLOSE_WAIT:
1544 /* The server initiated a shutdown of the socket */
1545 xprt->connect_cookie++;
1546 clear_bit(XPRT_CONNECTED, &xprt->state);
1547 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1548 /* fall through */
1549 case TCP_CLOSING:
1550 /*
1551 * If the server closed down the connection, make sure that
1552 * we back off before reconnecting
1553 */
1554 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1555 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1556 break;
1557 case TCP_LAST_ACK:
1558 set_bit(XPRT_CLOSING, &xprt->state);
1559 smp_mb__before_atomic();
1560 clear_bit(XPRT_CONNECTED, &xprt->state);
1561 smp_mb__after_atomic();
1562 break;
1563 case TCP_CLOSE:
1564 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1565 &transport->sock_state))
1566 xprt_clear_connecting(xprt);
1567 clear_bit(XPRT_CLOSING, &xprt->state);
1568 /* Trigger the socket release */
1569 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1570 }
1571 out:
1572 read_unlock_bh(&sk->sk_callback_lock);
1573}
1574
1575static void xs_write_space(struct sock *sk)
1576{
1577 struct socket_wq *wq;
1578 struct sock_xprt *transport;
1579 struct rpc_xprt *xprt;
1580
1581 if (!sk->sk_socket)
1582 return;
1583 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1584
1585 if (unlikely(!(xprt = xprt_from_sock(sk))))
1586 return;
1587 transport = container_of(xprt, struct sock_xprt, xprt);
1588 rcu_read_lock();
1589 wq = rcu_dereference(sk->sk_wq);
1590 if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1591 goto out;
1592
1593 xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE);
1594 sk->sk_write_pending--;
1595out:
1596 rcu_read_unlock();
1597}
1598
1599/**
1600 * xs_udp_write_space - callback invoked when socket buffer space
1601 * becomes available
1602 * @sk: socket whose state has changed
1603 *
1604 * Called when more output buffer space is available for this socket.
1605 * We try not to wake our writers until they can make "significant"
1606 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1607 * with a bunch of small requests.
1608 */
1609static void xs_udp_write_space(struct sock *sk)
1610{
1611 read_lock_bh(&sk->sk_callback_lock);
1612
1613 /* from net/core/sock.c:sock_def_write_space */
1614 if (sock_writeable(sk))
1615 xs_write_space(sk);
1616
1617 read_unlock_bh(&sk->sk_callback_lock);
1618}
1619
1620/**
1621 * xs_tcp_write_space - callback invoked when socket buffer space
1622 * becomes available
1623 * @sk: socket whose state has changed
1624 *
1625 * Called when more output buffer space is available for this socket.
1626 * We try not to wake our writers until they can make "significant"
1627 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1628 * with a bunch of small requests.
1629 */
1630static void xs_tcp_write_space(struct sock *sk)
1631{
1632 read_lock_bh(&sk->sk_callback_lock);
1633
1634 /* from net/core/stream.c:sk_stream_write_space */
1635 if (sk_stream_is_writeable(sk))
1636 xs_write_space(sk);
1637
1638 read_unlock_bh(&sk->sk_callback_lock);
1639}
1640
1641static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1642{
1643 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1644 struct sock *sk = transport->inet;
1645
1646 if (transport->rcvsize) {
1647 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1648 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1649 }
1650 if (transport->sndsize) {
1651 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1652 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1653 sk->sk_write_space(sk);
1654 }
1655}
1656
1657/**
1658 * xs_udp_set_buffer_size - set send and receive limits
1659 * @xprt: generic transport
1660 * @sndsize: requested size of send buffer, in bytes
1661 * @rcvsize: requested size of receive buffer, in bytes
1662 *
1663 * Set socket send and receive buffer size limits.
1664 */
1665static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1666{
1667 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1668
1669 transport->sndsize = 0;
1670 if (sndsize)
1671 transport->sndsize = sndsize + 1024;
1672 transport->rcvsize = 0;
1673 if (rcvsize)
1674 transport->rcvsize = rcvsize + 1024;
1675
1676 xs_udp_do_set_buffer_size(xprt);
1677}
1678
1679/**
1680 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1681 * @xprt: controlling transport
1682 * @task: task that timed out
1683 *
1684 * Adjust the congestion window after a retransmit timeout has occurred.
1685 */
1686static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1687{
1688 spin_lock(&xprt->transport_lock);
1689 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1690 spin_unlock(&xprt->transport_lock);
1691}
1692
1693static int xs_get_random_port(void)
1694{
1695 unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1696 unsigned short range;
1697 unsigned short rand;
1698
1699 if (max < min)
1700 return -EADDRINUSE;
1701 range = max - min + 1;
1702 rand = (unsigned short) prandom_u32() % range;
1703 return rand + min;
1704}
1705
1706/**
1707 * xs_set_reuseaddr_port - set the socket's port and address reuse options
1708 * @sock: socket
1709 *
1710 * Note that this function has to be called on all sockets that share the
1711 * same port, and it must be called before binding.
1712 */
1713static void xs_sock_set_reuseport(struct socket *sock)
1714{
1715 int opt = 1;
1716
1717 kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1718 (char *)&opt, sizeof(opt));
1719}
1720
1721static unsigned short xs_sock_getport(struct socket *sock)
1722{
1723 struct sockaddr_storage buf;
1724 unsigned short port = 0;
1725
1726 if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1727 goto out;
1728 switch (buf.ss_family) {
1729 case AF_INET6:
1730 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1731 break;
1732 case AF_INET:
1733 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1734 }
1735out:
1736 return port;
1737}
1738
1739/**
1740 * xs_set_port - reset the port number in the remote endpoint address
1741 * @xprt: generic transport
1742 * @port: new port number
1743 *
1744 */
1745static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1746{
1747 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1748
1749 rpc_set_port(xs_addr(xprt), port);
1750 xs_update_peer_port(xprt);
1751}
1752
1753static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1754{
1755 if (transport->srcport == 0)
1756 transport->srcport = xs_sock_getport(sock);
1757}
1758
1759static int xs_get_srcport(struct sock_xprt *transport)
1760{
1761 int port = transport->srcport;
1762
1763 if (port == 0 && transport->xprt.resvport)
1764 port = xs_get_random_port();
1765 return port;
1766}
1767
1768static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1769{
1770 if (transport->srcport != 0)
1771 transport->srcport = 0;
1772 if (!transport->xprt.resvport)
1773 return 0;
1774 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1775 return xprt_max_resvport;
1776 return --port;
1777}
1778static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1779{
1780 struct sockaddr_storage myaddr;
1781 int err, nloop = 0;
1782 int port = xs_get_srcport(transport);
1783 unsigned short last;
1784
1785 /*
1786 * If we are asking for any ephemeral port (i.e. port == 0 &&
1787 * transport->xprt.resvport == 0), don't bind. Let the local
1788 * port selection happen implicitly when the socket is used
1789 * (for example at connect time).
1790 *
1791 * This ensures that we can continue to establish TCP
1792 * connections even when all local ephemeral ports are already
1793 * a part of some TCP connection. This makes no difference
1794 * for UDP sockets, but also doens't harm them.
1795 *
1796 * If we're asking for any reserved port (i.e. port == 0 &&
1797 * transport->xprt.resvport == 1) xs_get_srcport above will
1798 * ensure that port is non-zero and we will bind as needed.
1799 */
1800 if (port <= 0)
1801 return port;
1802
1803 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1804 do {
1805 rpc_set_port((struct sockaddr *)&myaddr, port);
1806 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1807 transport->xprt.addrlen);
1808 if (err == 0) {
1809 transport->srcport = port;
1810 break;
1811 }
1812 last = port;
1813 port = xs_next_srcport(transport, port);
1814 if (port > last)
1815 nloop++;
1816 } while (err == -EADDRINUSE && nloop != 2);
1817
1818 if (myaddr.ss_family == AF_INET)
1819 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1820 &((struct sockaddr_in *)&myaddr)->sin_addr,
1821 port, err ? "failed" : "ok", err);
1822 else
1823 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1824 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1825 port, err ? "failed" : "ok", err);
1826 return err;
1827}
1828
1829/*
1830 * We don't support autobind on AF_LOCAL sockets
1831 */
1832static void xs_local_rpcbind(struct rpc_task *task)
1833{
1834 xprt_set_bound(task->tk_xprt);
1835}
1836
1837static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1838{
1839}
1840
1841#ifdef CONFIG_DEBUG_LOCK_ALLOC
1842static struct lock_class_key xs_key[2];
1843static struct lock_class_key xs_slock_key[2];
1844
1845static inline void xs_reclassify_socketu(struct socket *sock)
1846{
1847 struct sock *sk = sock->sk;
1848
1849 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1850 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1851}
1852
1853static inline void xs_reclassify_socket4(struct socket *sock)
1854{
1855 struct sock *sk = sock->sk;
1856
1857 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1858 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1859}
1860
1861static inline void xs_reclassify_socket6(struct socket *sock)
1862{
1863 struct sock *sk = sock->sk;
1864
1865 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1866 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1867}
1868
1869static inline void xs_reclassify_socket(int family, struct socket *sock)
1870{
1871 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1872 return;
1873
1874 switch (family) {
1875 case AF_LOCAL:
1876 xs_reclassify_socketu(sock);
1877 break;
1878 case AF_INET:
1879 xs_reclassify_socket4(sock);
1880 break;
1881 case AF_INET6:
1882 xs_reclassify_socket6(sock);
1883 break;
1884 }
1885}
1886#else
1887static inline void xs_reclassify_socket(int family, struct socket *sock)
1888{
1889}
1890#endif
1891
1892static void xs_dummy_setup_socket(struct work_struct *work)
1893{
1894}
1895
1896static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1897 struct sock_xprt *transport, int family, int type,
1898 int protocol, bool reuseport)
1899{
1900 struct file *filp;
1901 struct socket *sock;
1902 int err;
1903
1904 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1905 if (err < 0) {
1906 dprintk("RPC: can't create %d transport socket (%d).\n",
1907 protocol, -err);
1908 goto out;
1909 }
1910 xs_reclassify_socket(family, sock);
1911
1912 if (reuseport)
1913 xs_sock_set_reuseport(sock);
1914
1915 err = xs_bind(transport, sock);
1916 if (err) {
1917 sock_release(sock);
1918 goto out;
1919 }
1920
1921 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1922 if (IS_ERR(filp))
1923 return ERR_CAST(filp);
1924 transport->file = filp;
1925
1926 return sock;
1927out:
1928 return ERR_PTR(err);
1929}
1930
1931static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1932 struct socket *sock)
1933{
1934 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1935 xprt);
1936
1937 if (!transport->inet) {
1938 struct sock *sk = sock->sk;
1939
1940 write_lock_bh(&sk->sk_callback_lock);
1941
1942 xs_save_old_callbacks(transport, sk);
1943
1944 sk->sk_user_data = xprt;
1945 sk->sk_data_ready = xs_data_ready;
1946 sk->sk_write_space = xs_udp_write_space;
1947 sock_set_flag(sk, SOCK_FASYNC);
1948 sk->sk_error_report = xs_error_report;
1949
1950 xprt_clear_connected(xprt);
1951
1952 /* Reset to new socket */
1953 transport->sock = sock;
1954 transport->inet = sk;
1955
1956 write_unlock_bh(&sk->sk_callback_lock);
1957 }
1958
1959 xs_stream_start_connect(transport);
1960
1961 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1962}
1963
1964/**
1965 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1966 * @transport: socket transport to connect
1967 */
1968static int xs_local_setup_socket(struct sock_xprt *transport)
1969{
1970 struct rpc_xprt *xprt = &transport->xprt;
1971 struct file *filp;
1972 struct socket *sock;
1973 int status = -EIO;
1974
1975 status = __sock_create(xprt->xprt_net, AF_LOCAL,
1976 SOCK_STREAM, 0, &sock, 1);
1977 if (status < 0) {
1978 dprintk("RPC: can't create AF_LOCAL "
1979 "transport socket (%d).\n", -status);
1980 goto out;
1981 }
1982 xs_reclassify_socket(AF_LOCAL, sock);
1983
1984 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1985 if (IS_ERR(filp)) {
1986 status = PTR_ERR(filp);
1987 goto out;
1988 }
1989 transport->file = filp;
1990
1991 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
1992 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1993
1994 status = xs_local_finish_connecting(xprt, sock);
1995 trace_rpc_socket_connect(xprt, sock, status);
1996 switch (status) {
1997 case 0:
1998 dprintk("RPC: xprt %p connected to %s\n",
1999 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2000 xprt->stat.connect_count++;
2001 xprt->stat.connect_time += (long)jiffies -
2002 xprt->stat.connect_start;
2003 xprt_set_connected(xprt);
2004 case -ENOBUFS:
2005 break;
2006 case -ENOENT:
2007 dprintk("RPC: xprt %p: socket %s does not exist\n",
2008 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2009 break;
2010 case -ECONNREFUSED:
2011 dprintk("RPC: xprt %p: connection refused for %s\n",
2012 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2013 break;
2014 default:
2015 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2016 __func__, -status,
2017 xprt->address_strings[RPC_DISPLAY_ADDR]);
2018 }
2019
2020out:
2021 xprt_clear_connecting(xprt);
2022 xprt_wake_pending_tasks(xprt, status);
2023 return status;
2024}
2025
2026static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2027{
2028 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2029 int ret;
2030
2031 if (RPC_IS_ASYNC(task)) {
2032 /*
2033 * We want the AF_LOCAL connect to be resolved in the
2034 * filesystem namespace of the process making the rpc
2035 * call. Thus we connect synchronously.
2036 *
2037 * If we want to support asynchronous AF_LOCAL calls,
2038 * we'll need to figure out how to pass a namespace to
2039 * connect.
2040 */
2041 task->tk_rpc_status = -ENOTCONN;
2042 rpc_exit(task, -ENOTCONN);
2043 return;
2044 }
2045 ret = xs_local_setup_socket(transport);
2046 if (ret && !RPC_IS_SOFTCONN(task))
2047 msleep_interruptible(15000);
2048}
2049
2050#if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2051/*
2052 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2053 * know that we have exclusive access to the socket), to guard against
2054 * races with xs_reset_transport.
2055 */
2056static void xs_set_memalloc(struct rpc_xprt *xprt)
2057{
2058 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2059 xprt);
2060
2061 /*
2062 * If there's no sock, then we have nothing to set. The
2063 * reconnecting process will get it for us.
2064 */
2065 if (!transport->inet)
2066 return;
2067 if (atomic_read(&xprt->swapper))
2068 sk_set_memalloc(transport->inet);
2069}
2070
2071/**
2072 * xs_enable_swap - Tag this transport as being used for swap.
2073 * @xprt: transport to tag
2074 *
2075 * Take a reference to this transport on behalf of the rpc_clnt, and
2076 * optionally mark it for swapping if it wasn't already.
2077 */
2078static int
2079xs_enable_swap(struct rpc_xprt *xprt)
2080{
2081 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2082
2083 if (atomic_inc_return(&xprt->swapper) != 1)
2084 return 0;
2085 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2086 return -ERESTARTSYS;
2087 if (xs->inet)
2088 sk_set_memalloc(xs->inet);
2089 xprt_release_xprt(xprt, NULL);
2090 return 0;
2091}
2092
2093/**
2094 * xs_disable_swap - Untag this transport as being used for swap.
2095 * @xprt: transport to tag
2096 *
2097 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2098 * swapper refcount goes to 0, untag the socket as a memalloc socket.
2099 */
2100static void
2101xs_disable_swap(struct rpc_xprt *xprt)
2102{
2103 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2104
2105 if (!atomic_dec_and_test(&xprt->swapper))
2106 return;
2107 if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2108 return;
2109 if (xs->inet)
2110 sk_clear_memalloc(xs->inet);
2111 xprt_release_xprt(xprt, NULL);
2112}
2113#else
2114static void xs_set_memalloc(struct rpc_xprt *xprt)
2115{
2116}
2117
2118static int
2119xs_enable_swap(struct rpc_xprt *xprt)
2120{
2121 return -EINVAL;
2122}
2123
2124static void
2125xs_disable_swap(struct rpc_xprt *xprt)
2126{
2127}
2128#endif
2129
2130static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2131{
2132 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2133
2134 if (!transport->inet) {
2135 struct sock *sk = sock->sk;
2136
2137 write_lock_bh(&sk->sk_callback_lock);
2138
2139 xs_save_old_callbacks(transport, sk);
2140
2141 sk->sk_user_data = xprt;
2142 sk->sk_data_ready = xs_data_ready;
2143 sk->sk_write_space = xs_udp_write_space;
2144 sock_set_flag(sk, SOCK_FASYNC);
2145
2146 xprt_set_connected(xprt);
2147
2148 /* Reset to new socket */
2149 transport->sock = sock;
2150 transport->inet = sk;
2151
2152 xs_set_memalloc(xprt);
2153
2154 write_unlock_bh(&sk->sk_callback_lock);
2155 }
2156 xs_udp_do_set_buffer_size(xprt);
2157
2158 xprt->stat.connect_start = jiffies;
2159}
2160
2161static void xs_udp_setup_socket(struct work_struct *work)
2162{
2163 struct sock_xprt *transport =
2164 container_of(work, struct sock_xprt, connect_worker.work);
2165 struct rpc_xprt *xprt = &transport->xprt;
2166 struct socket *sock;
2167 int status = -EIO;
2168
2169 sock = xs_create_sock(xprt, transport,
2170 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2171 IPPROTO_UDP, false);
2172 if (IS_ERR(sock))
2173 goto out;
2174
2175 dprintk("RPC: worker connecting xprt %p via %s to "
2176 "%s (port %s)\n", xprt,
2177 xprt->address_strings[RPC_DISPLAY_PROTO],
2178 xprt->address_strings[RPC_DISPLAY_ADDR],
2179 xprt->address_strings[RPC_DISPLAY_PORT]);
2180
2181 xs_udp_finish_connecting(xprt, sock);
2182 trace_rpc_socket_connect(xprt, sock, 0);
2183 status = 0;
2184out:
2185 xprt_clear_connecting(xprt);
2186 xprt_unlock_connect(xprt, transport);
2187 xprt_wake_pending_tasks(xprt, status);
2188}
2189
2190/**
2191 * xs_tcp_shutdown - gracefully shut down a TCP socket
2192 * @xprt: transport
2193 *
2194 * Initiates a graceful shutdown of the TCP socket by calling the
2195 * equivalent of shutdown(SHUT_RDWR);
2196 */
2197static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2198{
2199 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2200 struct socket *sock = transport->sock;
2201 int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2202
2203 if (sock == NULL)
2204 return;
2205 switch (skst) {
2206 default:
2207 kernel_sock_shutdown(sock, SHUT_RDWR);
2208 trace_rpc_socket_shutdown(xprt, sock);
2209 break;
2210 case TCP_CLOSE:
2211 case TCP_TIME_WAIT:
2212 xs_reset_transport(transport);
2213 }
2214}
2215
2216static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2217 struct socket *sock)
2218{
2219 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2220 unsigned int keepidle;
2221 unsigned int keepcnt;
2222 unsigned int opt_on = 1;
2223 unsigned int timeo;
2224
2225 spin_lock(&xprt->transport_lock);
2226 keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2227 keepcnt = xprt->timeout->to_retries + 1;
2228 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2229 (xprt->timeout->to_retries + 1);
2230 clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2231 spin_unlock(&xprt->transport_lock);
2232
2233 /* TCP Keepalive options */
2234 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2235 (char *)&opt_on, sizeof(opt_on));
2236 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2237 (char *)&keepidle, sizeof(keepidle));
2238 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2239 (char *)&keepidle, sizeof(keepidle));
2240 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2241 (char *)&keepcnt, sizeof(keepcnt));
2242
2243 /* TCP user timeout (see RFC5482) */
2244 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2245 (char *)&timeo, sizeof(timeo));
2246}
2247
2248static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2249 unsigned long connect_timeout,
2250 unsigned long reconnect_timeout)
2251{
2252 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2253 struct rpc_timeout to;
2254 unsigned long initval;
2255
2256 spin_lock(&xprt->transport_lock);
2257 if (reconnect_timeout < xprt->max_reconnect_timeout)
2258 xprt->max_reconnect_timeout = reconnect_timeout;
2259 if (connect_timeout < xprt->connect_timeout) {
2260 memcpy(&to, xprt->timeout, sizeof(to));
2261 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2262 /* Arbitrary lower limit */
2263 if (initval < XS_TCP_INIT_REEST_TO << 1)
2264 initval = XS_TCP_INIT_REEST_TO << 1;
2265 to.to_initval = initval;
2266 to.to_maxval = initval;
2267 memcpy(&transport->tcp_timeout, &to,
2268 sizeof(transport->tcp_timeout));
2269 xprt->timeout = &transport->tcp_timeout;
2270 xprt->connect_timeout = connect_timeout;
2271 }
2272 set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2273 spin_unlock(&xprt->transport_lock);
2274}
2275
2276static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2277{
2278 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2279 int ret = -ENOTCONN;
2280
2281 if (!transport->inet) {
2282 struct sock *sk = sock->sk;
2283 unsigned int addr_pref = IPV6_PREFER_SRC_PUBLIC;
2284
2285 /* Avoid temporary address, they are bad for long-lived
2286 * connections such as NFS mounts.
2287 * RFC4941, section 3.6 suggests that:
2288 * Individual applications, which have specific
2289 * knowledge about the normal duration of connections,
2290 * MAY override this as appropriate.
2291 */
2292 kernel_setsockopt(sock, SOL_IPV6, IPV6_ADDR_PREFERENCES,
2293 (char *)&addr_pref, sizeof(addr_pref));
2294
2295 xs_tcp_set_socket_timeouts(xprt, sock);
2296
2297 write_lock_bh(&sk->sk_callback_lock);
2298
2299 xs_save_old_callbacks(transport, sk);
2300
2301 sk->sk_user_data = xprt;
2302 sk->sk_data_ready = xs_data_ready;
2303 sk->sk_state_change = xs_tcp_state_change;
2304 sk->sk_write_space = xs_tcp_write_space;
2305 sock_set_flag(sk, SOCK_FASYNC);
2306 sk->sk_error_report = xs_error_report;
2307
2308 /* socket options */
2309 sock_reset_flag(sk, SOCK_LINGER);
2310 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2311
2312 xprt_clear_connected(xprt);
2313
2314 /* Reset to new socket */
2315 transport->sock = sock;
2316 transport->inet = sk;
2317
2318 write_unlock_bh(&sk->sk_callback_lock);
2319 }
2320
2321 if (!xprt_bound(xprt))
2322 goto out;
2323
2324 xs_set_memalloc(xprt);
2325
2326 xs_stream_start_connect(transport);
2327
2328 /* Tell the socket layer to start connecting... */
2329 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2330 ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2331 switch (ret) {
2332 case 0:
2333 xs_set_srcport(transport, sock);
2334 /* fall through */
2335 case -EINPROGRESS:
2336 /* SYN_SENT! */
2337 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2338 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2339 break;
2340 case -EADDRNOTAVAIL:
2341 /* Source port number is unavailable. Try a new one! */
2342 transport->srcport = 0;
2343 }
2344out:
2345 return ret;
2346}
2347
2348/**
2349 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2350 * @work: queued work item
2351 *
2352 * Invoked by a work queue tasklet.
2353 */
2354static void xs_tcp_setup_socket(struct work_struct *work)
2355{
2356 struct sock_xprt *transport =
2357 container_of(work, struct sock_xprt, connect_worker.work);
2358 struct socket *sock = transport->sock;
2359 struct rpc_xprt *xprt = &transport->xprt;
2360 int status = -EIO;
2361
2362 if (!sock) {
2363 sock = xs_create_sock(xprt, transport,
2364 xs_addr(xprt)->sa_family, SOCK_STREAM,
2365 IPPROTO_TCP, true);
2366 if (IS_ERR(sock)) {
2367 status = PTR_ERR(sock);
2368 goto out;
2369 }
2370 }
2371
2372 dprintk("RPC: worker connecting xprt %p via %s to "
2373 "%s (port %s)\n", xprt,
2374 xprt->address_strings[RPC_DISPLAY_PROTO],
2375 xprt->address_strings[RPC_DISPLAY_ADDR],
2376 xprt->address_strings[RPC_DISPLAY_PORT]);
2377
2378 status = xs_tcp_finish_connecting(xprt, sock);
2379 trace_rpc_socket_connect(xprt, sock, status);
2380 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2381 xprt, -status, xprt_connected(xprt),
2382 sock->sk->sk_state);
2383 switch (status) {
2384 default:
2385 printk("%s: connect returned unhandled error %d\n",
2386 __func__, status);
2387 /* fall through */
2388 case -EADDRNOTAVAIL:
2389 /* We're probably in TIME_WAIT. Get rid of existing socket,
2390 * and retry
2391 */
2392 xs_tcp_force_close(xprt);
2393 break;
2394 case 0:
2395 case -EINPROGRESS:
2396 case -EALREADY:
2397 xprt_unlock_connect(xprt, transport);
2398 return;
2399 case -EINVAL:
2400 /* Happens, for instance, if the user specified a link
2401 * local IPv6 address without a scope-id.
2402 */
2403 case -ECONNREFUSED:
2404 case -ECONNRESET:
2405 case -ENETDOWN:
2406 case -ENETUNREACH:
2407 case -EHOSTUNREACH:
2408 case -EADDRINUSE:
2409 case -ENOBUFS:
2410 /*
2411 * xs_tcp_force_close() wakes tasks with -EIO.
2412 * We need to wake them first to ensure the
2413 * correct error code.
2414 */
2415 xprt_wake_pending_tasks(xprt, status);
2416 xs_tcp_force_close(xprt);
2417 goto out;
2418 }
2419 status = -EAGAIN;
2420out:
2421 xprt_clear_connecting(xprt);
2422 xprt_unlock_connect(xprt, transport);
2423 xprt_wake_pending_tasks(xprt, status);
2424}
2425
2426/**
2427 * xs_connect - connect a socket to a remote endpoint
2428 * @xprt: pointer to transport structure
2429 * @task: address of RPC task that manages state of connect request
2430 *
2431 * TCP: If the remote end dropped the connection, delay reconnecting.
2432 *
2433 * UDP socket connects are synchronous, but we use a work queue anyway
2434 * to guarantee that even unprivileged user processes can set up a
2435 * socket on a privileged port.
2436 *
2437 * If a UDP socket connect fails, the delay behavior here prevents
2438 * retry floods (hard mounts).
2439 */
2440static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2441{
2442 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2443 unsigned long delay = 0;
2444
2445 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2446
2447 if (transport->sock != NULL) {
2448 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2449 "seconds\n",
2450 xprt, xprt->reestablish_timeout / HZ);
2451
2452 /* Start by resetting any existing state */
2453 xs_reset_transport(transport);
2454
2455 delay = xprt_reconnect_delay(xprt);
2456 xprt_reconnect_backoff(xprt, XS_TCP_INIT_REEST_TO);
2457
2458 } else
2459 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2460
2461 queue_delayed_work(xprtiod_workqueue,
2462 &transport->connect_worker,
2463 delay);
2464}
2465
2466static void xs_wake_disconnect(struct sock_xprt *transport)
2467{
2468 if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state))
2469 xs_tcp_force_close(&transport->xprt);
2470}
2471
2472static void xs_wake_write(struct sock_xprt *transport)
2473{
2474 if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state))
2475 xprt_write_space(&transport->xprt);
2476}
2477
2478static void xs_wake_error(struct sock_xprt *transport)
2479{
2480 int sockerr;
2481
2482 if (!test_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2483 return;
2484 mutex_lock(&transport->recv_mutex);
2485 if (transport->sock == NULL)
2486 goto out;
2487 if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2488 goto out;
2489 sockerr = xchg(&transport->xprt_err, 0);
2490 if (sockerr < 0)
2491 xprt_wake_pending_tasks(&transport->xprt, sockerr);
2492out:
2493 mutex_unlock(&transport->recv_mutex);
2494}
2495
2496static void xs_wake_pending(struct sock_xprt *transport)
2497{
2498 if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING, &transport->sock_state))
2499 xprt_wake_pending_tasks(&transport->xprt, -EAGAIN);
2500}
2501
2502static void xs_error_handle(struct work_struct *work)
2503{
2504 struct sock_xprt *transport = container_of(work,
2505 struct sock_xprt, error_worker);
2506
2507 xs_wake_disconnect(transport);
2508 xs_wake_write(transport);
2509 xs_wake_error(transport);
2510 xs_wake_pending(transport);
2511}
2512
2513/**
2514 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2515 * @xprt: rpc_xprt struct containing statistics
2516 * @seq: output file
2517 *
2518 */
2519static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2520{
2521 long idle_time = 0;
2522
2523 if (xprt_connected(xprt))
2524 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2525
2526 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2527 "%llu %llu %lu %llu %llu\n",
2528 xprt->stat.bind_count,
2529 xprt->stat.connect_count,
2530 xprt->stat.connect_time / HZ,
2531 idle_time,
2532 xprt->stat.sends,
2533 xprt->stat.recvs,
2534 xprt->stat.bad_xids,
2535 xprt->stat.req_u,
2536 xprt->stat.bklog_u,
2537 xprt->stat.max_slots,
2538 xprt->stat.sending_u,
2539 xprt->stat.pending_u);
2540}
2541
2542/**
2543 * xs_udp_print_stats - display UDP socket-specifc stats
2544 * @xprt: rpc_xprt struct containing statistics
2545 * @seq: output file
2546 *
2547 */
2548static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2549{
2550 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2551
2552 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2553 "%lu %llu %llu\n",
2554 transport->srcport,
2555 xprt->stat.bind_count,
2556 xprt->stat.sends,
2557 xprt->stat.recvs,
2558 xprt->stat.bad_xids,
2559 xprt->stat.req_u,
2560 xprt->stat.bklog_u,
2561 xprt->stat.max_slots,
2562 xprt->stat.sending_u,
2563 xprt->stat.pending_u);
2564}
2565
2566/**
2567 * xs_tcp_print_stats - display TCP socket-specifc stats
2568 * @xprt: rpc_xprt struct containing statistics
2569 * @seq: output file
2570 *
2571 */
2572static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2573{
2574 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2575 long idle_time = 0;
2576
2577 if (xprt_connected(xprt))
2578 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2579
2580 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2581 "%llu %llu %lu %llu %llu\n",
2582 transport->srcport,
2583 xprt->stat.bind_count,
2584 xprt->stat.connect_count,
2585 xprt->stat.connect_time / HZ,
2586 idle_time,
2587 xprt->stat.sends,
2588 xprt->stat.recvs,
2589 xprt->stat.bad_xids,
2590 xprt->stat.req_u,
2591 xprt->stat.bklog_u,
2592 xprt->stat.max_slots,
2593 xprt->stat.sending_u,
2594 xprt->stat.pending_u);
2595}
2596
2597/*
2598 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2599 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2600 * to use the server side send routines.
2601 */
2602static int bc_malloc(struct rpc_task *task)
2603{
2604 struct rpc_rqst *rqst = task->tk_rqstp;
2605 size_t size = rqst->rq_callsize;
2606 struct page *page;
2607 struct rpc_buffer *buf;
2608
2609 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2610 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2611 size);
2612 return -EINVAL;
2613 }
2614
2615 page = alloc_page(GFP_KERNEL);
2616 if (!page)
2617 return -ENOMEM;
2618
2619 buf = page_address(page);
2620 buf->len = PAGE_SIZE;
2621
2622 rqst->rq_buffer = buf->data;
2623 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2624 return 0;
2625}
2626
2627/*
2628 * Free the space allocated in the bc_alloc routine
2629 */
2630static void bc_free(struct rpc_task *task)
2631{
2632 void *buffer = task->tk_rqstp->rq_buffer;
2633 struct rpc_buffer *buf;
2634
2635 buf = container_of(buffer, struct rpc_buffer, data);
2636 free_page((unsigned long)buf);
2637}
2638
2639/*
2640 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2641 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2642 */
2643static int bc_sendto(struct rpc_rqst *req)
2644{
2645 int len;
2646 struct xdr_buf *xbufp = &req->rq_snd_buf;
2647 struct sock_xprt *transport =
2648 container_of(req->rq_xprt, struct sock_xprt, xprt);
2649 unsigned long headoff;
2650 unsigned long tailoff;
2651 struct page *tailpage;
2652 struct msghdr msg = {
2653 .msg_flags = MSG_MORE
2654 };
2655 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
2656 (u32)xbufp->len);
2657 struct kvec iov = {
2658 .iov_base = &marker,
2659 .iov_len = sizeof(marker),
2660 };
2661
2662 len = kernel_sendmsg(transport->sock, &msg, &iov, 1, iov.iov_len);
2663 if (len != iov.iov_len)
2664 return -EAGAIN;
2665
2666 tailpage = NULL;
2667 if (xbufp->tail[0].iov_len)
2668 tailpage = virt_to_page(xbufp->tail[0].iov_base);
2669 tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2670 headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2671 len = svc_send_common(transport->sock, xbufp,
2672 virt_to_page(xbufp->head[0].iov_base), headoff,
2673 tailpage, tailoff);
2674 if (len != xbufp->len)
2675 return -EAGAIN;
2676 return len;
2677}
2678
2679/*
2680 * The send routine. Borrows from svc_send
2681 */
2682static int bc_send_request(struct rpc_rqst *req)
2683{
2684 struct svc_xprt *xprt;
2685 int len;
2686
2687 dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2688 /*
2689 * Get the server socket associated with this callback xprt
2690 */
2691 xprt = req->rq_xprt->bc_xprt;
2692
2693 /*
2694 * Grab the mutex to serialize data as the connection is shared
2695 * with the fore channel
2696 */
2697 mutex_lock(&xprt->xpt_mutex);
2698 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2699 len = -ENOTCONN;
2700 else
2701 len = bc_sendto(req);
2702 mutex_unlock(&xprt->xpt_mutex);
2703
2704 if (len > 0)
2705 len = 0;
2706
2707 return len;
2708}
2709
2710/*
2711 * The close routine. Since this is client initiated, we do nothing
2712 */
2713
2714static void bc_close(struct rpc_xprt *xprt)
2715{
2716}
2717
2718/*
2719 * The xprt destroy routine. Again, because this connection is client
2720 * initiated, we do nothing
2721 */
2722
2723static void bc_destroy(struct rpc_xprt *xprt)
2724{
2725 dprintk("RPC: bc_destroy xprt %p\n", xprt);
2726
2727 xs_xprt_free(xprt);
2728 module_put(THIS_MODULE);
2729}
2730
2731static const struct rpc_xprt_ops xs_local_ops = {
2732 .reserve_xprt = xprt_reserve_xprt,
2733 .release_xprt = xprt_release_xprt,
2734 .alloc_slot = xprt_alloc_slot,
2735 .free_slot = xprt_free_slot,
2736 .rpcbind = xs_local_rpcbind,
2737 .set_port = xs_local_set_port,
2738 .connect = xs_local_connect,
2739 .buf_alloc = rpc_malloc,
2740 .buf_free = rpc_free,
2741 .prepare_request = xs_stream_prepare_request,
2742 .send_request = xs_local_send_request,
2743 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2744 .close = xs_close,
2745 .destroy = xs_destroy,
2746 .print_stats = xs_local_print_stats,
2747 .enable_swap = xs_enable_swap,
2748 .disable_swap = xs_disable_swap,
2749};
2750
2751static const struct rpc_xprt_ops xs_udp_ops = {
2752 .set_buffer_size = xs_udp_set_buffer_size,
2753 .reserve_xprt = xprt_reserve_xprt_cong,
2754 .release_xprt = xprt_release_xprt_cong,
2755 .alloc_slot = xprt_alloc_slot,
2756 .free_slot = xprt_free_slot,
2757 .rpcbind = rpcb_getport_async,
2758 .set_port = xs_set_port,
2759 .connect = xs_connect,
2760 .buf_alloc = rpc_malloc,
2761 .buf_free = rpc_free,
2762 .send_request = xs_udp_send_request,
2763 .wait_for_reply_request = xprt_wait_for_reply_request_rtt,
2764 .timer = xs_udp_timer,
2765 .release_request = xprt_release_rqst_cong,
2766 .close = xs_close,
2767 .destroy = xs_destroy,
2768 .print_stats = xs_udp_print_stats,
2769 .enable_swap = xs_enable_swap,
2770 .disable_swap = xs_disable_swap,
2771 .inject_disconnect = xs_inject_disconnect,
2772};
2773
2774static const struct rpc_xprt_ops xs_tcp_ops = {
2775 .reserve_xprt = xprt_reserve_xprt,
2776 .release_xprt = xprt_release_xprt,
2777 .alloc_slot = xprt_alloc_slot,
2778 .free_slot = xprt_free_slot,
2779 .rpcbind = rpcb_getport_async,
2780 .set_port = xs_set_port,
2781 .connect = xs_connect,
2782 .buf_alloc = rpc_malloc,
2783 .buf_free = rpc_free,
2784 .prepare_request = xs_stream_prepare_request,
2785 .send_request = xs_tcp_send_request,
2786 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2787 .close = xs_tcp_shutdown,
2788 .destroy = xs_destroy,
2789 .set_connect_timeout = xs_tcp_set_connect_timeout,
2790 .print_stats = xs_tcp_print_stats,
2791 .enable_swap = xs_enable_swap,
2792 .disable_swap = xs_disable_swap,
2793 .inject_disconnect = xs_inject_disconnect,
2794#ifdef CONFIG_SUNRPC_BACKCHANNEL
2795 .bc_setup = xprt_setup_bc,
2796 .bc_maxpayload = xs_tcp_bc_maxpayload,
2797 .bc_num_slots = xprt_bc_max_slots,
2798 .bc_free_rqst = xprt_free_bc_rqst,
2799 .bc_destroy = xprt_destroy_bc,
2800#endif
2801};
2802
2803/*
2804 * The rpc_xprt_ops for the server backchannel
2805 */
2806
2807static const struct rpc_xprt_ops bc_tcp_ops = {
2808 .reserve_xprt = xprt_reserve_xprt,
2809 .release_xprt = xprt_release_xprt,
2810 .alloc_slot = xprt_alloc_slot,
2811 .free_slot = xprt_free_slot,
2812 .buf_alloc = bc_malloc,
2813 .buf_free = bc_free,
2814 .send_request = bc_send_request,
2815 .wait_for_reply_request = xprt_wait_for_reply_request_def,
2816 .close = bc_close,
2817 .destroy = bc_destroy,
2818 .print_stats = xs_tcp_print_stats,
2819 .enable_swap = xs_enable_swap,
2820 .disable_swap = xs_disable_swap,
2821 .inject_disconnect = xs_inject_disconnect,
2822};
2823
2824static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2825{
2826 static const struct sockaddr_in sin = {
2827 .sin_family = AF_INET,
2828 .sin_addr.s_addr = htonl(INADDR_ANY),
2829 };
2830 static const struct sockaddr_in6 sin6 = {
2831 .sin6_family = AF_INET6,
2832 .sin6_addr = IN6ADDR_ANY_INIT,
2833 };
2834
2835 switch (family) {
2836 case AF_LOCAL:
2837 break;
2838 case AF_INET:
2839 memcpy(sap, &sin, sizeof(sin));
2840 break;
2841 case AF_INET6:
2842 memcpy(sap, &sin6, sizeof(sin6));
2843 break;
2844 default:
2845 dprintk("RPC: %s: Bad address family\n", __func__);
2846 return -EAFNOSUPPORT;
2847 }
2848 return 0;
2849}
2850
2851static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2852 unsigned int slot_table_size,
2853 unsigned int max_slot_table_size)
2854{
2855 struct rpc_xprt *xprt;
2856 struct sock_xprt *new;
2857
2858 if (args->addrlen > sizeof(xprt->addr)) {
2859 dprintk("RPC: xs_setup_xprt: address too large\n");
2860 return ERR_PTR(-EBADF);
2861 }
2862
2863 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2864 max_slot_table_size);
2865 if (xprt == NULL) {
2866 dprintk("RPC: xs_setup_xprt: couldn't allocate "
2867 "rpc_xprt\n");
2868 return ERR_PTR(-ENOMEM);
2869 }
2870
2871 new = container_of(xprt, struct sock_xprt, xprt);
2872 mutex_init(&new->recv_mutex);
2873 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2874 xprt->addrlen = args->addrlen;
2875 if (args->srcaddr)
2876 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2877 else {
2878 int err;
2879 err = xs_init_anyaddr(args->dstaddr->sa_family,
2880 (struct sockaddr *)&new->srcaddr);
2881 if (err != 0) {
2882 xprt_free(xprt);
2883 return ERR_PTR(err);
2884 }
2885 }
2886
2887 return xprt;
2888}
2889
2890static const struct rpc_timeout xs_local_default_timeout = {
2891 .to_initval = 10 * HZ,
2892 .to_maxval = 10 * HZ,
2893 .to_retries = 2,
2894};
2895
2896/**
2897 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2898 * @args: rpc transport creation arguments
2899 *
2900 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2901 */
2902static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2903{
2904 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2905 struct sock_xprt *transport;
2906 struct rpc_xprt *xprt;
2907 struct rpc_xprt *ret;
2908
2909 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2910 xprt_max_tcp_slot_table_entries);
2911 if (IS_ERR(xprt))
2912 return xprt;
2913 transport = container_of(xprt, struct sock_xprt, xprt);
2914
2915 xprt->prot = 0;
2916 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2917
2918 xprt->bind_timeout = XS_BIND_TO;
2919 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2920 xprt->idle_timeout = XS_IDLE_DISC_TO;
2921
2922 xprt->ops = &xs_local_ops;
2923 xprt->timeout = &xs_local_default_timeout;
2924
2925 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2926 INIT_WORK(&transport->error_worker, xs_error_handle);
2927 INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket);
2928
2929 switch (sun->sun_family) {
2930 case AF_LOCAL:
2931 if (sun->sun_path[0] != '/') {
2932 dprintk("RPC: bad AF_LOCAL address: %s\n",
2933 sun->sun_path);
2934 ret = ERR_PTR(-EINVAL);
2935 goto out_err;
2936 }
2937 xprt_set_bound(xprt);
2938 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2939 ret = ERR_PTR(xs_local_setup_socket(transport));
2940 if (ret)
2941 goto out_err;
2942 break;
2943 default:
2944 ret = ERR_PTR(-EAFNOSUPPORT);
2945 goto out_err;
2946 }
2947
2948 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
2949 xprt->address_strings[RPC_DISPLAY_ADDR]);
2950
2951 if (try_module_get(THIS_MODULE))
2952 return xprt;
2953 ret = ERR_PTR(-EINVAL);
2954out_err:
2955 xs_xprt_free(xprt);
2956 return ret;
2957}
2958
2959static const struct rpc_timeout xs_udp_default_timeout = {
2960 .to_initval = 5 * HZ,
2961 .to_maxval = 30 * HZ,
2962 .to_increment = 5 * HZ,
2963 .to_retries = 5,
2964};
2965
2966/**
2967 * xs_setup_udp - Set up transport to use a UDP socket
2968 * @args: rpc transport creation arguments
2969 *
2970 */
2971static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2972{
2973 struct sockaddr *addr = args->dstaddr;
2974 struct rpc_xprt *xprt;
2975 struct sock_xprt *transport;
2976 struct rpc_xprt *ret;
2977
2978 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2979 xprt_udp_slot_table_entries);
2980 if (IS_ERR(xprt))
2981 return xprt;
2982 transport = container_of(xprt, struct sock_xprt, xprt);
2983
2984 xprt->prot = IPPROTO_UDP;
2985 /* XXX: header size can vary due to auth type, IPv6, etc. */
2986 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2987
2988 xprt->bind_timeout = XS_BIND_TO;
2989 xprt->reestablish_timeout = XS_UDP_REEST_TO;
2990 xprt->idle_timeout = XS_IDLE_DISC_TO;
2991
2992 xprt->ops = &xs_udp_ops;
2993
2994 xprt->timeout = &xs_udp_default_timeout;
2995
2996 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
2997 INIT_WORK(&transport->error_worker, xs_error_handle);
2998 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2999
3000 switch (addr->sa_family) {
3001 case AF_INET:
3002 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3003 xprt_set_bound(xprt);
3004
3005 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
3006 break;
3007 case AF_INET6:
3008 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3009 xprt_set_bound(xprt);
3010
3011 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
3012 break;
3013 default:
3014 ret = ERR_PTR(-EAFNOSUPPORT);
3015 goto out_err;
3016 }
3017
3018 if (xprt_bound(xprt))
3019 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3020 xprt->address_strings[RPC_DISPLAY_ADDR],
3021 xprt->address_strings[RPC_DISPLAY_PORT],
3022 xprt->address_strings[RPC_DISPLAY_PROTO]);
3023 else
3024 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
3025 xprt->address_strings[RPC_DISPLAY_ADDR],
3026 xprt->address_strings[RPC_DISPLAY_PROTO]);
3027
3028 if (try_module_get(THIS_MODULE))
3029 return xprt;
3030 ret = ERR_PTR(-EINVAL);
3031out_err:
3032 xs_xprt_free(xprt);
3033 return ret;
3034}
3035
3036static const struct rpc_timeout xs_tcp_default_timeout = {
3037 .to_initval = 60 * HZ,
3038 .to_maxval = 60 * HZ,
3039 .to_retries = 2,
3040};
3041
3042/**
3043 * xs_setup_tcp - Set up transport to use a TCP socket
3044 * @args: rpc transport creation arguments
3045 *
3046 */
3047static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
3048{
3049 struct sockaddr *addr = args->dstaddr;
3050 struct rpc_xprt *xprt;
3051 struct sock_xprt *transport;
3052 struct rpc_xprt *ret;
3053 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
3054
3055 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
3056 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
3057
3058 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3059 max_slot_table_size);
3060 if (IS_ERR(xprt))
3061 return xprt;
3062 transport = container_of(xprt, struct sock_xprt, xprt);
3063
3064 xprt->prot = IPPROTO_TCP;
3065 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3066
3067 xprt->bind_timeout = XS_BIND_TO;
3068 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3069 xprt->idle_timeout = XS_IDLE_DISC_TO;
3070
3071 xprt->ops = &xs_tcp_ops;
3072 xprt->timeout = &xs_tcp_default_timeout;
3073
3074 xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3075 xprt->connect_timeout = xprt->timeout->to_initval *
3076 (xprt->timeout->to_retries + 1);
3077
3078 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3079 INIT_WORK(&transport->error_worker, xs_error_handle);
3080 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
3081
3082 switch (addr->sa_family) {
3083 case AF_INET:
3084 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3085 xprt_set_bound(xprt);
3086
3087 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3088 break;
3089 case AF_INET6:
3090 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3091 xprt_set_bound(xprt);
3092
3093 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3094 break;
3095 default:
3096 ret = ERR_PTR(-EAFNOSUPPORT);
3097 goto out_err;
3098 }
3099
3100 if (xprt_bound(xprt))
3101 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3102 xprt->address_strings[RPC_DISPLAY_ADDR],
3103 xprt->address_strings[RPC_DISPLAY_PORT],
3104 xprt->address_strings[RPC_DISPLAY_PROTO]);
3105 else
3106 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
3107 xprt->address_strings[RPC_DISPLAY_ADDR],
3108 xprt->address_strings[RPC_DISPLAY_PROTO]);
3109
3110 if (try_module_get(THIS_MODULE))
3111 return xprt;
3112 ret = ERR_PTR(-EINVAL);
3113out_err:
3114 xs_xprt_free(xprt);
3115 return ret;
3116}
3117
3118/**
3119 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3120 * @args: rpc transport creation arguments
3121 *
3122 */
3123static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3124{
3125 struct sockaddr *addr = args->dstaddr;
3126 struct rpc_xprt *xprt;
3127 struct sock_xprt *transport;
3128 struct svc_sock *bc_sock;
3129 struct rpc_xprt *ret;
3130
3131 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3132 xprt_tcp_slot_table_entries);
3133 if (IS_ERR(xprt))
3134 return xprt;
3135 transport = container_of(xprt, struct sock_xprt, xprt);
3136
3137 xprt->prot = IPPROTO_TCP;
3138 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3139 xprt->timeout = &xs_tcp_default_timeout;
3140
3141 /* backchannel */
3142 xprt_set_bound(xprt);
3143 xprt->bind_timeout = 0;
3144 xprt->reestablish_timeout = 0;
3145 xprt->idle_timeout = 0;
3146
3147 xprt->ops = &bc_tcp_ops;
3148
3149 switch (addr->sa_family) {
3150 case AF_INET:
3151 xs_format_peer_addresses(xprt, "tcp",
3152 RPCBIND_NETID_TCP);
3153 break;
3154 case AF_INET6:
3155 xs_format_peer_addresses(xprt, "tcp",
3156 RPCBIND_NETID_TCP6);
3157 break;
3158 default:
3159 ret = ERR_PTR(-EAFNOSUPPORT);
3160 goto out_err;
3161 }
3162
3163 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3164 xprt->address_strings[RPC_DISPLAY_ADDR],
3165 xprt->address_strings[RPC_DISPLAY_PORT],
3166 xprt->address_strings[RPC_DISPLAY_PROTO]);
3167
3168 /*
3169 * Once we've associated a backchannel xprt with a connection,
3170 * we want to keep it around as long as the connection lasts,
3171 * in case we need to start using it for a backchannel again;
3172 * this reference won't be dropped until bc_xprt is destroyed.
3173 */
3174 xprt_get(xprt);
3175 args->bc_xprt->xpt_bc_xprt = xprt;
3176 xprt->bc_xprt = args->bc_xprt;
3177 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3178 transport->sock = bc_sock->sk_sock;
3179 transport->inet = bc_sock->sk_sk;
3180
3181 /*
3182 * Since we don't want connections for the backchannel, we set
3183 * the xprt status to connected
3184 */
3185 xprt_set_connected(xprt);
3186
3187 if (try_module_get(THIS_MODULE))
3188 return xprt;
3189
3190 args->bc_xprt->xpt_bc_xprt = NULL;
3191 args->bc_xprt->xpt_bc_xps = NULL;
3192 xprt_put(xprt);
3193 ret = ERR_PTR(-EINVAL);
3194out_err:
3195 xs_xprt_free(xprt);
3196 return ret;
3197}
3198
3199static struct xprt_class xs_local_transport = {
3200 .list = LIST_HEAD_INIT(xs_local_transport.list),
3201 .name = "named UNIX socket",
3202 .owner = THIS_MODULE,
3203 .ident = XPRT_TRANSPORT_LOCAL,
3204 .setup = xs_setup_local,
3205};
3206
3207static struct xprt_class xs_udp_transport = {
3208 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3209 .name = "udp",
3210 .owner = THIS_MODULE,
3211 .ident = XPRT_TRANSPORT_UDP,
3212 .setup = xs_setup_udp,
3213};
3214
3215static struct xprt_class xs_tcp_transport = {
3216 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3217 .name = "tcp",
3218 .owner = THIS_MODULE,
3219 .ident = XPRT_TRANSPORT_TCP,
3220 .setup = xs_setup_tcp,
3221};
3222
3223static struct xprt_class xs_bc_tcp_transport = {
3224 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3225 .name = "tcp NFSv4.1 backchannel",
3226 .owner = THIS_MODULE,
3227 .ident = XPRT_TRANSPORT_BC_TCP,
3228 .setup = xs_setup_bc_tcp,
3229};
3230
3231/**
3232 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3233 *
3234 */
3235int init_socket_xprt(void)
3236{
3237 if (!sunrpc_table_header)
3238 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3239
3240 xprt_register_transport(&xs_local_transport);
3241 xprt_register_transport(&xs_udp_transport);
3242 xprt_register_transport(&xs_tcp_transport);
3243 xprt_register_transport(&xs_bc_tcp_transport);
3244
3245 return 0;
3246}
3247
3248/**
3249 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3250 *
3251 */
3252void cleanup_socket_xprt(void)
3253{
3254 if (sunrpc_table_header) {
3255 unregister_sysctl_table(sunrpc_table_header);
3256 sunrpc_table_header = NULL;
3257 }
3258
3259 xprt_unregister_transport(&xs_local_transport);
3260 xprt_unregister_transport(&xs_udp_transport);
3261 xprt_unregister_transport(&xs_tcp_transport);
3262 xprt_unregister_transport(&xs_bc_tcp_transport);
3263}
3264
3265static int param_set_uint_minmax(const char *val,
3266 const struct kernel_param *kp,
3267 unsigned int min, unsigned int max)
3268{
3269 unsigned int num;
3270 int ret;
3271
3272 if (!val)
3273 return -EINVAL;
3274 ret = kstrtouint(val, 0, &num);
3275 if (ret)
3276 return ret;
3277 if (num < min || num > max)
3278 return -EINVAL;
3279 *((unsigned int *)kp->arg) = num;
3280 return 0;
3281}
3282
3283static int param_set_portnr(const char *val, const struct kernel_param *kp)
3284{
3285 return param_set_uint_minmax(val, kp,
3286 RPC_MIN_RESVPORT,
3287 RPC_MAX_RESVPORT);
3288}
3289
3290static const struct kernel_param_ops param_ops_portnr = {
3291 .set = param_set_portnr,
3292 .get = param_get_uint,
3293};
3294
3295#define param_check_portnr(name, p) \
3296 __param_check(name, p, unsigned int);
3297
3298module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3299module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3300
3301static int param_set_slot_table_size(const char *val,
3302 const struct kernel_param *kp)
3303{
3304 return param_set_uint_minmax(val, kp,
3305 RPC_MIN_SLOT_TABLE,
3306 RPC_MAX_SLOT_TABLE);
3307}
3308
3309static const struct kernel_param_ops param_ops_slot_table_size = {
3310 .set = param_set_slot_table_size,
3311 .get = param_get_uint,
3312};
3313
3314#define param_check_slot_table_size(name, p) \
3315 __param_check(name, p, unsigned int);
3316
3317static int param_set_max_slot_table_size(const char *val,
3318 const struct kernel_param *kp)
3319{
3320 return param_set_uint_minmax(val, kp,
3321 RPC_MIN_SLOT_TABLE,
3322 RPC_MAX_SLOT_TABLE_LIMIT);
3323}
3324
3325static const struct kernel_param_ops param_ops_max_slot_table_size = {
3326 .set = param_set_max_slot_table_size,
3327 .get = param_get_uint,
3328};
3329
3330#define param_check_max_slot_table_size(name, p) \
3331 __param_check(name, p, unsigned int);
3332
3333module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3334 slot_table_size, 0644);
3335module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3336 max_slot_table_size, 0644);
3337module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3338 slot_table_size, 0644);
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/net/sunrpc/xprtsock.c
4 *
5 * Client-side transport implementation for sockets.
6 *
7 * TCP callback races fixes (C) 1998 Red Hat
8 * TCP send fixes (C) 1998 Red Hat
9 * TCP NFS related read + write fixes
10 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
11 *
12 * Rewrite of larges part of the code in order to stabilize TCP stuff.
13 * Fix behaviour when socket buffer is full.
14 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
15 *
16 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
17 *
18 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
19 * <gilles.quillard@bull.net>
20 */
21
22#include <linux/types.h>
23#include <linux/string.h>
24#include <linux/slab.h>
25#include <linux/module.h>
26#include <linux/capability.h>
27#include <linux/pagemap.h>
28#include <linux/errno.h>
29#include <linux/socket.h>
30#include <linux/in.h>
31#include <linux/net.h>
32#include <linux/mm.h>
33#include <linux/un.h>
34#include <linux/udp.h>
35#include <linux/tcp.h>
36#include <linux/sunrpc/clnt.h>
37#include <linux/sunrpc/addr.h>
38#include <linux/sunrpc/sched.h>
39#include <linux/sunrpc/svcsock.h>
40#include <linux/sunrpc/xprtsock.h>
41#include <linux/file.h>
42#ifdef CONFIG_SUNRPC_BACKCHANNEL
43#include <linux/sunrpc/bc_xprt.h>
44#endif
45
46#include <net/sock.h>
47#include <net/checksum.h>
48#include <net/udp.h>
49#include <net/tcp.h>
50#include <net/tls_prot.h>
51#include <net/handshake.h>
52
53#include <linux/bvec.h>
54#include <linux/highmem.h>
55#include <linux/uio.h>
56#include <linux/sched/mm.h>
57
58#include <trace/events/sock.h>
59#include <trace/events/sunrpc.h>
60
61#include "socklib.h"
62#include "sunrpc.h"
63
64static void xs_close(struct rpc_xprt *xprt);
65static void xs_reset_srcport(struct sock_xprt *transport);
66static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock);
67static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
68 struct socket *sock);
69
70/*
71 * xprtsock tunables
72 */
73static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
74static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
75static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
76
77static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
78static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
79
80#define XS_TCP_LINGER_TO (15U * HZ)
81static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
82
83/*
84 * We can register our own files under /proc/sys/sunrpc by
85 * calling register_sysctl() again. The files in that
86 * directory become the union of all files registered there.
87 *
88 * We simply need to make sure that we don't collide with
89 * someone else's file names!
90 */
91
92static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
93static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
94static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
95static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
96static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
97
98static struct ctl_table_header *sunrpc_table_header;
99
100static struct xprt_class xs_local_transport;
101static struct xprt_class xs_udp_transport;
102static struct xprt_class xs_tcp_transport;
103static struct xprt_class xs_tcp_tls_transport;
104static struct xprt_class xs_bc_tcp_transport;
105
106/*
107 * FIXME: changing the UDP slot table size should also resize the UDP
108 * socket buffers for existing UDP transports
109 */
110static struct ctl_table xs_tunables_table[] = {
111 {
112 .procname = "udp_slot_table_entries",
113 .data = &xprt_udp_slot_table_entries,
114 .maxlen = sizeof(unsigned int),
115 .mode = 0644,
116 .proc_handler = proc_dointvec_minmax,
117 .extra1 = &min_slot_table_size,
118 .extra2 = &max_slot_table_size
119 },
120 {
121 .procname = "tcp_slot_table_entries",
122 .data = &xprt_tcp_slot_table_entries,
123 .maxlen = sizeof(unsigned int),
124 .mode = 0644,
125 .proc_handler = proc_dointvec_minmax,
126 .extra1 = &min_slot_table_size,
127 .extra2 = &max_slot_table_size
128 },
129 {
130 .procname = "tcp_max_slot_table_entries",
131 .data = &xprt_max_tcp_slot_table_entries,
132 .maxlen = sizeof(unsigned int),
133 .mode = 0644,
134 .proc_handler = proc_dointvec_minmax,
135 .extra1 = &min_slot_table_size,
136 .extra2 = &max_tcp_slot_table_limit
137 },
138 {
139 .procname = "min_resvport",
140 .data = &xprt_min_resvport,
141 .maxlen = sizeof(unsigned int),
142 .mode = 0644,
143 .proc_handler = proc_dointvec_minmax,
144 .extra1 = &xprt_min_resvport_limit,
145 .extra2 = &xprt_max_resvport_limit
146 },
147 {
148 .procname = "max_resvport",
149 .data = &xprt_max_resvport,
150 .maxlen = sizeof(unsigned int),
151 .mode = 0644,
152 .proc_handler = proc_dointvec_minmax,
153 .extra1 = &xprt_min_resvport_limit,
154 .extra2 = &xprt_max_resvport_limit
155 },
156 {
157 .procname = "tcp_fin_timeout",
158 .data = &xs_tcp_fin_timeout,
159 .maxlen = sizeof(xs_tcp_fin_timeout),
160 .mode = 0644,
161 .proc_handler = proc_dointvec_jiffies,
162 },
163};
164
165/*
166 * Wait duration for a reply from the RPC portmapper.
167 */
168#define XS_BIND_TO (60U * HZ)
169
170/*
171 * Delay if a UDP socket connect error occurs. This is most likely some
172 * kind of resource problem on the local host.
173 */
174#define XS_UDP_REEST_TO (2U * HZ)
175
176/*
177 * The reestablish timeout allows clients to delay for a bit before attempting
178 * to reconnect to a server that just dropped our connection.
179 *
180 * We implement an exponential backoff when trying to reestablish a TCP
181 * transport connection with the server. Some servers like to drop a TCP
182 * connection when they are overworked, so we start with a short timeout and
183 * increase over time if the server is down or not responding.
184 */
185#define XS_TCP_INIT_REEST_TO (3U * HZ)
186
187/*
188 * TCP idle timeout; client drops the transport socket if it is idle
189 * for this long. Note that we also timeout UDP sockets to prevent
190 * holding port numbers when there is no RPC traffic.
191 */
192#define XS_IDLE_DISC_TO (5U * 60 * HZ)
193
194/*
195 * TLS handshake timeout.
196 */
197#define XS_TLS_HANDSHAKE_TO (10U * HZ)
198
199#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
200# undef RPC_DEBUG_DATA
201# define RPCDBG_FACILITY RPCDBG_TRANS
202#endif
203
204#ifdef RPC_DEBUG_DATA
205static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
206{
207 u8 *buf = (u8 *) packet;
208 int j;
209
210 dprintk("RPC: %s\n", msg);
211 for (j = 0; j < count && j < 128; j += 4) {
212 if (!(j & 31)) {
213 if (j)
214 dprintk("\n");
215 dprintk("0x%04x ", j);
216 }
217 dprintk("%02x%02x%02x%02x ",
218 buf[j], buf[j+1], buf[j+2], buf[j+3]);
219 }
220 dprintk("\n");
221}
222#else
223static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
224{
225 /* NOP */
226}
227#endif
228
229static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
230{
231 return (struct rpc_xprt *) sk->sk_user_data;
232}
233
234static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
235{
236 return (struct sockaddr *) &xprt->addr;
237}
238
239static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
240{
241 return (struct sockaddr_un *) &xprt->addr;
242}
243
244static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
245{
246 return (struct sockaddr_in *) &xprt->addr;
247}
248
249static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
250{
251 return (struct sockaddr_in6 *) &xprt->addr;
252}
253
254static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
255{
256 struct sockaddr *sap = xs_addr(xprt);
257 struct sockaddr_in6 *sin6;
258 struct sockaddr_in *sin;
259 struct sockaddr_un *sun;
260 char buf[128];
261
262 switch (sap->sa_family) {
263 case AF_LOCAL:
264 sun = xs_addr_un(xprt);
265 if (sun->sun_path[0]) {
266 strscpy(buf, sun->sun_path, sizeof(buf));
267 } else {
268 buf[0] = '@';
269 strscpy(buf+1, sun->sun_path+1, sizeof(buf)-1);
270 }
271 xprt->address_strings[RPC_DISPLAY_ADDR] =
272 kstrdup(buf, GFP_KERNEL);
273 break;
274 case AF_INET:
275 (void)rpc_ntop(sap, buf, sizeof(buf));
276 xprt->address_strings[RPC_DISPLAY_ADDR] =
277 kstrdup(buf, GFP_KERNEL);
278 sin = xs_addr_in(xprt);
279 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
280 break;
281 case AF_INET6:
282 (void)rpc_ntop(sap, buf, sizeof(buf));
283 xprt->address_strings[RPC_DISPLAY_ADDR] =
284 kstrdup(buf, GFP_KERNEL);
285 sin6 = xs_addr_in6(xprt);
286 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
287 break;
288 default:
289 BUG();
290 }
291
292 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
293}
294
295static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
296{
297 struct sockaddr *sap = xs_addr(xprt);
298 char buf[128];
299
300 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
301 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
302
303 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
304 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
305}
306
307static void xs_format_peer_addresses(struct rpc_xprt *xprt,
308 const char *protocol,
309 const char *netid)
310{
311 xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
312 xprt->address_strings[RPC_DISPLAY_NETID] = netid;
313 xs_format_common_peer_addresses(xprt);
314 xs_format_common_peer_ports(xprt);
315}
316
317static void xs_update_peer_port(struct rpc_xprt *xprt)
318{
319 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
320 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
321
322 xs_format_common_peer_ports(xprt);
323}
324
325static void xs_free_peer_addresses(struct rpc_xprt *xprt)
326{
327 unsigned int i;
328
329 for (i = 0; i < RPC_DISPLAY_MAX; i++)
330 switch (i) {
331 case RPC_DISPLAY_PROTO:
332 case RPC_DISPLAY_NETID:
333 continue;
334 default:
335 kfree(xprt->address_strings[i]);
336 }
337}
338
339static size_t
340xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp)
341{
342 size_t i,n;
343
344 if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES))
345 return want;
346 n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT;
347 for (i = 0; i < n; i++) {
348 if (buf->pages[i])
349 continue;
350 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp);
351 if (!buf->pages[i]) {
352 i *= PAGE_SIZE;
353 return i > buf->page_base ? i - buf->page_base : 0;
354 }
355 }
356 return want;
357}
358
359static int
360xs_sock_process_cmsg(struct socket *sock, struct msghdr *msg,
361 struct cmsghdr *cmsg, int ret)
362{
363 u8 content_type = tls_get_record_type(sock->sk, cmsg);
364 u8 level, description;
365
366 switch (content_type) {
367 case 0:
368 break;
369 case TLS_RECORD_TYPE_DATA:
370 /* TLS sets EOR at the end of each application data
371 * record, even though there might be more frames
372 * waiting to be decrypted.
373 */
374 msg->msg_flags &= ~MSG_EOR;
375 break;
376 case TLS_RECORD_TYPE_ALERT:
377 tls_alert_recv(sock->sk, msg, &level, &description);
378 ret = (level == TLS_ALERT_LEVEL_FATAL) ?
379 -EACCES : -EAGAIN;
380 break;
381 default:
382 /* discard this record type */
383 ret = -EAGAIN;
384 }
385 return ret;
386}
387
388static int
389xs_sock_recv_cmsg(struct socket *sock, struct msghdr *msg, int flags)
390{
391 union {
392 struct cmsghdr cmsg;
393 u8 buf[CMSG_SPACE(sizeof(u8))];
394 } u;
395 int ret;
396
397 msg->msg_control = &u;
398 msg->msg_controllen = sizeof(u);
399 ret = sock_recvmsg(sock, msg, flags);
400 if (msg->msg_controllen != sizeof(u))
401 ret = xs_sock_process_cmsg(sock, msg, &u.cmsg, ret);
402 return ret;
403}
404
405static ssize_t
406xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek)
407{
408 ssize_t ret;
409 if (seek != 0)
410 iov_iter_advance(&msg->msg_iter, seek);
411 ret = xs_sock_recv_cmsg(sock, msg, flags);
412 return ret > 0 ? ret + seek : ret;
413}
414
415static ssize_t
416xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags,
417 struct kvec *kvec, size_t count, size_t seek)
418{
419 iov_iter_kvec(&msg->msg_iter, ITER_DEST, kvec, 1, count);
420 return xs_sock_recvmsg(sock, msg, flags, seek);
421}
422
423static ssize_t
424xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags,
425 struct bio_vec *bvec, unsigned long nr, size_t count,
426 size_t seek)
427{
428 iov_iter_bvec(&msg->msg_iter, ITER_DEST, bvec, nr, count);
429 return xs_sock_recvmsg(sock, msg, flags, seek);
430}
431
432static ssize_t
433xs_read_discard(struct socket *sock, struct msghdr *msg, int flags,
434 size_t count)
435{
436 iov_iter_discard(&msg->msg_iter, ITER_DEST, count);
437 return xs_sock_recv_cmsg(sock, msg, flags);
438}
439
440#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
441static void
442xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
443{
444 struct bvec_iter bi = {
445 .bi_size = count,
446 };
447 struct bio_vec bv;
448
449 bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
450 for_each_bvec(bv, bvec, bi, bi)
451 flush_dcache_page(bv.bv_page);
452}
453#else
454static inline void
455xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
456{
457}
458#endif
459
460static ssize_t
461xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
462 struct xdr_buf *buf, size_t count, size_t seek, size_t *read)
463{
464 size_t want, seek_init = seek, offset = 0;
465 ssize_t ret;
466
467 want = min_t(size_t, count, buf->head[0].iov_len);
468 if (seek < want) {
469 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek);
470 if (ret <= 0)
471 goto sock_err;
472 offset += ret;
473 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
474 goto out;
475 if (ret != want)
476 goto out;
477 seek = 0;
478 } else {
479 seek -= want;
480 offset += want;
481 }
482
483 want = xs_alloc_sparse_pages(
484 buf, min_t(size_t, count - offset, buf->page_len),
485 GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN);
486 if (seek < want) {
487 ret = xs_read_bvec(sock, msg, flags, buf->bvec,
488 xdr_buf_pagecount(buf),
489 want + buf->page_base,
490 seek + buf->page_base);
491 if (ret <= 0)
492 goto sock_err;
493 xs_flush_bvec(buf->bvec, ret, seek + buf->page_base);
494 ret -= buf->page_base;
495 offset += ret;
496 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
497 goto out;
498 if (ret != want)
499 goto out;
500 seek = 0;
501 } else {
502 seek -= want;
503 offset += want;
504 }
505
506 want = min_t(size_t, count - offset, buf->tail[0].iov_len);
507 if (seek < want) {
508 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek);
509 if (ret <= 0)
510 goto sock_err;
511 offset += ret;
512 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
513 goto out;
514 if (ret != want)
515 goto out;
516 } else if (offset < seek_init)
517 offset = seek_init;
518 ret = -EMSGSIZE;
519out:
520 *read = offset - seek_init;
521 return ret;
522sock_err:
523 offset += seek;
524 goto out;
525}
526
527static void
528xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf)
529{
530 if (!transport->recv.copied) {
531 if (buf->head[0].iov_len >= transport->recv.offset)
532 memcpy(buf->head[0].iov_base,
533 &transport->recv.xid,
534 transport->recv.offset);
535 transport->recv.copied = transport->recv.offset;
536 }
537}
538
539static bool
540xs_read_stream_request_done(struct sock_xprt *transport)
541{
542 return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT);
543}
544
545static void
546xs_read_stream_check_eor(struct sock_xprt *transport,
547 struct msghdr *msg)
548{
549 if (xs_read_stream_request_done(transport))
550 msg->msg_flags |= MSG_EOR;
551}
552
553static ssize_t
554xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg,
555 int flags, struct rpc_rqst *req)
556{
557 struct xdr_buf *buf = &req->rq_private_buf;
558 size_t want, read;
559 ssize_t ret;
560
561 xs_read_header(transport, buf);
562
563 want = transport->recv.len - transport->recv.offset;
564 if (want != 0) {
565 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf,
566 transport->recv.copied + want,
567 transport->recv.copied,
568 &read);
569 transport->recv.offset += read;
570 transport->recv.copied += read;
571 }
572
573 if (transport->recv.offset == transport->recv.len)
574 xs_read_stream_check_eor(transport, msg);
575
576 if (want == 0)
577 return 0;
578
579 switch (ret) {
580 default:
581 break;
582 case -EFAULT:
583 case -EMSGSIZE:
584 msg->msg_flags |= MSG_TRUNC;
585 return read;
586 case 0:
587 return -ESHUTDOWN;
588 }
589 return ret < 0 ? ret : read;
590}
591
592static size_t
593xs_read_stream_headersize(bool isfrag)
594{
595 if (isfrag)
596 return sizeof(__be32);
597 return 3 * sizeof(__be32);
598}
599
600static ssize_t
601xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg,
602 int flags, size_t want, size_t seek)
603{
604 struct kvec kvec = {
605 .iov_base = &transport->recv.fraghdr,
606 .iov_len = want,
607 };
608 return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek);
609}
610
611#if defined(CONFIG_SUNRPC_BACKCHANNEL)
612static ssize_t
613xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
614{
615 struct rpc_xprt *xprt = &transport->xprt;
616 struct rpc_rqst *req;
617 ssize_t ret;
618
619 /* Is this transport associated with the backchannel? */
620 if (!xprt->bc_serv)
621 return -ESHUTDOWN;
622
623 /* Look up and lock the request corresponding to the given XID */
624 req = xprt_lookup_bc_request(xprt, transport->recv.xid);
625 if (!req) {
626 printk(KERN_WARNING "Callback slot table overflowed\n");
627 return -ESHUTDOWN;
628 }
629 if (transport->recv.copied && !req->rq_private_buf.len)
630 return -ESHUTDOWN;
631
632 ret = xs_read_stream_request(transport, msg, flags, req);
633 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
634 xprt_complete_bc_request(req, transport->recv.copied);
635 else
636 req->rq_private_buf.len = transport->recv.copied;
637
638 return ret;
639}
640#else /* CONFIG_SUNRPC_BACKCHANNEL */
641static ssize_t
642xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
643{
644 return -ESHUTDOWN;
645}
646#endif /* CONFIG_SUNRPC_BACKCHANNEL */
647
648static ssize_t
649xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags)
650{
651 struct rpc_xprt *xprt = &transport->xprt;
652 struct rpc_rqst *req;
653 ssize_t ret = 0;
654
655 /* Look up and lock the request corresponding to the given XID */
656 spin_lock(&xprt->queue_lock);
657 req = xprt_lookup_rqst(xprt, transport->recv.xid);
658 if (!req || (transport->recv.copied && !req->rq_private_buf.len)) {
659 msg->msg_flags |= MSG_TRUNC;
660 goto out;
661 }
662 xprt_pin_rqst(req);
663 spin_unlock(&xprt->queue_lock);
664
665 ret = xs_read_stream_request(transport, msg, flags, req);
666
667 spin_lock(&xprt->queue_lock);
668 if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
669 xprt_complete_rqst(req->rq_task, transport->recv.copied);
670 else
671 req->rq_private_buf.len = transport->recv.copied;
672 xprt_unpin_rqst(req);
673out:
674 spin_unlock(&xprt->queue_lock);
675 return ret;
676}
677
678static ssize_t
679xs_read_stream(struct sock_xprt *transport, int flags)
680{
681 struct msghdr msg = { 0 };
682 size_t want, read = 0;
683 ssize_t ret = 0;
684
685 if (transport->recv.len == 0) {
686 want = xs_read_stream_headersize(transport->recv.copied != 0);
687 ret = xs_read_stream_header(transport, &msg, flags, want,
688 transport->recv.offset);
689 if (ret <= 0)
690 goto out_err;
691 transport->recv.offset = ret;
692 if (transport->recv.offset != want)
693 return transport->recv.offset;
694 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) &
695 RPC_FRAGMENT_SIZE_MASK;
696 transport->recv.offset -= sizeof(transport->recv.fraghdr);
697 read = ret;
698 }
699
700 switch (be32_to_cpu(transport->recv.calldir)) {
701 default:
702 msg.msg_flags |= MSG_TRUNC;
703 break;
704 case RPC_CALL:
705 ret = xs_read_stream_call(transport, &msg, flags);
706 break;
707 case RPC_REPLY:
708 ret = xs_read_stream_reply(transport, &msg, flags);
709 }
710 if (msg.msg_flags & MSG_TRUNC) {
711 transport->recv.calldir = cpu_to_be32(-1);
712 transport->recv.copied = -1;
713 }
714 if (ret < 0)
715 goto out_err;
716 read += ret;
717 if (transport->recv.offset < transport->recv.len) {
718 if (!(msg.msg_flags & MSG_TRUNC))
719 return read;
720 msg.msg_flags = 0;
721 ret = xs_read_discard(transport->sock, &msg, flags,
722 transport->recv.len - transport->recv.offset);
723 if (ret <= 0)
724 goto out_err;
725 transport->recv.offset += ret;
726 read += ret;
727 if (transport->recv.offset != transport->recv.len)
728 return read;
729 }
730 if (xs_read_stream_request_done(transport)) {
731 trace_xs_stream_read_request(transport);
732 transport->recv.copied = 0;
733 }
734 transport->recv.offset = 0;
735 transport->recv.len = 0;
736 return read;
737out_err:
738 return ret != 0 ? ret : -ESHUTDOWN;
739}
740
741static __poll_t xs_poll_socket(struct sock_xprt *transport)
742{
743 return transport->sock->ops->poll(transport->file, transport->sock,
744 NULL);
745}
746
747static bool xs_poll_socket_readable(struct sock_xprt *transport)
748{
749 __poll_t events = xs_poll_socket(transport);
750
751 return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP);
752}
753
754static void xs_poll_check_readable(struct sock_xprt *transport)
755{
756
757 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
758 if (test_bit(XPRT_SOCK_IGNORE_RECV, &transport->sock_state))
759 return;
760 if (!xs_poll_socket_readable(transport))
761 return;
762 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
763 queue_work(xprtiod_workqueue, &transport->recv_worker);
764}
765
766static void xs_stream_data_receive(struct sock_xprt *transport)
767{
768 size_t read = 0;
769 ssize_t ret = 0;
770
771 mutex_lock(&transport->recv_mutex);
772 if (transport->sock == NULL)
773 goto out;
774 for (;;) {
775 ret = xs_read_stream(transport, MSG_DONTWAIT);
776 if (ret < 0)
777 break;
778 read += ret;
779 cond_resched();
780 }
781 if (ret == -ESHUTDOWN)
782 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
783 else if (ret == -EACCES)
784 xprt_wake_pending_tasks(&transport->xprt, -EACCES);
785 else
786 xs_poll_check_readable(transport);
787out:
788 mutex_unlock(&transport->recv_mutex);
789 trace_xs_stream_read_data(&transport->xprt, ret, read);
790}
791
792static void xs_stream_data_receive_workfn(struct work_struct *work)
793{
794 struct sock_xprt *transport =
795 container_of(work, struct sock_xprt, recv_worker);
796 unsigned int pflags = memalloc_nofs_save();
797
798 xs_stream_data_receive(transport);
799 memalloc_nofs_restore(pflags);
800}
801
802static void
803xs_stream_reset_connect(struct sock_xprt *transport)
804{
805 transport->recv.offset = 0;
806 transport->recv.len = 0;
807 transport->recv.copied = 0;
808 transport->xmit.offset = 0;
809}
810
811static void
812xs_stream_start_connect(struct sock_xprt *transport)
813{
814 transport->xprt.stat.connect_count++;
815 transport->xprt.stat.connect_start = jiffies;
816}
817
818#define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
819
820/**
821 * xs_nospace - handle transmit was incomplete
822 * @req: pointer to RPC request
823 * @transport: pointer to struct sock_xprt
824 *
825 */
826static int xs_nospace(struct rpc_rqst *req, struct sock_xprt *transport)
827{
828 struct rpc_xprt *xprt = &transport->xprt;
829 struct sock *sk = transport->inet;
830 int ret = -EAGAIN;
831
832 trace_rpc_socket_nospace(req, transport);
833
834 /* Protect against races with write_space */
835 spin_lock(&xprt->transport_lock);
836
837 /* Don't race with disconnect */
838 if (xprt_connected(xprt)) {
839 /* wait for more buffer space */
840 set_bit(XPRT_SOCK_NOSPACE, &transport->sock_state);
841 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
842 sk->sk_write_pending++;
843 xprt_wait_for_buffer_space(xprt);
844 } else
845 ret = -ENOTCONN;
846
847 spin_unlock(&xprt->transport_lock);
848 return ret;
849}
850
851static int xs_sock_nospace(struct rpc_rqst *req)
852{
853 struct sock_xprt *transport =
854 container_of(req->rq_xprt, struct sock_xprt, xprt);
855 struct sock *sk = transport->inet;
856 int ret = -EAGAIN;
857
858 lock_sock(sk);
859 if (!sock_writeable(sk))
860 ret = xs_nospace(req, transport);
861 release_sock(sk);
862 return ret;
863}
864
865static int xs_stream_nospace(struct rpc_rqst *req, bool vm_wait)
866{
867 struct sock_xprt *transport =
868 container_of(req->rq_xprt, struct sock_xprt, xprt);
869 struct sock *sk = transport->inet;
870 int ret = -EAGAIN;
871
872 if (vm_wait)
873 return -ENOBUFS;
874 lock_sock(sk);
875 if (!sk_stream_memory_free(sk))
876 ret = xs_nospace(req, transport);
877 release_sock(sk);
878 return ret;
879}
880
881static int xs_stream_prepare_request(struct rpc_rqst *req, struct xdr_buf *buf)
882{
883 return xdr_alloc_bvec(buf, rpc_task_gfp_mask());
884}
885
886static void xs_stream_abort_send_request(struct rpc_rqst *req)
887{
888 struct rpc_xprt *xprt = req->rq_xprt;
889 struct sock_xprt *transport =
890 container_of(xprt, struct sock_xprt, xprt);
891
892 if (transport->xmit.offset != 0 &&
893 !test_bit(XPRT_CLOSE_WAIT, &xprt->state))
894 xprt_force_disconnect(xprt);
895}
896
897/*
898 * Determine if the previous message in the stream was aborted before it
899 * could complete transmission.
900 */
901static bool
902xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req)
903{
904 return transport->xmit.offset != 0 && req->rq_bytes_sent == 0;
905}
906
907/*
908 * Return the stream record marker field for a record of length < 2^31-1
909 */
910static rpc_fraghdr
911xs_stream_record_marker(struct xdr_buf *xdr)
912{
913 if (!xdr->len)
914 return 0;
915 return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len);
916}
917
918/**
919 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
920 * @req: pointer to RPC request
921 *
922 * Return values:
923 * 0: The request has been sent
924 * EAGAIN: The socket was blocked, please call again later to
925 * complete the request
926 * ENOTCONN: Caller needs to invoke connect logic then call again
927 * other: Some other error occurred, the request was not sent
928 */
929static int xs_local_send_request(struct rpc_rqst *req)
930{
931 struct rpc_xprt *xprt = req->rq_xprt;
932 struct sock_xprt *transport =
933 container_of(xprt, struct sock_xprt, xprt);
934 struct xdr_buf *xdr = &req->rq_snd_buf;
935 rpc_fraghdr rm = xs_stream_record_marker(xdr);
936 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
937 struct msghdr msg = {
938 .msg_flags = XS_SENDMSG_FLAGS,
939 };
940 bool vm_wait;
941 unsigned int sent;
942 int status;
943
944 /* Close the stream if the previous transmission was incomplete */
945 if (xs_send_request_was_aborted(transport, req)) {
946 xprt_force_disconnect(xprt);
947 return -ENOTCONN;
948 }
949
950 xs_pktdump("packet data:",
951 req->rq_svec->iov_base, req->rq_svec->iov_len);
952
953 vm_wait = sk_stream_is_writeable(transport->inet) ? true : false;
954
955 req->rq_xtime = ktime_get();
956 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
957 transport->xmit.offset, rm, &sent);
958 dprintk("RPC: %s(%u) = %d\n",
959 __func__, xdr->len - transport->xmit.offset, status);
960
961 if (likely(sent > 0) || status == 0) {
962 transport->xmit.offset += sent;
963 req->rq_bytes_sent = transport->xmit.offset;
964 if (likely(req->rq_bytes_sent >= msglen)) {
965 req->rq_xmit_bytes_sent += transport->xmit.offset;
966 transport->xmit.offset = 0;
967 return 0;
968 }
969 status = -EAGAIN;
970 vm_wait = false;
971 }
972
973 switch (status) {
974 case -EAGAIN:
975 status = xs_stream_nospace(req, vm_wait);
976 break;
977 default:
978 dprintk("RPC: sendmsg returned unrecognized error %d\n",
979 -status);
980 fallthrough;
981 case -EPIPE:
982 xprt_force_disconnect(xprt);
983 status = -ENOTCONN;
984 }
985
986 return status;
987}
988
989/**
990 * xs_udp_send_request - write an RPC request to a UDP socket
991 * @req: pointer to RPC request
992 *
993 * Return values:
994 * 0: The request has been sent
995 * EAGAIN: The socket was blocked, please call again later to
996 * complete the request
997 * ENOTCONN: Caller needs to invoke connect logic then call again
998 * other: Some other error occurred, the request was not sent
999 */
1000static int xs_udp_send_request(struct rpc_rqst *req)
1001{
1002 struct rpc_xprt *xprt = req->rq_xprt;
1003 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1004 struct xdr_buf *xdr = &req->rq_snd_buf;
1005 struct msghdr msg = {
1006 .msg_name = xs_addr(xprt),
1007 .msg_namelen = xprt->addrlen,
1008 .msg_flags = XS_SENDMSG_FLAGS,
1009 };
1010 unsigned int sent;
1011 int status;
1012
1013 xs_pktdump("packet data:",
1014 req->rq_svec->iov_base,
1015 req->rq_svec->iov_len);
1016
1017 if (!xprt_bound(xprt))
1018 return -ENOTCONN;
1019
1020 if (!xprt_request_get_cong(xprt, req))
1021 return -EBADSLT;
1022
1023 status = xdr_alloc_bvec(xdr, rpc_task_gfp_mask());
1024 if (status < 0)
1025 return status;
1026 req->rq_xtime = ktime_get();
1027 status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, 0, &sent);
1028
1029 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
1030 xdr->len, status);
1031
1032 /* firewall is blocking us, don't return -EAGAIN or we end up looping */
1033 if (status == -EPERM)
1034 goto process_status;
1035
1036 if (status == -EAGAIN && sock_writeable(transport->inet))
1037 status = -ENOBUFS;
1038
1039 if (sent > 0 || status == 0) {
1040 req->rq_xmit_bytes_sent += sent;
1041 if (sent >= req->rq_slen)
1042 return 0;
1043 /* Still some bytes left; set up for a retry later. */
1044 status = -EAGAIN;
1045 }
1046
1047process_status:
1048 switch (status) {
1049 case -ENOTSOCK:
1050 status = -ENOTCONN;
1051 /* Should we call xs_close() here? */
1052 break;
1053 case -EAGAIN:
1054 status = xs_sock_nospace(req);
1055 break;
1056 case -ENETUNREACH:
1057 case -ENOBUFS:
1058 case -EPIPE:
1059 case -ECONNREFUSED:
1060 case -EPERM:
1061 /* When the server has died, an ICMP port unreachable message
1062 * prompts ECONNREFUSED. */
1063 break;
1064 default:
1065 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1066 -status);
1067 }
1068
1069 return status;
1070}
1071
1072/**
1073 * xs_tcp_send_request - write an RPC request to a TCP socket
1074 * @req: pointer to RPC request
1075 *
1076 * Return values:
1077 * 0: The request has been sent
1078 * EAGAIN: The socket was blocked, please call again later to
1079 * complete the request
1080 * ENOTCONN: Caller needs to invoke connect logic then call again
1081 * other: Some other error occurred, the request was not sent
1082 *
1083 * XXX: In the case of soft timeouts, should we eventually give up
1084 * if sendmsg is not able to make progress?
1085 */
1086static int xs_tcp_send_request(struct rpc_rqst *req)
1087{
1088 struct rpc_xprt *xprt = req->rq_xprt;
1089 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1090 struct xdr_buf *xdr = &req->rq_snd_buf;
1091 rpc_fraghdr rm = xs_stream_record_marker(xdr);
1092 unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
1093 struct msghdr msg = {
1094 .msg_flags = XS_SENDMSG_FLAGS,
1095 };
1096 bool vm_wait;
1097 unsigned int sent;
1098 int status;
1099
1100 /* Close the stream if the previous transmission was incomplete */
1101 if (xs_send_request_was_aborted(transport, req)) {
1102 if (transport->sock != NULL)
1103 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
1104 return -ENOTCONN;
1105 }
1106 if (!transport->inet)
1107 return -ENOTCONN;
1108
1109 xs_pktdump("packet data:",
1110 req->rq_svec->iov_base,
1111 req->rq_svec->iov_len);
1112
1113 if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
1114 xs_tcp_set_socket_timeouts(xprt, transport->sock);
1115
1116 xs_set_srcport(transport, transport->sock);
1117
1118 /* Continue transmitting the packet/record. We must be careful
1119 * to cope with writespace callbacks arriving _after_ we have
1120 * called sendmsg(). */
1121 req->rq_xtime = ktime_get();
1122 tcp_sock_set_cork(transport->inet, true);
1123
1124 vm_wait = sk_stream_is_writeable(transport->inet) ? true : false;
1125
1126 do {
1127 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
1128 transport->xmit.offset, rm, &sent);
1129
1130 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
1131 xdr->len - transport->xmit.offset, status);
1132
1133 /* If we've sent the entire packet, immediately
1134 * reset the count of bytes sent. */
1135 transport->xmit.offset += sent;
1136 req->rq_bytes_sent = transport->xmit.offset;
1137 if (likely(req->rq_bytes_sent >= msglen)) {
1138 req->rq_xmit_bytes_sent += transport->xmit.offset;
1139 transport->xmit.offset = 0;
1140 if (atomic_long_read(&xprt->xmit_queuelen) == 1)
1141 tcp_sock_set_cork(transport->inet, false);
1142 return 0;
1143 }
1144
1145 WARN_ON_ONCE(sent == 0 && status == 0);
1146
1147 if (sent > 0)
1148 vm_wait = false;
1149
1150 } while (status == 0);
1151
1152 switch (status) {
1153 case -ENOTSOCK:
1154 status = -ENOTCONN;
1155 /* Should we call xs_close() here? */
1156 break;
1157 case -EAGAIN:
1158 status = xs_stream_nospace(req, vm_wait);
1159 break;
1160 case -ECONNRESET:
1161 case -ECONNREFUSED:
1162 case -ENOTCONN:
1163 case -EADDRINUSE:
1164 case -ENOBUFS:
1165 case -EPIPE:
1166 break;
1167 default:
1168 dprintk("RPC: sendmsg returned unrecognized error %d\n",
1169 -status);
1170 }
1171
1172 return status;
1173}
1174
1175static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1176{
1177 transport->old_data_ready = sk->sk_data_ready;
1178 transport->old_state_change = sk->sk_state_change;
1179 transport->old_write_space = sk->sk_write_space;
1180 transport->old_error_report = sk->sk_error_report;
1181}
1182
1183static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1184{
1185 sk->sk_data_ready = transport->old_data_ready;
1186 sk->sk_state_change = transport->old_state_change;
1187 sk->sk_write_space = transport->old_write_space;
1188 sk->sk_error_report = transport->old_error_report;
1189}
1190
1191static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
1192{
1193 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1194
1195 transport->xprt_err = 0;
1196 clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1197 clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state);
1198 clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state);
1199 clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state);
1200 clear_bit(XPRT_SOCK_NOSPACE, &transport->sock_state);
1201 clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
1202}
1203
1204static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr)
1205{
1206 set_bit(nr, &transport->sock_state);
1207 queue_work(xprtiod_workqueue, &transport->error_worker);
1208}
1209
1210static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1211{
1212 xprt->connect_cookie++;
1213 smp_mb__before_atomic();
1214 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1215 clear_bit(XPRT_CLOSING, &xprt->state);
1216 xs_sock_reset_state_flags(xprt);
1217 smp_mb__after_atomic();
1218}
1219
1220/**
1221 * xs_error_report - callback to handle TCP socket state errors
1222 * @sk: socket
1223 *
1224 * Note: we don't call sock_error() since there may be a rpc_task
1225 * using the socket, and so we don't want to clear sk->sk_err.
1226 */
1227static void xs_error_report(struct sock *sk)
1228{
1229 struct sock_xprt *transport;
1230 struct rpc_xprt *xprt;
1231
1232 if (!(xprt = xprt_from_sock(sk)))
1233 return;
1234
1235 transport = container_of(xprt, struct sock_xprt, xprt);
1236 transport->xprt_err = -sk->sk_err;
1237 if (transport->xprt_err == 0)
1238 return;
1239 dprintk("RPC: xs_error_report client %p, error=%d...\n",
1240 xprt, -transport->xprt_err);
1241 trace_rpc_socket_error(xprt, sk->sk_socket, transport->xprt_err);
1242
1243 /* barrier ensures xprt_err is set before XPRT_SOCK_WAKE_ERROR */
1244 smp_mb__before_atomic();
1245 xs_run_error_worker(transport, XPRT_SOCK_WAKE_ERROR);
1246}
1247
1248static void xs_reset_transport(struct sock_xprt *transport)
1249{
1250 struct socket *sock = transport->sock;
1251 struct sock *sk = transport->inet;
1252 struct rpc_xprt *xprt = &transport->xprt;
1253 struct file *filp = transport->file;
1254
1255 if (sk == NULL)
1256 return;
1257 /*
1258 * Make sure we're calling this in a context from which it is safe
1259 * to call __fput_sync(). In practice that means rpciod and the
1260 * system workqueue.
1261 */
1262 if (!(current->flags & PF_WQ_WORKER)) {
1263 WARN_ON_ONCE(1);
1264 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
1265 return;
1266 }
1267
1268 if (atomic_read(&transport->xprt.swapper))
1269 sk_clear_memalloc(sk);
1270
1271 tls_handshake_cancel(sk);
1272
1273 kernel_sock_shutdown(sock, SHUT_RDWR);
1274
1275 mutex_lock(&transport->recv_mutex);
1276 lock_sock(sk);
1277 transport->inet = NULL;
1278 transport->sock = NULL;
1279 transport->file = NULL;
1280
1281 sk->sk_user_data = NULL;
1282 sk->sk_sndtimeo = 0;
1283
1284 xs_restore_old_callbacks(transport, sk);
1285 xprt_clear_connected(xprt);
1286 xs_sock_reset_connection_flags(xprt);
1287 /* Reset stream record info */
1288 xs_stream_reset_connect(transport);
1289 release_sock(sk);
1290 mutex_unlock(&transport->recv_mutex);
1291
1292 trace_rpc_socket_close(xprt, sock);
1293 __fput_sync(filp);
1294
1295 xprt_disconnect_done(xprt);
1296}
1297
1298/**
1299 * xs_close - close a socket
1300 * @xprt: transport
1301 *
1302 * This is used when all requests are complete; ie, no DRC state remains
1303 * on the server we want to save.
1304 *
1305 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1306 * xs_reset_transport() zeroing the socket from underneath a writer.
1307 */
1308static void xs_close(struct rpc_xprt *xprt)
1309{
1310 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1311
1312 dprintk("RPC: xs_close xprt %p\n", xprt);
1313
1314 if (transport->sock)
1315 tls_handshake_close(transport->sock);
1316 xs_reset_transport(transport);
1317 xprt->reestablish_timeout = 0;
1318}
1319
1320static void xs_inject_disconnect(struct rpc_xprt *xprt)
1321{
1322 dprintk("RPC: injecting transport disconnect on xprt=%p\n",
1323 xprt);
1324 xprt_disconnect_done(xprt);
1325}
1326
1327static void xs_xprt_free(struct rpc_xprt *xprt)
1328{
1329 xs_free_peer_addresses(xprt);
1330 xprt_free(xprt);
1331}
1332
1333/**
1334 * xs_destroy - prepare to shutdown a transport
1335 * @xprt: doomed transport
1336 *
1337 */
1338static void xs_destroy(struct rpc_xprt *xprt)
1339{
1340 struct sock_xprt *transport = container_of(xprt,
1341 struct sock_xprt, xprt);
1342 dprintk("RPC: xs_destroy xprt %p\n", xprt);
1343
1344 cancel_delayed_work_sync(&transport->connect_worker);
1345 xs_close(xprt);
1346 cancel_work_sync(&transport->recv_worker);
1347 cancel_work_sync(&transport->error_worker);
1348 xs_xprt_free(xprt);
1349 module_put(THIS_MODULE);
1350}
1351
1352/**
1353 * xs_udp_data_read_skb - receive callback for UDP sockets
1354 * @xprt: transport
1355 * @sk: socket
1356 * @skb: skbuff
1357 *
1358 */
1359static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1360 struct sock *sk,
1361 struct sk_buff *skb)
1362{
1363 struct rpc_task *task;
1364 struct rpc_rqst *rovr;
1365 int repsize, copied;
1366 u32 _xid;
1367 __be32 *xp;
1368
1369 repsize = skb->len;
1370 if (repsize < 4) {
1371 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
1372 return;
1373 }
1374
1375 /* Copy the XID from the skb... */
1376 xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1377 if (xp == NULL)
1378 return;
1379
1380 /* Look up and lock the request corresponding to the given XID */
1381 spin_lock(&xprt->queue_lock);
1382 rovr = xprt_lookup_rqst(xprt, *xp);
1383 if (!rovr)
1384 goto out_unlock;
1385 xprt_pin_rqst(rovr);
1386 xprt_update_rtt(rovr->rq_task);
1387 spin_unlock(&xprt->queue_lock);
1388 task = rovr->rq_task;
1389
1390 if ((copied = rovr->rq_private_buf.buflen) > repsize)
1391 copied = repsize;
1392
1393 /* Suck it into the iovec, verify checksum if not done by hw. */
1394 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1395 spin_lock(&xprt->queue_lock);
1396 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1397 goto out_unpin;
1398 }
1399
1400
1401 spin_lock(&xprt->transport_lock);
1402 xprt_adjust_cwnd(xprt, task, copied);
1403 spin_unlock(&xprt->transport_lock);
1404 spin_lock(&xprt->queue_lock);
1405 xprt_complete_rqst(task, copied);
1406 __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1407out_unpin:
1408 xprt_unpin_rqst(rovr);
1409 out_unlock:
1410 spin_unlock(&xprt->queue_lock);
1411}
1412
1413static void xs_udp_data_receive(struct sock_xprt *transport)
1414{
1415 struct sk_buff *skb;
1416 struct sock *sk;
1417 int err;
1418
1419 mutex_lock(&transport->recv_mutex);
1420 sk = transport->inet;
1421 if (sk == NULL)
1422 goto out;
1423 for (;;) {
1424 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err);
1425 if (skb == NULL)
1426 break;
1427 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1428 consume_skb(skb);
1429 cond_resched();
1430 }
1431 xs_poll_check_readable(transport);
1432out:
1433 mutex_unlock(&transport->recv_mutex);
1434}
1435
1436static void xs_udp_data_receive_workfn(struct work_struct *work)
1437{
1438 struct sock_xprt *transport =
1439 container_of(work, struct sock_xprt, recv_worker);
1440 unsigned int pflags = memalloc_nofs_save();
1441
1442 xs_udp_data_receive(transport);
1443 memalloc_nofs_restore(pflags);
1444}
1445
1446/**
1447 * xs_data_ready - "data ready" callback for sockets
1448 * @sk: socket with data to read
1449 *
1450 */
1451static void xs_data_ready(struct sock *sk)
1452{
1453 struct rpc_xprt *xprt;
1454
1455 trace_sk_data_ready(sk);
1456
1457 xprt = xprt_from_sock(sk);
1458 if (xprt != NULL) {
1459 struct sock_xprt *transport = container_of(xprt,
1460 struct sock_xprt, xprt);
1461
1462 trace_xs_data_ready(xprt);
1463
1464 transport->old_data_ready(sk);
1465
1466 if (test_bit(XPRT_SOCK_IGNORE_RECV, &transport->sock_state))
1467 return;
1468
1469 /* Any data means we had a useful conversation, so
1470 * then we don't need to delay the next reconnect
1471 */
1472 if (xprt->reestablish_timeout)
1473 xprt->reestablish_timeout = 0;
1474 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1475 queue_work(xprtiod_workqueue, &transport->recv_worker);
1476 }
1477}
1478
1479/*
1480 * Helper function to force a TCP close if the server is sending
1481 * junk and/or it has put us in CLOSE_WAIT
1482 */
1483static void xs_tcp_force_close(struct rpc_xprt *xprt)
1484{
1485 xprt_force_disconnect(xprt);
1486}
1487
1488#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1489static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1490{
1491 return PAGE_SIZE;
1492}
1493#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1494
1495/**
1496 * xs_local_state_change - callback to handle AF_LOCAL socket state changes
1497 * @sk: socket whose state has changed
1498 *
1499 */
1500static void xs_local_state_change(struct sock *sk)
1501{
1502 struct rpc_xprt *xprt;
1503 struct sock_xprt *transport;
1504
1505 if (!(xprt = xprt_from_sock(sk)))
1506 return;
1507 transport = container_of(xprt, struct sock_xprt, xprt);
1508 if (sk->sk_shutdown & SHUTDOWN_MASK) {
1509 clear_bit(XPRT_CONNECTED, &xprt->state);
1510 /* Trigger the socket release */
1511 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1512 }
1513}
1514
1515/**
1516 * xs_tcp_state_change - callback to handle TCP socket state changes
1517 * @sk: socket whose state has changed
1518 *
1519 */
1520static void xs_tcp_state_change(struct sock *sk)
1521{
1522 struct rpc_xprt *xprt;
1523 struct sock_xprt *transport;
1524
1525 if (!(xprt = xprt_from_sock(sk)))
1526 return;
1527 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
1528 dprintk("RPC: state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1529 sk->sk_state, xprt_connected(xprt),
1530 sock_flag(sk, SOCK_DEAD),
1531 sock_flag(sk, SOCK_ZAPPED),
1532 sk->sk_shutdown);
1533
1534 transport = container_of(xprt, struct sock_xprt, xprt);
1535 trace_rpc_socket_state_change(xprt, sk->sk_socket);
1536 switch (sk->sk_state) {
1537 case TCP_ESTABLISHED:
1538 if (!xprt_test_and_set_connected(xprt)) {
1539 xprt->connect_cookie++;
1540 clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1541 xprt_clear_connecting(xprt);
1542
1543 xprt->stat.connect_count++;
1544 xprt->stat.connect_time += (long)jiffies -
1545 xprt->stat.connect_start;
1546 xs_run_error_worker(transport, XPRT_SOCK_WAKE_PENDING);
1547 }
1548 break;
1549 case TCP_FIN_WAIT1:
1550 /* The client initiated a shutdown of the socket */
1551 xprt->connect_cookie++;
1552 xprt->reestablish_timeout = 0;
1553 set_bit(XPRT_CLOSING, &xprt->state);
1554 smp_mb__before_atomic();
1555 clear_bit(XPRT_CONNECTED, &xprt->state);
1556 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1557 smp_mb__after_atomic();
1558 break;
1559 case TCP_CLOSE_WAIT:
1560 /* The server initiated a shutdown of the socket */
1561 xprt->connect_cookie++;
1562 clear_bit(XPRT_CONNECTED, &xprt->state);
1563 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1564 fallthrough;
1565 case TCP_CLOSING:
1566 /*
1567 * If the server closed down the connection, make sure that
1568 * we back off before reconnecting
1569 */
1570 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1571 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1572 break;
1573 case TCP_LAST_ACK:
1574 set_bit(XPRT_CLOSING, &xprt->state);
1575 smp_mb__before_atomic();
1576 clear_bit(XPRT_CONNECTED, &xprt->state);
1577 smp_mb__after_atomic();
1578 break;
1579 case TCP_CLOSE:
1580 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1581 &transport->sock_state)) {
1582 xs_reset_srcport(transport);
1583 xprt_clear_connecting(xprt);
1584 }
1585 clear_bit(XPRT_CLOSING, &xprt->state);
1586 /* Trigger the socket release */
1587 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1588 }
1589}
1590
1591static void xs_write_space(struct sock *sk)
1592{
1593 struct sock_xprt *transport;
1594 struct rpc_xprt *xprt;
1595
1596 if (!sk->sk_socket)
1597 return;
1598 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1599
1600 if (unlikely(!(xprt = xprt_from_sock(sk))))
1601 return;
1602 transport = container_of(xprt, struct sock_xprt, xprt);
1603 if (!test_and_clear_bit(XPRT_SOCK_NOSPACE, &transport->sock_state))
1604 return;
1605 xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE);
1606 sk->sk_write_pending--;
1607}
1608
1609/**
1610 * xs_udp_write_space - callback invoked when socket buffer space
1611 * becomes available
1612 * @sk: socket whose state has changed
1613 *
1614 * Called when more output buffer space is available for this socket.
1615 * We try not to wake our writers until they can make "significant"
1616 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1617 * with a bunch of small requests.
1618 */
1619static void xs_udp_write_space(struct sock *sk)
1620{
1621 /* from net/core/sock.c:sock_def_write_space */
1622 if (sock_writeable(sk))
1623 xs_write_space(sk);
1624}
1625
1626/**
1627 * xs_tcp_write_space - callback invoked when socket buffer space
1628 * becomes available
1629 * @sk: socket whose state has changed
1630 *
1631 * Called when more output buffer space is available for this socket.
1632 * We try not to wake our writers until they can make "significant"
1633 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1634 * with a bunch of small requests.
1635 */
1636static void xs_tcp_write_space(struct sock *sk)
1637{
1638 /* from net/core/stream.c:sk_stream_write_space */
1639 if (sk_stream_is_writeable(sk))
1640 xs_write_space(sk);
1641}
1642
1643static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1644{
1645 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1646 struct sock *sk = transport->inet;
1647
1648 if (transport->rcvsize) {
1649 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1650 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1651 }
1652 if (transport->sndsize) {
1653 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1654 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1655 sk->sk_write_space(sk);
1656 }
1657}
1658
1659/**
1660 * xs_udp_set_buffer_size - set send and receive limits
1661 * @xprt: generic transport
1662 * @sndsize: requested size of send buffer, in bytes
1663 * @rcvsize: requested size of receive buffer, in bytes
1664 *
1665 * Set socket send and receive buffer size limits.
1666 */
1667static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1668{
1669 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1670
1671 transport->sndsize = 0;
1672 if (sndsize)
1673 transport->sndsize = sndsize + 1024;
1674 transport->rcvsize = 0;
1675 if (rcvsize)
1676 transport->rcvsize = rcvsize + 1024;
1677
1678 xs_udp_do_set_buffer_size(xprt);
1679}
1680
1681/**
1682 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1683 * @xprt: controlling transport
1684 * @task: task that timed out
1685 *
1686 * Adjust the congestion window after a retransmit timeout has occurred.
1687 */
1688static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1689{
1690 spin_lock(&xprt->transport_lock);
1691 xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1692 spin_unlock(&xprt->transport_lock);
1693}
1694
1695static int xs_get_random_port(void)
1696{
1697 unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1698 unsigned short range;
1699 unsigned short rand;
1700
1701 if (max < min)
1702 return -EADDRINUSE;
1703 range = max - min + 1;
1704 rand = get_random_u32_below(range);
1705 return rand + min;
1706}
1707
1708static unsigned short xs_sock_getport(struct socket *sock)
1709{
1710 struct sockaddr_storage buf;
1711 unsigned short port = 0;
1712
1713 if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1714 goto out;
1715 switch (buf.ss_family) {
1716 case AF_INET6:
1717 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1718 break;
1719 case AF_INET:
1720 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1721 }
1722out:
1723 return port;
1724}
1725
1726/**
1727 * xs_set_port - reset the port number in the remote endpoint address
1728 * @xprt: generic transport
1729 * @port: new port number
1730 *
1731 */
1732static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1733{
1734 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
1735
1736 rpc_set_port(xs_addr(xprt), port);
1737 xs_update_peer_port(xprt);
1738}
1739
1740static void xs_reset_srcport(struct sock_xprt *transport)
1741{
1742 transport->srcport = 0;
1743}
1744
1745static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1746{
1747 if (transport->srcport == 0 && transport->xprt.reuseport)
1748 transport->srcport = xs_sock_getport(sock);
1749}
1750
1751static int xs_get_srcport(struct sock_xprt *transport)
1752{
1753 int port = transport->srcport;
1754
1755 if (port == 0 && transport->xprt.resvport)
1756 port = xs_get_random_port();
1757 return port;
1758}
1759
1760static unsigned short xs_sock_srcport(struct rpc_xprt *xprt)
1761{
1762 struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt);
1763 unsigned short ret = 0;
1764 mutex_lock(&sock->recv_mutex);
1765 if (sock->sock)
1766 ret = xs_sock_getport(sock->sock);
1767 mutex_unlock(&sock->recv_mutex);
1768 return ret;
1769}
1770
1771static int xs_sock_srcaddr(struct rpc_xprt *xprt, char *buf, size_t buflen)
1772{
1773 struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt);
1774 union {
1775 struct sockaddr sa;
1776 struct sockaddr_storage st;
1777 } saddr;
1778 int ret = -ENOTCONN;
1779
1780 mutex_lock(&sock->recv_mutex);
1781 if (sock->sock) {
1782 ret = kernel_getsockname(sock->sock, &saddr.sa);
1783 if (ret >= 0)
1784 ret = snprintf(buf, buflen, "%pISc", &saddr.sa);
1785 }
1786 mutex_unlock(&sock->recv_mutex);
1787 return ret;
1788}
1789
1790static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1791{
1792 if (transport->srcport != 0)
1793 transport->srcport = 0;
1794 if (!transport->xprt.resvport)
1795 return 0;
1796 if (port <= xprt_min_resvport || port > xprt_max_resvport)
1797 return xprt_max_resvport;
1798 return --port;
1799}
1800static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1801{
1802 struct sockaddr_storage myaddr;
1803 int err, nloop = 0;
1804 int port = xs_get_srcport(transport);
1805 unsigned short last;
1806
1807 /*
1808 * If we are asking for any ephemeral port (i.e. port == 0 &&
1809 * transport->xprt.resvport == 0), don't bind. Let the local
1810 * port selection happen implicitly when the socket is used
1811 * (for example at connect time).
1812 *
1813 * This ensures that we can continue to establish TCP
1814 * connections even when all local ephemeral ports are already
1815 * a part of some TCP connection. This makes no difference
1816 * for UDP sockets, but also doesn't harm them.
1817 *
1818 * If we're asking for any reserved port (i.e. port == 0 &&
1819 * transport->xprt.resvport == 1) xs_get_srcport above will
1820 * ensure that port is non-zero and we will bind as needed.
1821 */
1822 if (port <= 0)
1823 return port;
1824
1825 memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1826 do {
1827 rpc_set_port((struct sockaddr *)&myaddr, port);
1828 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1829 transport->xprt.addrlen);
1830 if (err == 0) {
1831 if (transport->xprt.reuseport)
1832 transport->srcport = port;
1833 break;
1834 }
1835 last = port;
1836 port = xs_next_srcport(transport, port);
1837 if (port > last)
1838 nloop++;
1839 } while (err == -EADDRINUSE && nloop != 2);
1840
1841 if (myaddr.ss_family == AF_INET)
1842 dprintk("RPC: %s %pI4:%u: %s (%d)\n", __func__,
1843 &((struct sockaddr_in *)&myaddr)->sin_addr,
1844 port, err ? "failed" : "ok", err);
1845 else
1846 dprintk("RPC: %s %pI6:%u: %s (%d)\n", __func__,
1847 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1848 port, err ? "failed" : "ok", err);
1849 return err;
1850}
1851
1852/*
1853 * We don't support autobind on AF_LOCAL sockets
1854 */
1855static void xs_local_rpcbind(struct rpc_task *task)
1856{
1857 xprt_set_bound(task->tk_xprt);
1858}
1859
1860static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1861{
1862}
1863
1864#ifdef CONFIG_DEBUG_LOCK_ALLOC
1865static struct lock_class_key xs_key[3];
1866static struct lock_class_key xs_slock_key[3];
1867
1868static inline void xs_reclassify_socketu(struct socket *sock)
1869{
1870 struct sock *sk = sock->sk;
1871
1872 sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1873 &xs_slock_key[0], "sk_lock-AF_LOCAL-RPC", &xs_key[0]);
1874}
1875
1876static inline void xs_reclassify_socket4(struct socket *sock)
1877{
1878 struct sock *sk = sock->sk;
1879
1880 sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1881 &xs_slock_key[1], "sk_lock-AF_INET-RPC", &xs_key[1]);
1882}
1883
1884static inline void xs_reclassify_socket6(struct socket *sock)
1885{
1886 struct sock *sk = sock->sk;
1887
1888 sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1889 &xs_slock_key[2], "sk_lock-AF_INET6-RPC", &xs_key[2]);
1890}
1891
1892static inline void xs_reclassify_socket(int family, struct socket *sock)
1893{
1894 if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1895 return;
1896
1897 switch (family) {
1898 case AF_LOCAL:
1899 xs_reclassify_socketu(sock);
1900 break;
1901 case AF_INET:
1902 xs_reclassify_socket4(sock);
1903 break;
1904 case AF_INET6:
1905 xs_reclassify_socket6(sock);
1906 break;
1907 }
1908}
1909#else
1910static inline void xs_reclassify_socket(int family, struct socket *sock)
1911{
1912}
1913#endif
1914
1915static void xs_dummy_setup_socket(struct work_struct *work)
1916{
1917}
1918
1919static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1920 struct sock_xprt *transport, int family, int type,
1921 int protocol, bool reuseport)
1922{
1923 struct file *filp;
1924 struct socket *sock;
1925 int err;
1926
1927 err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1928 if (err < 0) {
1929 dprintk("RPC: can't create %d transport socket (%d).\n",
1930 protocol, -err);
1931 goto out;
1932 }
1933 xs_reclassify_socket(family, sock);
1934
1935 if (reuseport)
1936 sock_set_reuseport(sock->sk);
1937
1938 err = xs_bind(transport, sock);
1939 if (err) {
1940 sock_release(sock);
1941 goto out;
1942 }
1943
1944 if (protocol == IPPROTO_TCP)
1945 sk_net_refcnt_upgrade(sock->sk);
1946
1947 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1948 if (IS_ERR(filp))
1949 return ERR_CAST(filp);
1950 transport->file = filp;
1951
1952 return sock;
1953out:
1954 return ERR_PTR(err);
1955}
1956
1957static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1958 struct socket *sock)
1959{
1960 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1961 xprt);
1962
1963 if (!transport->inet) {
1964 struct sock *sk = sock->sk;
1965
1966 lock_sock(sk);
1967
1968 xs_save_old_callbacks(transport, sk);
1969
1970 sk->sk_user_data = xprt;
1971 sk->sk_data_ready = xs_data_ready;
1972 sk->sk_write_space = xs_udp_write_space;
1973 sk->sk_state_change = xs_local_state_change;
1974 sk->sk_error_report = xs_error_report;
1975 sk->sk_use_task_frag = false;
1976
1977 xprt_clear_connected(xprt);
1978
1979 /* Reset to new socket */
1980 transport->sock = sock;
1981 transport->inet = sk;
1982
1983 release_sock(sk);
1984 }
1985
1986 xs_stream_start_connect(transport);
1987
1988 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1989}
1990
1991/**
1992 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1993 * @transport: socket transport to connect
1994 */
1995static int xs_local_setup_socket(struct sock_xprt *transport)
1996{
1997 struct rpc_xprt *xprt = &transport->xprt;
1998 struct file *filp;
1999 struct socket *sock;
2000 int status;
2001
2002 status = __sock_create(xprt->xprt_net, AF_LOCAL,
2003 SOCK_STREAM, 0, &sock, 1);
2004 if (status < 0) {
2005 dprintk("RPC: can't create AF_LOCAL "
2006 "transport socket (%d).\n", -status);
2007 goto out;
2008 }
2009 xs_reclassify_socket(AF_LOCAL, sock);
2010
2011 filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
2012 if (IS_ERR(filp)) {
2013 status = PTR_ERR(filp);
2014 goto out;
2015 }
2016 transport->file = filp;
2017
2018 dprintk("RPC: worker connecting xprt %p via AF_LOCAL to %s\n",
2019 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2020
2021 status = xs_local_finish_connecting(xprt, sock);
2022 trace_rpc_socket_connect(xprt, sock, status);
2023 switch (status) {
2024 case 0:
2025 dprintk("RPC: xprt %p connected to %s\n",
2026 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2027 xprt->stat.connect_count++;
2028 xprt->stat.connect_time += (long)jiffies -
2029 xprt->stat.connect_start;
2030 xprt_set_connected(xprt);
2031 break;
2032 case -ENOBUFS:
2033 break;
2034 case -ENOENT:
2035 dprintk("RPC: xprt %p: socket %s does not exist\n",
2036 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2037 break;
2038 case -ECONNREFUSED:
2039 dprintk("RPC: xprt %p: connection refused for %s\n",
2040 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2041 break;
2042 default:
2043 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2044 __func__, -status,
2045 xprt->address_strings[RPC_DISPLAY_ADDR]);
2046 }
2047
2048out:
2049 xprt_clear_connecting(xprt);
2050 xprt_wake_pending_tasks(xprt, status);
2051 return status;
2052}
2053
2054static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2055{
2056 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2057 int ret;
2058
2059 if (transport->file)
2060 goto force_disconnect;
2061
2062 if (RPC_IS_ASYNC(task)) {
2063 /*
2064 * We want the AF_LOCAL connect to be resolved in the
2065 * filesystem namespace of the process making the rpc
2066 * call. Thus we connect synchronously.
2067 *
2068 * If we want to support asynchronous AF_LOCAL calls,
2069 * we'll need to figure out how to pass a namespace to
2070 * connect.
2071 */
2072 rpc_task_set_rpc_status(task, -ENOTCONN);
2073 goto out_wake;
2074 }
2075 ret = xs_local_setup_socket(transport);
2076 if (ret && !RPC_IS_SOFTCONN(task))
2077 msleep_interruptible(15000);
2078 return;
2079force_disconnect:
2080 xprt_force_disconnect(xprt);
2081out_wake:
2082 xprt_clear_connecting(xprt);
2083 xprt_wake_pending_tasks(xprt, -ENOTCONN);
2084}
2085
2086#if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2087/*
2088 * Note that this should be called with XPRT_LOCKED held, or recv_mutex
2089 * held, or when we otherwise know that we have exclusive access to the
2090 * socket, to guard against races with xs_reset_transport.
2091 */
2092static void xs_set_memalloc(struct rpc_xprt *xprt)
2093{
2094 struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2095 xprt);
2096
2097 /*
2098 * If there's no sock, then we have nothing to set. The
2099 * reconnecting process will get it for us.
2100 */
2101 if (!transport->inet)
2102 return;
2103 if (atomic_read(&xprt->swapper))
2104 sk_set_memalloc(transport->inet);
2105}
2106
2107/**
2108 * xs_enable_swap - Tag this transport as being used for swap.
2109 * @xprt: transport to tag
2110 *
2111 * Take a reference to this transport on behalf of the rpc_clnt, and
2112 * optionally mark it for swapping if it wasn't already.
2113 */
2114static int
2115xs_enable_swap(struct rpc_xprt *xprt)
2116{
2117 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2118
2119 mutex_lock(&xs->recv_mutex);
2120 if (atomic_inc_return(&xprt->swapper) == 1 &&
2121 xs->inet)
2122 sk_set_memalloc(xs->inet);
2123 mutex_unlock(&xs->recv_mutex);
2124 return 0;
2125}
2126
2127/**
2128 * xs_disable_swap - Untag this transport as being used for swap.
2129 * @xprt: transport to tag
2130 *
2131 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2132 * swapper refcount goes to 0, untag the socket as a memalloc socket.
2133 */
2134static void
2135xs_disable_swap(struct rpc_xprt *xprt)
2136{
2137 struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2138
2139 mutex_lock(&xs->recv_mutex);
2140 if (atomic_dec_and_test(&xprt->swapper) &&
2141 xs->inet)
2142 sk_clear_memalloc(xs->inet);
2143 mutex_unlock(&xs->recv_mutex);
2144}
2145#else
2146static void xs_set_memalloc(struct rpc_xprt *xprt)
2147{
2148}
2149
2150static int
2151xs_enable_swap(struct rpc_xprt *xprt)
2152{
2153 return -EINVAL;
2154}
2155
2156static void
2157xs_disable_swap(struct rpc_xprt *xprt)
2158{
2159}
2160#endif
2161
2162static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2163{
2164 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2165
2166 if (!transport->inet) {
2167 struct sock *sk = sock->sk;
2168
2169 lock_sock(sk);
2170
2171 xs_save_old_callbacks(transport, sk);
2172
2173 sk->sk_user_data = xprt;
2174 sk->sk_data_ready = xs_data_ready;
2175 sk->sk_write_space = xs_udp_write_space;
2176 sk->sk_use_task_frag = false;
2177
2178 xprt_set_connected(xprt);
2179
2180 /* Reset to new socket */
2181 transport->sock = sock;
2182 transport->inet = sk;
2183
2184 xs_set_memalloc(xprt);
2185
2186 release_sock(sk);
2187 }
2188 xs_udp_do_set_buffer_size(xprt);
2189
2190 xprt->stat.connect_start = jiffies;
2191}
2192
2193static void xs_udp_setup_socket(struct work_struct *work)
2194{
2195 struct sock_xprt *transport =
2196 container_of(work, struct sock_xprt, connect_worker.work);
2197 struct rpc_xprt *xprt = &transport->xprt;
2198 struct socket *sock;
2199 int status = -EIO;
2200 unsigned int pflags = current->flags;
2201
2202 if (atomic_read(&xprt->swapper))
2203 current->flags |= PF_MEMALLOC;
2204 sock = xs_create_sock(xprt, transport,
2205 xs_addr(xprt)->sa_family, SOCK_DGRAM,
2206 IPPROTO_UDP, false);
2207 if (IS_ERR(sock))
2208 goto out;
2209
2210 dprintk("RPC: worker connecting xprt %p via %s to "
2211 "%s (port %s)\n", xprt,
2212 xprt->address_strings[RPC_DISPLAY_PROTO],
2213 xprt->address_strings[RPC_DISPLAY_ADDR],
2214 xprt->address_strings[RPC_DISPLAY_PORT]);
2215
2216 xs_udp_finish_connecting(xprt, sock);
2217 trace_rpc_socket_connect(xprt, sock, 0);
2218 status = 0;
2219out:
2220 xprt_clear_connecting(xprt);
2221 xprt_unlock_connect(xprt, transport);
2222 xprt_wake_pending_tasks(xprt, status);
2223 current_restore_flags(pflags, PF_MEMALLOC);
2224}
2225
2226/**
2227 * xs_tcp_shutdown - gracefully shut down a TCP socket
2228 * @xprt: transport
2229 *
2230 * Initiates a graceful shutdown of the TCP socket by calling the
2231 * equivalent of shutdown(SHUT_RDWR);
2232 */
2233static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2234{
2235 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2236 struct socket *sock = transport->sock;
2237 int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2238
2239 if (sock == NULL)
2240 return;
2241 if (!xprt->reuseport) {
2242 xs_close(xprt);
2243 return;
2244 }
2245 switch (skst) {
2246 case TCP_FIN_WAIT1:
2247 case TCP_FIN_WAIT2:
2248 case TCP_LAST_ACK:
2249 break;
2250 case TCP_ESTABLISHED:
2251 case TCP_CLOSE_WAIT:
2252 kernel_sock_shutdown(sock, SHUT_RDWR);
2253 trace_rpc_socket_shutdown(xprt, sock);
2254 break;
2255 default:
2256 xs_reset_transport(transport);
2257 }
2258}
2259
2260static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2261 struct socket *sock)
2262{
2263 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2264 struct net *net = sock_net(sock->sk);
2265 unsigned long connect_timeout;
2266 unsigned long syn_retries;
2267 unsigned int keepidle;
2268 unsigned int keepcnt;
2269 unsigned int timeo;
2270 unsigned long t;
2271
2272 spin_lock(&xprt->transport_lock);
2273 keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2274 keepcnt = xprt->timeout->to_retries + 1;
2275 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2276 (xprt->timeout->to_retries + 1);
2277 clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2278 spin_unlock(&xprt->transport_lock);
2279
2280 /* TCP Keepalive options */
2281 sock_set_keepalive(sock->sk);
2282 tcp_sock_set_keepidle(sock->sk, keepidle);
2283 tcp_sock_set_keepintvl(sock->sk, keepidle);
2284 tcp_sock_set_keepcnt(sock->sk, keepcnt);
2285
2286 /* TCP user timeout (see RFC5482) */
2287 tcp_sock_set_user_timeout(sock->sk, timeo);
2288
2289 /* Connect timeout */
2290 connect_timeout = max_t(unsigned long,
2291 DIV_ROUND_UP(xprt->connect_timeout, HZ), 1);
2292 syn_retries = max_t(unsigned long,
2293 READ_ONCE(net->ipv4.sysctl_tcp_syn_retries), 1);
2294 for (t = 0; t <= syn_retries && (1UL << t) < connect_timeout; t++)
2295 ;
2296 if (t <= syn_retries)
2297 tcp_sock_set_syncnt(sock->sk, t - 1);
2298}
2299
2300static void xs_tcp_do_set_connect_timeout(struct rpc_xprt *xprt,
2301 unsigned long connect_timeout)
2302{
2303 struct sock_xprt *transport =
2304 container_of(xprt, struct sock_xprt, xprt);
2305 struct rpc_timeout to;
2306 unsigned long initval;
2307
2308 memcpy(&to, xprt->timeout, sizeof(to));
2309 /* Arbitrary lower limit */
2310 initval = max_t(unsigned long, connect_timeout, XS_TCP_INIT_REEST_TO);
2311 to.to_initval = initval;
2312 to.to_maxval = initval;
2313 to.to_retries = 0;
2314 memcpy(&transport->tcp_timeout, &to, sizeof(transport->tcp_timeout));
2315 xprt->timeout = &transport->tcp_timeout;
2316 xprt->connect_timeout = connect_timeout;
2317}
2318
2319static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2320 unsigned long connect_timeout,
2321 unsigned long reconnect_timeout)
2322{
2323 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2324
2325 spin_lock(&xprt->transport_lock);
2326 if (reconnect_timeout < xprt->max_reconnect_timeout)
2327 xprt->max_reconnect_timeout = reconnect_timeout;
2328 if (connect_timeout < xprt->connect_timeout)
2329 xs_tcp_do_set_connect_timeout(xprt, connect_timeout);
2330 set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2331 spin_unlock(&xprt->transport_lock);
2332}
2333
2334static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2335{
2336 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2337
2338 if (!transport->inet) {
2339 struct sock *sk = sock->sk;
2340
2341 /* Avoid temporary address, they are bad for long-lived
2342 * connections such as NFS mounts.
2343 * RFC4941, section 3.6 suggests that:
2344 * Individual applications, which have specific
2345 * knowledge about the normal duration of connections,
2346 * MAY override this as appropriate.
2347 */
2348 if (xs_addr(xprt)->sa_family == PF_INET6) {
2349 ip6_sock_set_addr_preferences(sk,
2350 IPV6_PREFER_SRC_PUBLIC);
2351 }
2352
2353 xs_tcp_set_socket_timeouts(xprt, sock);
2354 tcp_sock_set_nodelay(sk);
2355
2356 lock_sock(sk);
2357
2358 xs_save_old_callbacks(transport, sk);
2359
2360 sk->sk_user_data = xprt;
2361 sk->sk_data_ready = xs_data_ready;
2362 sk->sk_state_change = xs_tcp_state_change;
2363 sk->sk_write_space = xs_tcp_write_space;
2364 sk->sk_error_report = xs_error_report;
2365 sk->sk_use_task_frag = false;
2366
2367 /* socket options */
2368 sock_reset_flag(sk, SOCK_LINGER);
2369
2370 xprt_clear_connected(xprt);
2371
2372 /* Reset to new socket */
2373 transport->sock = sock;
2374 transport->inet = sk;
2375
2376 release_sock(sk);
2377 }
2378
2379 if (!xprt_bound(xprt))
2380 return -ENOTCONN;
2381
2382 xs_set_memalloc(xprt);
2383
2384 xs_stream_start_connect(transport);
2385
2386 /* Tell the socket layer to start connecting... */
2387 set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2388 return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2389}
2390
2391/**
2392 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2393 * @work: queued work item
2394 *
2395 * Invoked by a work queue tasklet.
2396 */
2397static void xs_tcp_setup_socket(struct work_struct *work)
2398{
2399 struct sock_xprt *transport =
2400 container_of(work, struct sock_xprt, connect_worker.work);
2401 struct socket *sock = transport->sock;
2402 struct rpc_xprt *xprt = &transport->xprt;
2403 int status;
2404 unsigned int pflags = current->flags;
2405
2406 if (atomic_read(&xprt->swapper))
2407 current->flags |= PF_MEMALLOC;
2408
2409 if (xprt_connected(xprt))
2410 goto out;
2411 if (test_and_clear_bit(XPRT_SOCK_CONNECT_SENT,
2412 &transport->sock_state) ||
2413 !sock) {
2414 xs_reset_transport(transport);
2415 sock = xs_create_sock(xprt, transport, xs_addr(xprt)->sa_family,
2416 SOCK_STREAM, IPPROTO_TCP, true);
2417 if (IS_ERR(sock)) {
2418 xprt_wake_pending_tasks(xprt, PTR_ERR(sock));
2419 goto out;
2420 }
2421 }
2422
2423 dprintk("RPC: worker connecting xprt %p via %s to "
2424 "%s (port %s)\n", xprt,
2425 xprt->address_strings[RPC_DISPLAY_PROTO],
2426 xprt->address_strings[RPC_DISPLAY_ADDR],
2427 xprt->address_strings[RPC_DISPLAY_PORT]);
2428
2429 status = xs_tcp_finish_connecting(xprt, sock);
2430 trace_rpc_socket_connect(xprt, sock, status);
2431 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
2432 xprt, -status, xprt_connected(xprt),
2433 sock->sk->sk_state);
2434 switch (status) {
2435 case 0:
2436 case -EINPROGRESS:
2437 /* SYN_SENT! */
2438 set_bit(XPRT_SOCK_CONNECT_SENT, &transport->sock_state);
2439 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2440 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2441 fallthrough;
2442 case -EALREADY:
2443 goto out_unlock;
2444 case -EADDRNOTAVAIL:
2445 /* Source port number is unavailable. Try a new one! */
2446 transport->srcport = 0;
2447 status = -EAGAIN;
2448 break;
2449 case -EPERM:
2450 /* Happens, for instance, if a BPF program is preventing
2451 * the connect. Remap the error so upper layers can better
2452 * deal with it.
2453 */
2454 status = -ECONNREFUSED;
2455 fallthrough;
2456 case -EINVAL:
2457 /* Happens, for instance, if the user specified a link
2458 * local IPv6 address without a scope-id.
2459 */
2460 case -ECONNREFUSED:
2461 case -ECONNRESET:
2462 case -ENETDOWN:
2463 case -ENETUNREACH:
2464 case -EHOSTUNREACH:
2465 case -EADDRINUSE:
2466 case -ENOBUFS:
2467 case -ENOTCONN:
2468 break;
2469 default:
2470 printk("%s: connect returned unhandled error %d\n",
2471 __func__, status);
2472 status = -EAGAIN;
2473 }
2474
2475 /* xs_tcp_force_close() wakes tasks with a fixed error code.
2476 * We need to wake them first to ensure the correct error code.
2477 */
2478 xprt_wake_pending_tasks(xprt, status);
2479 xs_tcp_force_close(xprt);
2480out:
2481 xprt_clear_connecting(xprt);
2482out_unlock:
2483 xprt_unlock_connect(xprt, transport);
2484 current_restore_flags(pflags, PF_MEMALLOC);
2485}
2486
2487/*
2488 * Transfer the connected socket to @upper_transport, then mark that
2489 * xprt CONNECTED.
2490 */
2491static int xs_tcp_tls_finish_connecting(struct rpc_xprt *lower_xprt,
2492 struct sock_xprt *upper_transport)
2493{
2494 struct sock_xprt *lower_transport =
2495 container_of(lower_xprt, struct sock_xprt, xprt);
2496 struct rpc_xprt *upper_xprt = &upper_transport->xprt;
2497
2498 if (!upper_transport->inet) {
2499 struct socket *sock = lower_transport->sock;
2500 struct sock *sk = sock->sk;
2501
2502 /* Avoid temporary address, they are bad for long-lived
2503 * connections such as NFS mounts.
2504 * RFC4941, section 3.6 suggests that:
2505 * Individual applications, which have specific
2506 * knowledge about the normal duration of connections,
2507 * MAY override this as appropriate.
2508 */
2509 if (xs_addr(upper_xprt)->sa_family == PF_INET6)
2510 ip6_sock_set_addr_preferences(sk, IPV6_PREFER_SRC_PUBLIC);
2511
2512 xs_tcp_set_socket_timeouts(upper_xprt, sock);
2513 tcp_sock_set_nodelay(sk);
2514
2515 lock_sock(sk);
2516
2517 /* @sk is already connected, so it now has the RPC callbacks.
2518 * Reach into @lower_transport to save the original ones.
2519 */
2520 upper_transport->old_data_ready = lower_transport->old_data_ready;
2521 upper_transport->old_state_change = lower_transport->old_state_change;
2522 upper_transport->old_write_space = lower_transport->old_write_space;
2523 upper_transport->old_error_report = lower_transport->old_error_report;
2524 sk->sk_user_data = upper_xprt;
2525
2526 /* socket options */
2527 sock_reset_flag(sk, SOCK_LINGER);
2528
2529 xprt_clear_connected(upper_xprt);
2530
2531 upper_transport->sock = sock;
2532 upper_transport->inet = sk;
2533 upper_transport->file = lower_transport->file;
2534
2535 release_sock(sk);
2536
2537 /* Reset lower_transport before shutting down its clnt */
2538 mutex_lock(&lower_transport->recv_mutex);
2539 lower_transport->inet = NULL;
2540 lower_transport->sock = NULL;
2541 lower_transport->file = NULL;
2542
2543 xprt_clear_connected(lower_xprt);
2544 xs_sock_reset_connection_flags(lower_xprt);
2545 xs_stream_reset_connect(lower_transport);
2546 mutex_unlock(&lower_transport->recv_mutex);
2547 }
2548
2549 if (!xprt_bound(upper_xprt))
2550 return -ENOTCONN;
2551
2552 xs_set_memalloc(upper_xprt);
2553
2554 if (!xprt_test_and_set_connected(upper_xprt)) {
2555 upper_xprt->connect_cookie++;
2556 clear_bit(XPRT_SOCK_CONNECTING, &upper_transport->sock_state);
2557 xprt_clear_connecting(upper_xprt);
2558
2559 upper_xprt->stat.connect_count++;
2560 upper_xprt->stat.connect_time += (long)jiffies -
2561 upper_xprt->stat.connect_start;
2562 xs_run_error_worker(upper_transport, XPRT_SOCK_WAKE_PENDING);
2563 }
2564 return 0;
2565}
2566
2567/**
2568 * xs_tls_handshake_done - TLS handshake completion handler
2569 * @data: address of xprt to wake
2570 * @status: status of handshake
2571 * @peerid: serial number of key containing the remote's identity
2572 *
2573 */
2574static void xs_tls_handshake_done(void *data, int status, key_serial_t peerid)
2575{
2576 struct rpc_xprt *lower_xprt = data;
2577 struct sock_xprt *lower_transport =
2578 container_of(lower_xprt, struct sock_xprt, xprt);
2579
2580 switch (status) {
2581 case 0:
2582 case -EACCES:
2583 case -ETIMEDOUT:
2584 lower_transport->xprt_err = status;
2585 break;
2586 default:
2587 lower_transport->xprt_err = -EACCES;
2588 }
2589 complete(&lower_transport->handshake_done);
2590 xprt_put(lower_xprt);
2591}
2592
2593static int xs_tls_handshake_sync(struct rpc_xprt *lower_xprt, struct xprtsec_parms *xprtsec)
2594{
2595 struct sock_xprt *lower_transport =
2596 container_of(lower_xprt, struct sock_xprt, xprt);
2597 struct tls_handshake_args args = {
2598 .ta_sock = lower_transport->sock,
2599 .ta_done = xs_tls_handshake_done,
2600 .ta_data = xprt_get(lower_xprt),
2601 .ta_peername = lower_xprt->servername,
2602 };
2603 struct sock *sk = lower_transport->inet;
2604 int rc;
2605
2606 init_completion(&lower_transport->handshake_done);
2607 set_bit(XPRT_SOCK_IGNORE_RECV, &lower_transport->sock_state);
2608 lower_transport->xprt_err = -ETIMEDOUT;
2609 switch (xprtsec->policy) {
2610 case RPC_XPRTSEC_TLS_ANON:
2611 rc = tls_client_hello_anon(&args, GFP_KERNEL);
2612 if (rc)
2613 goto out_put_xprt;
2614 break;
2615 case RPC_XPRTSEC_TLS_X509:
2616 args.ta_my_cert = xprtsec->cert_serial;
2617 args.ta_my_privkey = xprtsec->privkey_serial;
2618 rc = tls_client_hello_x509(&args, GFP_KERNEL);
2619 if (rc)
2620 goto out_put_xprt;
2621 break;
2622 default:
2623 rc = -EACCES;
2624 goto out_put_xprt;
2625 }
2626
2627 rc = wait_for_completion_interruptible_timeout(&lower_transport->handshake_done,
2628 XS_TLS_HANDSHAKE_TO);
2629 if (rc <= 0) {
2630 tls_handshake_cancel(sk);
2631 if (rc == 0)
2632 rc = -ETIMEDOUT;
2633 goto out_put_xprt;
2634 }
2635
2636 rc = lower_transport->xprt_err;
2637
2638out:
2639 xs_stream_reset_connect(lower_transport);
2640 clear_bit(XPRT_SOCK_IGNORE_RECV, &lower_transport->sock_state);
2641 return rc;
2642
2643out_put_xprt:
2644 xprt_put(lower_xprt);
2645 goto out;
2646}
2647
2648/**
2649 * xs_tcp_tls_setup_socket - establish a TLS session on a TCP socket
2650 * @work: queued work item
2651 *
2652 * Invoked by a work queue tasklet.
2653 *
2654 * For RPC-with-TLS, there is a two-stage connection process.
2655 *
2656 * The "upper-layer xprt" is visible to the RPC consumer. Once it has
2657 * been marked connected, the consumer knows that a TCP connection and
2658 * a TLS session have been established.
2659 *
2660 * A "lower-layer xprt", created in this function, handles the mechanics
2661 * of connecting the TCP socket, performing the RPC_AUTH_TLS probe, and
2662 * then driving the TLS handshake. Once all that is complete, the upper
2663 * layer xprt is marked connected.
2664 */
2665static void xs_tcp_tls_setup_socket(struct work_struct *work)
2666{
2667 struct sock_xprt *upper_transport =
2668 container_of(work, struct sock_xprt, connect_worker.work);
2669 struct rpc_clnt *upper_clnt = upper_transport->clnt;
2670 struct rpc_xprt *upper_xprt = &upper_transport->xprt;
2671 struct rpc_create_args args = {
2672 .net = upper_xprt->xprt_net,
2673 .protocol = upper_xprt->prot,
2674 .address = (struct sockaddr *)&upper_xprt->addr,
2675 .addrsize = upper_xprt->addrlen,
2676 .timeout = upper_clnt->cl_timeout,
2677 .servername = upper_xprt->servername,
2678 .program = upper_clnt->cl_program,
2679 .prognumber = upper_clnt->cl_prog,
2680 .version = upper_clnt->cl_vers,
2681 .authflavor = RPC_AUTH_TLS,
2682 .cred = upper_clnt->cl_cred,
2683 .xprtsec = {
2684 .policy = RPC_XPRTSEC_NONE,
2685 },
2686 .stats = upper_clnt->cl_stats,
2687 };
2688 unsigned int pflags = current->flags;
2689 struct rpc_clnt *lower_clnt;
2690 struct rpc_xprt *lower_xprt;
2691 int status;
2692
2693 if (atomic_read(&upper_xprt->swapper))
2694 current->flags |= PF_MEMALLOC;
2695
2696 xs_stream_start_connect(upper_transport);
2697
2698 /* This implicitly sends an RPC_AUTH_TLS probe */
2699 lower_clnt = rpc_create(&args);
2700 if (IS_ERR(lower_clnt)) {
2701 trace_rpc_tls_unavailable(upper_clnt, upper_xprt);
2702 clear_bit(XPRT_SOCK_CONNECTING, &upper_transport->sock_state);
2703 xprt_clear_connecting(upper_xprt);
2704 xprt_wake_pending_tasks(upper_xprt, PTR_ERR(lower_clnt));
2705 xs_run_error_worker(upper_transport, XPRT_SOCK_WAKE_PENDING);
2706 goto out_unlock;
2707 }
2708
2709 /* RPC_AUTH_TLS probe was successful. Try a TLS handshake on
2710 * the lower xprt.
2711 */
2712 rcu_read_lock();
2713 lower_xprt = rcu_dereference(lower_clnt->cl_xprt);
2714 rcu_read_unlock();
2715
2716 if (wait_on_bit_lock(&lower_xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2717 goto out_unlock;
2718
2719 status = xs_tls_handshake_sync(lower_xprt, &upper_xprt->xprtsec);
2720 if (status) {
2721 trace_rpc_tls_not_started(upper_clnt, upper_xprt);
2722 goto out_close;
2723 }
2724
2725 status = xs_tcp_tls_finish_connecting(lower_xprt, upper_transport);
2726 if (status)
2727 goto out_close;
2728 xprt_release_write(lower_xprt, NULL);
2729
2730 trace_rpc_socket_connect(upper_xprt, upper_transport->sock, 0);
2731 if (!xprt_test_and_set_connected(upper_xprt)) {
2732 upper_xprt->connect_cookie++;
2733 clear_bit(XPRT_SOCK_CONNECTING, &upper_transport->sock_state);
2734 xprt_clear_connecting(upper_xprt);
2735
2736 upper_xprt->stat.connect_count++;
2737 upper_xprt->stat.connect_time += (long)jiffies -
2738 upper_xprt->stat.connect_start;
2739 xs_run_error_worker(upper_transport, XPRT_SOCK_WAKE_PENDING);
2740 }
2741 rpc_shutdown_client(lower_clnt);
2742
2743out_unlock:
2744 current_restore_flags(pflags, PF_MEMALLOC);
2745 upper_transport->clnt = NULL;
2746 xprt_unlock_connect(upper_xprt, upper_transport);
2747 return;
2748
2749out_close:
2750 xprt_release_write(lower_xprt, NULL);
2751 rpc_shutdown_client(lower_clnt);
2752
2753 /* xprt_force_disconnect() wakes tasks with a fixed tk_status code.
2754 * Wake them first here to ensure they get our tk_status code.
2755 */
2756 xprt_wake_pending_tasks(upper_xprt, status);
2757 xs_tcp_force_close(upper_xprt);
2758 xprt_clear_connecting(upper_xprt);
2759 goto out_unlock;
2760}
2761
2762/**
2763 * xs_connect - connect a socket to a remote endpoint
2764 * @xprt: pointer to transport structure
2765 * @task: address of RPC task that manages state of connect request
2766 *
2767 * TCP: If the remote end dropped the connection, delay reconnecting.
2768 *
2769 * UDP socket connects are synchronous, but we use a work queue anyway
2770 * to guarantee that even unprivileged user processes can set up a
2771 * socket on a privileged port.
2772 *
2773 * If a UDP socket connect fails, the delay behavior here prevents
2774 * retry floods (hard mounts).
2775 */
2776static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2777{
2778 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2779 unsigned long delay = 0;
2780
2781 WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2782
2783 if (transport->sock != NULL) {
2784 dprintk("RPC: xs_connect delayed xprt %p for %lu "
2785 "seconds\n", xprt, xprt->reestablish_timeout / HZ);
2786
2787 delay = xprt_reconnect_delay(xprt);
2788 xprt_reconnect_backoff(xprt, XS_TCP_INIT_REEST_TO);
2789
2790 } else
2791 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
2792
2793 transport->clnt = task->tk_client;
2794 queue_delayed_work(xprtiod_workqueue,
2795 &transport->connect_worker,
2796 delay);
2797}
2798
2799static void xs_wake_disconnect(struct sock_xprt *transport)
2800{
2801 if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state))
2802 xs_tcp_force_close(&transport->xprt);
2803}
2804
2805static void xs_wake_write(struct sock_xprt *transport)
2806{
2807 if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state))
2808 xprt_write_space(&transport->xprt);
2809}
2810
2811static void xs_wake_error(struct sock_xprt *transport)
2812{
2813 int sockerr;
2814
2815 if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2816 return;
2817 sockerr = xchg(&transport->xprt_err, 0);
2818 if (sockerr < 0) {
2819 xprt_wake_pending_tasks(&transport->xprt, sockerr);
2820 xs_tcp_force_close(&transport->xprt);
2821 }
2822}
2823
2824static void xs_wake_pending(struct sock_xprt *transport)
2825{
2826 if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING, &transport->sock_state))
2827 xprt_wake_pending_tasks(&transport->xprt, -EAGAIN);
2828}
2829
2830static void xs_error_handle(struct work_struct *work)
2831{
2832 struct sock_xprt *transport = container_of(work,
2833 struct sock_xprt, error_worker);
2834
2835 xs_wake_disconnect(transport);
2836 xs_wake_write(transport);
2837 xs_wake_error(transport);
2838 xs_wake_pending(transport);
2839}
2840
2841/**
2842 * xs_local_print_stats - display AF_LOCAL socket-specific stats
2843 * @xprt: rpc_xprt struct containing statistics
2844 * @seq: output file
2845 *
2846 */
2847static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2848{
2849 long idle_time = 0;
2850
2851 if (xprt_connected(xprt))
2852 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2853
2854 seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2855 "%llu %llu %lu %llu %llu\n",
2856 xprt->stat.bind_count,
2857 xprt->stat.connect_count,
2858 xprt->stat.connect_time / HZ,
2859 idle_time,
2860 xprt->stat.sends,
2861 xprt->stat.recvs,
2862 xprt->stat.bad_xids,
2863 xprt->stat.req_u,
2864 xprt->stat.bklog_u,
2865 xprt->stat.max_slots,
2866 xprt->stat.sending_u,
2867 xprt->stat.pending_u);
2868}
2869
2870/**
2871 * xs_udp_print_stats - display UDP socket-specific stats
2872 * @xprt: rpc_xprt struct containing statistics
2873 * @seq: output file
2874 *
2875 */
2876static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2877{
2878 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2879
2880 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2881 "%lu %llu %llu\n",
2882 transport->srcport,
2883 xprt->stat.bind_count,
2884 xprt->stat.sends,
2885 xprt->stat.recvs,
2886 xprt->stat.bad_xids,
2887 xprt->stat.req_u,
2888 xprt->stat.bklog_u,
2889 xprt->stat.max_slots,
2890 xprt->stat.sending_u,
2891 xprt->stat.pending_u);
2892}
2893
2894/**
2895 * xs_tcp_print_stats - display TCP socket-specific stats
2896 * @xprt: rpc_xprt struct containing statistics
2897 * @seq: output file
2898 *
2899 */
2900static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2901{
2902 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2903 long idle_time = 0;
2904
2905 if (xprt_connected(xprt))
2906 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2907
2908 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2909 "%llu %llu %lu %llu %llu\n",
2910 transport->srcport,
2911 xprt->stat.bind_count,
2912 xprt->stat.connect_count,
2913 xprt->stat.connect_time / HZ,
2914 idle_time,
2915 xprt->stat.sends,
2916 xprt->stat.recvs,
2917 xprt->stat.bad_xids,
2918 xprt->stat.req_u,
2919 xprt->stat.bklog_u,
2920 xprt->stat.max_slots,
2921 xprt->stat.sending_u,
2922 xprt->stat.pending_u);
2923}
2924
2925/*
2926 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2927 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2928 * to use the server side send routines.
2929 */
2930static int bc_malloc(struct rpc_task *task)
2931{
2932 struct rpc_rqst *rqst = task->tk_rqstp;
2933 size_t size = rqst->rq_callsize;
2934 struct page *page;
2935 struct rpc_buffer *buf;
2936
2937 if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2938 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2939 size);
2940 return -EINVAL;
2941 }
2942
2943 page = alloc_page(GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN);
2944 if (!page)
2945 return -ENOMEM;
2946
2947 buf = page_address(page);
2948 buf->len = PAGE_SIZE;
2949
2950 rqst->rq_buffer = buf->data;
2951 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2952 return 0;
2953}
2954
2955/*
2956 * Free the space allocated in the bc_alloc routine
2957 */
2958static void bc_free(struct rpc_task *task)
2959{
2960 void *buffer = task->tk_rqstp->rq_buffer;
2961 struct rpc_buffer *buf;
2962
2963 buf = container_of(buffer, struct rpc_buffer, data);
2964 free_page((unsigned long)buf);
2965}
2966
2967static int bc_sendto(struct rpc_rqst *req)
2968{
2969 struct xdr_buf *xdr = &req->rq_snd_buf;
2970 struct sock_xprt *transport =
2971 container_of(req->rq_xprt, struct sock_xprt, xprt);
2972 struct msghdr msg = {
2973 .msg_flags = 0,
2974 };
2975 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
2976 (u32)xdr->len);
2977 unsigned int sent = 0;
2978 int err;
2979
2980 req->rq_xtime = ktime_get();
2981 err = xdr_alloc_bvec(xdr, rpc_task_gfp_mask());
2982 if (err < 0)
2983 return err;
2984 err = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, marker, &sent);
2985 xdr_free_bvec(xdr);
2986 if (err < 0 || sent != (xdr->len + sizeof(marker)))
2987 return -EAGAIN;
2988 return sent;
2989}
2990
2991/**
2992 * bc_send_request - Send a backchannel Call on a TCP socket
2993 * @req: rpc_rqst containing Call message to be sent
2994 *
2995 * xpt_mutex ensures @rqstp's whole message is written to the socket
2996 * without interruption.
2997 *
2998 * Return values:
2999 * %0 if the message was sent successfully
3000 * %ENOTCONN if the message was not sent
3001 */
3002static int bc_send_request(struct rpc_rqst *req)
3003{
3004 struct svc_xprt *xprt;
3005 int len;
3006
3007 /*
3008 * Get the server socket associated with this callback xprt
3009 */
3010 xprt = req->rq_xprt->bc_xprt;
3011
3012 /*
3013 * Grab the mutex to serialize data as the connection is shared
3014 * with the fore channel
3015 */
3016 mutex_lock(&xprt->xpt_mutex);
3017 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
3018 len = -ENOTCONN;
3019 else
3020 len = bc_sendto(req);
3021 mutex_unlock(&xprt->xpt_mutex);
3022
3023 if (len > 0)
3024 len = 0;
3025
3026 return len;
3027}
3028
3029static void bc_close(struct rpc_xprt *xprt)
3030{
3031 xprt_disconnect_done(xprt);
3032}
3033
3034static void bc_destroy(struct rpc_xprt *xprt)
3035{
3036 dprintk("RPC: bc_destroy xprt %p\n", xprt);
3037
3038 xs_xprt_free(xprt);
3039 module_put(THIS_MODULE);
3040}
3041
3042static const struct rpc_xprt_ops xs_local_ops = {
3043 .reserve_xprt = xprt_reserve_xprt,
3044 .release_xprt = xprt_release_xprt,
3045 .alloc_slot = xprt_alloc_slot,
3046 .free_slot = xprt_free_slot,
3047 .rpcbind = xs_local_rpcbind,
3048 .set_port = xs_local_set_port,
3049 .connect = xs_local_connect,
3050 .buf_alloc = rpc_malloc,
3051 .buf_free = rpc_free,
3052 .prepare_request = xs_stream_prepare_request,
3053 .send_request = xs_local_send_request,
3054 .abort_send_request = xs_stream_abort_send_request,
3055 .wait_for_reply_request = xprt_wait_for_reply_request_def,
3056 .close = xs_close,
3057 .destroy = xs_destroy,
3058 .print_stats = xs_local_print_stats,
3059 .enable_swap = xs_enable_swap,
3060 .disable_swap = xs_disable_swap,
3061};
3062
3063static const struct rpc_xprt_ops xs_udp_ops = {
3064 .set_buffer_size = xs_udp_set_buffer_size,
3065 .reserve_xprt = xprt_reserve_xprt_cong,
3066 .release_xprt = xprt_release_xprt_cong,
3067 .alloc_slot = xprt_alloc_slot,
3068 .free_slot = xprt_free_slot,
3069 .rpcbind = rpcb_getport_async,
3070 .set_port = xs_set_port,
3071 .connect = xs_connect,
3072 .get_srcaddr = xs_sock_srcaddr,
3073 .get_srcport = xs_sock_srcport,
3074 .buf_alloc = rpc_malloc,
3075 .buf_free = rpc_free,
3076 .send_request = xs_udp_send_request,
3077 .wait_for_reply_request = xprt_wait_for_reply_request_rtt,
3078 .timer = xs_udp_timer,
3079 .release_request = xprt_release_rqst_cong,
3080 .close = xs_close,
3081 .destroy = xs_destroy,
3082 .print_stats = xs_udp_print_stats,
3083 .enable_swap = xs_enable_swap,
3084 .disable_swap = xs_disable_swap,
3085 .inject_disconnect = xs_inject_disconnect,
3086};
3087
3088static const struct rpc_xprt_ops xs_tcp_ops = {
3089 .reserve_xprt = xprt_reserve_xprt,
3090 .release_xprt = xprt_release_xprt,
3091 .alloc_slot = xprt_alloc_slot,
3092 .free_slot = xprt_free_slot,
3093 .rpcbind = rpcb_getport_async,
3094 .set_port = xs_set_port,
3095 .connect = xs_connect,
3096 .get_srcaddr = xs_sock_srcaddr,
3097 .get_srcport = xs_sock_srcport,
3098 .buf_alloc = rpc_malloc,
3099 .buf_free = rpc_free,
3100 .prepare_request = xs_stream_prepare_request,
3101 .send_request = xs_tcp_send_request,
3102 .abort_send_request = xs_stream_abort_send_request,
3103 .wait_for_reply_request = xprt_wait_for_reply_request_def,
3104 .close = xs_tcp_shutdown,
3105 .destroy = xs_destroy,
3106 .set_connect_timeout = xs_tcp_set_connect_timeout,
3107 .print_stats = xs_tcp_print_stats,
3108 .enable_swap = xs_enable_swap,
3109 .disable_swap = xs_disable_swap,
3110 .inject_disconnect = xs_inject_disconnect,
3111#ifdef CONFIG_SUNRPC_BACKCHANNEL
3112 .bc_setup = xprt_setup_bc,
3113 .bc_maxpayload = xs_tcp_bc_maxpayload,
3114 .bc_num_slots = xprt_bc_max_slots,
3115 .bc_free_rqst = xprt_free_bc_rqst,
3116 .bc_destroy = xprt_destroy_bc,
3117#endif
3118};
3119
3120/*
3121 * The rpc_xprt_ops for the server backchannel
3122 */
3123
3124static const struct rpc_xprt_ops bc_tcp_ops = {
3125 .reserve_xprt = xprt_reserve_xprt,
3126 .release_xprt = xprt_release_xprt,
3127 .alloc_slot = xprt_alloc_slot,
3128 .free_slot = xprt_free_slot,
3129 .buf_alloc = bc_malloc,
3130 .buf_free = bc_free,
3131 .send_request = bc_send_request,
3132 .wait_for_reply_request = xprt_wait_for_reply_request_def,
3133 .close = bc_close,
3134 .destroy = bc_destroy,
3135 .print_stats = xs_tcp_print_stats,
3136 .enable_swap = xs_enable_swap,
3137 .disable_swap = xs_disable_swap,
3138 .inject_disconnect = xs_inject_disconnect,
3139};
3140
3141static int xs_init_anyaddr(const int family, struct sockaddr *sap)
3142{
3143 static const struct sockaddr_in sin = {
3144 .sin_family = AF_INET,
3145 .sin_addr.s_addr = htonl(INADDR_ANY),
3146 };
3147 static const struct sockaddr_in6 sin6 = {
3148 .sin6_family = AF_INET6,
3149 .sin6_addr = IN6ADDR_ANY_INIT,
3150 };
3151
3152 switch (family) {
3153 case AF_LOCAL:
3154 break;
3155 case AF_INET:
3156 memcpy(sap, &sin, sizeof(sin));
3157 break;
3158 case AF_INET6:
3159 memcpy(sap, &sin6, sizeof(sin6));
3160 break;
3161 default:
3162 dprintk("RPC: %s: Bad address family\n", __func__);
3163 return -EAFNOSUPPORT;
3164 }
3165 return 0;
3166}
3167
3168static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
3169 unsigned int slot_table_size,
3170 unsigned int max_slot_table_size)
3171{
3172 struct rpc_xprt *xprt;
3173 struct sock_xprt *new;
3174
3175 if (args->addrlen > sizeof(xprt->addr)) {
3176 dprintk("RPC: xs_setup_xprt: address too large\n");
3177 return ERR_PTR(-EBADF);
3178 }
3179
3180 xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
3181 max_slot_table_size);
3182 if (xprt == NULL) {
3183 dprintk("RPC: xs_setup_xprt: couldn't allocate "
3184 "rpc_xprt\n");
3185 return ERR_PTR(-ENOMEM);
3186 }
3187
3188 new = container_of(xprt, struct sock_xprt, xprt);
3189 mutex_init(&new->recv_mutex);
3190 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
3191 xprt->addrlen = args->addrlen;
3192 if (args->srcaddr)
3193 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
3194 else {
3195 int err;
3196 err = xs_init_anyaddr(args->dstaddr->sa_family,
3197 (struct sockaddr *)&new->srcaddr);
3198 if (err != 0) {
3199 xprt_free(xprt);
3200 return ERR_PTR(err);
3201 }
3202 }
3203
3204 return xprt;
3205}
3206
3207static const struct rpc_timeout xs_local_default_timeout = {
3208 .to_initval = 10 * HZ,
3209 .to_maxval = 10 * HZ,
3210 .to_retries = 2,
3211};
3212
3213/**
3214 * xs_setup_local - Set up transport to use an AF_LOCAL socket
3215 * @args: rpc transport creation arguments
3216 *
3217 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
3218 */
3219static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
3220{
3221 struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
3222 struct sock_xprt *transport;
3223 struct rpc_xprt *xprt;
3224 struct rpc_xprt *ret;
3225
3226 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3227 xprt_max_tcp_slot_table_entries);
3228 if (IS_ERR(xprt))
3229 return xprt;
3230 transport = container_of(xprt, struct sock_xprt, xprt);
3231
3232 xprt->prot = 0;
3233 xprt->xprt_class = &xs_local_transport;
3234 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3235
3236 xprt->bind_timeout = XS_BIND_TO;
3237 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3238 xprt->idle_timeout = XS_IDLE_DISC_TO;
3239
3240 xprt->ops = &xs_local_ops;
3241 xprt->timeout = &xs_local_default_timeout;
3242
3243 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3244 INIT_WORK(&transport->error_worker, xs_error_handle);
3245 INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket);
3246
3247 switch (sun->sun_family) {
3248 case AF_LOCAL:
3249 if (sun->sun_path[0] != '/' && sun->sun_path[0] != '\0') {
3250 dprintk("RPC: bad AF_LOCAL address: %s\n",
3251 sun->sun_path);
3252 ret = ERR_PTR(-EINVAL);
3253 goto out_err;
3254 }
3255 xprt_set_bound(xprt);
3256 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
3257 break;
3258 default:
3259 ret = ERR_PTR(-EAFNOSUPPORT);
3260 goto out_err;
3261 }
3262
3263 dprintk("RPC: set up xprt to %s via AF_LOCAL\n",
3264 xprt->address_strings[RPC_DISPLAY_ADDR]);
3265
3266 if (try_module_get(THIS_MODULE))
3267 return xprt;
3268 ret = ERR_PTR(-EINVAL);
3269out_err:
3270 xs_xprt_free(xprt);
3271 return ret;
3272}
3273
3274static const struct rpc_timeout xs_udp_default_timeout = {
3275 .to_initval = 5 * HZ,
3276 .to_maxval = 30 * HZ,
3277 .to_increment = 5 * HZ,
3278 .to_retries = 5,
3279};
3280
3281/**
3282 * xs_setup_udp - Set up transport to use a UDP socket
3283 * @args: rpc transport creation arguments
3284 *
3285 */
3286static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
3287{
3288 struct sockaddr *addr = args->dstaddr;
3289 struct rpc_xprt *xprt;
3290 struct sock_xprt *transport;
3291 struct rpc_xprt *ret;
3292
3293 xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
3294 xprt_udp_slot_table_entries);
3295 if (IS_ERR(xprt))
3296 return xprt;
3297 transport = container_of(xprt, struct sock_xprt, xprt);
3298
3299 xprt->prot = IPPROTO_UDP;
3300 xprt->xprt_class = &xs_udp_transport;
3301 /* XXX: header size can vary due to auth type, IPv6, etc. */
3302 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
3303
3304 xprt->bind_timeout = XS_BIND_TO;
3305 xprt->reestablish_timeout = XS_UDP_REEST_TO;
3306 xprt->idle_timeout = XS_IDLE_DISC_TO;
3307
3308 xprt->ops = &xs_udp_ops;
3309
3310 xprt->timeout = &xs_udp_default_timeout;
3311
3312 INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
3313 INIT_WORK(&transport->error_worker, xs_error_handle);
3314 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
3315
3316 switch (addr->sa_family) {
3317 case AF_INET:
3318 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3319 xprt_set_bound(xprt);
3320
3321 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
3322 break;
3323 case AF_INET6:
3324 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3325 xprt_set_bound(xprt);
3326
3327 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
3328 break;
3329 default:
3330 ret = ERR_PTR(-EAFNOSUPPORT);
3331 goto out_err;
3332 }
3333
3334 if (xprt_bound(xprt))
3335 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3336 xprt->address_strings[RPC_DISPLAY_ADDR],
3337 xprt->address_strings[RPC_DISPLAY_PORT],
3338 xprt->address_strings[RPC_DISPLAY_PROTO]);
3339 else
3340 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
3341 xprt->address_strings[RPC_DISPLAY_ADDR],
3342 xprt->address_strings[RPC_DISPLAY_PROTO]);
3343
3344 if (try_module_get(THIS_MODULE))
3345 return xprt;
3346 ret = ERR_PTR(-EINVAL);
3347out_err:
3348 xs_xprt_free(xprt);
3349 return ret;
3350}
3351
3352static const struct rpc_timeout xs_tcp_default_timeout = {
3353 .to_initval = 60 * HZ,
3354 .to_maxval = 60 * HZ,
3355 .to_retries = 2,
3356};
3357
3358/**
3359 * xs_setup_tcp - Set up transport to use a TCP socket
3360 * @args: rpc transport creation arguments
3361 *
3362 */
3363static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
3364{
3365 struct sockaddr *addr = args->dstaddr;
3366 struct rpc_xprt *xprt;
3367 struct sock_xprt *transport;
3368 struct rpc_xprt *ret;
3369 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
3370
3371 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
3372 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
3373
3374 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3375 max_slot_table_size);
3376 if (IS_ERR(xprt))
3377 return xprt;
3378 transport = container_of(xprt, struct sock_xprt, xprt);
3379
3380 xprt->prot = IPPROTO_TCP;
3381 xprt->xprt_class = &xs_tcp_transport;
3382 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3383
3384 xprt->bind_timeout = XS_BIND_TO;
3385 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3386 xprt->idle_timeout = XS_IDLE_DISC_TO;
3387
3388 xprt->ops = &xs_tcp_ops;
3389 xprt->timeout = &xs_tcp_default_timeout;
3390
3391 xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3392 if (args->reconnect_timeout)
3393 xprt->max_reconnect_timeout = args->reconnect_timeout;
3394
3395 xprt->connect_timeout = xprt->timeout->to_initval *
3396 (xprt->timeout->to_retries + 1);
3397 if (args->connect_timeout)
3398 xs_tcp_do_set_connect_timeout(xprt, args->connect_timeout);
3399
3400 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3401 INIT_WORK(&transport->error_worker, xs_error_handle);
3402 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
3403
3404 switch (addr->sa_family) {
3405 case AF_INET:
3406 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3407 xprt_set_bound(xprt);
3408
3409 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3410 break;
3411 case AF_INET6:
3412 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3413 xprt_set_bound(xprt);
3414
3415 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3416 break;
3417 default:
3418 ret = ERR_PTR(-EAFNOSUPPORT);
3419 goto out_err;
3420 }
3421
3422 if (xprt_bound(xprt))
3423 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3424 xprt->address_strings[RPC_DISPLAY_ADDR],
3425 xprt->address_strings[RPC_DISPLAY_PORT],
3426 xprt->address_strings[RPC_DISPLAY_PROTO]);
3427 else
3428 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
3429 xprt->address_strings[RPC_DISPLAY_ADDR],
3430 xprt->address_strings[RPC_DISPLAY_PROTO]);
3431
3432 if (try_module_get(THIS_MODULE))
3433 return xprt;
3434 ret = ERR_PTR(-EINVAL);
3435out_err:
3436 xs_xprt_free(xprt);
3437 return ret;
3438}
3439
3440/**
3441 * xs_setup_tcp_tls - Set up transport to use a TCP with TLS
3442 * @args: rpc transport creation arguments
3443 *
3444 */
3445static struct rpc_xprt *xs_setup_tcp_tls(struct xprt_create *args)
3446{
3447 struct sockaddr *addr = args->dstaddr;
3448 struct rpc_xprt *xprt;
3449 struct sock_xprt *transport;
3450 struct rpc_xprt *ret;
3451 unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
3452
3453 if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
3454 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
3455
3456 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3457 max_slot_table_size);
3458 if (IS_ERR(xprt))
3459 return xprt;
3460 transport = container_of(xprt, struct sock_xprt, xprt);
3461
3462 xprt->prot = IPPROTO_TCP;
3463 xprt->xprt_class = &xs_tcp_transport;
3464 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3465
3466 xprt->bind_timeout = XS_BIND_TO;
3467 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3468 xprt->idle_timeout = XS_IDLE_DISC_TO;
3469
3470 xprt->ops = &xs_tcp_ops;
3471 xprt->timeout = &xs_tcp_default_timeout;
3472
3473 xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3474 xprt->connect_timeout = xprt->timeout->to_initval *
3475 (xprt->timeout->to_retries + 1);
3476
3477 INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3478 INIT_WORK(&transport->error_worker, xs_error_handle);
3479
3480 switch (args->xprtsec.policy) {
3481 case RPC_XPRTSEC_TLS_ANON:
3482 case RPC_XPRTSEC_TLS_X509:
3483 xprt->xprtsec = args->xprtsec;
3484 INIT_DELAYED_WORK(&transport->connect_worker,
3485 xs_tcp_tls_setup_socket);
3486 break;
3487 default:
3488 ret = ERR_PTR(-EACCES);
3489 goto out_err;
3490 }
3491
3492 switch (addr->sa_family) {
3493 case AF_INET:
3494 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3495 xprt_set_bound(xprt);
3496
3497 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3498 break;
3499 case AF_INET6:
3500 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3501 xprt_set_bound(xprt);
3502
3503 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3504 break;
3505 default:
3506 ret = ERR_PTR(-EAFNOSUPPORT);
3507 goto out_err;
3508 }
3509
3510 if (xprt_bound(xprt))
3511 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3512 xprt->address_strings[RPC_DISPLAY_ADDR],
3513 xprt->address_strings[RPC_DISPLAY_PORT],
3514 xprt->address_strings[RPC_DISPLAY_PROTO]);
3515 else
3516 dprintk("RPC: set up xprt to %s (autobind) via %s\n",
3517 xprt->address_strings[RPC_DISPLAY_ADDR],
3518 xprt->address_strings[RPC_DISPLAY_PROTO]);
3519
3520 if (try_module_get(THIS_MODULE))
3521 return xprt;
3522 ret = ERR_PTR(-EINVAL);
3523out_err:
3524 xs_xprt_free(xprt);
3525 return ret;
3526}
3527
3528/**
3529 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3530 * @args: rpc transport creation arguments
3531 *
3532 */
3533static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3534{
3535 struct sockaddr *addr = args->dstaddr;
3536 struct rpc_xprt *xprt;
3537 struct sock_xprt *transport;
3538 struct svc_sock *bc_sock;
3539 struct rpc_xprt *ret;
3540
3541 xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3542 xprt_tcp_slot_table_entries);
3543 if (IS_ERR(xprt))
3544 return xprt;
3545 transport = container_of(xprt, struct sock_xprt, xprt);
3546
3547 xprt->prot = IPPROTO_TCP;
3548 xprt->xprt_class = &xs_bc_tcp_transport;
3549 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3550 xprt->timeout = &xs_tcp_default_timeout;
3551
3552 /* backchannel */
3553 xprt_set_bound(xprt);
3554 xprt->bind_timeout = 0;
3555 xprt->reestablish_timeout = 0;
3556 xprt->idle_timeout = 0;
3557
3558 xprt->ops = &bc_tcp_ops;
3559
3560 switch (addr->sa_family) {
3561 case AF_INET:
3562 xs_format_peer_addresses(xprt, "tcp",
3563 RPCBIND_NETID_TCP);
3564 break;
3565 case AF_INET6:
3566 xs_format_peer_addresses(xprt, "tcp",
3567 RPCBIND_NETID_TCP6);
3568 break;
3569 default:
3570 ret = ERR_PTR(-EAFNOSUPPORT);
3571 goto out_err;
3572 }
3573
3574 dprintk("RPC: set up xprt to %s (port %s) via %s\n",
3575 xprt->address_strings[RPC_DISPLAY_ADDR],
3576 xprt->address_strings[RPC_DISPLAY_PORT],
3577 xprt->address_strings[RPC_DISPLAY_PROTO]);
3578
3579 /*
3580 * Once we've associated a backchannel xprt with a connection,
3581 * we want to keep it around as long as the connection lasts,
3582 * in case we need to start using it for a backchannel again;
3583 * this reference won't be dropped until bc_xprt is destroyed.
3584 */
3585 xprt_get(xprt);
3586 args->bc_xprt->xpt_bc_xprt = xprt;
3587 xprt->bc_xprt = args->bc_xprt;
3588 bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3589 transport->sock = bc_sock->sk_sock;
3590 transport->inet = bc_sock->sk_sk;
3591
3592 /*
3593 * Since we don't want connections for the backchannel, we set
3594 * the xprt status to connected
3595 */
3596 xprt_set_connected(xprt);
3597
3598 if (try_module_get(THIS_MODULE))
3599 return xprt;
3600
3601 args->bc_xprt->xpt_bc_xprt = NULL;
3602 args->bc_xprt->xpt_bc_xps = NULL;
3603 xprt_put(xprt);
3604 ret = ERR_PTR(-EINVAL);
3605out_err:
3606 xs_xprt_free(xprt);
3607 return ret;
3608}
3609
3610static struct xprt_class xs_local_transport = {
3611 .list = LIST_HEAD_INIT(xs_local_transport.list),
3612 .name = "named UNIX socket",
3613 .owner = THIS_MODULE,
3614 .ident = XPRT_TRANSPORT_LOCAL,
3615 .setup = xs_setup_local,
3616 .netid = { "" },
3617};
3618
3619static struct xprt_class xs_udp_transport = {
3620 .list = LIST_HEAD_INIT(xs_udp_transport.list),
3621 .name = "udp",
3622 .owner = THIS_MODULE,
3623 .ident = XPRT_TRANSPORT_UDP,
3624 .setup = xs_setup_udp,
3625 .netid = { "udp", "udp6", "" },
3626};
3627
3628static struct xprt_class xs_tcp_transport = {
3629 .list = LIST_HEAD_INIT(xs_tcp_transport.list),
3630 .name = "tcp",
3631 .owner = THIS_MODULE,
3632 .ident = XPRT_TRANSPORT_TCP,
3633 .setup = xs_setup_tcp,
3634 .netid = { "tcp", "tcp6", "" },
3635};
3636
3637static struct xprt_class xs_tcp_tls_transport = {
3638 .list = LIST_HEAD_INIT(xs_tcp_tls_transport.list),
3639 .name = "tcp-with-tls",
3640 .owner = THIS_MODULE,
3641 .ident = XPRT_TRANSPORT_TCP_TLS,
3642 .setup = xs_setup_tcp_tls,
3643 .netid = { "tcp", "tcp6", "" },
3644};
3645
3646static struct xprt_class xs_bc_tcp_transport = {
3647 .list = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3648 .name = "tcp NFSv4.1 backchannel",
3649 .owner = THIS_MODULE,
3650 .ident = XPRT_TRANSPORT_BC_TCP,
3651 .setup = xs_setup_bc_tcp,
3652 .netid = { "" },
3653};
3654
3655/**
3656 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3657 *
3658 */
3659int init_socket_xprt(void)
3660{
3661 if (!sunrpc_table_header)
3662 sunrpc_table_header = register_sysctl("sunrpc", xs_tunables_table);
3663
3664 xprt_register_transport(&xs_local_transport);
3665 xprt_register_transport(&xs_udp_transport);
3666 xprt_register_transport(&xs_tcp_transport);
3667 xprt_register_transport(&xs_tcp_tls_transport);
3668 xprt_register_transport(&xs_bc_tcp_transport);
3669
3670 return 0;
3671}
3672
3673/**
3674 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3675 *
3676 */
3677void cleanup_socket_xprt(void)
3678{
3679 if (sunrpc_table_header) {
3680 unregister_sysctl_table(sunrpc_table_header);
3681 sunrpc_table_header = NULL;
3682 }
3683
3684 xprt_unregister_transport(&xs_local_transport);
3685 xprt_unregister_transport(&xs_udp_transport);
3686 xprt_unregister_transport(&xs_tcp_transport);
3687 xprt_unregister_transport(&xs_tcp_tls_transport);
3688 xprt_unregister_transport(&xs_bc_tcp_transport);
3689}
3690
3691static int param_set_portnr(const char *val, const struct kernel_param *kp)
3692{
3693 return param_set_uint_minmax(val, kp,
3694 RPC_MIN_RESVPORT,
3695 RPC_MAX_RESVPORT);
3696}
3697
3698static const struct kernel_param_ops param_ops_portnr = {
3699 .set = param_set_portnr,
3700 .get = param_get_uint,
3701};
3702
3703#define param_check_portnr(name, p) \
3704 __param_check(name, p, unsigned int);
3705
3706module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3707module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3708
3709static int param_set_slot_table_size(const char *val,
3710 const struct kernel_param *kp)
3711{
3712 return param_set_uint_minmax(val, kp,
3713 RPC_MIN_SLOT_TABLE,
3714 RPC_MAX_SLOT_TABLE);
3715}
3716
3717static const struct kernel_param_ops param_ops_slot_table_size = {
3718 .set = param_set_slot_table_size,
3719 .get = param_get_uint,
3720};
3721
3722#define param_check_slot_table_size(name, p) \
3723 __param_check(name, p, unsigned int);
3724
3725static int param_set_max_slot_table_size(const char *val,
3726 const struct kernel_param *kp)
3727{
3728 return param_set_uint_minmax(val, kp,
3729 RPC_MIN_SLOT_TABLE,
3730 RPC_MAX_SLOT_TABLE_LIMIT);
3731}
3732
3733static const struct kernel_param_ops param_ops_max_slot_table_size = {
3734 .set = param_set_max_slot_table_size,
3735 .get = param_get_uint,
3736};
3737
3738#define param_check_max_slot_table_size(name, p) \
3739 __param_check(name, p, unsigned int);
3740
3741module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3742 slot_table_size, 0644);
3743module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3744 max_slot_table_size, 0644);
3745module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3746 slot_table_size, 0644);