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